xref: /linux/fs/exec.c (revision ccea15f45eb0ab12d658f88b5d4be005cb2bb1a7)
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/config.h>
26 #include <linux/slab.h>
27 #include <linux/file.h>
28 #include <linux/mman.h>
29 #include <linux/a.out.h>
30 #include <linux/stat.h>
31 #include <linux/fcntl.h>
32 #include <linux/smp_lock.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/module.h>
43 #include <linux/namei.h>
44 #include <linux/proc_fs.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/rmap.h>
50 #include <linux/acct.h>
51 #include <linux/cn_proc.h>
52 
53 #include <asm/uaccess.h>
54 #include <asm/mmu_context.h>
55 
56 #ifdef CONFIG_KMOD
57 #include <linux/kmod.h>
58 #endif
59 
60 int core_uses_pid;
61 char core_pattern[65] = "core";
62 int suid_dumpable = 0;
63 
64 EXPORT_SYMBOL(suid_dumpable);
65 /* The maximal length of core_pattern is also specified in sysctl.c */
66 
67 static struct linux_binfmt *formats;
68 static DEFINE_RWLOCK(binfmt_lock);
69 
70 int register_binfmt(struct linux_binfmt * fmt)
71 {
72 	struct linux_binfmt ** tmp = &formats;
73 
74 	if (!fmt)
75 		return -EINVAL;
76 	if (fmt->next)
77 		return -EBUSY;
78 	write_lock(&binfmt_lock);
79 	while (*tmp) {
80 		if (fmt == *tmp) {
81 			write_unlock(&binfmt_lock);
82 			return -EBUSY;
83 		}
84 		tmp = &(*tmp)->next;
85 	}
86 	fmt->next = formats;
87 	formats = fmt;
88 	write_unlock(&binfmt_lock);
89 	return 0;
90 }
91 
92 EXPORT_SYMBOL(register_binfmt);
93 
94 int unregister_binfmt(struct linux_binfmt * fmt)
95 {
96 	struct linux_binfmt ** tmp = &formats;
97 
98 	write_lock(&binfmt_lock);
99 	while (*tmp) {
100 		if (fmt == *tmp) {
101 			*tmp = fmt->next;
102 			write_unlock(&binfmt_lock);
103 			return 0;
104 		}
105 		tmp = &(*tmp)->next;
106 	}
107 	write_unlock(&binfmt_lock);
108 	return -EINVAL;
109 }
110 
111 EXPORT_SYMBOL(unregister_binfmt);
112 
113 static inline void put_binfmt(struct linux_binfmt * fmt)
114 {
115 	module_put(fmt->module);
116 }
117 
118 /*
119  * Note that a shared library must be both readable and executable due to
120  * security reasons.
121  *
122  * Also note that we take the address to load from from the file itself.
123  */
124 asmlinkage long sys_uselib(const char __user * library)
125 {
126 	struct file * file;
127 	struct nameidata nd;
128 	int error;
129 
130 	error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
131 	if (error)
132 		goto out;
133 
134 	error = -EINVAL;
135 	if (!S_ISREG(nd.dentry->d_inode->i_mode))
136 		goto exit;
137 
138 	error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
139 	if (error)
140 		goto exit;
141 
142 	file = nameidata_to_filp(&nd, O_RDONLY);
143 	error = PTR_ERR(file);
144 	if (IS_ERR(file))
145 		goto out;
146 
147 	error = -ENOEXEC;
148 	if(file->f_op) {
149 		struct linux_binfmt * fmt;
150 
151 		read_lock(&binfmt_lock);
152 		for (fmt = formats ; fmt ; fmt = fmt->next) {
153 			if (!fmt->load_shlib)
154 				continue;
155 			if (!try_module_get(fmt->module))
156 				continue;
157 			read_unlock(&binfmt_lock);
158 			error = fmt->load_shlib(file);
159 			read_lock(&binfmt_lock);
160 			put_binfmt(fmt);
161 			if (error != -ENOEXEC)
162 				break;
163 		}
164 		read_unlock(&binfmt_lock);
165 	}
166 	fput(file);
167 out:
168   	return error;
169 exit:
170 	release_open_intent(&nd);
171 	path_release(&nd);
172 	goto out;
173 }
174 
175 /*
176  * count() counts the number of strings in array ARGV.
177  */
178 static int count(char __user * __user * argv, int max)
179 {
180 	int i = 0;
181 
182 	if (argv != NULL) {
183 		for (;;) {
184 			char __user * p;
185 
186 			if (get_user(p, argv))
187 				return -EFAULT;
188 			if (!p)
189 				break;
190 			argv++;
191 			if(++i > max)
192 				return -E2BIG;
193 			cond_resched();
194 		}
195 	}
196 	return i;
197 }
198 
199 /*
200  * 'copy_strings()' copies argument/environment strings from user
201  * memory to free pages in kernel mem. These are in a format ready
202  * to be put directly into the top of new user memory.
203  */
204 static int copy_strings(int argc, char __user * __user * argv,
205 			struct linux_binprm *bprm)
206 {
207 	struct page *kmapped_page = NULL;
208 	char *kaddr = NULL;
209 	int ret;
210 
211 	while (argc-- > 0) {
212 		char __user *str;
213 		int len;
214 		unsigned long pos;
215 
216 		if (get_user(str, argv+argc) ||
217 				!(len = strnlen_user(str, bprm->p))) {
218 			ret = -EFAULT;
219 			goto out;
220 		}
221 
222 		if (bprm->p < len)  {
223 			ret = -E2BIG;
224 			goto out;
225 		}
226 
227 		bprm->p -= len;
228 		/* XXX: add architecture specific overflow check here. */
229 		pos = bprm->p;
230 
231 		while (len > 0) {
232 			int i, new, err;
233 			int offset, bytes_to_copy;
234 			struct page *page;
235 
236 			offset = pos % PAGE_SIZE;
237 			i = pos/PAGE_SIZE;
238 			page = bprm->page[i];
239 			new = 0;
240 			if (!page) {
241 				page = alloc_page(GFP_HIGHUSER);
242 				bprm->page[i] = page;
243 				if (!page) {
244 					ret = -ENOMEM;
245 					goto out;
246 				}
247 				new = 1;
248 			}
249 
250 			if (page != kmapped_page) {
251 				if (kmapped_page)
252 					kunmap(kmapped_page);
253 				kmapped_page = page;
254 				kaddr = kmap(kmapped_page);
255 			}
256 			if (new && offset)
257 				memset(kaddr, 0, offset);
258 			bytes_to_copy = PAGE_SIZE - offset;
259 			if (bytes_to_copy > len) {
260 				bytes_to_copy = len;
261 				if (new)
262 					memset(kaddr+offset+len, 0,
263 						PAGE_SIZE-offset-len);
264 			}
265 			err = copy_from_user(kaddr+offset, str, bytes_to_copy);
266 			if (err) {
267 				ret = -EFAULT;
268 				goto out;
269 			}
270 
271 			pos += bytes_to_copy;
272 			str += bytes_to_copy;
273 			len -= bytes_to_copy;
274 		}
275 	}
276 	ret = 0;
277 out:
278 	if (kmapped_page)
279 		kunmap(kmapped_page);
280 	return ret;
281 }
282 
283 /*
284  * Like copy_strings, but get argv and its values from kernel memory.
285  */
286 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
287 {
288 	int r;
289 	mm_segment_t oldfs = get_fs();
290 	set_fs(KERNEL_DS);
291 	r = copy_strings(argc, (char __user * __user *)argv, bprm);
292 	set_fs(oldfs);
293 	return r;
294 }
295 
296 EXPORT_SYMBOL(copy_strings_kernel);
297 
298 #ifdef CONFIG_MMU
299 /*
300  * This routine is used to map in a page into an address space: needed by
301  * execve() for the initial stack and environment pages.
302  *
303  * vma->vm_mm->mmap_sem is held for writing.
304  */
305 void install_arg_page(struct vm_area_struct *vma,
306 			struct page *page, unsigned long address)
307 {
308 	struct mm_struct *mm = vma->vm_mm;
309 	pte_t * pte;
310 	spinlock_t *ptl;
311 
312 	if (unlikely(anon_vma_prepare(vma)))
313 		goto out;
314 
315 	flush_dcache_page(page);
316 	pte = get_locked_pte(mm, address, &ptl);
317 	if (!pte)
318 		goto out;
319 	if (!pte_none(*pte)) {
320 		pte_unmap_unlock(pte, ptl);
321 		goto out;
322 	}
323 	inc_mm_counter(mm, anon_rss);
324 	lru_cache_add_active(page);
325 	set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
326 					page, vma->vm_page_prot))));
327 	page_add_new_anon_rmap(page, vma, address);
328 	pte_unmap_unlock(pte, ptl);
329 
330 	/* no need for flush_tlb */
331 	return;
332 out:
333 	__free_page(page);
334 	force_sig(SIGKILL, current);
335 }
336 
337 #define EXTRA_STACK_VM_PAGES	20	/* random */
338 
339 int setup_arg_pages(struct linux_binprm *bprm,
340 		    unsigned long stack_top,
341 		    int executable_stack)
342 {
343 	unsigned long stack_base;
344 	struct vm_area_struct *mpnt;
345 	struct mm_struct *mm = current->mm;
346 	int i, ret;
347 	long arg_size;
348 
349 #ifdef CONFIG_STACK_GROWSUP
350 	/* Move the argument and environment strings to the bottom of the
351 	 * stack space.
352 	 */
353 	int offset, j;
354 	char *to, *from;
355 
356 	/* Start by shifting all the pages down */
357 	i = 0;
358 	for (j = 0; j < MAX_ARG_PAGES; j++) {
359 		struct page *page = bprm->page[j];
360 		if (!page)
361 			continue;
362 		bprm->page[i++] = page;
363 	}
364 
365 	/* Now move them within their pages */
366 	offset = bprm->p % PAGE_SIZE;
367 	to = kmap(bprm->page[0]);
368 	for (j = 1; j < i; j++) {
369 		memmove(to, to + offset, PAGE_SIZE - offset);
370 		from = kmap(bprm->page[j]);
371 		memcpy(to + PAGE_SIZE - offset, from, offset);
372 		kunmap(bprm->page[j - 1]);
373 		to = from;
374 	}
375 	memmove(to, to + offset, PAGE_SIZE - offset);
376 	kunmap(bprm->page[j - 1]);
377 
378 	/* Limit stack size to 1GB */
379 	stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
380 	if (stack_base > (1 << 30))
381 		stack_base = 1 << 30;
382 	stack_base = PAGE_ALIGN(stack_top - stack_base);
383 
384 	/* Adjust bprm->p to point to the end of the strings. */
385 	bprm->p = stack_base + PAGE_SIZE * i - offset;
386 
387 	mm->arg_start = stack_base;
388 	arg_size = i << PAGE_SHIFT;
389 
390 	/* zero pages that were copied above */
391 	while (i < MAX_ARG_PAGES)
392 		bprm->page[i++] = NULL;
393 #else
394 	stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
395 	stack_base = PAGE_ALIGN(stack_base);
396 	bprm->p += stack_base;
397 	mm->arg_start = bprm->p;
398 	arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
399 #endif
400 
401 	arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
402 
403 	if (bprm->loader)
404 		bprm->loader += stack_base;
405 	bprm->exec += stack_base;
406 
407 	mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
408 	if (!mpnt)
409 		return -ENOMEM;
410 
411 	memset(mpnt, 0, sizeof(*mpnt));
412 
413 	down_write(&mm->mmap_sem);
414 	{
415 		mpnt->vm_mm = mm;
416 #ifdef CONFIG_STACK_GROWSUP
417 		mpnt->vm_start = stack_base;
418 		mpnt->vm_end = stack_base + arg_size;
419 #else
420 		mpnt->vm_end = stack_top;
421 		mpnt->vm_start = mpnt->vm_end - arg_size;
422 #endif
423 		/* Adjust stack execute permissions; explicitly enable
424 		 * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
425 		 * and leave alone (arch default) otherwise. */
426 		if (unlikely(executable_stack == EXSTACK_ENABLE_X))
427 			mpnt->vm_flags = VM_STACK_FLAGS |  VM_EXEC;
428 		else if (executable_stack == EXSTACK_DISABLE_X)
429 			mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
430 		else
431 			mpnt->vm_flags = VM_STACK_FLAGS;
432 		mpnt->vm_flags |= mm->def_flags;
433 		mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
434 		if ((ret = insert_vm_struct(mm, mpnt))) {
435 			up_write(&mm->mmap_sem);
436 			kmem_cache_free(vm_area_cachep, mpnt);
437 			return ret;
438 		}
439 		mm->stack_vm = mm->total_vm = vma_pages(mpnt);
440 	}
441 
442 	for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
443 		struct page *page = bprm->page[i];
444 		if (page) {
445 			bprm->page[i] = NULL;
446 			install_arg_page(mpnt, page, stack_base);
447 		}
448 		stack_base += PAGE_SIZE;
449 	}
450 	up_write(&mm->mmap_sem);
451 
452 	return 0;
453 }
454 
455 EXPORT_SYMBOL(setup_arg_pages);
456 
457 #define free_arg_pages(bprm) do { } while (0)
458 
459 #else
460 
461 static inline void free_arg_pages(struct linux_binprm *bprm)
462 {
463 	int i;
464 
465 	for (i = 0; i < MAX_ARG_PAGES; i++) {
466 		if (bprm->page[i])
467 			__free_page(bprm->page[i]);
468 		bprm->page[i] = NULL;
469 	}
470 }
471 
472 #endif /* CONFIG_MMU */
473 
474 struct file *open_exec(const char *name)
475 {
476 	struct nameidata nd;
477 	int err;
478 	struct file *file;
479 
480 	err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
481 	file = ERR_PTR(err);
482 
483 	if (!err) {
484 		struct inode *inode = nd.dentry->d_inode;
485 		file = ERR_PTR(-EACCES);
486 		if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
487 		    S_ISREG(inode->i_mode)) {
488 			int err = vfs_permission(&nd, MAY_EXEC);
489 			if (!err && !(inode->i_mode & 0111))
490 				err = -EACCES;
491 			file = ERR_PTR(err);
492 			if (!err) {
493 				file = nameidata_to_filp(&nd, O_RDONLY);
494 				if (!IS_ERR(file)) {
495 					err = deny_write_access(file);
496 					if (err) {
497 						fput(file);
498 						file = ERR_PTR(err);
499 					}
500 				}
501 out:
502 				return file;
503 			}
504 		}
505 		release_open_intent(&nd);
506 		path_release(&nd);
507 	}
508 	goto out;
509 }
510 
511 EXPORT_SYMBOL(open_exec);
512 
513 int kernel_read(struct file *file, unsigned long offset,
514 	char *addr, unsigned long count)
515 {
516 	mm_segment_t old_fs;
517 	loff_t pos = offset;
518 	int result;
519 
520 	old_fs = get_fs();
521 	set_fs(get_ds());
522 	/* The cast to a user pointer is valid due to the set_fs() */
523 	result = vfs_read(file, (void __user *)addr, count, &pos);
524 	set_fs(old_fs);
525 	return result;
526 }
527 
528 EXPORT_SYMBOL(kernel_read);
529 
530 static int exec_mmap(struct mm_struct *mm)
531 {
532 	struct task_struct *tsk;
533 	struct mm_struct * old_mm, *active_mm;
534 
535 	/* Notify parent that we're no longer interested in the old VM */
536 	tsk = current;
537 	old_mm = current->mm;
538 	mm_release(tsk, old_mm);
539 
540 	if (old_mm) {
541 		/*
542 		 * Make sure that if there is a core dump in progress
543 		 * for the old mm, we get out and die instead of going
544 		 * through with the exec.  We must hold mmap_sem around
545 		 * checking core_waiters and changing tsk->mm.  The
546 		 * core-inducing thread will increment core_waiters for
547 		 * each thread whose ->mm == old_mm.
548 		 */
549 		down_read(&old_mm->mmap_sem);
550 		if (unlikely(old_mm->core_waiters)) {
551 			up_read(&old_mm->mmap_sem);
552 			return -EINTR;
553 		}
554 	}
555 	task_lock(tsk);
556 	active_mm = tsk->active_mm;
557 	tsk->mm = mm;
558 	tsk->active_mm = mm;
559 	activate_mm(active_mm, mm);
560 	task_unlock(tsk);
561 	arch_pick_mmap_layout(mm);
562 	if (old_mm) {
563 		up_read(&old_mm->mmap_sem);
564 		BUG_ON(active_mm != old_mm);
565 		mmput(old_mm);
566 		return 0;
567 	}
568 	mmdrop(active_mm);
569 	return 0;
570 }
571 
572 /*
573  * This function makes sure the current process has its own signal table,
574  * so that flush_signal_handlers can later reset the handlers without
575  * disturbing other processes.  (Other processes might share the signal
576  * table via the CLONE_SIGHAND option to clone().)
577  */
578 static int de_thread(struct task_struct *tsk)
579 {
580 	struct signal_struct *sig = tsk->signal;
581 	struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
582 	spinlock_t *lock = &oldsighand->siglock;
583 	struct task_struct *leader = NULL;
584 	int count;
585 
586 	/*
587 	 * If we don't share sighandlers, then we aren't sharing anything
588 	 * and we can just re-use it all.
589 	 */
590 	if (atomic_read(&oldsighand->count) <= 1) {
591 		BUG_ON(atomic_read(&sig->count) != 1);
592 		exit_itimers(sig);
593 		return 0;
594 	}
595 
596 	newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
597 	if (!newsighand)
598 		return -ENOMEM;
599 
600 	if (thread_group_empty(current))
601 		goto no_thread_group;
602 
603 	/*
604 	 * Kill all other threads in the thread group.
605 	 * We must hold tasklist_lock to call zap_other_threads.
606 	 */
607 	read_lock(&tasklist_lock);
608 	spin_lock_irq(lock);
609 	if (sig->flags & SIGNAL_GROUP_EXIT) {
610 		/*
611 		 * Another group action in progress, just
612 		 * return so that the signal is processed.
613 		 */
614 		spin_unlock_irq(lock);
615 		read_unlock(&tasklist_lock);
616 		kmem_cache_free(sighand_cachep, newsighand);
617 		return -EAGAIN;
618 	}
619 
620 	/*
621 	 * child_reaper ignores SIGKILL, change it now.
622 	 * Reparenting needs write_lock on tasklist_lock,
623 	 * so it is safe to do it under read_lock.
624 	 */
625 	if (unlikely(current->group_leader == child_reaper))
626 		child_reaper = current;
627 
628 	zap_other_threads(current);
629 	read_unlock(&tasklist_lock);
630 
631 	/*
632 	 * Account for the thread group leader hanging around:
633 	 */
634 	count = 1;
635 	if (!thread_group_leader(current)) {
636 		count = 2;
637 		/*
638 		 * The SIGALRM timer survives the exec, but needs to point
639 		 * at us as the new group leader now.  We have a race with
640 		 * a timer firing now getting the old leader, so we need to
641 		 * synchronize with any firing (by calling del_timer_sync)
642 		 * before we can safely let the old group leader die.
643 		 */
644 		sig->tsk = current;
645 		spin_unlock_irq(lock);
646 		if (hrtimer_cancel(&sig->real_timer))
647 			hrtimer_restart(&sig->real_timer);
648 		spin_lock_irq(lock);
649 	}
650 	while (atomic_read(&sig->count) > count) {
651 		sig->group_exit_task = current;
652 		sig->notify_count = count;
653 		__set_current_state(TASK_UNINTERRUPTIBLE);
654 		spin_unlock_irq(lock);
655 		schedule();
656 		spin_lock_irq(lock);
657 	}
658 	sig->group_exit_task = NULL;
659 	sig->notify_count = 0;
660 	spin_unlock_irq(lock);
661 
662 	/*
663 	 * At this point all other threads have exited, all we have to
664 	 * do is to wait for the thread group leader to become inactive,
665 	 * and to assume its PID:
666 	 */
667 	if (!thread_group_leader(current)) {
668 		struct task_struct *parent;
669 		struct dentry *proc_dentry1, *proc_dentry2;
670 		unsigned long ptrace;
671 
672 		/*
673 		 * Wait for the thread group leader to be a zombie.
674 		 * It should already be zombie at this point, most
675 		 * of the time.
676 		 */
677 		leader = current->group_leader;
678 		while (leader->exit_state != EXIT_ZOMBIE)
679 			yield();
680 
681 		spin_lock(&leader->proc_lock);
682 		spin_lock(&current->proc_lock);
683 		proc_dentry1 = proc_pid_unhash(current);
684 		proc_dentry2 = proc_pid_unhash(leader);
685 		write_lock_irq(&tasklist_lock);
686 
687 		BUG_ON(leader->tgid != current->tgid);
688 		BUG_ON(current->pid == current->tgid);
689 		/*
690 		 * An exec() starts a new thread group with the
691 		 * TGID of the previous thread group. Rehash the
692 		 * two threads with a switched PID, and release
693 		 * the former thread group leader:
694 		 */
695 		ptrace = leader->ptrace;
696 		parent = leader->parent;
697 		if (unlikely(ptrace) && unlikely(parent == current)) {
698 			/*
699 			 * Joker was ptracing his own group leader,
700 			 * and now he wants to be his own parent!
701 			 * We can't have that.
702 			 */
703 			ptrace = 0;
704 		}
705 
706 		ptrace_unlink(current);
707 		ptrace_unlink(leader);
708 		remove_parent(current);
709 		remove_parent(leader);
710 
711 
712 		/* Become a process group leader with the old leader's pid.
713 		 * Note: The old leader also uses thispid until release_task
714 		 *       is called.  Odd but simple and correct.
715 		 */
716 		detach_pid(current, PIDTYPE_PID);
717 		current->pid = leader->pid;
718 		attach_pid(current, PIDTYPE_PID,  current->pid);
719 		attach_pid(current, PIDTYPE_PGID, current->signal->pgrp);
720 		attach_pid(current, PIDTYPE_SID,  current->signal->session);
721 		list_add_tail(&current->tasks, &init_task.tasks);
722 
723 		current->parent = current->real_parent = leader->real_parent;
724 		leader->parent = leader->real_parent = child_reaper;
725 		current->group_leader = current;
726 		leader->group_leader = current;
727 
728 		/* Reduce leader to a thread */
729 		detach_pid(leader, PIDTYPE_PGID);
730 		detach_pid(leader, PIDTYPE_SID);
731 		list_del_init(&leader->tasks);
732 
733 		add_parent(current);
734 		add_parent(leader);
735 		if (ptrace) {
736 			current->ptrace = ptrace;
737 			__ptrace_link(current, parent);
738 		}
739 
740 		current->exit_signal = SIGCHLD;
741 
742 		BUG_ON(leader->exit_state != EXIT_ZOMBIE);
743 		leader->exit_state = EXIT_DEAD;
744 
745 		write_unlock_irq(&tasklist_lock);
746 		spin_unlock(&leader->proc_lock);
747 		spin_unlock(&current->proc_lock);
748 		proc_pid_flush(proc_dentry1);
749 		proc_pid_flush(proc_dentry2);
750         }
751 
752 	/*
753 	 * There may be one thread left which is just exiting,
754 	 * but it's safe to stop telling the group to kill themselves.
755 	 */
756 	sig->flags = 0;
757 
758 no_thread_group:
759 	exit_itimers(sig);
760 	if (leader)
761 		release_task(leader);
762 
763 	BUG_ON(atomic_read(&sig->count) != 1);
764 
765 	if (atomic_read(&oldsighand->count) == 1) {
766 		/*
767 		 * Now that we nuked the rest of the thread group,
768 		 * it turns out we are not sharing sighand any more either.
769 		 * So we can just keep it.
770 		 */
771 		kmem_cache_free(sighand_cachep, newsighand);
772 	} else {
773 		/*
774 		 * Move our state over to newsighand and switch it in.
775 		 */
776 		atomic_set(&newsighand->count, 1);
777 		memcpy(newsighand->action, oldsighand->action,
778 		       sizeof(newsighand->action));
779 
780 		write_lock_irq(&tasklist_lock);
781 		spin_lock(&oldsighand->siglock);
782 		spin_lock(&newsighand->siglock);
783 
784 		rcu_assign_pointer(current->sighand, newsighand);
785 		recalc_sigpending();
786 
787 		spin_unlock(&newsighand->siglock);
788 		spin_unlock(&oldsighand->siglock);
789 		write_unlock_irq(&tasklist_lock);
790 
791 		if (atomic_dec_and_test(&oldsighand->count))
792 			kmem_cache_free(sighand_cachep, oldsighand);
793 	}
794 
795 	BUG_ON(!thread_group_leader(current));
796 	return 0;
797 }
798 
799 /*
800  * These functions flushes out all traces of the currently running executable
801  * so that a new one can be started
802  */
803 
804 static void flush_old_files(struct files_struct * files)
805 {
806 	long j = -1;
807 	struct fdtable *fdt;
808 
809 	spin_lock(&files->file_lock);
810 	for (;;) {
811 		unsigned long set, i;
812 
813 		j++;
814 		i = j * __NFDBITS;
815 		fdt = files_fdtable(files);
816 		if (i >= fdt->max_fds || i >= fdt->max_fdset)
817 			break;
818 		set = fdt->close_on_exec->fds_bits[j];
819 		if (!set)
820 			continue;
821 		fdt->close_on_exec->fds_bits[j] = 0;
822 		spin_unlock(&files->file_lock);
823 		for ( ; set ; i++,set >>= 1) {
824 			if (set & 1) {
825 				sys_close(i);
826 			}
827 		}
828 		spin_lock(&files->file_lock);
829 
830 	}
831 	spin_unlock(&files->file_lock);
832 }
833 
834 void get_task_comm(char *buf, struct task_struct *tsk)
835 {
836 	/* buf must be at least sizeof(tsk->comm) in size */
837 	task_lock(tsk);
838 	strncpy(buf, tsk->comm, sizeof(tsk->comm));
839 	task_unlock(tsk);
840 }
841 
842 void set_task_comm(struct task_struct *tsk, char *buf)
843 {
844 	task_lock(tsk);
845 	strlcpy(tsk->comm, buf, sizeof(tsk->comm));
846 	task_unlock(tsk);
847 }
848 
849 int flush_old_exec(struct linux_binprm * bprm)
850 {
851 	char * name;
852 	int i, ch, retval;
853 	struct files_struct *files;
854 	char tcomm[sizeof(current->comm)];
855 
856 	/*
857 	 * Make sure we have a private signal table and that
858 	 * we are unassociated from the previous thread group.
859 	 */
860 	retval = de_thread(current);
861 	if (retval)
862 		goto out;
863 
864 	/*
865 	 * Make sure we have private file handles. Ask the
866 	 * fork helper to do the work for us and the exit
867 	 * helper to do the cleanup of the old one.
868 	 */
869 	files = current->files;		/* refcounted so safe to hold */
870 	retval = unshare_files();
871 	if (retval)
872 		goto out;
873 	/*
874 	 * Release all of the old mmap stuff
875 	 */
876 	retval = exec_mmap(bprm->mm);
877 	if (retval)
878 		goto mmap_failed;
879 
880 	bprm->mm = NULL;		/* We're using it now */
881 
882 	/* This is the point of no return */
883 	steal_locks(files);
884 	put_files_struct(files);
885 
886 	current->sas_ss_sp = current->sas_ss_size = 0;
887 
888 	if (current->euid == current->uid && current->egid == current->gid)
889 		current->mm->dumpable = 1;
890 	else
891 		current->mm->dumpable = suid_dumpable;
892 
893 	name = bprm->filename;
894 
895 	/* Copies the binary name from after last slash */
896 	for (i=0; (ch = *(name++)) != '\0';) {
897 		if (ch == '/')
898 			i = 0; /* overwrite what we wrote */
899 		else
900 			if (i < (sizeof(tcomm) - 1))
901 				tcomm[i++] = ch;
902 	}
903 	tcomm[i] = '\0';
904 	set_task_comm(current, tcomm);
905 
906 	current->flags &= ~PF_RANDOMIZE;
907 	flush_thread();
908 
909 	/* Set the new mm task size. We have to do that late because it may
910 	 * depend on TIF_32BIT which is only updated in flush_thread() on
911 	 * some architectures like powerpc
912 	 */
913 	current->mm->task_size = TASK_SIZE;
914 
915 	if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
916 	    file_permission(bprm->file, MAY_READ) ||
917 	    (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
918 		suid_keys(current);
919 		current->mm->dumpable = suid_dumpable;
920 	}
921 
922 	/* An exec changes our domain. We are no longer part of the thread
923 	   group */
924 
925 	current->self_exec_id++;
926 
927 	flush_signal_handlers(current, 0);
928 	flush_old_files(current->files);
929 
930 	return 0;
931 
932 mmap_failed:
933 	put_files_struct(current->files);
934 	current->files = files;
935 out:
936 	return retval;
937 }
938 
939 EXPORT_SYMBOL(flush_old_exec);
940 
941 /*
942  * Fill the binprm structure from the inode.
943  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
944  */
945 int prepare_binprm(struct linux_binprm *bprm)
946 {
947 	int mode;
948 	struct inode * inode = bprm->file->f_dentry->d_inode;
949 	int retval;
950 
951 	mode = inode->i_mode;
952 	/*
953 	 * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
954 	 * generic_permission lets a non-executable through
955 	 */
956 	if (!(mode & 0111))	/* with at least _one_ execute bit set */
957 		return -EACCES;
958 	if (bprm->file->f_op == NULL)
959 		return -EACCES;
960 
961 	bprm->e_uid = current->euid;
962 	bprm->e_gid = current->egid;
963 
964 	if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
965 		/* Set-uid? */
966 		if (mode & S_ISUID) {
967 			current->personality &= ~PER_CLEAR_ON_SETID;
968 			bprm->e_uid = inode->i_uid;
969 		}
970 
971 		/* Set-gid? */
972 		/*
973 		 * If setgid is set but no group execute bit then this
974 		 * is a candidate for mandatory locking, not a setgid
975 		 * executable.
976 		 */
977 		if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
978 			current->personality &= ~PER_CLEAR_ON_SETID;
979 			bprm->e_gid = inode->i_gid;
980 		}
981 	}
982 
983 	/* fill in binprm security blob */
984 	retval = security_bprm_set(bprm);
985 	if (retval)
986 		return retval;
987 
988 	memset(bprm->buf,0,BINPRM_BUF_SIZE);
989 	return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
990 }
991 
992 EXPORT_SYMBOL(prepare_binprm);
993 
994 static int unsafe_exec(struct task_struct *p)
995 {
996 	int unsafe = 0;
997 	if (p->ptrace & PT_PTRACED) {
998 		if (p->ptrace & PT_PTRACE_CAP)
999 			unsafe |= LSM_UNSAFE_PTRACE_CAP;
1000 		else
1001 			unsafe |= LSM_UNSAFE_PTRACE;
1002 	}
1003 	if (atomic_read(&p->fs->count) > 1 ||
1004 	    atomic_read(&p->files->count) > 1 ||
1005 	    atomic_read(&p->sighand->count) > 1)
1006 		unsafe |= LSM_UNSAFE_SHARE;
1007 
1008 	return unsafe;
1009 }
1010 
1011 void compute_creds(struct linux_binprm *bprm)
1012 {
1013 	int unsafe;
1014 
1015 	if (bprm->e_uid != current->uid)
1016 		suid_keys(current);
1017 	exec_keys(current);
1018 
1019 	task_lock(current);
1020 	unsafe = unsafe_exec(current);
1021 	security_bprm_apply_creds(bprm, unsafe);
1022 	task_unlock(current);
1023 	security_bprm_post_apply_creds(bprm);
1024 }
1025 
1026 EXPORT_SYMBOL(compute_creds);
1027 
1028 void remove_arg_zero(struct linux_binprm *bprm)
1029 {
1030 	if (bprm->argc) {
1031 		unsigned long offset;
1032 		char * kaddr;
1033 		struct page *page;
1034 
1035 		offset = bprm->p % PAGE_SIZE;
1036 		goto inside;
1037 
1038 		while (bprm->p++, *(kaddr+offset++)) {
1039 			if (offset != PAGE_SIZE)
1040 				continue;
1041 			offset = 0;
1042 			kunmap_atomic(kaddr, KM_USER0);
1043 inside:
1044 			page = bprm->page[bprm->p/PAGE_SIZE];
1045 			kaddr = kmap_atomic(page, KM_USER0);
1046 		}
1047 		kunmap_atomic(kaddr, KM_USER0);
1048 		bprm->argc--;
1049 	}
1050 }
1051 
1052 EXPORT_SYMBOL(remove_arg_zero);
1053 
1054 /*
1055  * cycle the list of binary formats handler, until one recognizes the image
1056  */
1057 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1058 {
1059 	int try,retval;
1060 	struct linux_binfmt *fmt;
1061 #ifdef __alpha__
1062 	/* handle /sbin/loader.. */
1063 	{
1064 	    struct exec * eh = (struct exec *) bprm->buf;
1065 
1066 	    if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1067 		(eh->fh.f_flags & 0x3000) == 0x3000)
1068 	    {
1069 		struct file * file;
1070 		unsigned long loader;
1071 
1072 		allow_write_access(bprm->file);
1073 		fput(bprm->file);
1074 		bprm->file = NULL;
1075 
1076 	        loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1077 
1078 		file = open_exec("/sbin/loader");
1079 		retval = PTR_ERR(file);
1080 		if (IS_ERR(file))
1081 			return retval;
1082 
1083 		/* Remember if the application is TASO.  */
1084 		bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1085 
1086 		bprm->file = file;
1087 		bprm->loader = loader;
1088 		retval = prepare_binprm(bprm);
1089 		if (retval<0)
1090 			return retval;
1091 		/* should call search_binary_handler recursively here,
1092 		   but it does not matter */
1093 	    }
1094 	}
1095 #endif
1096 	retval = security_bprm_check(bprm);
1097 	if (retval)
1098 		return retval;
1099 
1100 	/* kernel module loader fixup */
1101 	/* so we don't try to load run modprobe in kernel space. */
1102 	set_fs(USER_DS);
1103 	retval = -ENOENT;
1104 	for (try=0; try<2; try++) {
1105 		read_lock(&binfmt_lock);
1106 		for (fmt = formats ; fmt ; fmt = fmt->next) {
1107 			int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1108 			if (!fn)
1109 				continue;
1110 			if (!try_module_get(fmt->module))
1111 				continue;
1112 			read_unlock(&binfmt_lock);
1113 			retval = fn(bprm, regs);
1114 			if (retval >= 0) {
1115 				put_binfmt(fmt);
1116 				allow_write_access(bprm->file);
1117 				if (bprm->file)
1118 					fput(bprm->file);
1119 				bprm->file = NULL;
1120 				current->did_exec = 1;
1121 				proc_exec_connector(current);
1122 				return retval;
1123 			}
1124 			read_lock(&binfmt_lock);
1125 			put_binfmt(fmt);
1126 			if (retval != -ENOEXEC || bprm->mm == NULL)
1127 				break;
1128 			if (!bprm->file) {
1129 				read_unlock(&binfmt_lock);
1130 				return retval;
1131 			}
1132 		}
1133 		read_unlock(&binfmt_lock);
1134 		if (retval != -ENOEXEC || bprm->mm == NULL) {
1135 			break;
1136 #ifdef CONFIG_KMOD
1137 		}else{
1138 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1139 			if (printable(bprm->buf[0]) &&
1140 			    printable(bprm->buf[1]) &&
1141 			    printable(bprm->buf[2]) &&
1142 			    printable(bprm->buf[3]))
1143 				break; /* -ENOEXEC */
1144 			request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1145 #endif
1146 		}
1147 	}
1148 	return retval;
1149 }
1150 
1151 EXPORT_SYMBOL(search_binary_handler);
1152 
1153 /*
1154  * sys_execve() executes a new program.
1155  */
1156 int do_execve(char * filename,
1157 	char __user *__user *argv,
1158 	char __user *__user *envp,
1159 	struct pt_regs * regs)
1160 {
1161 	struct linux_binprm *bprm;
1162 	struct file *file;
1163 	int retval;
1164 	int i;
1165 
1166 	retval = -ENOMEM;
1167 	bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1168 	if (!bprm)
1169 		goto out_ret;
1170 
1171 	file = open_exec(filename);
1172 	retval = PTR_ERR(file);
1173 	if (IS_ERR(file))
1174 		goto out_kfree;
1175 
1176 	sched_exec();
1177 
1178 	bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1179 
1180 	bprm->file = file;
1181 	bprm->filename = filename;
1182 	bprm->interp = filename;
1183 	bprm->mm = mm_alloc();
1184 	retval = -ENOMEM;
1185 	if (!bprm->mm)
1186 		goto out_file;
1187 
1188 	retval = init_new_context(current, bprm->mm);
1189 	if (retval < 0)
1190 		goto out_mm;
1191 
1192 	bprm->argc = count(argv, bprm->p / sizeof(void *));
1193 	if ((retval = bprm->argc) < 0)
1194 		goto out_mm;
1195 
1196 	bprm->envc = count(envp, bprm->p / sizeof(void *));
1197 	if ((retval = bprm->envc) < 0)
1198 		goto out_mm;
1199 
1200 	retval = security_bprm_alloc(bprm);
1201 	if (retval)
1202 		goto out;
1203 
1204 	retval = prepare_binprm(bprm);
1205 	if (retval < 0)
1206 		goto out;
1207 
1208 	retval = copy_strings_kernel(1, &bprm->filename, bprm);
1209 	if (retval < 0)
1210 		goto out;
1211 
1212 	bprm->exec = bprm->p;
1213 	retval = copy_strings(bprm->envc, envp, bprm);
1214 	if (retval < 0)
1215 		goto out;
1216 
1217 	retval = copy_strings(bprm->argc, argv, bprm);
1218 	if (retval < 0)
1219 		goto out;
1220 
1221 	retval = search_binary_handler(bprm,regs);
1222 	if (retval >= 0) {
1223 		free_arg_pages(bprm);
1224 
1225 		/* execve success */
1226 		security_bprm_free(bprm);
1227 		acct_update_integrals(current);
1228 		kfree(bprm);
1229 		return retval;
1230 	}
1231 
1232 out:
1233 	/* Something went wrong, return the inode and free the argument pages*/
1234 	for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1235 		struct page * page = bprm->page[i];
1236 		if (page)
1237 			__free_page(page);
1238 	}
1239 
1240 	if (bprm->security)
1241 		security_bprm_free(bprm);
1242 
1243 out_mm:
1244 	if (bprm->mm)
1245 		mmdrop(bprm->mm);
1246 
1247 out_file:
1248 	if (bprm->file) {
1249 		allow_write_access(bprm->file);
1250 		fput(bprm->file);
1251 	}
1252 
1253 out_kfree:
1254 	kfree(bprm);
1255 
1256 out_ret:
1257 	return retval;
1258 }
1259 
1260 int set_binfmt(struct linux_binfmt *new)
1261 {
1262 	struct linux_binfmt *old = current->binfmt;
1263 
1264 	if (new) {
1265 		if (!try_module_get(new->module))
1266 			return -1;
1267 	}
1268 	current->binfmt = new;
1269 	if (old)
1270 		module_put(old->module);
1271 	return 0;
1272 }
1273 
1274 EXPORT_SYMBOL(set_binfmt);
1275 
1276 #define CORENAME_MAX_SIZE 64
1277 
1278 /* format_corename will inspect the pattern parameter, and output a
1279  * name into corename, which must have space for at least
1280  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1281  */
1282 static void format_corename(char *corename, const char *pattern, long signr)
1283 {
1284 	const char *pat_ptr = pattern;
1285 	char *out_ptr = corename;
1286 	char *const out_end = corename + CORENAME_MAX_SIZE;
1287 	int rc;
1288 	int pid_in_pattern = 0;
1289 
1290 	/* Repeat as long as we have more pattern to process and more output
1291 	   space */
1292 	while (*pat_ptr) {
1293 		if (*pat_ptr != '%') {
1294 			if (out_ptr == out_end)
1295 				goto out;
1296 			*out_ptr++ = *pat_ptr++;
1297 		} else {
1298 			switch (*++pat_ptr) {
1299 			case 0:
1300 				goto out;
1301 			/* Double percent, output one percent */
1302 			case '%':
1303 				if (out_ptr == out_end)
1304 					goto out;
1305 				*out_ptr++ = '%';
1306 				break;
1307 			/* pid */
1308 			case 'p':
1309 				pid_in_pattern = 1;
1310 				rc = snprintf(out_ptr, out_end - out_ptr,
1311 					      "%d", current->tgid);
1312 				if (rc > out_end - out_ptr)
1313 					goto out;
1314 				out_ptr += rc;
1315 				break;
1316 			/* uid */
1317 			case 'u':
1318 				rc = snprintf(out_ptr, out_end - out_ptr,
1319 					      "%d", current->uid);
1320 				if (rc > out_end - out_ptr)
1321 					goto out;
1322 				out_ptr += rc;
1323 				break;
1324 			/* gid */
1325 			case 'g':
1326 				rc = snprintf(out_ptr, out_end - out_ptr,
1327 					      "%d", current->gid);
1328 				if (rc > out_end - out_ptr)
1329 					goto out;
1330 				out_ptr += rc;
1331 				break;
1332 			/* signal that caused the coredump */
1333 			case 's':
1334 				rc = snprintf(out_ptr, out_end - out_ptr,
1335 					      "%ld", signr);
1336 				if (rc > out_end - out_ptr)
1337 					goto out;
1338 				out_ptr += rc;
1339 				break;
1340 			/* UNIX time of coredump */
1341 			case 't': {
1342 				struct timeval tv;
1343 				do_gettimeofday(&tv);
1344 				rc = snprintf(out_ptr, out_end - out_ptr,
1345 					      "%lu", tv.tv_sec);
1346 				if (rc > out_end - out_ptr)
1347 					goto out;
1348 				out_ptr += rc;
1349 				break;
1350 			}
1351 			/* hostname */
1352 			case 'h':
1353 				down_read(&uts_sem);
1354 				rc = snprintf(out_ptr, out_end - out_ptr,
1355 					      "%s", system_utsname.nodename);
1356 				up_read(&uts_sem);
1357 				if (rc > out_end - out_ptr)
1358 					goto out;
1359 				out_ptr += rc;
1360 				break;
1361 			/* executable */
1362 			case 'e':
1363 				rc = snprintf(out_ptr, out_end - out_ptr,
1364 					      "%s", current->comm);
1365 				if (rc > out_end - out_ptr)
1366 					goto out;
1367 				out_ptr += rc;
1368 				break;
1369 			default:
1370 				break;
1371 			}
1372 			++pat_ptr;
1373 		}
1374 	}
1375 	/* Backward compatibility with core_uses_pid:
1376 	 *
1377 	 * If core_pattern does not include a %p (as is the default)
1378 	 * and core_uses_pid is set, then .%pid will be appended to
1379 	 * the filename */
1380 	if (!pid_in_pattern
1381             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1382 		rc = snprintf(out_ptr, out_end - out_ptr,
1383 			      ".%d", current->tgid);
1384 		if (rc > out_end - out_ptr)
1385 			goto out;
1386 		out_ptr += rc;
1387 	}
1388       out:
1389 	*out_ptr = 0;
1390 }
1391 
1392 static void zap_threads (struct mm_struct *mm)
1393 {
1394 	struct task_struct *g, *p;
1395 	struct task_struct *tsk = current;
1396 	struct completion *vfork_done = tsk->vfork_done;
1397 	int traced = 0;
1398 
1399 	/*
1400 	 * Make sure nobody is waiting for us to release the VM,
1401 	 * otherwise we can deadlock when we wait on each other
1402 	 */
1403 	if (vfork_done) {
1404 		tsk->vfork_done = NULL;
1405 		complete(vfork_done);
1406 	}
1407 
1408 	read_lock(&tasklist_lock);
1409 	do_each_thread(g,p)
1410 		if (mm == p->mm && p != tsk) {
1411 			force_sig_specific(SIGKILL, p);
1412 			mm->core_waiters++;
1413 			if (unlikely(p->ptrace) &&
1414 			    unlikely(p->parent->mm == mm))
1415 				traced = 1;
1416 		}
1417 	while_each_thread(g,p);
1418 
1419 	read_unlock(&tasklist_lock);
1420 
1421 	if (unlikely(traced)) {
1422 		/*
1423 		 * We are zapping a thread and the thread it ptraces.
1424 		 * If the tracee went into a ptrace stop for exit tracing,
1425 		 * we could deadlock since the tracer is waiting for this
1426 		 * coredump to finish.  Detach them so they can both die.
1427 		 */
1428 		write_lock_irq(&tasklist_lock);
1429 		do_each_thread(g,p) {
1430 			if (mm == p->mm && p != tsk &&
1431 			    p->ptrace && p->parent->mm == mm) {
1432 				__ptrace_detach(p, 0);
1433 			}
1434 		} while_each_thread(g,p);
1435 		write_unlock_irq(&tasklist_lock);
1436 	}
1437 }
1438 
1439 static void coredump_wait(struct mm_struct *mm)
1440 {
1441 	DECLARE_COMPLETION(startup_done);
1442 	int core_waiters;
1443 
1444 	mm->core_startup_done = &startup_done;
1445 
1446 	zap_threads(mm);
1447 	core_waiters = mm->core_waiters;
1448 	up_write(&mm->mmap_sem);
1449 
1450 	if (core_waiters)
1451 		wait_for_completion(&startup_done);
1452 	BUG_ON(mm->core_waiters);
1453 }
1454 
1455 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1456 {
1457 	char corename[CORENAME_MAX_SIZE + 1];
1458 	struct mm_struct *mm = current->mm;
1459 	struct linux_binfmt * binfmt;
1460 	struct inode * inode;
1461 	struct file * file;
1462 	int retval = 0;
1463 	int fsuid = current->fsuid;
1464 	int flag = 0;
1465 
1466 	binfmt = current->binfmt;
1467 	if (!binfmt || !binfmt->core_dump)
1468 		goto fail;
1469 	down_write(&mm->mmap_sem);
1470 	if (!mm->dumpable) {
1471 		up_write(&mm->mmap_sem);
1472 		goto fail;
1473 	}
1474 
1475 	/*
1476 	 *	We cannot trust fsuid as being the "true" uid of the
1477 	 *	process nor do we know its entire history. We only know it
1478 	 *	was tainted so we dump it as root in mode 2.
1479 	 */
1480 	if (mm->dumpable == 2) {	/* Setuid core dump mode */
1481 		flag = O_EXCL;		/* Stop rewrite attacks */
1482 		current->fsuid = 0;	/* Dump root private */
1483 	}
1484 	mm->dumpable = 0;
1485 
1486 	retval = -EAGAIN;
1487 	spin_lock_irq(&current->sighand->siglock);
1488 	if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
1489 		current->signal->flags = SIGNAL_GROUP_EXIT;
1490 		current->signal->group_exit_code = exit_code;
1491 		current->signal->group_stop_count = 0;
1492 		retval = 0;
1493 	}
1494 	spin_unlock_irq(&current->sighand->siglock);
1495 	if (retval) {
1496 		up_write(&mm->mmap_sem);
1497 		goto fail;
1498 	}
1499 
1500 	init_completion(&mm->core_done);
1501 	coredump_wait(mm);
1502 
1503 	/*
1504 	 * Clear any false indication of pending signals that might
1505 	 * be seen by the filesystem code called to write the core file.
1506 	 */
1507 	clear_thread_flag(TIF_SIGPENDING);
1508 
1509 	if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1510 		goto fail_unlock;
1511 
1512 	/*
1513 	 * lock_kernel() because format_corename() is controlled by sysctl, which
1514 	 * uses lock_kernel()
1515 	 */
1516  	lock_kernel();
1517 	format_corename(corename, core_pattern, signr);
1518 	unlock_kernel();
1519 	file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
1520 	if (IS_ERR(file))
1521 		goto fail_unlock;
1522 	inode = file->f_dentry->d_inode;
1523 	if (inode->i_nlink > 1)
1524 		goto close_fail;	/* multiple links - don't dump */
1525 	if (d_unhashed(file->f_dentry))
1526 		goto close_fail;
1527 
1528 	if (!S_ISREG(inode->i_mode))
1529 		goto close_fail;
1530 	if (!file->f_op)
1531 		goto close_fail;
1532 	if (!file->f_op->write)
1533 		goto close_fail;
1534 	if (do_truncate(file->f_dentry, 0, 0, file) != 0)
1535 		goto close_fail;
1536 
1537 	retval = binfmt->core_dump(signr, regs, file);
1538 
1539 	if (retval)
1540 		current->signal->group_exit_code |= 0x80;
1541 close_fail:
1542 	filp_close(file, NULL);
1543 fail_unlock:
1544 	current->fsuid = fsuid;
1545 	complete_all(&mm->core_done);
1546 fail:
1547 	return retval;
1548 }
1549