projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branches 'release' and 'ppc-workaround' into release
[deliverable/linux.git] / fs / exec.c
1 /*
2 * linux/fs/exec.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7 /*
8 * #!-checking implemented by tytso.
9 */
10 /*
11 * Demand-loading implemented 01.12.91 - no need to read anything but
12 * the header into memory. The inode of the executable is put into
13 * "current->executable", and page faults do the actual loading. Clean.
14 *
15 * Once more I can proudly say that linux stood up to being changed: it
16 * was less than 2 hours work to get demand-loading completely implemented.
17 *
18 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
19 * current->executable is only used by the procfs. This allows a dispatch
20 * table to check for several different types of binary formats. We keep
21 * trying until we recognize the file or we run out of supported binary
22 * formats.
23 */
24
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/mman.h>
28 #include <linux/a.out.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/highmem.h>
36 #include <linux/spinlock.h>
37 #include <linux/key.h>
38 #include <linux/personality.h>
39 #include <linux/binfmts.h>
40 #include <linux/swap.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/rmap.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
57 #include <asm/tlb.h>
58
59 #ifdef CONFIG_KMOD
60 #include <linux/kmod.h>
61 #endif
62
63 int core_uses_pid;
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 int suid_dumpable = 0;
66
67 /* The maximal length of core_pattern is also specified in sysctl.c */
68
69 static LIST_HEAD(formats);
70 static DEFINE_RWLOCK(binfmt_lock);
71
72 int register_binfmt(struct linux_binfmt * fmt)
73 {
74 if (!fmt)
75 return -EINVAL;
76 write_lock(&binfmt_lock);
77 list_add(&fmt->lh, &formats);
78 write_unlock(&binfmt_lock);
79 return 0;
80 }
81
82 EXPORT_SYMBOL(register_binfmt);
83
84 void unregister_binfmt(struct linux_binfmt * fmt)
85 {
86 write_lock(&binfmt_lock);
87 list_del(&fmt->lh);
88 write_unlock(&binfmt_lock);
89 }
90
91 EXPORT_SYMBOL(unregister_binfmt);
92
93 static inline void put_binfmt(struct linux_binfmt * fmt)
94 {
95 module_put(fmt->module);
96 }
97
98 /*
99 * Note that a shared library must be both readable and executable due to
100 * security reasons.
101 *
102 * Also note that we take the address to load from from the file itself.
103 */
104 asmlinkage long sys_uselib(const char __user * library)
105 {
106 struct file * file;
107 struct nameidata nd;
108 int error;
109
110 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
111 if (error)
112 goto out;
113
114 error = -EINVAL;
115 if (!S_ISREG(nd.dentry->d_inode->i_mode))
116 goto exit;
117
118 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
119 if (error)
120 goto exit;
121
122 file = nameidata_to_filp(&nd, O_RDONLY);
123 error = PTR_ERR(file);
124 if (IS_ERR(file))
125 goto out;
126
127 error = -ENOEXEC;
128 if(file->f_op) {
129 struct linux_binfmt * fmt;
130
131 read_lock(&binfmt_lock);
132 list_for_each_entry(fmt, &formats, lh) {
133 if (!fmt->load_shlib)
134 continue;
135 if (!try_module_get(fmt->module))
136 continue;
137 read_unlock(&binfmt_lock);
138 error = fmt->load_shlib(file);
139 read_lock(&binfmt_lock);
140 put_binfmt(fmt);
141 if (error != -ENOEXEC)
142 break;
143 }
144 read_unlock(&binfmt_lock);
145 }
146 fput(file);
147 out:
148 return error;
149 exit:
150 release_open_intent(&nd);
151 path_release(&nd);
152 goto out;
153 }
154
155 #ifdef CONFIG_MMU
156
157 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
158 int write)
159 {
160 struct page *page;
161 int ret;
162
163 #ifdef CONFIG_STACK_GROWSUP
164 if (write) {
165 ret = expand_stack_downwards(bprm->vma, pos);
166 if (ret < 0)
167 return NULL;
168 }
169 #endif
170 ret = get_user_pages(current, bprm->mm, pos,
171 1, write, 1, &page, NULL);
172 if (ret <= 0)
173 return NULL;
174
175 if (write) {
176 struct rlimit *rlim = current->signal->rlim;
177 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
178
179 /*
180 * Limit to 1/4-th the stack size for the argv+env strings.
181 * This ensures that:
182 * - the remaining binfmt code will not run out of stack space,
183 * - the program will have a reasonable amount of stack left
184 * to work from.
185 */
186 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
187 put_page(page);
188 return NULL;
189 }
190 }
191
192 return page;
193 }
194
195 static void put_arg_page(struct page *page)
196 {
197 put_page(page);
198 }
199
200 static void free_arg_page(struct linux_binprm *bprm, int i)
201 {
202 }
203
204 static void free_arg_pages(struct linux_binprm *bprm)
205 {
206 }
207
208 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
209 struct page *page)
210 {
211 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
212 }
213
214 static int __bprm_mm_init(struct linux_binprm *bprm)
215 {
216 int err = -ENOMEM;
217 struct vm_area_struct *vma = NULL;
218 struct mm_struct *mm = bprm->mm;
219
220 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
221 if (!vma)
222 goto err;
223
224 down_write(&mm->mmap_sem);
225 vma->vm_mm = mm;
226
227 /*
228 * Place the stack at the largest stack address the architecture
229 * supports. Later, we'll move this to an appropriate place. We don't
230 * use STACK_TOP because that can depend on attributes which aren't
231 * configured yet.
232 */
233 vma->vm_end = STACK_TOP_MAX;
234 vma->vm_start = vma->vm_end - PAGE_SIZE;
235
236 vma->vm_flags = VM_STACK_FLAGS;
237 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
238 err = insert_vm_struct(mm, vma);
239 if (err) {
240 up_write(&mm->mmap_sem);
241 goto err;
242 }
243
244 mm->stack_vm = mm->total_vm = 1;
245 up_write(&mm->mmap_sem);
246
247 bprm->p = vma->vm_end - sizeof(void *);
248
249 return 0;
250
251 err:
252 if (vma) {
253 bprm->vma = NULL;
254 kmem_cache_free(vm_area_cachep, vma);
255 }
256
257 return err;
258 }
259
260 static bool valid_arg_len(struct linux_binprm *bprm, long len)
261 {
262 return len <= MAX_ARG_STRLEN;
263 }
264
265 #else
266
267 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
268 int write)
269 {
270 struct page *page;
271
272 page = bprm->page[pos / PAGE_SIZE];
273 if (!page && write) {
274 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
275 if (!page)
276 return NULL;
277 bprm->page[pos / PAGE_SIZE] = page;
278 }
279
280 return page;
281 }
282
283 static void put_arg_page(struct page *page)
284 {
285 }
286
287 static void free_arg_page(struct linux_binprm *bprm, int i)
288 {
289 if (bprm->page[i]) {
290 __free_page(bprm->page[i]);
291 bprm->page[i] = NULL;
292 }
293 }
294
295 static void free_arg_pages(struct linux_binprm *bprm)
296 {
297 int i;
298
299 for (i = 0; i < MAX_ARG_PAGES; i++)
300 free_arg_page(bprm, i);
301 }
302
303 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
304 struct page *page)
305 {
306 }
307
308 static int __bprm_mm_init(struct linux_binprm *bprm)
309 {
310 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
311 return 0;
312 }
313
314 static bool valid_arg_len(struct linux_binprm *bprm, long len)
315 {
316 return len <= bprm->p;
317 }
318
319 #endif /* CONFIG_MMU */
320
321 /*
322 * Create a new mm_struct and populate it with a temporary stack
323 * vm_area_struct. We don't have enough context at this point to set the stack
324 * flags, permissions, and offset, so we use temporary values. We'll update
325 * them later in setup_arg_pages().
326 */
327 int bprm_mm_init(struct linux_binprm *bprm)
328 {
329 int err;
330 struct mm_struct *mm = NULL;
331
332 bprm->mm = mm = mm_alloc();
333 err = -ENOMEM;
334 if (!mm)
335 goto err;
336
337 err = init_new_context(current, mm);
338 if (err)
339 goto err;
340
341 err = __bprm_mm_init(bprm);
342 if (err)
343 goto err;
344
345 return 0;
346
347 err:
348 if (mm) {
349 bprm->mm = NULL;
350 mmdrop(mm);
351 }
352
353 return err;
354 }
355
356 /*
357 * count() counts the number of strings in array ARGV.
358 */
359 static int count(char __user * __user * argv, int max)
360 {
361 int i = 0;
362
363 if (argv != NULL) {
364 for (;;) {
365 char __user * p;
366
367 if (get_user(p, argv))
368 return -EFAULT;
369 if (!p)
370 break;
371 argv++;
372 if(++i > max)
373 return -E2BIG;
374 cond_resched();
375 }
376 }
377 return i;
378 }
379
380 /*
381 * 'copy_strings()' copies argument/environment strings from the old
382 * processes's memory to the new process's stack. The call to get_user_pages()
383 * ensures the destination page is created and not swapped out.
384 */
385 static int copy_strings(int argc, char __user * __user * argv,
386 struct linux_binprm *bprm)
387 {
388 struct page *kmapped_page = NULL;
389 char *kaddr = NULL;
390 unsigned long kpos = 0;
391 int ret;
392
393 while (argc-- > 0) {
394 char __user *str;
395 int len;
396 unsigned long pos;
397
398 if (get_user(str, argv+argc) ||
399 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
400 ret = -EFAULT;
401 goto out;
402 }
403
404 if (!valid_arg_len(bprm, len)) {
405 ret = -E2BIG;
406 goto out;
407 }
408
409 /* We're going to work our way backwords. */
410 pos = bprm->p;
411 str += len;
412 bprm->p -= len;
413
414 while (len > 0) {
415 int offset, bytes_to_copy;
416
417 offset = pos % PAGE_SIZE;
418 if (offset == 0)
419 offset = PAGE_SIZE;
420
421 bytes_to_copy = offset;
422 if (bytes_to_copy > len)
423 bytes_to_copy = len;
424
425 offset -= bytes_to_copy;
426 pos -= bytes_to_copy;
427 str -= bytes_to_copy;
428 len -= bytes_to_copy;
429
430 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
431 struct page *page;
432
433 page = get_arg_page(bprm, pos, 1);
434 if (!page) {
435 ret = -E2BIG;
436 goto out;
437 }
438
439 if (kmapped_page) {
440 flush_kernel_dcache_page(kmapped_page);
441 kunmap(kmapped_page);
442 put_arg_page(kmapped_page);
443 }
444 kmapped_page = page;
445 kaddr = kmap(kmapped_page);
446 kpos = pos & PAGE_MASK;
447 flush_arg_page(bprm, kpos, kmapped_page);
448 }
449 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
450 ret = -EFAULT;
451 goto out;
452 }
453 }
454 }
455 ret = 0;
456 out:
457 if (kmapped_page) {
458 flush_kernel_dcache_page(kmapped_page);
459 kunmap(kmapped_page);
460 put_arg_page(kmapped_page);
461 }
462 return ret;
463 }
464
465 /*
466 * Like copy_strings, but get argv and its values from kernel memory.
467 */
468 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
469 {
470 int r;
471 mm_segment_t oldfs = get_fs();
472 set_fs(KERNEL_DS);
473 r = copy_strings(argc, (char __user * __user *)argv, bprm);
474 set_fs(oldfs);
475 return r;
476 }
477 EXPORT_SYMBOL(copy_strings_kernel);
478
479 #ifdef CONFIG_MMU
480
481 /*
482 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
483 * the binfmt code determines where the new stack should reside, we shift it to
484 * its final location. The process proceeds as follows:
485 *
486 * 1) Use shift to calculate the new vma endpoints.
487 * 2) Extend vma to cover both the old and new ranges. This ensures the
488 * arguments passed to subsequent functions are consistent.
489 * 3) Move vma's page tables to the new range.
490 * 4) Free up any cleared pgd range.
491 * 5) Shrink the vma to cover only the new range.
492 */
493 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
494 {
495 struct mm_struct *mm = vma->vm_mm;
496 unsigned long old_start = vma->vm_start;
497 unsigned long old_end = vma->vm_end;
498 unsigned long length = old_end - old_start;
499 unsigned long new_start = old_start - shift;
500 unsigned long new_end = old_end - shift;
501 struct mmu_gather *tlb;
502
503 BUG_ON(new_start > new_end);
504
505 /*
506 * ensure there are no vmas between where we want to go
507 * and where we are
508 */
509 if (vma != find_vma(mm, new_start))
510 return -EFAULT;
511
512 /*
513 * cover the whole range: [new_start, old_end)
514 */
515 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
516
517 /*
518 * move the page tables downwards, on failure we rely on
519 * process cleanup to remove whatever mess we made.
520 */
521 if (length != move_page_tables(vma, old_start,
522 vma, new_start, length))
523 return -ENOMEM;
524
525 lru_add_drain();
526 tlb = tlb_gather_mmu(mm, 0);
527 if (new_end > old_start) {
528 /*
529 * when the old and new regions overlap clear from new_end.
530 */
531 free_pgd_range(&tlb, new_end, old_end, new_end,
532 vma->vm_next ? vma->vm_next->vm_start : 0);
533 } else {
534 /*
535 * otherwise, clean from old_start; this is done to not touch
536 * the address space in [new_end, old_start) some architectures
537 * have constraints on va-space that make this illegal (IA64) -
538 * for the others its just a little faster.
539 */
540 free_pgd_range(&tlb, old_start, old_end, new_end,
541 vma->vm_next ? vma->vm_next->vm_start : 0);
542 }
543 tlb_finish_mmu(tlb, new_end, old_end);
544
545 /*
546 * shrink the vma to just the new range.
547 */
548 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
549
550 return 0;
551 }
552
553 #define EXTRA_STACK_VM_PAGES 20 /* random */
554
555 /*
556 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
557 * the stack is optionally relocated, and some extra space is added.
558 */
559 int setup_arg_pages(struct linux_binprm *bprm,
560 unsigned long stack_top,
561 int executable_stack)
562 {
563 unsigned long ret;
564 unsigned long stack_shift;
565 struct mm_struct *mm = current->mm;
566 struct vm_area_struct *vma = bprm->vma;
567 struct vm_area_struct *prev = NULL;
568 unsigned long vm_flags;
569 unsigned long stack_base;
570
571 #ifdef CONFIG_STACK_GROWSUP
572 /* Limit stack size to 1GB */
573 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
574 if (stack_base > (1 << 30))
575 stack_base = 1 << 30;
576
577 /* Make sure we didn't let the argument array grow too large. */
578 if (vma->vm_end - vma->vm_start > stack_base)
579 return -ENOMEM;
580
581 stack_base = PAGE_ALIGN(stack_top - stack_base);
582
583 stack_shift = vma->vm_start - stack_base;
584 mm->arg_start = bprm->p - stack_shift;
585 bprm->p = vma->vm_end - stack_shift;
586 #else
587 stack_top = arch_align_stack(stack_top);
588 stack_top = PAGE_ALIGN(stack_top);
589 stack_shift = vma->vm_end - stack_top;
590
591 bprm->p -= stack_shift;
592 mm->arg_start = bprm->p;
593 #endif
594
595 if (bprm->loader)
596 bprm->loader -= stack_shift;
597 bprm->exec -= stack_shift;
598
599 down_write(&mm->mmap_sem);
600 vm_flags = vma->vm_flags;
601
602 /*
603 * Adjust stack execute permissions; explicitly enable for
604 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
605 * (arch default) otherwise.
606 */
607 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
608 vm_flags |= VM_EXEC;
609 else if (executable_stack == EXSTACK_DISABLE_X)
610 vm_flags &= ~VM_EXEC;
611 vm_flags |= mm->def_flags;
612
613 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
614 vm_flags);
615 if (ret)
616 goto out_unlock;
617 BUG_ON(prev != vma);
618
619 /* Move stack pages down in memory. */
620 if (stack_shift) {
621 ret = shift_arg_pages(vma, stack_shift);
622 if (ret) {
623 up_write(&mm->mmap_sem);
624 return ret;
625 }
626 }
627
628 #ifdef CONFIG_STACK_GROWSUP
629 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
630 #else
631 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
632 #endif
633 ret = expand_stack(vma, stack_base);
634 if (ret)
635 ret = -EFAULT;
636
637 out_unlock:
638 up_write(&mm->mmap_sem);
639 return 0;
640 }
641 EXPORT_SYMBOL(setup_arg_pages);
642
643 #endif /* CONFIG_MMU */
644
645 struct file *open_exec(const char *name)
646 {
647 struct nameidata nd;
648 int err;
649 struct file *file;
650
651 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
652 file = ERR_PTR(err);
653
654 if (!err) {
655 struct inode *inode = nd.dentry->d_inode;
656 file = ERR_PTR(-EACCES);
657 if (S_ISREG(inode->i_mode)) {
658 int err = vfs_permission(&nd, MAY_EXEC);
659 file = ERR_PTR(err);
660 if (!err) {
661 file = nameidata_to_filp(&nd, O_RDONLY);
662 if (!IS_ERR(file)) {
663 err = deny_write_access(file);
664 if (err) {
665 fput(file);
666 file = ERR_PTR(err);
667 }
668 }
669 out:
670 return file;
671 }
672 }
673 release_open_intent(&nd);
674 path_release(&nd);
675 }
676 goto out;
677 }
678
679 EXPORT_SYMBOL(open_exec);
680
681 int kernel_read(struct file *file, unsigned long offset,
682 char *addr, unsigned long count)
683 {
684 mm_segment_t old_fs;
685 loff_t pos = offset;
686 int result;
687
688 old_fs = get_fs();
689 set_fs(get_ds());
690 /* The cast to a user pointer is valid due to the set_fs() */
691 result = vfs_read(file, (void __user *)addr, count, &pos);
692 set_fs(old_fs);
693 return result;
694 }
695
696 EXPORT_SYMBOL(kernel_read);
697
698 static int exec_mmap(struct mm_struct *mm)
699 {
700 struct task_struct *tsk;
701 struct mm_struct * old_mm, *active_mm;
702
703 /* Notify parent that we're no longer interested in the old VM */
704 tsk = current;
705 old_mm = current->mm;
706 mm_release(tsk, old_mm);
707
708 if (old_mm) {
709 /*
710 * Make sure that if there is a core dump in progress
711 * for the old mm, we get out and die instead of going
712 * through with the exec. We must hold mmap_sem around
713 * checking core_waiters and changing tsk->mm. The
714 * core-inducing thread will increment core_waiters for
715 * each thread whose ->mm == old_mm.
716 */
717 down_read(&old_mm->mmap_sem);
718 if (unlikely(old_mm->core_waiters)) {
719 up_read(&old_mm->mmap_sem);
720 return -EINTR;
721 }
722 }
723 task_lock(tsk);
724 active_mm = tsk->active_mm;
725 tsk->mm = mm;
726 tsk->active_mm = mm;
727 activate_mm(active_mm, mm);
728 task_unlock(tsk);
729 arch_pick_mmap_layout(mm);
730 if (old_mm) {
731 up_read(&old_mm->mmap_sem);
732 BUG_ON(active_mm != old_mm);
733 mmput(old_mm);
734 return 0;
735 }
736 mmdrop(active_mm);
737 return 0;
738 }
739
740 /*
741 * This function makes sure the current process has its own signal table,
742 * so that flush_signal_handlers can later reset the handlers without
743 * disturbing other processes. (Other processes might share the signal
744 * table via the CLONE_SIGHAND option to clone().)
745 */
746 static int de_thread(struct task_struct *tsk)
747 {
748 struct signal_struct *sig = tsk->signal;
749 struct sighand_struct *oldsighand = tsk->sighand;
750 spinlock_t *lock = &oldsighand->siglock;
751 struct task_struct *leader = NULL;
752 int count;
753
754 if (thread_group_empty(tsk))
755 goto no_thread_group;
756
757 /*
758 * Kill all other threads in the thread group.
759 * We must hold tasklist_lock to call zap_other_threads.
760 */
761 read_lock(&tasklist_lock);
762 spin_lock_irq(lock);
763 if (signal_group_exit(sig)) {
764 /*
765 * Another group action in progress, just
766 * return so that the signal is processed.
767 */
768 spin_unlock_irq(lock);
769 read_unlock(&tasklist_lock);
770 return -EAGAIN;
771 }
772
773 /*
774 * child_reaper ignores SIGKILL, change it now.
775 * Reparenting needs write_lock on tasklist_lock,
776 * so it is safe to do it under read_lock.
777 */
778 if (unlikely(tsk->group_leader == task_child_reaper(tsk)))
779 task_active_pid_ns(tsk)->child_reaper = tsk;
780
781 sig->group_exit_task = tsk;
782 zap_other_threads(tsk);
783 read_unlock(&tasklist_lock);
784
785 /*
786 * Account for the thread group leader hanging around:
787 */
788 count = 1;
789 if (!thread_group_leader(tsk)) {
790 count = 2;
791 /*
792 * The SIGALRM timer survives the exec, but needs to point
793 * at us as the new group leader now. We have a race with
794 * a timer firing now getting the old leader, so we need to
795 * synchronize with any firing (by calling del_timer_sync)
796 * before we can safely let the old group leader die.
797 */
798 sig->tsk = tsk;
799 spin_unlock_irq(lock);
800 if (hrtimer_cancel(&sig->real_timer))
801 hrtimer_restart(&sig->real_timer);
802 spin_lock_irq(lock);
803 }
804
805 sig->notify_count = count;
806 while (atomic_read(&sig->count) > count) {
807 __set_current_state(TASK_UNINTERRUPTIBLE);
808 spin_unlock_irq(lock);
809 schedule();
810 spin_lock_irq(lock);
811 }
812 spin_unlock_irq(lock);
813
814 /*
815 * At this point all other threads have exited, all we have to
816 * do is to wait for the thread group leader to become inactive,
817 * and to assume its PID:
818 */
819 if (!thread_group_leader(tsk)) {
820 leader = tsk->group_leader;
821
822 sig->notify_count = -1;
823 for (;;) {
824 write_lock_irq(&tasklist_lock);
825 if (likely(leader->exit_state))
826 break;
827 __set_current_state(TASK_UNINTERRUPTIBLE);
828 write_unlock_irq(&tasklist_lock);
829 schedule();
830 }
831
832 /*
833 * The only record we have of the real-time age of a
834 * process, regardless of execs it's done, is start_time.
835 * All the past CPU time is accumulated in signal_struct
836 * from sister threads now dead. But in this non-leader
837 * exec, nothing survives from the original leader thread,
838 * whose birth marks the true age of this process now.
839 * When we take on its identity by switching to its PID, we
840 * also take its birthdate (always earlier than our own).
841 */
842 tsk->start_time = leader->start_time;
843
844 BUG_ON(!same_thread_group(leader, tsk));
845 BUG_ON(has_group_leader_pid(tsk));
846 /*
847 * An exec() starts a new thread group with the
848 * TGID of the previous thread group. Rehash the
849 * two threads with a switched PID, and release
850 * the former thread group leader:
851 */
852
853 /* Become a process group leader with the old leader's pid.
854 * The old leader becomes a thread of the this thread group.
855 * Note: The old leader also uses this pid until release_task
856 * is called. Odd but simple and correct.
857 */
858 detach_pid(tsk, PIDTYPE_PID);
859 tsk->pid = leader->pid;
860 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
861 transfer_pid(leader, tsk, PIDTYPE_PGID);
862 transfer_pid(leader, tsk, PIDTYPE_SID);
863 list_replace_rcu(&leader->tasks, &tsk->tasks);
864
865 tsk->group_leader = tsk;
866 leader->group_leader = tsk;
867
868 tsk->exit_signal = SIGCHLD;
869
870 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
871 leader->exit_state = EXIT_DEAD;
872
873 write_unlock_irq(&tasklist_lock);
874 }
875
876 sig->group_exit_task = NULL;
877 sig->notify_count = 0;
878
879 no_thread_group:
880 exit_itimers(sig);
881 if (leader)
882 release_task(leader);
883
884 if (atomic_read(&oldsighand->count) != 1) {
885 struct sighand_struct *newsighand;
886 /*
887 * This ->sighand is shared with the CLONE_SIGHAND
888 * but not CLONE_THREAD task, switch to the new one.
889 */
890 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
891 if (!newsighand)
892 return -ENOMEM;
893
894 atomic_set(&newsighand->count, 1);
895 memcpy(newsighand->action, oldsighand->action,
896 sizeof(newsighand->action));
897
898 write_lock_irq(&tasklist_lock);
899 spin_lock(&oldsighand->siglock);
900 rcu_assign_pointer(tsk->sighand, newsighand);
901 spin_unlock(&oldsighand->siglock);
902 write_unlock_irq(&tasklist_lock);
903
904 __cleanup_sighand(oldsighand);
905 }
906
907 BUG_ON(!thread_group_leader(tsk));
908 return 0;
909 }
910
911 /*
912 * These functions flushes out all traces of the currently running executable
913 * so that a new one can be started
914 */
915 static void flush_old_files(struct files_struct * files)
916 {
917 long j = -1;
918 struct fdtable *fdt;
919
920 spin_lock(&files->file_lock);
921 for (;;) {
922 unsigned long set, i;
923
924 j++;
925 i = j * __NFDBITS;
926 fdt = files_fdtable(files);
927 if (i >= fdt->max_fds)
928 break;
929 set = fdt->close_on_exec->fds_bits[j];
930 if (!set)
931 continue;
932 fdt->close_on_exec->fds_bits[j] = 0;
933 spin_unlock(&files->file_lock);
934 for ( ; set ; i++,set >>= 1) {
935 if (set & 1) {
936 sys_close(i);
937 }
938 }
939 spin_lock(&files->file_lock);
940
941 }
942 spin_unlock(&files->file_lock);
943 }
944
945 char *get_task_comm(char *buf, struct task_struct *tsk)
946 {
947 /* buf must be at least sizeof(tsk->comm) in size */
948 task_lock(tsk);
949 strncpy(buf, tsk->comm, sizeof(tsk->comm));
950 task_unlock(tsk);
951 return buf;
952 }
953
954 void set_task_comm(struct task_struct *tsk, char *buf)
955 {
956 task_lock(tsk);
957 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
958 task_unlock(tsk);
959 }
960
961 int flush_old_exec(struct linux_binprm * bprm)
962 {
963 char * name;
964 int i, ch, retval;
965 struct files_struct *files;
966 char tcomm[sizeof(current->comm)];
967
968 /*
969 * Make sure we have a private signal table and that
970 * we are unassociated from the previous thread group.
971 */
972 retval = de_thread(current);
973 if (retval)
974 goto out;
975
976 /*
977 * Make sure we have private file handles. Ask the
978 * fork helper to do the work for us and the exit
979 * helper to do the cleanup of the old one.
980 */
981 files = current->files; /* refcounted so safe to hold */
982 retval = unshare_files();
983 if (retval)
984 goto out;
985 /*
986 * Release all of the old mmap stuff
987 */
988 retval = exec_mmap(bprm->mm);
989 if (retval)
990 goto mmap_failed;
991
992 bprm->mm = NULL; /* We're using it now */
993
994 /* This is the point of no return */
995 put_files_struct(files);
996
997 current->sas_ss_sp = current->sas_ss_size = 0;
998
999 if (current->euid == current->uid && current->egid == current->gid)
1000 set_dumpable(current->mm, 1);
1001 else
1002 set_dumpable(current->mm, suid_dumpable);
1003
1004 name = bprm->filename;
1005
1006 /* Copies the binary name from after last slash */
1007 for (i=0; (ch = *(name++)) != '\0';) {
1008 if (ch == '/')
1009 i = 0; /* overwrite what we wrote */
1010 else
1011 if (i < (sizeof(tcomm) - 1))
1012 tcomm[i++] = ch;
1013 }
1014 tcomm[i] = '\0';
1015 set_task_comm(current, tcomm);
1016
1017 current->flags &= ~PF_RANDOMIZE;
1018 flush_thread();
1019
1020 /* Set the new mm task size. We have to do that late because it may
1021 * depend on TIF_32BIT which is only updated in flush_thread() on
1022 * some architectures like powerpc
1023 */
1024 current->mm->task_size = TASK_SIZE;
1025
1026 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1027 suid_keys(current);
1028 set_dumpable(current->mm, suid_dumpable);
1029 current->pdeath_signal = 0;
1030 } else if (file_permission(bprm->file, MAY_READ) ||
1031 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1032 suid_keys(current);
1033 set_dumpable(current->mm, suid_dumpable);
1034 }
1035
1036 /* An exec changes our domain. We are no longer part of the thread
1037 group */
1038
1039 current->self_exec_id++;
1040
1041 flush_signal_handlers(current, 0);
1042 flush_old_files(current->files);
1043
1044 return 0;
1045
1046 mmap_failed:
1047 reset_files_struct(current, files);
1048 out:
1049 return retval;
1050 }
1051
1052 EXPORT_SYMBOL(flush_old_exec);
1053
1054 /*
1055 * Fill the binprm structure from the inode.
1056 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1057 */
1058 int prepare_binprm(struct linux_binprm *bprm)
1059 {
1060 int mode;
1061 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1062 int retval;
1063
1064 mode = inode->i_mode;
1065 if (bprm->file->f_op == NULL)
1066 return -EACCES;
1067
1068 bprm->e_uid = current->euid;
1069 bprm->e_gid = current->egid;
1070
1071 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1072 /* Set-uid? */
1073 if (mode & S_ISUID) {
1074 current->personality &= ~PER_CLEAR_ON_SETID;
1075 bprm->e_uid = inode->i_uid;
1076 }
1077
1078 /* Set-gid? */
1079 /*
1080 * If setgid is set but no group execute bit then this
1081 * is a candidate for mandatory locking, not a setgid
1082 * executable.
1083 */
1084 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1085 current->personality &= ~PER_CLEAR_ON_SETID;
1086 bprm->e_gid = inode->i_gid;
1087 }
1088 }
1089
1090 /* fill in binprm security blob */
1091 retval = security_bprm_set(bprm);
1092 if (retval)
1093 return retval;
1094
1095 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1096 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1097 }
1098
1099 EXPORT_SYMBOL(prepare_binprm);
1100
1101 static int unsafe_exec(struct task_struct *p)
1102 {
1103 int unsafe = 0;
1104 if (p->ptrace & PT_PTRACED) {
1105 if (p->ptrace & PT_PTRACE_CAP)
1106 unsafe |= LSM_UNSAFE_PTRACE_CAP;
1107 else
1108 unsafe |= LSM_UNSAFE_PTRACE;
1109 }
1110 if (atomic_read(&p->fs->count) > 1 ||
1111 atomic_read(&p->files->count) > 1 ||
1112 atomic_read(&p->sighand->count) > 1)
1113 unsafe |= LSM_UNSAFE_SHARE;
1114
1115 return unsafe;
1116 }
1117
1118 void compute_creds(struct linux_binprm *bprm)
1119 {
1120 int unsafe;
1121
1122 if (bprm->e_uid != current->uid) {
1123 suid_keys(current);
1124 current->pdeath_signal = 0;
1125 }
1126 exec_keys(current);
1127
1128 task_lock(current);
1129 unsafe = unsafe_exec(current);
1130 security_bprm_apply_creds(bprm, unsafe);
1131 task_unlock(current);
1132 security_bprm_post_apply_creds(bprm);
1133 }
1134 EXPORT_SYMBOL(compute_creds);
1135
1136 /*
1137 * Arguments are '\0' separated strings found at the location bprm->p
1138 * points to; chop off the first by relocating brpm->p to right after
1139 * the first '\0' encountered.
1140 */
1141 int remove_arg_zero(struct linux_binprm *bprm)
1142 {
1143 int ret = 0;
1144 unsigned long offset;
1145 char *kaddr;
1146 struct page *page;
1147
1148 if (!bprm->argc)
1149 return 0;
1150
1151 do {
1152 offset = bprm->p & ~PAGE_MASK;
1153 page = get_arg_page(bprm, bprm->p, 0);
1154 if (!page) {
1155 ret = -EFAULT;
1156 goto out;
1157 }
1158 kaddr = kmap_atomic(page, KM_USER0);
1159
1160 for (; offset < PAGE_SIZE && kaddr[offset];
1161 offset++, bprm->p++)
1162 ;
1163
1164 kunmap_atomic(kaddr, KM_USER0);
1165 put_arg_page(page);
1166
1167 if (offset == PAGE_SIZE)
1168 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1169 } while (offset == PAGE_SIZE);
1170
1171 bprm->p++;
1172 bprm->argc--;
1173 ret = 0;
1174
1175 out:
1176 return ret;
1177 }
1178 EXPORT_SYMBOL(remove_arg_zero);
1179
1180 /*
1181 * cycle the list of binary formats handler, until one recognizes the image
1182 */
1183 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1184 {
1185 int try,retval;
1186 struct linux_binfmt *fmt;
1187 #ifdef __alpha__
1188 /* handle /sbin/loader.. */
1189 {
1190 struct exec * eh = (struct exec *) bprm->buf;
1191
1192 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1193 (eh->fh.f_flags & 0x3000) == 0x3000)
1194 {
1195 struct file * file;
1196 unsigned long loader;
1197
1198 allow_write_access(bprm->file);
1199 fput(bprm->file);
1200 bprm->file = NULL;
1201
1202 loader = bprm->vma->vm_end - sizeof(void *);
1203
1204 file = open_exec("/sbin/loader");
1205 retval = PTR_ERR(file);
1206 if (IS_ERR(file))
1207 return retval;
1208
1209 /* Remember if the application is TASO. */
1210 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1211
1212 bprm->file = file;
1213 bprm->loader = loader;
1214 retval = prepare_binprm(bprm);
1215 if (retval<0)
1216 return retval;
1217 /* should call search_binary_handler recursively here,
1218 but it does not matter */
1219 }
1220 }
1221 #endif
1222 retval = security_bprm_check(bprm);
1223 if (retval)
1224 return retval;
1225
1226 /* kernel module loader fixup */
1227 /* so we don't try to load run modprobe in kernel space. */
1228 set_fs(USER_DS);
1229
1230 retval = audit_bprm(bprm);
1231 if (retval)
1232 return retval;
1233
1234 retval = -ENOENT;
1235 for (try=0; try<2; try++) {
1236 read_lock(&binfmt_lock);
1237 list_for_each_entry(fmt, &formats, lh) {
1238 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1239 if (!fn)
1240 continue;
1241 if (!try_module_get(fmt->module))
1242 continue;
1243 read_unlock(&binfmt_lock);
1244 retval = fn(bprm, regs);
1245 if (retval >= 0) {
1246 put_binfmt(fmt);
1247 allow_write_access(bprm->file);
1248 if (bprm->file)
1249 fput(bprm->file);
1250 bprm->file = NULL;
1251 current->did_exec = 1;
1252 proc_exec_connector(current);
1253 return retval;
1254 }
1255 read_lock(&binfmt_lock);
1256 put_binfmt(fmt);
1257 if (retval != -ENOEXEC || bprm->mm == NULL)
1258 break;
1259 if (!bprm->file) {
1260 read_unlock(&binfmt_lock);
1261 return retval;
1262 }
1263 }
1264 read_unlock(&binfmt_lock);
1265 if (retval != -ENOEXEC || bprm->mm == NULL) {
1266 break;
1267 #ifdef CONFIG_KMOD
1268 }else{
1269 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1270 if (printable(bprm->buf[0]) &&
1271 printable(bprm->buf[1]) &&
1272 printable(bprm->buf[2]) &&
1273 printable(bprm->buf[3]))
1274 break; /* -ENOEXEC */
1275 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1276 #endif
1277 }
1278 }
1279 return retval;
1280 }
1281
1282 EXPORT_SYMBOL(search_binary_handler);
1283
1284 /*
1285 * sys_execve() executes a new program.
1286 */
1287 int do_execve(char * filename,
1288 char __user *__user *argv,
1289 char __user *__user *envp,
1290 struct pt_regs * regs)
1291 {
1292 struct linux_binprm *bprm;
1293 struct file *file;
1294 unsigned long env_p;
1295 int retval;
1296
1297 retval = -ENOMEM;
1298 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1299 if (!bprm)
1300 goto out_ret;
1301
1302 file = open_exec(filename);
1303 retval = PTR_ERR(file);
1304 if (IS_ERR(file))
1305 goto out_kfree;
1306
1307 sched_exec();
1308
1309 bprm->file = file;
1310 bprm->filename = filename;
1311 bprm->interp = filename;
1312
1313 retval = bprm_mm_init(bprm);
1314 if (retval)
1315 goto out_file;
1316
1317 bprm->argc = count(argv, MAX_ARG_STRINGS);
1318 if ((retval = bprm->argc) < 0)
1319 goto out_mm;
1320
1321 bprm->envc = count(envp, MAX_ARG_STRINGS);
1322 if ((retval = bprm->envc) < 0)
1323 goto out_mm;
1324
1325 retval = security_bprm_alloc(bprm);
1326 if (retval)
1327 goto out;
1328
1329 retval = prepare_binprm(bprm);
1330 if (retval < 0)
1331 goto out;
1332
1333 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1334 if (retval < 0)
1335 goto out;
1336
1337 bprm->exec = bprm->p;
1338 retval = copy_strings(bprm->envc, envp, bprm);
1339 if (retval < 0)
1340 goto out;
1341
1342 env_p = bprm->p;
1343 retval = copy_strings(bprm->argc, argv, bprm);
1344 if (retval < 0)
1345 goto out;
1346 bprm->argv_len = env_p - bprm->p;
1347
1348 retval = search_binary_handler(bprm,regs);
1349 if (retval >= 0) {
1350 /* execve success */
1351 free_arg_pages(bprm);
1352 security_bprm_free(bprm);
1353 acct_update_integrals(current);
1354 kfree(bprm);
1355 return retval;
1356 }
1357
1358 out:
1359 free_arg_pages(bprm);
1360 if (bprm->security)
1361 security_bprm_free(bprm);
1362
1363 out_mm:
1364 if (bprm->mm)
1365 mmput (bprm->mm);
1366
1367 out_file:
1368 if (bprm->file) {
1369 allow_write_access(bprm->file);
1370 fput(bprm->file);
1371 }
1372 out_kfree:
1373 kfree(bprm);
1374
1375 out_ret:
1376 return retval;
1377 }
1378
1379 int set_binfmt(struct linux_binfmt *new)
1380 {
1381 struct linux_binfmt *old = current->binfmt;
1382
1383 if (new) {
1384 if (!try_module_get(new->module))
1385 return -1;
1386 }
1387 current->binfmt = new;
1388 if (old)
1389 module_put(old->module);
1390 return 0;
1391 }
1392
1393 EXPORT_SYMBOL(set_binfmt);
1394
1395 /* format_corename will inspect the pattern parameter, and output a
1396 * name into corename, which must have space for at least
1397 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1398 */
1399 static int format_corename(char *corename, const char *pattern, long signr)
1400 {
1401 const char *pat_ptr = pattern;
1402 char *out_ptr = corename;
1403 char *const out_end = corename + CORENAME_MAX_SIZE;
1404 int rc;
1405 int pid_in_pattern = 0;
1406 int ispipe = 0;
1407
1408 if (*pattern == '|')
1409 ispipe = 1;
1410
1411 /* Repeat as long as we have more pattern to process and more output
1412 space */
1413 while (*pat_ptr) {
1414 if (*pat_ptr != '%') {
1415 if (out_ptr == out_end)
1416 goto out;
1417 *out_ptr++ = *pat_ptr++;
1418 } else {
1419 switch (*++pat_ptr) {
1420 case 0:
1421 goto out;
1422 /* Double percent, output one percent */
1423 case '%':
1424 if (out_ptr == out_end)
1425 goto out;
1426 *out_ptr++ = '%';
1427 break;
1428 /* pid */
1429 case 'p':
1430 pid_in_pattern = 1;
1431 rc = snprintf(out_ptr, out_end - out_ptr,
1432 "%d", task_tgid_vnr(current));
1433 if (rc > out_end - out_ptr)
1434 goto out;
1435 out_ptr += rc;
1436 break;
1437 /* uid */
1438 case 'u':
1439 rc = snprintf(out_ptr, out_end - out_ptr,
1440 "%d", current->uid);
1441 if (rc > out_end - out_ptr)
1442 goto out;
1443 out_ptr += rc;
1444 break;
1445 /* gid */
1446 case 'g':
1447 rc = snprintf(out_ptr, out_end - out_ptr,
1448 "%d", current->gid);
1449 if (rc > out_end - out_ptr)
1450 goto out;
1451 out_ptr += rc;
1452 break;
1453 /* signal that caused the coredump */
1454 case 's':
1455 rc = snprintf(out_ptr, out_end - out_ptr,
1456 "%ld", signr);
1457 if (rc > out_end - out_ptr)
1458 goto out;
1459 out_ptr += rc;
1460 break;
1461 /* UNIX time of coredump */
1462 case 't': {
1463 struct timeval tv;
1464 do_gettimeofday(&tv);
1465 rc = snprintf(out_ptr, out_end - out_ptr,
1466 "%lu", tv.tv_sec);
1467 if (rc > out_end - out_ptr)
1468 goto out;
1469 out_ptr += rc;
1470 break;
1471 }
1472 /* hostname */
1473 case 'h':
1474 down_read(&uts_sem);
1475 rc = snprintf(out_ptr, out_end - out_ptr,
1476 "%s", utsname()->nodename);
1477 up_read(&uts_sem);
1478 if (rc > out_end - out_ptr)
1479 goto out;
1480 out_ptr += rc;
1481 break;
1482 /* executable */
1483 case 'e':
1484 rc = snprintf(out_ptr, out_end - out_ptr,
1485 "%s", current->comm);
1486 if (rc > out_end - out_ptr)
1487 goto out;
1488 out_ptr += rc;
1489 break;
1490 /* core limit size */
1491 case 'c':
1492 rc = snprintf(out_ptr, out_end - out_ptr,
1493 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1494 if (rc > out_end - out_ptr)
1495 goto out;
1496 out_ptr += rc;
1497 break;
1498 default:
1499 break;
1500 }
1501 ++pat_ptr;
1502 }
1503 }
1504 /* Backward compatibility with core_uses_pid:
1505 *
1506 * If core_pattern does not include a %p (as is the default)
1507 * and core_uses_pid is set, then .%pid will be appended to
1508 * the filename. Do not do this for piped commands. */
1509 if (!ispipe && !pid_in_pattern
1510 && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1511 rc = snprintf(out_ptr, out_end - out_ptr,
1512 ".%d", task_tgid_vnr(current));
1513 if (rc > out_end - out_ptr)
1514 goto out;
1515 out_ptr += rc;
1516 }
1517 out:
1518 *out_ptr = 0;
1519 return ispipe;
1520 }
1521
1522 static void zap_process(struct task_struct *start)
1523 {
1524 struct task_struct *t;
1525
1526 start->signal->flags = SIGNAL_GROUP_EXIT;
1527 start->signal->group_stop_count = 0;
1528
1529 t = start;
1530 do {
1531 if (t != current && t->mm) {
1532 t->mm->core_waiters++;
1533 sigaddset(&t->pending.signal, SIGKILL);
1534 signal_wake_up(t, 1);
1535 }
1536 } while ((t = next_thread(t)) != start);
1537 }
1538
1539 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1540 int exit_code)
1541 {
1542 struct task_struct *g, *p;
1543 unsigned long flags;
1544 int err = -EAGAIN;
1545
1546 spin_lock_irq(&tsk->sighand->siglock);
1547 if (!signal_group_exit(tsk->signal)) {
1548 tsk->signal->group_exit_code = exit_code;
1549 zap_process(tsk);
1550 err = 0;
1551 }
1552 spin_unlock_irq(&tsk->sighand->siglock);
1553 if (err)
1554 return err;
1555
1556 if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1557 goto done;
1558
1559 rcu_read_lock();
1560 for_each_process(g) {
1561 if (g == tsk->group_leader)
1562 continue;
1563
1564 p = g;
1565 do {
1566 if (p->mm) {
1567 if (p->mm == mm) {
1568 /*
1569 * p->sighand can't disappear, but
1570 * may be changed by de_thread()
1571 */
1572 lock_task_sighand(p, &flags);
1573 zap_process(p);
1574 unlock_task_sighand(p, &flags);
1575 }
1576 break;
1577 }
1578 } while ((p = next_thread(p)) != g);
1579 }
1580 rcu_read_unlock();
1581 done:
1582 return mm->core_waiters;
1583 }
1584
1585 static int coredump_wait(int exit_code)
1586 {
1587 struct task_struct *tsk = current;
1588 struct mm_struct *mm = tsk->mm;
1589 struct completion startup_done;
1590 struct completion *vfork_done;
1591 int core_waiters;
1592
1593 init_completion(&mm->core_done);
1594 init_completion(&startup_done);
1595 mm->core_startup_done = &startup_done;
1596
1597 core_waiters = zap_threads(tsk, mm, exit_code);
1598 up_write(&mm->mmap_sem);
1599
1600 if (unlikely(core_waiters < 0))
1601 goto fail;
1602
1603 /*
1604 * Make sure nobody is waiting for us to release the VM,
1605 * otherwise we can deadlock when we wait on each other
1606 */
1607 vfork_done = tsk->vfork_done;
1608 if (vfork_done) {
1609 tsk->vfork_done = NULL;
1610 complete(vfork_done);
1611 }
1612
1613 if (core_waiters)
1614 wait_for_completion(&startup_done);
1615 fail:
1616 BUG_ON(mm->core_waiters);
1617 return core_waiters;
1618 }
1619
1620 /*
1621 * set_dumpable converts traditional three-value dumpable to two flags and
1622 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1623 * these bits are not changed atomically. So get_dumpable can observe the
1624 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1625 * return either old dumpable or new one by paying attention to the order of
1626 * modifying the bits.
1627 *
1628 * dumpable | mm->flags (binary)
1629 * old new | initial interim final
1630 * ---------+-----------------------
1631 * 0 1 | 00 01 01
1632 * 0 2 | 00 10(*) 11
1633 * 1 0 | 01 00 00
1634 * 1 2 | 01 11 11
1635 * 2 0 | 11 10(*) 00
1636 * 2 1 | 11 11 01
1637 *
1638 * (*) get_dumpable regards interim value of 10 as 11.
1639 */
1640 void set_dumpable(struct mm_struct *mm, int value)
1641 {
1642 switch (value) {
1643 case 0:
1644 clear_bit(MMF_DUMPABLE, &mm->flags);
1645 smp_wmb();
1646 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1647 break;
1648 case 1:
1649 set_bit(MMF_DUMPABLE, &mm->flags);
1650 smp_wmb();
1651 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1652 break;
1653 case 2:
1654 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1655 smp_wmb();
1656 set_bit(MMF_DUMPABLE, &mm->flags);
1657 break;
1658 }
1659 }
1660
1661 int get_dumpable(struct mm_struct *mm)
1662 {
1663 int ret;
1664
1665 ret = mm->flags & 0x3;
1666 return (ret >= 2) ? 2 : ret;
1667 }
1668
1669 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1670 {
1671 char corename[CORENAME_MAX_SIZE + 1];
1672 struct mm_struct *mm = current->mm;
1673 struct linux_binfmt * binfmt;
1674 struct inode * inode;
1675 struct file * file;
1676 int retval = 0;
1677 int fsuid = current->fsuid;
1678 int flag = 0;
1679 int ispipe = 0;
1680 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1681 char **helper_argv = NULL;
1682 int helper_argc = 0;
1683 char *delimit;
1684
1685 audit_core_dumps(signr);
1686
1687 binfmt = current->binfmt;
1688 if (!binfmt || !binfmt->core_dump)
1689 goto fail;
1690 down_write(&mm->mmap_sem);
1691 /*
1692 * If another thread got here first, or we are not dumpable, bail out.
1693 */
1694 if (mm->core_waiters || !get_dumpable(mm)) {
1695 up_write(&mm->mmap_sem);
1696 goto fail;
1697 }
1698
1699 /*
1700 * We cannot trust fsuid as being the "true" uid of the
1701 * process nor do we know its entire history. We only know it
1702 * was tainted so we dump it as root in mode 2.
1703 */
1704 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1705 flag = O_EXCL; /* Stop rewrite attacks */
1706 current->fsuid = 0; /* Dump root private */
1707 }
1708
1709 retval = coredump_wait(exit_code);
1710 if (retval < 0)
1711 goto fail;
1712
1713 /*
1714 * Clear any false indication of pending signals that might
1715 * be seen by the filesystem code called to write the core file.
1716 */
1717 clear_thread_flag(TIF_SIGPENDING);
1718
1719 /*
1720 * lock_kernel() because format_corename() is controlled by sysctl, which
1721 * uses lock_kernel()
1722 */
1723 lock_kernel();
1724 ispipe = format_corename(corename, core_pattern, signr);
1725 unlock_kernel();
1726 /*
1727 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1728 * to a pipe. Since we're not writing directly to the filesystem
1729 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1730 * created unless the pipe reader choses to write out the core file
1731 * at which point file size limits and permissions will be imposed
1732 * as it does with any other process
1733 */
1734 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1735 goto fail_unlock;
1736
1737 if (ispipe) {
1738 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1739 /* Terminate the string before the first option */
1740 delimit = strchr(corename, ' ');
1741 if (delimit)
1742 *delimit = '\0';
1743 delimit = strrchr(helper_argv[0], '/');
1744 if (delimit)
1745 delimit++;
1746 else
1747 delimit = helper_argv[0];
1748 if (!strcmp(delimit, current->comm)) {
1749 printk(KERN_NOTICE "Recursive core dump detected, "
1750 "aborting\n");
1751 goto fail_unlock;
1752 }
1753
1754 core_limit = RLIM_INFINITY;
1755
1756 /* SIGPIPE can happen, but it's just never processed */
1757 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1758 &file)) {
1759 printk(KERN_INFO "Core dump to %s pipe failed\n",
1760 corename);
1761 goto fail_unlock;
1762 }
1763 } else
1764 file = filp_open(corename,
1765 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1766 0600);
1767 if (IS_ERR(file))
1768 goto fail_unlock;
1769 inode = file->f_path.dentry->d_inode;
1770 if (inode->i_nlink > 1)
1771 goto close_fail; /* multiple links - don't dump */
1772 if (!ispipe && d_unhashed(file->f_path.dentry))
1773 goto close_fail;
1774
1775 /* AK: actually i see no reason to not allow this for named pipes etc.,
1776 but keep the previous behaviour for now. */
1777 if (!ispipe && !S_ISREG(inode->i_mode))
1778 goto close_fail;
1779 /*
1780 * Dont allow local users get cute and trick others to coredump
1781 * into their pre-created files:
1782 */
1783 if (inode->i_uid != current->fsuid)
1784 goto close_fail;
1785 if (!file->f_op)
1786 goto close_fail;
1787 if (!file->f_op->write)
1788 goto close_fail;
1789 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1790 goto close_fail;
1791
1792 retval = binfmt->core_dump(signr, regs, file, core_limit);
1793
1794 if (retval)
1795 current->signal->group_exit_code |= 0x80;
1796 close_fail:
1797 filp_close(file, NULL);
1798 fail_unlock:
1799 if (helper_argv)
1800 argv_free(helper_argv);
1801
1802 current->fsuid = fsuid;
1803 complete_all(&mm->core_done);
1804 fail:
1805 return retval;
1806 }
Linux kernel - Deliverables
This page took 0.064545 seconds and 6 git commands to generate.