| 1 | /* Unexec for Sunos 4 using shared libraries. |
| 2 | Copyright (C) 1990, 1994 Free Software Foundation, Inc. |
| 3 | |
| 4 | This file is part of GNU Emacs. |
| 5 | |
| 6 | GNU Emacs is free software; you can redistribute it and/or modify |
| 7 | it under the terms of the GNU General Public License as published by |
| 8 | the Free Software Foundation; either version 2, or (at your option) |
| 9 | any later version. |
| 10 | |
| 11 | GNU Emacs is distributed in the hope that it will be useful, |
| 12 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | GNU General Public License for more details. |
| 15 | |
| 16 | You should have received a copy of the GNU General Public License |
| 17 | along with GNU Emacs; see the file COPYING. If not, write to |
| 18 | the Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 19 | Boston, MA 02111-1307, USA. */ |
| 20 | |
| 21 | /* Contributed by Viktor Dukhovni. */ |
| 22 | /* |
| 23 | * Unexec for Berkeley a.out format + SUNOS shared libraries |
| 24 | * The unexeced executable contains the __DYNAMIC area from the |
| 25 | * original text file, and then the rest of data + bss + malloced area of |
| 26 | * the current process. (The __DYNAMIC area is at the top of the process |
| 27 | * data segment, we use "data_start" defined externally to mark the start |
| 28 | * of the "real" data segment.) |
| 29 | * |
| 30 | * For programs that want to remap some of the data segment read only |
| 31 | * a run_time_remap is provided. This attempts to remap largest area starting |
| 32 | * and ending on page boundaries between "data_start" and "bndry" |
| 33 | * For this it to figure out where the text file is located. A path search |
| 34 | * is attempted after trying argv[0] and if all fails we simply do not remap |
| 35 | * |
| 36 | * One feature of run_time_remap () is mandatory: reseting the break. |
| 37 | * |
| 38 | * Note that we can no longer map data into the text segment, as this causes |
| 39 | * the __DYNAMIC struct to become read only, breaking the runtime loader. |
| 40 | * Thus we no longer need to mess with a private crt0.c, the standard one |
| 41 | * will do just fine, since environ can live in the writable area between |
| 42 | * __DYNAMIC and data_start, just make sure that pre-crt0.o (the name |
| 43 | * is somewhat abused here) is loaded first! |
| 44 | * |
| 45 | */ |
| 46 | #include <sys/param.h> |
| 47 | #include <sys/mman.h> |
| 48 | #include <sys/file.h> |
| 49 | #include <sys/stat.h> |
| 50 | #include <string.h> |
| 51 | #include <stdio.h> |
| 52 | #include <a.out.h> |
| 53 | |
| 54 | /* Do this after the above #include's in case a configuration file wants |
| 55 | to define things for this file based on what <a.out.h> defines. */ |
| 56 | #ifdef emacs |
| 57 | #include <config.h> |
| 58 | #endif |
| 59 | |
| 60 | #if defined (SUNOS4) || defined (__FreeBSD__) || defined (__NetBSD__) |
| 61 | #define UNDO_RELOCATION |
| 62 | #endif |
| 63 | |
| 64 | #ifdef UNDO_RELOCATION |
| 65 | #include <link.h> |
| 66 | #endif |
| 67 | |
| 68 | #ifdef HAVE_UNISTD_H |
| 69 | #include <unistd.h> |
| 70 | #endif |
| 71 | |
| 72 | /* NetBSD needs this bit, but SunOS does not have it. */ |
| 73 | #ifndef MAP_FILE |
| 74 | #define MAP_FILE 0 |
| 75 | #endif |
| 76 | |
| 77 | |
| 78 | /* |
| 79 | * for programs other than emacs |
| 80 | * define data_start + initialized here, and make sure |
| 81 | * this object is loaded first! |
| 82 | * emacs will define these elsewhere, and load the object containing |
| 83 | * data_start (pre-crt0.o or firstfile.o?) first! |
| 84 | * The custom crt0.o *must not* be loaded! |
| 85 | */ |
| 86 | #ifndef emacs |
| 87 | static int data_start = 0; |
| 88 | static int initialized = 0; |
| 89 | #else |
| 90 | extern int initialized; |
| 91 | extern unsigned data_start; |
| 92 | extern int pureptr; |
| 93 | #endif |
| 94 | |
| 95 | extern char *getenv (); |
| 96 | static unsigned brk_value; |
| 97 | static struct exec nhdr; |
| 98 | static int rd_only_len; |
| 99 | static long cookie; |
| 100 | |
| 101 | |
| 102 | unexec (new_name, a_name, bndry, bss_start, entry) |
| 103 | char *new_name, *a_name; |
| 104 | unsigned bndry, bss_start, entry; |
| 105 | { |
| 106 | int fd, new; |
| 107 | char *old; |
| 108 | struct exec ohdr; /* Allocate on the stack, not needed in the next life */ |
| 109 | struct stat stat; |
| 110 | |
| 111 | if ((fd = open (a_name, O_RDONLY)) < 0) |
| 112 | { |
| 113 | fprintf (stderr, "%s: open: ", a_name); |
| 114 | perror (a_name); |
| 115 | exit (1); |
| 116 | } |
| 117 | if ((new = open (new_name, O_WRONLY | O_CREAT, 0666)) == -1) |
| 118 | { |
| 119 | fprintf (stderr, "%s: open: ", a_name); |
| 120 | perror (new_name); |
| 121 | exit (1); |
| 122 | } |
| 123 | |
| 124 | if ((fstat (fd, &stat) == -1)) |
| 125 | { |
| 126 | fprintf (stderr, "%s: ", a_name); |
| 127 | perror ("fstat"); |
| 128 | exit (1); |
| 129 | } |
| 130 | |
| 131 | old = (char *)mmap (0, stat.st_size, PROT_READ, MAP_FILE|MAP_SHARED, fd, 0); |
| 132 | if (old == (char *)-1) |
| 133 | { |
| 134 | fprintf (stderr, "%s: ", a_name); |
| 135 | perror ("mmap"); |
| 136 | exit (1); |
| 137 | } |
| 138 | close (fd); |
| 139 | |
| 140 | nhdr = ohdr = (*(struct exec *)old); |
| 141 | |
| 142 | |
| 143 | /* |
| 144 | * Remember a magic cookie so we know we've got the right binary |
| 145 | * when remapping. |
| 146 | */ |
| 147 | cookie = time (0); |
| 148 | |
| 149 | /* Save the break, it is reset to &_end (by ld.so?). */ |
| 150 | brk_value = (unsigned) sbrk (0); |
| 151 | |
| 152 | /* |
| 153 | * Round up data start to a page boundary (Lose if not a 2 power!) |
| 154 | */ |
| 155 | data_start = ((((int)&data_start) - 1) & ~(N_PAGSIZ (nhdr) - 1)) + N_PAGSIZ (nhdr); |
| 156 | |
| 157 | /* |
| 158 | * Round down read only pages to a multiple of the page size |
| 159 | */ |
| 160 | if (bndry) |
| 161 | rd_only_len = ((int)bndry & ~(N_PAGSIZ (nhdr) - 1)) - data_start; |
| 162 | |
| 163 | #ifndef emacs |
| 164 | /* Have to do this some time before dumping the data */ |
| 165 | initialized = 1; |
| 166 | #endif |
| 167 | |
| 168 | /* Handle new data and bss sizes and optional new entry point. |
| 169 | No one actually uses bss_start and entry, but tradition compels |
| 170 | one to support them. |
| 171 | Could complain if bss_start > brk_value, |
| 172 | but the caller is *supposed* to know what she is doing. */ |
| 173 | nhdr.a_data = (bss_start ? bss_start : brk_value) - N_DATADDR (nhdr); |
| 174 | nhdr.a_bss = bss_start ? brk_value - bss_start : 0; |
| 175 | if (entry) |
| 176 | nhdr.a_entry = entry; |
| 177 | |
| 178 | /* |
| 179 | * Write out the text segment with new header |
| 180 | * Dynamic executables are ZMAGIC with N_TXTOFF==0 and the header |
| 181 | * part of the text segment, but no need to rely on this. |
| 182 | * So write the TEXT first, then go back replace the header. |
| 183 | * Doing it in the other order is less general! |
| 184 | */ |
| 185 | lseek (new, N_TXTOFF (nhdr), L_SET); |
| 186 | write (new, old + N_TXTOFF (ohdr), N_TXTOFF (ohdr) + ohdr.a_text); |
| 187 | lseek (new, 0L, L_SET); |
| 188 | write (new, &nhdr, sizeof (nhdr)); |
| 189 | |
| 190 | /* |
| 191 | * Write out the head of the old data segment from the file not |
| 192 | * from core, this has the unresolved __DYNAMIC relocation data |
| 193 | * we need to reload |
| 194 | */ |
| 195 | lseek (new, N_DATOFF (nhdr), L_SET); |
| 196 | write (new, old + N_DATOFF (ohdr), (int)&data_start - N_DATADDR (ohdr)); |
| 197 | |
| 198 | /* |
| 199 | * Copy the rest of the data from core |
| 200 | */ |
| 201 | write (new, &data_start, N_BSSADDR (nhdr) - (int)&data_start); |
| 202 | |
| 203 | /* |
| 204 | * Copy the symbol table and line numbers |
| 205 | */ |
| 206 | lseek (new, N_TRELOFF (nhdr), L_SET); |
| 207 | write (new, old + N_TRELOFF (ohdr), stat.st_size - N_TRELOFF (ohdr)); |
| 208 | |
| 209 | /* Some other BSD systems use this file. |
| 210 | We don't know whether this change is right for them. */ |
| 211 | #ifdef UNDO_RELOCATION |
| 212 | /* Undo the relocations done at startup by ld.so. |
| 213 | It will do these relocations again when we start the dumped Emacs. |
| 214 | Doing them twice gives incorrect results. */ |
| 215 | { |
| 216 | unsigned long daddr = N_DATADDR (ohdr); |
| 217 | unsigned long rel, erel; |
| 218 | #ifdef SUNOS4 |
| 219 | #ifdef SUNOS4_SHARED_LIBRARIES |
| 220 | extern struct link_dynamic _DYNAMIC; |
| 221 | |
| 222 | /* SunOS4.x's ld_rel is relative to N_TXTADDR. */ |
| 223 | if (!ohdr.a_dynamic) |
| 224 | /* This was statically linked. */ |
| 225 | rel = erel = 0; |
| 226 | else if (_DYNAMIC.ld_version < 2) |
| 227 | { |
| 228 | rel = _DYNAMIC.ld_un.ld_1->ld_rel + N_TXTADDR (ohdr); |
| 229 | erel = _DYNAMIC.ld_un.ld_1->ld_hash + N_TXTADDR (ohdr); |
| 230 | } |
| 231 | else |
| 232 | { |
| 233 | rel = _DYNAMIC.ld_un.ld_2->ld_rel + N_TXTADDR (ohdr); |
| 234 | erel = _DYNAMIC.ld_un.ld_2->ld_hash + N_TXTADDR (ohdr); |
| 235 | } |
| 236 | #else /* not SUNOS4_SHARED_LIBRARIES */ |
| 237 | rel = erel = 0; |
| 238 | #endif /* not SUNOS4_SHARED_LIBRARIES */ |
| 239 | #ifdef sparc |
| 240 | #define REL_INFO_TYPE struct reloc_info_sparc |
| 241 | #else |
| 242 | #define REL_INFO_TYPE struct relocation_info |
| 243 | #endif /* sparc */ |
| 244 | #define REL_TARGET_ADDRESS(r) (((REL_INFO_TYPE *)(r))->r_address) |
| 245 | #endif /* SUNOS4 */ |
| 246 | #if defined (__FreeBSD__) || defined (__NetBSD__) |
| 247 | extern struct _dynamic _DYNAMIC; |
| 248 | |
| 249 | /* FreeBSD's LD_REL is a virtual address itself. */ |
| 250 | rel = LD_REL (&_DYNAMIC); |
| 251 | erel = rel + LD_RELSZ (&_DYNAMIC); |
| 252 | #define REL_INFO_TYPE struct relocation_info |
| 253 | #define REL_TARGET_ADDRESS(r) (((REL_INFO_TYPE *)(r))->r_address) |
| 254 | #endif |
| 255 | |
| 256 | for (; rel < erel; rel += sizeof (REL_INFO_TYPE)) |
| 257 | { |
| 258 | /* This is the virtual address where ld.so will do relocation. */ |
| 259 | unsigned long target = REL_TARGET_ADDRESS (rel); |
| 260 | /* This is the offset in the data segment. */ |
| 261 | unsigned long segoffset = target - daddr; |
| 262 | |
| 263 | /* If it is located below data_start, we have to do nothing here, |
| 264 | because the old data has been already written to the location. */ |
| 265 | if (target < (unsigned long)&data_start) |
| 266 | continue; |
| 267 | |
| 268 | lseek (new, N_DATOFF (nhdr) + segoffset, L_SET); |
| 269 | write (new, old + N_DATOFF (ohdr) + segoffset, sizeof (unsigned long)); |
| 270 | } |
| 271 | } |
| 272 | #endif /* UNDO_RELOCATION */ |
| 273 | |
| 274 | fchmod (new, 0755); |
| 275 | } |
| 276 | |
| 277 | void |
| 278 | run_time_remap (progname) |
| 279 | char *progname; |
| 280 | { |
| 281 | char aout[MAXPATHLEN]; |
| 282 | register char *path, *p; |
| 283 | |
| 284 | /* Just in case */ |
| 285 | if (!initialized) |
| 286 | return; |
| 287 | |
| 288 | /* Restore the break */ |
| 289 | brk ((char *) brk_value); |
| 290 | |
| 291 | /* If nothing to remap: we are done! */ |
| 292 | if (rd_only_len == 0) |
| 293 | return; |
| 294 | |
| 295 | /* |
| 296 | * Attempt to find the executable |
| 297 | * First try argv[0], will almost always succeed as shells tend to give |
| 298 | * the full path from the hash list rather than using execvp () |
| 299 | */ |
| 300 | if (is_it (progname)) |
| 301 | return; |
| 302 | |
| 303 | /* |
| 304 | * If argv[0] is a full path and does not exist, not much sense in |
| 305 | * searching further |
| 306 | */ |
| 307 | if (strchr (progname, '/')) |
| 308 | return; |
| 309 | |
| 310 | /* |
| 311 | * Try to search for argv[0] on the PATH |
| 312 | */ |
| 313 | path = getenv ("PATH"); |
| 314 | if (path == NULL) |
| 315 | return; |
| 316 | |
| 317 | while (*path) |
| 318 | { |
| 319 | /* copy through ':' or end */ |
| 320 | for (p = aout; *p = *path; ++p, ++path) |
| 321 | if (*p == ':') |
| 322 | { |
| 323 | ++path; /* move past ':' */ |
| 324 | break; |
| 325 | } |
| 326 | *p++ = '/'; |
| 327 | strcpy (p, progname); |
| 328 | /* |
| 329 | * aout is a candidate full path name |
| 330 | */ |
| 331 | if (is_it (aout)) |
| 332 | return; |
| 333 | } |
| 334 | } |
| 335 | |
| 336 | is_it (filename) |
| 337 | char *filename; |
| 338 | { |
| 339 | int fd; |
| 340 | long filenames_cookie; |
| 341 | struct exec hdr; |
| 342 | |
| 343 | /* |
| 344 | * Open an executable and check for a valid header! |
| 345 | * Can't bcmp the header with what we had, it may have been stripped! |
| 346 | * so we may save looking at non executables with the same name, mostly |
| 347 | * directories. |
| 348 | */ |
| 349 | fd = open (filename, O_RDONLY); |
| 350 | if (fd != -1) |
| 351 | { |
| 352 | if (read (fd, &hdr, sizeof (hdr)) == sizeof (hdr) |
| 353 | && !N_BADMAG (hdr) && N_DATOFF (hdr) == N_DATOFF (nhdr) |
| 354 | && N_TRELOFF (hdr) == N_TRELOFF (nhdr)) |
| 355 | { |
| 356 | /* compare cookies */ |
| 357 | lseek (fd, N_DATOFF (hdr) + (int)&cookie - N_DATADDR (hdr), L_SET); |
| 358 | read (fd, &filenames_cookie, sizeof (filenames_cookie)); |
| 359 | if (filenames_cookie == cookie) |
| 360 | { /* Eureka */ |
| 361 | |
| 362 | /* |
| 363 | * Do the mapping |
| 364 | * The PROT_EXEC may not be needed, but it is safer this way. |
| 365 | * should the shared library decide to indirect through |
| 366 | * addresses in the data segment not part of __DYNAMIC |
| 367 | */ |
| 368 | mmap ((char *) data_start, rd_only_len, PROT_READ | PROT_EXEC, |
| 369 | MAP_FILE | MAP_SHARED | MAP_FIXED, fd, |
| 370 | N_DATOFF (hdr) + data_start - N_DATADDR (hdr)); |
| 371 | close (fd); |
| 372 | return 1; |
| 373 | } |
| 374 | } |
| 375 | close (fd); |
| 376 | } |
| 377 | return 0; |
| 378 | } |