?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[NETFILTER]: x_tables: struct xt_table_info diet
[linux-2.6/openmoko-kernel/knife-kernel.git] / fs / binfmt_elf.c
blobf0b3171842f22e75796d2aa68a7713f30af2f111
1 /*
2 * linux/fs/binfmt_elf.c
3 *
4 * These are the functions used to load ELF format executables as used
5 * on SVr4 machines. Information on the format may be found in the book
6 * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
7 * Tools".
8 *
9 * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
10 */
11
12 #include <linux/module.h>
13 #include <linux/kernel.h>
14 #include <linux/fs.h>
15 #include <linux/stat.h>
16 #include <linux/time.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/a.out.h>
20 #include <linux/errno.h>
21 #include <linux/signal.h>
22 #include <linux/binfmts.h>
23 #include <linux/string.h>
24 #include <linux/file.h>
25 #include <linux/fcntl.h>
26 #include <linux/ptrace.h>
27 #include <linux/slab.h>
28 #include <linux/shm.h>
29 #include <linux/personality.h>
30 #include <linux/elfcore.h>
31 #include <linux/init.h>
32 #include <linux/highuid.h>
33 #include <linux/smp.h>
34 #include <linux/compiler.h>
35 #include <linux/highmem.h>
36 #include <linux/pagemap.h>
37 #include <linux/security.h>
38 #include <linux/syscalls.h>
39 #include <linux/random.h>
40 #include <linux/elf.h>
41 #include <linux/utsname.h>
42 #include <asm/uaccess.h>
43 #include <asm/param.h>
44 #include <asm/page.h>
45
46 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
47 static int load_elf_library(struct file *);
48 static unsigned long elf_map (struct file *, unsigned long, struct elf_phdr *, int, int);
49
50 /*
51 * If we don't support core dumping, then supply a NULL so we
52 * don't even try.
53 */
54 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
55 static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, unsigned long limit);
56 #else
57 #define elf_core_dump NULL
58 #endif
59
60 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
61 #define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
62 #else
63 #define ELF_MIN_ALIGN PAGE_SIZE
64 #endif
65
66 #ifndef ELF_CORE_EFLAGS
67 #define ELF_CORE_EFLAGS 0
68 #endif
69
70 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
71 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
72 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
73
74 static struct linux_binfmt elf_format = {
75 .module = THIS_MODULE,
76 .load_binary = load_elf_binary,
77 .load_shlib = load_elf_library,
78 .core_dump = elf_core_dump,
79 .min_coredump = ELF_EXEC_PAGESIZE,
80 .hasvdso = 1
81 };
82
83 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
84
85 static int set_brk(unsigned long start, unsigned long end)
86 {
87 start = ELF_PAGEALIGN(start);
88 end = ELF_PAGEALIGN(end);
89 if (end > start) {
90 unsigned long addr;
91 down_write(&current->mm->mmap_sem);
92 addr = do_brk(start, end - start);
93 up_write(&current->mm->mmap_sem);
94 if (BAD_ADDR(addr))
95 return addr;
96 }
97 current->mm->start_brk = current->mm->brk = end;
98 return 0;
99 }
100
101 /* We need to explicitly zero any fractional pages
102 after the data section (i.e. bss). This would
103 contain the junk from the file that should not
104 be in memory
105 */
106 static int padzero(unsigned long elf_bss)
107 {
108 unsigned long nbyte;
109
110 nbyte = ELF_PAGEOFFSET(elf_bss);
111 if (nbyte) {
112 nbyte = ELF_MIN_ALIGN - nbyte;
113 if (clear_user((void __user *) elf_bss, nbyte))
114 return -EFAULT;
115 }
116 return 0;
117 }
118
119 /* Let's use some macros to make this stack manipulation a litle clearer */
120 #ifdef CONFIG_STACK_GROWSUP
121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
122 #define STACK_ROUND(sp, items) \
123 ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
124 #define STACK_ALLOC(sp, len) ({ \
125 elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
126 old_sp; })
127 #else
128 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
129 #define STACK_ROUND(sp, items) \
130 (((unsigned long) (sp - items)) &~ 15UL)
131 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
132 #endif
133
134 static int
135 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
136 int interp_aout, unsigned long load_addr,
137 unsigned long interp_load_addr)
138 {
139 unsigned long p = bprm->p;
140 int argc = bprm->argc;
141 int envc = bprm->envc;
142 elf_addr_t __user *argv;
143 elf_addr_t __user *envp;
144 elf_addr_t __user *sp;
145 elf_addr_t __user *u_platform;
146 const char *k_platform = ELF_PLATFORM;
147 int items;
148 elf_addr_t *elf_info;
149 int ei_index = 0;
150 struct task_struct *tsk = current;
151 struct vm_area_struct *vma;
152
153 /*
154 * In some cases (e.g. Hyper-Threading), we want to avoid L1
155 * evictions by the processes running on the same package. One
156 * thing we can do is to shuffle the initial stack for them.
157 */
158
159 p = arch_align_stack(p);
160
161 /*
162 * If this architecture has a platform capability string, copy it
163 * to userspace. In some cases (Sparc), this info is impossible
164 * for userspace to get any other way, in others (i386) it is
165 * merely difficult.
166 */
167 u_platform = NULL;
168 if (k_platform) {
169 size_t len = strlen(k_platform) + 1;
170
171 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
172 if (__copy_to_user(u_platform, k_platform, len))
173 return -EFAULT;
174 }
175
176 /* Create the ELF interpreter info */
177 elf_info = (elf_addr_t *)current->mm->saved_auxv;
178 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
179 #define NEW_AUX_ENT(id, val) \
180 do { \
181 elf_info[ei_index++] = id; \
182 elf_info[ei_index++] = val; \
183 } while (0)
184
185 #ifdef ARCH_DLINFO
186 /*
187 * ARCH_DLINFO must come first so PPC can do its special alignment of
188 * AUXV.
189 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
190 * ARCH_DLINFO changes
191 */
192 ARCH_DLINFO;
193 #endif
194 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
195 NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
196 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
197 NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
198 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
199 NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
200 NEW_AUX_ENT(AT_BASE, interp_load_addr);
201 NEW_AUX_ENT(AT_FLAGS, 0);
202 NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
203 NEW_AUX_ENT(AT_UID, tsk->uid);
204 NEW_AUX_ENT(AT_EUID, tsk->euid);
205 NEW_AUX_ENT(AT_GID, tsk->gid);
206 NEW_AUX_ENT(AT_EGID, tsk->egid);
207 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
208 if (k_platform) {
209 NEW_AUX_ENT(AT_PLATFORM,
210 (elf_addr_t)(unsigned long)u_platform);
211 }
212 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
213 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
214 }
215 #undef NEW_AUX_ENT
216 /* AT_NULL is zero; clear the rest too */
217 memset(&elf_info[ei_index], 0,
218 sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
219
220 /* And advance past the AT_NULL entry. */
221 ei_index += 2;
222
223 sp = STACK_ADD(p, ei_index);
224
225 items = (argc + 1) + (envc + 1);
226 if (interp_aout) {
227 items += 3; /* a.out interpreters require argv & envp too */
228 } else {
229 items += 1; /* ELF interpreters only put argc on the stack */
230 }
231 bprm->p = STACK_ROUND(sp, items);
232
233 /* Point sp at the lowest address on the stack */
234 #ifdef CONFIG_STACK_GROWSUP
235 sp = (elf_addr_t __user *)bprm->p - items - ei_index;
236 bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
237 #else
238 sp = (elf_addr_t __user *)bprm->p;
239 #endif
240
241
242 /*
243 * Grow the stack manually; some architectures have a limit on how
244 * far ahead a user-space access may be in order to grow the stack.
245 */
246 vma = find_extend_vma(current->mm, bprm->p);
247 if (!vma)
248 return -EFAULT;
249
250 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
251 if (__put_user(argc, sp++))
252 return -EFAULT;
253 if (interp_aout) {
254 argv = sp + 2;
255 envp = argv + argc + 1;
256 if (__put_user((elf_addr_t)(unsigned long)argv, sp++) ||
257 __put_user((elf_addr_t)(unsigned long)envp, sp++))
258 return -EFAULT;
259 } else {
260 argv = sp;
261 envp = argv + argc + 1;
262 }
263
264 /* Populate argv and envp */
265 p = current->mm->arg_end = current->mm->arg_start;
266 while (argc-- > 0) {
267 size_t len;
268 if (__put_user((elf_addr_t)p, argv++))
269 return -EFAULT;
270 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
271 if (!len || len > MAX_ARG_STRLEN)
272 return 0;
273 p += len;
274 }
275 if (__put_user(0, argv))
276 return -EFAULT;
277 current->mm->arg_end = current->mm->env_start = p;
278 while (envc-- > 0) {
279 size_t len;
280 if (__put_user((elf_addr_t)p, envp++))
281 return -EFAULT;
282 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
283 if (!len || len > MAX_ARG_STRLEN)
284 return 0;
285 p += len;
286 }
287 if (__put_user(0, envp))
288 return -EFAULT;
289 current->mm->env_end = p;
290
291 /* Put the elf_info on the stack in the right place. */
292 sp = (elf_addr_t __user *)envp + 1;
293 if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
294 return -EFAULT;
295 return 0;
296 }
297
298 #ifndef elf_map
299
300 static unsigned long elf_map(struct file *filep, unsigned long addr,
301 struct elf_phdr *eppnt, int prot, int type)
302 {
303 unsigned long map_addr;
304 unsigned long pageoffset = ELF_PAGEOFFSET(eppnt->p_vaddr);
305
306 down_write(&current->mm->mmap_sem);
307 /* mmap() will return -EINVAL if given a zero size, but a
308 * segment with zero filesize is perfectly valid */
309 if (eppnt->p_filesz + pageoffset)
310 map_addr = do_mmap(filep, ELF_PAGESTART(addr),
311 eppnt->p_filesz + pageoffset, prot, type,
312 eppnt->p_offset - pageoffset);
313 else
314 map_addr = ELF_PAGESTART(addr);
315 up_write(&current->mm->mmap_sem);
316 return(map_addr);
317 }
318
319 #endif /* !elf_map */
320
321 /* This is much more generalized than the library routine read function,
322 so we keep this separate. Technically the library read function
323 is only provided so that we can read a.out libraries that have
324 an ELF header */
325
326 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
327 struct file *interpreter, unsigned long *interp_load_addr)
328 {
329 struct elf_phdr *elf_phdata;
330 struct elf_phdr *eppnt;
331 unsigned long load_addr = 0;
332 int load_addr_set = 0;
333 unsigned long last_bss = 0, elf_bss = 0;
334 unsigned long error = ~0UL;
335 int retval, i, size;
336
337 /* First of all, some simple consistency checks */
338 if (interp_elf_ex->e_type != ET_EXEC &&
339 interp_elf_ex->e_type != ET_DYN)
340 goto out;
341 if (!elf_check_arch(interp_elf_ex))
342 goto out;
343 if (!interpreter->f_op || !interpreter->f_op->mmap)
344 goto out;
345
346 /*
347 * If the size of this structure has changed, then punt, since
348 * we will be doing the wrong thing.
349 */
350 if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
351 goto out;
352 if (interp_elf_ex->e_phnum < 1 ||
353 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
354 goto out;
355
356 /* Now read in all of the header information */
357 size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
358 if (size > ELF_MIN_ALIGN)
359 goto out;
360 elf_phdata = kmalloc(size, GFP_KERNEL);
361 if (!elf_phdata)
362 goto out;
363
364 retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
365 (char *)elf_phdata,size);
366 error = -EIO;
367 if (retval != size) {
368 if (retval < 0)
369 error = retval;
370 goto out_close;
371 }
372
373 eppnt = elf_phdata;
374 for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
375 if (eppnt->p_type == PT_LOAD) {
376 int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
377 int elf_prot = 0;
378 unsigned long vaddr = 0;
379 unsigned long k, map_addr;
380
381 if (eppnt->p_flags & PF_R)
382 elf_prot = PROT_READ;
383 if (eppnt->p_flags & PF_W)
384 elf_prot |= PROT_WRITE;
385 if (eppnt->p_flags & PF_X)
386 elf_prot |= PROT_EXEC;
387 vaddr = eppnt->p_vaddr;
388 if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
389 elf_type |= MAP_FIXED;
390
391 map_addr = elf_map(interpreter, load_addr + vaddr,
392 eppnt, elf_prot, elf_type);
393 error = map_addr;
394 if (BAD_ADDR(map_addr))
395 goto out_close;
396
397 if (!load_addr_set &&
398 interp_elf_ex->e_type == ET_DYN) {
399 load_addr = map_addr - ELF_PAGESTART(vaddr);
400 load_addr_set = 1;
401 }
402
403 /*
404 * Check to see if the section's size will overflow the
405 * allowed task size. Note that p_filesz must always be
406 * <= p_memsize so it's only necessary to check p_memsz.
407 */
408 k = load_addr + eppnt->p_vaddr;
409 if (BAD_ADDR(k) ||
410 eppnt->p_filesz > eppnt->p_memsz ||
411 eppnt->p_memsz > TASK_SIZE ||
412 TASK_SIZE - eppnt->p_memsz < k) {
413 error = -ENOMEM;
414 goto out_close;
415 }
416
417 /*
418 * Find the end of the file mapping for this phdr, and
419 * keep track of the largest address we see for this.
420 */
421 k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
422 if (k > elf_bss)
423 elf_bss = k;
424
425 /*
426 * Do the same thing for the memory mapping - between
427 * elf_bss and last_bss is the bss section.
428 */
429 k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
430 if (k > last_bss)
431 last_bss = k;
432 }
433 }
434
435 /*
436 * Now fill out the bss section. First pad the last page up
437 * to the page boundary, and then perform a mmap to make sure
438 * that there are zero-mapped pages up to and including the
439 * last bss page.
440 */
441 if (padzero(elf_bss)) {
442 error = -EFAULT;
443 goto out_close;
444 }
445
446 /* What we have mapped so far */
447 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
448
449 /* Map the last of the bss segment */
450 if (last_bss > elf_bss) {
451 down_write(&current->mm->mmap_sem);
452 error = do_brk(elf_bss, last_bss - elf_bss);
453 up_write(&current->mm->mmap_sem);
454 if (BAD_ADDR(error))
455 goto out_close;
456 }
457
458 *interp_load_addr = load_addr;
459 error = ((unsigned long)interp_elf_ex->e_entry) + load_addr;
460
461 out_close:
462 kfree(elf_phdata);
463 out:
464 return error;
465 }
466
467 static unsigned long load_aout_interp(struct exec *interp_ex,
468 struct file *interpreter)
469 {
470 unsigned long text_data, elf_entry = ~0UL;
471 char __user * addr;
472 loff_t offset;
473
474 current->mm->end_code = interp_ex->a_text;
475 text_data = interp_ex->a_text + interp_ex->a_data;
476 current->mm->end_data = text_data;
477 current->mm->brk = interp_ex->a_bss + text_data;
478
479 switch (N_MAGIC(*interp_ex)) {
480 case OMAGIC:
481 offset = 32;
482 addr = (char __user *)0;
483 break;
484 case ZMAGIC:
485 case QMAGIC:
486 offset = N_TXTOFF(*interp_ex);
487 addr = (char __user *)N_TXTADDR(*interp_ex);
488 break;
489 default:
490 goto out;
491 }
492
493 down_write(&current->mm->mmap_sem);
494 do_brk(0, text_data);
495 up_write(&current->mm->mmap_sem);
496 if (!interpreter->f_op || !interpreter->f_op->read)
497 goto out;
498 if (interpreter->f_op->read(interpreter, addr, text_data, &offset) < 0)
499 goto out;
500 flush_icache_range((unsigned long)addr,
501 (unsigned long)addr + text_data);
502
503 down_write(&current->mm->mmap_sem);
504 do_brk(ELF_PAGESTART(text_data + ELF_MIN_ALIGN - 1),
505 interp_ex->a_bss);
506 up_write(&current->mm->mmap_sem);
507 elf_entry = interp_ex->a_entry;
508
509 out:
510 return elf_entry;
511 }
512
513 /*
514 * These are the functions used to load ELF style executables and shared
515 * libraries. There is no binary dependent code anywhere else.
516 */
517
518 #define INTERPRETER_NONE 0
519 #define INTERPRETER_AOUT 1
520 #define INTERPRETER_ELF 2
521
522 #ifndef STACK_RND_MASK
523 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
524 #endif
525
526 static unsigned long randomize_stack_top(unsigned long stack_top)
527 {
528 unsigned int random_variable = 0;
529
530 if ((current->flags & PF_RANDOMIZE) &&
531 !(current->personality & ADDR_NO_RANDOMIZE)) {
532 random_variable = get_random_int() & STACK_RND_MASK;
533 random_variable <<= PAGE_SHIFT;
534 }
535 #ifdef CONFIG_STACK_GROWSUP
536 return PAGE_ALIGN(stack_top) + random_variable;
537 #else
538 return PAGE_ALIGN(stack_top) - random_variable;
539 #endif
540 }
541
542 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
543 {
544 struct file *interpreter = NULL; /* to shut gcc up */
545 unsigned long load_addr = 0, load_bias = 0;
546 int load_addr_set = 0;
547 char * elf_interpreter = NULL;
548 unsigned int interpreter_type = INTERPRETER_NONE;
549 unsigned char ibcs2_interpreter = 0;
550 unsigned long error;
551 struct elf_phdr *elf_ppnt, *elf_phdata;
552 unsigned long elf_bss, elf_brk;
553 int elf_exec_fileno;
554 int retval, i;
555 unsigned int size;
556 unsigned long elf_entry, interp_load_addr = 0;
557 unsigned long start_code, end_code, start_data, end_data;
558 unsigned long reloc_func_desc = 0;
559 char passed_fileno[6];
560 struct files_struct *files;
561 int executable_stack = EXSTACK_DEFAULT;
562 unsigned long def_flags = 0;
563 struct {
564 struct elfhdr elf_ex;
565 struct elfhdr interp_elf_ex;
566 struct exec interp_ex;
567 } *loc;
568
569 loc = kmalloc(sizeof(*loc), GFP_KERNEL);
570 if (!loc) {
571 retval = -ENOMEM;
572 goto out_ret;
573 }
574
575 /* Get the exec-header */
576 loc->elf_ex = *((struct elfhdr *)bprm->buf);
577
578 retval = -ENOEXEC;
579 /* First of all, some simple consistency checks */
580 if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
581 goto out;
582
583 if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
584 goto out;
585 if (!elf_check_arch(&loc->elf_ex))
586 goto out;
587 if (!bprm->file->f_op||!bprm->file->f_op->mmap)
588 goto out;
589
590 /* Now read in all of the header information */
591 if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
592 goto out;
593 if (loc->elf_ex.e_phnum < 1 ||
594 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
595 goto out;
596 size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
597 retval = -ENOMEM;
598 elf_phdata = kmalloc(size, GFP_KERNEL);
599 if (!elf_phdata)
600 goto out;
601
602 retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
603 (char *)elf_phdata, size);
604 if (retval != size) {
605 if (retval >= 0)
606 retval = -EIO;
607 goto out_free_ph;
608 }
609
610 files = current->files; /* Refcounted so ok */
611 retval = unshare_files();
612 if (retval < 0)
613 goto out_free_ph;
614 if (files == current->files) {
615 put_files_struct(files);
616 files = NULL;
617 }
618
619 /* exec will make our files private anyway, but for the a.out
620 loader stuff we need to do it earlier */
621 retval = get_unused_fd();
622 if (retval < 0)
623 goto out_free_fh;
624 get_file(bprm->file);
625 fd_install(elf_exec_fileno = retval, bprm->file);
626
627 elf_ppnt = elf_phdata;
628 elf_bss = 0;
629 elf_brk = 0;
630
631 start_code = ~0UL;
632 end_code = 0;
633 start_data = 0;
634 end_data = 0;
635
636 for (i = 0; i < loc->elf_ex.e_phnum; i++) {
637 if (elf_ppnt->p_type == PT_INTERP) {
638 /* This is the program interpreter used for
639 * shared libraries - for now assume that this
640 * is an a.out format binary
641 */
642 retval = -ENOEXEC;
643 if (elf_ppnt->p_filesz > PATH_MAX ||
644 elf_ppnt->p_filesz < 2)
645 goto out_free_file;
646
647 retval = -ENOMEM;
648 elf_interpreter = kmalloc(elf_ppnt->p_filesz,
649 GFP_KERNEL);
650 if (!elf_interpreter)
651 goto out_free_file;
652
653 retval = kernel_read(bprm->file, elf_ppnt->p_offset,
654 elf_interpreter,
655 elf_ppnt->p_filesz);
656 if (retval != elf_ppnt->p_filesz) {
657 if (retval >= 0)
658 retval = -EIO;
659 goto out_free_interp;
660 }
661 /* make sure path is NULL terminated */
662 retval = -ENOEXEC;
663 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
664 goto out_free_interp;
665
666 /* If the program interpreter is one of these two,
667 * then assume an iBCS2 image. Otherwise assume
668 * a native linux image.
669 */
670 if (strcmp(elf_interpreter,"/usr/lib/libc.so.1") == 0 ||
671 strcmp(elf_interpreter,"/usr/lib/ld.so.1") == 0)
672 ibcs2_interpreter = 1;
673
674 /*
675 * The early SET_PERSONALITY here is so that the lookup
676 * for the interpreter happens in the namespace of the
677 * to-be-execed image. SET_PERSONALITY can select an
678 * alternate root.
679 *
680 * However, SET_PERSONALITY is NOT allowed to switch
681 * this task into the new images's memory mapping
682 * policy - that is, TASK_SIZE must still evaluate to
683 * that which is appropriate to the execing application.
684 * This is because exit_mmap() needs to have TASK_SIZE
685 * evaluate to the size of the old image.
686 *
687 * So if (say) a 64-bit application is execing a 32-bit
688 * application it is the architecture's responsibility
689 * to defer changing the value of TASK_SIZE until the
690 * switch really is going to happen - do this in
691 * flush_thread(). - akpm
692 */
693 SET_PERSONALITY(loc->elf_ex, ibcs2_interpreter);
694
695 interpreter = open_exec(elf_interpreter);
696 retval = PTR_ERR(interpreter);
697 if (IS_ERR(interpreter))
698 goto out_free_interp;
699
700 /*
701 * If the binary is not readable then enforce
702 * mm->dumpable = 0 regardless of the interpreter's
703 * permissions.
704 */
705 if (file_permission(interpreter, MAY_READ) < 0)
706 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
707
708 retval = kernel_read(interpreter, 0, bprm->buf,
709 BINPRM_BUF_SIZE);
710 if (retval != BINPRM_BUF_SIZE) {
711 if (retval >= 0)
712 retval = -EIO;
713 goto out_free_dentry;
714 }
715
716 /* Get the exec headers */
717 loc->interp_ex = *((struct exec *)bprm->buf);
718 loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
719 break;
720 }
721 elf_ppnt++;
722 }
723
724 elf_ppnt = elf_phdata;
725 for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
726 if (elf_ppnt->p_type == PT_GNU_STACK) {
727 if (elf_ppnt->p_flags & PF_X)
728 executable_stack = EXSTACK_ENABLE_X;
729 else
730 executable_stack = EXSTACK_DISABLE_X;
731 break;
732 }
733
734 /* Some simple consistency checks for the interpreter */
735 if (elf_interpreter) {
736 static int warn;
737 interpreter_type = INTERPRETER_ELF | INTERPRETER_AOUT;
738
739 /* Now figure out which format our binary is */
740 if ((N_MAGIC(loc->interp_ex) != OMAGIC) &&
741 (N_MAGIC(loc->interp_ex) != ZMAGIC) &&
742 (N_MAGIC(loc->interp_ex) != QMAGIC))
743 interpreter_type = INTERPRETER_ELF;
744
745 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
746 interpreter_type &= ~INTERPRETER_ELF;
747
748 if (interpreter_type == INTERPRETER_AOUT && warn < 10) {
749 printk(KERN_WARNING "a.out ELF interpreter %s is "
750 "deprecated and will not be supported "
751 "after Linux 2.6.25\n", elf_interpreter);
752 warn++;
753 }
754
755 retval = -ELIBBAD;
756 if (!interpreter_type)
757 goto out_free_dentry;
758
759 /* Make sure only one type was selected */
760 if ((interpreter_type & INTERPRETER_ELF) &&
761 interpreter_type != INTERPRETER_ELF) {
762 // FIXME - ratelimit this before re-enabling
763 // printk(KERN_WARNING "ELF: Ambiguous type, using ELF\n");
764 interpreter_type = INTERPRETER_ELF;
765 }
766 /* Verify the interpreter has a valid arch */
767 if ((interpreter_type == INTERPRETER_ELF) &&
768 !elf_check_arch(&loc->interp_elf_ex))
769 goto out_free_dentry;
770 } else {
771 /* Executables without an interpreter also need a personality */
772 SET_PERSONALITY(loc->elf_ex, ibcs2_interpreter);
773 }
774
775 /* OK, we are done with that, now set up the arg stuff,
776 and then start this sucker up */
777 if ((!bprm->sh_bang) && (interpreter_type == INTERPRETER_AOUT)) {
778 char *passed_p = passed_fileno;
779 sprintf(passed_fileno, "%d", elf_exec_fileno);
780
781 if (elf_interpreter) {
782 retval = copy_strings_kernel(1, &passed_p, bprm);
783 if (retval)
784 goto out_free_dentry;
785 bprm->argc++;
786 }
787 }
788
789 /* Flush all traces of the currently running executable */
790 retval = flush_old_exec(bprm);
791 if (retval)
792 goto out_free_dentry;
793
794 /* Discard our unneeded old files struct */
795 if (files) {
796 put_files_struct(files);
797 files = NULL;
798 }
799
800 /* OK, This is the point of no return */
801 current->flags &= ~PF_FORKNOEXEC;
802 current->mm->def_flags = def_flags;
803
804 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
805 may depend on the personality. */
806 SET_PERSONALITY(loc->elf_ex, ibcs2_interpreter);
807 if (elf_read_implies_exec(loc->elf_ex, executable_stack))
808 current->personality |= READ_IMPLIES_EXEC;
809
810 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
811 current->flags |= PF_RANDOMIZE;
812 arch_pick_mmap_layout(current->mm);
813
814 /* Do this so that we can load the interpreter, if need be. We will
815 change some of these later */
816 current->mm->free_area_cache = current->mm->mmap_base;
817 current->mm->cached_hole_size = 0;
818 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
819 executable_stack);
820 if (retval < 0) {
821 send_sig(SIGKILL, current, 0);
822 goto out_free_dentry;
823 }
824
825 current->mm->start_stack = bprm->p;
826
827 /* Now we do a little grungy work by mmaping the ELF image into
828 the correct location in memory. At this point, we assume that
829 the image should be loaded at fixed address, not at a variable
830 address. */
831 for(i = 0, elf_ppnt = elf_phdata;
832 i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
833 int elf_prot = 0, elf_flags;
834 unsigned long k, vaddr;
835
836 if (elf_ppnt->p_type != PT_LOAD)
837 continue;
838
839 if (unlikely (elf_brk > elf_bss)) {
840 unsigned long nbyte;
841
842 /* There was a PT_LOAD segment with p_memsz > p_filesz
843 before this one. Map anonymous pages, if needed,
844 and clear the area. */
845 retval = set_brk (elf_bss + load_bias,
846 elf_brk + load_bias);
847 if (retval) {
848 send_sig(SIGKILL, current, 0);
849 goto out_free_dentry;
850 }
851 nbyte = ELF_PAGEOFFSET(elf_bss);
852 if (nbyte) {
853 nbyte = ELF_MIN_ALIGN - nbyte;
854 if (nbyte > elf_brk - elf_bss)
855 nbyte = elf_brk - elf_bss;
856 if (clear_user((void __user *)elf_bss +
857 load_bias, nbyte)) {
858 /*
859 * This bss-zeroing can fail if the ELF
860 * file specifies odd protections. So
861 * we don't check the return value
862 */
863 }
864 }
865 }
866
867 if (elf_ppnt->p_flags & PF_R)
868 elf_prot |= PROT_READ;
869 if (elf_ppnt->p_flags & PF_W)
870 elf_prot |= PROT_WRITE;
871 if (elf_ppnt->p_flags & PF_X)
872 elf_prot |= PROT_EXEC;
873
874 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
875
876 vaddr = elf_ppnt->p_vaddr;
877 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
878 elf_flags |= MAP_FIXED;
879 } else if (loc->elf_ex.e_type == ET_DYN) {
880 /* Try and get dynamic programs out of the way of the
881 * default mmap base, as well as whatever program they
882 * might try to exec. This is because the brk will
883 * follow the loader, and is not movable. */
884 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
885 }
886
887 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
888 elf_prot, elf_flags);
889 if (BAD_ADDR(error)) {
890 send_sig(SIGKILL, current, 0);
891 retval = IS_ERR((void *)error) ?
892 PTR_ERR((void*)error) : -EINVAL;
893 goto out_free_dentry;
894 }
895
896 if (!load_addr_set) {
897 load_addr_set = 1;
898 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
899 if (loc->elf_ex.e_type == ET_DYN) {
900 load_bias += error -
901 ELF_PAGESTART(load_bias + vaddr);
902 load_addr += load_bias;
903 reloc_func_desc = load_bias;
904 }
905 }
906 k = elf_ppnt->p_vaddr;
907 if (k < start_code)
908 start_code = k;
909 if (start_data < k)
910 start_data = k;
911
912 /*
913 * Check to see if the section's size will overflow the
914 * allowed task size. Note that p_filesz must always be
915 * <= p_memsz so it is only necessary to check p_memsz.
916 */
917 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
918 elf_ppnt->p_memsz > TASK_SIZE ||
919 TASK_SIZE - elf_ppnt->p_memsz < k) {
920 /* set_brk can never work. Avoid overflows. */
921 send_sig(SIGKILL, current, 0);
922 retval = -EINVAL;
923 goto out_free_dentry;
924 }
925
926 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
927
928 if (k > elf_bss)
929 elf_bss = k;
930 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
931 end_code = k;
932 if (end_data < k)
933 end_data = k;
934 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
935 if (k > elf_brk)
936 elf_brk = k;
937 }
938
939 loc->elf_ex.e_entry += load_bias;
940 elf_bss += load_bias;
941 elf_brk += load_bias;
942 start_code += load_bias;
943 end_code += load_bias;
944 start_data += load_bias;
945 end_data += load_bias;
946
947 /* Calling set_brk effectively mmaps the pages that we need
948 * for the bss and break sections. We must do this before
949 * mapping in the interpreter, to make sure it doesn't wind
950 * up getting placed where the bss needs to go.
951 */
952 retval = set_brk(elf_bss, elf_brk);
953 if (retval) {
954 send_sig(SIGKILL, current, 0);
955 goto out_free_dentry;
956 }
957 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
958 send_sig(SIGSEGV, current, 0);
959 retval = -EFAULT; /* Nobody gets to see this, but.. */
960 goto out_free_dentry;
961 }
962
963 if (elf_interpreter) {
964 if (interpreter_type == INTERPRETER_AOUT)
965 elf_entry = load_aout_interp(&loc->interp_ex,
966 interpreter);
967 else
968 elf_entry = load_elf_interp(&loc->interp_elf_ex,
969 interpreter,
970 &interp_load_addr);
971 if (BAD_ADDR(elf_entry)) {
972 force_sig(SIGSEGV, current);
973 retval = IS_ERR((void *)elf_entry) ?
974 (int)elf_entry : -EINVAL;
975 goto out_free_dentry;
976 }
977 reloc_func_desc = interp_load_addr;
978
979 allow_write_access(interpreter);
980 fput(interpreter);
981 kfree(elf_interpreter);
982 } else {
983 elf_entry = loc->elf_ex.e_entry;
984 if (BAD_ADDR(elf_entry)) {
985 force_sig(SIGSEGV, current);
986 retval = -EINVAL;
987 goto out_free_dentry;
988 }
989 }
990
991 kfree(elf_phdata);
992
993 if (interpreter_type != INTERPRETER_AOUT)
994 sys_close(elf_exec_fileno);
995
996 set_binfmt(&elf_format);
997
998 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
999 retval = arch_setup_additional_pages(bprm, executable_stack);
1000 if (retval < 0) {
1001 send_sig(SIGKILL, current, 0);
1002 goto out;
1003 }
1004 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
1005
1006 compute_creds(bprm);
1007 current->flags &= ~PF_FORKNOEXEC;
1008 retval = create_elf_tables(bprm, &loc->elf_ex,
1009 (interpreter_type == INTERPRETER_AOUT),
1010 load_addr, interp_load_addr);
1011 if (retval < 0) {
1012 send_sig(SIGKILL, current, 0);
1013 goto out;
1014 }
1015 /* N.B. passed_fileno might not be initialized? */
1016 if (interpreter_type == INTERPRETER_AOUT)
1017 current->mm->arg_start += strlen(passed_fileno) + 1;
1018 current->mm->end_code = end_code;
1019 current->mm->start_code = start_code;
1020 current->mm->start_data = start_data;
1021 current->mm->end_data = end_data;
1022 current->mm->start_stack = bprm->p;
1023
1024 if (current->personality & MMAP_PAGE_ZERO) {
1025 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
1026 and some applications "depend" upon this behavior.
1027 Since we do not have the power to recompile these, we
1028 emulate the SVr4 behavior. Sigh. */
1029 down_write(&current->mm->mmap_sem);
1030 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
1031 MAP_FIXED | MAP_PRIVATE, 0);
1032 up_write(&current->mm->mmap_sem);
1033 }
1034
1035 #ifdef ELF_PLAT_INIT
1036 /*
1037 * The ABI may specify that certain registers be set up in special
1038 * ways (on i386 %edx is the address of a DT_FINI function, for
1039 * example. In addition, it may also specify (eg, PowerPC64 ELF)
1040 * that the e_entry field is the address of the function descriptor
1041 * for the startup routine, rather than the address of the startup
1042 * routine itself. This macro performs whatever initialization to
1043 * the regs structure is required as well as any relocations to the
1044 * function descriptor entries when executing dynamically links apps.
1045 */
1046 ELF_PLAT_INIT(regs, reloc_func_desc);
1047 #endif
1048
1049 start_thread(regs, elf_entry, bprm->p);
1050 if (unlikely(current->ptrace & PT_PTRACED)) {
1051 if (current->ptrace & PT_TRACE_EXEC)
1052 ptrace_notify ((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
1053 else
1054 send_sig(SIGTRAP, current, 0);
1055 }
1056 retval = 0;
1057 out:
1058 kfree(loc);
1059 out_ret:
1060 return retval;
1061
1062 /* error cleanup */
1063 out_free_dentry:
1064 allow_write_access(interpreter);
1065 if (interpreter)
1066 fput(interpreter);
1067 out_free_interp:
1068 kfree(elf_interpreter);
1069 out_free_file:
1070 sys_close(elf_exec_fileno);
1071 out_free_fh:
1072 if (files)
1073 reset_files_struct(current, files);
1074 out_free_ph:
1075 kfree(elf_phdata);
1076 goto out;
1077 }
1078
1079 /* This is really simpleminded and specialized - we are loading an
1080 a.out library that is given an ELF header. */
1081 static int load_elf_library(struct file *file)
1082 {
1083 struct elf_phdr *elf_phdata;
1084 struct elf_phdr *eppnt;
1085 unsigned long elf_bss, bss, len;
1086 int retval, error, i, j;
1087 struct elfhdr elf_ex;
1088
1089 error = -ENOEXEC;
1090 retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1091 if (retval != sizeof(elf_ex))
1092 goto out;
1093
1094 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1095 goto out;
1096
1097 /* First of all, some simple consistency checks */
1098 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1099 !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1100 goto out;
1101
1102 /* Now read in all of the header information */
1103
1104 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1105 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1106
1107 error = -ENOMEM;
1108 elf_phdata = kmalloc(j, GFP_KERNEL);
1109 if (!elf_phdata)
1110 goto out;
1111
1112 eppnt = elf_phdata;
1113 error = -ENOEXEC;
1114 retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1115 if (retval != j)
1116 goto out_free_ph;
1117
1118 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1119 if ((eppnt + i)->p_type == PT_LOAD)
1120 j++;
1121 if (j != 1)
1122 goto out_free_ph;
1123
1124 while (eppnt->p_type != PT_LOAD)
1125 eppnt++;
1126
1127 /* Now use mmap to map the library into memory. */
1128 down_write(&current->mm->mmap_sem);
1129 error = do_mmap(file,
1130 ELF_PAGESTART(eppnt->p_vaddr),
1131 (eppnt->p_filesz +
1132 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1133 PROT_READ | PROT_WRITE | PROT_EXEC,
1134 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1135 (eppnt->p_offset -
1136 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1137 up_write(&current->mm->mmap_sem);
1138 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1139 goto out_free_ph;
1140
1141 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1142 if (padzero(elf_bss)) {
1143 error = -EFAULT;
1144 goto out_free_ph;
1145 }
1146
1147 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1148 ELF_MIN_ALIGN - 1);
1149 bss = eppnt->p_memsz + eppnt->p_vaddr;
1150 if (bss > len) {
1151 down_write(&current->mm->mmap_sem);
1152 do_brk(len, bss - len);
1153 up_write(&current->mm->mmap_sem);
1154 }
1155 error = 0;
1156
1157 out_free_ph:
1158 kfree(elf_phdata);
1159 out:
1160 return error;
1161 }
1162
1163 /*
1164 * Note that some platforms still use traditional core dumps and not
1165 * the ELF core dump. Each platform can select it as appropriate.
1166 */
1167 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1168
1169 /*
1170 * ELF core dumper
1171 *
1172 * Modelled on fs/exec.c:aout_core_dump()
1173 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1174 */
1175 /*
1176 * These are the only things you should do on a core-file: use only these
1177 * functions to write out all the necessary info.
1178 */
1179 static int dump_write(struct file *file, const void *addr, int nr)
1180 {
1181 return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1182 }
1183
1184 static int dump_seek(struct file *file, loff_t off)
1185 {
1186 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
1187 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
1188 return 0;
1189 } else {
1190 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
1191 if (!buf)
1192 return 0;
1193 while (off > 0) {
1194 unsigned long n = off;
1195 if (n > PAGE_SIZE)
1196 n = PAGE_SIZE;
1197 if (!dump_write(file, buf, n))
1198 return 0;
1199 off -= n;
1200 }
1201 free_page((unsigned long)buf);
1202 }
1203 return 1;
1204 }
1205
1206 /*
1207 * Decide what to dump of a segment, part, all or none.
1208 */
1209 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1210 unsigned long mm_flags)
1211 {
1212 /* The vma can be set up to tell us the answer directly. */
1213 if (vma->vm_flags & VM_ALWAYSDUMP)
1214 goto whole;
1215
1216 /* Do not dump I/O mapped devices or special mappings */
1217 if (vma->vm_flags & (VM_IO | VM_RESERVED))
1218 return 0;
1219
1220 #define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1221
1222 /* By default, dump shared memory if mapped from an anonymous file. */
1223 if (vma->vm_flags & VM_SHARED) {
1224 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1225 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1226 goto whole;
1227 return 0;
1228 }
1229
1230 /* Dump segments that have been written to. */
1231 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1232 goto whole;
1233 if (vma->vm_file == NULL)
1234 return 0;
1235
1236 if (FILTER(MAPPED_PRIVATE))
1237 goto whole;
1238
1239 /*
1240 * If this looks like the beginning of a DSO or executable mapping,
1241 * check for an ELF header. If we find one, dump the first page to
1242 * aid in determining what was mapped here.
1243 */
1244 if (FILTER(ELF_HEADERS) && vma->vm_file != NULL && vma->vm_pgoff == 0) {
1245 u32 __user *header = (u32 __user *) vma->vm_start;
1246 u32 word;
1247 /*
1248 * Doing it this way gets the constant folded by GCC.
1249 */
1250 union {
1251 u32 cmp;
1252 char elfmag[SELFMAG];
1253 } magic;
1254 BUILD_BUG_ON(SELFMAG != sizeof word);
1255 magic.elfmag[EI_MAG0] = ELFMAG0;
1256 magic.elfmag[EI_MAG1] = ELFMAG1;
1257 magic.elfmag[EI_MAG2] = ELFMAG2;
1258 magic.elfmag[EI_MAG3] = ELFMAG3;
1259 if (get_user(word, header) == 0 && word == magic.cmp)
1260 return PAGE_SIZE;
1261 }
1262
1263 #undef FILTER
1264
1265 return 0;
1266
1267 whole:
1268 return vma->vm_end - vma->vm_start;
1269 }
1270
1271 /* An ELF note in memory */
1272 struct memelfnote
1273 {
1274 const char *name;
1275 int type;
1276 unsigned int datasz;
1277 void *data;
1278 };
1279
1280 static int notesize(struct memelfnote *en)
1281 {
1282 int sz;
1283
1284 sz = sizeof(struct elf_note);
1285 sz += roundup(strlen(en->name) + 1, 4);
1286 sz += roundup(en->datasz, 4);
1287
1288 return sz;
1289 }
1290
1291 #define DUMP_WRITE(addr, nr, foffset) \
1292 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1293
1294 static int alignfile(struct file *file, loff_t *foffset)
1295 {
1296 static const char buf[4] = { 0, };
1297 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1298 return 1;
1299 }
1300
1301 static int writenote(struct memelfnote *men, struct file *file,
1302 loff_t *foffset)
1303 {
1304 struct elf_note en;
1305 en.n_namesz = strlen(men->name) + 1;
1306 en.n_descsz = men->datasz;
1307 en.n_type = men->type;
1308
1309 DUMP_WRITE(&en, sizeof(en), foffset);
1310 DUMP_WRITE(men->name, en.n_namesz, foffset);
1311 if (!alignfile(file, foffset))
1312 return 0;
1313 DUMP_WRITE(men->data, men->datasz, foffset);
1314 if (!alignfile(file, foffset))
1315 return 0;
1316
1317 return 1;
1318 }
1319 #undef DUMP_WRITE
1320
1321 #define DUMP_WRITE(addr, nr) \
1322 if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1323 goto end_coredump;
1324 #define DUMP_SEEK(off) \
1325 if (!dump_seek(file, (off))) \
1326 goto end_coredump;
1327
1328 static void fill_elf_header(struct elfhdr *elf, int segs)
1329 {
1330 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1331 elf->e_ident[EI_CLASS] = ELF_CLASS;
1332 elf->e_ident[EI_DATA] = ELF_DATA;
1333 elf->e_ident[EI_VERSION] = EV_CURRENT;
1334 elf->e_ident[EI_OSABI] = ELF_OSABI;
1335 memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
1336
1337 elf->e_type = ET_CORE;
1338 elf->e_machine = ELF_ARCH;
1339 elf->e_version = EV_CURRENT;
1340 elf->e_entry = 0;
1341 elf->e_phoff = sizeof(struct elfhdr);
1342 elf->e_shoff = 0;
1343 elf->e_flags = ELF_CORE_EFLAGS;
1344 elf->e_ehsize = sizeof(struct elfhdr);
1345 elf->e_phentsize = sizeof(struct elf_phdr);
1346 elf->e_phnum = segs;
1347 elf->e_shentsize = 0;
1348 elf->e_shnum = 0;
1349 elf->e_shstrndx = 0;
1350 return;
1351 }
1352
1353 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1354 {
1355 phdr->p_type = PT_NOTE;
1356 phdr->p_offset = offset;
1357 phdr->p_vaddr = 0;
1358 phdr->p_paddr = 0;
1359 phdr->p_filesz = sz;
1360 phdr->p_memsz = 0;
1361 phdr->p_flags = 0;
1362 phdr->p_align = 0;
1363 return;
1364 }
1365
1366 static void fill_note(struct memelfnote *note, const char *name, int type,
1367 unsigned int sz, void *data)
1368 {
1369 note->name = name;
1370 note->type = type;
1371 note->datasz = sz;
1372 note->data = data;
1373 return;
1374 }
1375
1376 /*
1377 * fill up all the fields in prstatus from the given task struct, except
1378 * registers which need to be filled up separately.
1379 */
1380 static void fill_prstatus(struct elf_prstatus *prstatus,
1381 struct task_struct *p, long signr)
1382 {
1383 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1384 prstatus->pr_sigpend = p->pending.signal.sig[0];
1385 prstatus->pr_sighold = p->blocked.sig[0];
1386 prstatus->pr_pid = task_pid_vnr(p);
1387 prstatus->pr_ppid = task_pid_vnr(p->real_parent);
1388 prstatus->pr_pgrp = task_pgrp_vnr(p);
1389 prstatus->pr_sid = task_session_vnr(p);
1390 if (thread_group_leader(p)) {
1391 /*
1392 * This is the record for the group leader. Add in the
1393 * cumulative times of previous dead threads. This total
1394 * won't include the time of each live thread whose state
1395 * is included in the core dump. The final total reported
1396 * to our parent process when it calls wait4 will include
1397 * those sums as well as the little bit more time it takes
1398 * this and each other thread to finish dying after the
1399 * core dump synchronization phase.
1400 */
1401 cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1402 &prstatus->pr_utime);
1403 cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1404 &prstatus->pr_stime);
1405 } else {
1406 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1407 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1408 }
1409 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1410 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1411 }
1412
1413 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1414 struct mm_struct *mm)
1415 {
1416 unsigned int i, len;
1417
1418 /* first copy the parameters from user space */
1419 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1420
1421 len = mm->arg_end - mm->arg_start;
1422 if (len >= ELF_PRARGSZ)
1423 len = ELF_PRARGSZ-1;
1424 if (copy_from_user(&psinfo->pr_psargs,
1425 (const char __user *)mm->arg_start, len))
1426 return -EFAULT;
1427 for(i = 0; i < len; i++)
1428 if (psinfo->pr_psargs[i] == 0)
1429 psinfo->pr_psargs[i] = ' ';
1430 psinfo->pr_psargs[len] = 0;
1431
1432 psinfo->pr_pid = task_pid_vnr(p);
1433 psinfo->pr_ppid = task_pid_vnr(p->real_parent);
1434 psinfo->pr_pgrp = task_pgrp_vnr(p);
1435 psinfo->pr_sid = task_session_vnr(p);
1436
1437 i = p->state ? ffz(~p->state) + 1 : 0;
1438 psinfo->pr_state = i;
1439 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1440 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1441 psinfo->pr_nice = task_nice(p);
1442 psinfo->pr_flag = p->flags;
1443 SET_UID(psinfo->pr_uid, p->uid);
1444 SET_GID(psinfo->pr_gid, p->gid);
1445 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1446
1447 return 0;
1448 }
1449
1450 /* Here is the structure in which status of each thread is captured. */
1451 struct elf_thread_status
1452 {
1453 struct list_head list;
1454 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1455 elf_fpregset_t fpu; /* NT_PRFPREG */
1456 struct task_struct *thread;
1457 #ifdef ELF_CORE_COPY_XFPREGS
1458 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
1459 #endif
1460 struct memelfnote notes[3];
1461 int num_notes;
1462 };
1463
1464 /*
1465 * In order to add the specific thread information for the elf file format,
1466 * we need to keep a linked list of every threads pr_status and then create
1467 * a single section for them in the final core file.
1468 */
1469 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1470 {
1471 int sz = 0;
1472 struct task_struct *p = t->thread;
1473 t->num_notes = 0;
1474
1475 fill_prstatus(&t->prstatus, p, signr);
1476 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1477
1478 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1479 &(t->prstatus));
1480 t->num_notes++;
1481 sz += notesize(&t->notes[0]);
1482
1483 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1484 &t->fpu))) {
1485 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1486 &(t->fpu));
1487 t->num_notes++;
1488 sz += notesize(&t->notes[1]);
1489 }
1490
1491 #ifdef ELF_CORE_COPY_XFPREGS
1492 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1493 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1494 sizeof(t->xfpu), &t->xfpu);
1495 t->num_notes++;
1496 sz += notesize(&t->notes[2]);
1497 }
1498 #endif
1499 return sz;
1500 }
1501
1502 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1503 struct vm_area_struct *gate_vma)
1504 {
1505 struct vm_area_struct *ret = tsk->mm->mmap;
1506
1507 if (ret)
1508 return ret;
1509 return gate_vma;
1510 }
1511 /*
1512 * Helper function for iterating across a vma list. It ensures that the caller
1513 * will visit `gate_vma' prior to terminating the search.
1514 */
1515 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1516 struct vm_area_struct *gate_vma)
1517 {
1518 struct vm_area_struct *ret;
1519
1520 ret = this_vma->vm_next;
1521 if (ret)
1522 return ret;
1523 if (this_vma == gate_vma)
1524 return NULL;
1525 return gate_vma;
1526 }
1527
1528 /*
1529 * Actual dumper
1530 *
1531 * This is a two-pass process; first we find the offsets of the bits,
1532 * and then they are actually written out. If we run out of core limit
1533 * we just truncate.
1534 */
1535 static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, unsigned long limit)
1536 {
1537 #define NUM_NOTES 6
1538 int has_dumped = 0;
1539 mm_segment_t fs;
1540 int segs;
1541 size_t size = 0;
1542 int i;
1543 struct vm_area_struct *vma, *gate_vma;
1544 struct elfhdr *elf = NULL;
1545 loff_t offset = 0, dataoff, foffset;
1546 int numnote;
1547 struct memelfnote *notes = NULL;
1548 struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */
1549 struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
1550 struct task_struct *g, *p;
1551 LIST_HEAD(thread_list);
1552 struct list_head *t;
1553 elf_fpregset_t *fpu = NULL;
1554 #ifdef ELF_CORE_COPY_XFPREGS
1555 elf_fpxregset_t *xfpu = NULL;
1556 #endif
1557 int thread_status_size = 0;
1558 elf_addr_t *auxv;
1559 unsigned long mm_flags;
1560
1561 /*
1562 * We no longer stop all VM operations.
1563 *
1564 * This is because those proceses that could possibly change map_count
1565 * or the mmap / vma pages are now blocked in do_exit on current
1566 * finishing this core dump.
1567 *
1568 * Only ptrace can touch these memory addresses, but it doesn't change
1569 * the map_count or the pages allocated. So no possibility of crashing
1570 * exists while dumping the mm->vm_next areas to the core file.
1571 */
1572
1573 /* alloc memory for large data structures: too large to be on stack */
1574 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1575 if (!elf)
1576 goto cleanup;
1577 prstatus = kmalloc(sizeof(*prstatus), GFP_KERNEL);
1578 if (!prstatus)
1579 goto cleanup;
1580 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1581 if (!psinfo)
1582 goto cleanup;
1583 notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
1584 if (!notes)
1585 goto cleanup;
1586 fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
1587 if (!fpu)
1588 goto cleanup;
1589 #ifdef ELF_CORE_COPY_XFPREGS
1590 xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
1591 if (!xfpu)
1592 goto cleanup;
1593 #endif
1594
1595 if (signr) {
1596 struct elf_thread_status *tmp;
1597 rcu_read_lock();
1598 do_each_thread(g,p)
1599 if (current->mm == p->mm && current != p) {
1600 tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC);
1601 if (!tmp) {
1602 rcu_read_unlock();
1603 goto cleanup;
1604 }
1605 tmp->thread = p;
1606 list_add(&tmp->list, &thread_list);
1607 }
1608 while_each_thread(g,p);
1609 rcu_read_unlock();
1610 list_for_each(t, &thread_list) {
1611 struct elf_thread_status *tmp;
1612 int sz;
1613
1614 tmp = list_entry(t, struct elf_thread_status, list);
1615 sz = elf_dump_thread_status(signr, tmp);
1616 thread_status_size += sz;
1617 }
1618 }
1619 /* now collect the dump for the current */
1620 memset(prstatus, 0, sizeof(*prstatus));
1621 fill_prstatus(prstatus, current, signr);
1622 elf_core_copy_regs(&prstatus->pr_reg, regs);
1623
1624 segs = current->mm->map_count;
1625 #ifdef ELF_CORE_EXTRA_PHDRS
1626 segs += ELF_CORE_EXTRA_PHDRS;
1627 #endif
1628
1629 gate_vma = get_gate_vma(current);
1630 if (gate_vma != NULL)
1631 segs++;
1632
1633 /* Set up header */
1634 fill_elf_header(elf, segs + 1); /* including notes section */
1635
1636 has_dumped = 1;
1637 current->flags |= PF_DUMPCORE;
1638
1639 /*
1640 * Set up the notes in similar form to SVR4 core dumps made
1641 * with info from their /proc.
1642 */
1643
1644 fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
1645 fill_psinfo(psinfo, current->group_leader, current->mm);
1646 fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1647
1648 numnote = 2;
1649
1650 auxv = (elf_addr_t *)current->mm->saved_auxv;
1651
1652 i = 0;
1653 do
1654 i += 2;
1655 while (auxv[i - 2] != AT_NULL);
1656 fill_note(&notes[numnote++], "CORE", NT_AUXV,
1657 i * sizeof(elf_addr_t), auxv);
1658
1659 /* Try to dump the FPU. */
1660 if ((prstatus->pr_fpvalid =
1661 elf_core_copy_task_fpregs(current, regs, fpu)))
1662 fill_note(notes + numnote++,
1663 "CORE", NT_PRFPREG, sizeof(*fpu), fpu);
1664 #ifdef ELF_CORE_COPY_XFPREGS
1665 if (elf_core_copy_task_xfpregs(current, xfpu))
1666 fill_note(notes + numnote++,
1667 "LINUX", ELF_CORE_XFPREG_TYPE, sizeof(*xfpu), xfpu);
1668 #endif
1669
1670 fs = get_fs();
1671 set_fs(KERNEL_DS);
1672
1673 DUMP_WRITE(elf, sizeof(*elf));
1674 offset += sizeof(*elf); /* Elf header */
1675 offset += (segs + 1) * sizeof(struct elf_phdr); /* Program headers */
1676 foffset = offset;
1677
1678 /* Write notes phdr entry */
1679 {
1680 struct elf_phdr phdr;
1681 int sz = 0;
1682
1683 for (i = 0; i < numnote; i++)
1684 sz += notesize(notes + i);
1685
1686 sz += thread_status_size;
1687
1688 sz += elf_coredump_extra_notes_size();
1689
1690 fill_elf_note_phdr(&phdr, sz, offset);
1691 offset += sz;
1692 DUMP_WRITE(&phdr, sizeof(phdr));
1693 }
1694
1695 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1696
1697 /*
1698 * We must use the same mm->flags while dumping core to avoid
1699 * inconsistency between the program headers and bodies, otherwise an
1700 * unusable core file can be generated.
1701 */
1702 mm_flags = current->mm->flags;
1703
1704 /* Write program headers for segments dump */
1705 for (vma = first_vma(current, gate_vma); vma != NULL;
1706 vma = next_vma(vma, gate_vma)) {
1707 struct elf_phdr phdr;
1708
1709 phdr.p_type = PT_LOAD;
1710 phdr.p_offset = offset;
1711 phdr.p_vaddr = vma->vm_start;
1712 phdr.p_paddr = 0;
1713 phdr.p_filesz = vma_dump_size(vma, mm_flags);
1714 phdr.p_memsz = vma->vm_end - vma->vm_start;
1715 offset += phdr.p_filesz;
1716 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1717 if (vma->vm_flags & VM_WRITE)
1718 phdr.p_flags |= PF_W;
1719 if (vma->vm_flags & VM_EXEC)
1720 phdr.p_flags |= PF_X;
1721 phdr.p_align = ELF_EXEC_PAGESIZE;
1722
1723 DUMP_WRITE(&phdr, sizeof(phdr));
1724 }
1725
1726 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1727 ELF_CORE_WRITE_EXTRA_PHDRS;
1728 #endif
1729
1730 /* write out the notes section */
1731 for (i = 0; i < numnote; i++)
1732 if (!writenote(notes + i, file, &foffset))
1733 goto end_coredump;
1734
1735 if (elf_coredump_extra_notes_write(file, &foffset))
1736 goto end_coredump;
1737
1738 /* write out the thread status notes section */
1739 list_for_each(t, &thread_list) {
1740 struct elf_thread_status *tmp =
1741 list_entry(t, struct elf_thread_status, list);
1742
1743 for (i = 0; i < tmp->num_notes; i++)
1744 if (!writenote(&tmp->notes[i], file, &foffset))
1745 goto end_coredump;
1746 }
1747
1748 /* Align to page */
1749 DUMP_SEEK(dataoff - foffset);
1750
1751 for (vma = first_vma(current, gate_vma); vma != NULL;
1752 vma = next_vma(vma, gate_vma)) {
1753 unsigned long addr;
1754 unsigned long end;
1755
1756 end = vma->vm_start + vma_dump_size(vma, mm_flags);
1757
1758 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
1759 struct page *page;
1760 struct vm_area_struct *vma;
1761
1762 if (get_user_pages(current, current->mm, addr, 1, 0, 1,
1763 &page, &vma) <= 0) {
1764 DUMP_SEEK(PAGE_SIZE);
1765 } else {
1766 if (page == ZERO_PAGE(0)) {
1767 if (!dump_seek(file, PAGE_SIZE)) {
1768 page_cache_release(page);
1769 goto end_coredump;
1770 }
1771 } else {
1772 void *kaddr;
1773 flush_cache_page(vma, addr,
1774 page_to_pfn(page));
1775 kaddr = kmap(page);
1776 if ((size += PAGE_SIZE) > limit ||
1777 !dump_write(file, kaddr,
1778 PAGE_SIZE)) {
1779 kunmap(page);
1780 page_cache_release(page);
1781 goto end_coredump;
1782 }
1783 kunmap(page);
1784 }
1785 page_cache_release(page);
1786 }
1787 }
1788 }
1789
1790 #ifdef ELF_CORE_WRITE_EXTRA_DATA
1791 ELF_CORE_WRITE_EXTRA_DATA;
1792 #endif
1793
1794 end_coredump:
1795 set_fs(fs);
1796
1797 cleanup:
1798 while (!list_empty(&thread_list)) {
1799 struct list_head *tmp = thread_list.next;
1800 list_del(tmp);
1801 kfree(list_entry(tmp, struct elf_thread_status, list));
1802 }
1803
1804 kfree(elf);
1805 kfree(prstatus);
1806 kfree(psinfo);
1807 kfree(notes);
1808 kfree(fpu);
1809 #ifdef ELF_CORE_COPY_XFPREGS
1810 kfree(xfpu);
1811 #endif
1812 return has_dumped;
1813 #undef NUM_NOTES
1814 }
1815
1816 #endif /* USE_ELF_CORE_DUMP */
1817
1818 static int __init init_elf_binfmt(void)
1819 {
1820 return register_binfmt(&elf_format);
1821 }
1822
1823 static void __exit exit_elf_binfmt(void)
1824 {
1825 /* Remove the COFF and ELF loaders. */
1826 unregister_binfmt(&elf_format);
1827 }
1828
1829 core_initcall(init_elf_binfmt);
1830 module_exit(exit_elf_binfmt);
1831 MODULE_LICENSE("GPL");
Community contributions to the Openmoko Kernel