1 /* $OpenBSD: crunchide.c,v 1.3 2009/07/21 17:19:13 deraadt Exp $ */ 2 3 /* 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 #include <unistd.h> 62 #include <stdio.h> 63 #include <stdlib.h> 64 #include <string.h> 65 #include <errno.h> 66 #include <fcntl.h> 67 #include <a.out.h> 68 #include <sys/types.h> 69 #ifdef _NLIST_DO_ECOFF 70 #include <sys/exec_ecoff.h> 71 #endif 72 #include <sys/mman.h> 73 #include <sys/stat.h> 74 75 /* 76 * if __ELF__ is defined, do not bother supporting AOUT. 77 */ 78 #if defined(_NLIST_DO_AOUT) && !(defined(__ELF__)) 79 #define DO_AOUT 80 #endif 81 82 void usage(void); 83 84 void add_to_keep_list(char *); 85 void add_file_to_keep_list(char *); 86 87 void hide_syms(char *); 88 #ifdef _NLIST_DO_ECOFF 89 void ecoff_hide(int, char *); 90 #endif 91 #ifdef _NLIST_DO_ELF 92 void elf_hide(int, char *); 93 #endif 94 95 extern char *__progname; 96 97 int 98 crunchide_main(int argc, char *argv[]) 99 { 100 int ch; 101 102 while ((ch = getopt(argc, argv, "hk:f:")) != -1) 103 switch (ch) { 104 case 'h': 105 break; 106 case 'k': 107 add_to_keep_list(optarg); 108 break; 109 case 'f': 110 add_file_to_keep_list(optarg); 111 break; 112 default: 113 usage(); 114 } 115 116 argc -= optind; 117 argv += optind; 118 119 if (argc == 0) 120 usage(); 121 122 while (argc) { 123 hide_syms(*argv); 124 argc--; 125 argv++; 126 } 127 128 return 0; 129 } 130 131 struct keep { 132 struct keep *next; 133 char *sym; 134 } *keep_list; 135 136 void 137 add_to_keep_list(char *symbol) 138 { 139 struct keep *newp, *prevp, *curp; 140 int cmp = 0; 141 142 for (curp = keep_list, prevp = NULL; curp; prevp = curp, curp = curp->next) 143 if ((cmp = strcmp(symbol, curp->sym)) <= 0) 144 break; 145 146 if (curp && cmp == 0) 147 return; /* already in table */ 148 149 newp = (struct keep *) calloc(1, sizeof(struct keep)); 150 if (newp) 151 newp->sym = strdup(symbol); 152 if (newp == NULL || newp->sym == NULL) { 153 fprintf(stderr, "%s: out of memory for keep list\n", __progname); 154 exit(1); 155 } 156 newp->next = curp; 157 if (prevp) 158 prevp->next = newp; 159 else 160 keep_list = newp; 161 } 162 163 int 164 in_keep_list(char *symbol) 165 { 166 struct keep *curp; 167 int cmp = 0; 168 169 for (curp = keep_list; curp; curp = curp->next) 170 if ((cmp = strcmp(symbol, curp->sym)) <= 0) 171 break; 172 173 return curp && cmp == 0; 174 } 175 176 void 177 add_file_to_keep_list(char *filename) 178 { 179 FILE *keepf; 180 char symbol[1024]; 181 int len; 182 183 if ((keepf = fopen(filename, "r")) == NULL) { 184 perror(filename); 185 usage(); 186 } 187 while (fgets(symbol, sizeof(symbol), keepf)) { 188 len = strlen(symbol); 189 if (len && symbol[len - 1] == '\n') 190 symbol[len - 1] = '\0'; 191 192 add_to_keep_list(symbol); 193 } 194 fclose(keepf); 195 } 196 197 int nsyms, ntextrel, ndatarel; 198 struct exec *hdrp; 199 char *aoutdata, *strbase; 200 struct relocation_info *textrel, *datarel; 201 struct nlist *symbase; 202 203 #define SYMSTR(sp) &strbase[(sp)->n_un.n_strx] 204 205 /* is the symbol a global symbol defined in the current file? */ 206 #define IS_GLOBAL_DEFINED(sp) \ 207 (((sp)->n_type & N_EXT) && ((sp)->n_type & N_TYPE) != N_UNDF) 208 209 #ifdef DO_AOUT 210 #if defined(__sparc__) 211 /* is the relocation entry dependent on a symbol? */ 212 #define IS_SYMBOL_RELOC(rp) \ 213 ((rp)->r_extern || \ 214 ((rp)->r_type >= RELOC_BASE10 && (rp)->r_type <= RELOC_BASE22) || \ 215 (rp)->r_type == RELOC_JMP_TBL) 216 #else 217 /* is the relocation entry dependent on a symbol? */ 218 #define IS_SYMBOL_RELOC(rp) \ 219 ((rp)->r_extern||(rp)->r_baserel||(rp)->r_jmptable) 220 #endif 221 #endif 222 223 void check_reloc(char *filename, struct relocation_info * relp); 224 225 void 226 hide_syms(char *filename) 227 { 228 int inf; 229 struct stat infstat; 230 #ifdef DO_AOUT 231 struct relocation_info *relp; 232 struct nlist *symp; 233 u_char zero = 0; 234 #endif 235 char *buf; 236 237 /* 238 * Open the file and do some error checking. 239 */ 240 241 if ((inf = open(filename, O_RDWR)) == -1) { 242 perror(filename); 243 return; 244 } 245 if (fstat(inf, &infstat) == -1) { 246 perror(filename); 247 close(inf); 248 return; 249 } 250 if (infstat.st_size < sizeof(struct exec)) { 251 fprintf(stderr, "%s: short file\n", filename); 252 close(inf); 253 return; 254 } 255 if ((buf = mmap(NULL, infstat.st_size, PROT_READ | PROT_WRITE, 256 MAP_FILE | MAP_SHARED, inf, 0)) == MAP_FAILED) { 257 fprintf(stderr, "%s: cannot map\n", filename); 258 close(inf); 259 return; 260 } 261 262 #ifdef _NLIST_DO_ELF 263 if (buf[0] == 0x7f && (buf[1] == 'E' || buf[1] == 'O') && 264 buf[2] == 'L' && buf[3] == 'F') { 265 elf_hide(inf, buf); 266 return; 267 } 268 #endif /* _NLIST_DO_ELF */ 269 270 #ifdef _NLIST_DO_ECOFF 271 if (!ECOFF_BADMAG((struct ecoff_exechdr *) buf)) { 272 ecoff_hide(inf, buf); 273 return; 274 } 275 #endif /* _NLIST_DO_ECOFF */ 276 277 #ifdef DO_AOUT 278 aoutdata = buf; 279 280 /* 281 * Check the header and calculate offsets and sizes from it. 282 */ 283 hdrp = (struct exec *) aoutdata; 284 285 if (N_BADMAG(*hdrp)) { 286 fprintf(stderr, "%s: bad magic: not an a.out, ecoff or elf file\n", 287 filename); 288 close(inf); 289 return; 290 } 291 textrel = (struct relocation_info *) (aoutdata + N_TRELOFF(*hdrp)); 292 datarel = (struct relocation_info *) (aoutdata + N_DRELOFF(*hdrp)); 293 symbase = (struct nlist *) (aoutdata + N_SYMOFF(*hdrp)); 294 strbase = (char *) (aoutdata + N_STROFF(*hdrp)); 295 296 ntextrel = hdrp->a_trsize / sizeof(struct relocation_info); 297 ndatarel = hdrp->a_drsize / sizeof(struct relocation_info); 298 nsyms = hdrp->a_syms / sizeof(struct nlist); 299 300 /* 301 * Zap the type field of all globally-defined symbols. The linker will 302 * subsequently ignore these entries. Don't zap any symbols in the 303 * keep list. 304 */ 305 for (symp = symbase; symp < symbase + nsyms; symp++) 306 if (IS_GLOBAL_DEFINED(symp) && !in_keep_list(SYMSTR(symp))) { 307 /* 308 * XXX Our VM system has some problems, so 309 * avoid the VM system.... 310 */ 311 lseek(inf, (off_t) ((void *) &symp->n_type - 312 (void *) buf), SEEK_SET); 313 write(inf, &zero, sizeof zero); 314 symp->n_type = 0; 315 } 316 /* 317 * Check whether the relocation entries reference any symbols that we 318 * just zapped. I don't know whether ld can handle this case, but I 319 * haven't encountered it yet. These checks are here so that the program 320 * doesn't fail silently should such symbols be encountered. 321 */ 322 for (relp = textrel; relp < textrel + ntextrel; relp++) 323 check_reloc(filename, relp); 324 for (relp = datarel; relp < datarel + ndatarel; relp++) 325 check_reloc(filename, relp); 326 327 msync(buf, infstat.st_size, MS_SYNC); 328 munmap(buf, infstat.st_size); 329 close(inf); 330 #endif /* DO_AOUT */ 331 } 332 333 #ifdef DO_AOUT 334 void 335 check_reloc(char *filename, struct relocation_info * relp) 336 { 337 /* bail out if we zapped a symbol that is needed */ 338 if (IS_SYMBOL_RELOC(relp) && symbase[relp->r_symbolnum].n_type == 0) { 339 fprintf(stderr, 340 "%s: oops, have hanging relocation for %s: bailing out!\n", 341 filename, SYMSTR(&symbase[relp->r_symbolnum])); 342 exit(1); 343 } 344 } 345 #endif /* DO_AOUT */ 346