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/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/perf_counter.h>
37 #include <linux/highmem.h>
38 #include <linux/spinlock.h>
39 #include <linux/key.h>
40 #include <linux/personality.h>
41 #include <linux/binfmts.h>
42 #include <linux/utsname.h>
43 #include <linux/pid_namespace.h>
44 #include <linux/module.h>
45 #include <linux/namei.h>
46 #include <linux/proc_fs.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/ima.h>
50 #include <linux/syscalls.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54 #include <linux/tracehook.h>
55 #include <linux/kmod.h>
56 #include <linux/fsnotify.h>
57 #include <linux/fs_struct.h>
58
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
61 #include <asm/tlb.h>
62 #include "internal.h"
63
64 int core_uses_pid;
65 char core_pattern[CORENAME_MAX_SIZE] = "core";
66 int suid_dumpable = 0;
67
68 /* The maximal length of core_pattern is also specified in sysctl.c */
69
70 static LIST_HEAD(formats);
71 static DEFINE_RWLOCK(binfmt_lock);
72
73 int __register_binfmt(struct linux_binfmt * fmt, int insert)
74 {
75 if (!fmt)
76 return -EINVAL;
77 write_lock(&binfmt_lock);
78 insert ? list_add(&fmt->lh, &formats) :
79 list_add_tail(&fmt->lh, &formats);
80 write_unlock(&binfmt_lock);
81 return 0;
82 }
83
84 EXPORT_SYMBOL(__register_binfmt);
85
86 void unregister_binfmt(struct linux_binfmt * fmt)
87 {
88 write_lock(&binfmt_lock);
89 list_del(&fmt->lh);
90 write_unlock(&binfmt_lock);
91 }
92
93 EXPORT_SYMBOL(unregister_binfmt);
94
95 static inline void put_binfmt(struct linux_binfmt * fmt)
96 {
97 module_put(fmt->module);
98 }
99
100 /*
101 * Note that a shared library must be both readable and executable due to
102 * security reasons.
103 *
104 * Also note that we take the address to load from from the file itself.
105 */
106 SYSCALL_DEFINE1(uselib, const char __user *, library)
107 {
108 struct file *file;
109 char *tmp = getname(library);
110 int error = PTR_ERR(tmp);
111
112 if (IS_ERR(tmp))
113 goto out;
114
115 file = do_filp_open(AT_FDCWD, tmp,
116 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
117 MAY_READ | MAY_EXEC | MAY_OPEN);
118 putname(tmp);
119 error = PTR_ERR(file);
120 if (IS_ERR(file))
121 goto out;
122
123 error = -EINVAL;
124 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
125 goto exit;
126
127 error = -EACCES;
128 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
129 goto exit;
130
131 fsnotify_open(file->f_path.dentry);
132
133 error = -ENOEXEC;
134 if(file->f_op) {
135 struct linux_binfmt * fmt;
136
137 read_lock(&binfmt_lock);
138 list_for_each_entry(fmt, &formats, lh) {
139 if (!fmt->load_shlib)
140 continue;
141 if (!try_module_get(fmt->module))
142 continue;
143 read_unlock(&binfmt_lock);
144 error = fmt->load_shlib(file);
145 read_lock(&binfmt_lock);
146 put_binfmt(fmt);
147 if (error != -ENOEXEC)
148 break;
149 }
150 read_unlock(&binfmt_lock);
151 }
152 exit:
153 fput(file);
154 out:
155 return error;
156 }
157
158 #ifdef CONFIG_MMU
159
160 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
161 int write)
162 {
163 struct page *page;
164 int ret;
165
166 #ifdef CONFIG_STACK_GROWSUP
167 if (write) {
168 ret = expand_stack_downwards(bprm->vma, pos);
169 if (ret < 0)
170 return NULL;
171 }
172 #endif
173 ret = get_user_pages(current, bprm->mm, pos,
174 1, write, 1, &page, NULL);
175 if (ret <= 0)
176 return NULL;
177
178 if (write) {
179 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
180 struct rlimit *rlim;
181
182 /*
183 * We've historically supported up to 32 pages (ARG_MAX)
184 * of argument strings even with small stacks
185 */
186 if (size <= ARG_MAX)
187 return page;
188
189 /*
190 * Limit to 1/4-th the stack size for the argv+env strings.
191 * This ensures that:
192 * - the remaining binfmt code will not run out of stack space,
193 * - the program will have a reasonable amount of stack left
194 * to work from.
195 */
196 rlim = current->signal->rlim;
197 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
198 put_page(page);
199 return NULL;
200 }
201 }
202
203 return page;
204 }
205
206 static void put_arg_page(struct page *page)
207 {
208 put_page(page);
209 }
210
211 static void free_arg_page(struct linux_binprm *bprm, int i)
212 {
213 }
214
215 static void free_arg_pages(struct linux_binprm *bprm)
216 {
217 }
218
219 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
220 struct page *page)
221 {
222 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
223 }
224
225 static int __bprm_mm_init(struct linux_binprm *bprm)
226 {
227 int err;
228 struct vm_area_struct *vma = NULL;
229 struct mm_struct *mm = bprm->mm;
230
231 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
232 if (!vma)
233 return -ENOMEM;
234
235 down_write(&mm->mmap_sem);
236 vma->vm_mm = mm;
237
238 /*
239 * Place the stack at the largest stack address the architecture
240 * supports. Later, we'll move this to an appropriate place. We don't
241 * use STACK_TOP because that can depend on attributes which aren't
242 * configured yet.
243 */
244 vma->vm_end = STACK_TOP_MAX;
245 vma->vm_start = vma->vm_end - PAGE_SIZE;
246 vma->vm_flags = VM_STACK_FLAGS;
247 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
248 err = insert_vm_struct(mm, vma);
249 if (err)
250 goto err;
251
252 mm->stack_vm = mm->total_vm = 1;
253 up_write(&mm->mmap_sem);
254 bprm->p = vma->vm_end - sizeof(void *);
255 return 0;
256 err:
257 up_write(&mm->mmap_sem);
258 bprm->vma = NULL;
259 kmem_cache_free(vm_area_cachep, vma);
260 return err;
261 }
262
263 static bool valid_arg_len(struct linux_binprm *bprm, long len)
264 {
265 return len <= MAX_ARG_STRLEN;
266 }
267
268 #else
269
270 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
271 int write)
272 {
273 struct page *page;
274
275 page = bprm->page[pos / PAGE_SIZE];
276 if (!page && write) {
277 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
278 if (!page)
279 return NULL;
280 bprm->page[pos / PAGE_SIZE] = page;
281 }
282
283 return page;
284 }
285
286 static void put_arg_page(struct page *page)
287 {
288 }
289
290 static void free_arg_page(struct linux_binprm *bprm, int i)
291 {
292 if (bprm->page[i]) {
293 __free_page(bprm->page[i]);
294 bprm->page[i] = NULL;
295 }
296 }
297
298 static void free_arg_pages(struct linux_binprm *bprm)
299 {
300 int i;
301
302 for (i = 0; i < MAX_ARG_PAGES; i++)
303 free_arg_page(bprm, i);
304 }
305
306 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
307 struct page *page)
308 {
309 }
310
311 static int __bprm_mm_init(struct linux_binprm *bprm)
312 {
313 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
314 return 0;
315 }
316
317 static bool valid_arg_len(struct linux_binprm *bprm, long len)
318 {
319 return len <= bprm->p;
320 }
321
322 #endif /* CONFIG_MMU */
323
324 /*
325 * Create a new mm_struct and populate it with a temporary stack
326 * vm_area_struct. We don't have enough context at this point to set the stack
327 * flags, permissions, and offset, so we use temporary values. We'll update
328 * them later in setup_arg_pages().
329 */
330 int bprm_mm_init(struct linux_binprm *bprm)
331 {
332 int err;
333 struct mm_struct *mm = NULL;
334
335 bprm->mm = mm = mm_alloc();
336 err = -ENOMEM;
337 if (!mm)
338 goto err;
339
340 err = init_new_context(current, mm);
341 if (err)
342 goto err;
343
344 err = __bprm_mm_init(bprm);
345 if (err)
346 goto err;
347
348 return 0;
349
350 err:
351 if (mm) {
352 bprm->mm = NULL;
353 mmdrop(mm);
354 }
355
356 return err;
357 }
358
359 /*
360 * count() counts the number of strings in array ARGV.
361 */
362 static int count(char __user * __user * argv, int max)
363 {
364 int i = 0;
365
366 if (argv != NULL) {
367 for (;;) {
368 char __user * p;
369
370 if (get_user(p, argv))
371 return -EFAULT;
372 if (!p)
373 break;
374 argv++;
375 if (i++ >= max)
376 return -E2BIG;
377 cond_resched();
378 }
379 }
380 return i;
381 }
382
383 /*
384 * 'copy_strings()' copies argument/environment strings from the old
385 * processes's memory to the new process's stack. The call to get_user_pages()
386 * ensures the destination page is created and not swapped out.
387 */
388 static int copy_strings(int argc, char __user * __user * argv,
389 struct linux_binprm *bprm)
390 {
391 struct page *kmapped_page = NULL;
392 char *kaddr = NULL;
393 unsigned long kpos = 0;
394 int ret;
395
396 while (argc-- > 0) {
397 char __user *str;
398 int len;
399 unsigned long pos;
400
401 if (get_user(str, argv+argc) ||
402 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
403 ret = -EFAULT;
404 goto out;
405 }
406
407 if (!valid_arg_len(bprm, len)) {
408 ret = -E2BIG;
409 goto out;
410 }
411
412 /* We're going to work our way backwords. */
413 pos = bprm->p;
414 str += len;
415 bprm->p -= len;
416
417 while (len > 0) {
418 int offset, bytes_to_copy;
419
420 offset = pos % PAGE_SIZE;
421 if (offset == 0)
422 offset = PAGE_SIZE;
423
424 bytes_to_copy = offset;
425 if (bytes_to_copy > len)
426 bytes_to_copy = len;
427
428 offset -= bytes_to_copy;
429 pos -= bytes_to_copy;
430 str -= bytes_to_copy;
431 len -= bytes_to_copy;
432
433 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
434 struct page *page;
435
436 page = get_arg_page(bprm, pos, 1);
437 if (!page) {
438 ret = -E2BIG;
439 goto out;
440 }
441
442 if (kmapped_page) {
443 flush_kernel_dcache_page(kmapped_page);
444 kunmap(kmapped_page);
445 put_arg_page(kmapped_page);
446 }
447 kmapped_page = page;
448 kaddr = kmap(kmapped_page);
449 kpos = pos & PAGE_MASK;
450 flush_arg_page(bprm, kpos, kmapped_page);
451 }
452 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
453 ret = -EFAULT;
454 goto out;
455 }
456 }
457 }
458 ret = 0;
459 out:
460 if (kmapped_page) {
461 flush_kernel_dcache_page(kmapped_page);
462 kunmap(kmapped_page);
463 put_arg_page(kmapped_page);
464 }
465 return ret;
466 }
467
468 /*
469 * Like copy_strings, but get argv and its values from kernel memory.
470 */
471 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
472 {
473 int r;
474 mm_segment_t oldfs = get_fs();
475 set_fs(KERNEL_DS);
476 r = copy_strings(argc, (char __user * __user *)argv, bprm);
477 set_fs(oldfs);
478 return r;
479 }
480 EXPORT_SYMBOL(copy_strings_kernel);
481
482 #ifdef CONFIG_MMU
483
484 /*
485 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
486 * the binfmt code determines where the new stack should reside, we shift it to
487 * its final location. The process proceeds as follows:
488 *
489 * 1) Use shift to calculate the new vma endpoints.
490 * 2) Extend vma to cover both the old and new ranges. This ensures the
491 * arguments passed to subsequent functions are consistent.
492 * 3) Move vma's page tables to the new range.
493 * 4) Free up any cleared pgd range.
494 * 5) Shrink the vma to cover only the new range.
495 */
496 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
497 {
498 struct mm_struct *mm = vma->vm_mm;
499 unsigned long old_start = vma->vm_start;
500 unsigned long old_end = vma->vm_end;
501 unsigned long length = old_end - old_start;
502 unsigned long new_start = old_start - shift;
503 unsigned long new_end = old_end - shift;
504 struct mmu_gather *tlb;
505
506 BUG_ON(new_start > new_end);
507
508 /*
509 * ensure there are no vmas between where we want to go
510 * and where we are
511 */
512 if (vma != find_vma(mm, new_start))
513 return -EFAULT;
514
515 /*
516 * cover the whole range: [new_start, old_end)
517 */
518 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
519
520 /*
521 * move the page tables downwards, on failure we rely on
522 * process cleanup to remove whatever mess we made.
523 */
524 if (length != move_page_tables(vma, old_start,
525 vma, new_start, length))
526 return -ENOMEM;
527
528 lru_add_drain();
529 tlb = tlb_gather_mmu(mm, 0);
530 if (new_end > old_start) {
531 /*
532 * when the old and new regions overlap clear from new_end.
533 */
534 free_pgd_range(tlb, new_end, old_end, new_end,
535 vma->vm_next ? vma->vm_next->vm_start : 0);
536 } else {
537 /*
538 * otherwise, clean from old_start; this is done to not touch
539 * the address space in [new_end, old_start) some architectures
540 * have constraints on va-space that make this illegal (IA64) -
541 * for the others its just a little faster.
542 */
543 free_pgd_range(tlb, old_start, old_end, new_end,
544 vma->vm_next ? vma->vm_next->vm_start : 0);
545 }
546 tlb_finish_mmu(tlb, new_end, old_end);
547
548 /*
549 * shrink the vma to just the new range.
550 */
551 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
552
553 return 0;
554 }
555
556 #define EXTRA_STACK_VM_PAGES 20 /* random */
557
558 /*
559 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
560 * the stack is optionally relocated, and some extra space is added.
561 */
562 int setup_arg_pages(struct linux_binprm *bprm,
563 unsigned long stack_top,
564 int executable_stack)
565 {
566 unsigned long ret;
567 unsigned long stack_shift;
568 struct mm_struct *mm = current->mm;
569 struct vm_area_struct *vma = bprm->vma;
570 struct vm_area_struct *prev = NULL;
571 unsigned long vm_flags;
572 unsigned long stack_base;
573
574 #ifdef CONFIG_STACK_GROWSUP
575 /* Limit stack size to 1GB */
576 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
577 if (stack_base > (1 << 30))
578 stack_base = 1 << 30;
579
580 /* Make sure we didn't let the argument array grow too large. */
581 if (vma->vm_end - vma->vm_start > stack_base)
582 return -ENOMEM;
583
584 stack_base = PAGE_ALIGN(stack_top - stack_base);
585
586 stack_shift = vma->vm_start - stack_base;
587 mm->arg_start = bprm->p - stack_shift;
588 bprm->p = vma->vm_end - stack_shift;
589 #else
590 stack_top = arch_align_stack(stack_top);
591 stack_top = PAGE_ALIGN(stack_top);
592 stack_shift = vma->vm_end - stack_top;
593
594 bprm->p -= stack_shift;
595 mm->arg_start = bprm->p;
596 #endif
597
598 if (bprm->loader)
599 bprm->loader -= stack_shift;
600 bprm->exec -= stack_shift;
601
602 down_write(&mm->mmap_sem);
603 vm_flags = VM_STACK_FLAGS;
604
605 /*
606 * Adjust stack execute permissions; explicitly enable for
607 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
608 * (arch default) otherwise.
609 */
610 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
611 vm_flags |= VM_EXEC;
612 else if (executable_stack == EXSTACK_DISABLE_X)
613 vm_flags &= ~VM_EXEC;
614 vm_flags |= mm->def_flags;
615
616 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
617 vm_flags);
618 if (ret)
619 goto out_unlock;
620 BUG_ON(prev != vma);
621
622 /* Move stack pages down in memory. */
623 if (stack_shift) {
624 ret = shift_arg_pages(vma, stack_shift);
625 if (ret) {
626 up_write(&mm->mmap_sem);
627 return ret;
628 }
629 }
630
631 #ifdef CONFIG_STACK_GROWSUP
632 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
633 #else
634 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
635 #endif
636 ret = expand_stack(vma, stack_base);
637 if (ret)
638 ret = -EFAULT;
639
640 out_unlock:
641 up_write(&mm->mmap_sem);
642 return 0;
643 }
644 EXPORT_SYMBOL(setup_arg_pages);
645
646 #endif /* CONFIG_MMU */
647
648 struct file *open_exec(const char *name)
649 {
650 struct file *file;
651 int err;
652
653 file = do_filp_open(AT_FDCWD, name,
654 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
655 MAY_EXEC | MAY_OPEN);
656 if (IS_ERR(file))
657 goto out;
658
659 err = -EACCES;
660 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
661 goto exit;
662
663 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
664 goto exit;
665
666 fsnotify_open(file->f_path.dentry);
667
668 err = deny_write_access(file);
669 if (err)
670 goto exit;
671
672 out:
673 return file;
674
675 exit:
676 fput(file);
677 return ERR_PTR(err);
678 }
679 EXPORT_SYMBOL(open_exec);
680
681 int kernel_read(struct file *file, loff_t offset,
682 char *addr, unsigned long count)
683 {
684 mm_segment_t old_fs;
685 loff_t pos = offset;
686 int result;
687
688 old_fs = get_fs();
689 set_fs(get_ds());
690 /* The cast to a user pointer is valid due to the set_fs() */
691 result = vfs_read(file, (void __user *)addr, count, &pos);
692 set_fs(old_fs);
693 return result;
694 }
695
696 EXPORT_SYMBOL(kernel_read);
697
698 static int exec_mmap(struct mm_struct *mm)
699 {
700 struct task_struct *tsk;
701 struct mm_struct * old_mm, *active_mm;
702
703 /* Notify parent that we're no longer interested in the old VM */
704 tsk = current;
705 old_mm = current->mm;
706 mm_release(tsk, old_mm);
707
708 if (old_mm) {
709 /*
710 * Make sure that if there is a core dump in progress
711 * for the old mm, we get out and die instead of going
712 * through with the exec. We must hold mmap_sem around
713 * checking core_state and changing tsk->mm.
714 */
715 down_read(&old_mm->mmap_sem);
716 if (unlikely(old_mm->core_state)) {
717 up_read(&old_mm->mmap_sem);
718 return -EINTR;
719 }
720 }
721 task_lock(tsk);
722 active_mm = tsk->active_mm;
723 tsk->mm = mm;
724 tsk->active_mm = mm;
725 activate_mm(active_mm, mm);
726 task_unlock(tsk);
727 arch_pick_mmap_layout(mm);
728 if (old_mm) {
729 up_read(&old_mm->mmap_sem);
730 BUG_ON(active_mm != old_mm);
731 mm_update_next_owner(old_mm);
732 mmput(old_mm);
733 return 0;
734 }
735 mmdrop(active_mm);
736 return 0;
737 }
738
739 /*
740 * This function makes sure the current process has its own signal table,
741 * so that flush_signal_handlers can later reset the handlers without
742 * disturbing other processes. (Other processes might share the signal
743 * table via the CLONE_SIGHAND option to clone().)
744 */
745 static int de_thread(struct task_struct *tsk)
746 {
747 struct signal_struct *sig = tsk->signal;
748 struct sighand_struct *oldsighand = tsk->sighand;
749 spinlock_t *lock = &oldsighand->siglock;
750 int count;
751
752 if (thread_group_empty(tsk))
753 goto no_thread_group;
754
755 /*
756 * Kill all other threads in the thread group.
757 */
758 spin_lock_irq(lock);
759 if (signal_group_exit(sig)) {
760 /*
761 * Another group action in progress, just
762 * return so that the signal is processed.
763 */
764 spin_unlock_irq(lock);
765 return -EAGAIN;
766 }
767 sig->group_exit_task = tsk;
768 zap_other_threads(tsk);
769
770 /* Account for the thread group leader hanging around: */
771 count = thread_group_leader(tsk) ? 1 : 2;
772 sig->notify_count = count;
773 while (atomic_read(&sig->count) > count) {
774 __set_current_state(TASK_UNINTERRUPTIBLE);
775 spin_unlock_irq(lock);
776 schedule();
777 spin_lock_irq(lock);
778 }
779 spin_unlock_irq(lock);
780
781 /*
782 * At this point all other threads have exited, all we have to
783 * do is to wait for the thread group leader to become inactive,
784 * and to assume its PID:
785 */
786 if (!thread_group_leader(tsk)) {
787 struct task_struct *leader = tsk->group_leader;
788
789 sig->notify_count = -1; /* for exit_notify() */
790 for (;;) {
791 write_lock_irq(&tasklist_lock);
792 if (likely(leader->exit_state))
793 break;
794 __set_current_state(TASK_UNINTERRUPTIBLE);
795 write_unlock_irq(&tasklist_lock);
796 schedule();
797 }
798
799 /*
800 * The only record we have of the real-time age of a
801 * process, regardless of execs it's done, is start_time.
802 * All the past CPU time is accumulated in signal_struct
803 * from sister threads now dead. But in this non-leader
804 * exec, nothing survives from the original leader thread,
805 * whose birth marks the true age of this process now.
806 * When we take on its identity by switching to its PID, we
807 * also take its birthdate (always earlier than our own).
808 */
809 tsk->start_time = leader->start_time;
810
811 BUG_ON(!same_thread_group(leader, tsk));
812 BUG_ON(has_group_leader_pid(tsk));
813 /*
814 * An exec() starts a new thread group with the
815 * TGID of the previous thread group. Rehash the
816 * two threads with a switched PID, and release
817 * the former thread group leader:
818 */
819
820 /* Become a process group leader with the old leader's pid.
821 * The old leader becomes a thread of the this thread group.
822 * Note: The old leader also uses this pid until release_task
823 * is called. Odd but simple and correct.
824 */
825 detach_pid(tsk, PIDTYPE_PID);
826 tsk->pid = leader->pid;
827 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
828 transfer_pid(leader, tsk, PIDTYPE_PGID);
829 transfer_pid(leader, tsk, PIDTYPE_SID);
830 list_replace_rcu(&leader->tasks, &tsk->tasks);
831
832 tsk->group_leader = tsk;
833 leader->group_leader = tsk;
834
835 tsk->exit_signal = SIGCHLD;
836
837 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
838 leader->exit_state = EXIT_DEAD;
839 write_unlock_irq(&tasklist_lock);
840
841 release_task(leader);
842 }
843
844 sig->group_exit_task = NULL;
845 sig->notify_count = 0;
846
847 no_thread_group:
848 exit_itimers(sig);
849 flush_itimer_signals();
850
851 if (atomic_read(&oldsighand->count) != 1) {
852 struct sighand_struct *newsighand;
853 /*
854 * This ->sighand is shared with the CLONE_SIGHAND
855 * but not CLONE_THREAD task, switch to the new one.
856 */
857 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
858 if (!newsighand)
859 return -ENOMEM;
860
861 atomic_set(&newsighand->count, 1);
862 memcpy(newsighand->action, oldsighand->action,
863 sizeof(newsighand->action));
864
865 write_lock_irq(&tasklist_lock);
866 spin_lock(&oldsighand->siglock);
867 rcu_assign_pointer(tsk->sighand, newsighand);
868 spin_unlock(&oldsighand->siglock);
869 write_unlock_irq(&tasklist_lock);
870
871 __cleanup_sighand(oldsighand);
872 }
873
874 BUG_ON(!thread_group_leader(tsk));
875 return 0;
876 }
877
878 /*
879 * These functions flushes out all traces of the currently running executable
880 * so that a new one can be started
881 */
882 static void flush_old_files(struct files_struct * files)
883 {
884 long j = -1;
885 struct fdtable *fdt;
886
887 spin_lock(&files->file_lock);
888 for (;;) {
889 unsigned long set, i;
890
891 j++;
892 i = j * __NFDBITS;
893 fdt = files_fdtable(files);
894 if (i >= fdt->max_fds)
895 break;
896 set = fdt->close_on_exec->fds_bits[j];
897 if (!set)
898 continue;
899 fdt->close_on_exec->fds_bits[j] = 0;
900 spin_unlock(&files->file_lock);
901 for ( ; set ; i++,set >>= 1) {
902 if (set & 1) {
903 sys_close(i);
904 }
905 }
906 spin_lock(&files->file_lock);
907
908 }
909 spin_unlock(&files->file_lock);
910 }
911
912 char *get_task_comm(char *buf, struct task_struct *tsk)
913 {
914 /* buf must be at least sizeof(tsk->comm) in size */
915 task_lock(tsk);
916 strncpy(buf, tsk->comm, sizeof(tsk->comm));
917 task_unlock(tsk);
918 return buf;
919 }
920
921 void set_task_comm(struct task_struct *tsk, char *buf)
922 {
923 task_lock(tsk);
924 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
925 task_unlock(tsk);
926 perf_counter_comm(tsk);
927 }
928
929 int flush_old_exec(struct linux_binprm * bprm)
930 {
931 int retval;
932
933 /*
934 * Make sure we have a private signal table and that
935 * we are unassociated from the previous thread group.
936 */
937 retval = de_thread(current);
938 if (retval)
939 goto out;
940
941 set_mm_exe_file(bprm->mm, bprm->file);
942
943 /*
944 * Release all of the old mmap stuff
945 */
946 retval = exec_mmap(bprm->mm);
947 if (retval)
948 goto out;
949
950 bprm->mm = NULL; /* We're using it now */
951
952 current->flags &= ~PF_RANDOMIZE;
953 flush_thread();
954 current->personality &= ~bprm->per_clear;
955
956 return 0;
957
958 out:
959 return retval;
960 }
961 EXPORT_SYMBOL(flush_old_exec);
962
963 void setup_new_exec(struct linux_binprm * bprm)
964 {
965 int i, ch;
966 char * name;
967 char tcomm[sizeof(current->comm)];
968
969 arch_pick_mmap_layout(current->mm);
970
971 /* This is the point of no return */
972 current->sas_ss_sp = current->sas_ss_size = 0;
973
974 if (current_euid() == current_uid() && current_egid() == current_gid())
975 set_dumpable(current->mm, 1);
976 else
977 set_dumpable(current->mm, suid_dumpable);
978
979 name = bprm->filename;
980
981 /* Copies the binary name from after last slash */
982 for (i=0; (ch = *(name++)) != '\0';) {
983 if (ch == '/')
984 i = 0; /* overwrite what we wrote */
985 else
986 if (i < (sizeof(tcomm) - 1))
987 tcomm[i++] = ch;
988 }
989 tcomm[i] = '\0';
990 set_task_comm(current, tcomm);
991
992 /* Set the new mm task size. We have to do that late because it may
993 * depend on TIF_32BIT which is only updated in flush_thread() on
994 * some architectures like powerpc
995 */
996 current->mm->task_size = TASK_SIZE;
997
998 /* install the new credentials */
999 if (bprm->cred->uid != current_euid() ||
1000 bprm->cred->gid != current_egid()) {
1001 current->pdeath_signal = 0;
1002 } else if (file_permission(bprm->file, MAY_READ) ||
1003 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1004 set_dumpable(current->mm, suid_dumpable);
1005 }
1006
1007 /*
1008 * Flush performance counters when crossing a
1009 * security domain:
1010 */
1011 if (!get_dumpable(current->mm))
1012 perf_counter_exit_task(current);
1013
1014 /* An exec changes our domain. We are no longer part of the thread
1015 group */
1016
1017 current->self_exec_id++;
1018
1019 flush_signal_handlers(current, 0);
1020 flush_old_files(current->files);
1021 }
1022 EXPORT_SYMBOL(setup_new_exec);
1023
1024 /*
1025 * Prepare credentials and lock ->cred_guard_mutex.
1026 * install_exec_creds() commits the new creds and drops the lock.
1027 * Or, if exec fails before, free_bprm() should release ->cred and
1028 * and unlock.
1029 */
1030 int prepare_bprm_creds(struct linux_binprm *bprm)
1031 {
1032 if (mutex_lock_interruptible(¤t->cred_guard_mutex))
1033 return -ERESTARTNOINTR;
1034
1035 bprm->cred = prepare_exec_creds();
1036 if (likely(bprm->cred))
1037 return 0;
1038
1039 mutex_unlock(¤t->cred_guard_mutex);
1040 return -ENOMEM;
1041 }
1042
1043 void free_bprm(struct linux_binprm *bprm)
1044 {
1045 free_arg_pages(bprm);
1046 if (bprm->cred) {
1047 mutex_unlock(¤t->cred_guard_mutex);
1048 abort_creds(bprm->cred);
1049 }
1050 kfree(bprm);
1051 }
1052
1053 /*
1054 * install the new credentials for this executable
1055 */
1056 void install_exec_creds(struct linux_binprm *bprm)
1057 {
1058 security_bprm_committing_creds(bprm);
1059
1060 commit_creds(bprm->cred);
1061 bprm->cred = NULL;
1062 /*
1063 * cred_guard_mutex must be held at least to this point to prevent
1064 * ptrace_attach() from altering our determination of the task's
1065 * credentials; any time after this it may be unlocked.
1066 */
1067 security_bprm_committed_creds(bprm);
1068 mutex_unlock(¤t->cred_guard_mutex);
1069 }
1070 EXPORT_SYMBOL(install_exec_creds);
1071
1072 /*
1073 * determine how safe it is to execute the proposed program
1074 * - the caller must hold current->cred_guard_mutex to protect against
1075 * PTRACE_ATTACH
1076 */
1077 int check_unsafe_exec(struct linux_binprm *bprm)
1078 {
1079 struct task_struct *p = current, *t;
1080 unsigned n_fs;
1081 int res = 0;
1082
1083 bprm->unsafe = tracehook_unsafe_exec(p);
1084
1085 n_fs = 1;
1086 write_lock(&p->fs->lock);
1087 rcu_read_lock();
1088 for (t = next_thread(p); t != p; t = next_thread(t)) {
1089 if (t->fs == p->fs)
1090 n_fs++;
1091 }
1092 rcu_read_unlock();
1093
1094 if (p->fs->users > n_fs) {
1095 bprm->unsafe |= LSM_UNSAFE_SHARE;
1096 } else {
1097 res = -EAGAIN;
1098 if (!p->fs->in_exec) {
1099 p->fs->in_exec = 1;
1100 res = 1;
1101 }
1102 }
1103 write_unlock(&p->fs->lock);
1104
1105 return res;
1106 }
1107
1108 /*
1109 * Fill the binprm structure from the inode.
1110 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1111 *
1112 * This may be called multiple times for binary chains (scripts for example).
1113 */
1114 int prepare_binprm(struct linux_binprm *bprm)
1115 {
1116 umode_t mode;
1117 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1118 int retval;
1119
1120 mode = inode->i_mode;
1121 if (bprm->file->f_op == NULL)
1122 return -EACCES;
1123
1124 /* clear any previous set[ug]id data from a previous binary */
1125 bprm->cred->euid = current_euid();
1126 bprm->cred->egid = current_egid();
1127
1128 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1129 /* Set-uid? */
1130 if (mode & S_ISUID) {
1131 bprm->per_clear |= PER_CLEAR_ON_SETID;
1132 bprm->cred->euid = inode->i_uid;
1133 }
1134
1135 /* Set-gid? */
1136 /*
1137 * If setgid is set but no group execute bit then this
1138 * is a candidate for mandatory locking, not a setgid
1139 * executable.
1140 */
1141 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1142 bprm->per_clear |= PER_CLEAR_ON_SETID;
1143 bprm->cred->egid = inode->i_gid;
1144 }
1145 }
1146
1147 /* fill in binprm security blob */
1148 retval = security_bprm_set_creds(bprm);
1149 if (retval)
1150 return retval;
1151 bprm->cred_prepared = 1;
1152
1153 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1154 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1155 }
1156
1157 EXPORT_SYMBOL(prepare_binprm);
1158
1159 /*
1160 * Arguments are '\0' separated strings found at the location bprm->p
1161 * points to; chop off the first by relocating brpm->p to right after
1162 * the first '\0' encountered.
1163 */
1164 int remove_arg_zero(struct linux_binprm *bprm)
1165 {
1166 int ret = 0;
1167 unsigned long offset;
1168 char *kaddr;
1169 struct page *page;
1170
1171 if (!bprm->argc)
1172 return 0;
1173
1174 do {
1175 offset = bprm->p & ~PAGE_MASK;
1176 page = get_arg_page(bprm, bprm->p, 0);
1177 if (!page) {
1178 ret = -EFAULT;
1179 goto out;
1180 }
1181 kaddr = kmap_atomic(page, KM_USER0);
1182
1183 for (; offset < PAGE_SIZE && kaddr[offset];
1184 offset++, bprm->p++)
1185 ;
1186
1187 kunmap_atomic(kaddr, KM_USER0);
1188 put_arg_page(page);
1189
1190 if (offset == PAGE_SIZE)
1191 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1192 } while (offset == PAGE_SIZE);
1193
1194 bprm->p++;
1195 bprm->argc--;
1196 ret = 0;
1197
1198 out:
1199 return ret;
1200 }
1201 EXPORT_SYMBOL(remove_arg_zero);
1202
1203 /*
1204 * cycle the list of binary formats handler, until one recognizes the image
1205 */
1206 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1207 {
1208 unsigned int depth = bprm->recursion_depth;
1209 int try,retval;
1210 struct linux_binfmt *fmt;
1211
1212 retval = security_bprm_check(bprm);
1213 if (retval)
1214 return retval;
1215 retval = ima_bprm_check(bprm);
1216 if (retval)
1217 return retval;
1218
1219 /* kernel module loader fixup */
1220 /* so we don't try to load run modprobe in kernel space. */
1221 set_fs(USER_DS);
1222
1223 retval = audit_bprm(bprm);
1224 if (retval)
1225 return retval;
1226
1227 retval = -ENOENT;
1228 for (try=0; try<2; try++) {
1229 read_lock(&binfmt_lock);
1230 list_for_each_entry(fmt, &formats, lh) {
1231 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1232 if (!fn)
1233 continue;
1234 if (!try_module_get(fmt->module))
1235 continue;
1236 read_unlock(&binfmt_lock);
1237 retval = fn(bprm, regs);
1238 /*
1239 * Restore the depth counter to its starting value
1240 * in this call, so we don't have to rely on every
1241 * load_binary function to restore it on return.
1242 */
1243 bprm->recursion_depth = depth;
1244 if (retval >= 0) {
1245 if (depth == 0)
1246 tracehook_report_exec(fmt, bprm, regs);
1247 put_binfmt(fmt);
1248 allow_write_access(bprm->file);
1249 if (bprm->file)
1250 fput(bprm->file);
1251 bprm->file = NULL;
1252 current->did_exec = 1;
1253 proc_exec_connector(current);
1254 return retval;
1255 }
1256 read_lock(&binfmt_lock);
1257 put_binfmt(fmt);
1258 if (retval != -ENOEXEC || bprm->mm == NULL)
1259 break;
1260 if (!bprm->file) {
1261 read_unlock(&binfmt_lock);
1262 return retval;
1263 }
1264 }
1265 read_unlock(&binfmt_lock);
1266 if (retval != -ENOEXEC || bprm->mm == NULL) {
1267 break;
1268 #ifdef CONFIG_MODULES
1269 } else {
1270 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1271 if (printable(bprm->buf[0]) &&
1272 printable(bprm->buf[1]) &&
1273 printable(bprm->buf[2]) &&
1274 printable(bprm->buf[3]))
1275 break; /* -ENOEXEC */
1276 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1277 #endif
1278 }
1279 }
1280 return retval;
1281 }
1282
1283 EXPORT_SYMBOL(search_binary_handler);
1284
1285 /*
1286 * sys_execve() executes a new program.
1287 */
1288 int do_execve(char * filename,
1289 char __user *__user *argv,
1290 char __user *__user *envp,
1291 struct pt_regs * regs)
1292 {
1293 struct linux_binprm *bprm;
1294 struct file *file;
1295 struct files_struct *displaced;
1296 bool clear_in_exec;
1297 int retval;
1298
1299 retval = unshare_files(&displaced);
1300 if (retval)
1301 goto out_ret;
1302
1303 retval = -ENOMEM;
1304 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1305 if (!bprm)
1306 goto out_files;
1307
1308 retval = prepare_bprm_creds(bprm);
1309 if (retval)
1310 goto out_free;
1311
1312 retval = check_unsafe_exec(bprm);
1313 if (retval < 0)
1314 goto out_free;
1315 clear_in_exec = retval;
1316 current->in_execve = 1;
1317
1318 file = open_exec(filename);
1319 retval = PTR_ERR(file);
1320 if (IS_ERR(file))
1321 goto out_unmark;
1322
1323 sched_exec();
1324
1325 bprm->file = file;
1326 bprm->filename = filename;
1327 bprm->interp = filename;
1328
1329 retval = bprm_mm_init(bprm);
1330 if (retval)
1331 goto out_file;
1332
1333 bprm->argc = count(argv, MAX_ARG_STRINGS);
1334 if ((retval = bprm->argc) < 0)
1335 goto out;
1336
1337 bprm->envc = count(envp, MAX_ARG_STRINGS);
1338 if ((retval = bprm->envc) < 0)
1339 goto out;
1340
1341 retval = prepare_binprm(bprm);
1342 if (retval < 0)
1343 goto out;
1344
1345 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1346 if (retval < 0)
1347 goto out;
1348
1349 bprm->exec = bprm->p;
1350 retval = copy_strings(bprm->envc, envp, bprm);
1351 if (retval < 0)
1352 goto out;
1353
1354 retval = copy_strings(bprm->argc, argv, bprm);
1355 if (retval < 0)
1356 goto out;
1357
1358 current->flags &= ~PF_KTHREAD;
1359 retval = search_binary_handler(bprm,regs);
1360 if (retval < 0)
1361 goto out;
1362
1363 /* execve succeeded */
1364 current->fs->in_exec = 0;
1365 current->in_execve = 0;
1366 acct_update_integrals(current);
1367 free_bprm(bprm);
1368 if (displaced)
1369 put_files_struct(displaced);
1370 return retval;
1371
1372 out:
1373 if (bprm->mm)
1374 mmput (bprm->mm);
1375
1376 out_file:
1377 if (bprm->file) {
1378 allow_write_access(bprm->file);
1379 fput(bprm->file);
1380 }
1381
1382 out_unmark:
1383 if (clear_in_exec)
1384 current->fs->in_exec = 0;
1385 current->in_execve = 0;
1386
1387 out_free:
1388 free_bprm(bprm);
1389
1390 out_files:
1391 if (displaced)
1392 reset_files_struct(displaced);
1393 out_ret:
1394 return retval;
1395 }
1396
1397 int set_binfmt(struct linux_binfmt *new)
1398 {
1399 struct linux_binfmt *old = current->binfmt;
1400
1401 if (new) {
1402 if (!try_module_get(new->module))
1403 return -1;
1404 }
1405 current->binfmt = new;
1406 if (old)
1407 module_put(old->module);
1408 return 0;
1409 }
1410
1411 EXPORT_SYMBOL(set_binfmt);
1412
1413 /* format_corename will inspect the pattern parameter, and output a
1414 * name into corename, which must have space for at least
1415 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1416 */
1417 static int format_corename(char *corename, long signr)
1418 {
1419 const struct cred *cred = current_cred();
1420 const char *pat_ptr = core_pattern;
1421 int ispipe = (*pat_ptr == '|');
1422 char *out_ptr = corename;
1423 char *const out_end = corename + CORENAME_MAX_SIZE;
1424 int rc;
1425 int pid_in_pattern = 0;
1426
1427 /* Repeat as long as we have more pattern to process and more output
1428 space */
1429 while (*pat_ptr) {
1430 if (*pat_ptr != '%') {
1431 if (out_ptr == out_end)
1432 goto out;
1433 *out_ptr++ = *pat_ptr++;
1434 } else {
1435 switch (*++pat_ptr) {
1436 case 0:
1437 goto out;
1438 /* Double percent, output one percent */
1439 case '%':
1440 if (out_ptr == out_end)
1441 goto out;
1442 *out_ptr++ = '%';
1443 break;
1444 /* pid */
1445 case 'p':
1446 pid_in_pattern = 1;
1447 rc = snprintf(out_ptr, out_end - out_ptr,
1448 "%d", task_tgid_vnr(current));
1449 if (rc > out_end - out_ptr)
1450 goto out;
1451 out_ptr += rc;
1452 break;
1453 /* uid */
1454 case 'u':
1455 rc = snprintf(out_ptr, out_end - out_ptr,
1456 "%d", cred->uid);
1457 if (rc > out_end - out_ptr)
1458 goto out;
1459 out_ptr += rc;
1460 break;
1461 /* gid */
1462 case 'g':
1463 rc = snprintf(out_ptr, out_end - out_ptr,
1464 "%d", cred->gid);
1465 if (rc > out_end - out_ptr)
1466 goto out;
1467 out_ptr += rc;
1468 break;
1469 /* signal that caused the coredump */
1470 case 's':
1471 rc = snprintf(out_ptr, out_end - out_ptr,
1472 "%ld", signr);
1473 if (rc > out_end - out_ptr)
1474 goto out;
1475 out_ptr += rc;
1476 break;
1477 /* UNIX time of coredump */
1478 case 't': {
1479 struct timeval tv;
1480 do_gettimeofday(&tv);
1481 rc = snprintf(out_ptr, out_end - out_ptr,
1482 "%lu", tv.tv_sec);
1483 if (rc > out_end - out_ptr)
1484 goto out;
1485 out_ptr += rc;
1486 break;
1487 }
1488 /* hostname */
1489 case 'h':
1490 down_read(&uts_sem);
1491 rc = snprintf(out_ptr, out_end - out_ptr,
1492 "%s", utsname()->nodename);
1493 up_read(&uts_sem);
1494 if (rc > out_end - out_ptr)
1495 goto out;
1496 out_ptr += rc;
1497 break;
1498 /* executable */
1499 case 'e':
1500 rc = snprintf(out_ptr, out_end - out_ptr,
1501 "%s", current->comm);
1502 if (rc > out_end - out_ptr)
1503 goto out;
1504 out_ptr += rc;
1505 break;
1506 /* core limit size */
1507 case 'c':
1508 rc = snprintf(out_ptr, out_end - out_ptr,
1509 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1510 if (rc > out_end - out_ptr)
1511 goto out;
1512 out_ptr += rc;
1513 break;
1514 default:
1515 break;
1516 }
1517 ++pat_ptr;
1518 }
1519 }
1520 /* Backward compatibility with core_uses_pid:
1521 *
1522 * If core_pattern does not include a %p (as is the default)
1523 * and core_uses_pid is set, then .%pid will be appended to
1524 * the filename. Do not do this for piped commands. */
1525 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1526 rc = snprintf(out_ptr, out_end - out_ptr,
1527 ".%d", task_tgid_vnr(current));
1528 if (rc > out_end - out_ptr)
1529 goto out;
1530 out_ptr += rc;
1531 }
1532 out:
1533 *out_ptr = 0;
1534 return ispipe;
1535 }
1536
1537 static int zap_process(struct task_struct *start)
1538 {
1539 struct task_struct *t;
1540 int nr = 0;
1541
1542 start->signal->flags = SIGNAL_GROUP_EXIT;
1543 start->signal->group_stop_count = 0;
1544
1545 t = start;
1546 do {
1547 if (t != current && t->mm) {
1548 sigaddset(&t->pending.signal, SIGKILL);
1549 signal_wake_up(t, 1);
1550 nr++;
1551 }
1552 } while_each_thread(start, t);
1553
1554 return nr;
1555 }
1556
1557 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1558 struct core_state *core_state, int exit_code)
1559 {
1560 struct task_struct *g, *p;
1561 unsigned long flags;
1562 int nr = -EAGAIN;
1563
1564 spin_lock_irq(&tsk->sighand->siglock);
1565 if (!signal_group_exit(tsk->signal)) {
1566 mm->core_state = core_state;
1567 tsk->signal->group_exit_code = exit_code;
1568 nr = zap_process(tsk);
1569 }
1570 spin_unlock_irq(&tsk->sighand->siglock);
1571 if (unlikely(nr < 0))
1572 return nr;
1573
1574 if (atomic_read(&mm->mm_users) == nr + 1)
1575 goto done;
1576 /*
1577 * We should find and kill all tasks which use this mm, and we should
1578 * count them correctly into ->nr_threads. We don't take tasklist
1579 * lock, but this is safe wrt:
1580 *
1581 * fork:
1582 * None of sub-threads can fork after zap_process(leader). All
1583 * processes which were created before this point should be
1584 * visible to zap_threads() because copy_process() adds the new
1585 * process to the tail of init_task.tasks list, and lock/unlock
1586 * of ->siglock provides a memory barrier.
1587 *
1588 * do_exit:
1589 * The caller holds mm->mmap_sem. This means that the task which
1590 * uses this mm can't pass exit_mm(), so it can't exit or clear
1591 * its ->mm.
1592 *
1593 * de_thread:
1594 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1595 * we must see either old or new leader, this does not matter.
1596 * However, it can change p->sighand, so lock_task_sighand(p)
1597 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1598 * it can't fail.
1599 *
1600 * Note also that "g" can be the old leader with ->mm == NULL
1601 * and already unhashed and thus removed from ->thread_group.
1602 * This is OK, __unhash_process()->list_del_rcu() does not
1603 * clear the ->next pointer, we will find the new leader via
1604 * next_thread().
1605 */
1606 rcu_read_lock();
1607 for_each_process(g) {
1608 if (g == tsk->group_leader)
1609 continue;
1610 if (g->flags & PF_KTHREAD)
1611 continue;
1612 p = g;
1613 do {
1614 if (p->mm) {
1615 if (unlikely(p->mm == mm)) {
1616 lock_task_sighand(p, &flags);
1617 nr += zap_process(p);
1618 unlock_task_sighand(p, &flags);
1619 }
1620 break;
1621 }
1622 } while_each_thread(g, p);
1623 }
1624 rcu_read_unlock();
1625 done:
1626 atomic_set(&core_state->nr_threads, nr);
1627 return nr;
1628 }
1629
1630 static int coredump_wait(int exit_code, struct core_state *core_state)
1631 {
1632 struct task_struct *tsk = current;
1633 struct mm_struct *mm = tsk->mm;
1634 struct completion *vfork_done;
1635 int core_waiters;
1636
1637 init_completion(&core_state->startup);
1638 core_state->dumper.task = tsk;
1639 core_state->dumper.next = NULL;
1640 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1641 up_write(&mm->mmap_sem);
1642
1643 if (unlikely(core_waiters < 0))
1644 goto fail;
1645
1646 /*
1647 * Make sure nobody is waiting for us to release the VM,
1648 * otherwise we can deadlock when we wait on each other
1649 */
1650 vfork_done = tsk->vfork_done;
1651 if (vfork_done) {
1652 tsk->vfork_done = NULL;
1653 complete(vfork_done);
1654 }
1655
1656 if (core_waiters)
1657 wait_for_completion(&core_state->startup);
1658 fail:
1659 return core_waiters;
1660 }
1661
1662 static void coredump_finish(struct mm_struct *mm)
1663 {
1664 struct core_thread *curr, *next;
1665 struct task_struct *task;
1666
1667 next = mm->core_state->dumper.next;
1668 while ((curr = next) != NULL) {
1669 next = curr->next;
1670 task = curr->task;
1671 /*
1672 * see exit_mm(), curr->task must not see
1673 * ->task == NULL before we read ->next.
1674 */
1675 smp_mb();
1676 curr->task = NULL;
1677 wake_up_process(task);
1678 }
1679
1680 mm->core_state = NULL;
1681 }
1682
1683 /*
1684 * set_dumpable converts traditional three-value dumpable to two flags and
1685 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1686 * these bits are not changed atomically. So get_dumpable can observe the
1687 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1688 * return either old dumpable or new one by paying attention to the order of
1689 * modifying the bits.
1690 *
1691 * dumpable | mm->flags (binary)
1692 * old new | initial interim final
1693 * ---------+-----------------------
1694 * 0 1 | 00 01 01
1695 * 0 2 | 00 10(*) 11
1696 * 1 0 | 01 00 00
1697 * 1 2 | 01 11 11
1698 * 2 0 | 11 10(*) 00
1699 * 2 1 | 11 11 01
1700 *
1701 * (*) get_dumpable regards interim value of 10 as 11.
1702 */
1703 void set_dumpable(struct mm_struct *mm, int value)
1704 {
1705 switch (value) {
1706 case 0:
1707 clear_bit(MMF_DUMPABLE, &mm->flags);
1708 smp_wmb();
1709 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1710 break;
1711 case 1:
1712 set_bit(MMF_DUMPABLE, &mm->flags);
1713 smp_wmb();
1714 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1715 break;
1716 case 2:
1717 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1718 smp_wmb();
1719 set_bit(MMF_DUMPABLE, &mm->flags);
1720 break;
1721 }
1722 }
1723
1724 int get_dumpable(struct mm_struct *mm)
1725 {
1726 int ret;
1727
1728 ret = mm->flags & 0x3;
1729 return (ret >= 2) ? 2 : ret;
1730 }
1731
1732 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1733 {
1734 struct core_state core_state;
1735 char corename[CORENAME_MAX_SIZE + 1];
1736 struct mm_struct *mm = current->mm;
1737 struct linux_binfmt * binfmt;
1738 struct inode * inode;
1739 struct file * file;
1740 const struct cred *old_cred;
1741 struct cred *cred;
1742 int retval = 0;
1743 int flag = 0;
1744 int ispipe = 0;
1745 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1746 char **helper_argv = NULL;
1747 int helper_argc = 0;
1748 char *delimit;
1749
1750 audit_core_dumps(signr);
1751
1752 binfmt = current->binfmt;
1753 if (!binfmt || !binfmt->core_dump)
1754 goto fail;
1755
1756 cred = prepare_creds();
1757 if (!cred) {
1758 retval = -ENOMEM;
1759 goto fail;
1760 }
1761
1762 down_write(&mm->mmap_sem);
1763 /*
1764 * If another thread got here first, or we are not dumpable, bail out.
1765 */
1766 if (mm->core_state || !get_dumpable(mm)) {
1767 up_write(&mm->mmap_sem);
1768 put_cred(cred);
1769 goto fail;
1770 }
1771
1772 /*
1773 * We cannot trust fsuid as being the "true" uid of the
1774 * process nor do we know its entire history. We only know it
1775 * was tainted so we dump it as root in mode 2.
1776 */
1777 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1778 flag = O_EXCL; /* Stop rewrite attacks */
1779 cred->fsuid = 0; /* Dump root private */
1780 }
1781
1782 retval = coredump_wait(exit_code, &core_state);
1783 if (retval < 0) {
1784 put_cred(cred);
1785 goto fail;
1786 }
1787
1788 old_cred = override_creds(cred);
1789
1790 /*
1791 * Clear any false indication of pending signals that might
1792 * be seen by the filesystem code called to write the core file.
1793 */
1794 clear_thread_flag(TIF_SIGPENDING);
1795
1796 /*
1797 * lock_kernel() because format_corename() is controlled by sysctl, which
1798 * uses lock_kernel()
1799 */
1800 lock_kernel();
1801 ispipe = format_corename(corename, signr);
1802 unlock_kernel();
1803 /*
1804 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1805 * to a pipe. Since we're not writing directly to the filesystem
1806 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1807 * created unless the pipe reader choses to write out the core file
1808 * at which point file size limits and permissions will be imposed
1809 * as it does with any other process
1810 */
1811 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1812 goto fail_unlock;
1813
1814 if (ispipe) {
1815 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1816 if (!helper_argv) {
1817 printk(KERN_WARNING "%s failed to allocate memory\n",
1818 __func__);
1819 goto fail_unlock;
1820 }
1821 /* Terminate the string before the first option */
1822 delimit = strchr(corename, ' ');
1823 if (delimit)
1824 *delimit = '\0';
1825 delimit = strrchr(helper_argv[0], '/');
1826 if (delimit)
1827 delimit++;
1828 else
1829 delimit = helper_argv[0];
1830 if (!strcmp(delimit, current->comm)) {
1831 printk(KERN_NOTICE "Recursive core dump detected, "
1832 "aborting\n");
1833 goto fail_unlock;
1834 }
1835
1836 core_limit = RLIM_INFINITY;
1837
1838 /* SIGPIPE can happen, but it's just never processed */
1839 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1840 &file)) {
1841 printk(KERN_INFO "Core dump to %s pipe failed\n",
1842 corename);
1843 goto fail_unlock;
1844 }
1845 } else
1846 file = filp_open(corename,
1847 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1848 0600);
1849 if (IS_ERR(file))
1850 goto fail_unlock;
1851 inode = file->f_path.dentry->d_inode;
1852 if (inode->i_nlink > 1)
1853 goto close_fail; /* multiple links - don't dump */
1854 if (!ispipe && d_unhashed(file->f_path.dentry))
1855 goto close_fail;
1856
1857 /* AK: actually i see no reason to not allow this for named pipes etc.,
1858 but keep the previous behaviour for now. */
1859 if (!ispipe && !S_ISREG(inode->i_mode))
1860 goto close_fail;
1861 /*
1862 * Dont allow local users get cute and trick others to coredump
1863 * into their pre-created files:
1864 * Note, this is not relevant for pipes
1865 */
1866 if (!ispipe && (inode->i_uid != current_fsuid()))
1867 goto close_fail;
1868 if (!file->f_op)
1869 goto close_fail;
1870 if (!file->f_op->write)
1871 goto close_fail;
1872 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1873 goto close_fail;
1874
1875 retval = binfmt->core_dump(signr, regs, file, core_limit);
1876
1877 if (retval)
1878 current->signal->group_exit_code |= 0x80;
1879 close_fail:
1880 filp_close(file, NULL);
1881 fail_unlock:
1882 if (helper_argv)
1883 argv_free(helper_argv);
1884
1885 revert_creds(old_cred);
1886 put_cred(cred);
1887 coredump_finish(mm);
1888 fail:
1889 return;
1890 }
1891
|
This page was automatically generated by the
LXR engine.
|