spi-topcliff-pch: supports a spi mode setup and bit order setup by IO control
[zen-stable.git] / fs / exec.c
blobae422777b7e857948fde0c0dcc9d3d1a55522133
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/mm.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/swap.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/perf_event.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/mount.h>
46 #include <linux/security.h>
47 #include <linux/syscalls.h>
48 #include <linux/tsacct_kern.h>
49 #include <linux/cn_proc.h>
50 #include <linux/audit.h>
51 #include <linux/tracehook.h>
52 #include <linux/kmod.h>
53 #include <linux/fsnotify.h>
54 #include <linux/fs_struct.h>
55 #include <linux/pipe_fs_i.h>
56 #include <linux/oom.h>
57 #include <linux/compat.h>
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
61 #include <asm/tlb.h>
63 #include <trace/events/task.h>
64 #include "internal.h"
66 int core_uses_pid;
67 char core_pattern[CORENAME_MAX_SIZE] = "core";
68 unsigned int core_pipe_limit;
69 int suid_dumpable = 0;
71 struct core_name {
72 char *corename;
73 int used, size;
75 static atomic_t call_count = ATOMIC_INIT(1);
77 /* The maximal length of core_pattern is also specified in sysctl.c */
79 static LIST_HEAD(formats);
80 static DEFINE_RWLOCK(binfmt_lock);
82 int __register_binfmt(struct linux_binfmt * fmt, int insert)
84 if (!fmt)
85 return -EINVAL;
86 write_lock(&binfmt_lock);
87 insert ? list_add(&fmt->lh, &formats) :
88 list_add_tail(&fmt->lh, &formats);
89 write_unlock(&binfmt_lock);
90 return 0;
93 EXPORT_SYMBOL(__register_binfmt);
95 void unregister_binfmt(struct linux_binfmt * fmt)
97 write_lock(&binfmt_lock);
98 list_del(&fmt->lh);
99 write_unlock(&binfmt_lock);
102 EXPORT_SYMBOL(unregister_binfmt);
104 static inline void put_binfmt(struct linux_binfmt * fmt)
106 module_put(fmt->module);
110 * Note that a shared library must be both readable and executable due to
111 * security reasons.
113 * Also note that we take the address to load from from the file itself.
115 SYSCALL_DEFINE1(uselib, const char __user *, library)
117 struct file *file;
118 char *tmp = getname(library);
119 int error = PTR_ERR(tmp);
120 static const struct open_flags uselib_flags = {
121 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
122 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
123 .intent = LOOKUP_OPEN
126 if (IS_ERR(tmp))
127 goto out;
129 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
130 putname(tmp);
131 error = PTR_ERR(file);
132 if (IS_ERR(file))
133 goto out;
135 error = -EINVAL;
136 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
137 goto exit;
139 error = -EACCES;
140 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
141 goto exit;
143 fsnotify_open(file);
145 error = -ENOEXEC;
146 if(file->f_op) {
147 struct linux_binfmt * fmt;
149 read_lock(&binfmt_lock);
150 list_for_each_entry(fmt, &formats, lh) {
151 if (!fmt->load_shlib)
152 continue;
153 if (!try_module_get(fmt->module))
154 continue;
155 read_unlock(&binfmt_lock);
156 error = fmt->load_shlib(file);
157 read_lock(&binfmt_lock);
158 put_binfmt(fmt);
159 if (error != -ENOEXEC)
160 break;
162 read_unlock(&binfmt_lock);
164 exit:
165 fput(file);
166 out:
167 return error;
170 #ifdef CONFIG_MMU
172 * The nascent bprm->mm is not visible until exec_mmap() but it can
173 * use a lot of memory, account these pages in current->mm temporary
174 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
175 * change the counter back via acct_arg_size(0).
177 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
179 struct mm_struct *mm = current->mm;
180 long diff = (long)(pages - bprm->vma_pages);
182 if (!mm || !diff)
183 return;
185 bprm->vma_pages = pages;
186 add_mm_counter(mm, MM_ANONPAGES, diff);
189 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
190 int write)
192 struct page *page;
193 int ret;
195 #ifdef CONFIG_STACK_GROWSUP
196 if (write) {
197 ret = expand_downwards(bprm->vma, pos);
198 if (ret < 0)
199 return NULL;
201 #endif
202 ret = get_user_pages(current, bprm->mm, pos,
203 1, write, 1, &page, NULL);
204 if (ret <= 0)
205 return NULL;
207 if (write) {
208 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
209 struct rlimit *rlim;
211 acct_arg_size(bprm, size / PAGE_SIZE);
214 * We've historically supported up to 32 pages (ARG_MAX)
215 * of argument strings even with small stacks
217 if (size <= ARG_MAX)
218 return page;
221 * Limit to 1/4-th the stack size for the argv+env strings.
222 * This ensures that:
223 * - the remaining binfmt code will not run out of stack space,
224 * - the program will have a reasonable amount of stack left
225 * to work from.
227 rlim = current->signal->rlim;
228 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
229 put_page(page);
230 return NULL;
234 return page;
237 static void put_arg_page(struct page *page)
239 put_page(page);
242 static void free_arg_page(struct linux_binprm *bprm, int i)
246 static void free_arg_pages(struct linux_binprm *bprm)
250 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
251 struct page *page)
253 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
256 static int __bprm_mm_init(struct linux_binprm *bprm)
258 int err;
259 struct vm_area_struct *vma = NULL;
260 struct mm_struct *mm = bprm->mm;
262 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
263 if (!vma)
264 return -ENOMEM;
266 down_write(&mm->mmap_sem);
267 vma->vm_mm = mm;
270 * Place the stack at the largest stack address the architecture
271 * supports. Later, we'll move this to an appropriate place. We don't
272 * use STACK_TOP because that can depend on attributes which aren't
273 * configured yet.
275 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
276 vma->vm_end = STACK_TOP_MAX;
277 vma->vm_start = vma->vm_end - PAGE_SIZE;
278 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
279 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
280 INIT_LIST_HEAD(&vma->anon_vma_chain);
282 err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
283 if (err)
284 goto err;
286 err = insert_vm_struct(mm, vma);
287 if (err)
288 goto err;
290 mm->stack_vm = mm->total_vm = 1;
291 up_write(&mm->mmap_sem);
292 bprm->p = vma->vm_end - sizeof(void *);
293 return 0;
294 err:
295 up_write(&mm->mmap_sem);
296 bprm->vma = NULL;
297 kmem_cache_free(vm_area_cachep, vma);
298 return err;
301 static bool valid_arg_len(struct linux_binprm *bprm, long len)
303 return len <= MAX_ARG_STRLEN;
306 #else
308 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
312 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
313 int write)
315 struct page *page;
317 page = bprm->page[pos / PAGE_SIZE];
318 if (!page && write) {
319 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
320 if (!page)
321 return NULL;
322 bprm->page[pos / PAGE_SIZE] = page;
325 return page;
328 static void put_arg_page(struct page *page)
332 static void free_arg_page(struct linux_binprm *bprm, int i)
334 if (bprm->page[i]) {
335 __free_page(bprm->page[i]);
336 bprm->page[i] = NULL;
340 static void free_arg_pages(struct linux_binprm *bprm)
342 int i;
344 for (i = 0; i < MAX_ARG_PAGES; i++)
345 free_arg_page(bprm, i);
348 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
349 struct page *page)
353 static int __bprm_mm_init(struct linux_binprm *bprm)
355 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
356 return 0;
359 static bool valid_arg_len(struct linux_binprm *bprm, long len)
361 return len <= bprm->p;
364 #endif /* CONFIG_MMU */
367 * Create a new mm_struct and populate it with a temporary stack
368 * vm_area_struct. We don't have enough context at this point to set the stack
369 * flags, permissions, and offset, so we use temporary values. We'll update
370 * them later in setup_arg_pages().
372 int bprm_mm_init(struct linux_binprm *bprm)
374 int err;
375 struct mm_struct *mm = NULL;
377 bprm->mm = mm = mm_alloc();
378 err = -ENOMEM;
379 if (!mm)
380 goto err;
382 err = init_new_context(current, mm);
383 if (err)
384 goto err;
386 err = __bprm_mm_init(bprm);
387 if (err)
388 goto err;
390 return 0;
392 err:
393 if (mm) {
394 bprm->mm = NULL;
395 mmdrop(mm);
398 return err;
401 struct user_arg_ptr {
402 #ifdef CONFIG_COMPAT
403 bool is_compat;
404 #endif
405 union {
406 const char __user *const __user *native;
407 #ifdef CONFIG_COMPAT
408 compat_uptr_t __user *compat;
409 #endif
410 } ptr;
413 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
415 const char __user *native;
417 #ifdef CONFIG_COMPAT
418 if (unlikely(argv.is_compat)) {
419 compat_uptr_t compat;
421 if (get_user(compat, argv.ptr.compat + nr))
422 return ERR_PTR(-EFAULT);
424 return compat_ptr(compat);
426 #endif
428 if (get_user(native, argv.ptr.native + nr))
429 return ERR_PTR(-EFAULT);
431 return native;
435 * count() counts the number of strings in array ARGV.
437 static int count(struct user_arg_ptr argv, int max)
439 int i = 0;
441 if (argv.ptr.native != NULL) {
442 for (;;) {
443 const char __user *p = get_user_arg_ptr(argv, i);
445 if (!p)
446 break;
448 if (IS_ERR(p))
449 return -EFAULT;
451 if (i++ >= max)
452 return -E2BIG;
454 if (fatal_signal_pending(current))
455 return -ERESTARTNOHAND;
456 cond_resched();
459 return i;
463 * 'copy_strings()' copies argument/environment strings from the old
464 * processes's memory to the new process's stack. The call to get_user_pages()
465 * ensures the destination page is created and not swapped out.
467 static int copy_strings(int argc, struct user_arg_ptr argv,
468 struct linux_binprm *bprm)
470 struct page *kmapped_page = NULL;
471 char *kaddr = NULL;
472 unsigned long kpos = 0;
473 int ret;
475 while (argc-- > 0) {
476 const char __user *str;
477 int len;
478 unsigned long pos;
480 ret = -EFAULT;
481 str = get_user_arg_ptr(argv, argc);
482 if (IS_ERR(str))
483 goto out;
485 len = strnlen_user(str, MAX_ARG_STRLEN);
486 if (!len)
487 goto out;
489 ret = -E2BIG;
490 if (!valid_arg_len(bprm, len))
491 goto out;
493 /* We're going to work our way backwords. */
494 pos = bprm->p;
495 str += len;
496 bprm->p -= len;
498 while (len > 0) {
499 int offset, bytes_to_copy;
501 if (fatal_signal_pending(current)) {
502 ret = -ERESTARTNOHAND;
503 goto out;
505 cond_resched();
507 offset = pos % PAGE_SIZE;
508 if (offset == 0)
509 offset = PAGE_SIZE;
511 bytes_to_copy = offset;
512 if (bytes_to_copy > len)
513 bytes_to_copy = len;
515 offset -= bytes_to_copy;
516 pos -= bytes_to_copy;
517 str -= bytes_to_copy;
518 len -= bytes_to_copy;
520 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
521 struct page *page;
523 page = get_arg_page(bprm, pos, 1);
524 if (!page) {
525 ret = -E2BIG;
526 goto out;
529 if (kmapped_page) {
530 flush_kernel_dcache_page(kmapped_page);
531 kunmap(kmapped_page);
532 put_arg_page(kmapped_page);
534 kmapped_page = page;
535 kaddr = kmap(kmapped_page);
536 kpos = pos & PAGE_MASK;
537 flush_arg_page(bprm, kpos, kmapped_page);
539 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
540 ret = -EFAULT;
541 goto out;
545 ret = 0;
546 out:
547 if (kmapped_page) {
548 flush_kernel_dcache_page(kmapped_page);
549 kunmap(kmapped_page);
550 put_arg_page(kmapped_page);
552 return ret;
556 * Like copy_strings, but get argv and its values from kernel memory.
558 int copy_strings_kernel(int argc, const char *const *__argv,
559 struct linux_binprm *bprm)
561 int r;
562 mm_segment_t oldfs = get_fs();
563 struct user_arg_ptr argv = {
564 .ptr.native = (const char __user *const __user *)__argv,
567 set_fs(KERNEL_DS);
568 r = copy_strings(argc, argv, bprm);
569 set_fs(oldfs);
571 return r;
573 EXPORT_SYMBOL(copy_strings_kernel);
575 #ifdef CONFIG_MMU
578 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
579 * the binfmt code determines where the new stack should reside, we shift it to
580 * its final location. The process proceeds as follows:
582 * 1) Use shift to calculate the new vma endpoints.
583 * 2) Extend vma to cover both the old and new ranges. This ensures the
584 * arguments passed to subsequent functions are consistent.
585 * 3) Move vma's page tables to the new range.
586 * 4) Free up any cleared pgd range.
587 * 5) Shrink the vma to cover only the new range.
589 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
591 struct mm_struct *mm = vma->vm_mm;
592 unsigned long old_start = vma->vm_start;
593 unsigned long old_end = vma->vm_end;
594 unsigned long length = old_end - old_start;
595 unsigned long new_start = old_start - shift;
596 unsigned long new_end = old_end - shift;
597 struct mmu_gather tlb;
599 BUG_ON(new_start > new_end);
602 * ensure there are no vmas between where we want to go
603 * and where we are
605 if (vma != find_vma(mm, new_start))
606 return -EFAULT;
609 * cover the whole range: [new_start, old_end)
611 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
612 return -ENOMEM;
615 * move the page tables downwards, on failure we rely on
616 * process cleanup to remove whatever mess we made.
618 if (length != move_page_tables(vma, old_start,
619 vma, new_start, length))
620 return -ENOMEM;
622 lru_add_drain();
623 tlb_gather_mmu(&tlb, mm, 0);
624 if (new_end > old_start) {
626 * when the old and new regions overlap clear from new_end.
628 free_pgd_range(&tlb, new_end, old_end, new_end,
629 vma->vm_next ? vma->vm_next->vm_start : 0);
630 } else {
632 * otherwise, clean from old_start; this is done to not touch
633 * the address space in [new_end, old_start) some architectures
634 * have constraints on va-space that make this illegal (IA64) -
635 * for the others its just a little faster.
637 free_pgd_range(&tlb, old_start, old_end, new_end,
638 vma->vm_next ? vma->vm_next->vm_start : 0);
640 tlb_finish_mmu(&tlb, new_end, old_end);
643 * Shrink the vma to just the new range. Always succeeds.
645 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
647 return 0;
651 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
652 * the stack is optionally relocated, and some extra space is added.
654 int setup_arg_pages(struct linux_binprm *bprm,
655 unsigned long stack_top,
656 int executable_stack)
658 unsigned long ret;
659 unsigned long stack_shift;
660 struct mm_struct *mm = current->mm;
661 struct vm_area_struct *vma = bprm->vma;
662 struct vm_area_struct *prev = NULL;
663 unsigned long vm_flags;
664 unsigned long stack_base;
665 unsigned long stack_size;
666 unsigned long stack_expand;
667 unsigned long rlim_stack;
669 #ifdef CONFIG_STACK_GROWSUP
670 /* Limit stack size to 1GB */
671 stack_base = rlimit_max(RLIMIT_STACK);
672 if (stack_base > (1 << 30))
673 stack_base = 1 << 30;
675 /* Make sure we didn't let the argument array grow too large. */
676 if (vma->vm_end - vma->vm_start > stack_base)
677 return -ENOMEM;
679 stack_base = PAGE_ALIGN(stack_top - stack_base);
681 stack_shift = vma->vm_start - stack_base;
682 mm->arg_start = bprm->p - stack_shift;
683 bprm->p = vma->vm_end - stack_shift;
684 #else
685 stack_top = arch_align_stack(stack_top);
686 stack_top = PAGE_ALIGN(stack_top);
688 if (unlikely(stack_top < mmap_min_addr) ||
689 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
690 return -ENOMEM;
692 stack_shift = vma->vm_end - stack_top;
694 bprm->p -= stack_shift;
695 mm->arg_start = bprm->p;
696 #endif
698 if (bprm->loader)
699 bprm->loader -= stack_shift;
700 bprm->exec -= stack_shift;
702 down_write(&mm->mmap_sem);
703 vm_flags = VM_STACK_FLAGS;
706 * Adjust stack execute permissions; explicitly enable for
707 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
708 * (arch default) otherwise.
710 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
711 vm_flags |= VM_EXEC;
712 else if (executable_stack == EXSTACK_DISABLE_X)
713 vm_flags &= ~VM_EXEC;
714 vm_flags |= mm->def_flags;
715 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
717 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
718 vm_flags);
719 if (ret)
720 goto out_unlock;
721 BUG_ON(prev != vma);
723 /* Move stack pages down in memory. */
724 if (stack_shift) {
725 ret = shift_arg_pages(vma, stack_shift);
726 if (ret)
727 goto out_unlock;
730 /* mprotect_fixup is overkill to remove the temporary stack flags */
731 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
733 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
734 stack_size = vma->vm_end - vma->vm_start;
736 * Align this down to a page boundary as expand_stack
737 * will align it up.
739 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
740 #ifdef CONFIG_STACK_GROWSUP
741 if (stack_size + stack_expand > rlim_stack)
742 stack_base = vma->vm_start + rlim_stack;
743 else
744 stack_base = vma->vm_end + stack_expand;
745 #else
746 if (stack_size + stack_expand > rlim_stack)
747 stack_base = vma->vm_end - rlim_stack;
748 else
749 stack_base = vma->vm_start - stack_expand;
750 #endif
751 current->mm->start_stack = bprm->p;
752 ret = expand_stack(vma, stack_base);
753 if (ret)
754 ret = -EFAULT;
756 out_unlock:
757 up_write(&mm->mmap_sem);
758 return ret;
760 EXPORT_SYMBOL(setup_arg_pages);
762 #endif /* CONFIG_MMU */
764 struct file *open_exec(const char *name)
766 struct file *file;
767 int err;
768 static const struct open_flags open_exec_flags = {
769 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
770 .acc_mode = MAY_EXEC | MAY_OPEN,
771 .intent = LOOKUP_OPEN
774 file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
775 if (IS_ERR(file))
776 goto out;
778 err = -EACCES;
779 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
780 goto exit;
782 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
783 goto exit;
785 fsnotify_open(file);
787 err = deny_write_access(file);
788 if (err)
789 goto exit;
791 out:
792 return file;
794 exit:
795 fput(file);
796 return ERR_PTR(err);
798 EXPORT_SYMBOL(open_exec);
800 int kernel_read(struct file *file, loff_t offset,
801 char *addr, unsigned long count)
803 mm_segment_t old_fs;
804 loff_t pos = offset;
805 int result;
807 old_fs = get_fs();
808 set_fs(get_ds());
809 /* The cast to a user pointer is valid due to the set_fs() */
810 result = vfs_read(file, (void __user *)addr, count, &pos);
811 set_fs(old_fs);
812 return result;
815 EXPORT_SYMBOL(kernel_read);
817 static int exec_mmap(struct mm_struct *mm)
819 struct task_struct *tsk;
820 struct mm_struct * old_mm, *active_mm;
822 /* Notify parent that we're no longer interested in the old VM */
823 tsk = current;
824 old_mm = current->mm;
825 sync_mm_rss(tsk, old_mm);
826 mm_release(tsk, old_mm);
828 if (old_mm) {
830 * Make sure that if there is a core dump in progress
831 * for the old mm, we get out and die instead of going
832 * through with the exec. We must hold mmap_sem around
833 * checking core_state and changing tsk->mm.
835 down_read(&old_mm->mmap_sem);
836 if (unlikely(old_mm->core_state)) {
837 up_read(&old_mm->mmap_sem);
838 return -EINTR;
841 task_lock(tsk);
842 active_mm = tsk->active_mm;
843 tsk->mm = mm;
844 tsk->active_mm = mm;
845 activate_mm(active_mm, mm);
846 task_unlock(tsk);
847 arch_pick_mmap_layout(mm);
848 if (old_mm) {
849 up_read(&old_mm->mmap_sem);
850 BUG_ON(active_mm != old_mm);
851 mm_update_next_owner(old_mm);
852 mmput(old_mm);
853 return 0;
855 mmdrop(active_mm);
856 return 0;
860 * This function makes sure the current process has its own signal table,
861 * so that flush_signal_handlers can later reset the handlers without
862 * disturbing other processes. (Other processes might share the signal
863 * table via the CLONE_SIGHAND option to clone().)
865 static int de_thread(struct task_struct *tsk)
867 struct signal_struct *sig = tsk->signal;
868 struct sighand_struct *oldsighand = tsk->sighand;
869 spinlock_t *lock = &oldsighand->siglock;
871 if (thread_group_empty(tsk))
872 goto no_thread_group;
875 * Kill all other threads in the thread group.
877 spin_lock_irq(lock);
878 if (signal_group_exit(sig)) {
880 * Another group action in progress, just
881 * return so that the signal is processed.
883 spin_unlock_irq(lock);
884 return -EAGAIN;
887 sig->group_exit_task = tsk;
888 sig->notify_count = zap_other_threads(tsk);
889 if (!thread_group_leader(tsk))
890 sig->notify_count--;
892 while (sig->notify_count) {
893 __set_current_state(TASK_UNINTERRUPTIBLE);
894 spin_unlock_irq(lock);
895 schedule();
896 spin_lock_irq(lock);
898 spin_unlock_irq(lock);
901 * At this point all other threads have exited, all we have to
902 * do is to wait for the thread group leader to become inactive,
903 * and to assume its PID:
905 if (!thread_group_leader(tsk)) {
906 struct task_struct *leader = tsk->group_leader;
908 sig->notify_count = -1; /* for exit_notify() */
909 for (;;) {
910 write_lock_irq(&tasklist_lock);
911 if (likely(leader->exit_state))
912 break;
913 __set_current_state(TASK_UNINTERRUPTIBLE);
914 write_unlock_irq(&tasklist_lock);
915 schedule();
919 * The only record we have of the real-time age of a
920 * process, regardless of execs it's done, is start_time.
921 * All the past CPU time is accumulated in signal_struct
922 * from sister threads now dead. But in this non-leader
923 * exec, nothing survives from the original leader thread,
924 * whose birth marks the true age of this process now.
925 * When we take on its identity by switching to its PID, we
926 * also take its birthdate (always earlier than our own).
928 tsk->start_time = leader->start_time;
930 BUG_ON(!same_thread_group(leader, tsk));
931 BUG_ON(has_group_leader_pid(tsk));
933 * An exec() starts a new thread group with the
934 * TGID of the previous thread group. Rehash the
935 * two threads with a switched PID, and release
936 * the former thread group leader:
939 /* Become a process group leader with the old leader's pid.
940 * The old leader becomes a thread of the this thread group.
941 * Note: The old leader also uses this pid until release_task
942 * is called. Odd but simple and correct.
944 detach_pid(tsk, PIDTYPE_PID);
945 tsk->pid = leader->pid;
946 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
947 transfer_pid(leader, tsk, PIDTYPE_PGID);
948 transfer_pid(leader, tsk, PIDTYPE_SID);
950 list_replace_rcu(&leader->tasks, &tsk->tasks);
951 list_replace_init(&leader->sibling, &tsk->sibling);
953 tsk->group_leader = tsk;
954 leader->group_leader = tsk;
956 tsk->exit_signal = SIGCHLD;
957 leader->exit_signal = -1;
959 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
960 leader->exit_state = EXIT_DEAD;
963 * We are going to release_task()->ptrace_unlink() silently,
964 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
965 * the tracer wont't block again waiting for this thread.
967 if (unlikely(leader->ptrace))
968 __wake_up_parent(leader, leader->parent);
969 write_unlock_irq(&tasklist_lock);
971 release_task(leader);
974 sig->group_exit_task = NULL;
975 sig->notify_count = 0;
977 no_thread_group:
978 /* we have changed execution domain */
979 tsk->exit_signal = SIGCHLD;
981 if (current->mm)
982 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
984 exit_itimers(sig);
985 flush_itimer_signals();
987 if (atomic_read(&oldsighand->count) != 1) {
988 struct sighand_struct *newsighand;
990 * This ->sighand is shared with the CLONE_SIGHAND
991 * but not CLONE_THREAD task, switch to the new one.
993 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
994 if (!newsighand)
995 return -ENOMEM;
997 atomic_set(&newsighand->count, 1);
998 memcpy(newsighand->action, oldsighand->action,
999 sizeof(newsighand->action));
1001 write_lock_irq(&tasklist_lock);
1002 spin_lock(&oldsighand->siglock);
1003 rcu_assign_pointer(tsk->sighand, newsighand);
1004 spin_unlock(&oldsighand->siglock);
1005 write_unlock_irq(&tasklist_lock);
1007 __cleanup_sighand(oldsighand);
1010 BUG_ON(!thread_group_leader(tsk));
1011 return 0;
1015 * These functions flushes out all traces of the currently running executable
1016 * so that a new one can be started
1018 static void flush_old_files(struct files_struct * files)
1020 long j = -1;
1021 struct fdtable *fdt;
1023 spin_lock(&files->file_lock);
1024 for (;;) {
1025 unsigned long set, i;
1027 j++;
1028 i = j * __NFDBITS;
1029 fdt = files_fdtable(files);
1030 if (i >= fdt->max_fds)
1031 break;
1032 set = fdt->close_on_exec->fds_bits[j];
1033 if (!set)
1034 continue;
1035 fdt->close_on_exec->fds_bits[j] = 0;
1036 spin_unlock(&files->file_lock);
1037 for ( ; set ; i++,set >>= 1) {
1038 if (set & 1) {
1039 sys_close(i);
1042 spin_lock(&files->file_lock);
1045 spin_unlock(&files->file_lock);
1048 char *get_task_comm(char *buf, struct task_struct *tsk)
1050 /* buf must be at least sizeof(tsk->comm) in size */
1051 task_lock(tsk);
1052 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1053 task_unlock(tsk);
1054 return buf;
1056 EXPORT_SYMBOL_GPL(get_task_comm);
1058 void set_task_comm(struct task_struct *tsk, char *buf)
1060 task_lock(tsk);
1062 trace_task_rename(tsk, buf);
1065 * Threads may access current->comm without holding
1066 * the task lock, so write the string carefully.
1067 * Readers without a lock may see incomplete new
1068 * names but are safe from non-terminating string reads.
1070 memset(tsk->comm, 0, TASK_COMM_LEN);
1071 wmb();
1072 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1073 task_unlock(tsk);
1074 perf_event_comm(tsk);
1077 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1079 int i, ch;
1081 /* Copies the binary name from after last slash */
1082 for (i = 0; (ch = *(fn++)) != '\0';) {
1083 if (ch == '/')
1084 i = 0; /* overwrite what we wrote */
1085 else
1086 if (i < len - 1)
1087 tcomm[i++] = ch;
1089 tcomm[i] = '\0';
1092 int flush_old_exec(struct linux_binprm * bprm)
1094 int retval;
1097 * Make sure we have a private signal table and that
1098 * we are unassociated from the previous thread group.
1100 retval = de_thread(current);
1101 if (retval)
1102 goto out;
1104 set_mm_exe_file(bprm->mm, bprm->file);
1106 filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1108 * Release all of the old mmap stuff
1110 acct_arg_size(bprm, 0);
1111 retval = exec_mmap(bprm->mm);
1112 if (retval)
1113 goto out;
1115 bprm->mm = NULL; /* We're using it now */
1117 set_fs(USER_DS);
1118 current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1119 flush_thread();
1120 current->personality &= ~bprm->per_clear;
1122 return 0;
1124 out:
1125 return retval;
1127 EXPORT_SYMBOL(flush_old_exec);
1129 void would_dump(struct linux_binprm *bprm, struct file *file)
1131 if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1132 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1134 EXPORT_SYMBOL(would_dump);
1136 void setup_new_exec(struct linux_binprm * bprm)
1138 arch_pick_mmap_layout(current->mm);
1140 /* This is the point of no return */
1141 current->sas_ss_sp = current->sas_ss_size = 0;
1143 if (current_euid() == current_uid() && current_egid() == current_gid())
1144 set_dumpable(current->mm, 1);
1145 else
1146 set_dumpable(current->mm, suid_dumpable);
1148 set_task_comm(current, bprm->tcomm);
1150 /* Set the new mm task size. We have to do that late because it may
1151 * depend on TIF_32BIT which is only updated in flush_thread() on
1152 * some architectures like powerpc
1154 current->mm->task_size = TASK_SIZE;
1156 /* install the new credentials */
1157 if (bprm->cred->uid != current_euid() ||
1158 bprm->cred->gid != current_egid()) {
1159 current->pdeath_signal = 0;
1160 } else {
1161 would_dump(bprm, bprm->file);
1162 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1163 set_dumpable(current->mm, suid_dumpable);
1167 * Flush performance counters when crossing a
1168 * security domain:
1170 if (!get_dumpable(current->mm))
1171 perf_event_exit_task(current);
1173 /* An exec changes our domain. We are no longer part of the thread
1174 group */
1176 current->self_exec_id++;
1178 flush_signal_handlers(current, 0);
1179 flush_old_files(current->files);
1181 EXPORT_SYMBOL(setup_new_exec);
1184 * Prepare credentials and lock ->cred_guard_mutex.
1185 * install_exec_creds() commits the new creds and drops the lock.
1186 * Or, if exec fails before, free_bprm() should release ->cred and
1187 * and unlock.
1189 int prepare_bprm_creds(struct linux_binprm *bprm)
1191 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1192 return -ERESTARTNOINTR;
1194 bprm->cred = prepare_exec_creds();
1195 if (likely(bprm->cred))
1196 return 0;
1198 mutex_unlock(&current->signal->cred_guard_mutex);
1199 return -ENOMEM;
1202 void free_bprm(struct linux_binprm *bprm)
1204 free_arg_pages(bprm);
1205 if (bprm->cred) {
1206 mutex_unlock(&current->signal->cred_guard_mutex);
1207 abort_creds(bprm->cred);
1209 kfree(bprm);
1213 * install the new credentials for this executable
1215 void install_exec_creds(struct linux_binprm *bprm)
1217 security_bprm_committing_creds(bprm);
1219 commit_creds(bprm->cred);
1220 bprm->cred = NULL;
1222 * cred_guard_mutex must be held at least to this point to prevent
1223 * ptrace_attach() from altering our determination of the task's
1224 * credentials; any time after this it may be unlocked.
1226 security_bprm_committed_creds(bprm);
1227 mutex_unlock(&current->signal->cred_guard_mutex);
1229 EXPORT_SYMBOL(install_exec_creds);
1232 * determine how safe it is to execute the proposed program
1233 * - the caller must hold ->cred_guard_mutex to protect against
1234 * PTRACE_ATTACH
1236 static int check_unsafe_exec(struct linux_binprm *bprm)
1238 struct task_struct *p = current, *t;
1239 unsigned n_fs;
1240 int res = 0;
1242 if (p->ptrace) {
1243 if (p->ptrace & PT_PTRACE_CAP)
1244 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1245 else
1246 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1249 n_fs = 1;
1250 spin_lock(&p->fs->lock);
1251 rcu_read_lock();
1252 for (t = next_thread(p); t != p; t = next_thread(t)) {
1253 if (t->fs == p->fs)
1254 n_fs++;
1256 rcu_read_unlock();
1258 if (p->fs->users > n_fs) {
1259 bprm->unsafe |= LSM_UNSAFE_SHARE;
1260 } else {
1261 res = -EAGAIN;
1262 if (!p->fs->in_exec) {
1263 p->fs->in_exec = 1;
1264 res = 1;
1267 spin_unlock(&p->fs->lock);
1269 return res;
1273 * Fill the binprm structure from the inode.
1274 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1276 * This may be called multiple times for binary chains (scripts for example).
1278 int prepare_binprm(struct linux_binprm *bprm)
1280 umode_t mode;
1281 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1282 int retval;
1284 mode = inode->i_mode;
1285 if (bprm->file->f_op == NULL)
1286 return -EACCES;
1288 /* clear any previous set[ug]id data from a previous binary */
1289 bprm->cred->euid = current_euid();
1290 bprm->cred->egid = current_egid();
1292 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1293 /* Set-uid? */
1294 if (mode & S_ISUID) {
1295 bprm->per_clear |= PER_CLEAR_ON_SETID;
1296 bprm->cred->euid = inode->i_uid;
1299 /* Set-gid? */
1301 * If setgid is set but no group execute bit then this
1302 * is a candidate for mandatory locking, not a setgid
1303 * executable.
1305 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1306 bprm->per_clear |= PER_CLEAR_ON_SETID;
1307 bprm->cred->egid = inode->i_gid;
1311 /* fill in binprm security blob */
1312 retval = security_bprm_set_creds(bprm);
1313 if (retval)
1314 return retval;
1315 bprm->cred_prepared = 1;
1317 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1318 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1321 EXPORT_SYMBOL(prepare_binprm);
1324 * Arguments are '\0' separated strings found at the location bprm->p
1325 * points to; chop off the first by relocating brpm->p to right after
1326 * the first '\0' encountered.
1328 int remove_arg_zero(struct linux_binprm *bprm)
1330 int ret = 0;
1331 unsigned long offset;
1332 char *kaddr;
1333 struct page *page;
1335 if (!bprm->argc)
1336 return 0;
1338 do {
1339 offset = bprm->p & ~PAGE_MASK;
1340 page = get_arg_page(bprm, bprm->p, 0);
1341 if (!page) {
1342 ret = -EFAULT;
1343 goto out;
1345 kaddr = kmap_atomic(page, KM_USER0);
1347 for (; offset < PAGE_SIZE && kaddr[offset];
1348 offset++, bprm->p++)
1351 kunmap_atomic(kaddr, KM_USER0);
1352 put_arg_page(page);
1354 if (offset == PAGE_SIZE)
1355 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1356 } while (offset == PAGE_SIZE);
1358 bprm->p++;
1359 bprm->argc--;
1360 ret = 0;
1362 out:
1363 return ret;
1365 EXPORT_SYMBOL(remove_arg_zero);
1368 * cycle the list of binary formats handler, until one recognizes the image
1370 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1372 unsigned int depth = bprm->recursion_depth;
1373 int try,retval;
1374 struct linux_binfmt *fmt;
1375 pid_t old_pid;
1377 retval = security_bprm_check(bprm);
1378 if (retval)
1379 return retval;
1381 retval = audit_bprm(bprm);
1382 if (retval)
1383 return retval;
1385 /* Need to fetch pid before load_binary changes it */
1386 rcu_read_lock();
1387 old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1388 rcu_read_unlock();
1390 retval = -ENOENT;
1391 for (try=0; try<2; try++) {
1392 read_lock(&binfmt_lock);
1393 list_for_each_entry(fmt, &formats, lh) {
1394 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1395 if (!fn)
1396 continue;
1397 if (!try_module_get(fmt->module))
1398 continue;
1399 read_unlock(&binfmt_lock);
1400 retval = fn(bprm, regs);
1402 * Restore the depth counter to its starting value
1403 * in this call, so we don't have to rely on every
1404 * load_binary function to restore it on return.
1406 bprm->recursion_depth = depth;
1407 if (retval >= 0) {
1408 if (depth == 0)
1409 ptrace_event(PTRACE_EVENT_EXEC,
1410 old_pid);
1411 put_binfmt(fmt);
1412 allow_write_access(bprm->file);
1413 if (bprm->file)
1414 fput(bprm->file);
1415 bprm->file = NULL;
1416 current->did_exec = 1;
1417 proc_exec_connector(current);
1418 return retval;
1420 read_lock(&binfmt_lock);
1421 put_binfmt(fmt);
1422 if (retval != -ENOEXEC || bprm->mm == NULL)
1423 break;
1424 if (!bprm->file) {
1425 read_unlock(&binfmt_lock);
1426 return retval;
1429 read_unlock(&binfmt_lock);
1430 #ifdef CONFIG_MODULES
1431 if (retval != -ENOEXEC || bprm->mm == NULL) {
1432 break;
1433 } else {
1434 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1435 if (printable(bprm->buf[0]) &&
1436 printable(bprm->buf[1]) &&
1437 printable(bprm->buf[2]) &&
1438 printable(bprm->buf[3]))
1439 break; /* -ENOEXEC */
1440 if (try)
1441 break; /* -ENOEXEC */
1442 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1444 #else
1445 break;
1446 #endif
1448 return retval;
1451 EXPORT_SYMBOL(search_binary_handler);
1454 * sys_execve() executes a new program.
1456 static int do_execve_common(const char *filename,
1457 struct user_arg_ptr argv,
1458 struct user_arg_ptr envp,
1459 struct pt_regs *regs)
1461 struct linux_binprm *bprm;
1462 struct file *file;
1463 struct files_struct *displaced;
1464 bool clear_in_exec;
1465 int retval;
1466 const struct cred *cred = current_cred();
1469 * We move the actual failure in case of RLIMIT_NPROC excess from
1470 * set*uid() to execve() because too many poorly written programs
1471 * don't check setuid() return code. Here we additionally recheck
1472 * whether NPROC limit is still exceeded.
1474 if ((current->flags & PF_NPROC_EXCEEDED) &&
1475 atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1476 retval = -EAGAIN;
1477 goto out_ret;
1480 /* We're below the limit (still or again), so we don't want to make
1481 * further execve() calls fail. */
1482 current->flags &= ~PF_NPROC_EXCEEDED;
1484 retval = unshare_files(&displaced);
1485 if (retval)
1486 goto out_ret;
1488 retval = -ENOMEM;
1489 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1490 if (!bprm)
1491 goto out_files;
1493 retval = prepare_bprm_creds(bprm);
1494 if (retval)
1495 goto out_free;
1497 retval = check_unsafe_exec(bprm);
1498 if (retval < 0)
1499 goto out_free;
1500 clear_in_exec = retval;
1501 current->in_execve = 1;
1503 file = open_exec(filename);
1504 retval = PTR_ERR(file);
1505 if (IS_ERR(file))
1506 goto out_unmark;
1508 sched_exec();
1510 bprm->file = file;
1511 bprm->filename = filename;
1512 bprm->interp = filename;
1514 retval = bprm_mm_init(bprm);
1515 if (retval)
1516 goto out_file;
1518 bprm->argc = count(argv, MAX_ARG_STRINGS);
1519 if ((retval = bprm->argc) < 0)
1520 goto out;
1522 bprm->envc = count(envp, MAX_ARG_STRINGS);
1523 if ((retval = bprm->envc) < 0)
1524 goto out;
1526 retval = prepare_binprm(bprm);
1527 if (retval < 0)
1528 goto out;
1530 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1531 if (retval < 0)
1532 goto out;
1534 bprm->exec = bprm->p;
1535 retval = copy_strings(bprm->envc, envp, bprm);
1536 if (retval < 0)
1537 goto out;
1539 retval = copy_strings(bprm->argc, argv, bprm);
1540 if (retval < 0)
1541 goto out;
1543 retval = search_binary_handler(bprm,regs);
1544 if (retval < 0)
1545 goto out;
1547 /* execve succeeded */
1548 current->fs->in_exec = 0;
1549 current->in_execve = 0;
1550 acct_update_integrals(current);
1551 free_bprm(bprm);
1552 if (displaced)
1553 put_files_struct(displaced);
1554 return retval;
1556 out:
1557 if (bprm->mm) {
1558 acct_arg_size(bprm, 0);
1559 mmput(bprm->mm);
1562 out_file:
1563 if (bprm->file) {
1564 allow_write_access(bprm->file);
1565 fput(bprm->file);
1568 out_unmark:
1569 if (clear_in_exec)
1570 current->fs->in_exec = 0;
1571 current->in_execve = 0;
1573 out_free:
1574 free_bprm(bprm);
1576 out_files:
1577 if (displaced)
1578 reset_files_struct(displaced);
1579 out_ret:
1580 return retval;
1583 int do_execve(const char *filename,
1584 const char __user *const __user *__argv,
1585 const char __user *const __user *__envp,
1586 struct pt_regs *regs)
1588 struct user_arg_ptr argv = { .ptr.native = __argv };
1589 struct user_arg_ptr envp = { .ptr.native = __envp };
1590 return do_execve_common(filename, argv, envp, regs);
1593 #ifdef CONFIG_COMPAT
1594 int compat_do_execve(char *filename,
1595 compat_uptr_t __user *__argv,
1596 compat_uptr_t __user *__envp,
1597 struct pt_regs *regs)
1599 struct user_arg_ptr argv = {
1600 .is_compat = true,
1601 .ptr.compat = __argv,
1603 struct user_arg_ptr envp = {
1604 .is_compat = true,
1605 .ptr.compat = __envp,
1607 return do_execve_common(filename, argv, envp, regs);
1609 #endif
1611 void set_binfmt(struct linux_binfmt *new)
1613 struct mm_struct *mm = current->mm;
1615 if (mm->binfmt)
1616 module_put(mm->binfmt->module);
1618 mm->binfmt = new;
1619 if (new)
1620 __module_get(new->module);
1623 EXPORT_SYMBOL(set_binfmt);
1625 static int expand_corename(struct core_name *cn)
1627 char *old_corename = cn->corename;
1629 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1630 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1632 if (!cn->corename) {
1633 kfree(old_corename);
1634 return -ENOMEM;
1637 return 0;
1640 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1642 char *cur;
1643 int need;
1644 int ret;
1645 va_list arg;
1647 va_start(arg, fmt);
1648 need = vsnprintf(NULL, 0, fmt, arg);
1649 va_end(arg);
1651 if (likely(need < cn->size - cn->used - 1))
1652 goto out_printf;
1654 ret = expand_corename(cn);
1655 if (ret)
1656 goto expand_fail;
1658 out_printf:
1659 cur = cn->corename + cn->used;
1660 va_start(arg, fmt);
1661 vsnprintf(cur, need + 1, fmt, arg);
1662 va_end(arg);
1663 cn->used += need;
1664 return 0;
1666 expand_fail:
1667 return ret;
1670 static void cn_escape(char *str)
1672 for (; *str; str++)
1673 if (*str == '/')
1674 *str = '!';
1677 static int cn_print_exe_file(struct core_name *cn)
1679 struct file *exe_file;
1680 char *pathbuf, *path;
1681 int ret;
1683 exe_file = get_mm_exe_file(current->mm);
1684 if (!exe_file) {
1685 char *commstart = cn->corename + cn->used;
1686 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1687 cn_escape(commstart);
1688 return ret;
1691 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1692 if (!pathbuf) {
1693 ret = -ENOMEM;
1694 goto put_exe_file;
1697 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1698 if (IS_ERR(path)) {
1699 ret = PTR_ERR(path);
1700 goto free_buf;
1703 cn_escape(path);
1705 ret = cn_printf(cn, "%s", path);
1707 free_buf:
1708 kfree(pathbuf);
1709 put_exe_file:
1710 fput(exe_file);
1711 return ret;
1714 /* format_corename will inspect the pattern parameter, and output a
1715 * name into corename, which must have space for at least
1716 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1718 static int format_corename(struct core_name *cn, long signr)
1720 const struct cred *cred = current_cred();
1721 const char *pat_ptr = core_pattern;
1722 int ispipe = (*pat_ptr == '|');
1723 int pid_in_pattern = 0;
1724 int err = 0;
1726 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1727 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1728 cn->used = 0;
1730 if (!cn->corename)
1731 return -ENOMEM;
1733 /* Repeat as long as we have more pattern to process and more output
1734 space */
1735 while (*pat_ptr) {
1736 if (*pat_ptr != '%') {
1737 if (*pat_ptr == 0)
1738 goto out;
1739 err = cn_printf(cn, "%c", *pat_ptr++);
1740 } else {
1741 switch (*++pat_ptr) {
1742 /* single % at the end, drop that */
1743 case 0:
1744 goto out;
1745 /* Double percent, output one percent */
1746 case '%':
1747 err = cn_printf(cn, "%c", '%');
1748 break;
1749 /* pid */
1750 case 'p':
1751 pid_in_pattern = 1;
1752 err = cn_printf(cn, "%d",
1753 task_tgid_vnr(current));
1754 break;
1755 /* uid */
1756 case 'u':
1757 err = cn_printf(cn, "%d", cred->uid);
1758 break;
1759 /* gid */
1760 case 'g':
1761 err = cn_printf(cn, "%d", cred->gid);
1762 break;
1763 /* signal that caused the coredump */
1764 case 's':
1765 err = cn_printf(cn, "%ld", signr);
1766 break;
1767 /* UNIX time of coredump */
1768 case 't': {
1769 struct timeval tv;
1770 do_gettimeofday(&tv);
1771 err = cn_printf(cn, "%lu", tv.tv_sec);
1772 break;
1774 /* hostname */
1775 case 'h': {
1776 char *namestart = cn->corename + cn->used;
1777 down_read(&uts_sem);
1778 err = cn_printf(cn, "%s",
1779 utsname()->nodename);
1780 up_read(&uts_sem);
1781 cn_escape(namestart);
1782 break;
1784 /* executable */
1785 case 'e': {
1786 char *commstart = cn->corename + cn->used;
1787 err = cn_printf(cn, "%s", current->comm);
1788 cn_escape(commstart);
1789 break;
1791 case 'E':
1792 err = cn_print_exe_file(cn);
1793 break;
1794 /* core limit size */
1795 case 'c':
1796 err = cn_printf(cn, "%lu",
1797 rlimit(RLIMIT_CORE));
1798 break;
1799 default:
1800 break;
1802 ++pat_ptr;
1805 if (err)
1806 return err;
1809 /* Backward compatibility with core_uses_pid:
1811 * If core_pattern does not include a %p (as is the default)
1812 * and core_uses_pid is set, then .%pid will be appended to
1813 * the filename. Do not do this for piped commands. */
1814 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1815 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1816 if (err)
1817 return err;
1819 out:
1820 return ispipe;
1823 static int zap_process(struct task_struct *start, int exit_code)
1825 struct task_struct *t;
1826 int nr = 0;
1828 start->signal->flags = SIGNAL_GROUP_EXIT;
1829 start->signal->group_exit_code = exit_code;
1830 start->signal->group_stop_count = 0;
1832 t = start;
1833 do {
1834 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1835 if (t != current && t->mm) {
1836 sigaddset(&t->pending.signal, SIGKILL);
1837 signal_wake_up(t, 1);
1838 nr++;
1840 } while_each_thread(start, t);
1842 return nr;
1845 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1846 struct core_state *core_state, int exit_code)
1848 struct task_struct *g, *p;
1849 unsigned long flags;
1850 int nr = -EAGAIN;
1852 spin_lock_irq(&tsk->sighand->siglock);
1853 if (!signal_group_exit(tsk->signal)) {
1854 mm->core_state = core_state;
1855 nr = zap_process(tsk, exit_code);
1857 spin_unlock_irq(&tsk->sighand->siglock);
1858 if (unlikely(nr < 0))
1859 return nr;
1861 if (atomic_read(&mm->mm_users) == nr + 1)
1862 goto done;
1864 * We should find and kill all tasks which use this mm, and we should
1865 * count them correctly into ->nr_threads. We don't take tasklist
1866 * lock, but this is safe wrt:
1868 * fork:
1869 * None of sub-threads can fork after zap_process(leader). All
1870 * processes which were created before this point should be
1871 * visible to zap_threads() because copy_process() adds the new
1872 * process to the tail of init_task.tasks list, and lock/unlock
1873 * of ->siglock provides a memory barrier.
1875 * do_exit:
1876 * The caller holds mm->mmap_sem. This means that the task which
1877 * uses this mm can't pass exit_mm(), so it can't exit or clear
1878 * its ->mm.
1880 * de_thread:
1881 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1882 * we must see either old or new leader, this does not matter.
1883 * However, it can change p->sighand, so lock_task_sighand(p)
1884 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1885 * it can't fail.
1887 * Note also that "g" can be the old leader with ->mm == NULL
1888 * and already unhashed and thus removed from ->thread_group.
1889 * This is OK, __unhash_process()->list_del_rcu() does not
1890 * clear the ->next pointer, we will find the new leader via
1891 * next_thread().
1893 rcu_read_lock();
1894 for_each_process(g) {
1895 if (g == tsk->group_leader)
1896 continue;
1897 if (g->flags & PF_KTHREAD)
1898 continue;
1899 p = g;
1900 do {
1901 if (p->mm) {
1902 if (unlikely(p->mm == mm)) {
1903 lock_task_sighand(p, &flags);
1904 nr += zap_process(p, exit_code);
1905 unlock_task_sighand(p, &flags);
1907 break;
1909 } while_each_thread(g, p);
1911 rcu_read_unlock();
1912 done:
1913 atomic_set(&core_state->nr_threads, nr);
1914 return nr;
1917 static int coredump_wait(int exit_code, struct core_state *core_state)
1919 struct task_struct *tsk = current;
1920 struct mm_struct *mm = tsk->mm;
1921 int core_waiters = -EBUSY;
1923 init_completion(&core_state->startup);
1924 core_state->dumper.task = tsk;
1925 core_state->dumper.next = NULL;
1927 down_write(&mm->mmap_sem);
1928 if (!mm->core_state)
1929 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1930 up_write(&mm->mmap_sem);
1932 if (core_waiters > 0)
1933 wait_for_completion(&core_state->startup);
1935 return core_waiters;
1938 static void coredump_finish(struct mm_struct *mm)
1940 struct core_thread *curr, *next;
1941 struct task_struct *task;
1943 next = mm->core_state->dumper.next;
1944 while ((curr = next) != NULL) {
1945 next = curr->next;
1946 task = curr->task;
1948 * see exit_mm(), curr->task must not see
1949 * ->task == NULL before we read ->next.
1951 smp_mb();
1952 curr->task = NULL;
1953 wake_up_process(task);
1956 mm->core_state = NULL;
1960 * set_dumpable converts traditional three-value dumpable to two flags and
1961 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1962 * these bits are not changed atomically. So get_dumpable can observe the
1963 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1964 * return either old dumpable or new one by paying attention to the order of
1965 * modifying the bits.
1967 * dumpable | mm->flags (binary)
1968 * old new | initial interim final
1969 * ---------+-----------------------
1970 * 0 1 | 00 01 01
1971 * 0 2 | 00 10(*) 11
1972 * 1 0 | 01 00 00
1973 * 1 2 | 01 11 11
1974 * 2 0 | 11 10(*) 00
1975 * 2 1 | 11 11 01
1977 * (*) get_dumpable regards interim value of 10 as 11.
1979 void set_dumpable(struct mm_struct *mm, int value)
1981 switch (value) {
1982 case 0:
1983 clear_bit(MMF_DUMPABLE, &mm->flags);
1984 smp_wmb();
1985 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1986 break;
1987 case 1:
1988 set_bit(MMF_DUMPABLE, &mm->flags);
1989 smp_wmb();
1990 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1991 break;
1992 case 2:
1993 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1994 smp_wmb();
1995 set_bit(MMF_DUMPABLE, &mm->flags);
1996 break;
2000 static int __get_dumpable(unsigned long mm_flags)
2002 int ret;
2004 ret = mm_flags & MMF_DUMPABLE_MASK;
2005 return (ret >= 2) ? 2 : ret;
2008 int get_dumpable(struct mm_struct *mm)
2010 return __get_dumpable(mm->flags);
2013 static void wait_for_dump_helpers(struct file *file)
2015 struct pipe_inode_info *pipe;
2017 pipe = file->f_path.dentry->d_inode->i_pipe;
2019 pipe_lock(pipe);
2020 pipe->readers++;
2021 pipe->writers--;
2023 while ((pipe->readers > 1) && (!signal_pending(current))) {
2024 wake_up_interruptible_sync(&pipe->wait);
2025 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2026 pipe_wait(pipe);
2029 pipe->readers--;
2030 pipe->writers++;
2031 pipe_unlock(pipe);
2037 * umh_pipe_setup
2038 * helper function to customize the process used
2039 * to collect the core in userspace. Specifically
2040 * it sets up a pipe and installs it as fd 0 (stdin)
2041 * for the process. Returns 0 on success, or
2042 * PTR_ERR on failure.
2043 * Note that it also sets the core limit to 1. This
2044 * is a special value that we use to trap recursive
2045 * core dumps
2047 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2049 struct file *rp, *wp;
2050 struct fdtable *fdt;
2051 struct coredump_params *cp = (struct coredump_params *)info->data;
2052 struct files_struct *cf = current->files;
2054 wp = create_write_pipe(0);
2055 if (IS_ERR(wp))
2056 return PTR_ERR(wp);
2058 rp = create_read_pipe(wp, 0);
2059 if (IS_ERR(rp)) {
2060 free_write_pipe(wp);
2061 return PTR_ERR(rp);
2064 cp->file = wp;
2066 sys_close(0);
2067 fd_install(0, rp);
2068 spin_lock(&cf->file_lock);
2069 fdt = files_fdtable(cf);
2070 FD_SET(0, fdt->open_fds);
2071 FD_CLR(0, fdt->close_on_exec);
2072 spin_unlock(&cf->file_lock);
2074 /* and disallow core files too */
2075 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2077 return 0;
2080 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2082 struct core_state core_state;
2083 struct core_name cn;
2084 struct mm_struct *mm = current->mm;
2085 struct linux_binfmt * binfmt;
2086 const struct cred *old_cred;
2087 struct cred *cred;
2088 int retval = 0;
2089 int flag = 0;
2090 int ispipe;
2091 static atomic_t core_dump_count = ATOMIC_INIT(0);
2092 struct coredump_params cprm = {
2093 .signr = signr,
2094 .regs = regs,
2095 .limit = rlimit(RLIMIT_CORE),
2097 * We must use the same mm->flags while dumping core to avoid
2098 * inconsistency of bit flags, since this flag is not protected
2099 * by any locks.
2101 .mm_flags = mm->flags,
2104 audit_core_dumps(signr);
2106 binfmt = mm->binfmt;
2107 if (!binfmt || !binfmt->core_dump)
2108 goto fail;
2109 if (!__get_dumpable(cprm.mm_flags))
2110 goto fail;
2112 cred = prepare_creds();
2113 if (!cred)
2114 goto fail;
2116 * We cannot trust fsuid as being the "true" uid of the
2117 * process nor do we know its entire history. We only know it
2118 * was tainted so we dump it as root in mode 2.
2120 if (__get_dumpable(cprm.mm_flags) == 2) {
2121 /* Setuid core dump mode */
2122 flag = O_EXCL; /* Stop rewrite attacks */
2123 cred->fsuid = 0; /* Dump root private */
2126 retval = coredump_wait(exit_code, &core_state);
2127 if (retval < 0)
2128 goto fail_creds;
2130 old_cred = override_creds(cred);
2133 * Clear any false indication of pending signals that might
2134 * be seen by the filesystem code called to write the core file.
2136 clear_thread_flag(TIF_SIGPENDING);
2138 ispipe = format_corename(&cn, signr);
2140 if (ispipe) {
2141 int dump_count;
2142 char **helper_argv;
2144 if (ispipe < 0) {
2145 printk(KERN_WARNING "format_corename failed\n");
2146 printk(KERN_WARNING "Aborting core\n");
2147 goto fail_corename;
2150 if (cprm.limit == 1) {
2152 * Normally core limits are irrelevant to pipes, since
2153 * we're not writing to the file system, but we use
2154 * cprm.limit of 1 here as a speacial value. Any
2155 * non-1 limit gets set to RLIM_INFINITY below, but
2156 * a limit of 0 skips the dump. This is a consistent
2157 * way to catch recursive crashes. We can still crash
2158 * if the core_pattern binary sets RLIM_CORE = !1
2159 * but it runs as root, and can do lots of stupid things
2160 * Note that we use task_tgid_vnr here to grab the pid
2161 * of the process group leader. That way we get the
2162 * right pid if a thread in a multi-threaded
2163 * core_pattern process dies.
2165 printk(KERN_WARNING
2166 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2167 task_tgid_vnr(current), current->comm);
2168 printk(KERN_WARNING "Aborting core\n");
2169 goto fail_unlock;
2171 cprm.limit = RLIM_INFINITY;
2173 dump_count = atomic_inc_return(&core_dump_count);
2174 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2175 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2176 task_tgid_vnr(current), current->comm);
2177 printk(KERN_WARNING "Skipping core dump\n");
2178 goto fail_dropcount;
2181 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2182 if (!helper_argv) {
2183 printk(KERN_WARNING "%s failed to allocate memory\n",
2184 __func__);
2185 goto fail_dropcount;
2188 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2189 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2190 NULL, &cprm);
2191 argv_free(helper_argv);
2192 if (retval) {
2193 printk(KERN_INFO "Core dump to %s pipe failed\n",
2194 cn.corename);
2195 goto close_fail;
2197 } else {
2198 struct inode *inode;
2200 if (cprm.limit < binfmt->min_coredump)
2201 goto fail_unlock;
2203 cprm.file = filp_open(cn.corename,
2204 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2205 0600);
2206 if (IS_ERR(cprm.file))
2207 goto fail_unlock;
2209 inode = cprm.file->f_path.dentry->d_inode;
2210 if (inode->i_nlink > 1)
2211 goto close_fail;
2212 if (d_unhashed(cprm.file->f_path.dentry))
2213 goto close_fail;
2215 * AK: actually i see no reason to not allow this for named
2216 * pipes etc, but keep the previous behaviour for now.
2218 if (!S_ISREG(inode->i_mode))
2219 goto close_fail;
2221 * Dont allow local users get cute and trick others to coredump
2222 * into their pre-created files.
2224 if (inode->i_uid != current_fsuid())
2225 goto close_fail;
2226 if (!cprm.file->f_op || !cprm.file->f_op->write)
2227 goto close_fail;
2228 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2229 goto close_fail;
2232 retval = binfmt->core_dump(&cprm);
2233 if (retval)
2234 current->signal->group_exit_code |= 0x80;
2236 if (ispipe && core_pipe_limit)
2237 wait_for_dump_helpers(cprm.file);
2238 close_fail:
2239 if (cprm.file)
2240 filp_close(cprm.file, NULL);
2241 fail_dropcount:
2242 if (ispipe)
2243 atomic_dec(&core_dump_count);
2244 fail_unlock:
2245 kfree(cn.corename);
2246 fail_corename:
2247 coredump_finish(mm);
2248 revert_creds(old_cred);
2249 fail_creds:
2250 put_cred(cred);
2251 fail:
2252 return;
2256 * Core dumping helper functions. These are the only things you should
2257 * do on a core-file: use only these functions to write out all the
2258 * necessary info.
2260 int dump_write(struct file *file, const void *addr, int nr)
2262 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2264 EXPORT_SYMBOL(dump_write);
2266 int dump_seek(struct file *file, loff_t off)
2268 int ret = 1;
2270 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2271 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2272 return 0;
2273 } else {
2274 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2276 if (!buf)
2277 return 0;
2278 while (off > 0) {
2279 unsigned long n = off;
2281 if (n > PAGE_SIZE)
2282 n = PAGE_SIZE;
2283 if (!dump_write(file, buf, n)) {
2284 ret = 0;
2285 break;
2287 off -= n;
2289 free_page((unsigned long)buf);
2291 return ret;
2293 EXPORT_SYMBOL(dump_seek);