4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * #!-checking implemented by tytso.
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
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/freezer.h>
30 #include <linux/stat.h>
31 #include <linux/fcntl.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/perf_event.h>
37 #include <linux/highmem.h>
38 #include <linux/spinlock.h>
39 #include <linux/key.h>
40 #include <linux/personality.h>
41 #include <linux/binfmts.h>
42 #include <linux/utsname.h>
43 #include <linux/pid_namespace.h>
44 #include <linux/module.h>
45 #include <linux/namei.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
53 #include <linux/kmod.h>
54 #include <linux/fsnotify.h>
55 #include <linux/fs_struct.h>
56 #include <linux/pipe_fs_i.h>
57 #include <linux/oom.h>
58 #include <linux/compat.h>
59 #include <linux/sched.h>
61 #include <linux/path.h>
63 #include <asm/uaccess.h>
64 #include <asm/mmu_context.h>
69 char core_pattern[CORENAME_MAX_SIZE] = "core";
70 unsigned int core_pipe_limit;
71 int suid_dumpable = 0;
77 static atomic_t call_count = ATOMIC_INIT(1);
79 /* The maximal length of core_pattern is also specified in sysctl.c */
81 static LIST_HEAD(formats);
82 static DEFINE_RWLOCK(binfmt_lock);
84 int __register_binfmt(struct linux_binfmt * fmt, int insert)
88 write_lock(&binfmt_lock);
89 insert ? list_add(&fmt->lh, &formats) :
90 list_add_tail(&fmt->lh, &formats);
91 write_unlock(&binfmt_lock);
95 EXPORT_SYMBOL(__register_binfmt);
97 void unregister_binfmt(struct linux_binfmt * fmt)
99 write_lock(&binfmt_lock);
101 write_unlock(&binfmt_lock);
104 EXPORT_SYMBOL(unregister_binfmt);
106 static inline void put_binfmt(struct linux_binfmt * fmt)
108 module_put(fmt->module);
112 * Note that a shared library must be both readable and executable due to
115 * Also note that we take the address to load from from the file itself.
117 SYSCALL_DEFINE1(uselib, const char __user *, library)
120 char *tmp = getname(library);
121 int error = PTR_ERR(tmp);
122 static const struct open_flags uselib_flags = {
123 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
124 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
125 .intent = LOOKUP_OPEN
131 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
133 error = PTR_ERR(file);
138 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
142 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
149 struct linux_binfmt * fmt;
151 read_lock(&binfmt_lock);
152 list_for_each_entry(fmt, &formats, lh) {
153 if (!fmt->load_shlib)
155 if (!try_module_get(fmt->module))
157 read_unlock(&binfmt_lock);
158 error = fmt->load_shlib(file);
159 read_lock(&binfmt_lock);
161 if (error != -ENOEXEC)
164 read_unlock(&binfmt_lock);
174 * The nascent bprm->mm is not visible until exec_mmap() but it can
175 * use a lot of memory, account these pages in current->mm temporary
176 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
177 * change the counter back via acct_arg_size(0).
179 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
181 struct mm_struct *mm = current->mm;
182 long diff = (long)(pages - bprm->vma_pages);
187 bprm->vma_pages = pages;
188 add_mm_counter(mm, MM_ANONPAGES, diff);
191 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
197 #ifdef CONFIG_STACK_GROWSUP
199 ret = expand_downwards(bprm->vma, pos);
204 ret = get_user_pages(current, bprm->mm, pos,
205 1, write, 1, &page, NULL);
210 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
211 unsigned long ptr_size;
215 * Since the stack will hold pointers to the strings, we
216 * must account for them as well.
218 * The size calculation is the entire vma while each arg page is
219 * built, so each time we get here it's calculating how far it
220 * is currently (rather than each call being just the newly
221 * added size from the arg page). As a result, we need to
222 * always add the entire size of the pointers, so that on the
223 * last call to get_arg_page() we'll actually have the entire
226 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
227 if (ptr_size > ULONG_MAX - size)
231 acct_arg_size(bprm, size / PAGE_SIZE);
234 * We've historically supported up to 32 pages (ARG_MAX)
235 * of argument strings even with small stacks
241 * Limit to 1/4-th the stack size for the argv+env strings.
243 * - the remaining binfmt code will not run out of stack space,
244 * - the program will have a reasonable amount of stack left
247 rlim = current->signal->rlim;
248 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4)
259 static void put_arg_page(struct page *page)
264 static void free_arg_page(struct linux_binprm *bprm, int i)
268 static void free_arg_pages(struct linux_binprm *bprm)
272 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
275 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
278 static int __bprm_mm_init(struct linux_binprm *bprm)
281 struct vm_area_struct *vma = NULL;
282 struct mm_struct *mm = bprm->mm;
284 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
288 down_write(&mm->mmap_sem);
292 * Place the stack at the largest stack address the architecture
293 * supports. Later, we'll move this to an appropriate place. We don't
294 * use STACK_TOP because that can depend on attributes which aren't
297 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
298 vma->vm_end = STACK_TOP_MAX;
299 vma->vm_start = vma->vm_end - PAGE_SIZE;
300 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
301 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
302 INIT_LIST_HEAD(&vma->anon_vma_chain);
304 err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
308 err = insert_vm_struct(mm, vma);
312 mm->stack_vm = mm->total_vm = 1;
313 up_write(&mm->mmap_sem);
314 bprm->p = vma->vm_end - sizeof(void *);
317 up_write(&mm->mmap_sem);
319 kmem_cache_free(vm_area_cachep, vma);
323 static bool valid_arg_len(struct linux_binprm *bprm, long len)
325 return len <= MAX_ARG_STRLEN;
330 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
334 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
339 page = bprm->page[pos / PAGE_SIZE];
340 if (!page && write) {
341 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
344 bprm->page[pos / PAGE_SIZE] = page;
350 static void put_arg_page(struct page *page)
354 static void free_arg_page(struct linux_binprm *bprm, int i)
357 __free_page(bprm->page[i]);
358 bprm->page[i] = NULL;
362 static void free_arg_pages(struct linux_binprm *bprm)
366 for (i = 0; i < MAX_ARG_PAGES; i++)
367 free_arg_page(bprm, i);
370 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
375 static int __bprm_mm_init(struct linux_binprm *bprm)
377 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
381 static bool valid_arg_len(struct linux_binprm *bprm, long len)
383 return len <= bprm->p;
386 #endif /* CONFIG_MMU */
389 * Create a new mm_struct and populate it with a temporary stack
390 * vm_area_struct. We don't have enough context at this point to set the stack
391 * flags, permissions, and offset, so we use temporary values. We'll update
392 * them later in setup_arg_pages().
394 int bprm_mm_init(struct linux_binprm *bprm)
397 struct mm_struct *mm = NULL;
399 bprm->mm = mm = mm_alloc();
404 err = init_new_context(current, mm);
408 err = __bprm_mm_init(bprm);
423 struct user_arg_ptr {
428 const char __user *const __user *native;
430 compat_uptr_t __user *compat;
435 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
437 const char __user *native;
440 if (unlikely(argv.is_compat)) {
441 compat_uptr_t compat;
443 if (get_user(compat, argv.ptr.compat + nr))
444 return ERR_PTR(-EFAULT);
446 return compat_ptr(compat);
450 if (get_user(native, argv.ptr.native + nr))
451 return ERR_PTR(-EFAULT);
457 * count() counts the number of strings in array ARGV.
459 static int count(struct user_arg_ptr argv, int max)
463 if (argv.ptr.native != NULL) {
465 const char __user *p = get_user_arg_ptr(argv, i);
476 if (fatal_signal_pending(current))
477 return -ERESTARTNOHAND;
485 * 'copy_strings()' copies argument/environment strings from the old
486 * processes's memory to the new process's stack. The call to get_user_pages()
487 * ensures the destination page is created and not swapped out.
489 static int copy_strings(int argc, struct user_arg_ptr argv,
490 struct linux_binprm *bprm)
492 struct page *kmapped_page = NULL;
494 unsigned long kpos = 0;
498 const char __user *str;
503 str = get_user_arg_ptr(argv, argc);
507 len = strnlen_user(str, MAX_ARG_STRLEN);
512 if (!valid_arg_len(bprm, len))
515 /* We're going to work our way backwords. */
521 int offset, bytes_to_copy;
523 if (fatal_signal_pending(current)) {
524 ret = -ERESTARTNOHAND;
529 offset = pos % PAGE_SIZE;
533 bytes_to_copy = offset;
534 if (bytes_to_copy > len)
537 offset -= bytes_to_copy;
538 pos -= bytes_to_copy;
539 str -= bytes_to_copy;
540 len -= bytes_to_copy;
542 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
545 page = get_arg_page(bprm, pos, 1);
552 flush_kernel_dcache_page(kmapped_page);
553 kunmap(kmapped_page);
554 put_arg_page(kmapped_page);
557 kaddr = kmap(kmapped_page);
558 kpos = pos & PAGE_MASK;
559 flush_arg_page(bprm, kpos, kmapped_page);
561 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
570 flush_kernel_dcache_page(kmapped_page);
571 kunmap(kmapped_page);
572 put_arg_page(kmapped_page);
578 * Like copy_strings, but get argv and its values from kernel memory.
580 int copy_strings_kernel(int argc, const char *const *__argv,
581 struct linux_binprm *bprm)
584 mm_segment_t oldfs = get_fs();
585 struct user_arg_ptr argv = {
586 .ptr.native = (const char __user *const __user *)__argv,
590 r = copy_strings(argc, argv, bprm);
595 EXPORT_SYMBOL(copy_strings_kernel);
600 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
601 * the binfmt code determines where the new stack should reside, we shift it to
602 * its final location. The process proceeds as follows:
604 * 1) Use shift to calculate the new vma endpoints.
605 * 2) Extend vma to cover both the old and new ranges. This ensures the
606 * arguments passed to subsequent functions are consistent.
607 * 3) Move vma's page tables to the new range.
608 * 4) Free up any cleared pgd range.
609 * 5) Shrink the vma to cover only the new range.
611 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
613 struct mm_struct *mm = vma->vm_mm;
614 unsigned long old_start = vma->vm_start;
615 unsigned long old_end = vma->vm_end;
616 unsigned long length = old_end - old_start;
617 unsigned long new_start = old_start - shift;
618 unsigned long new_end = old_end - shift;
619 struct mmu_gather tlb;
621 BUG_ON(new_start > new_end);
624 * ensure there are no vmas between where we want to go
627 if (vma != find_vma(mm, new_start))
631 * cover the whole range: [new_start, old_end)
633 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
637 * move the page tables downwards, on failure we rely on
638 * process cleanup to remove whatever mess we made.
640 if (length != move_page_tables(vma, old_start,
641 vma, new_start, length))
645 tlb_gather_mmu(&tlb, mm, 0);
646 if (new_end > old_start) {
648 * when the old and new regions overlap clear from new_end.
650 free_pgd_range(&tlb, new_end, old_end, new_end,
651 vma->vm_next ? vma->vm_next->vm_start : 0);
654 * otherwise, clean from old_start; this is done to not touch
655 * the address space in [new_end, old_start) some architectures
656 * have constraints on va-space that make this illegal (IA64) -
657 * for the others its just a little faster.
659 free_pgd_range(&tlb, old_start, old_end, new_end,
660 vma->vm_next ? vma->vm_next->vm_start : 0);
662 tlb_finish_mmu(&tlb, new_end, old_end);
665 * Shrink the vma to just the new range. Always succeeds.
667 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
673 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
674 * the stack is optionally relocated, and some extra space is added.
676 int setup_arg_pages(struct linux_binprm *bprm,
677 unsigned long stack_top,
678 int executable_stack)
681 unsigned long stack_shift;
682 struct mm_struct *mm = current->mm;
683 struct vm_area_struct *vma = bprm->vma;
684 struct vm_area_struct *prev = NULL;
685 unsigned long vm_flags;
686 unsigned long stack_base;
687 unsigned long stack_size;
688 unsigned long stack_expand;
689 unsigned long rlim_stack;
691 #ifdef CONFIG_STACK_GROWSUP
692 /* Limit stack size to 1GB */
693 stack_base = rlimit_max(RLIMIT_STACK);
694 if (stack_base > (1 << 30))
695 stack_base = 1 << 30;
697 /* Make sure we didn't let the argument array grow too large. */
698 if (vma->vm_end - vma->vm_start > stack_base)
701 stack_base = PAGE_ALIGN(stack_top - stack_base);
703 stack_shift = vma->vm_start - stack_base;
704 mm->arg_start = bprm->p - stack_shift;
705 bprm->p = vma->vm_end - stack_shift;
707 stack_top = arch_align_stack(stack_top);
708 stack_top = PAGE_ALIGN(stack_top);
710 if (unlikely(stack_top < mmap_min_addr) ||
711 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
714 stack_shift = vma->vm_end - stack_top;
716 bprm->p -= stack_shift;
717 mm->arg_start = bprm->p;
721 bprm->loader -= stack_shift;
722 bprm->exec -= stack_shift;
724 down_write(&mm->mmap_sem);
725 vm_flags = VM_STACK_FLAGS;
728 * Adjust stack execute permissions; explicitly enable for
729 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
730 * (arch default) otherwise.
732 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
734 else if (executable_stack == EXSTACK_DISABLE_X)
735 vm_flags &= ~VM_EXEC;
736 vm_flags |= mm->def_flags;
737 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
739 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
745 /* Move stack pages down in memory. */
747 ret = shift_arg_pages(vma, stack_shift);
752 /* mprotect_fixup is overkill to remove the temporary stack flags */
753 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
755 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
756 stack_size = vma->vm_end - vma->vm_start;
758 * Align this down to a page boundary as expand_stack
761 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
762 #ifdef CONFIG_STACK_GROWSUP
763 if (stack_size + stack_expand > rlim_stack)
764 stack_base = vma->vm_start + rlim_stack;
766 stack_base = vma->vm_end + stack_expand;
768 if (stack_size + stack_expand > rlim_stack)
769 stack_base = vma->vm_end - rlim_stack;
771 stack_base = vma->vm_start - stack_expand;
773 current->mm->start_stack = bprm->p;
774 ret = expand_stack(vma, stack_base);
779 up_write(&mm->mmap_sem);
782 EXPORT_SYMBOL(setup_arg_pages);
784 #endif /* CONFIG_MMU */
786 struct file *open_exec(const char *name)
790 static const struct open_flags open_exec_flags = {
791 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
792 .acc_mode = MAY_EXEC | MAY_OPEN,
793 .intent = LOOKUP_OPEN
796 file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
801 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
804 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
809 err = deny_write_access(file);
820 EXPORT_SYMBOL(open_exec);
822 int kernel_read(struct file *file, loff_t offset,
823 char *addr, unsigned long count)
831 /* The cast to a user pointer is valid due to the set_fs() */
832 result = vfs_read(file, (void __user *)addr, count, &pos);
837 EXPORT_SYMBOL(kernel_read);
839 static int exec_mmap(struct mm_struct *mm)
841 struct task_struct *tsk;
842 struct mm_struct * old_mm, *active_mm;
844 /* Notify parent that we're no longer interested in the old VM */
846 old_mm = current->mm;
847 sync_mm_rss(tsk, old_mm);
848 mm_release(tsk, old_mm);
852 * Make sure that if there is a core dump in progress
853 * for the old mm, we get out and die instead of going
854 * through with the exec. We must hold mmap_sem around
855 * checking core_state and changing tsk->mm.
857 down_read(&old_mm->mmap_sem);
858 if (unlikely(old_mm->core_state)) {
859 up_read(&old_mm->mmap_sem);
864 active_mm = tsk->active_mm;
867 activate_mm(active_mm, mm);
869 arch_pick_mmap_layout(mm);
871 up_read(&old_mm->mmap_sem);
872 BUG_ON(active_mm != old_mm);
873 mm_update_next_owner(old_mm);
882 * This function makes sure the current process has its own signal table,
883 * so that flush_signal_handlers can later reset the handlers without
884 * disturbing other processes. (Other processes might share the signal
885 * table via the CLONE_SIGHAND option to clone().)
887 static int de_thread(struct task_struct *tsk)
889 struct signal_struct *sig = tsk->signal;
890 struct sighand_struct *oldsighand = tsk->sighand;
891 spinlock_t *lock = &oldsighand->siglock;
893 if (thread_group_empty(tsk))
894 goto no_thread_group;
897 * Kill all other threads in the thread group.
900 if (signal_group_exit(sig)) {
902 * Another group action in progress, just
903 * return so that the signal is processed.
905 spin_unlock_irq(lock);
909 sig->group_exit_task = tsk;
910 sig->notify_count = zap_other_threads(tsk);
911 if (!thread_group_leader(tsk))
914 while (sig->notify_count) {
915 __set_current_state(TASK_UNINTERRUPTIBLE);
916 spin_unlock_irq(lock);
920 spin_unlock_irq(lock);
923 * At this point all other threads have exited, all we have to
924 * do is to wait for the thread group leader to become inactive,
925 * and to assume its PID:
927 if (!thread_group_leader(tsk)) {
928 struct task_struct *leader = tsk->group_leader;
930 sig->notify_count = -1; /* for exit_notify() */
932 write_lock_irq(&tasklist_lock);
933 if (likely(leader->exit_state))
935 __set_current_state(TASK_UNINTERRUPTIBLE);
936 write_unlock_irq(&tasklist_lock);
941 * The only record we have of the real-time age of a
942 * process, regardless of execs it's done, is start_time.
943 * All the past CPU time is accumulated in signal_struct
944 * from sister threads now dead. But in this non-leader
945 * exec, nothing survives from the original leader thread,
946 * whose birth marks the true age of this process now.
947 * When we take on its identity by switching to its PID, we
948 * also take its birthdate (always earlier than our own).
950 tsk->start_time = leader->start_time;
952 BUG_ON(!same_thread_group(leader, tsk));
953 BUG_ON(has_group_leader_pid(tsk));
955 * An exec() starts a new thread group with the
956 * TGID of the previous thread group. Rehash the
957 * two threads with a switched PID, and release
958 * the former thread group leader:
961 /* Become a process group leader with the old leader's pid.
962 * The old leader becomes a thread of the this thread group.
963 * Note: The old leader also uses this pid until release_task
964 * is called. Odd but simple and correct.
966 detach_pid(tsk, PIDTYPE_PID);
967 tsk->pid = leader->pid;
968 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
969 transfer_pid(leader, tsk, PIDTYPE_PGID);
970 transfer_pid(leader, tsk, PIDTYPE_SID);
972 list_replace_rcu(&leader->tasks, &tsk->tasks);
973 list_replace_init(&leader->sibling, &tsk->sibling);
975 tsk->group_leader = tsk;
976 leader->group_leader = tsk;
978 tsk->exit_signal = SIGCHLD;
979 leader->exit_signal = -1;
981 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
982 leader->exit_state = EXIT_DEAD;
985 * We are going to release_task()->ptrace_unlink() silently,
986 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
987 * the tracer wont't block again waiting for this thread.
989 if (unlikely(leader->ptrace))
990 __wake_up_parent(leader, leader->parent);
991 write_unlock_irq(&tasklist_lock);
993 release_task(leader);
996 sig->group_exit_task = NULL;
997 sig->notify_count = 0;
1000 /* we have changed execution domain */
1001 tsk->exit_signal = SIGCHLD;
1004 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
1007 flush_itimer_signals();
1009 if (atomic_read(&oldsighand->count) != 1) {
1010 struct sighand_struct *newsighand;
1012 * This ->sighand is shared with the CLONE_SIGHAND
1013 * but not CLONE_THREAD task, switch to the new one.
1015 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1019 atomic_set(&newsighand->count, 1);
1020 memcpy(newsighand->action, oldsighand->action,
1021 sizeof(newsighand->action));
1023 write_lock_irq(&tasklist_lock);
1024 spin_lock(&oldsighand->siglock);
1025 rcu_assign_pointer(tsk->sighand, newsighand);
1026 spin_unlock(&oldsighand->siglock);
1027 write_unlock_irq(&tasklist_lock);
1029 __cleanup_sighand(oldsighand);
1032 BUG_ON(!thread_group_leader(tsk));
1037 * These functions flushes out all traces of the currently running executable
1038 * so that a new one can be started
1040 static void flush_old_files(struct files_struct * files)
1043 struct fdtable *fdt;
1045 spin_lock(&files->file_lock);
1047 unsigned long set, i;
1051 fdt = files_fdtable(files);
1052 if (i >= fdt->max_fds)
1054 set = fdt->close_on_exec->fds_bits[j];
1057 fdt->close_on_exec->fds_bits[j] = 0;
1058 spin_unlock(&files->file_lock);
1059 for ( ; set ; i++,set >>= 1) {
1064 spin_lock(&files->file_lock);
1067 spin_unlock(&files->file_lock);
1070 char *get_task_comm(char *buf, struct task_struct *tsk)
1072 /* buf must be at least sizeof(tsk->comm) in size */
1074 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1078 EXPORT_SYMBOL_GPL(get_task_comm);
1080 void set_task_comm(struct task_struct *tsk, char *buf)
1085 * Threads may access current->comm without holding
1086 * the task lock, so write the string carefully.
1087 * Readers without a lock may see incomplete new
1088 * names but are safe from non-terminating string reads.
1090 memset(tsk->comm, 0, TASK_COMM_LEN);
1092 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1094 perf_event_comm(tsk);
1097 int flush_old_exec(struct linux_binprm * bprm)
1102 * Make sure we have a private signal table and that
1103 * we are unassociated from the previous thread group.
1105 retval = de_thread(current);
1109 set_mm_exe_file(bprm->mm, bprm->file);
1112 * Release all of the old mmap stuff
1114 acct_arg_size(bprm, 0);
1115 retval = exec_mmap(bprm->mm);
1119 bprm->mm = NULL; /* We're using it now */
1123 ~(PF_RANDOMIZE | PF_KTHREAD | PF_NOFREEZE | PF_FREEZER_NOSIG);
1125 current->personality &= ~bprm->per_clear;
1132 EXPORT_SYMBOL(flush_old_exec);
1134 void would_dump(struct linux_binprm *bprm, struct file *file)
1136 if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1137 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1139 EXPORT_SYMBOL(would_dump);
1141 void setup_new_exec(struct linux_binprm * bprm)
1145 char tcomm[sizeof(current->comm)];
1147 arch_pick_mmap_layout(current->mm);
1149 /* This is the point of no return */
1150 current->sas_ss_sp = current->sas_ss_size = 0;
1152 if (current_euid() == current_uid() && current_egid() == current_gid())
1153 set_dumpable(current->mm, 1);
1155 set_dumpable(current->mm, suid_dumpable);
1157 name = bprm->filename;
1159 /* Copies the binary name from after last slash */
1160 for (i=0; (ch = *(name++)) != '\0';) {
1162 i = 0; /* overwrite what we wrote */
1164 if (i < (sizeof(tcomm) - 1))
1168 set_task_comm(current, tcomm);
1170 /* Set the new mm task size. We have to do that late because it may
1171 * depend on TIF_32BIT which is only updated in flush_thread() on
1172 * some architectures like powerpc
1174 current->mm->task_size = TASK_SIZE;
1176 /* install the new credentials */
1177 if (bprm->cred->uid != current_euid() ||
1178 bprm->cred->gid != current_egid()) {
1179 current->pdeath_signal = 0;
1181 would_dump(bprm, bprm->file);
1182 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1183 set_dumpable(current->mm, suid_dumpable);
1186 /* An exec changes our domain. We are no longer part of the thread
1189 current->self_exec_id++;
1191 flush_signal_handlers(current, 0);
1192 flush_old_files(current->files);
1194 EXPORT_SYMBOL(setup_new_exec);
1197 * Prepare credentials and lock ->cred_guard_mutex.
1198 * install_exec_creds() commits the new creds and drops the lock.
1199 * Or, if exec fails before, free_bprm() should release ->cred and
1202 int prepare_bprm_creds(struct linux_binprm *bprm)
1204 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1205 return -ERESTARTNOINTR;
1207 bprm->cred = prepare_exec_creds();
1208 if (likely(bprm->cred))
1211 mutex_unlock(¤t->signal->cred_guard_mutex);
1215 void free_bprm(struct linux_binprm *bprm)
1217 free_arg_pages(bprm);
1219 mutex_unlock(¤t->signal->cred_guard_mutex);
1220 abort_creds(bprm->cred);
1222 /* If a binfmt changed the interp, free it. */
1223 if (bprm->interp != bprm->filename)
1224 kfree(bprm->interp);
1228 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1230 /* If a binfmt changed the interp, free it first. */
1231 if (bprm->interp != bprm->filename)
1232 kfree(bprm->interp);
1233 bprm->interp = kstrdup(interp, GFP_KERNEL);
1238 EXPORT_SYMBOL(bprm_change_interp);
1241 * install the new credentials for this executable
1243 void install_exec_creds(struct linux_binprm *bprm)
1245 security_bprm_committing_creds(bprm);
1247 commit_creds(bprm->cred);
1251 * Disable monitoring for regular users
1252 * when executing setuid binaries. Must
1253 * wait until new credentials are committed
1254 * by commit_creds() above
1256 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1257 perf_event_exit_task(current);
1259 * cred_guard_mutex must be held at least to this point to prevent
1260 * ptrace_attach() from altering our determination of the task's
1261 * credentials; any time after this it may be unlocked.
1263 security_bprm_committed_creds(bprm);
1264 mutex_unlock(¤t->signal->cred_guard_mutex);
1266 EXPORT_SYMBOL(install_exec_creds);
1269 * determine how safe it is to execute the proposed program
1270 * - the caller must hold ->cred_guard_mutex to protect against
1273 int check_unsafe_exec(struct linux_binprm *bprm)
1275 struct task_struct *p = current, *t;
1280 if (p->ptrace & PT_PTRACE_CAP)
1281 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1283 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1287 spin_lock(&p->fs->lock);
1289 for (t = next_thread(p); t != p; t = next_thread(t)) {
1295 if (p->fs->users > n_fs) {
1296 bprm->unsafe |= LSM_UNSAFE_SHARE;
1299 if (!p->fs->in_exec) {
1304 spin_unlock(&p->fs->lock);
1309 static void bprm_fill_uid(struct linux_binprm *bprm)
1311 struct inode *inode;
1316 /* clear any previous set[ug]id data from a previous binary */
1317 bprm->cred->euid = current_euid();
1318 bprm->cred->egid = current_egid();
1320 if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1323 inode = bprm->file->f_path.dentry->d_inode;
1324 mode = ACCESS_ONCE(inode->i_mode);
1325 if (!(mode & (S_ISUID|S_ISGID)))
1328 /* Be careful if suid/sgid is set */
1329 mutex_lock(&inode->i_mutex);
1331 /* reload atomically mode/uid/gid now that lock held */
1332 mode = inode->i_mode;
1335 mutex_unlock(&inode->i_mutex);
1337 if (mode & S_ISUID) {
1338 bprm->per_clear |= PER_CLEAR_ON_SETID;
1339 bprm->cred->euid = uid;
1342 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1343 bprm->per_clear |= PER_CLEAR_ON_SETID;
1344 bprm->cred->egid = gid;
1349 * Fill the binprm structure from the inode.
1350 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1352 * This may be called multiple times for binary chains (scripts for example).
1354 int prepare_binprm(struct linux_binprm *bprm)
1358 if (bprm->file->f_op == NULL)
1361 bprm_fill_uid(bprm);
1363 /* fill in binprm security blob */
1364 retval = security_bprm_set_creds(bprm);
1367 bprm->cred_prepared = 1;
1369 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1370 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1373 EXPORT_SYMBOL(prepare_binprm);
1376 * Arguments are '\0' separated strings found at the location bprm->p
1377 * points to; chop off the first by relocating brpm->p to right after
1378 * the first '\0' encountered.
1380 int remove_arg_zero(struct linux_binprm *bprm)
1383 unsigned long offset;
1391 offset = bprm->p & ~PAGE_MASK;
1392 page = get_arg_page(bprm, bprm->p, 0);
1397 kaddr = kmap_atomic(page, KM_USER0);
1399 for (; offset < PAGE_SIZE && kaddr[offset];
1400 offset++, bprm->p++)
1403 kunmap_atomic(kaddr, KM_USER0);
1406 if (offset == PAGE_SIZE)
1407 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1408 } while (offset == PAGE_SIZE);
1417 EXPORT_SYMBOL(remove_arg_zero);
1420 * cycle the list of binary formats handler, until one recognizes the image
1422 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1424 unsigned int depth = bprm->recursion_depth;
1426 struct linux_binfmt *fmt;
1429 /* This allows 4 levels of binfmt rewrites before failing hard. */
1433 retval = security_bprm_check(bprm);
1437 retval = audit_bprm(bprm);
1441 /* Need to fetch pid before load_binary changes it */
1443 old_pid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1447 for (try=0; try<2; try++) {
1448 read_lock(&binfmt_lock);
1449 list_for_each_entry(fmt, &formats, lh) {
1450 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1453 if (!try_module_get(fmt->module))
1455 read_unlock(&binfmt_lock);
1456 bprm->recursion_depth = depth + 1;
1457 retval = fn(bprm, regs);
1458 bprm->recursion_depth = depth;
1461 ptrace_event(PTRACE_EVENT_EXEC,
1464 allow_write_access(bprm->file);
1468 current->did_exec = 1;
1469 proc_exec_connector(current);
1472 read_lock(&binfmt_lock);
1474 if (retval != -ENOEXEC || bprm->mm == NULL)
1477 read_unlock(&binfmt_lock);
1481 read_unlock(&binfmt_lock);
1482 #ifdef CONFIG_MODULES
1483 if (retval != -ENOEXEC || bprm->mm == NULL) {
1486 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1487 if (printable(bprm->buf[0]) &&
1488 printable(bprm->buf[1]) &&
1489 printable(bprm->buf[2]) &&
1490 printable(bprm->buf[3]))
1491 break; /* -ENOEXEC */
1493 break; /* -ENOEXEC */
1494 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1503 EXPORT_SYMBOL(search_binary_handler);
1506 * sys_execve() executes a new program.
1508 static int do_execve_common(const char *filename,
1509 struct user_arg_ptr argv,
1510 struct user_arg_ptr envp,
1511 struct pt_regs *regs)
1513 struct linux_binprm *bprm;
1515 struct files_struct *displaced;
1518 const struct cred *cred = current_cred();
1521 * We move the actual failure in case of RLIMIT_NPROC excess from
1522 * set*uid() to execve() because too many poorly written programs
1523 * don't check setuid() return code. Here we additionally recheck
1524 * whether NPROC limit is still exceeded.
1526 if ((current->flags & PF_NPROC_EXCEEDED) &&
1527 atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1532 /* We're below the limit (still or again), so we don't want to make
1533 * further execve() calls fail. */
1534 current->flags &= ~PF_NPROC_EXCEEDED;
1536 retval = unshare_files(&displaced);
1541 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1545 retval = prepare_bprm_creds(bprm);
1549 retval = check_unsafe_exec(bprm);
1552 clear_in_exec = retval;
1553 current->in_execve = 1;
1555 file = open_exec(filename);
1556 retval = PTR_ERR(file);
1563 bprm->filename = filename;
1564 bprm->interp = filename;
1566 retval = bprm_mm_init(bprm);
1570 bprm->argc = count(argv, MAX_ARG_STRINGS);
1571 if ((retval = bprm->argc) < 0)
1574 bprm->envc = count(envp, MAX_ARG_STRINGS);
1575 if ((retval = bprm->envc) < 0)
1578 retval = prepare_binprm(bprm);
1582 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1586 bprm->exec = bprm->p;
1587 retval = copy_strings(bprm->envc, envp, bprm);
1591 retval = copy_strings(bprm->argc, argv, bprm);
1595 retval = search_binary_handler(bprm,regs);
1599 /* execve succeeded */
1600 current->fs->in_exec = 0;
1601 current->in_execve = 0;
1602 acct_update_integrals(current);
1605 put_files_struct(displaced);
1610 acct_arg_size(bprm, 0);
1616 allow_write_access(bprm->file);
1622 current->fs->in_exec = 0;
1623 current->in_execve = 0;
1630 reset_files_struct(displaced);
1635 int do_execve(const char *filename,
1636 const char __user *const __user *__argv,
1637 const char __user *const __user *__envp,
1638 struct pt_regs *regs)
1640 struct user_arg_ptr argv = { .ptr.native = __argv };
1641 struct user_arg_ptr envp = { .ptr.native = __envp };
1642 return do_execve_common(filename, argv, envp, regs);
1645 #ifdef CONFIG_COMPAT
1646 int compat_do_execve(char *filename,
1647 compat_uptr_t __user *__argv,
1648 compat_uptr_t __user *__envp,
1649 struct pt_regs *regs)
1651 struct user_arg_ptr argv = {
1653 .ptr.compat = __argv,
1655 struct user_arg_ptr envp = {
1657 .ptr.compat = __envp,
1659 return do_execve_common(filename, argv, envp, regs);
1663 void set_binfmt(struct linux_binfmt *new)
1665 struct mm_struct *mm = current->mm;
1668 module_put(mm->binfmt->module);
1672 __module_get(new->module);
1675 EXPORT_SYMBOL(set_binfmt);
1677 static int expand_corename(struct core_name *cn)
1679 char *old_corename = cn->corename;
1681 cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1682 cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1684 if (!cn->corename) {
1685 kfree(old_corename);
1692 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1700 need = vsnprintf(NULL, 0, fmt, arg);
1703 if (likely(need < cn->size - cn->used - 1))
1706 ret = expand_corename(cn);
1711 cur = cn->corename + cn->used;
1713 vsnprintf(cur, need + 1, fmt, arg);
1722 static void cn_escape(char *str)
1729 static int cn_print_exe_file(struct core_name *cn)
1731 struct file *exe_file;
1732 char *pathbuf, *path;
1735 exe_file = get_mm_exe_file(current->mm);
1737 char *commstart = cn->corename + cn->used;
1738 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1739 cn_escape(commstart);
1743 pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1749 path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1751 ret = PTR_ERR(path);
1757 ret = cn_printf(cn, "%s", path);
1766 /* format_corename will inspect the pattern parameter, and output a
1767 * name into corename, which must have space for at least
1768 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1770 static int format_corename(struct core_name *cn, long signr)
1772 const struct cred *cred = current_cred();
1773 const char *pat_ptr = core_pattern;
1774 int ispipe = (*pat_ptr == '|');
1775 int pid_in_pattern = 0;
1778 cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1779 cn->corename = kmalloc(cn->size, GFP_KERNEL);
1785 /* Repeat as long as we have more pattern to process and more output
1788 if (*pat_ptr != '%') {
1791 err = cn_printf(cn, "%c", *pat_ptr++);
1793 switch (*++pat_ptr) {
1794 /* single % at the end, drop that */
1797 /* Double percent, output one percent */
1799 err = cn_printf(cn, "%c", '%');
1804 err = cn_printf(cn, "%d",
1805 task_tgid_vnr(current));
1809 err = cn_printf(cn, "%d", cred->uid);
1813 err = cn_printf(cn, "%d", cred->gid);
1815 /* signal that caused the coredump */
1817 err = cn_printf(cn, "%ld", signr);
1819 /* UNIX time of coredump */
1822 do_gettimeofday(&tv);
1823 err = cn_printf(cn, "%lu", tv.tv_sec);
1828 char *namestart = cn->corename + cn->used;
1829 down_read(&uts_sem);
1830 err = cn_printf(cn, "%s",
1831 utsname()->nodename);
1833 cn_escape(namestart);
1838 char *commstart = cn->corename + cn->used;
1839 err = cn_printf(cn, "%s", current->comm);
1840 cn_escape(commstart);
1844 err = cn_print_exe_file(cn);
1846 /* core limit size */
1848 err = cn_printf(cn, "%lu",
1849 rlimit(RLIMIT_CORE));
1861 /* Backward compatibility with core_uses_pid:
1863 * If core_pattern does not include a %p (as is the default)
1864 * and core_uses_pid is set, then .%pid will be appended to
1865 * the filename. Do not do this for piped commands. */
1866 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1867 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1875 static int zap_process(struct task_struct *start, int exit_code)
1877 struct task_struct *t;
1880 start->signal->flags = SIGNAL_GROUP_EXIT;
1881 start->signal->group_exit_code = exit_code;
1882 start->signal->group_stop_count = 0;
1886 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1887 if (t != current && t->mm) {
1888 sigaddset(&t->pending.signal, SIGKILL);
1889 signal_wake_up(t, 1);
1892 } while_each_thread(start, t);
1897 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1898 struct core_state *core_state, int exit_code)
1900 struct task_struct *g, *p;
1901 unsigned long flags;
1904 spin_lock_irq(&tsk->sighand->siglock);
1905 if (!signal_group_exit(tsk->signal)) {
1906 mm->core_state = core_state;
1907 nr = zap_process(tsk, exit_code);
1909 spin_unlock_irq(&tsk->sighand->siglock);
1910 if (unlikely(nr < 0))
1913 if (atomic_read(&mm->mm_users) == nr + 1)
1916 * We should find and kill all tasks which use this mm, and we should
1917 * count them correctly into ->nr_threads. We don't take tasklist
1918 * lock, but this is safe wrt:
1921 * None of sub-threads can fork after zap_process(leader). All
1922 * processes which were created before this point should be
1923 * visible to zap_threads() because copy_process() adds the new
1924 * process to the tail of init_task.tasks list, and lock/unlock
1925 * of ->siglock provides a memory barrier.
1928 * The caller holds mm->mmap_sem. This means that the task which
1929 * uses this mm can't pass exit_mm(), so it can't exit or clear
1933 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1934 * we must see either old or new leader, this does not matter.
1935 * However, it can change p->sighand, so lock_task_sighand(p)
1936 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1939 * Note also that "g" can be the old leader with ->mm == NULL
1940 * and already unhashed and thus removed from ->thread_group.
1941 * This is OK, __unhash_process()->list_del_rcu() does not
1942 * clear the ->next pointer, we will find the new leader via
1946 for_each_process(g) {
1947 if (g == tsk->group_leader)
1949 if (g->flags & PF_KTHREAD)
1954 if (unlikely(p->mm == mm)) {
1955 lock_task_sighand(p, &flags);
1956 nr += zap_process(p, exit_code);
1957 unlock_task_sighand(p, &flags);
1961 } while_each_thread(g, p);
1965 atomic_set(&core_state->nr_threads, nr);
1969 static int coredump_wait(int exit_code, struct core_state *core_state)
1971 struct task_struct *tsk = current;
1972 struct mm_struct *mm = tsk->mm;
1973 struct completion *vfork_done;
1974 int core_waiters = -EBUSY;
1976 init_completion(&core_state->startup);
1977 core_state->dumper.task = tsk;
1978 core_state->dumper.next = NULL;
1980 down_write(&mm->mmap_sem);
1981 if (!mm->core_state)
1982 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1983 up_write(&mm->mmap_sem);
1985 if (unlikely(core_waiters < 0))
1989 * Make sure nobody is waiting for us to release the VM,
1990 * otherwise we can deadlock when we wait on each other
1992 vfork_done = tsk->vfork_done;
1994 tsk->vfork_done = NULL;
1995 complete(vfork_done);
1998 if (core_waiters > 0) {
1999 freezer_do_not_count();
2000 wait_for_completion(&core_state->startup);
2004 return core_waiters;
2007 static void coredump_finish(struct mm_struct *mm)
2009 struct core_thread *curr, *next;
2010 struct task_struct *task;
2012 next = mm->core_state->dumper.next;
2013 while ((curr = next) != NULL) {
2017 * see exit_mm(), curr->task must not see
2018 * ->task == NULL before we read ->next.
2022 wake_up_process(task);
2025 mm->core_state = NULL;
2029 * set_dumpable converts traditional three-value dumpable to two flags and
2030 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
2031 * these bits are not changed atomically. So get_dumpable can observe the
2032 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
2033 * return either old dumpable or new one by paying attention to the order of
2034 * modifying the bits.
2036 * dumpable | mm->flags (binary)
2037 * old new | initial interim final
2038 * ---------+-----------------------
2046 * (*) get_dumpable regards interim value of 10 as 11.
2048 void set_dumpable(struct mm_struct *mm, int value)
2052 clear_bit(MMF_DUMPABLE, &mm->flags);
2054 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2057 set_bit(MMF_DUMPABLE, &mm->flags);
2059 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2062 set_bit(MMF_DUMP_SECURELY, &mm->flags);
2064 set_bit(MMF_DUMPABLE, &mm->flags);
2069 static int __get_dumpable(unsigned long mm_flags)
2073 ret = mm_flags & MMF_DUMPABLE_MASK;
2074 return (ret >= 2) ? 2 : ret;
2078 * This returns the actual value of the suid_dumpable flag. For things
2079 * that are using this for checking for privilege transitions, it must
2080 * test against SUID_DUMP_USER rather than treating it as a boolean
2083 int get_dumpable(struct mm_struct *mm)
2085 return __get_dumpable(mm->flags);
2088 static void wait_for_dump_helpers(struct file *file)
2090 struct pipe_inode_info *pipe;
2092 pipe = file->f_path.dentry->d_inode->i_pipe;
2098 while ((pipe->readers > 1) && (!signal_pending(current))) {
2099 wake_up_interruptible_sync(&pipe->wait);
2100 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2113 * helper function to customize the process used
2114 * to collect the core in userspace. Specifically
2115 * it sets up a pipe and installs it as fd 0 (stdin)
2116 * for the process. Returns 0 on success, or
2117 * PTR_ERR on failure.
2118 * Note that it also sets the core limit to 1. This
2119 * is a special value that we use to trap recursive
2122 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2124 struct file *rp, *wp;
2125 struct fdtable *fdt;
2126 struct coredump_params *cp = (struct coredump_params *)info->data;
2127 struct files_struct *cf = current->files;
2129 wp = create_write_pipe(0);
2133 rp = create_read_pipe(wp, 0);
2135 free_write_pipe(wp);
2143 spin_lock(&cf->file_lock);
2144 fdt = files_fdtable(cf);
2145 __set_open_fd(0, fdt);
2146 __clear_close_on_exec(0, fdt);
2147 spin_unlock(&cf->file_lock);
2149 /* and disallow core files too */
2150 current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2155 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2157 struct core_state core_state;
2158 struct core_name cn;
2159 struct mm_struct *mm = current->mm;
2160 struct linux_binfmt * binfmt;
2161 const struct cred *old_cred;
2165 /* require nonrelative corefile path and be extra careful */
2166 bool need_suid_safe = false;
2167 static atomic_t core_dump_count = ATOMIC_INIT(0);
2168 struct coredump_params cprm = {
2171 .limit = rlimit(RLIMIT_CORE),
2173 * We must use the same mm->flags while dumping core to avoid
2174 * inconsistency of bit flags, since this flag is not protected
2177 .mm_flags = mm->flags,
2180 audit_core_dumps(signr);
2182 binfmt = mm->binfmt;
2183 if (!binfmt || !binfmt->core_dump)
2185 if (!__get_dumpable(cprm.mm_flags))
2188 cred = prepare_creds();
2192 * We cannot trust fsuid as being the "true" uid of the process
2193 * nor do we know its entire history. We only know it was tainted
2194 * so we dump it as root in mode 2, and only into a controlled
2195 * environment (pipe handler or fully qualified path).
2197 if (__get_dumpable(cprm.mm_flags) == 2) {
2198 /* Setuid core dump mode */
2199 cred->fsuid = 0; /* Dump root private */
2200 need_suid_safe = true;
2203 retval = coredump_wait(exit_code, &core_state);
2207 old_cred = override_creds(cred);
2210 * Clear any false indication of pending signals that might
2211 * be seen by the filesystem code called to write the core file.
2213 clear_thread_flag(TIF_SIGPENDING);
2215 ispipe = format_corename(&cn, signr);
2222 printk(KERN_WARNING "format_corename failed\n");
2223 printk(KERN_WARNING "Aborting core\n");
2227 if (cprm.limit == 1) {
2229 * Normally core limits are irrelevant to pipes, since
2230 * we're not writing to the file system, but we use
2231 * cprm.limit of 1 here as a speacial value. Any
2232 * non-1 limit gets set to RLIM_INFINITY below, but
2233 * a limit of 0 skips the dump. This is a consistent
2234 * way to catch recursive crashes. We can still crash
2235 * if the core_pattern binary sets RLIM_CORE = !1
2236 * but it runs as root, and can do lots of stupid things
2237 * Note that we use task_tgid_vnr here to grab the pid
2238 * of the process group leader. That way we get the
2239 * right pid if a thread in a multi-threaded
2240 * core_pattern process dies.
2243 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2244 task_tgid_vnr(current), current->comm);
2245 printk(KERN_WARNING "Aborting core\n");
2248 cprm.limit = RLIM_INFINITY;
2250 dump_count = atomic_inc_return(&core_dump_count);
2251 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2252 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2253 task_tgid_vnr(current), current->comm);
2254 printk(KERN_WARNING "Skipping core dump\n");
2255 goto fail_dropcount;
2258 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2260 printk(KERN_WARNING "%s failed to allocate memory\n",
2262 goto fail_dropcount;
2265 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2266 NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2268 argv_free(helper_argv);
2270 printk(KERN_INFO "Core dump to %s pipe failed\n",
2275 struct inode *inode;
2276 int open_flags = O_CREAT | O_RDWR | O_NOFOLLOW |
2277 O_LARGEFILE | O_EXCL;
2279 if (cprm.limit < binfmt->min_coredump)
2282 if (need_suid_safe && cn.corename[0] != '/') {
2283 printk(KERN_WARNING "Pid %d(%s) can only dump core "\
2284 "to fully qualified path!\n",
2285 task_tgid_vnr(current), current->comm);
2286 printk(KERN_WARNING "Skipping core dump\n");
2291 * Unlink the file if it exists unless this is a SUID
2292 * binary - in that case, we're running around with root
2293 * privs and don't want to unlink another user's coredump.
2295 if (!need_suid_safe) {
2296 mm_segment_t old_fs;
2301 * If it doesn't exist, that's fine. If there's some
2302 * other problem, we'll catch it at the filp_open().
2304 (void) sys_unlink((const char __user *)cn.corename);
2309 * There is a race between unlinking and creating the
2310 * file, but if that causes an EEXIST here, that's
2311 * fine - another process raced with us while creating
2312 * the corefile, and the other process won. To userspace,
2313 * what matters is that at least one of the two processes
2314 * writes its coredump successfully, not which one.
2316 if (need_suid_safe) {
2318 * Using user namespaces, normal user tasks can change
2319 * their current->fs->root to point to arbitrary
2320 * directories. Since the intention of the "only dump
2321 * with a fully qualified path" rule is to control where
2322 * coredumps may be placed using root privileges,
2323 * current->fs->root must not be used. Instead, use the
2324 * root directory of init_task.
2328 task_lock(&init_task);
2329 get_fs_root(init_task.fs, &root);
2330 task_unlock(&init_task);
2331 cprm.file = file_open_root(root.dentry, root.mnt,
2332 cn.corename, open_flags, 0600);
2335 cprm.file = filp_open(cn.corename, open_flags, 0600);
2337 if (IS_ERR(cprm.file))
2340 inode = cprm.file->f_path.dentry->d_inode;
2341 if (inode->i_nlink > 1)
2343 if (d_unhashed(cprm.file->f_path.dentry))
2346 * AK: actually i see no reason to not allow this for named
2347 * pipes etc, but keep the previous behaviour for now.
2349 if (!S_ISREG(inode->i_mode))
2352 * Dont allow local users get cute and trick others to coredump
2353 * into their pre-created files.
2355 if (inode->i_uid != current_fsuid())
2357 if (!cprm.file->f_op || !cprm.file->f_op->write)
2359 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2363 retval = binfmt->core_dump(&cprm);
2365 current->signal->group_exit_code |= 0x80;
2367 if (ispipe && core_pipe_limit)
2368 wait_for_dump_helpers(cprm.file);
2371 filp_close(cprm.file, NULL);
2374 atomic_dec(&core_dump_count);
2378 coredump_finish(mm);
2379 revert_creds(old_cred);
2387 * Core dumping helper functions. These are the only things you should
2388 * do on a core-file: use only these functions to write out all the
2391 int dump_write(struct file *file, const void *addr, int nr)
2393 return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2395 EXPORT_SYMBOL(dump_write);
2397 int dump_seek(struct file *file, loff_t off)
2401 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2402 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2405 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2410 unsigned long n = off;
2414 if (!dump_write(file, buf, n)) {
2420 free_page((unsigned long)buf);
2424 EXPORT_SYMBOL(dump_seek);