1 /*	$NetBSD: crunchide.c,v 1.8 1997/11/01 06:51:45 lukem Exp $	*/
2 /*
3  * Copyright (c) 1997 Christopher G. Demetriou.  All rights reserved.
4  * Copyright (c) 1994 University of Maryland
5  * All Rights Reserved.
6  *
7  * Permission to use, copy, modify, distribute, and sell this software and its
8  * documentation for any purpose is hereby granted without fee, provided that
9  * the above copyright notice appear in all copies and that both that
10  * copyright notice and this permission notice appear in supporting
11  * documentation, and that the name of U.M. not be used in advertising or
12  * publicity pertaining to distribution of the software without specific,
13  * written prior permission.  U.M. makes no representations about the
14  * suitability of this software for any purpose.  It is provided "as is"
15  * without express or implied warranty.
16  *
17  * U.M. DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING ALL
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO EVENT SHALL U.M.
19  * BE LIABLE FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
20  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
21  * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
22  * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
23  *
24  * Author: James da Silva, Systems Design and Analysis Group
25  *			   Computer Science Department
26  *			   University of Maryland at College Park
27  */
28 /*
29  * crunchide.c - tiptoes through an a.out symbol table, hiding all defined
30  *	global symbols.  Allows the user to supply a "keep list" of symbols
31  *	that are not to be hidden.  This program relies on the use of the
32  * 	linker's -dc flag to actually put global bss data into the file's
33  * 	bss segment (rather than leaving it as undefined "common" data).
34  *
35  * 	The point of all this is to allow multiple programs to be linked
36  *	together without getting multiple-defined errors.
37  *
38  *	For example, consider a program "foo.c".  It can be linked with a
39  *	small stub routine, called "foostub.c", eg:
40  *	    int foo_main(int argc, char **argv){ return main(argc, argv); }
41  *      like so:
42  *	    cc -c foo.c foostub.c
43  *	    ld -dc -r foo.o foostub.o -o foo.combined.o
44  *	    crunchide -k _foo_main foo.combined.o
45  *	at this point, foo.combined.o can be linked with another program
46  * 	and invoked with "foo_main(argc, argv)".  foo's main() and any
47  * 	other globals are hidden and will not conflict with other symbols.
48  *
49  * TODO:
50  *	- resolve the theoretical hanging reloc problem (see check_reloc()
51  *	  below). I have yet to see this problem actually occur in any real
52  *	  program. In what cases will gcc/gas generate code that needs a
53  *	  relative reloc from a global symbol, other than PIC?  The
54  *	  solution is to not hide the symbol from the linker in this case,
55  *	  but to generate some random name for it so that it doesn't link
56  *	  with anything but holds the place for the reloc.
57  *      - arrange that all the BSS segments start at the same address, so
58  *	  that the final crunched binary BSS size is the max of all the
59  *	  component programs' BSS sizes, rather than their sum.
60  */
61 
62 #include <sys/types.h>
63 #include <sys/stat.h>
64 #include <sys/errno.h>
65 #include <unistd.h>
66 #include <stdio.h>
67 #include <stdlib.h>
68 #include <string.h>
69 #include <fcntl.h>
70 #include <a.out.h>
71 
72 #include "extern.h"
73 
74 static char *pname = "crunchide";
75 
76 static void usage(void) __dead2;
77 
78 static void add_to_keep_list(char *symbol);
79 static void add_file_to_keep_list(char *filename);
80 
81 static int hide_syms(const char *filename);
82 
83 static int verbose;
84 
85 int
86 main(int argc, char **argv)
87 {
88     int ch, errors;
89 
90     if(argc > 0) pname = argv[0];
91 
92     while ((ch = getopt(argc, argv, "k:f:v")) != -1)
93 	switch(ch) {
94 	case 'k':
95 	    add_to_keep_list(optarg);
96 	    break;
97 	case 'f':
98 	    add_file_to_keep_list(optarg);
99 	    break;
100 	case 'v':
101 	    verbose = 1;
102 	    break;
103 	default:
104 	    usage();
105 	}
106 
107     argc -= optind;
108     argv += optind;
109 
110     if(argc == 0) usage();
111 
112     errors = 0;
113     while(argc) {
114 	if (hide_syms(*argv))
115 		errors = 1;
116 	argc--, argv++;
117     }
118 
119     return errors;
120 }
121 
122 static void
123 usage(void)
124 {
125     fprintf(stderr,
126 	    "usage: %s [-k <symbol-name>] [-f <keep-list-file>] <files> ...\n",
127 	    pname);
128     exit(1);
129 }
130 
131 /* ---------------------------- */
132 
133 static struct keep {
134     struct keep *next;
135     char *sym;
136 } *keep_list;
137 
138 static void
139 add_to_keep_list(char *symbol)
140 {
141     struct keep *newp, *prevp, *curp;
142     int cmp;
143 
144     cmp = 0;
145 
146     for(curp = keep_list, prevp = NULL; curp; prevp = curp, curp = curp->next)
147 	if((cmp = strcmp(symbol, curp->sym)) <= 0) break;
148 
149     if(curp && cmp == 0)
150 	return;	/* already in table */
151 
152     newp = (struct keep *) malloc(sizeof(struct keep));
153     if(newp) newp->sym = strdup(symbol);
154     if(newp == NULL || newp->sym == NULL) {
155 	fprintf(stderr, "%s: out of memory for keep list\n", pname);
156 	exit(1);
157     }
158 
159     newp->next = curp;
160     if(prevp) prevp->next = newp;
161     else keep_list = newp;
162 }
163 
164 int
165 in_keep_list(const char *symbol)
166 {
167     struct keep *curp;
168     int cmp;
169 
170     cmp = 0;
171 
172     for(curp = keep_list; curp; curp = curp->next)
173 	if((cmp = strcmp(symbol, curp->sym)) <= 0) break;
174 
175     return curp && cmp == 0;
176 }
177 
178 static void
179 add_file_to_keep_list(char *filename)
180 {
181     FILE *keepf;
182     char symbol[1024];
183     int len;
184 
185     if((keepf = fopen(filename, "r")) == NULL) {
186 	perror(filename);
187 	usage();
188     }
189 
190     while(fgets(symbol, sizeof(symbol), keepf)) {
191 	len = strlen(symbol);
192 	if(len && symbol[len-1] == '\n')
193 	    symbol[len-1] = '\0';
194 
195 	add_to_keep_list(symbol);
196     }
197     fclose(keepf);
198 }
199 
200 /* ---------------------------- */
201 
202 static struct {
203 	const char *name;
204 	int	(*check)(int, const char *);	/* 1 if match, zero if not */
205 	int	(*hide)(int, const char *);	/* non-zero if error */
206 } exec_formats[] = {
207 #ifdef NLIST_AOUT
208 	{	"a.out",	check_aout,	hide_aout,	},
209 #endif
210 #ifdef NLIST_ECOFF
211 	{	"ECOFF",	check_elf64,	hide_elf64,	},
212 #endif
213 #ifdef NLIST_ELF32
214 	{	"ELF32",	check_elf32,	hide_elf32,	},
215 #endif
216 #ifdef NLIST_ELF64
217 	{	"ELF64",	check_elf64,	hide_elf64,	},
218 #endif
219 };
220 
221 static int
222 hide_syms(const char *filename)
223 {
224 	int fd, i, n, rv;
225 
226 	fd = open(filename, O_RDWR, 0);
227 	if (fd == -1) {
228 		perror(filename);
229 		return 1;
230 	}
231 
232 	rv = 0;
233 
234         n = sizeof exec_formats / sizeof exec_formats[0];
235         for (i = 0; i < n; i++) {
236 		if (lseek(fd, 0, SEEK_SET) != 0) {
237 			perror(filename);
238 			goto err;
239 		}
240                 if ((*exec_formats[i].check)(fd, filename) != 0)
241                         break;
242 	}
243 	if (i == n) {
244 		fprintf(stderr, "%s: unknown executable format\n", filename);
245 		goto err;
246 	}
247 
248 	if (verbose)
249 		fprintf(stderr, "%s is an %s binary\n", filename,
250 		    exec_formats[i].name);
251 
252 	if (lseek(fd, 0, SEEK_SET) != 0) {
253 		perror(filename);
254 		goto err;
255 	}
256 	rv = (*exec_formats[i].hide)(fd, filename);
257 
258 out:
259 	close (fd);
260 	return (rv);
261 
262 err:
263 	rv = 1;
264 	goto out;
265 }
266