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