netlink: Unneeded local variable
[linux-2.6/mini2440.git] / fs / exec.c
blob9448f1b50b4a07a8969d8587d044094bbfcfce8a
1 /*
2 * linux/fs/exec.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
7 /*
8 * #!-checking implemented by tytso.
9 */
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.
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.
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.
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/mman.h>
29 #include <linux/a.out.h>
30 #include <linux/stat.h>
31 #include <linux/fcntl.h>
32 #include <linux/smp_lock.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/swap.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/ptrace.h>
48 #include <linux/mount.h>
49 #include <linux/security.h>
50 #include <linux/syscalls.h>
51 #include <linux/rmap.h>
52 #include <linux/tsacct_kern.h>
53 #include <linux/cn_proc.h>
54 #include <linux/audit.h>
56 #include <asm/uaccess.h>
57 #include <asm/mmu_context.h>
58 #include <asm/tlb.h>
60 #ifdef CONFIG_KMOD
61 #include <linux/kmod.h>
62 #endif
64 int core_uses_pid;
65 char core_pattern[CORENAME_MAX_SIZE] = "core";
66 int suid_dumpable = 0;
68 /* The maximal length of core_pattern is also specified in sysctl.c */
70 static LIST_HEAD(formats);
71 static DEFINE_RWLOCK(binfmt_lock);
73 int register_binfmt(struct linux_binfmt * fmt)
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;
83 EXPORT_SYMBOL(register_binfmt);
85 void unregister_binfmt(struct linux_binfmt * fmt)
87 write_lock(&binfmt_lock);
88 list_del(&fmt->lh);
89 write_unlock(&binfmt_lock);
92 EXPORT_SYMBOL(unregister_binfmt);
94 static inline void put_binfmt(struct linux_binfmt * fmt)
96 module_put(fmt->module);
100 * Note that a shared library must be both readable and executable due to
101 * security reasons.
103 * Also note that we take the address to load from from the file itself.
105 asmlinkage long sys_uselib(const char __user * library)
107 struct file * file;
108 struct nameidata nd;
109 int error;
111 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
112 if (error)
113 goto out;
115 error = -EINVAL;
116 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
117 goto exit;
119 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
120 if (error)
121 goto exit;
123 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
124 error = PTR_ERR(file);
125 if (IS_ERR(file))
126 goto out;
128 error = -ENOEXEC;
129 if(file->f_op) {
130 struct linux_binfmt * fmt;
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;
145 read_unlock(&binfmt_lock);
147 fput(file);
148 out:
149 return error;
150 exit:
151 release_open_intent(&nd);
152 path_put(&nd.path);
153 goto out;
156 #ifdef CONFIG_MMU
158 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
159 int write)
161 struct page *page;
162 int ret;
164 #ifdef CONFIG_STACK_GROWSUP
165 if (write) {
166 ret = expand_stack_downwards(bprm->vma, pos);
167 if (ret < 0)
168 return NULL;
170 #endif
171 ret = get_user_pages(current, bprm->mm, pos,
172 1, write, 1, &page, NULL);
173 if (ret <= 0)
174 return NULL;
176 if (write) {
177 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
178 struct rlimit *rlim;
181 * We've historically supported up to 32 pages (ARG_MAX)
182 * of argument strings even with small stacks
184 if (size <= ARG_MAX)
185 return page;
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.
194 rlim = current->signal->rlim;
195 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
196 put_page(page);
197 return NULL;
201 return page;
204 static void put_arg_page(struct page *page)
206 put_page(page);
209 static void free_arg_page(struct linux_binprm *bprm, int i)
213 static void free_arg_pages(struct linux_binprm *bprm)
217 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
218 struct page *page)
220 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
223 static int __bprm_mm_init(struct linux_binprm *bprm)
225 int err = -ENOMEM;
226 struct vm_area_struct *vma = NULL;
227 struct mm_struct *mm = bprm->mm;
229 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
230 if (!vma)
231 goto err;
233 down_write(&mm->mmap_sem);
234 vma->vm_mm = mm;
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.
242 vma->vm_end = STACK_TOP_MAX;
243 vma->vm_start = vma->vm_end - PAGE_SIZE;
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;
253 mm->stack_vm = mm->total_vm = 1;
254 up_write(&mm->mmap_sem);
256 bprm->p = vma->vm_end - sizeof(void *);
258 return 0;
260 err:
261 if (vma) {
262 bprm->vma = NULL;
263 kmem_cache_free(vm_area_cachep, vma);
266 return err;
269 static bool valid_arg_len(struct linux_binprm *bprm, long len)
271 return len <= MAX_ARG_STRLEN;
274 #else
276 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
277 int write)
279 struct page *page;
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;
289 return page;
292 static void put_arg_page(struct page *page)
296 static void free_arg_page(struct linux_binprm *bprm, int i)
298 if (bprm->page[i]) {
299 __free_page(bprm->page[i]);
300 bprm->page[i] = NULL;
304 static void free_arg_pages(struct linux_binprm *bprm)
306 int i;
308 for (i = 0; i < MAX_ARG_PAGES; i++)
309 free_arg_page(bprm, i);
312 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
313 struct page *page)
317 static int __bprm_mm_init(struct linux_binprm *bprm)
319 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
320 return 0;
323 static bool valid_arg_len(struct linux_binprm *bprm, long len)
325 return len <= bprm->p;
328 #endif /* CONFIG_MMU */
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().
336 int bprm_mm_init(struct linux_binprm *bprm)
338 int err;
339 struct mm_struct *mm = NULL;
341 bprm->mm = mm = mm_alloc();
342 err = -ENOMEM;
343 if (!mm)
344 goto err;
346 err = init_new_context(current, mm);
347 if (err)
348 goto err;
350 err = __bprm_mm_init(bprm);
351 if (err)
352 goto err;
354 return 0;
356 err:
357 if (mm) {
358 bprm->mm = NULL;
359 mmdrop(mm);
362 return err;
366 * count() counts the number of strings in array ARGV.
368 static int count(char __user * __user * argv, int max)
370 int i = 0;
372 if (argv != NULL) {
373 for (;;) {
374 char __user * p;
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();
386 return i;
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.
394 static int copy_strings(int argc, char __user * __user * argv,
395 struct linux_binprm *bprm)
397 struct page *kmapped_page = NULL;
398 char *kaddr = NULL;
399 unsigned long kpos = 0;
400 int ret;
402 while (argc-- > 0) {
403 char __user *str;
404 int len;
405 unsigned long pos;
407 if (get_user(str, argv+argc) ||
408 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
409 ret = -EFAULT;
410 goto out;
413 if (!valid_arg_len(bprm, len)) {
414 ret = -E2BIG;
415 goto out;
418 /* We're going to work our way backwords. */
419 pos = bprm->p;
420 str += len;
421 bprm->p -= len;
423 while (len > 0) {
424 int offset, bytes_to_copy;
426 offset = pos % PAGE_SIZE;
427 if (offset == 0)
428 offset = PAGE_SIZE;
430 bytes_to_copy = offset;
431 if (bytes_to_copy > len)
432 bytes_to_copy = len;
434 offset -= bytes_to_copy;
435 pos -= bytes_to_copy;
436 str -= bytes_to_copy;
437 len -= bytes_to_copy;
439 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
440 struct page *page;
442 page = get_arg_page(bprm, pos, 1);
443 if (!page) {
444 ret = -E2BIG;
445 goto out;
448 if (kmapped_page) {
449 flush_kernel_dcache_page(kmapped_page);
450 kunmap(kmapped_page);
451 put_arg_page(kmapped_page);
453 kmapped_page = page;
454 kaddr = kmap(kmapped_page);
455 kpos = pos & PAGE_MASK;
456 flush_arg_page(bprm, kpos, kmapped_page);
458 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
459 ret = -EFAULT;
460 goto out;
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);
471 return ret;
475 * Like copy_strings, but get argv and its values from kernel memory.
477 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
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;
486 EXPORT_SYMBOL(copy_strings_kernel);
488 #ifdef CONFIG_MMU
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:
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.
502 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
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;
512 BUG_ON(new_start > new_end);
515 * ensure there are no vmas between where we want to go
516 * and where we are
518 if (vma != find_vma(mm, new_start))
519 return -EFAULT;
522 * cover the whole range: [new_start, old_end)
524 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
527 * move the page tables downwards, on failure we rely on
528 * process cleanup to remove whatever mess we made.
530 if (length != move_page_tables(vma, old_start,
531 vma, new_start, length))
532 return -ENOMEM;
534 lru_add_drain();
535 tlb = tlb_gather_mmu(mm, 0);
536 if (new_end > old_start) {
538 * when the old and new regions overlap clear from new_end.
540 free_pgd_range(&tlb, new_end, old_end, new_end,
541 vma->vm_next ? vma->vm_next->vm_start : 0);
542 } else {
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.
549 free_pgd_range(&tlb, old_start, old_end, new_end,
550 vma->vm_next ? vma->vm_next->vm_start : 0);
552 tlb_finish_mmu(tlb, new_end, old_end);
555 * shrink the vma to just the new range.
557 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
559 return 0;
562 #define EXTRA_STACK_VM_PAGES 20 /* random */
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.
568 int setup_arg_pages(struct linux_binprm *bprm,
569 unsigned long stack_top,
570 int executable_stack)
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;
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;
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;
590 stack_base = PAGE_ALIGN(stack_top - stack_base);
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;
600 bprm->p -= stack_shift;
601 mm->arg_start = bprm->p;
602 #endif
604 if (bprm->loader)
605 bprm->loader -= stack_shift;
606 bprm->exec -= stack_shift;
608 down_write(&mm->mmap_sem);
609 vm_flags = vma->vm_flags;
612 * Adjust stack execute permissions; explicitly enable for
613 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
614 * (arch default) otherwise.
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;
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);
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;
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;
646 out_unlock:
647 up_write(&mm->mmap_sem);
648 return 0;
650 EXPORT_SYMBOL(setup_arg_pages);
652 #endif /* CONFIG_MMU */
654 struct file *open_exec(const char *name)
656 struct nameidata nd;
657 int err;
658 struct file *file;
660 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
661 file = ERR_PTR(err);
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);
679 out:
680 return file;
683 release_open_intent(&nd);
684 path_put(&nd.path);
686 goto out;
689 EXPORT_SYMBOL(open_exec);
691 int kernel_read(struct file *file, unsigned long offset,
692 char *addr, unsigned long count)
694 mm_segment_t old_fs;
695 loff_t pos = offset;
696 int result;
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;
706 EXPORT_SYMBOL(kernel_read);
708 static int exec_mmap(struct mm_struct *mm)
710 struct task_struct *tsk;
711 struct mm_struct * old_mm, *active_mm;
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);
718 if (old_mm) {
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.
727 down_read(&old_mm->mmap_sem);
728 if (unlikely(old_mm->core_waiters)) {
729 up_read(&old_mm->mmap_sem);
730 return -EINTR;
733 task_lock(tsk);
734 active_mm = tsk->active_mm;
735 tsk->mm = mm;
736 tsk->active_mm = mm;
737 activate_mm(active_mm, mm);
738 task_unlock(tsk);
739 mm_update_next_owner(old_mm);
740 arch_pick_mmap_layout(mm);
741 if (old_mm) {
742 up_read(&old_mm->mmap_sem);
743 BUG_ON(active_mm != old_mm);
744 mmput(old_mm);
745 return 0;
747 mmdrop(active_mm);
748 return 0;
752 * This function makes sure the current process has its own signal table,
753 * so that flush_signal_handlers can later reset the handlers without
754 * disturbing other processes. (Other processes might share the signal
755 * table via the CLONE_SIGHAND option to clone().)
757 static int de_thread(struct task_struct *tsk)
759 struct signal_struct *sig = tsk->signal;
760 struct sighand_struct *oldsighand = tsk->sighand;
761 spinlock_t *lock = &oldsighand->siglock;
762 struct task_struct *leader = NULL;
763 int count;
765 if (thread_group_empty(tsk))
766 goto no_thread_group;
769 * Kill all other threads in the thread group.
771 spin_lock_irq(lock);
772 if (signal_group_exit(sig)) {
774 * Another group action in progress, just
775 * return so that the signal is processed.
777 spin_unlock_irq(lock);
778 return -EAGAIN;
780 sig->group_exit_task = tsk;
781 zap_other_threads(tsk);
783 /* Account for the thread group leader hanging around: */
784 count = thread_group_leader(tsk) ? 1 : 2;
785 sig->notify_count = count;
786 while (atomic_read(&sig->count) > count) {
787 __set_current_state(TASK_UNINTERRUPTIBLE);
788 spin_unlock_irq(lock);
789 schedule();
790 spin_lock_irq(lock);
792 spin_unlock_irq(lock);
795 * At this point all other threads have exited, all we have to
796 * do is to wait for the thread group leader to become inactive,
797 * and to assume its PID:
799 if (!thread_group_leader(tsk)) {
800 leader = tsk->group_leader;
802 sig->notify_count = -1; /* for exit_notify() */
803 for (;;) {
804 write_lock_irq(&tasklist_lock);
805 if (likely(leader->exit_state))
806 break;
807 __set_current_state(TASK_UNINTERRUPTIBLE);
808 write_unlock_irq(&tasklist_lock);
809 schedule();
812 if (unlikely(task_child_reaper(tsk) == leader))
813 task_active_pid_ns(tsk)->child_reaper = tsk;
815 * The only record we have of the real-time age of a
816 * process, regardless of execs it's done, is start_time.
817 * All the past CPU time is accumulated in signal_struct
818 * from sister threads now dead. But in this non-leader
819 * exec, nothing survives from the original leader thread,
820 * whose birth marks the true age of this process now.
821 * When we take on its identity by switching to its PID, we
822 * also take its birthdate (always earlier than our own).
824 tsk->start_time = leader->start_time;
826 BUG_ON(!same_thread_group(leader, tsk));
827 BUG_ON(has_group_leader_pid(tsk));
829 * An exec() starts a new thread group with the
830 * TGID of the previous thread group. Rehash the
831 * two threads with a switched PID, and release
832 * the former thread group leader:
835 /* Become a process group leader with the old leader's pid.
836 * The old leader becomes a thread of the this thread group.
837 * Note: The old leader also uses this pid until release_task
838 * is called. Odd but simple and correct.
840 detach_pid(tsk, PIDTYPE_PID);
841 tsk->pid = leader->pid;
842 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
843 transfer_pid(leader, tsk, PIDTYPE_PGID);
844 transfer_pid(leader, tsk, PIDTYPE_SID);
845 list_replace_rcu(&leader->tasks, &tsk->tasks);
847 tsk->group_leader = tsk;
848 leader->group_leader = tsk;
850 tsk->exit_signal = SIGCHLD;
852 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
853 leader->exit_state = EXIT_DEAD;
855 write_unlock_irq(&tasklist_lock);
858 sig->group_exit_task = NULL;
859 sig->notify_count = 0;
861 no_thread_group:
862 exit_itimers(sig);
863 flush_itimer_signals();
864 if (leader)
865 release_task(leader);
867 if (atomic_read(&oldsighand->count) != 1) {
868 struct sighand_struct *newsighand;
870 * This ->sighand is shared with the CLONE_SIGHAND
871 * but not CLONE_THREAD task, switch to the new one.
873 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
874 if (!newsighand)
875 return -ENOMEM;
877 atomic_set(&newsighand->count, 1);
878 memcpy(newsighand->action, oldsighand->action,
879 sizeof(newsighand->action));
881 write_lock_irq(&tasklist_lock);
882 spin_lock(&oldsighand->siglock);
883 rcu_assign_pointer(tsk->sighand, newsighand);
884 spin_unlock(&oldsighand->siglock);
885 write_unlock_irq(&tasklist_lock);
887 __cleanup_sighand(oldsighand);
890 BUG_ON(!thread_group_leader(tsk));
891 return 0;
895 * These functions flushes out all traces of the currently running executable
896 * so that a new one can be started
898 static void flush_old_files(struct files_struct * files)
900 long j = -1;
901 struct fdtable *fdt;
903 spin_lock(&files->file_lock);
904 for (;;) {
905 unsigned long set, i;
907 j++;
908 i = j * __NFDBITS;
909 fdt = files_fdtable(files);
910 if (i >= fdt->max_fds)
911 break;
912 set = fdt->close_on_exec->fds_bits[j];
913 if (!set)
914 continue;
915 fdt->close_on_exec->fds_bits[j] = 0;
916 spin_unlock(&files->file_lock);
917 for ( ; set ; i++,set >>= 1) {
918 if (set & 1) {
919 sys_close(i);
922 spin_lock(&files->file_lock);
925 spin_unlock(&files->file_lock);
928 char *get_task_comm(char *buf, struct task_struct *tsk)
930 /* buf must be at least sizeof(tsk->comm) in size */
931 task_lock(tsk);
932 strncpy(buf, tsk->comm, sizeof(tsk->comm));
933 task_unlock(tsk);
934 return buf;
937 void set_task_comm(struct task_struct *tsk, char *buf)
939 task_lock(tsk);
940 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
941 task_unlock(tsk);
944 int flush_old_exec(struct linux_binprm * bprm)
946 char * name;
947 int i, ch, retval;
948 char tcomm[sizeof(current->comm)];
951 * Make sure we have a private signal table and that
952 * we are unassociated from the previous thread group.
954 retval = de_thread(current);
955 if (retval)
956 goto out;
958 set_mm_exe_file(bprm->mm, bprm->file);
961 * Release all of the old mmap stuff
963 retval = exec_mmap(bprm->mm);
964 if (retval)
965 goto out;
967 bprm->mm = NULL; /* We're using it now */
969 /* This is the point of no return */
970 current->sas_ss_sp = current->sas_ss_size = 0;
972 if (current->euid == current->uid && current->egid == current->gid)
973 set_dumpable(current->mm, 1);
974 else
975 set_dumpable(current->mm, suid_dumpable);
977 name = bprm->filename;
979 /* Copies the binary name from after last slash */
980 for (i=0; (ch = *(name++)) != '\0';) {
981 if (ch == '/')
982 i = 0; /* overwrite what we wrote */
983 else
984 if (i < (sizeof(tcomm) - 1))
985 tcomm[i++] = ch;
987 tcomm[i] = '\0';
988 set_task_comm(current, tcomm);
990 current->flags &= ~PF_RANDOMIZE;
991 flush_thread();
993 /* Set the new mm task size. We have to do that late because it may
994 * depend on TIF_32BIT which is only updated in flush_thread() on
995 * some architectures like powerpc
997 current->mm->task_size = TASK_SIZE;
999 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1000 suid_keys(current);
1001 set_dumpable(current->mm, suid_dumpable);
1002 current->pdeath_signal = 0;
1003 } else if (file_permission(bprm->file, MAY_READ) ||
1004 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1005 suid_keys(current);
1006 set_dumpable(current->mm, suid_dumpable);
1009 /* An exec changes our domain. We are no longer part of the thread
1010 group */
1012 current->self_exec_id++;
1014 flush_signal_handlers(current, 0);
1015 flush_old_files(current->files);
1017 return 0;
1019 out:
1020 return retval;
1023 EXPORT_SYMBOL(flush_old_exec);
1026 * Fill the binprm structure from the inode.
1027 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1029 int prepare_binprm(struct linux_binprm *bprm)
1031 int mode;
1032 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1033 int retval;
1035 mode = inode->i_mode;
1036 if (bprm->file->f_op == NULL)
1037 return -EACCES;
1039 bprm->e_uid = current->euid;
1040 bprm->e_gid = current->egid;
1042 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1043 /* Set-uid? */
1044 if (mode & S_ISUID) {
1045 current->personality &= ~PER_CLEAR_ON_SETID;
1046 bprm->e_uid = inode->i_uid;
1049 /* Set-gid? */
1051 * If setgid is set but no group execute bit then this
1052 * is a candidate for mandatory locking, not a setgid
1053 * executable.
1055 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1056 current->personality &= ~PER_CLEAR_ON_SETID;
1057 bprm->e_gid = inode->i_gid;
1061 /* fill in binprm security blob */
1062 retval = security_bprm_set(bprm);
1063 if (retval)
1064 return retval;
1066 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1067 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1070 EXPORT_SYMBOL(prepare_binprm);
1072 static int unsafe_exec(struct task_struct *p)
1074 int unsafe = 0;
1075 if (p->ptrace & PT_PTRACED) {
1076 if (p->ptrace & PT_PTRACE_CAP)
1077 unsafe |= LSM_UNSAFE_PTRACE_CAP;
1078 else
1079 unsafe |= LSM_UNSAFE_PTRACE;
1081 if (atomic_read(&p->fs->count) > 1 ||
1082 atomic_read(&p->files->count) > 1 ||
1083 atomic_read(&p->sighand->count) > 1)
1084 unsafe |= LSM_UNSAFE_SHARE;
1086 return unsafe;
1089 void compute_creds(struct linux_binprm *bprm)
1091 int unsafe;
1093 if (bprm->e_uid != current->uid) {
1094 suid_keys(current);
1095 current->pdeath_signal = 0;
1097 exec_keys(current);
1099 task_lock(current);
1100 unsafe = unsafe_exec(current);
1101 security_bprm_apply_creds(bprm, unsafe);
1102 task_unlock(current);
1103 security_bprm_post_apply_creds(bprm);
1105 EXPORT_SYMBOL(compute_creds);
1108 * Arguments are '\0' separated strings found at the location bprm->p
1109 * points to; chop off the first by relocating brpm->p to right after
1110 * the first '\0' encountered.
1112 int remove_arg_zero(struct linux_binprm *bprm)
1114 int ret = 0;
1115 unsigned long offset;
1116 char *kaddr;
1117 struct page *page;
1119 if (!bprm->argc)
1120 return 0;
1122 do {
1123 offset = bprm->p & ~PAGE_MASK;
1124 page = get_arg_page(bprm, bprm->p, 0);
1125 if (!page) {
1126 ret = -EFAULT;
1127 goto out;
1129 kaddr = kmap_atomic(page, KM_USER0);
1131 for (; offset < PAGE_SIZE && kaddr[offset];
1132 offset++, bprm->p++)
1135 kunmap_atomic(kaddr, KM_USER0);
1136 put_arg_page(page);
1138 if (offset == PAGE_SIZE)
1139 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1140 } while (offset == PAGE_SIZE);
1142 bprm->p++;
1143 bprm->argc--;
1144 ret = 0;
1146 out:
1147 return ret;
1149 EXPORT_SYMBOL(remove_arg_zero);
1152 * cycle the list of binary formats handler, until one recognizes the image
1154 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1156 int try,retval;
1157 struct linux_binfmt *fmt;
1158 #if defined(__alpha__) && defined(CONFIG_ARCH_SUPPORTS_AOUT)
1159 /* handle /sbin/loader.. */
1161 struct exec * eh = (struct exec *) bprm->buf;
1163 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1164 (eh->fh.f_flags & 0x3000) == 0x3000)
1166 struct file * file;
1167 unsigned long loader;
1169 allow_write_access(bprm->file);
1170 fput(bprm->file);
1171 bprm->file = NULL;
1173 loader = bprm->vma->vm_end - sizeof(void *);
1175 file = open_exec("/sbin/loader");
1176 retval = PTR_ERR(file);
1177 if (IS_ERR(file))
1178 return retval;
1180 /* Remember if the application is TASO. */
1181 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1183 bprm->file = file;
1184 bprm->loader = loader;
1185 retval = prepare_binprm(bprm);
1186 if (retval<0)
1187 return retval;
1188 /* should call search_binary_handler recursively here,
1189 but it does not matter */
1192 #endif
1193 retval = security_bprm_check(bprm);
1194 if (retval)
1195 return retval;
1197 /* kernel module loader fixup */
1198 /* so we don't try to load run modprobe in kernel space. */
1199 set_fs(USER_DS);
1201 retval = audit_bprm(bprm);
1202 if (retval)
1203 return retval;
1205 retval = -ENOENT;
1206 for (try=0; try<2; try++) {
1207 read_lock(&binfmt_lock);
1208 list_for_each_entry(fmt, &formats, lh) {
1209 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1210 if (!fn)
1211 continue;
1212 if (!try_module_get(fmt->module))
1213 continue;
1214 read_unlock(&binfmt_lock);
1215 retval = fn(bprm, regs);
1216 if (retval >= 0) {
1217 put_binfmt(fmt);
1218 allow_write_access(bprm->file);
1219 if (bprm->file)
1220 fput(bprm->file);
1221 bprm->file = NULL;
1222 current->did_exec = 1;
1223 proc_exec_connector(current);
1224 return retval;
1226 read_lock(&binfmt_lock);
1227 put_binfmt(fmt);
1228 if (retval != -ENOEXEC || bprm->mm == NULL)
1229 break;
1230 if (!bprm->file) {
1231 read_unlock(&binfmt_lock);
1232 return retval;
1235 read_unlock(&binfmt_lock);
1236 if (retval != -ENOEXEC || bprm->mm == NULL) {
1237 break;
1238 #ifdef CONFIG_KMOD
1239 }else{
1240 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1241 if (printable(bprm->buf[0]) &&
1242 printable(bprm->buf[1]) &&
1243 printable(bprm->buf[2]) &&
1244 printable(bprm->buf[3]))
1245 break; /* -ENOEXEC */
1246 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1247 #endif
1250 return retval;
1253 EXPORT_SYMBOL(search_binary_handler);
1255 void free_bprm(struct linux_binprm *bprm)
1257 free_arg_pages(bprm);
1258 kfree(bprm);
1262 * sys_execve() executes a new program.
1264 int do_execve(char * filename,
1265 char __user *__user *argv,
1266 char __user *__user *envp,
1267 struct pt_regs * regs)
1269 struct linux_binprm *bprm;
1270 struct file *file;
1271 struct files_struct *displaced;
1272 int retval;
1274 retval = unshare_files(&displaced);
1275 if (retval)
1276 goto out_ret;
1278 retval = -ENOMEM;
1279 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1280 if (!bprm)
1281 goto out_files;
1283 file = open_exec(filename);
1284 retval = PTR_ERR(file);
1285 if (IS_ERR(file))
1286 goto out_kfree;
1288 sched_exec();
1290 bprm->file = file;
1291 bprm->filename = filename;
1292 bprm->interp = filename;
1294 retval = bprm_mm_init(bprm);
1295 if (retval)
1296 goto out_file;
1298 bprm->argc = count(argv, MAX_ARG_STRINGS);
1299 if ((retval = bprm->argc) < 0)
1300 goto out_mm;
1302 bprm->envc = count(envp, MAX_ARG_STRINGS);
1303 if ((retval = bprm->envc) < 0)
1304 goto out_mm;
1306 retval = security_bprm_alloc(bprm);
1307 if (retval)
1308 goto out;
1310 retval = prepare_binprm(bprm);
1311 if (retval < 0)
1312 goto out;
1314 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1315 if (retval < 0)
1316 goto out;
1318 bprm->exec = bprm->p;
1319 retval = copy_strings(bprm->envc, envp, bprm);
1320 if (retval < 0)
1321 goto out;
1323 retval = copy_strings(bprm->argc, argv, bprm);
1324 if (retval < 0)
1325 goto out;
1327 retval = search_binary_handler(bprm,regs);
1328 if (retval >= 0) {
1329 /* execve success */
1330 security_bprm_free(bprm);
1331 acct_update_integrals(current);
1332 free_bprm(bprm);
1333 if (displaced)
1334 put_files_struct(displaced);
1335 return retval;
1338 out:
1339 if (bprm->security)
1340 security_bprm_free(bprm);
1342 out_mm:
1343 if (bprm->mm)
1344 mmput (bprm->mm);
1346 out_file:
1347 if (bprm->file) {
1348 allow_write_access(bprm->file);
1349 fput(bprm->file);
1351 out_kfree:
1352 free_bprm(bprm);
1354 out_files:
1355 if (displaced)
1356 reset_files_struct(displaced);
1357 out_ret:
1358 return retval;
1361 int set_binfmt(struct linux_binfmt *new)
1363 struct linux_binfmt *old = current->binfmt;
1365 if (new) {
1366 if (!try_module_get(new->module))
1367 return -1;
1369 current->binfmt = new;
1370 if (old)
1371 module_put(old->module);
1372 return 0;
1375 EXPORT_SYMBOL(set_binfmt);
1377 /* format_corename will inspect the pattern parameter, and output a
1378 * name into corename, which must have space for at least
1379 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1381 static int format_corename(char *corename, const char *pattern, long signr)
1383 const char *pat_ptr = pattern;
1384 char *out_ptr = corename;
1385 char *const out_end = corename + CORENAME_MAX_SIZE;
1386 int rc;
1387 int pid_in_pattern = 0;
1388 int ispipe = 0;
1390 if (*pattern == '|')
1391 ispipe = 1;
1393 /* Repeat as long as we have more pattern to process and more output
1394 space */
1395 while (*pat_ptr) {
1396 if (*pat_ptr != '%') {
1397 if (out_ptr == out_end)
1398 goto out;
1399 *out_ptr++ = *pat_ptr++;
1400 } else {
1401 switch (*++pat_ptr) {
1402 case 0:
1403 goto out;
1404 /* Double percent, output one percent */
1405 case '%':
1406 if (out_ptr == out_end)
1407 goto out;
1408 *out_ptr++ = '%';
1409 break;
1410 /* pid */
1411 case 'p':
1412 pid_in_pattern = 1;
1413 rc = snprintf(out_ptr, out_end - out_ptr,
1414 "%d", task_tgid_vnr(current));
1415 if (rc > out_end - out_ptr)
1416 goto out;
1417 out_ptr += rc;
1418 break;
1419 /* uid */
1420 case 'u':
1421 rc = snprintf(out_ptr, out_end - out_ptr,
1422 "%d", current->uid);
1423 if (rc > out_end - out_ptr)
1424 goto out;
1425 out_ptr += rc;
1426 break;
1427 /* gid */
1428 case 'g':
1429 rc = snprintf(out_ptr, out_end - out_ptr,
1430 "%d", current->gid);
1431 if (rc > out_end - out_ptr)
1432 goto out;
1433 out_ptr += rc;
1434 break;
1435 /* signal that caused the coredump */
1436 case 's':
1437 rc = snprintf(out_ptr, out_end - out_ptr,
1438 "%ld", signr);
1439 if (rc > out_end - out_ptr)
1440 goto out;
1441 out_ptr += rc;
1442 break;
1443 /* UNIX time of coredump */
1444 case 't': {
1445 struct timeval tv;
1446 do_gettimeofday(&tv);
1447 rc = snprintf(out_ptr, out_end - out_ptr,
1448 "%lu", tv.tv_sec);
1449 if (rc > out_end - out_ptr)
1450 goto out;
1451 out_ptr += rc;
1452 break;
1454 /* hostname */
1455 case 'h':
1456 down_read(&uts_sem);
1457 rc = snprintf(out_ptr, out_end - out_ptr,
1458 "%s", utsname()->nodename);
1459 up_read(&uts_sem);
1460 if (rc > out_end - out_ptr)
1461 goto out;
1462 out_ptr += rc;
1463 break;
1464 /* executable */
1465 case 'e':
1466 rc = snprintf(out_ptr, out_end - out_ptr,
1467 "%s", current->comm);
1468 if (rc > out_end - out_ptr)
1469 goto out;
1470 out_ptr += rc;
1471 break;
1472 /* core limit size */
1473 case 'c':
1474 rc = snprintf(out_ptr, out_end - out_ptr,
1475 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1476 if (rc > out_end - out_ptr)
1477 goto out;
1478 out_ptr += rc;
1479 break;
1480 default:
1481 break;
1483 ++pat_ptr;
1486 /* Backward compatibility with core_uses_pid:
1488 * If core_pattern does not include a %p (as is the default)
1489 * and core_uses_pid is set, then .%pid will be appended to
1490 * the filename. Do not do this for piped commands. */
1491 if (!ispipe && !pid_in_pattern
1492 && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1493 rc = snprintf(out_ptr, out_end - out_ptr,
1494 ".%d", task_tgid_vnr(current));
1495 if (rc > out_end - out_ptr)
1496 goto out;
1497 out_ptr += rc;
1499 out:
1500 *out_ptr = 0;
1501 return ispipe;
1504 static void zap_process(struct task_struct *start)
1506 struct task_struct *t;
1508 start->signal->flags = SIGNAL_GROUP_EXIT;
1509 start->signal->group_stop_count = 0;
1511 t = start;
1512 do {
1513 if (t != current && t->mm) {
1514 t->mm->core_waiters++;
1515 sigaddset(&t->pending.signal, SIGKILL);
1516 signal_wake_up(t, 1);
1518 } while ((t = next_thread(t)) != start);
1521 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1522 int exit_code)
1524 struct task_struct *g, *p;
1525 unsigned long flags;
1526 int err = -EAGAIN;
1528 spin_lock_irq(&tsk->sighand->siglock);
1529 if (!signal_group_exit(tsk->signal)) {
1530 tsk->signal->group_exit_code = exit_code;
1531 zap_process(tsk);
1532 err = 0;
1534 spin_unlock_irq(&tsk->sighand->siglock);
1535 if (err)
1536 return err;
1538 if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1539 goto done;
1541 rcu_read_lock();
1542 for_each_process(g) {
1543 if (g == tsk->group_leader)
1544 continue;
1546 p = g;
1547 do {
1548 if (p->mm) {
1549 if (p->mm == mm) {
1551 * p->sighand can't disappear, but
1552 * may be changed by de_thread()
1554 lock_task_sighand(p, &flags);
1555 zap_process(p);
1556 unlock_task_sighand(p, &flags);
1558 break;
1560 } while ((p = next_thread(p)) != g);
1562 rcu_read_unlock();
1563 done:
1564 return mm->core_waiters;
1567 static int coredump_wait(int exit_code)
1569 struct task_struct *tsk = current;
1570 struct mm_struct *mm = tsk->mm;
1571 struct completion startup_done;
1572 struct completion *vfork_done;
1573 int core_waiters;
1575 init_completion(&mm->core_done);
1576 init_completion(&startup_done);
1577 mm->core_startup_done = &startup_done;
1579 core_waiters = zap_threads(tsk, mm, exit_code);
1580 up_write(&mm->mmap_sem);
1582 if (unlikely(core_waiters < 0))
1583 goto fail;
1586 * Make sure nobody is waiting for us to release the VM,
1587 * otherwise we can deadlock when we wait on each other
1589 vfork_done = tsk->vfork_done;
1590 if (vfork_done) {
1591 tsk->vfork_done = NULL;
1592 complete(vfork_done);
1595 if (core_waiters)
1596 wait_for_completion(&startup_done);
1597 fail:
1598 BUG_ON(mm->core_waiters);
1599 return core_waiters;
1603 * set_dumpable converts traditional three-value dumpable to two flags and
1604 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1605 * these bits are not changed atomically. So get_dumpable can observe the
1606 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1607 * return either old dumpable or new one by paying attention to the order of
1608 * modifying the bits.
1610 * dumpable | mm->flags (binary)
1611 * old new | initial interim final
1612 * ---------+-----------------------
1613 * 0 1 | 00 01 01
1614 * 0 2 | 00 10(*) 11
1615 * 1 0 | 01 00 00
1616 * 1 2 | 01 11 11
1617 * 2 0 | 11 10(*) 00
1618 * 2 1 | 11 11 01
1620 * (*) get_dumpable regards interim value of 10 as 11.
1622 void set_dumpable(struct mm_struct *mm, int value)
1624 switch (value) {
1625 case 0:
1626 clear_bit(MMF_DUMPABLE, &mm->flags);
1627 smp_wmb();
1628 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1629 break;
1630 case 1:
1631 set_bit(MMF_DUMPABLE, &mm->flags);
1632 smp_wmb();
1633 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1634 break;
1635 case 2:
1636 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1637 smp_wmb();
1638 set_bit(MMF_DUMPABLE, &mm->flags);
1639 break;
1643 int get_dumpable(struct mm_struct *mm)
1645 int ret;
1647 ret = mm->flags & 0x3;
1648 return (ret >= 2) ? 2 : ret;
1651 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1653 char corename[CORENAME_MAX_SIZE + 1];
1654 struct mm_struct *mm = current->mm;
1655 struct linux_binfmt * binfmt;
1656 struct inode * inode;
1657 struct file * file;
1658 int retval = 0;
1659 int fsuid = current->fsuid;
1660 int flag = 0;
1661 int ispipe = 0;
1662 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1663 char **helper_argv = NULL;
1664 int helper_argc = 0;
1665 char *delimit;
1667 audit_core_dumps(signr);
1669 binfmt = current->binfmt;
1670 if (!binfmt || !binfmt->core_dump)
1671 goto fail;
1672 down_write(&mm->mmap_sem);
1674 * If another thread got here first, or we are not dumpable, bail out.
1676 if (mm->core_waiters || !get_dumpable(mm)) {
1677 up_write(&mm->mmap_sem);
1678 goto fail;
1682 * We cannot trust fsuid as being the "true" uid of the
1683 * process nor do we know its entire history. We only know it
1684 * was tainted so we dump it as root in mode 2.
1686 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1687 flag = O_EXCL; /* Stop rewrite attacks */
1688 current->fsuid = 0; /* Dump root private */
1691 retval = coredump_wait(exit_code);
1692 if (retval < 0)
1693 goto fail;
1696 * Clear any false indication of pending signals that might
1697 * be seen by the filesystem code called to write the core file.
1699 clear_thread_flag(TIF_SIGPENDING);
1702 * lock_kernel() because format_corename() is controlled by sysctl, which
1703 * uses lock_kernel()
1705 lock_kernel();
1706 ispipe = format_corename(corename, core_pattern, signr);
1707 unlock_kernel();
1709 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1710 * to a pipe. Since we're not writing directly to the filesystem
1711 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1712 * created unless the pipe reader choses to write out the core file
1713 * at which point file size limits and permissions will be imposed
1714 * as it does with any other process
1716 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1717 goto fail_unlock;
1719 if (ispipe) {
1720 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1721 /* Terminate the string before the first option */
1722 delimit = strchr(corename, ' ');
1723 if (delimit)
1724 *delimit = '\0';
1725 delimit = strrchr(helper_argv[0], '/');
1726 if (delimit)
1727 delimit++;
1728 else
1729 delimit = helper_argv[0];
1730 if (!strcmp(delimit, current->comm)) {
1731 printk(KERN_NOTICE "Recursive core dump detected, "
1732 "aborting\n");
1733 goto fail_unlock;
1736 core_limit = RLIM_INFINITY;
1738 /* SIGPIPE can happen, but it's just never processed */
1739 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1740 &file)) {
1741 printk(KERN_INFO "Core dump to %s pipe failed\n",
1742 corename);
1743 goto fail_unlock;
1745 } else
1746 file = filp_open(corename,
1747 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1748 0600);
1749 if (IS_ERR(file))
1750 goto fail_unlock;
1751 inode = file->f_path.dentry->d_inode;
1752 if (inode->i_nlink > 1)
1753 goto close_fail; /* multiple links - don't dump */
1754 if (!ispipe && d_unhashed(file->f_path.dentry))
1755 goto close_fail;
1757 /* AK: actually i see no reason to not allow this for named pipes etc.,
1758 but keep the previous behaviour for now. */
1759 if (!ispipe && !S_ISREG(inode->i_mode))
1760 goto close_fail;
1762 * Dont allow local users get cute and trick others to coredump
1763 * into their pre-created files:
1765 if (inode->i_uid != current->fsuid)
1766 goto close_fail;
1767 if (!file->f_op)
1768 goto close_fail;
1769 if (!file->f_op->write)
1770 goto close_fail;
1771 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1772 goto close_fail;
1774 retval = binfmt->core_dump(signr, regs, file, core_limit);
1776 if (retval)
1777 current->signal->group_exit_code |= 0x80;
1778 close_fail:
1779 filp_close(file, NULL);
1780 fail_unlock:
1781 if (helper_argv)
1782 argv_free(helper_argv);
1784 current->fsuid = fsuid;
1785 complete_all(&mm->core_done);
1786 fail:
1787 return retval;