xref: /linux/fs/exec.c (revision 50069a5851323ba5def0e414a21e234345016870)
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/fdtable.h>
28 #include <linux/mm.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/swap.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/perf_event.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.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/mount.h>
46 #include <linux/security.h>
47 #include <linux/syscalls.h>
48 #include <linux/tsacct_kern.h>
49 #include <linux/cn_proc.h>
50 #include <linux/audit.h>
51 #include <linux/tracehook.h>
52 #include <linux/kmod.h>
53 #include <linux/fsnotify.h>
54 #include <linux/fs_struct.h>
55 #include <linux/pipe_fs_i.h>
56 #include <linux/oom.h>
57 #include <linux/compat.h>
58 
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
61 #include <asm/tlb.h>
62 #include <asm/exec.h>
63 
64 #include <trace/events/task.h>
65 #include "internal.h"
66 
67 #include <trace/events/sched.h>
68 
69 int core_uses_pid;
70 char core_pattern[CORENAME_MAX_SIZE] = "core";
71 unsigned int core_pipe_limit;
72 int suid_dumpable = 0;
73 
74 struct core_name {
75 	char *corename;
76 	int used, size;
77 };
78 static atomic_t call_count = ATOMIC_INIT(1);
79 
80 /* The maximal length of core_pattern is also specified in sysctl.c */
81 
82 static LIST_HEAD(formats);
83 static DEFINE_RWLOCK(binfmt_lock);
84 
85 void __register_binfmt(struct linux_binfmt * fmt, int insert)
86 {
87 	BUG_ON(!fmt);
88 	write_lock(&binfmt_lock);
89 	insert ? list_add(&fmt->lh, &formats) :
90 		 list_add_tail(&fmt->lh, &formats);
91 	write_unlock(&binfmt_lock);
92 }
93 
94 EXPORT_SYMBOL(__register_binfmt);
95 
96 void unregister_binfmt(struct linux_binfmt * fmt)
97 {
98 	write_lock(&binfmt_lock);
99 	list_del(&fmt->lh);
100 	write_unlock(&binfmt_lock);
101 }
102 
103 EXPORT_SYMBOL(unregister_binfmt);
104 
105 static inline void put_binfmt(struct linux_binfmt * fmt)
106 {
107 	module_put(fmt->module);
108 }
109 
110 /*
111  * Note that a shared library must be both readable and executable due to
112  * security reasons.
113  *
114  * Also note that we take the address to load from from the file itself.
115  */
116 SYSCALL_DEFINE1(uselib, const char __user *, library)
117 {
118 	struct file *file;
119 	char *tmp = getname(library);
120 	int error = PTR_ERR(tmp);
121 	static const struct open_flags uselib_flags = {
122 		.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
123 		.acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
124 		.intent = LOOKUP_OPEN
125 	};
126 
127 	if (IS_ERR(tmp))
128 		goto out;
129 
130 	file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
131 	putname(tmp);
132 	error = PTR_ERR(file);
133 	if (IS_ERR(file))
134 		goto out;
135 
136 	error = -EINVAL;
137 	if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
138 		goto exit;
139 
140 	error = -EACCES;
141 	if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
142 		goto exit;
143 
144 	fsnotify_open(file);
145 
146 	error = -ENOEXEC;
147 	if(file->f_op) {
148 		struct linux_binfmt * fmt;
149 
150 		read_lock(&binfmt_lock);
151 		list_for_each_entry(fmt, &formats, lh) {
152 			if (!fmt->load_shlib)
153 				continue;
154 			if (!try_module_get(fmt->module))
155 				continue;
156 			read_unlock(&binfmt_lock);
157 			error = fmt->load_shlib(file);
158 			read_lock(&binfmt_lock);
159 			put_binfmt(fmt);
160 			if (error != -ENOEXEC)
161 				break;
162 		}
163 		read_unlock(&binfmt_lock);
164 	}
165 exit:
166 	fput(file);
167 out:
168   	return error;
169 }
170 
171 #ifdef CONFIG_MMU
172 /*
173  * The nascent bprm->mm is not visible until exec_mmap() but it can
174  * use a lot of memory, account these pages in current->mm temporary
175  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
176  * change the counter back via acct_arg_size(0).
177  */
178 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
179 {
180 	struct mm_struct *mm = current->mm;
181 	long diff = (long)(pages - bprm->vma_pages);
182 
183 	if (!mm || !diff)
184 		return;
185 
186 	bprm->vma_pages = pages;
187 	add_mm_counter(mm, MM_ANONPAGES, diff);
188 }
189 
190 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
191 		int write)
192 {
193 	struct page *page;
194 	int ret;
195 
196 #ifdef CONFIG_STACK_GROWSUP
197 	if (write) {
198 		ret = expand_downwards(bprm->vma, pos);
199 		if (ret < 0)
200 			return NULL;
201 	}
202 #endif
203 	ret = get_user_pages(current, bprm->mm, pos,
204 			1, write, 1, &page, NULL);
205 	if (ret <= 0)
206 		return NULL;
207 
208 	if (write) {
209 		unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
210 		struct rlimit *rlim;
211 
212 		acct_arg_size(bprm, size / PAGE_SIZE);
213 
214 		/*
215 		 * We've historically supported up to 32 pages (ARG_MAX)
216 		 * of argument strings even with small stacks
217 		 */
218 		if (size <= ARG_MAX)
219 			return page;
220 
221 		/*
222 		 * Limit to 1/4-th the stack size for the argv+env strings.
223 		 * This ensures that:
224 		 *  - the remaining binfmt code will not run out of stack space,
225 		 *  - the program will have a reasonable amount of stack left
226 		 *    to work from.
227 		 */
228 		rlim = current->signal->rlim;
229 		if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
230 			put_page(page);
231 			return NULL;
232 		}
233 	}
234 
235 	return page;
236 }
237 
238 static void put_arg_page(struct page *page)
239 {
240 	put_page(page);
241 }
242 
243 static void free_arg_page(struct linux_binprm *bprm, int i)
244 {
245 }
246 
247 static void free_arg_pages(struct linux_binprm *bprm)
248 {
249 }
250 
251 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
252 		struct page *page)
253 {
254 	flush_cache_page(bprm->vma, pos, page_to_pfn(page));
255 }
256 
257 static int __bprm_mm_init(struct linux_binprm *bprm)
258 {
259 	int err;
260 	struct vm_area_struct *vma = NULL;
261 	struct mm_struct *mm = bprm->mm;
262 
263 	bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
264 	if (!vma)
265 		return -ENOMEM;
266 
267 	down_write(&mm->mmap_sem);
268 	vma->vm_mm = mm;
269 
270 	/*
271 	 * Place the stack at the largest stack address the architecture
272 	 * supports. Later, we'll move this to an appropriate place. We don't
273 	 * use STACK_TOP because that can depend on attributes which aren't
274 	 * configured yet.
275 	 */
276 	BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
277 	vma->vm_end = STACK_TOP_MAX;
278 	vma->vm_start = vma->vm_end - PAGE_SIZE;
279 	vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
280 	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
281 	INIT_LIST_HEAD(&vma->anon_vma_chain);
282 
283 	err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
284 	if (err)
285 		goto err;
286 
287 	err = insert_vm_struct(mm, vma);
288 	if (err)
289 		goto err;
290 
291 	mm->stack_vm = mm->total_vm = 1;
292 	up_write(&mm->mmap_sem);
293 	bprm->p = vma->vm_end - sizeof(void *);
294 	return 0;
295 err:
296 	up_write(&mm->mmap_sem);
297 	bprm->vma = NULL;
298 	kmem_cache_free(vm_area_cachep, vma);
299 	return err;
300 }
301 
302 static bool valid_arg_len(struct linux_binprm *bprm, long len)
303 {
304 	return len <= MAX_ARG_STRLEN;
305 }
306 
307 #else
308 
309 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
310 {
311 }
312 
313 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
314 		int write)
315 {
316 	struct page *page;
317 
318 	page = bprm->page[pos / PAGE_SIZE];
319 	if (!page && write) {
320 		page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
321 		if (!page)
322 			return NULL;
323 		bprm->page[pos / PAGE_SIZE] = page;
324 	}
325 
326 	return page;
327 }
328 
329 static void put_arg_page(struct page *page)
330 {
331 }
332 
333 static void free_arg_page(struct linux_binprm *bprm, int i)
334 {
335 	if (bprm->page[i]) {
336 		__free_page(bprm->page[i]);
337 		bprm->page[i] = NULL;
338 	}
339 }
340 
341 static void free_arg_pages(struct linux_binprm *bprm)
342 {
343 	int i;
344 
345 	for (i = 0; i < MAX_ARG_PAGES; i++)
346 		free_arg_page(bprm, i);
347 }
348 
349 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
350 		struct page *page)
351 {
352 }
353 
354 static int __bprm_mm_init(struct linux_binprm *bprm)
355 {
356 	bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
357 	return 0;
358 }
359 
360 static bool valid_arg_len(struct linux_binprm *bprm, long len)
361 {
362 	return len <= bprm->p;
363 }
364 
365 #endif /* CONFIG_MMU */
366 
367 /*
368  * Create a new mm_struct and populate it with a temporary stack
369  * vm_area_struct.  We don't have enough context at this point to set the stack
370  * flags, permissions, and offset, so we use temporary values.  We'll update
371  * them later in setup_arg_pages().
372  */
373 int bprm_mm_init(struct linux_binprm *bprm)
374 {
375 	int err;
376 	struct mm_struct *mm = NULL;
377 
378 	bprm->mm = mm = mm_alloc();
379 	err = -ENOMEM;
380 	if (!mm)
381 		goto err;
382 
383 	err = init_new_context(current, mm);
384 	if (err)
385 		goto err;
386 
387 	err = __bprm_mm_init(bprm);
388 	if (err)
389 		goto err;
390 
391 	return 0;
392 
393 err:
394 	if (mm) {
395 		bprm->mm = NULL;
396 		mmdrop(mm);
397 	}
398 
399 	return err;
400 }
401 
402 struct user_arg_ptr {
403 #ifdef CONFIG_COMPAT
404 	bool is_compat;
405 #endif
406 	union {
407 		const char __user *const __user *native;
408 #ifdef CONFIG_COMPAT
409 		compat_uptr_t __user *compat;
410 #endif
411 	} ptr;
412 };
413 
414 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
415 {
416 	const char __user *native;
417 
418 #ifdef CONFIG_COMPAT
419 	if (unlikely(argv.is_compat)) {
420 		compat_uptr_t compat;
421 
422 		if (get_user(compat, argv.ptr.compat + nr))
423 			return ERR_PTR(-EFAULT);
424 
425 		return compat_ptr(compat);
426 	}
427 #endif
428 
429 	if (get_user(native, argv.ptr.native + nr))
430 		return ERR_PTR(-EFAULT);
431 
432 	return native;
433 }
434 
435 /*
436  * count() counts the number of strings in array ARGV.
437  */
438 static int count(struct user_arg_ptr argv, int max)
439 {
440 	int i = 0;
441 
442 	if (argv.ptr.native != NULL) {
443 		for (;;) {
444 			const char __user *p = get_user_arg_ptr(argv, i);
445 
446 			if (!p)
447 				break;
448 
449 			if (IS_ERR(p))
450 				return -EFAULT;
451 
452 			if (i++ >= max)
453 				return -E2BIG;
454 
455 			if (fatal_signal_pending(current))
456 				return -ERESTARTNOHAND;
457 			cond_resched();
458 		}
459 	}
460 	return i;
461 }
462 
463 /*
464  * 'copy_strings()' copies argument/environment strings from the old
465  * processes's memory to the new process's stack.  The call to get_user_pages()
466  * ensures the destination page is created and not swapped out.
467  */
468 static int copy_strings(int argc, struct user_arg_ptr argv,
469 			struct linux_binprm *bprm)
470 {
471 	struct page *kmapped_page = NULL;
472 	char *kaddr = NULL;
473 	unsigned long kpos = 0;
474 	int ret;
475 
476 	while (argc-- > 0) {
477 		const char __user *str;
478 		int len;
479 		unsigned long pos;
480 
481 		ret = -EFAULT;
482 		str = get_user_arg_ptr(argv, argc);
483 		if (IS_ERR(str))
484 			goto out;
485 
486 		len = strnlen_user(str, MAX_ARG_STRLEN);
487 		if (!len)
488 			goto out;
489 
490 		ret = -E2BIG;
491 		if (!valid_arg_len(bprm, len))
492 			goto out;
493 
494 		/* We're going to work our way backwords. */
495 		pos = bprm->p;
496 		str += len;
497 		bprm->p -= len;
498 
499 		while (len > 0) {
500 			int offset, bytes_to_copy;
501 
502 			if (fatal_signal_pending(current)) {
503 				ret = -ERESTARTNOHAND;
504 				goto out;
505 			}
506 			cond_resched();
507 
508 			offset = pos % PAGE_SIZE;
509 			if (offset == 0)
510 				offset = PAGE_SIZE;
511 
512 			bytes_to_copy = offset;
513 			if (bytes_to_copy > len)
514 				bytes_to_copy = len;
515 
516 			offset -= bytes_to_copy;
517 			pos -= bytes_to_copy;
518 			str -= bytes_to_copy;
519 			len -= bytes_to_copy;
520 
521 			if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
522 				struct page *page;
523 
524 				page = get_arg_page(bprm, pos, 1);
525 				if (!page) {
526 					ret = -E2BIG;
527 					goto out;
528 				}
529 
530 				if (kmapped_page) {
531 					flush_kernel_dcache_page(kmapped_page);
532 					kunmap(kmapped_page);
533 					put_arg_page(kmapped_page);
534 				}
535 				kmapped_page = page;
536 				kaddr = kmap(kmapped_page);
537 				kpos = pos & PAGE_MASK;
538 				flush_arg_page(bprm, kpos, kmapped_page);
539 			}
540 			if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
541 				ret = -EFAULT;
542 				goto out;
543 			}
544 		}
545 	}
546 	ret = 0;
547 out:
548 	if (kmapped_page) {
549 		flush_kernel_dcache_page(kmapped_page);
550 		kunmap(kmapped_page);
551 		put_arg_page(kmapped_page);
552 	}
553 	return ret;
554 }
555 
556 /*
557  * Like copy_strings, but get argv and its values from kernel memory.
558  */
559 int copy_strings_kernel(int argc, const char *const *__argv,
560 			struct linux_binprm *bprm)
561 {
562 	int r;
563 	mm_segment_t oldfs = get_fs();
564 	struct user_arg_ptr argv = {
565 		.ptr.native = (const char __user *const  __user *)__argv,
566 	};
567 
568 	set_fs(KERNEL_DS);
569 	r = copy_strings(argc, argv, bprm);
570 	set_fs(oldfs);
571 
572 	return r;
573 }
574 EXPORT_SYMBOL(copy_strings_kernel);
575 
576 #ifdef CONFIG_MMU
577 
578 /*
579  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
580  * the binfmt code determines where the new stack should reside, we shift it to
581  * its final location.  The process proceeds as follows:
582  *
583  * 1) Use shift to calculate the new vma endpoints.
584  * 2) Extend vma to cover both the old and new ranges.  This ensures the
585  *    arguments passed to subsequent functions are consistent.
586  * 3) Move vma's page tables to the new range.
587  * 4) Free up any cleared pgd range.
588  * 5) Shrink the vma to cover only the new range.
589  */
590 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
591 {
592 	struct mm_struct *mm = vma->vm_mm;
593 	unsigned long old_start = vma->vm_start;
594 	unsigned long old_end = vma->vm_end;
595 	unsigned long length = old_end - old_start;
596 	unsigned long new_start = old_start - shift;
597 	unsigned long new_end = old_end - shift;
598 	struct mmu_gather tlb;
599 
600 	BUG_ON(new_start > new_end);
601 
602 	/*
603 	 * ensure there are no vmas between where we want to go
604 	 * and where we are
605 	 */
606 	if (vma != find_vma(mm, new_start))
607 		return -EFAULT;
608 
609 	/*
610 	 * cover the whole range: [new_start, old_end)
611 	 */
612 	if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
613 		return -ENOMEM;
614 
615 	/*
616 	 * move the page tables downwards, on failure we rely on
617 	 * process cleanup to remove whatever mess we made.
618 	 */
619 	if (length != move_page_tables(vma, old_start,
620 				       vma, new_start, length))
621 		return -ENOMEM;
622 
623 	lru_add_drain();
624 	tlb_gather_mmu(&tlb, mm, 0);
625 	if (new_end > old_start) {
626 		/*
627 		 * when the old and new regions overlap clear from new_end.
628 		 */
629 		free_pgd_range(&tlb, new_end, old_end, new_end,
630 			vma->vm_next ? vma->vm_next->vm_start : 0);
631 	} else {
632 		/*
633 		 * otherwise, clean from old_start; this is done to not touch
634 		 * the address space in [new_end, old_start) some architectures
635 		 * have constraints on va-space that make this illegal (IA64) -
636 		 * for the others its just a little faster.
637 		 */
638 		free_pgd_range(&tlb, old_start, old_end, new_end,
639 			vma->vm_next ? vma->vm_next->vm_start : 0);
640 	}
641 	tlb_finish_mmu(&tlb, new_end, old_end);
642 
643 	/*
644 	 * Shrink the vma to just the new range.  Always succeeds.
645 	 */
646 	vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
647 
648 	return 0;
649 }
650 
651 /*
652  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
653  * the stack is optionally relocated, and some extra space is added.
654  */
655 int setup_arg_pages(struct linux_binprm *bprm,
656 		    unsigned long stack_top,
657 		    int executable_stack)
658 {
659 	unsigned long ret;
660 	unsigned long stack_shift;
661 	struct mm_struct *mm = current->mm;
662 	struct vm_area_struct *vma = bprm->vma;
663 	struct vm_area_struct *prev = NULL;
664 	unsigned long vm_flags;
665 	unsigned long stack_base;
666 	unsigned long stack_size;
667 	unsigned long stack_expand;
668 	unsigned long rlim_stack;
669 
670 #ifdef CONFIG_STACK_GROWSUP
671 	/* Limit stack size to 1GB */
672 	stack_base = rlimit_max(RLIMIT_STACK);
673 	if (stack_base > (1 << 30))
674 		stack_base = 1 << 30;
675 
676 	/* Make sure we didn't let the argument array grow too large. */
677 	if (vma->vm_end - vma->vm_start > stack_base)
678 		return -ENOMEM;
679 
680 	stack_base = PAGE_ALIGN(stack_top - stack_base);
681 
682 	stack_shift = vma->vm_start - stack_base;
683 	mm->arg_start = bprm->p - stack_shift;
684 	bprm->p = vma->vm_end - stack_shift;
685 #else
686 	stack_top = arch_align_stack(stack_top);
687 	stack_top = PAGE_ALIGN(stack_top);
688 
689 	if (unlikely(stack_top < mmap_min_addr) ||
690 	    unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
691 		return -ENOMEM;
692 
693 	stack_shift = vma->vm_end - stack_top;
694 
695 	bprm->p -= stack_shift;
696 	mm->arg_start = bprm->p;
697 #endif
698 
699 	if (bprm->loader)
700 		bprm->loader -= stack_shift;
701 	bprm->exec -= stack_shift;
702 
703 	down_write(&mm->mmap_sem);
704 	vm_flags = VM_STACK_FLAGS;
705 
706 	/*
707 	 * Adjust stack execute permissions; explicitly enable for
708 	 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
709 	 * (arch default) otherwise.
710 	 */
711 	if (unlikely(executable_stack == EXSTACK_ENABLE_X))
712 		vm_flags |= VM_EXEC;
713 	else if (executable_stack == EXSTACK_DISABLE_X)
714 		vm_flags &= ~VM_EXEC;
715 	vm_flags |= mm->def_flags;
716 	vm_flags |= VM_STACK_INCOMPLETE_SETUP;
717 
718 	ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
719 			vm_flags);
720 	if (ret)
721 		goto out_unlock;
722 	BUG_ON(prev != vma);
723 
724 	/* Move stack pages down in memory. */
725 	if (stack_shift) {
726 		ret = shift_arg_pages(vma, stack_shift);
727 		if (ret)
728 			goto out_unlock;
729 	}
730 
731 	/* mprotect_fixup is overkill to remove the temporary stack flags */
732 	vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
733 
734 	stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
735 	stack_size = vma->vm_end - vma->vm_start;
736 	/*
737 	 * Align this down to a page boundary as expand_stack
738 	 * will align it up.
739 	 */
740 	rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
741 #ifdef CONFIG_STACK_GROWSUP
742 	if (stack_size + stack_expand > rlim_stack)
743 		stack_base = vma->vm_start + rlim_stack;
744 	else
745 		stack_base = vma->vm_end + stack_expand;
746 #else
747 	if (stack_size + stack_expand > rlim_stack)
748 		stack_base = vma->vm_end - rlim_stack;
749 	else
750 		stack_base = vma->vm_start - stack_expand;
751 #endif
752 	current->mm->start_stack = bprm->p;
753 	ret = expand_stack(vma, stack_base);
754 	if (ret)
755 		ret = -EFAULT;
756 
757 out_unlock:
758 	up_write(&mm->mmap_sem);
759 	return ret;
760 }
761 EXPORT_SYMBOL(setup_arg_pages);
762 
763 #endif /* CONFIG_MMU */
764 
765 struct file *open_exec(const char *name)
766 {
767 	struct file *file;
768 	int err;
769 	static const struct open_flags open_exec_flags = {
770 		.open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
771 		.acc_mode = MAY_EXEC | MAY_OPEN,
772 		.intent = LOOKUP_OPEN
773 	};
774 
775 	file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
776 	if (IS_ERR(file))
777 		goto out;
778 
779 	err = -EACCES;
780 	if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
781 		goto exit;
782 
783 	if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
784 		goto exit;
785 
786 	fsnotify_open(file);
787 
788 	err = deny_write_access(file);
789 	if (err)
790 		goto exit;
791 
792 out:
793 	return file;
794 
795 exit:
796 	fput(file);
797 	return ERR_PTR(err);
798 }
799 EXPORT_SYMBOL(open_exec);
800 
801 int kernel_read(struct file *file, loff_t offset,
802 		char *addr, unsigned long count)
803 {
804 	mm_segment_t old_fs;
805 	loff_t pos = offset;
806 	int result;
807 
808 	old_fs = get_fs();
809 	set_fs(get_ds());
810 	/* The cast to a user pointer is valid due to the set_fs() */
811 	result = vfs_read(file, (void __user *)addr, count, &pos);
812 	set_fs(old_fs);
813 	return result;
814 }
815 
816 EXPORT_SYMBOL(kernel_read);
817 
818 static int exec_mmap(struct mm_struct *mm)
819 {
820 	struct task_struct *tsk;
821 	struct mm_struct * old_mm, *active_mm;
822 
823 	/* Notify parent that we're no longer interested in the old VM */
824 	tsk = current;
825 	old_mm = current->mm;
826 	sync_mm_rss(old_mm);
827 	mm_release(tsk, old_mm);
828 
829 	if (old_mm) {
830 		/*
831 		 * Make sure that if there is a core dump in progress
832 		 * for the old mm, we get out and die instead of going
833 		 * through with the exec.  We must hold mmap_sem around
834 		 * checking core_state and changing tsk->mm.
835 		 */
836 		down_read(&old_mm->mmap_sem);
837 		if (unlikely(old_mm->core_state)) {
838 			up_read(&old_mm->mmap_sem);
839 			return -EINTR;
840 		}
841 	}
842 	task_lock(tsk);
843 	active_mm = tsk->active_mm;
844 	tsk->mm = mm;
845 	tsk->active_mm = mm;
846 	activate_mm(active_mm, mm);
847 	task_unlock(tsk);
848 	arch_pick_mmap_layout(mm);
849 	if (old_mm) {
850 		up_read(&old_mm->mmap_sem);
851 		BUG_ON(active_mm != old_mm);
852 		setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
853 		mm_update_next_owner(old_mm);
854 		mmput(old_mm);
855 		return 0;
856 	}
857 	mmdrop(active_mm);
858 	return 0;
859 }
860 
861 /*
862  * This function makes sure the current process has its own signal table,
863  * so that flush_signal_handlers can later reset the handlers without
864  * disturbing other processes.  (Other processes might share the signal
865  * table via the CLONE_SIGHAND option to clone().)
866  */
867 static int de_thread(struct task_struct *tsk)
868 {
869 	struct signal_struct *sig = tsk->signal;
870 	struct sighand_struct *oldsighand = tsk->sighand;
871 	spinlock_t *lock = &oldsighand->siglock;
872 
873 	if (thread_group_empty(tsk))
874 		goto no_thread_group;
875 
876 	/*
877 	 * Kill all other threads in the thread group.
878 	 */
879 	spin_lock_irq(lock);
880 	if (signal_group_exit(sig)) {
881 		/*
882 		 * Another group action in progress, just
883 		 * return so that the signal is processed.
884 		 */
885 		spin_unlock_irq(lock);
886 		return -EAGAIN;
887 	}
888 
889 	sig->group_exit_task = tsk;
890 	sig->notify_count = zap_other_threads(tsk);
891 	if (!thread_group_leader(tsk))
892 		sig->notify_count--;
893 
894 	while (sig->notify_count) {
895 		__set_current_state(TASK_UNINTERRUPTIBLE);
896 		spin_unlock_irq(lock);
897 		schedule();
898 		spin_lock_irq(lock);
899 	}
900 	spin_unlock_irq(lock);
901 
902 	/*
903 	 * At this point all other threads have exited, all we have to
904 	 * do is to wait for the thread group leader to become inactive,
905 	 * and to assume its PID:
906 	 */
907 	if (!thread_group_leader(tsk)) {
908 		struct task_struct *leader = tsk->group_leader;
909 
910 		sig->notify_count = -1;	/* for exit_notify() */
911 		for (;;) {
912 			write_lock_irq(&tasklist_lock);
913 			if (likely(leader->exit_state))
914 				break;
915 			__set_current_state(TASK_UNINTERRUPTIBLE);
916 			write_unlock_irq(&tasklist_lock);
917 			schedule();
918 		}
919 
920 		/*
921 		 * The only record we have of the real-time age of a
922 		 * process, regardless of execs it's done, is start_time.
923 		 * All the past CPU time is accumulated in signal_struct
924 		 * from sister threads now dead.  But in this non-leader
925 		 * exec, nothing survives from the original leader thread,
926 		 * whose birth marks the true age of this process now.
927 		 * When we take on its identity by switching to its PID, we
928 		 * also take its birthdate (always earlier than our own).
929 		 */
930 		tsk->start_time = leader->start_time;
931 
932 		BUG_ON(!same_thread_group(leader, tsk));
933 		BUG_ON(has_group_leader_pid(tsk));
934 		/*
935 		 * An exec() starts a new thread group with the
936 		 * TGID of the previous thread group. Rehash the
937 		 * two threads with a switched PID, and release
938 		 * the former thread group leader:
939 		 */
940 
941 		/* Become a process group leader with the old leader's pid.
942 		 * The old leader becomes a thread of the this thread group.
943 		 * Note: The old leader also uses this pid until release_task
944 		 *       is called.  Odd but simple and correct.
945 		 */
946 		detach_pid(tsk, PIDTYPE_PID);
947 		tsk->pid = leader->pid;
948 		attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
949 		transfer_pid(leader, tsk, PIDTYPE_PGID);
950 		transfer_pid(leader, tsk, PIDTYPE_SID);
951 
952 		list_replace_rcu(&leader->tasks, &tsk->tasks);
953 		list_replace_init(&leader->sibling, &tsk->sibling);
954 
955 		tsk->group_leader = tsk;
956 		leader->group_leader = tsk;
957 
958 		tsk->exit_signal = SIGCHLD;
959 		leader->exit_signal = -1;
960 
961 		BUG_ON(leader->exit_state != EXIT_ZOMBIE);
962 		leader->exit_state = EXIT_DEAD;
963 
964 		/*
965 		 * We are going to release_task()->ptrace_unlink() silently,
966 		 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
967 		 * the tracer wont't block again waiting for this thread.
968 		 */
969 		if (unlikely(leader->ptrace))
970 			__wake_up_parent(leader, leader->parent);
971 		write_unlock_irq(&tasklist_lock);
972 
973 		release_task(leader);
974 	}
975 
976 	sig->group_exit_task = NULL;
977 	sig->notify_count = 0;
978 
979 no_thread_group:
980 	/* we have changed execution domain */
981 	tsk->exit_signal = SIGCHLD;
982 
983 	exit_itimers(sig);
984 	flush_itimer_signals();
985 
986 	if (atomic_read(&oldsighand->count) != 1) {
987 		struct sighand_struct *newsighand;
988 		/*
989 		 * This ->sighand is shared with the CLONE_SIGHAND
990 		 * but not CLONE_THREAD task, switch to the new one.
991 		 */
992 		newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
993 		if (!newsighand)
994 			return -ENOMEM;
995 
996 		atomic_set(&newsighand->count, 1);
997 		memcpy(newsighand->action, oldsighand->action,
998 		       sizeof(newsighand->action));
999 
1000 		write_lock_irq(&tasklist_lock);
1001 		spin_lock(&oldsighand->siglock);
1002 		rcu_assign_pointer(tsk->sighand, newsighand);
1003 		spin_unlock(&oldsighand->siglock);
1004 		write_unlock_irq(&tasklist_lock);
1005 
1006 		__cleanup_sighand(oldsighand);
1007 	}
1008 
1009 	BUG_ON(!thread_group_leader(tsk));
1010 	return 0;
1011 }
1012 
1013 /*
1014  * These functions flushes out all traces of the currently running executable
1015  * so that a new one can be started
1016  */
1017 static void flush_old_files(struct files_struct * files)
1018 {
1019 	long j = -1;
1020 	struct fdtable *fdt;
1021 
1022 	spin_lock(&files->file_lock);
1023 	for (;;) {
1024 		unsigned long set, i;
1025 
1026 		j++;
1027 		i = j * __NFDBITS;
1028 		fdt = files_fdtable(files);
1029 		if (i >= fdt->max_fds)
1030 			break;
1031 		set = fdt->close_on_exec[j];
1032 		if (!set)
1033 			continue;
1034 		fdt->close_on_exec[j] = 0;
1035 		spin_unlock(&files->file_lock);
1036 		for ( ; set ; i++,set >>= 1) {
1037 			if (set & 1) {
1038 				sys_close(i);
1039 			}
1040 		}
1041 		spin_lock(&files->file_lock);
1042 
1043 	}
1044 	spin_unlock(&files->file_lock);
1045 }
1046 
1047 char *get_task_comm(char *buf, struct task_struct *tsk)
1048 {
1049 	/* buf must be at least sizeof(tsk->comm) in size */
1050 	task_lock(tsk);
1051 	strncpy(buf, tsk->comm, sizeof(tsk->comm));
1052 	task_unlock(tsk);
1053 	return buf;
1054 }
1055 EXPORT_SYMBOL_GPL(get_task_comm);
1056 
1057 void set_task_comm(struct task_struct *tsk, char *buf)
1058 {
1059 	task_lock(tsk);
1060 
1061 	trace_task_rename(tsk, buf);
1062 
1063 	/*
1064 	 * Threads may access current->comm without holding
1065 	 * the task lock, so write the string carefully.
1066 	 * Readers without a lock may see incomplete new
1067 	 * names but are safe from non-terminating string reads.
1068 	 */
1069 	memset(tsk->comm, 0, TASK_COMM_LEN);
1070 	wmb();
1071 	strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1072 	task_unlock(tsk);
1073 	perf_event_comm(tsk);
1074 }
1075 
1076 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1077 {
1078 	int i, ch;
1079 
1080 	/* Copies the binary name from after last slash */
1081 	for (i = 0; (ch = *(fn++)) != '\0';) {
1082 		if (ch == '/')
1083 			i = 0; /* overwrite what we wrote */
1084 		else
1085 			if (i < len - 1)
1086 				tcomm[i++] = ch;
1087 	}
1088 	tcomm[i] = '\0';
1089 }
1090 
1091 int flush_old_exec(struct linux_binprm * bprm)
1092 {
1093 	int retval;
1094 
1095 	/*
1096 	 * Make sure we have a private signal table and that
1097 	 * we are unassociated from the previous thread group.
1098 	 */
1099 	retval = de_thread(current);
1100 	if (retval)
1101 		goto out;
1102 
1103 	set_mm_exe_file(bprm->mm, bprm->file);
1104 
1105 	filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1106 	/*
1107 	 * Release all of the old mmap stuff
1108 	 */
1109 	acct_arg_size(bprm, 0);
1110 	retval = exec_mmap(bprm->mm);
1111 	if (retval)
1112 		goto out;
1113 
1114 	bprm->mm = NULL;		/* We're using it now */
1115 
1116 	set_fs(USER_DS);
1117 	current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD);
1118 	flush_thread();
1119 	current->personality &= ~bprm->per_clear;
1120 
1121 	return 0;
1122 
1123 out:
1124 	return retval;
1125 }
1126 EXPORT_SYMBOL(flush_old_exec);
1127 
1128 void would_dump(struct linux_binprm *bprm, struct file *file)
1129 {
1130 	if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1131 		bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1132 }
1133 EXPORT_SYMBOL(would_dump);
1134 
1135 void setup_new_exec(struct linux_binprm * bprm)
1136 {
1137 	arch_pick_mmap_layout(current->mm);
1138 
1139 	/* This is the point of no return */
1140 	current->sas_ss_sp = current->sas_ss_size = 0;
1141 
1142 	if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1143 		set_dumpable(current->mm, 1);
1144 	else
1145 		set_dumpable(current->mm, suid_dumpable);
1146 
1147 	set_task_comm(current, bprm->tcomm);
1148 
1149 	/* Set the new mm task size. We have to do that late because it may
1150 	 * depend on TIF_32BIT which is only updated in flush_thread() on
1151 	 * some architectures like powerpc
1152 	 */
1153 	current->mm->task_size = TASK_SIZE;
1154 
1155 	/* install the new credentials */
1156 	if (!uid_eq(bprm->cred->uid, current_euid()) ||
1157 	    !gid_eq(bprm->cred->gid, current_egid())) {
1158 		current->pdeath_signal = 0;
1159 	} else {
1160 		would_dump(bprm, bprm->file);
1161 		if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1162 			set_dumpable(current->mm, suid_dumpable);
1163 	}
1164 
1165 	/*
1166 	 * Flush performance counters when crossing a
1167 	 * security domain:
1168 	 */
1169 	if (!get_dumpable(current->mm))
1170 		perf_event_exit_task(current);
1171 
1172 	/* An exec changes our domain. We are no longer part of the thread
1173 	   group */
1174 
1175 	current->self_exec_id++;
1176 
1177 	flush_signal_handlers(current, 0);
1178 	flush_old_files(current->files);
1179 }
1180 EXPORT_SYMBOL(setup_new_exec);
1181 
1182 /*
1183  * Prepare credentials and lock ->cred_guard_mutex.
1184  * install_exec_creds() commits the new creds and drops the lock.
1185  * Or, if exec fails before, free_bprm() should release ->cred and
1186  * and unlock.
1187  */
1188 int prepare_bprm_creds(struct linux_binprm *bprm)
1189 {
1190 	if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1191 		return -ERESTARTNOINTR;
1192 
1193 	bprm->cred = prepare_exec_creds();
1194 	if (likely(bprm->cred))
1195 		return 0;
1196 
1197 	mutex_unlock(&current->signal->cred_guard_mutex);
1198 	return -ENOMEM;
1199 }
1200 
1201 void free_bprm(struct linux_binprm *bprm)
1202 {
1203 	free_arg_pages(bprm);
1204 	if (bprm->cred) {
1205 		mutex_unlock(&current->signal->cred_guard_mutex);
1206 		abort_creds(bprm->cred);
1207 	}
1208 	kfree(bprm);
1209 }
1210 
1211 /*
1212  * install the new credentials for this executable
1213  */
1214 void install_exec_creds(struct linux_binprm *bprm)
1215 {
1216 	security_bprm_committing_creds(bprm);
1217 
1218 	commit_creds(bprm->cred);
1219 	bprm->cred = NULL;
1220 	/*
1221 	 * cred_guard_mutex must be held at least to this point to prevent
1222 	 * ptrace_attach() from altering our determination of the task's
1223 	 * credentials; any time after this it may be unlocked.
1224 	 */
1225 	security_bprm_committed_creds(bprm);
1226 	mutex_unlock(&current->signal->cred_guard_mutex);
1227 }
1228 EXPORT_SYMBOL(install_exec_creds);
1229 
1230 /*
1231  * determine how safe it is to execute the proposed program
1232  * - the caller must hold ->cred_guard_mutex to protect against
1233  *   PTRACE_ATTACH
1234  */
1235 static int check_unsafe_exec(struct linux_binprm *bprm)
1236 {
1237 	struct task_struct *p = current, *t;
1238 	unsigned n_fs;
1239 	int res = 0;
1240 
1241 	if (p->ptrace) {
1242 		if (p->ptrace & PT_PTRACE_CAP)
1243 			bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1244 		else
1245 			bprm->unsafe |= LSM_UNSAFE_PTRACE;
1246 	}
1247 
1248 	/*
1249 	 * This isn't strictly necessary, but it makes it harder for LSMs to
1250 	 * mess up.
1251 	 */
1252 	if (current->no_new_privs)
1253 		bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1254 
1255 	n_fs = 1;
1256 	spin_lock(&p->fs->lock);
1257 	rcu_read_lock();
1258 	for (t = next_thread(p); t != p; t = next_thread(t)) {
1259 		if (t->fs == p->fs)
1260 			n_fs++;
1261 	}
1262 	rcu_read_unlock();
1263 
1264 	if (p->fs->users > n_fs) {
1265 		bprm->unsafe |= LSM_UNSAFE_SHARE;
1266 	} else {
1267 		res = -EAGAIN;
1268 		if (!p->fs->in_exec) {
1269 			p->fs->in_exec = 1;
1270 			res = 1;
1271 		}
1272 	}
1273 	spin_unlock(&p->fs->lock);
1274 
1275 	return res;
1276 }
1277 
1278 /*
1279  * Fill the binprm structure from the inode.
1280  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1281  *
1282  * This may be called multiple times for binary chains (scripts for example).
1283  */
1284 int prepare_binprm(struct linux_binprm *bprm)
1285 {
1286 	umode_t mode;
1287 	struct inode * inode = bprm->file->f_path.dentry->d_inode;
1288 	int retval;
1289 
1290 	mode = inode->i_mode;
1291 	if (bprm->file->f_op == NULL)
1292 		return -EACCES;
1293 
1294 	/* clear any previous set[ug]id data from a previous binary */
1295 	bprm->cred->euid = current_euid();
1296 	bprm->cred->egid = current_egid();
1297 
1298 	if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1299 	    !current->no_new_privs) {
1300 		/* Set-uid? */
1301 		if (mode & S_ISUID) {
1302 			if (!kuid_has_mapping(bprm->cred->user_ns, inode->i_uid))
1303 				return -EPERM;
1304 			bprm->per_clear |= PER_CLEAR_ON_SETID;
1305 			bprm->cred->euid = inode->i_uid;
1306 
1307 		}
1308 
1309 		/* Set-gid? */
1310 		/*
1311 		 * If setgid is set but no group execute bit then this
1312 		 * is a candidate for mandatory locking, not a setgid
1313 		 * executable.
1314 		 */
1315 		if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1316 			if (!kgid_has_mapping(bprm->cred->user_ns, inode->i_gid))
1317 				return -EPERM;
1318 			bprm->per_clear |= PER_CLEAR_ON_SETID;
1319 			bprm->cred->egid = inode->i_gid;
1320 		}
1321 	}
1322 
1323 	/* fill in binprm security blob */
1324 	retval = security_bprm_set_creds(bprm);
1325 	if (retval)
1326 		return retval;
1327 	bprm->cred_prepared = 1;
1328 
1329 	memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1330 	return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1331 }
1332 
1333 EXPORT_SYMBOL(prepare_binprm);
1334 
1335 /*
1336  * Arguments are '\0' separated strings found at the location bprm->p
1337  * points to; chop off the first by relocating brpm->p to right after
1338  * the first '\0' encountered.
1339  */
1340 int remove_arg_zero(struct linux_binprm *bprm)
1341 {
1342 	int ret = 0;
1343 	unsigned long offset;
1344 	char *kaddr;
1345 	struct page *page;
1346 
1347 	if (!bprm->argc)
1348 		return 0;
1349 
1350 	do {
1351 		offset = bprm->p & ~PAGE_MASK;
1352 		page = get_arg_page(bprm, bprm->p, 0);
1353 		if (!page) {
1354 			ret = -EFAULT;
1355 			goto out;
1356 		}
1357 		kaddr = kmap_atomic(page);
1358 
1359 		for (; offset < PAGE_SIZE && kaddr[offset];
1360 				offset++, bprm->p++)
1361 			;
1362 
1363 		kunmap_atomic(kaddr);
1364 		put_arg_page(page);
1365 
1366 		if (offset == PAGE_SIZE)
1367 			free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1368 	} while (offset == PAGE_SIZE);
1369 
1370 	bprm->p++;
1371 	bprm->argc--;
1372 	ret = 0;
1373 
1374 out:
1375 	return ret;
1376 }
1377 EXPORT_SYMBOL(remove_arg_zero);
1378 
1379 /*
1380  * cycle the list of binary formats handler, until one recognizes the image
1381  */
1382 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1383 {
1384 	unsigned int depth = bprm->recursion_depth;
1385 	int try,retval;
1386 	struct linux_binfmt *fmt;
1387 	pid_t old_pid, old_vpid;
1388 
1389 	retval = security_bprm_check(bprm);
1390 	if (retval)
1391 		return retval;
1392 
1393 	retval = audit_bprm(bprm);
1394 	if (retval)
1395 		return retval;
1396 
1397 	/* Need to fetch pid before load_binary changes it */
1398 	old_pid = current->pid;
1399 	rcu_read_lock();
1400 	old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1401 	rcu_read_unlock();
1402 
1403 	retval = -ENOENT;
1404 	for (try=0; try<2; try++) {
1405 		read_lock(&binfmt_lock);
1406 		list_for_each_entry(fmt, &formats, lh) {
1407 			int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1408 			if (!fn)
1409 				continue;
1410 			if (!try_module_get(fmt->module))
1411 				continue;
1412 			read_unlock(&binfmt_lock);
1413 			retval = fn(bprm, regs);
1414 			/*
1415 			 * Restore the depth counter to its starting value
1416 			 * in this call, so we don't have to rely on every
1417 			 * load_binary function to restore it on return.
1418 			 */
1419 			bprm->recursion_depth = depth;
1420 			if (retval >= 0) {
1421 				if (depth == 0) {
1422 					trace_sched_process_exec(current, old_pid, bprm);
1423 					ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1424 				}
1425 				put_binfmt(fmt);
1426 				allow_write_access(bprm->file);
1427 				if (bprm->file)
1428 					fput(bprm->file);
1429 				bprm->file = NULL;
1430 				current->did_exec = 1;
1431 				proc_exec_connector(current);
1432 				return retval;
1433 			}
1434 			read_lock(&binfmt_lock);
1435 			put_binfmt(fmt);
1436 			if (retval != -ENOEXEC || bprm->mm == NULL)
1437 				break;
1438 			if (!bprm->file) {
1439 				read_unlock(&binfmt_lock);
1440 				return retval;
1441 			}
1442 		}
1443 		read_unlock(&binfmt_lock);
1444 #ifdef CONFIG_MODULES
1445 		if (retval != -ENOEXEC || bprm->mm == NULL) {
1446 			break;
1447 		} else {
1448 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1449 			if (printable(bprm->buf[0]) &&
1450 			    printable(bprm->buf[1]) &&
1451 			    printable(bprm->buf[2]) &&
1452 			    printable(bprm->buf[3]))
1453 				break; /* -ENOEXEC */
1454 			if (try)
1455 				break; /* -ENOEXEC */
1456 			request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1457 		}
1458 #else
1459 		break;
1460 #endif
1461 	}
1462 	return retval;
1463 }
1464 
1465 EXPORT_SYMBOL(search_binary_handler);
1466 
1467 /*
1468  * sys_execve() executes a new program.
1469  */
1470 static int do_execve_common(const char *filename,
1471 				struct user_arg_ptr argv,
1472 				struct user_arg_ptr envp,
1473 				struct pt_regs *regs)
1474 {
1475 	struct linux_binprm *bprm;
1476 	struct file *file;
1477 	struct files_struct *displaced;
1478 	bool clear_in_exec;
1479 	int retval;
1480 	const struct cred *cred = current_cred();
1481 
1482 	/*
1483 	 * We move the actual failure in case of RLIMIT_NPROC excess from
1484 	 * set*uid() to execve() because too many poorly written programs
1485 	 * don't check setuid() return code.  Here we additionally recheck
1486 	 * whether NPROC limit is still exceeded.
1487 	 */
1488 	if ((current->flags & PF_NPROC_EXCEEDED) &&
1489 	    atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1490 		retval = -EAGAIN;
1491 		goto out_ret;
1492 	}
1493 
1494 	/* We're below the limit (still or again), so we don't want to make
1495 	 * further execve() calls fail. */
1496 	current->flags &= ~PF_NPROC_EXCEEDED;
1497 
1498 	retval = unshare_files(&displaced);
1499 	if (retval)
1500 		goto out_ret;
1501 
1502 	retval = -ENOMEM;
1503 	bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1504 	if (!bprm)
1505 		goto out_files;
1506 
1507 	retval = prepare_bprm_creds(bprm);
1508 	if (retval)
1509 		goto out_free;
1510 
1511 	retval = check_unsafe_exec(bprm);
1512 	if (retval < 0)
1513 		goto out_free;
1514 	clear_in_exec = retval;
1515 	current->in_execve = 1;
1516 
1517 	file = open_exec(filename);
1518 	retval = PTR_ERR(file);
1519 	if (IS_ERR(file))
1520 		goto out_unmark;
1521 
1522 	sched_exec();
1523 
1524 	bprm->file = file;
1525 	bprm->filename = filename;
1526 	bprm->interp = filename;
1527 
1528 	retval = bprm_mm_init(bprm);
1529 	if (retval)
1530 		goto out_file;
1531 
1532 	bprm->argc = count(argv, MAX_ARG_STRINGS);
1533 	if ((retval = bprm->argc) < 0)
1534 		goto out;
1535 
1536 	bprm->envc = count(envp, MAX_ARG_STRINGS);
1537 	if ((retval = bprm->envc) < 0)
1538 		goto out;
1539 
1540 	retval = prepare_binprm(bprm);
1541 	if (retval < 0)
1542 		goto out;
1543 
1544 	retval = copy_strings_kernel(1, &bprm->filename, bprm);
1545 	if (retval < 0)
1546 		goto out;
1547 
1548 	bprm->exec = bprm->p;
1549 	retval = copy_strings(bprm->envc, envp, bprm);
1550 	if (retval < 0)
1551 		goto out;
1552 
1553 	retval = copy_strings(bprm->argc, argv, bprm);
1554 	if (retval < 0)
1555 		goto out;
1556 
1557 	retval = search_binary_handler(bprm,regs);
1558 	if (retval < 0)
1559 		goto out;
1560 
1561 	/* execve succeeded */
1562 	current->fs->in_exec = 0;
1563 	current->in_execve = 0;
1564 	acct_update_integrals(current);
1565 	free_bprm(bprm);
1566 	if (displaced)
1567 		put_files_struct(displaced);
1568 	return retval;
1569 
1570 out:
1571 	if (bprm->mm) {
1572 		acct_arg_size(bprm, 0);
1573 		mmput(bprm->mm);
1574 	}
1575 
1576 out_file:
1577 	if (bprm->file) {
1578 		allow_write_access(bprm->file);
1579 		fput(bprm->file);
1580 	}
1581 
1582 out_unmark:
1583 	if (clear_in_exec)
1584 		current->fs->in_exec = 0;
1585 	current->in_execve = 0;
1586 
1587 out_free:
1588 	free_bprm(bprm);
1589 
1590 out_files:
1591 	if (displaced)
1592 		reset_files_struct(displaced);
1593 out_ret:
1594 	return retval;
1595 }
1596 
1597 int do_execve(const char *filename,
1598 	const char __user *const __user *__argv,
1599 	const char __user *const __user *__envp,
1600 	struct pt_regs *regs)
1601 {
1602 	struct user_arg_ptr argv = { .ptr.native = __argv };
1603 	struct user_arg_ptr envp = { .ptr.native = __envp };
1604 	return do_execve_common(filename, argv, envp, regs);
1605 }
1606 
1607 #ifdef CONFIG_COMPAT
1608 int compat_do_execve(char *filename,
1609 	compat_uptr_t __user *__argv,
1610 	compat_uptr_t __user *__envp,
1611 	struct pt_regs *regs)
1612 {
1613 	struct user_arg_ptr argv = {
1614 		.is_compat = true,
1615 		.ptr.compat = __argv,
1616 	};
1617 	struct user_arg_ptr envp = {
1618 		.is_compat = true,
1619 		.ptr.compat = __envp,
1620 	};
1621 	return do_execve_common(filename, argv, envp, regs);
1622 }
1623 #endif
1624 
1625 void set_binfmt(struct linux_binfmt *new)
1626 {
1627 	struct mm_struct *mm = current->mm;
1628 
1629 	if (mm->binfmt)
1630 		module_put(mm->binfmt->module);
1631 
1632 	mm->binfmt = new;
1633 	if (new)
1634 		__module_get(new->module);
1635 }
1636 
1637 EXPORT_SYMBOL(set_binfmt);
1638 
1639 static int expand_corename(struct core_name *cn)
1640 {
1641 	char *old_corename = cn->corename;
1642 
1643 	cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1644 	cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1645 
1646 	if (!cn->corename) {
1647 		kfree(old_corename);
1648 		return -ENOMEM;
1649 	}
1650 
1651 	return 0;
1652 }
1653 
1654 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1655 {
1656 	char *cur;
1657 	int need;
1658 	int ret;
1659 	va_list arg;
1660 
1661 	va_start(arg, fmt);
1662 	need = vsnprintf(NULL, 0, fmt, arg);
1663 	va_end(arg);
1664 
1665 	if (likely(need < cn->size - cn->used - 1))
1666 		goto out_printf;
1667 
1668 	ret = expand_corename(cn);
1669 	if (ret)
1670 		goto expand_fail;
1671 
1672 out_printf:
1673 	cur = cn->corename + cn->used;
1674 	va_start(arg, fmt);
1675 	vsnprintf(cur, need + 1, fmt, arg);
1676 	va_end(arg);
1677 	cn->used += need;
1678 	return 0;
1679 
1680 expand_fail:
1681 	return ret;
1682 }
1683 
1684 static void cn_escape(char *str)
1685 {
1686 	for (; *str; str++)
1687 		if (*str == '/')
1688 			*str = '!';
1689 }
1690 
1691 static int cn_print_exe_file(struct core_name *cn)
1692 {
1693 	struct file *exe_file;
1694 	char *pathbuf, *path;
1695 	int ret;
1696 
1697 	exe_file = get_mm_exe_file(current->mm);
1698 	if (!exe_file) {
1699 		char *commstart = cn->corename + cn->used;
1700 		ret = cn_printf(cn, "%s (path unknown)", current->comm);
1701 		cn_escape(commstart);
1702 		return ret;
1703 	}
1704 
1705 	pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1706 	if (!pathbuf) {
1707 		ret = -ENOMEM;
1708 		goto put_exe_file;
1709 	}
1710 
1711 	path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1712 	if (IS_ERR(path)) {
1713 		ret = PTR_ERR(path);
1714 		goto free_buf;
1715 	}
1716 
1717 	cn_escape(path);
1718 
1719 	ret = cn_printf(cn, "%s", path);
1720 
1721 free_buf:
1722 	kfree(pathbuf);
1723 put_exe_file:
1724 	fput(exe_file);
1725 	return ret;
1726 }
1727 
1728 /* format_corename will inspect the pattern parameter, and output a
1729  * name into corename, which must have space for at least
1730  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1731  */
1732 static int format_corename(struct core_name *cn, long signr)
1733 {
1734 	const struct cred *cred = current_cred();
1735 	const char *pat_ptr = core_pattern;
1736 	int ispipe = (*pat_ptr == '|');
1737 	int pid_in_pattern = 0;
1738 	int err = 0;
1739 
1740 	cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1741 	cn->corename = kmalloc(cn->size, GFP_KERNEL);
1742 	cn->used = 0;
1743 
1744 	if (!cn->corename)
1745 		return -ENOMEM;
1746 
1747 	/* Repeat as long as we have more pattern to process and more output
1748 	   space */
1749 	while (*pat_ptr) {
1750 		if (*pat_ptr != '%') {
1751 			if (*pat_ptr == 0)
1752 				goto out;
1753 			err = cn_printf(cn, "%c", *pat_ptr++);
1754 		} else {
1755 			switch (*++pat_ptr) {
1756 			/* single % at the end, drop that */
1757 			case 0:
1758 				goto out;
1759 			/* Double percent, output one percent */
1760 			case '%':
1761 				err = cn_printf(cn, "%c", '%');
1762 				break;
1763 			/* pid */
1764 			case 'p':
1765 				pid_in_pattern = 1;
1766 				err = cn_printf(cn, "%d",
1767 					      task_tgid_vnr(current));
1768 				break;
1769 			/* uid */
1770 			case 'u':
1771 				err = cn_printf(cn, "%d", cred->uid);
1772 				break;
1773 			/* gid */
1774 			case 'g':
1775 				err = cn_printf(cn, "%d", cred->gid);
1776 				break;
1777 			/* signal that caused the coredump */
1778 			case 's':
1779 				err = cn_printf(cn, "%ld", signr);
1780 				break;
1781 			/* UNIX time of coredump */
1782 			case 't': {
1783 				struct timeval tv;
1784 				do_gettimeofday(&tv);
1785 				err = cn_printf(cn, "%lu", tv.tv_sec);
1786 				break;
1787 			}
1788 			/* hostname */
1789 			case 'h': {
1790 				char *namestart = cn->corename + cn->used;
1791 				down_read(&uts_sem);
1792 				err = cn_printf(cn, "%s",
1793 					      utsname()->nodename);
1794 				up_read(&uts_sem);
1795 				cn_escape(namestart);
1796 				break;
1797 			}
1798 			/* executable */
1799 			case 'e': {
1800 				char *commstart = cn->corename + cn->used;
1801 				err = cn_printf(cn, "%s", current->comm);
1802 				cn_escape(commstart);
1803 				break;
1804 			}
1805 			case 'E':
1806 				err = cn_print_exe_file(cn);
1807 				break;
1808 			/* core limit size */
1809 			case 'c':
1810 				err = cn_printf(cn, "%lu",
1811 					      rlimit(RLIMIT_CORE));
1812 				break;
1813 			default:
1814 				break;
1815 			}
1816 			++pat_ptr;
1817 		}
1818 
1819 		if (err)
1820 			return err;
1821 	}
1822 
1823 	/* Backward compatibility with core_uses_pid:
1824 	 *
1825 	 * If core_pattern does not include a %p (as is the default)
1826 	 * and core_uses_pid is set, then .%pid will be appended to
1827 	 * the filename. Do not do this for piped commands. */
1828 	if (!ispipe && !pid_in_pattern && core_uses_pid) {
1829 		err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1830 		if (err)
1831 			return err;
1832 	}
1833 out:
1834 	return ispipe;
1835 }
1836 
1837 static int zap_process(struct task_struct *start, int exit_code)
1838 {
1839 	struct task_struct *t;
1840 	int nr = 0;
1841 
1842 	start->signal->flags = SIGNAL_GROUP_EXIT;
1843 	start->signal->group_exit_code = exit_code;
1844 	start->signal->group_stop_count = 0;
1845 
1846 	t = start;
1847 	do {
1848 		task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1849 		if (t != current && t->mm) {
1850 			sigaddset(&t->pending.signal, SIGKILL);
1851 			signal_wake_up(t, 1);
1852 			nr++;
1853 		}
1854 	} while_each_thread(start, t);
1855 
1856 	return nr;
1857 }
1858 
1859 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1860 				struct core_state *core_state, int exit_code)
1861 {
1862 	struct task_struct *g, *p;
1863 	unsigned long flags;
1864 	int nr = -EAGAIN;
1865 
1866 	spin_lock_irq(&tsk->sighand->siglock);
1867 	if (!signal_group_exit(tsk->signal)) {
1868 		mm->core_state = core_state;
1869 		nr = zap_process(tsk, exit_code);
1870 	}
1871 	spin_unlock_irq(&tsk->sighand->siglock);
1872 	if (unlikely(nr < 0))
1873 		return nr;
1874 
1875 	if (atomic_read(&mm->mm_users) == nr + 1)
1876 		goto done;
1877 	/*
1878 	 * We should find and kill all tasks which use this mm, and we should
1879 	 * count them correctly into ->nr_threads. We don't take tasklist
1880 	 * lock, but this is safe wrt:
1881 	 *
1882 	 * fork:
1883 	 *	None of sub-threads can fork after zap_process(leader). All
1884 	 *	processes which were created before this point should be
1885 	 *	visible to zap_threads() because copy_process() adds the new
1886 	 *	process to the tail of init_task.tasks list, and lock/unlock
1887 	 *	of ->siglock provides a memory barrier.
1888 	 *
1889 	 * do_exit:
1890 	 *	The caller holds mm->mmap_sem. This means that the task which
1891 	 *	uses this mm can't pass exit_mm(), so it can't exit or clear
1892 	 *	its ->mm.
1893 	 *
1894 	 * de_thread:
1895 	 *	It does list_replace_rcu(&leader->tasks, &current->tasks),
1896 	 *	we must see either old or new leader, this does not matter.
1897 	 *	However, it can change p->sighand, so lock_task_sighand(p)
1898 	 *	must be used. Since p->mm != NULL and we hold ->mmap_sem
1899 	 *	it can't fail.
1900 	 *
1901 	 *	Note also that "g" can be the old leader with ->mm == NULL
1902 	 *	and already unhashed and thus removed from ->thread_group.
1903 	 *	This is OK, __unhash_process()->list_del_rcu() does not
1904 	 *	clear the ->next pointer, we will find the new leader via
1905 	 *	next_thread().
1906 	 */
1907 	rcu_read_lock();
1908 	for_each_process(g) {
1909 		if (g == tsk->group_leader)
1910 			continue;
1911 		if (g->flags & PF_KTHREAD)
1912 			continue;
1913 		p = g;
1914 		do {
1915 			if (p->mm) {
1916 				if (unlikely(p->mm == mm)) {
1917 					lock_task_sighand(p, &flags);
1918 					nr += zap_process(p, exit_code);
1919 					unlock_task_sighand(p, &flags);
1920 				}
1921 				break;
1922 			}
1923 		} while_each_thread(g, p);
1924 	}
1925 	rcu_read_unlock();
1926 done:
1927 	atomic_set(&core_state->nr_threads, nr);
1928 	return nr;
1929 }
1930 
1931 static int coredump_wait(int exit_code, struct core_state *core_state)
1932 {
1933 	struct task_struct *tsk = current;
1934 	struct mm_struct *mm = tsk->mm;
1935 	int core_waiters = -EBUSY;
1936 
1937 	init_completion(&core_state->startup);
1938 	core_state->dumper.task = tsk;
1939 	core_state->dumper.next = NULL;
1940 
1941 	down_write(&mm->mmap_sem);
1942 	if (!mm->core_state)
1943 		core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1944 	up_write(&mm->mmap_sem);
1945 
1946 	if (core_waiters > 0) {
1947 		struct core_thread *ptr;
1948 
1949 		wait_for_completion(&core_state->startup);
1950 		/*
1951 		 * Wait for all the threads to become inactive, so that
1952 		 * all the thread context (extended register state, like
1953 		 * fpu etc) gets copied to the memory.
1954 		 */
1955 		ptr = core_state->dumper.next;
1956 		while (ptr != NULL) {
1957 			wait_task_inactive(ptr->task, 0);
1958 			ptr = ptr->next;
1959 		}
1960 	}
1961 
1962 	return core_waiters;
1963 }
1964 
1965 static void coredump_finish(struct mm_struct *mm)
1966 {
1967 	struct core_thread *curr, *next;
1968 	struct task_struct *task;
1969 
1970 	next = mm->core_state->dumper.next;
1971 	while ((curr = next) != NULL) {
1972 		next = curr->next;
1973 		task = curr->task;
1974 		/*
1975 		 * see exit_mm(), curr->task must not see
1976 		 * ->task == NULL before we read ->next.
1977 		 */
1978 		smp_mb();
1979 		curr->task = NULL;
1980 		wake_up_process(task);
1981 	}
1982 
1983 	mm->core_state = NULL;
1984 }
1985 
1986 /*
1987  * set_dumpable converts traditional three-value dumpable to two flags and
1988  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1989  * these bits are not changed atomically.  So get_dumpable can observe the
1990  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1991  * return either old dumpable or new one by paying attention to the order of
1992  * modifying the bits.
1993  *
1994  * dumpable |   mm->flags (binary)
1995  * old  new | initial interim  final
1996  * ---------+-----------------------
1997  *  0    1  |   00      01      01
1998  *  0    2  |   00      10(*)   11
1999  *  1    0  |   01      00      00
2000  *  1    2  |   01      11      11
2001  *  2    0  |   11      10(*)   00
2002  *  2    1  |   11      11      01
2003  *
2004  * (*) get_dumpable regards interim value of 10 as 11.
2005  */
2006 void set_dumpable(struct mm_struct *mm, int value)
2007 {
2008 	switch (value) {
2009 	case 0:
2010 		clear_bit(MMF_DUMPABLE, &mm->flags);
2011 		smp_wmb();
2012 		clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2013 		break;
2014 	case 1:
2015 		set_bit(MMF_DUMPABLE, &mm->flags);
2016 		smp_wmb();
2017 		clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2018 		break;
2019 	case 2:
2020 		set_bit(MMF_DUMP_SECURELY, &mm->flags);
2021 		smp_wmb();
2022 		set_bit(MMF_DUMPABLE, &mm->flags);
2023 		break;
2024 	}
2025 }
2026 
2027 static int __get_dumpable(unsigned long mm_flags)
2028 {
2029 	int ret;
2030 
2031 	ret = mm_flags & MMF_DUMPABLE_MASK;
2032 	return (ret >= 2) ? 2 : ret;
2033 }
2034 
2035 int get_dumpable(struct mm_struct *mm)
2036 {
2037 	return __get_dumpable(mm->flags);
2038 }
2039 
2040 static void wait_for_dump_helpers(struct file *file)
2041 {
2042 	struct pipe_inode_info *pipe;
2043 
2044 	pipe = file->f_path.dentry->d_inode->i_pipe;
2045 
2046 	pipe_lock(pipe);
2047 	pipe->readers++;
2048 	pipe->writers--;
2049 
2050 	while ((pipe->readers > 1) && (!signal_pending(current))) {
2051 		wake_up_interruptible_sync(&pipe->wait);
2052 		kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2053 		pipe_wait(pipe);
2054 	}
2055 
2056 	pipe->readers--;
2057 	pipe->writers++;
2058 	pipe_unlock(pipe);
2059 
2060 }
2061 
2062 
2063 /*
2064  * umh_pipe_setup
2065  * helper function to customize the process used
2066  * to collect the core in userspace.  Specifically
2067  * it sets up a pipe and installs it as fd 0 (stdin)
2068  * for the process.  Returns 0 on success, or
2069  * PTR_ERR on failure.
2070  * Note that it also sets the core limit to 1.  This
2071  * is a special value that we use to trap recursive
2072  * core dumps
2073  */
2074 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2075 {
2076 	struct file *rp, *wp;
2077 	struct fdtable *fdt;
2078 	struct coredump_params *cp = (struct coredump_params *)info->data;
2079 	struct files_struct *cf = current->files;
2080 
2081 	wp = create_write_pipe(0);
2082 	if (IS_ERR(wp))
2083 		return PTR_ERR(wp);
2084 
2085 	rp = create_read_pipe(wp, 0);
2086 	if (IS_ERR(rp)) {
2087 		free_write_pipe(wp);
2088 		return PTR_ERR(rp);
2089 	}
2090 
2091 	cp->file = wp;
2092 
2093 	sys_close(0);
2094 	fd_install(0, rp);
2095 	spin_lock(&cf->file_lock);
2096 	fdt = files_fdtable(cf);
2097 	__set_open_fd(0, fdt);
2098 	__clear_close_on_exec(0, fdt);
2099 	spin_unlock(&cf->file_lock);
2100 
2101 	/* and disallow core files too */
2102 	current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2103 
2104 	return 0;
2105 }
2106 
2107 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2108 {
2109 	struct core_state core_state;
2110 	struct core_name cn;
2111 	struct mm_struct *mm = current->mm;
2112 	struct linux_binfmt * binfmt;
2113 	const struct cred *old_cred;
2114 	struct cred *cred;
2115 	int retval = 0;
2116 	int flag = 0;
2117 	int ispipe;
2118 	static atomic_t core_dump_count = ATOMIC_INIT(0);
2119 	struct coredump_params cprm = {
2120 		.signr = signr,
2121 		.regs = regs,
2122 		.limit = rlimit(RLIMIT_CORE),
2123 		/*
2124 		 * We must use the same mm->flags while dumping core to avoid
2125 		 * inconsistency of bit flags, since this flag is not protected
2126 		 * by any locks.
2127 		 */
2128 		.mm_flags = mm->flags,
2129 	};
2130 
2131 	audit_core_dumps(signr);
2132 
2133 	binfmt = mm->binfmt;
2134 	if (!binfmt || !binfmt->core_dump)
2135 		goto fail;
2136 	if (!__get_dumpable(cprm.mm_flags))
2137 		goto fail;
2138 
2139 	cred = prepare_creds();
2140 	if (!cred)
2141 		goto fail;
2142 	/*
2143 	 *	We cannot trust fsuid as being the "true" uid of the
2144 	 *	process nor do we know its entire history. We only know it
2145 	 *	was tainted so we dump it as root in mode 2.
2146 	 */
2147 	if (__get_dumpable(cprm.mm_flags) == 2) {
2148 		/* Setuid core dump mode */
2149 		flag = O_EXCL;		/* Stop rewrite attacks */
2150 		cred->fsuid = GLOBAL_ROOT_UID;	/* Dump root private */
2151 	}
2152 
2153 	retval = coredump_wait(exit_code, &core_state);
2154 	if (retval < 0)
2155 		goto fail_creds;
2156 
2157 	old_cred = override_creds(cred);
2158 
2159 	/*
2160 	 * Clear any false indication of pending signals that might
2161 	 * be seen by the filesystem code called to write the core file.
2162 	 */
2163 	clear_thread_flag(TIF_SIGPENDING);
2164 
2165 	ispipe = format_corename(&cn, signr);
2166 
2167  	if (ispipe) {
2168 		int dump_count;
2169 		char **helper_argv;
2170 
2171 		if (ispipe < 0) {
2172 			printk(KERN_WARNING "format_corename failed\n");
2173 			printk(KERN_WARNING "Aborting core\n");
2174 			goto fail_corename;
2175 		}
2176 
2177 		if (cprm.limit == 1) {
2178 			/*
2179 			 * Normally core limits are irrelevant to pipes, since
2180 			 * we're not writing to the file system, but we use
2181 			 * cprm.limit of 1 here as a speacial value. Any
2182 			 * non-1 limit gets set to RLIM_INFINITY below, but
2183 			 * a limit of 0 skips the dump.  This is a consistent
2184 			 * way to catch recursive crashes.  We can still crash
2185 			 * if the core_pattern binary sets RLIM_CORE =  !1
2186 			 * but it runs as root, and can do lots of stupid things
2187 			 * Note that we use task_tgid_vnr here to grab the pid
2188 			 * of the process group leader.  That way we get the
2189 			 * right pid if a thread in a multi-threaded
2190 			 * core_pattern process dies.
2191 			 */
2192 			printk(KERN_WARNING
2193 				"Process %d(%s) has RLIMIT_CORE set to 1\n",
2194 				task_tgid_vnr(current), current->comm);
2195 			printk(KERN_WARNING "Aborting core\n");
2196 			goto fail_unlock;
2197 		}
2198 		cprm.limit = RLIM_INFINITY;
2199 
2200 		dump_count = atomic_inc_return(&core_dump_count);
2201 		if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2202 			printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2203 			       task_tgid_vnr(current), current->comm);
2204 			printk(KERN_WARNING "Skipping core dump\n");
2205 			goto fail_dropcount;
2206 		}
2207 
2208 		helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2209 		if (!helper_argv) {
2210 			printk(KERN_WARNING "%s failed to allocate memory\n",
2211 			       __func__);
2212 			goto fail_dropcount;
2213 		}
2214 
2215 		retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2216 					NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2217 					NULL, &cprm);
2218 		argv_free(helper_argv);
2219 		if (retval) {
2220  			printk(KERN_INFO "Core dump to %s pipe failed\n",
2221 			       cn.corename);
2222 			goto close_fail;
2223  		}
2224 	} else {
2225 		struct inode *inode;
2226 
2227 		if (cprm.limit < binfmt->min_coredump)
2228 			goto fail_unlock;
2229 
2230 		cprm.file = filp_open(cn.corename,
2231 				 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2232 				 0600);
2233 		if (IS_ERR(cprm.file))
2234 			goto fail_unlock;
2235 
2236 		inode = cprm.file->f_path.dentry->d_inode;
2237 		if (inode->i_nlink > 1)
2238 			goto close_fail;
2239 		if (d_unhashed(cprm.file->f_path.dentry))
2240 			goto close_fail;
2241 		/*
2242 		 * AK: actually i see no reason to not allow this for named
2243 		 * pipes etc, but keep the previous behaviour for now.
2244 		 */
2245 		if (!S_ISREG(inode->i_mode))
2246 			goto close_fail;
2247 		/*
2248 		 * Dont allow local users get cute and trick others to coredump
2249 		 * into their pre-created files.
2250 		 */
2251 		if (!uid_eq(inode->i_uid, current_fsuid()))
2252 			goto close_fail;
2253 		if (!cprm.file->f_op || !cprm.file->f_op->write)
2254 			goto close_fail;
2255 		if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2256 			goto close_fail;
2257 	}
2258 
2259 	retval = binfmt->core_dump(&cprm);
2260 	if (retval)
2261 		current->signal->group_exit_code |= 0x80;
2262 
2263 	if (ispipe && core_pipe_limit)
2264 		wait_for_dump_helpers(cprm.file);
2265 close_fail:
2266 	if (cprm.file)
2267 		filp_close(cprm.file, NULL);
2268 fail_dropcount:
2269 	if (ispipe)
2270 		atomic_dec(&core_dump_count);
2271 fail_unlock:
2272 	kfree(cn.corename);
2273 fail_corename:
2274 	coredump_finish(mm);
2275 	revert_creds(old_cred);
2276 fail_creds:
2277 	put_cred(cred);
2278 fail:
2279 	return;
2280 }
2281 
2282 /*
2283  * Core dumping helper functions.  These are the only things you should
2284  * do on a core-file: use only these functions to write out all the
2285  * necessary info.
2286  */
2287 int dump_write(struct file *file, const void *addr, int nr)
2288 {
2289 	return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2290 }
2291 EXPORT_SYMBOL(dump_write);
2292 
2293 int dump_seek(struct file *file, loff_t off)
2294 {
2295 	int ret = 1;
2296 
2297 	if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2298 		if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2299 			return 0;
2300 	} else {
2301 		char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2302 
2303 		if (!buf)
2304 			return 0;
2305 		while (off > 0) {
2306 			unsigned long n = off;
2307 
2308 			if (n > PAGE_SIZE)
2309 				n = PAGE_SIZE;
2310 			if (!dump_write(file, buf, n)) {
2311 				ret = 0;
2312 				break;
2313 			}
2314 			off -= n;
2315 		}
2316 		free_page((unsigned long)buf);
2317 	}
2318 	return ret;
2319 }
2320 EXPORT_SYMBOL(dump_seek);
2321