cleanup
[openwrt/svn-archive/archive.git] / tools / sstrip / src / sstrip.c
1 /* http://www.muppetlabs.com/~breadbox/software/elfkickers.html */
2
3 /* sstrip: Copyright (C) 1999-2001 by Brian Raiter, under the GNU
4 * General Public License. No warranty. See COPYING for details.
5 *
6 * Aug 23, 2004 Hacked by Manuel Novoa III <mjn3@codepoet.org> to
7 * handle targets of different endianness and/or elf class, making
8 * it more useful in a cross-devel environment.
9 */
10
11 /* ============== original README ===================
12 *
13 * sstrip is a small utility that removes the contents at the end of an
14 * ELF file that are not part of the program's memory image.
15 *
16 * Most ELF executables are built with both a program header table and a
17 * section header table. However, only the former is required in order
18 * for the OS to load, link and execute a program. sstrip attempts to
19 * extract the ELF header, the program header table, and its contents,
20 * leaving everything else in the bit bucket. It can only remove parts of
21 * the file that occur at the end, after the parts to be saved. However,
22 * this almost always includes the section header table, and occasionally
23 * a few random sections that are not used when running a program.
24 *
25 * It should be noted that the GNU bfd library is (understandably)
26 * dependent on the section header table as an index to the file's
27 * contents. Thus, an executable file that has no section header table
28 * cannot be used with gdb, objdump, or any other program based upon the
29 * bfd library, at all. In fact, the program will not even recognize the
30 * file as a valid executable. (This limitation is noted in the source
31 * code comments for bfd, and is marked "FIXME", so this may change at
32 * some future date. However, I would imagine that it is a pretty
33 * low-priority item, as executables without a section header table are
34 * rare in the extreme.) This probably also explains why strip doesn't
35 * offer the option to do this.
36 *
37 * Shared library files may also have their section header table removed.
38 * Such a library will still function; however, it will no longer be
39 * possible for a compiler to link a new program against it.
40 *
41 * As an added bonus, sstrip also tries to removes trailing zero bytes
42 * from the end of the file. (This normally cannot be done with an
43 * executable that has a section header table.)
44 *
45 * sstrip is a very simplistic program. It depends upon the common
46 * practice of putting the parts of the file that contribute to the
47 * memory image at the front, and the remaining material at the end. This
48 * permits it to discard the latter material without affecting file
49 * offsets and memory addresses in what remains. Of course, the ELF
50 * standard permits files to be organized in almost any order, so if a
51 * pathological linker decided to put its section headers at the top,
52 * sstrip would be useless on such executables.
53 */
54
55 #include <stdio.h>
56 #include <stdlib.h>
57 #include <string.h>
58 #include <errno.h>
59 #include <unistd.h>
60 #include <fcntl.h>
61 #include <elf.h>
62
63 #ifndef TRUE
64 #define TRUE 1
65 #define FALSE 0
66 #endif
67
68 /* The name of the program.
69 */
70 static char const *progname;
71
72 /* The name of the current file.
73 */
74 static char const *filename;
75
76
77 /* A simple error-handling function. FALSE is always returned for the
78 * convenience of the caller.
79 */
80 static int err(char const *errmsg)
81 {
82 fprintf(stderr, "%s: %s: %s\n", progname, filename, errmsg);
83 return FALSE;
84 }
85
86 /* A flag to signal the need for endian reversal.
87 */
88 static int do_reverse_endian;
89
90 /* Get a value from the elf header, compensating for endianness.
91 */
92 #define EGET(X) \
93 (__extension__ ({ \
94 uint64_t __res; \
95 if (!do_reverse_endian) { \
96 __res = (X); \
97 } else if (sizeof(X) == 1) { \
98 __res = (X); \
99 } else if (sizeof(X) == 2) { \
100 __res = bswap_16((X)); \
101 } else if (sizeof(X) == 4) { \
102 __res = bswap_32((X)); \
103 } else if (sizeof(X) == 8) { \
104 __res = bswap_64((X)); \
105 } else { \
106 fprintf(stderr, "%s: %s: EGET failed for size %d\n", \
107 progname, filename, sizeof(X)); \
108 exit(EXIT_FAILURE); \
109 } \
110 __res; \
111 }))
112
113 /* Set a value 'Y' in the elf header to 'X', compensating for endianness.
114 */
115 #define ESET(Y,X) \
116 do if (!do_reverse_endian) { \
117 Y = (X); \
118 } else if (sizeof(Y) == 1) { \
119 Y = (X); \
120 } else if (sizeof(Y) == 2) { \
121 Y = bswap_16((uint16_t)(X)); \
122 } else if (sizeof(Y) == 4) { \
123 Y = bswap_32((uint32_t)(X)); \
124 } else if (sizeof(Y) == 8) { \
125 Y = bswap_64((uint64_t)(X)); \
126 } else { \
127 fprintf(stderr, "%s: %s: ESET failed for size %d\n", \
128 progname, filename, sizeof(Y)); \
129 exit(EXIT_FAILURE); \
130 } while (0)
131
132
133 /* A macro for I/O errors: The given error message is used only when
134 * errno is not set.
135 */
136 #define ferr(msg) (err(errno ? strerror(errno) : (msg)))
137
138
139
140 #define HEADER_FUNCTIONS(CLASS) \
141 \
142 /* readelfheader() reads the ELF header into our global variable, and \
143 * checks to make sure that this is in fact a file that we should be \
144 * munging. \
145 */ \
146 static int readelfheader ## CLASS (int fd, Elf ## CLASS ## _Ehdr *ehdr) \
147 { \
148 if (read(fd, ((char *)ehdr)+EI_NIDENT, sizeof(*ehdr) - EI_NIDENT) \
149 != sizeof(*ehdr) - EI_NIDENT) \
150 return ferr("missing or incomplete ELF header."); \
151 \
152 /* Verify the sizes of the ELF header and the program segment \
153 * header table entries. \
154 */ \
155 if (EGET(ehdr->e_ehsize) != sizeof(Elf ## CLASS ## _Ehdr)) \
156 return err("unrecognized ELF header size."); \
157 if (EGET(ehdr->e_phentsize) != sizeof(Elf ## CLASS ## _Phdr)) \
158 return err("unrecognized program segment header size."); \
159 \
160 /* Finally, check the file type. \
161 */ \
162 if (EGET(ehdr->e_type) != ET_EXEC && EGET(ehdr->e_type) != ET_DYN) \
163 return err("not an executable or shared-object library."); \
164 \
165 return TRUE; \
166 } \
167 \
168 /* readphdrtable() loads the program segment header table into memory. \
169 */ \
170 static int readphdrtable ## CLASS (int fd, Elf ## CLASS ## _Ehdr const *ehdr, \
171 Elf ## CLASS ## _Phdr **phdrs) \
172 { \
173 size_t size; \
174 \
175 if (!EGET(ehdr->e_phoff) || !EGET(ehdr->e_phnum) \
176 ) return err("ELF file has no program header table."); \
177 \
178 size = EGET(ehdr->e_phnum) * sizeof **phdrs; \
179 if (!(*phdrs = malloc(size))) \
180 return err("Out of memory!"); \
181 \
182 errno = 0; \
183 if (read(fd, *phdrs, size) != (ssize_t)size) \
184 return ferr("missing or incomplete program segment header table."); \
185 \
186 return TRUE; \
187 } \
188 \
189 /* getmemorysize() determines the offset of the last byte of the file \
190 * that is referenced by an entry in the program segment header table. \
191 * (Anything in the file after that point is not used when the program \
192 * is executing, and thus can be safely discarded.) \
193 */ \
194 static int getmemorysize ## CLASS (Elf ## CLASS ## _Ehdr const *ehdr, \
195 Elf ## CLASS ## _Phdr const *phdrs, \
196 unsigned long *newsize) \
197 { \
198 Elf ## CLASS ## _Phdr const *phdr; \
199 unsigned long size, n; \
200 int i; \
201 \
202 /* Start by setting the size to include the ELF header and the \
203 * complete program segment header table. \
204 */ \
205 size = EGET(ehdr->e_phoff) + EGET(ehdr->e_phnum) * sizeof *phdrs; \
206 if (size < sizeof *ehdr) \
207 size = sizeof *ehdr; \
208 \
209 /* Then keep extending the size to include whatever data the \
210 * program segment header table references. \
211 */ \
212 for (i = 0, phdr = phdrs ; i < EGET(ehdr->e_phnum) ; ++i, ++phdr) { \
213 if (EGET(phdr->p_type) != PT_NULL) { \
214 n = EGET(phdr->p_offset) + EGET(phdr->p_filesz); \
215 if (n > size) \
216 size = n; \
217 } \
218 } \
219 \
220 *newsize = size; \
221 return TRUE; \
222 } \
223 \
224 /* modifyheaders() removes references to the section header table if \
225 * it was stripped, and reduces program header table entries that \
226 * included truncated bytes at the end of the file. \
227 */ \
228 static int modifyheaders ## CLASS (Elf ## CLASS ## _Ehdr *ehdr, \
229 Elf ## CLASS ## _Phdr *phdrs, \
230 unsigned long newsize) \
231 { \
232 Elf ## CLASS ## _Phdr *phdr; \
233 int i; \
234 \
235 /* If the section header table is gone, then remove all references \
236 * to it in the ELF header. \
237 */ \
238 if (EGET(ehdr->e_shoff) >= newsize) { \
239 ESET(ehdr->e_shoff,0); \
240 ESET(ehdr->e_shnum,0); \
241 ESET(ehdr->e_shentsize,0); \
242 ESET(ehdr->e_shstrndx,0); \
243 } \
244 \
245 /* The program adjusts the file size of any segment that was \
246 * truncated. The case of a segment being completely stripped out \
247 * is handled separately. \
248 */ \
249 for (i = 0, phdr = phdrs ; i < EGET(ehdr->e_phnum) ; ++i, ++phdr) { \
250 if (EGET(phdr->p_offset) >= newsize) { \
251 ESET(phdr->p_offset,newsize); \
252 ESET(phdr->p_filesz,0); \
253 } else if (EGET(phdr->p_offset) + EGET(phdr->p_filesz) > newsize) { \
254 newsize -= EGET(phdr->p_offset); \
255 ESET(phdr->p_filesz, newsize); \
256 } \
257 } \
258 \
259 return TRUE; \
260 } \
261 \
262 /* commitchanges() writes the new headers back to the original file \
263 * and sets the file to its new size. \
264 */ \
265 static int commitchanges ## CLASS (int fd, Elf ## CLASS ## _Ehdr const *ehdr, \
266 Elf ## CLASS ## _Phdr *phdrs, \
267 unsigned long newsize) \
268 { \
269 size_t n; \
270 \
271 /* Save the changes to the ELF header, if any. \
272 */ \
273 if (lseek(fd, 0, SEEK_SET)) \
274 return ferr("could not rewind file"); \
275 errno = 0; \
276 if (write(fd, ehdr, sizeof *ehdr) != sizeof *ehdr) \
277 return err("could not modify file"); \
278 \
279 /* Save the changes to the program segment header table, if any. \
280 */ \
281 if (lseek(fd, EGET(ehdr->e_phoff), SEEK_SET) == (off_t)-1) { \
282 err("could not seek in file."); \
283 goto warning; \
284 } \
285 n = EGET(ehdr->e_phnum) * sizeof *phdrs; \
286 if (write(fd, phdrs, n) != (ssize_t)n) { \
287 err("could not write to file"); \
288 goto warning; \
289 } \
290 \
291 /* Eleventh-hour sanity check: don't truncate before the end of \
292 * the program segment header table. \
293 */ \
294 if (newsize < EGET(ehdr->e_phoff) + n) \
295 newsize = EGET(ehdr->e_phoff) + n; \
296 \
297 /* Chop off the end of the file. \
298 */ \
299 if (ftruncate(fd, newsize)) { \
300 err("could not resize file"); \
301 goto warning; \
302 } \
303 \
304 return TRUE; \
305 \
306 warning: \
307 return err("ELF file may have been corrupted!"); \
308 }
309
310
311 /* First elements of Elf32_Ehdr and Elf64_Ehdr are common.
312 */
313 static int readelfheaderident(int fd, Elf32_Ehdr *ehdr)
314 {
315 errno = 0;
316 if (read(fd, ehdr, EI_NIDENT) != EI_NIDENT)
317 return ferr("missing or incomplete ELF header.");
318
319 /* Check the ELF signature.
320 */
321 if (!(ehdr->e_ident[EI_MAG0] == ELFMAG0 &&
322 ehdr->e_ident[EI_MAG1] == ELFMAG1 &&
323 ehdr->e_ident[EI_MAG2] == ELFMAG2 &&
324 ehdr->e_ident[EI_MAG3] == ELFMAG3))
325 {
326 err("missing ELF signature.");
327 return -1;
328 }
329
330 /* Compare the file's class and endianness with the program's.
331 */
332 #if __BYTE_ORDER == __LITTLE_ENDIAN
333 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
334 do_reverse_endian = 0;
335 } else if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
336 /* fprintf(stderr, "ELF file has different endianness.\n"); */
337 do_reverse_endian = 1;
338 }
339 #elif __BYTE_ORDER == __BIG_ENDIAN
340 if (ehdr->e_ident[EI_DATA] == ELFDATA2LSB) {
341 /* fprintf(stderr, "ELF file has different endianness.\n"); */
342 do_reverse_endian = 1;
343 } else if (ehdr->e_ident[EI_DATA] == ELFDATA2MSB) {
344 do_reverse_endian = 0;
345 }
346 #else
347 #error unkown endianness
348 #endif
349 else {
350 err("Unsupported endianness");
351 return -1;
352 }
353
354 /* Check the target architecture.
355 */
356 /* if (EGET(ehdr->e_machine) != ELF_ARCH) { */
357 /* /\* return err("ELF file created for different architecture."); *\/ */
358 /* fprintf(stderr, "ELF file created for different architecture.\n"); */
359 /* } */
360 return ehdr->e_ident[EI_CLASS];
361 }
362
363
364 HEADER_FUNCTIONS(32)
365
366 HEADER_FUNCTIONS(64)
367
368 /* truncatezeros() examines the bytes at the end of the file's
369 * size-to-be, and reduces the size to exclude any trailing zero
370 * bytes.
371 */
372 static int truncatezeros(int fd, unsigned long *newsize)
373 {
374 unsigned char contents[1024];
375 unsigned long size, n;
376
377 size = *newsize;
378 do {
379 n = sizeof contents;
380 if (n > size)
381 n = size;
382 if (lseek(fd, size - n, SEEK_SET) == (off_t)-1)
383 return ferr("cannot seek in file.");
384 if (read(fd, contents, n) != (ssize_t)n)
385 return ferr("cannot read file contents");
386 while (n && !contents[--n])
387 --size;
388 } while (size && !n);
389
390 /* Sanity check.
391 */
392 if (!size)
393 return err("ELF file is completely blank!");
394
395 *newsize = size;
396 return TRUE;
397 }
398
399 /* main() loops over the cmdline arguments, leaving all the real work
400 * to the other functions.
401 */
402 int main(int argc, char *argv[])
403 {
404 int fd;
405 union {
406 Elf32_Ehdr ehdr32;
407 Elf64_Ehdr ehdr64;
408 } e;
409 union {
410 Elf32_Phdr *phdrs32;
411 Elf64_Phdr *phdrs64;
412 } p;
413 unsigned long newsize;
414 char **arg;
415 int failures = 0;
416
417 if (argc < 2 || argv[1][0] == '-') {
418 printf("Usage: sstrip FILE...\n"
419 "sstrip discards all nonessential bytes from an executable.\n\n"
420 "Version 2.0-X Copyright (C) 2000,2001 Brian Raiter.\n"
421 "Cross-devel hacks Copyright (C) 2004 Manuel Novoa III.\n"
422 "This program is free software, licensed under the GNU\n"
423 "General Public License. There is absolutely no warranty.\n");
424 return EXIT_SUCCESS;
425 }
426
427 progname = argv[0];
428
429 for (arg = argv + 1 ; *arg != NULL ; ++arg) {
430 filename = *arg;
431
432 fd = open(*arg, O_RDWR);
433 if (fd < 0) {
434 ferr("can't open");
435 ++failures;
436 continue;
437 }
438
439 switch (readelfheaderident(fd, &e.ehdr32)) {
440 case ELFCLASS32:
441 if (!(readelfheader32(fd, &e.ehdr32) &&
442 readphdrtable32(fd, &e.ehdr32, &p.phdrs32) &&
443 getmemorysize32(&e.ehdr32, p.phdrs32, &newsize) &&
444 truncatezeros(fd, &newsize) &&
445 modifyheaders32(&e.ehdr32, p.phdrs32, newsize) &&
446 commitchanges32(fd, &e.ehdr32, p.phdrs32, newsize)))
447 ++failures;
448 break;
449 case ELFCLASS64:
450 if (!(readelfheader64(fd, &e.ehdr64) &&
451 readphdrtable64(fd, &e.ehdr64, &p.phdrs64) &&
452 getmemorysize64(&e.ehdr64, p.phdrs64, &newsize) &&
453 truncatezeros(fd, &newsize) &&
454 modifyheaders64(&e.ehdr64, p.phdrs64, newsize) &&
455 commitchanges64(fd, &e.ehdr64, p.phdrs64, newsize)))
456 ++failures;
457 break;
458 default:
459 ++failures;
460 break;
461 }
462 close(fd);
463 }
464
465 return failures ? EXIT_FAILURE : EXIT_SUCCESS;
466 }