nfsd4: typo logical vs bitwise negate for want_mask
[linux-btrfs-devel.git] / fs / exec.c
blobda80612a35f42e9bc1bb2d09fe79090e44b3808b
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>
62 #include "internal.h"
64 int core_uses_pid;
65 char core_pattern[CORENAME_MAX_SIZE] = "core";
66 unsigned int core_pipe_limit;
67 int suid_dumpable = 0;
69 struct core_name {
70 char *corename;
71 int used, size;
73 static atomic_t call_count = ATOMIC_INIT(1);
75 /* The maximal length of core_pattern is also specified in sysctl.c */
77 static LIST_HEAD(formats);
78 static DEFINE_RWLOCK(binfmt_lock);
80 int __register_binfmt(struct linux_binfmt * fmt, int insert)
82 if (!fmt)
83 return -EINVAL;
84 write_lock(&binfmt_lock);
85 insert ? list_add(&fmt->lh, &formats) :
86 list_add_tail(&fmt->lh, &formats);
87 write_unlock(&binfmt_lock);
88 return 0;
91 EXPORT_SYMBOL(__register_binfmt);
93 void unregister_binfmt(struct linux_binfmt * fmt)
95 write_lock(&binfmt_lock);
96 list_del(&fmt->lh);
97 write_unlock(&binfmt_lock);
100 EXPORT_SYMBOL(unregister_binfmt);
102 static inline void put_binfmt(struct linux_binfmt * fmt)
104 module_put(fmt->module);
108 * Note that a shared library must be both readable and executable due to
109 * security reasons.
111 * Also note that we take the address to load from from the file itself.
113 SYSCALL_DEFINE1(uselib, const char __user *, library)
115 struct file *file;
116 char *tmp = getname(library);
117 int error = PTR_ERR(tmp);
118 static const struct open_flags uselib_flags = {
119 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
120 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
121 .intent = LOOKUP_OPEN
124 if (IS_ERR(tmp))
125 goto out;
127 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
128 putname(tmp);
129 error = PTR_ERR(file);
130 if (IS_ERR(file))
131 goto out;
133 error = -EINVAL;
134 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
135 goto exit;
137 error = -EACCES;
138 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
139 goto exit;
141 fsnotify_open(file);
143 error = -ENOEXEC;
144 if(file->f_op) {
145 struct linux_binfmt * fmt;
147 read_lock(&binfmt_lock);
148 list_for_each_entry(fmt, &formats, lh) {
149 if (!fmt->load_shlib)
150 continue;
151 if (!try_module_get(fmt->module))
152 continue;
153 read_unlock(&binfmt_lock);
154 error = fmt->load_shlib(file);
155 read_lock(&binfmt_lock);
156 put_binfmt(fmt);
157 if (error != -ENOEXEC)
158 break;
160 read_unlock(&binfmt_lock);
162 exit:
163 fput(file);
164 out:
165 return error;
168 #ifdef CONFIG_MMU
170 * The nascent bprm->mm is not visible until exec_mmap() but it can
171 * use a lot of memory, account these pages in current->mm temporary
172 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
173 * change the counter back via acct_arg_size(0).
175 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
177 struct mm_struct *mm = current->mm;
178 long diff = (long)(pages - bprm->vma_pages);
180 if (!mm || !diff)
181 return;
183 bprm->vma_pages = pages;
184 add_mm_counter(mm, MM_ANONPAGES, diff);
187 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
188 int write)
190 struct page *page;
191 int ret;
193 #ifdef CONFIG_STACK_GROWSUP
194 if (write) {
195 ret = expand_downwards(bprm->vma, pos);
196 if (ret < 0)
197 return NULL;
199 #endif
200 ret = get_user_pages(current, bprm->mm, pos,
201 1, write, 1, &page, NULL);
202 if (ret <= 0)
203 return NULL;
205 if (write) {
206 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
207 struct rlimit *rlim;
209 acct_arg_size(bprm, size / PAGE_SIZE);
212 * We've historically supported up to 32 pages (ARG_MAX)
213 * of argument strings even with small stacks
215 if (size <= ARG_MAX)
216 return page;
219 * Limit to 1/4-th the stack size for the argv+env strings.
220 * This ensures that:
221 * - the remaining binfmt code will not run out of stack space,
222 * - the program will have a reasonable amount of stack left
223 * to work from.
225 rlim = current->signal->rlim;
226 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
227 put_page(page);
228 return NULL;
232 return page;
235 static void put_arg_page(struct page *page)
237 put_page(page);
240 static void free_arg_page(struct linux_binprm *bprm, int i)
244 static void free_arg_pages(struct linux_binprm *bprm)
248 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
249 struct page *page)
251 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
254 static int __bprm_mm_init(struct linux_binprm *bprm)
256 int err;
257 struct vm_area_struct *vma = NULL;
258 struct mm_struct *mm = bprm->mm;
260 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
261 if (!vma)
262 return -ENOMEM;
264 down_write(&mm->mmap_sem);
265 vma->vm_mm = mm;
268 * Place the stack at the largest stack address the architecture
269 * supports. Later, we'll move this to an appropriate place. We don't
270 * use STACK_TOP because that can depend on attributes which aren't
271 * configured yet.
273 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
274 vma->vm_end = STACK_TOP_MAX;
275 vma->vm_start = vma->vm_end - PAGE_SIZE;
276 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
277 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
278 INIT_LIST_HEAD(&vma->anon_vma_chain);
280 err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
281 if (err)
282 goto err;
284 err = insert_vm_struct(mm, vma);
285 if (err)
286 goto err;
288 mm->stack_vm = mm->total_vm = 1;
289 up_write(&mm->mmap_sem);
290 bprm->p = vma->vm_end - sizeof(void *);
291 return 0;
292 err:
293 up_write(&mm->mmap_sem);
294 bprm->vma = NULL;
295 kmem_cache_free(vm_area_cachep, vma);
296 return err;
299 static bool valid_arg_len(struct linux_binprm *bprm, long len)
301 return len <= MAX_ARG_STRLEN;
304 #else
306 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
310 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
311 int write)
313 struct page *page;
315 page = bprm->page[pos / PAGE_SIZE];
316 if (!page && write) {
317 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
318 if (!page)
319 return NULL;
320 bprm->page[pos / PAGE_SIZE] = page;
323 return page;
326 static void put_arg_page(struct page *page)
330 static void free_arg_page(struct linux_binprm *bprm, int i)
332 if (bprm->page[i]) {
333 __free_page(bprm->page[i]);
334 bprm->page[i] = NULL;
338 static void free_arg_pages(struct linux_binprm *bprm)
340 int i;
342 for (i = 0; i < MAX_ARG_PAGES; i++)
343 free_arg_page(bprm, i);
346 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
347 struct page *page)
351 static int __bprm_mm_init(struct linux_binprm *bprm)
353 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
354 return 0;
357 static bool valid_arg_len(struct linux_binprm *bprm, long len)
359 return len <= bprm->p;
362 #endif /* CONFIG_MMU */
365 * Create a new mm_struct and populate it with a temporary stack
366 * vm_area_struct. We don't have enough context at this point to set the stack
367 * flags, permissions, and offset, so we use temporary values. We'll update
368 * them later in setup_arg_pages().
370 int bprm_mm_init(struct linux_binprm *bprm)
372 int err;
373 struct mm_struct *mm = NULL;
375 bprm->mm = mm = mm_alloc();
376 err = -ENOMEM;
377 if (!mm)
378 goto err;
380 err = init_new_context(current, mm);
381 if (err)
382 goto err;
384 err = __bprm_mm_init(bprm);
385 if (err)
386 goto err;
388 return 0;
390 err:
391 if (mm) {
392 bprm->mm = NULL;
393 mmdrop(mm);
396 return err;
399 struct user_arg_ptr {
400 #ifdef CONFIG_COMPAT
401 bool is_compat;
402 #endif
403 union {
404 const char __user *const __user *native;
405 #ifdef CONFIG_COMPAT
406 compat_uptr_t __user *compat;
407 #endif
408 } ptr;
411 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
413 const char __user *native;
415 #ifdef CONFIG_COMPAT
416 if (unlikely(argv.is_compat)) {
417 compat_uptr_t compat;
419 if (get_user(compat, argv.ptr.compat + nr))
420 return ERR_PTR(-EFAULT);
422 return compat_ptr(compat);
424 #endif
426 if (get_user(native, argv.ptr.native + nr))
427 return ERR_PTR(-EFAULT);
429 return native;
433 * count() counts the number of strings in array ARGV.
435 static int count(struct user_arg_ptr argv, int max)
437 int i = 0;
439 if (argv.ptr.native != NULL) {
440 for (;;) {
441 const char __user *p = get_user_arg_ptr(argv, i);
443 if (!p)
444 break;
446 if (IS_ERR(p))
447 return -EFAULT;
449 if (i++ >= max)
450 return -E2BIG;
452 if (fatal_signal_pending(current))
453 return -ERESTARTNOHAND;
454 cond_resched();
457 return i;
461 * 'copy_strings()' copies argument/environment strings from the old
462 * processes's memory to the new process's stack. The call to get_user_pages()
463 * ensures the destination page is created and not swapped out.
465 static int copy_strings(int argc, struct user_arg_ptr argv,
466 struct linux_binprm *bprm)
468 struct page *kmapped_page = NULL;
469 char *kaddr = NULL;
470 unsigned long kpos = 0;
471 int ret;
473 while (argc-- > 0) {
474 const char __user *str;
475 int len;
476 unsigned long pos;
478 ret = -EFAULT;
479 str = get_user_arg_ptr(argv, argc);
480 if (IS_ERR(str))
481 goto out;
483 len = strnlen_user(str, MAX_ARG_STRLEN);
484 if (!len)
485 goto out;
487 ret = -E2BIG;
488 if (!valid_arg_len(bprm, len))
489 goto out;
491 /* We're going to work our way backwords. */
492 pos = bprm->p;
493 str += len;
494 bprm->p -= len;
496 while (len > 0) {
497 int offset, bytes_to_copy;
499 if (fatal_signal_pending(current)) {
500 ret = -ERESTARTNOHAND;
501 goto out;
503 cond_resched();
505 offset = pos % PAGE_SIZE;
506 if (offset == 0)
507 offset = PAGE_SIZE;
509 bytes_to_copy = offset;
510 if (bytes_to_copy > len)
511 bytes_to_copy = len;
513 offset -= bytes_to_copy;
514 pos -= bytes_to_copy;
515 str -= bytes_to_copy;
516 len -= bytes_to_copy;
518 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
519 struct page *page;
521 page = get_arg_page(bprm, pos, 1);
522 if (!page) {
523 ret = -E2BIG;
524 goto out;
527 if (kmapped_page) {
528 flush_kernel_dcache_page(kmapped_page);
529 kunmap(kmapped_page);
530 put_arg_page(kmapped_page);
532 kmapped_page = page;
533 kaddr = kmap(kmapped_page);
534 kpos = pos & PAGE_MASK;
535 flush_arg_page(bprm, kpos, kmapped_page);
537 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
538 ret = -EFAULT;
539 goto out;
543 ret = 0;
544 out:
545 if (kmapped_page) {
546 flush_kernel_dcache_page(kmapped_page);
547 kunmap(kmapped_page);
548 put_arg_page(kmapped_page);
550 return ret;
554 * Like copy_strings, but get argv and its values from kernel memory.
556 int copy_strings_kernel(int argc, const char *const *__argv,
557 struct linux_binprm *bprm)
559 int r;
560 mm_segment_t oldfs = get_fs();
561 struct user_arg_ptr argv = {
562 .ptr.native = (const char __user *const __user *)__argv,
565 set_fs(KERNEL_DS);
566 r = copy_strings(argc, argv, bprm);
567 set_fs(oldfs);
569 return r;
571 EXPORT_SYMBOL(copy_strings_kernel);
573 #ifdef CONFIG_MMU
576 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
577 * the binfmt code determines where the new stack should reside, we shift it to
578 * its final location. The process proceeds as follows:
580 * 1) Use shift to calculate the new vma endpoints.
581 * 2) Extend vma to cover both the old and new ranges. This ensures the
582 * arguments passed to subsequent functions are consistent.
583 * 3) Move vma's page tables to the new range.
584 * 4) Free up any cleared pgd range.
585 * 5) Shrink the vma to cover only the new range.
587 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
589 struct mm_struct *mm = vma->vm_mm;
590 unsigned long old_start = vma->vm_start;
591 unsigned long old_end = vma->vm_end;
592 unsigned long length = old_end - old_start;
593 unsigned long new_start = old_start - shift;
594 unsigned long new_end = old_end - shift;
595 struct mmu_gather tlb;
597 BUG_ON(new_start > new_end);
600 * ensure there are no vmas between where we want to go
601 * and where we are
603 if (vma != find_vma(mm, new_start))
604 return -EFAULT;
607 * cover the whole range: [new_start, old_end)
609 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
610 return -ENOMEM;
613 * move the page tables downwards, on failure we rely on
614 * process cleanup to remove whatever mess we made.
616 if (length != move_page_tables(vma, old_start,
617 vma, new_start, length))
618 return -ENOMEM;
620 lru_add_drain();
621 tlb_gather_mmu(&tlb, mm, 0);
622 if (new_end > old_start) {
624 * when the old and new regions overlap clear from new_end.
626 free_pgd_range(&tlb, new_end, old_end, new_end,
627 vma->vm_next ? vma->vm_next->vm_start : 0);
628 } else {
630 * otherwise, clean from old_start; this is done to not touch
631 * the address space in [new_end, old_start) some architectures
632 * have constraints on va-space that make this illegal (IA64) -
633 * for the others its just a little faster.
635 free_pgd_range(&tlb, old_start, old_end, new_end,
636 vma->vm_next ? vma->vm_next->vm_start : 0);
638 tlb_finish_mmu(&tlb, new_end, old_end);
641 * Shrink the vma to just the new range. Always succeeds.
643 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
645 return 0;
649 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
650 * the stack is optionally relocated, and some extra space is added.
652 int setup_arg_pages(struct linux_binprm *bprm,
653 unsigned long stack_top,
654 int executable_stack)
656 unsigned long ret;
657 unsigned long stack_shift;
658 struct mm_struct *mm = current->mm;
659 struct vm_area_struct *vma = bprm->vma;
660 struct vm_area_struct *prev = NULL;
661 unsigned long vm_flags;
662 unsigned long stack_base;
663 unsigned long stack_size;
664 unsigned long stack_expand;
665 unsigned long rlim_stack;
667 #ifdef CONFIG_STACK_GROWSUP
668 /* Limit stack size to 1GB */
669 stack_base = rlimit_max(RLIMIT_STACK);
670 if (stack_base > (1 << 30))
671 stack_base = 1 << 30;
673 /* Make sure we didn't let the argument array grow too large. */
674 if (vma->vm_end - vma->vm_start > stack_base)
675 return -ENOMEM;
677 stack_base = PAGE_ALIGN(stack_top - stack_base);
679 stack_shift = vma->vm_start - stack_base;
680 mm->arg_start = bprm->p - stack_shift;
681 bprm->p = vma->vm_end - stack_shift;
682 #else
683 stack_top = arch_align_stack(stack_top);
684 stack_top = PAGE_ALIGN(stack_top);
686 if (unlikely(stack_top < mmap_min_addr) ||
687 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
688 return -ENOMEM;
690 stack_shift = vma->vm_end - stack_top;
692 bprm->p -= stack_shift;
693 mm->arg_start = bprm->p;
694 #endif
696 if (bprm->loader)
697 bprm->loader -= stack_shift;
698 bprm->exec -= stack_shift;
700 down_write(&mm->mmap_sem);
701 vm_flags = VM_STACK_FLAGS;
704 * Adjust stack execute permissions; explicitly enable for
705 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
706 * (arch default) otherwise.
708 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
709 vm_flags |= VM_EXEC;
710 else if (executable_stack == EXSTACK_DISABLE_X)
711 vm_flags &= ~VM_EXEC;
712 vm_flags |= mm->def_flags;
713 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
715 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
716 vm_flags);
717 if (ret)
718 goto out_unlock;
719 BUG_ON(prev != vma);
721 /* Move stack pages down in memory. */
722 if (stack_shift) {
723 ret = shift_arg_pages(vma, stack_shift);
724 if (ret)
725 goto out_unlock;
728 /* mprotect_fixup is overkill to remove the temporary stack flags */
729 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
731 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
732 stack_size = vma->vm_end - vma->vm_start;
734 * Align this down to a page boundary as expand_stack
735 * will align it up.
737 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
738 #ifdef CONFIG_STACK_GROWSUP
739 if (stack_size + stack_expand > rlim_stack)
740 stack_base = vma->vm_start + rlim_stack;
741 else
742 stack_base = vma->vm_end + stack_expand;
743 #else
744 if (stack_size + stack_expand > rlim_stack)
745 stack_base = vma->vm_end - rlim_stack;
746 else
747 stack_base = vma->vm_start - stack_expand;
748 #endif
749 current->mm->start_stack = bprm->p;
750 ret = expand_stack(vma, stack_base);
751 if (ret)
752 ret = -EFAULT;
754 out_unlock:
755 up_write(&mm->mmap_sem);
756 return ret;
758 EXPORT_SYMBOL(setup_arg_pages);
760 #endif /* CONFIG_MMU */
762 struct file *open_exec(const char *name)
764 struct file *file;
765 int err;
766 static const struct open_flags open_exec_flags = {
767 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
768 .acc_mode = MAY_EXEC | MAY_OPEN,
769 .intent = LOOKUP_OPEN
772 file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
773 if (IS_ERR(file))
774 goto out;
776 err = -EACCES;
777 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
778 goto exit;
780 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
781 goto exit;
783 fsnotify_open(file);
785 err = deny_write_access(file);
786 if (err)
787 goto exit;
789 out:
790 return file;
792 exit:
793 fput(file);
794 return ERR_PTR(err);
796 EXPORT_SYMBOL(open_exec);
798 int kernel_read(struct file *file, loff_t offset,
799 char *addr, unsigned long count)
801 mm_segment_t old_fs;
802 loff_t pos = offset;
803 int result;
805 old_fs = get_fs();
806 set_fs(get_ds());
807 /* The cast to a user pointer is valid due to the set_fs() */
808 result = vfs_read(file, (void __user *)addr, count, &pos);
809 set_fs(old_fs);
810 return result;
813 EXPORT_SYMBOL(kernel_read);
815 static int exec_mmap(struct mm_struct *mm)
817 struct task_struct *tsk;
818 struct mm_struct * old_mm, *active_mm;
820 /* Notify parent that we're no longer interested in the old VM */
821 tsk = current;
822 old_mm = current->mm;
823 sync_mm_rss(tsk, old_mm);
824 mm_release(tsk, old_mm);
826 if (old_mm) {
828 * Make sure that if there is a core dump in progress
829 * for the old mm, we get out and die instead of going
830 * through with the exec. We must hold mmap_sem around
831 * checking core_state and changing tsk->mm.
833 down_read(&old_mm->mmap_sem);
834 if (unlikely(old_mm->core_state)) {
835 up_read(&old_mm->mmap_sem);
836 return -EINTR;
839 task_lock(tsk);
840 active_mm = tsk->active_mm;
841 tsk->mm = mm;
842 tsk->active_mm = mm;
843 activate_mm(active_mm, mm);
844 if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
845 atomic_dec(&old_mm->oom_disable_count);
846 atomic_inc(&tsk->mm->oom_disable_count);
848 task_unlock(tsk);
849 arch_pick_mmap_layout(mm);
850 if (old_mm) {
851 up_read(&old_mm->mmap_sem);
852 BUG_ON(active_mm != old_mm);
853 mm_update_next_owner(old_mm);
854 mmput(old_mm);
855 return 0;
857 mmdrop(active_mm);
858 return 0;
862 * This function makes sure the current process has its own signal table,
863 * so that flush_signal_handlers can later reset the handlers without
864 * disturbing other processes. (Other processes might share the signal
865 * table via the CLONE_SIGHAND option to clone().)
867 static int de_thread(struct task_struct *tsk)
869 struct signal_struct *sig = tsk->signal;
870 struct sighand_struct *oldsighand = tsk->sighand;
871 spinlock_t *lock = &oldsighand->siglock;
873 if (thread_group_empty(tsk))
874 goto no_thread_group;
877 * Kill all other threads in the thread group.
879 spin_lock_irq(lock);
880 if (signal_group_exit(sig)) {
882 * Another group action in progress, just
883 * return so that the signal is processed.
885 spin_unlock_irq(lock);
886 return -EAGAIN;
889 sig->group_exit_task = tsk;
890 sig->notify_count = zap_other_threads(tsk);
891 if (!thread_group_leader(tsk))
892 sig->notify_count--;
894 while (sig->notify_count) {
895 __set_current_state(TASK_UNINTERRUPTIBLE);
896 spin_unlock_irq(lock);
897 schedule();
898 spin_lock_irq(lock);
900 spin_unlock_irq(lock);
903 * At this point all other threads have exited, all we have to
904 * do is to wait for the thread group leader to become inactive,
905 * and to assume its PID:
907 if (!thread_group_leader(tsk)) {
908 struct task_struct *leader = tsk->group_leader;
910 sig->notify_count = -1; /* for exit_notify() */
911 for (;;) {
912 write_lock_irq(&tasklist_lock);
913 if (likely(leader->exit_state))
914 break;
915 __set_current_state(TASK_UNINTERRUPTIBLE);
916 write_unlock_irq(&tasklist_lock);
917 schedule();
921 * The only record we have of the real-time age of a
922 * process, regardless of execs it's done, is start_time.
923 * All the past CPU time is accumulated in signal_struct
924 * from sister threads now dead. But in this non-leader
925 * exec, nothing survives from the original leader thread,
926 * whose birth marks the true age of this process now.
927 * When we take on its identity by switching to its PID, we
928 * also take its birthdate (always earlier than our own).
930 tsk->start_time = leader->start_time;
932 BUG_ON(!same_thread_group(leader, tsk));
933 BUG_ON(has_group_leader_pid(tsk));
935 * An exec() starts a new thread group with the
936 * TGID of the previous thread group. Rehash the
937 * two threads with a switched PID, and release
938 * the former thread group leader:
941 /* Become a process group leader with the old leader's pid.
942 * The old leader becomes a thread of the this thread group.
943 * Note: The old leader also uses this pid until release_task
944 * is called. Odd but simple and correct.
946 detach_pid(tsk, PIDTYPE_PID);
947 tsk->pid = leader->pid;
948 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
949 transfer_pid(leader, tsk, PIDTYPE_PGID);
950 transfer_pid(leader, tsk, PIDTYPE_SID);
952 list_replace_rcu(&leader->tasks, &tsk->tasks);
953 list_replace_init(&leader->sibling, &tsk->sibling);
955 tsk->group_leader = tsk;
956 leader->group_leader = tsk;
958 tsk->exit_signal = SIGCHLD;
959 leader->exit_signal = -1;
961 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
962 leader->exit_state = EXIT_DEAD;
965 * We are going to release_task()->ptrace_unlink() silently,
966 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
967 * the tracer wont't block again waiting for this thread.
969 if (unlikely(leader->ptrace))
970 __wake_up_parent(leader, leader->parent);
971 write_unlock_irq(&tasklist_lock);
973 release_task(leader);
976 sig->group_exit_task = NULL;
977 sig->notify_count = 0;
979 no_thread_group:
980 if (current->mm)
981 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
983 exit_itimers(sig);
984 flush_itimer_signals();
986 if (atomic_read(&oldsighand->count) != 1) {
987 struct sighand_struct *newsighand;
989 * This ->sighand is shared with the CLONE_SIGHAND
990 * but not CLONE_THREAD task, switch to the new one.
992 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
993 if (!newsighand)
994 return -ENOMEM;
996 atomic_set(&newsighand->count, 1);
997 memcpy(newsighand->action, oldsighand->action,
998 sizeof(newsighand->action));
1000 write_lock_irq(&tasklist_lock);
1001 spin_lock(&oldsighand->siglock);
1002 rcu_assign_pointer(tsk->sighand, newsighand);
1003 spin_unlock(&oldsighand->siglock);
1004 write_unlock_irq(&tasklist_lock);
1006 __cleanup_sighand(oldsighand);
1009 BUG_ON(!thread_group_leader(tsk));
1010 return 0;
1014 * These functions flushes out all traces of the currently running executable
1015 * so that a new one can be started
1017 static void flush_old_files(struct files_struct * files)
1019 long j = -1;
1020 struct fdtable *fdt;
1022 spin_lock(&files->file_lock);
1023 for (;;) {
1024 unsigned long set, i;
1026 j++;
1027 i = j * __NFDBITS;
1028 fdt = files_fdtable(files);
1029 if (i >= fdt->max_fds)
1030 break;
1031 set = fdt->close_on_exec->fds_bits[j];
1032 if (!set)
1033 continue;
1034 fdt->close_on_exec->fds_bits[j] = 0;
1035 spin_unlock(&files->file_lock);
1036 for ( ; set ; i++,set >>= 1) {
1037 if (set & 1) {
1038 sys_close(i);
1041 spin_lock(&files->file_lock);
1044 spin_unlock(&files->file_lock);
1047 char *get_task_comm(char *buf, struct task_struct *tsk)
1049 /* buf must be at least sizeof(tsk->comm) in size */
1050 task_lock(tsk);
1051 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1052 task_unlock(tsk);
1053 return buf;
1055 EXPORT_SYMBOL_GPL(get_task_comm);
1057 void set_task_comm(struct task_struct *tsk, char *buf)
1059 task_lock(tsk);
1062 * Threads may access current->comm without holding
1063 * the task lock, so write the string carefully.
1064 * Readers without a lock may see incomplete new
1065 * names but are safe from non-terminating string reads.
1067 memset(tsk->comm, 0, TASK_COMM_LEN);
1068 wmb();
1069 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1070 task_unlock(tsk);
1071 perf_event_comm(tsk);
1074 int flush_old_exec(struct linux_binprm * bprm)
1076 int retval;
1079 * Make sure we have a private signal table and that
1080 * we are unassociated from the previous thread group.
1082 retval = de_thread(current);
1083 if (retval)
1084 goto out;
1086 set_mm_exe_file(bprm->mm, bprm->file);
1089 * Release all of the old mmap stuff
1091 acct_arg_size(bprm, 0);
1092 retval = exec_mmap(bprm->mm);
1093 if (retval)
1094 goto out;
1096 bprm->mm = NULL; /* We're using it now */
1098 set_fs(USER_DS);
1099 current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1100 flush_thread();
1101 current->personality &= ~bprm->per_clear;
1103 return 0;
1105 out:
1106 return retval;
1108 EXPORT_SYMBOL(flush_old_exec);
1110 void would_dump(struct linux_binprm *bprm, struct file *file)
1112 if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1113 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1115 EXPORT_SYMBOL(would_dump);
1117 void setup_new_exec(struct linux_binprm * bprm)
1119 int i, ch;
1120 const char *name;
1121 char tcomm[sizeof(current->comm)];
1123 arch_pick_mmap_layout(current->mm);
1125 /* This is the point of no return */
1126 current->sas_ss_sp = current->sas_ss_size = 0;
1128 if (current_euid() == current_uid() && current_egid() == current_gid())
1129 set_dumpable(current->mm, 1);
1130 else
1131 set_dumpable(current->mm, suid_dumpable);
1133 name = bprm->filename;
1135 /* Copies the binary name from after last slash */
1136 for (i=0; (ch = *(name++)) != '\0';) {
1137 if (ch == '/')
1138 i = 0; /* overwrite what we wrote */
1139 else
1140 if (i < (sizeof(tcomm) - 1))
1141 tcomm[i++] = ch;
1143 tcomm[i] = '\0';
1144 set_task_comm(current, tcomm);
1146 /* Set the new mm task size. We have to do that late because it may
1147 * depend on TIF_32BIT which is only updated in flush_thread() on
1148 * some architectures like powerpc
1150 current->mm->task_size = TASK_SIZE;
1152 /* install the new credentials */
1153 if (bprm->cred->uid != current_euid() ||
1154 bprm->cred->gid != current_egid()) {
1155 current->pdeath_signal = 0;
1156 } else {
1157 would_dump(bprm, bprm->file);
1158 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1159 set_dumpable(current->mm, suid_dumpable);
1163 * Flush performance counters when crossing a
1164 * security domain:
1166 if (!get_dumpable(current->mm))
1167 perf_event_exit_task(current);
1169 /* An exec changes our domain. We are no longer part of the thread
1170 group */
1172 current->self_exec_id++;
1174 flush_signal_handlers(current, 0);
1175 flush_old_files(current->files);
1177 EXPORT_SYMBOL(setup_new_exec);
1180 * Prepare credentials and lock ->cred_guard_mutex.
1181 * install_exec_creds() commits the new creds and drops the lock.
1182 * Or, if exec fails before, free_bprm() should release ->cred and
1183 * and unlock.
1185 int prepare_bprm_creds(struct linux_binprm *bprm)
1187 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1188 return -ERESTARTNOINTR;
1190 bprm->cred = prepare_exec_creds();
1191 if (likely(bprm->cred))
1192 return 0;
1194 mutex_unlock(&current->signal->cred_guard_mutex);
1195 return -ENOMEM;
1198 void free_bprm(struct linux_binprm *bprm)
1200 free_arg_pages(bprm);
1201 if (bprm->cred) {
1202 mutex_unlock(&current->signal->cred_guard_mutex);
1203 abort_creds(bprm->cred);
1205 kfree(bprm);
1209 * install the new credentials for this executable
1211 void install_exec_creds(struct linux_binprm *bprm)
1213 security_bprm_committing_creds(bprm);
1215 commit_creds(bprm->cred);
1216 bprm->cred = NULL;
1218 * cred_guard_mutex must be held at least to this point to prevent
1219 * ptrace_attach() from altering our determination of the task's
1220 * credentials; any time after this it may be unlocked.
1222 security_bprm_committed_creds(bprm);
1223 mutex_unlock(&current->signal->cred_guard_mutex);
1225 EXPORT_SYMBOL(install_exec_creds);
1228 * determine how safe it is to execute the proposed program
1229 * - the caller must hold ->cred_guard_mutex to protect against
1230 * PTRACE_ATTACH
1232 int check_unsafe_exec(struct linux_binprm *bprm)
1234 struct task_struct *p = current, *t;
1235 unsigned n_fs;
1236 int res = 0;
1238 if (p->ptrace) {
1239 if (p->ptrace & PT_PTRACE_CAP)
1240 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1241 else
1242 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1245 n_fs = 1;
1246 spin_lock(&p->fs->lock);
1247 rcu_read_lock();
1248 for (t = next_thread(p); t != p; t = next_thread(t)) {
1249 if (t->fs == p->fs)
1250 n_fs++;
1252 rcu_read_unlock();
1254 if (p->fs->users > n_fs) {
1255 bprm->unsafe |= LSM_UNSAFE_SHARE;
1256 } else {
1257 res = -EAGAIN;
1258 if (!p->fs->in_exec) {
1259 p->fs->in_exec = 1;
1260 res = 1;
1263 spin_unlock(&p->fs->lock);
1265 return res;
1269 * Fill the binprm structure from the inode.
1270 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1272 * This may be called multiple times for binary chains (scripts for example).
1274 int prepare_binprm(struct linux_binprm *bprm)
1276 umode_t mode;
1277 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1278 int retval;
1280 mode = inode->i_mode;
1281 if (bprm->file->f_op == NULL)
1282 return -EACCES;
1284 /* clear any previous set[ug]id data from a previous binary */
1285 bprm->cred->euid = current_euid();
1286 bprm->cred->egid = current_egid();
1288 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1289 /* Set-uid? */
1290 if (mode & S_ISUID) {
1291 bprm->per_clear |= PER_CLEAR_ON_SETID;
1292 bprm->cred->euid = inode->i_uid;
1295 /* Set-gid? */
1297 * If setgid is set but no group execute bit then this
1298 * is a candidate for mandatory locking, not a setgid
1299 * executable.
1301 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1302 bprm->per_clear |= PER_CLEAR_ON_SETID;
1303 bprm->cred->egid = inode->i_gid;
1307 /* fill in binprm security blob */
1308 retval = security_bprm_set_creds(bprm);
1309 if (retval)
1310 return retval;
1311 bprm->cred_prepared = 1;
1313 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1314 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1317 EXPORT_SYMBOL(prepare_binprm);
1320 * Arguments are '\0' separated strings found at the location bprm->p
1321 * points to; chop off the first by relocating brpm->p to right after
1322 * the first '\0' encountered.
1324 int remove_arg_zero(struct linux_binprm *bprm)
1326 int ret = 0;
1327 unsigned long offset;
1328 char *kaddr;
1329 struct page *page;
1331 if (!bprm->argc)
1332 return 0;
1334 do {
1335 offset = bprm->p & ~PAGE_MASK;
1336 page = get_arg_page(bprm, bprm->p, 0);
1337 if (!page) {
1338 ret = -EFAULT;
1339 goto out;
1341 kaddr = kmap_atomic(page, KM_USER0);
1343 for (; offset < PAGE_SIZE && kaddr[offset];
1344 offset++, bprm->p++)
1347 kunmap_atomic(kaddr, KM_USER0);
1348 put_arg_page(page);
1350 if (offset == PAGE_SIZE)
1351 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1352 } while (offset == PAGE_SIZE);
1354 bprm->p++;
1355 bprm->argc--;
1356 ret = 0;
1358 out:
1359 return ret;
1361 EXPORT_SYMBOL(remove_arg_zero);
1364 * cycle the list of binary formats handler, until one recognizes the image
1366 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1368 unsigned int depth = bprm->recursion_depth;
1369 int try,retval;
1370 struct linux_binfmt *fmt;
1371 pid_t old_pid;
1373 retval = security_bprm_check(bprm);
1374 if (retval)
1375 return retval;
1377 retval = audit_bprm(bprm);
1378 if (retval)
1379 return retval;
1381 /* Need to fetch pid before load_binary changes it */
1382 rcu_read_lock();
1383 old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1384 rcu_read_unlock();
1386 retval = -ENOENT;
1387 for (try=0; try<2; try++) {
1388 read_lock(&binfmt_lock);
1389 list_for_each_entry(fmt, &formats, lh) {
1390 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1391 if (!fn)
1392 continue;
1393 if (!try_module_get(fmt->module))
1394 continue;
1395 read_unlock(&binfmt_lock);
1396 retval = fn(bprm, regs);
1398 * Restore the depth counter to its starting value
1399 * in this call, so we don't have to rely on every
1400 * load_binary function to restore it on return.
1402 bprm->recursion_depth = depth;
1403 if (retval >= 0) {
1404 if (depth == 0)
1405 ptrace_event(PTRACE_EVENT_EXEC,
1406 old_pid);
1407 put_binfmt(fmt);
1408 allow_write_access(bprm->file);
1409 if (bprm->file)
1410 fput(bprm->file);
1411 bprm->file = NULL;
1412 current->did_exec = 1;
1413 proc_exec_connector(current);
1414 return retval;
1416 read_lock(&binfmt_lock);
1417 put_binfmt(fmt);
1418 if (retval != -ENOEXEC || bprm->mm == NULL)
1419 break;
1420 if (!bprm->file) {
1421 read_unlock(&binfmt_lock);
1422 return retval;
1425 read_unlock(&binfmt_lock);
1426 #ifdef CONFIG_MODULES
1427 if (retval != -ENOEXEC || bprm->mm == NULL) {
1428 break;
1429 } else {
1430 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1431 if (printable(bprm->buf[0]) &&
1432 printable(bprm->buf[1]) &&
1433 printable(bprm->buf[2]) &&
1434 printable(bprm->buf[3]))
1435 break; /* -ENOEXEC */
1436 if (try)
1437 break; /* -ENOEXEC */
1438 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1440 #else
1441 break;
1442 #endif
1444 return retval;
1447 EXPORT_SYMBOL(search_binary_handler);
1450 * sys_execve() executes a new program.
1452 static int do_execve_common(const char *filename,
1453 struct user_arg_ptr argv,
1454 struct user_arg_ptr envp,
1455 struct pt_regs *regs)
1457 struct linux_binprm *bprm;
1458 struct file *file;
1459 struct files_struct *displaced;
1460 bool clear_in_exec;
1461 int retval;
1463 retval = unshare_files(&displaced);
1464 if (retval)
1465 goto out_ret;
1467 retval = -ENOMEM;
1468 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1469 if (!bprm)
1470 goto out_files;
1472 retval = prepare_bprm_creds(bprm);
1473 if (retval)
1474 goto out_free;
1476 retval = check_unsafe_exec(bprm);
1477 if (retval < 0)
1478 goto out_free;
1479 clear_in_exec = retval;
1480 current->in_execve = 1;
1482 file = open_exec(filename);
1483 retval = PTR_ERR(file);
1484 if (IS_ERR(file))
1485 goto out_unmark;
1487 sched_exec();
1489 bprm->file = file;
1490 bprm->filename = filename;
1491 bprm->interp = filename;
1493 retval = bprm_mm_init(bprm);
1494 if (retval)
1495 goto out_file;
1497 bprm->argc = count(argv, MAX_ARG_STRINGS);
1498 if ((retval = bprm->argc) < 0)
1499 goto out;
1501 bprm->envc = count(envp, MAX_ARG_STRINGS);
1502 if ((retval = bprm->envc) < 0)
1503 goto out;
1505 retval = prepare_binprm(bprm);
1506 if (retval < 0)
1507 goto out;
1509 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1510 if (retval < 0)
1511 goto out;
1513 bprm->exec = bprm->p;
1514 retval = copy_strings(bprm->envc, envp, bprm);
1515 if (retval < 0)
1516 goto out;
1518 retval = copy_strings(bprm->argc, argv, bprm);
1519 if (retval < 0)
1520 goto out;
1522 retval = search_binary_handler(bprm,regs);
1523 if (retval < 0)
1524 goto out;
1526 /* execve succeeded */
1527 current->fs->in_exec = 0;
1528 current->in_execve = 0;
1529 acct_update_integrals(current);
1530 free_bprm(bprm);
1531 if (displaced)
1532 put_files_struct(displaced);
1533 return retval;
1535 out:
1536 if (bprm->mm) {
1537 acct_arg_size(bprm, 0);
1538 mmput(bprm->mm);
1541 out_file:
1542 if (bprm->file) {
1543 allow_write_access(bprm->file);
1544 fput(bprm->file);
1547 out_unmark:
1548 if (clear_in_exec)
1549 current->fs->in_exec = 0;
1550 current->in_execve = 0;
1552 out_free:
1553 free_bprm(bprm);
1555 out_files:
1556 if (displaced)
1557 reset_files_struct(displaced);
1558 out_ret:
1559 return retval;
1562 int do_execve(const char *filename,
1563 const char __user *const __user *__argv,
1564 const char __user *const __user *__envp,
1565 struct pt_regs *regs)
1567 struct user_arg_ptr argv = { .ptr.native = __argv };
1568 struct user_arg_ptr envp = { .ptr.native = __envp };
1569 return do_execve_common(filename, argv, envp, regs);
1572 #ifdef CONFIG_COMPAT
1573 int compat_do_execve(char *filename,
1574 compat_uptr_t __user *__argv,
1575 compat_uptr_t __user *__envp,
1576 struct pt_regs *regs)
1578 struct user_arg_ptr argv = {
1579 .is_compat = true,
1580 .ptr.compat = __argv,
1582 struct user_arg_ptr envp = {
1583 .is_compat = true,
1584 .ptr.compat = __envp,
1586 return do_execve_common(filename, argv, envp, regs);
1588 #endif
1590 void set_binfmt(struct linux_binfmt *new)
1592 struct mm_struct *mm = current->mm;
1594 if (mm->binfmt)
1595 module_put(mm->binfmt->module);
1597 mm->binfmt = new;
1598 if (new)
1599 __module_get(new->module);
1602 EXPORT_SYMBOL(set_binfmt);
1604 static int expand_corename(struct core_name *cn)
1606 char *old_corename = cn->corename;
1608 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1609 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1611 if (!cn->corename) {
1612 kfree(old_corename);
1613 return -ENOMEM;
1616 return 0;
1619 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1621 char *cur;
1622 int need;
1623 int ret;
1624 va_list arg;
1626 va_start(arg, fmt);
1627 need = vsnprintf(NULL, 0, fmt, arg);
1628 va_end(arg);
1630 if (likely(need < cn->size - cn->used - 1))
1631 goto out_printf;
1633 ret = expand_corename(cn);
1634 if (ret)
1635 goto expand_fail;
1637 out_printf:
1638 cur = cn->corename + cn->used;
1639 va_start(arg, fmt);
1640 vsnprintf(cur, need + 1, fmt, arg);
1641 va_end(arg);
1642 cn->used += need;
1643 return 0;
1645 expand_fail:
1646 return ret;
1649 static void cn_escape(char *str)
1651 for (; *str; str++)
1652 if (*str == '/')
1653 *str = '!';
1656 static int cn_print_exe_file(struct core_name *cn)
1658 struct file *exe_file;
1659 char *pathbuf, *path;
1660 int ret;
1662 exe_file = get_mm_exe_file(current->mm);
1663 if (!exe_file) {
1664 char *commstart = cn->corename + cn->used;
1665 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1666 cn_escape(commstart);
1667 return ret;
1670 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1671 if (!pathbuf) {
1672 ret = -ENOMEM;
1673 goto put_exe_file;
1676 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1677 if (IS_ERR(path)) {
1678 ret = PTR_ERR(path);
1679 goto free_buf;
1682 cn_escape(path);
1684 ret = cn_printf(cn, "%s", path);
1686 free_buf:
1687 kfree(pathbuf);
1688 put_exe_file:
1689 fput(exe_file);
1690 return ret;
1693 /* format_corename will inspect the pattern parameter, and output a
1694 * name into corename, which must have space for at least
1695 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1697 static int format_corename(struct core_name *cn, long signr)
1699 const struct cred *cred = current_cred();
1700 const char *pat_ptr = core_pattern;
1701 int ispipe = (*pat_ptr == '|');
1702 int pid_in_pattern = 0;
1703 int err = 0;
1705 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1706 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1707 cn->used = 0;
1709 if (!cn->corename)
1710 return -ENOMEM;
1712 /* Repeat as long as we have more pattern to process and more output
1713 space */
1714 while (*pat_ptr) {
1715 if (*pat_ptr != '%') {
1716 if (*pat_ptr == 0)
1717 goto out;
1718 err = cn_printf(cn, "%c", *pat_ptr++);
1719 } else {
1720 switch (*++pat_ptr) {
1721 /* single % at the end, drop that */
1722 case 0:
1723 goto out;
1724 /* Double percent, output one percent */
1725 case '%':
1726 err = cn_printf(cn, "%c", '%');
1727 break;
1728 /* pid */
1729 case 'p':
1730 pid_in_pattern = 1;
1731 err = cn_printf(cn, "%d",
1732 task_tgid_vnr(current));
1733 break;
1734 /* uid */
1735 case 'u':
1736 err = cn_printf(cn, "%d", cred->uid);
1737 break;
1738 /* gid */
1739 case 'g':
1740 err = cn_printf(cn, "%d", cred->gid);
1741 break;
1742 /* signal that caused the coredump */
1743 case 's':
1744 err = cn_printf(cn, "%ld", signr);
1745 break;
1746 /* UNIX time of coredump */
1747 case 't': {
1748 struct timeval tv;
1749 do_gettimeofday(&tv);
1750 err = cn_printf(cn, "%lu", tv.tv_sec);
1751 break;
1753 /* hostname */
1754 case 'h': {
1755 char *namestart = cn->corename + cn->used;
1756 down_read(&uts_sem);
1757 err = cn_printf(cn, "%s",
1758 utsname()->nodename);
1759 up_read(&uts_sem);
1760 cn_escape(namestart);
1761 break;
1763 /* executable */
1764 case 'e': {
1765 char *commstart = cn->corename + cn->used;
1766 err = cn_printf(cn, "%s", current->comm);
1767 cn_escape(commstart);
1768 break;
1770 case 'E':
1771 err = cn_print_exe_file(cn);
1772 break;
1773 /* core limit size */
1774 case 'c':
1775 err = cn_printf(cn, "%lu",
1776 rlimit(RLIMIT_CORE));
1777 break;
1778 default:
1779 break;
1781 ++pat_ptr;
1784 if (err)
1785 return err;
1788 /* Backward compatibility with core_uses_pid:
1790 * If core_pattern does not include a %p (as is the default)
1791 * and core_uses_pid is set, then .%pid will be appended to
1792 * the filename. Do not do this for piped commands. */
1793 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1794 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1795 if (err)
1796 return err;
1798 out:
1799 return ispipe;
1802 static int zap_process(struct task_struct *start, int exit_code)
1804 struct task_struct *t;
1805 int nr = 0;
1807 start->signal->flags = SIGNAL_GROUP_EXIT;
1808 start->signal->group_exit_code = exit_code;
1809 start->signal->group_stop_count = 0;
1811 t = start;
1812 do {
1813 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1814 if (t != current && t->mm) {
1815 sigaddset(&t->pending.signal, SIGKILL);
1816 signal_wake_up(t, 1);
1817 nr++;
1819 } while_each_thread(start, t);
1821 return nr;
1824 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1825 struct core_state *core_state, int exit_code)
1827 struct task_struct *g, *p;
1828 unsigned long flags;
1829 int nr = -EAGAIN;
1831 spin_lock_irq(&tsk->sighand->siglock);
1832 if (!signal_group_exit(tsk->signal)) {
1833 mm->core_state = core_state;
1834 nr = zap_process(tsk, exit_code);
1836 spin_unlock_irq(&tsk->sighand->siglock);
1837 if (unlikely(nr < 0))
1838 return nr;
1840 if (atomic_read(&mm->mm_users) == nr + 1)
1841 goto done;
1843 * We should find and kill all tasks which use this mm, and we should
1844 * count them correctly into ->nr_threads. We don't take tasklist
1845 * lock, but this is safe wrt:
1847 * fork:
1848 * None of sub-threads can fork after zap_process(leader). All
1849 * processes which were created before this point should be
1850 * visible to zap_threads() because copy_process() adds the new
1851 * process to the tail of init_task.tasks list, and lock/unlock
1852 * of ->siglock provides a memory barrier.
1854 * do_exit:
1855 * The caller holds mm->mmap_sem. This means that the task which
1856 * uses this mm can't pass exit_mm(), so it can't exit or clear
1857 * its ->mm.
1859 * de_thread:
1860 * It does list_replace_rcu(&leader->tasks, &current->tasks),
1861 * we must see either old or new leader, this does not matter.
1862 * However, it can change p->sighand, so lock_task_sighand(p)
1863 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1864 * it can't fail.
1866 * Note also that "g" can be the old leader with ->mm == NULL
1867 * and already unhashed and thus removed from ->thread_group.
1868 * This is OK, __unhash_process()->list_del_rcu() does not
1869 * clear the ->next pointer, we will find the new leader via
1870 * next_thread().
1872 rcu_read_lock();
1873 for_each_process(g) {
1874 if (g == tsk->group_leader)
1875 continue;
1876 if (g->flags & PF_KTHREAD)
1877 continue;
1878 p = g;
1879 do {
1880 if (p->mm) {
1881 if (unlikely(p->mm == mm)) {
1882 lock_task_sighand(p, &flags);
1883 nr += zap_process(p, exit_code);
1884 unlock_task_sighand(p, &flags);
1886 break;
1888 } while_each_thread(g, p);
1890 rcu_read_unlock();
1891 done:
1892 atomic_set(&core_state->nr_threads, nr);
1893 return nr;
1896 static int coredump_wait(int exit_code, struct core_state *core_state)
1898 struct task_struct *tsk = current;
1899 struct mm_struct *mm = tsk->mm;
1900 struct completion *vfork_done;
1901 int core_waiters = -EBUSY;
1903 init_completion(&core_state->startup);
1904 core_state->dumper.task = tsk;
1905 core_state->dumper.next = NULL;
1907 down_write(&mm->mmap_sem);
1908 if (!mm->core_state)
1909 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1910 up_write(&mm->mmap_sem);
1912 if (unlikely(core_waiters < 0))
1913 goto fail;
1916 * Make sure nobody is waiting for us to release the VM,
1917 * otherwise we can deadlock when we wait on each other
1919 vfork_done = tsk->vfork_done;
1920 if (vfork_done) {
1921 tsk->vfork_done = NULL;
1922 complete(vfork_done);
1925 if (core_waiters)
1926 wait_for_completion(&core_state->startup);
1927 fail:
1928 return core_waiters;
1931 static void coredump_finish(struct mm_struct *mm)
1933 struct core_thread *curr, *next;
1934 struct task_struct *task;
1936 next = mm->core_state->dumper.next;
1937 while ((curr = next) != NULL) {
1938 next = curr->next;
1939 task = curr->task;
1941 * see exit_mm(), curr->task must not see
1942 * ->task == NULL before we read ->next.
1944 smp_mb();
1945 curr->task = NULL;
1946 wake_up_process(task);
1949 mm->core_state = NULL;
1953 * set_dumpable converts traditional three-value dumpable to two flags and
1954 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1955 * these bits are not changed atomically. So get_dumpable can observe the
1956 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1957 * return either old dumpable or new one by paying attention to the order of
1958 * modifying the bits.
1960 * dumpable | mm->flags (binary)
1961 * old new | initial interim final
1962 * ---------+-----------------------
1963 * 0 1 | 00 01 01
1964 * 0 2 | 00 10(*) 11
1965 * 1 0 | 01 00 00
1966 * 1 2 | 01 11 11
1967 * 2 0 | 11 10(*) 00
1968 * 2 1 | 11 11 01
1970 * (*) get_dumpable regards interim value of 10 as 11.
1972 void set_dumpable(struct mm_struct *mm, int value)
1974 switch (value) {
1975 case 0:
1976 clear_bit(MMF_DUMPABLE, &mm->flags);
1977 smp_wmb();
1978 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1979 break;
1980 case 1:
1981 set_bit(MMF_DUMPABLE, &mm->flags);
1982 smp_wmb();
1983 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1984 break;
1985 case 2:
1986 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1987 smp_wmb();
1988 set_bit(MMF_DUMPABLE, &mm->flags);
1989 break;
1993 static int __get_dumpable(unsigned long mm_flags)
1995 int ret;
1997 ret = mm_flags & MMF_DUMPABLE_MASK;
1998 return (ret >= 2) ? 2 : ret;
2001 int get_dumpable(struct mm_struct *mm)
2003 return __get_dumpable(mm->flags);
2006 static void wait_for_dump_helpers(struct file *file)
2008 struct pipe_inode_info *pipe;
2010 pipe = file->f_path.dentry->d_inode->i_pipe;
2012 pipe_lock(pipe);
2013 pipe->readers++;
2014 pipe->writers--;
2016 while ((pipe->readers > 1) && (!signal_pending(current))) {
2017 wake_up_interruptible_sync(&pipe->wait);
2018 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2019 pipe_wait(pipe);
2022 pipe->readers--;
2023 pipe->writers++;
2024 pipe_unlock(pipe);
2030 * umh_pipe_setup
2031 * helper function to customize the process used
2032 * to collect the core in userspace. Specifically
2033 * it sets up a pipe and installs it as fd 0 (stdin)
2034 * for the process. Returns 0 on success, or
2035 * PTR_ERR on failure.
2036 * Note that it also sets the core limit to 1. This
2037 * is a special value that we use to trap recursive
2038 * core dumps
2040 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2042 struct file *rp, *wp;
2043 struct fdtable *fdt;
2044 struct coredump_params *cp = (struct coredump_params *)info->data;
2045 struct files_struct *cf = current->files;
2047 wp = create_write_pipe(0);
2048 if (IS_ERR(wp))
2049 return PTR_ERR(wp);
2051 rp = create_read_pipe(wp, 0);
2052 if (IS_ERR(rp)) {
2053 free_write_pipe(wp);
2054 return PTR_ERR(rp);
2057 cp->file = wp;
2059 sys_close(0);
2060 fd_install(0, rp);
2061 spin_lock(&cf->file_lock);
2062 fdt = files_fdtable(cf);
2063 FD_SET(0, fdt->open_fds);
2064 FD_CLR(0, fdt->close_on_exec);
2065 spin_unlock(&cf->file_lock);
2067 /* and disallow core files too */
2068 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2070 return 0;
2073 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2075 struct core_state core_state;
2076 struct core_name cn;
2077 struct mm_struct *mm = current->mm;
2078 struct linux_binfmt * binfmt;
2079 const struct cred *old_cred;
2080 struct cred *cred;
2081 int retval = 0;
2082 int flag = 0;
2083 int ispipe;
2084 static atomic_t core_dump_count = ATOMIC_INIT(0);
2085 struct coredump_params cprm = {
2086 .signr = signr,
2087 .regs = regs,
2088 .limit = rlimit(RLIMIT_CORE),
2090 * We must use the same mm->flags while dumping core to avoid
2091 * inconsistency of bit flags, since this flag is not protected
2092 * by any locks.
2094 .mm_flags = mm->flags,
2097 audit_core_dumps(signr);
2099 binfmt = mm->binfmt;
2100 if (!binfmt || !binfmt->core_dump)
2101 goto fail;
2102 if (!__get_dumpable(cprm.mm_flags))
2103 goto fail;
2105 cred = prepare_creds();
2106 if (!cred)
2107 goto fail;
2109 * We cannot trust fsuid as being the "true" uid of the
2110 * process nor do we know its entire history. We only know it
2111 * was tainted so we dump it as root in mode 2.
2113 if (__get_dumpable(cprm.mm_flags) == 2) {
2114 /* Setuid core dump mode */
2115 flag = O_EXCL; /* Stop rewrite attacks */
2116 cred->fsuid = 0; /* Dump root private */
2119 retval = coredump_wait(exit_code, &core_state);
2120 if (retval < 0)
2121 goto fail_creds;
2123 old_cred = override_creds(cred);
2126 * Clear any false indication of pending signals that might
2127 * be seen by the filesystem code called to write the core file.
2129 clear_thread_flag(TIF_SIGPENDING);
2131 ispipe = format_corename(&cn, signr);
2133 if (ispipe) {
2134 int dump_count;
2135 char **helper_argv;
2137 if (ispipe < 0) {
2138 printk(KERN_WARNING "format_corename failed\n");
2139 printk(KERN_WARNING "Aborting core\n");
2140 goto fail_corename;
2143 if (cprm.limit == 1) {
2145 * Normally core limits are irrelevant to pipes, since
2146 * we're not writing to the file system, but we use
2147 * cprm.limit of 1 here as a speacial value. Any
2148 * non-1 limit gets set to RLIM_INFINITY below, but
2149 * a limit of 0 skips the dump. This is a consistent
2150 * way to catch recursive crashes. We can still crash
2151 * if the core_pattern binary sets RLIM_CORE = !1
2152 * but it runs as root, and can do lots of stupid things
2153 * Note that we use task_tgid_vnr here to grab the pid
2154 * of the process group leader. That way we get the
2155 * right pid if a thread in a multi-threaded
2156 * core_pattern process dies.
2158 printk(KERN_WARNING
2159 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2160 task_tgid_vnr(current), current->comm);
2161 printk(KERN_WARNING "Aborting core\n");
2162 goto fail_unlock;
2164 cprm.limit = RLIM_INFINITY;
2166 dump_count = atomic_inc_return(&core_dump_count);
2167 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2168 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2169 task_tgid_vnr(current), current->comm);
2170 printk(KERN_WARNING "Skipping core dump\n");
2171 goto fail_dropcount;
2174 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2175 if (!helper_argv) {
2176 printk(KERN_WARNING "%s failed to allocate memory\n",
2177 __func__);
2178 goto fail_dropcount;
2181 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2182 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2183 NULL, &cprm);
2184 argv_free(helper_argv);
2185 if (retval) {
2186 printk(KERN_INFO "Core dump to %s pipe failed\n",
2187 cn.corename);
2188 goto close_fail;
2190 } else {
2191 struct inode *inode;
2193 if (cprm.limit < binfmt->min_coredump)
2194 goto fail_unlock;
2196 cprm.file = filp_open(cn.corename,
2197 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2198 0600);
2199 if (IS_ERR(cprm.file))
2200 goto fail_unlock;
2202 inode = cprm.file->f_path.dentry->d_inode;
2203 if (inode->i_nlink > 1)
2204 goto close_fail;
2205 if (d_unhashed(cprm.file->f_path.dentry))
2206 goto close_fail;
2208 * AK: actually i see no reason to not allow this for named
2209 * pipes etc, but keep the previous behaviour for now.
2211 if (!S_ISREG(inode->i_mode))
2212 goto close_fail;
2214 * Dont allow local users get cute and trick others to coredump
2215 * into their pre-created files.
2217 if (inode->i_uid != current_fsuid())
2218 goto close_fail;
2219 if (!cprm.file->f_op || !cprm.file->f_op->write)
2220 goto close_fail;
2221 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2222 goto close_fail;
2225 retval = binfmt->core_dump(&cprm);
2226 if (retval)
2227 current->signal->group_exit_code |= 0x80;
2229 if (ispipe && core_pipe_limit)
2230 wait_for_dump_helpers(cprm.file);
2231 close_fail:
2232 if (cprm.file)
2233 filp_close(cprm.file, NULL);
2234 fail_dropcount:
2235 if (ispipe)
2236 atomic_dec(&core_dump_count);
2237 fail_unlock:
2238 kfree(cn.corename);
2239 fail_corename:
2240 coredump_finish(mm);
2241 revert_creds(old_cred);
2242 fail_creds:
2243 put_cred(cred);
2244 fail:
2245 return;
2249 * Core dumping helper functions. These are the only things you should
2250 * do on a core-file: use only these functions to write out all the
2251 * necessary info.
2253 int dump_write(struct file *file, const void *addr, int nr)
2255 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2257 EXPORT_SYMBOL(dump_write);
2259 int dump_seek(struct file *file, loff_t off)
2261 int ret = 1;
2263 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2264 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2265 return 0;
2266 } else {
2267 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2269 if (!buf)
2270 return 0;
2271 while (off > 0) {
2272 unsigned long n = off;
2274 if (n > PAGE_SIZE)
2275 n = PAGE_SIZE;
2276 if (!dump_write(file, buf, n)) {
2277 ret = 0;
2278 break;
2280 off -= n;
2282 free_page((unsigned long)buf);
2284 return ret;
2286 EXPORT_SYMBOL(dump_seek);