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>
29 #include <linux/vmacache.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/vmalloc.h>
61 #include <asm/uaccess.h>
62 #include <asm/mmu_context.h>
65 #include <trace/events/task.h>
68 #include <trace/events/sched.h>
70 int suid_dumpable = 0;
72 static LIST_HEAD(formats);
73 static DEFINE_RWLOCK(binfmt_lock);
75 void __register_binfmt(struct linux_binfmt * fmt, int insert)
78 if (WARN_ON(!fmt->load_binary))
80 write_lock(&binfmt_lock);
81 insert ? list_add(&fmt->lh, &formats) :
82 list_add_tail(&fmt->lh, &formats);
83 write_unlock(&binfmt_lock);
86 EXPORT_SYMBOL(__register_binfmt);
88 void unregister_binfmt(struct linux_binfmt * fmt)
90 write_lock(&binfmt_lock);
92 write_unlock(&binfmt_lock);
95 EXPORT_SYMBOL(unregister_binfmt);
97 static inline void put_binfmt(struct linux_binfmt * fmt)
99 module_put(fmt->module);
102 bool path_noexec(const struct path *path)
104 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
105 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
110 * Note that a shared library must be both readable and executable due to
113 * Also note that we take the address to load from from the file itself.
115 SYSCALL_DEFINE1(uselib, const char __user *, library)
117 struct linux_binfmt *fmt;
119 struct filename *tmp = getname(library);
120 int error = PTR_ERR(tmp);
121 static const struct open_flags uselib_flags = {
122 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
123 .acc_mode = MAY_READ | MAY_EXEC,
124 .intent = LOOKUP_OPEN,
125 .lookup_flags = LOOKUP_FOLLOW,
131 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
133 error = PTR_ERR(file);
138 if (!S_ISREG(file_inode(file)->i_mode))
142 if (path_noexec(&file->f_path))
149 read_lock(&binfmt_lock);
150 list_for_each_entry(fmt, &formats, lh) {
151 if (!fmt->load_shlib)
153 if (!try_module_get(fmt->module))
155 read_unlock(&binfmt_lock);
156 error = fmt->load_shlib(file);
157 read_lock(&binfmt_lock);
159 if (error != -ENOEXEC)
162 read_unlock(&binfmt_lock);
168 #endif /* #ifdef CONFIG_USELIB */
172 * The nascent bprm->mm is not visible until exec_mmap() but it can
173 * use a lot of memory, account these pages in current->mm temporary
174 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
175 * change the counter back via acct_arg_size(0).
177 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
179 struct mm_struct *mm = current->mm;
180 long diff = (long)(pages - bprm->vma_pages);
185 bprm->vma_pages = pages;
186 add_mm_counter(mm, MM_ANONPAGES, diff);
189 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
195 #ifdef CONFIG_STACK_GROWSUP
197 ret = expand_downwards(bprm->vma, pos);
203 * We are doing an exec(). 'current' is the process
204 * doing the exec and bprm->mm is the new process's mm.
206 ret = get_user_pages_remote(current, bprm->mm, pos, 1, write,
212 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
215 acct_arg_size(bprm, size / PAGE_SIZE);
218 * We've historically supported up to 32 pages (ARG_MAX)
219 * of argument strings even with small stacks
225 * Limit to 1/4-th the stack size for the argv+env strings.
227 * - the remaining binfmt code will not run out of stack space,
228 * - the program will have a reasonable amount of stack left
231 rlim = current->signal->rlim;
232 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
241 static void put_arg_page(struct page *page)
246 static void free_arg_pages(struct linux_binprm *bprm)
250 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
253 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
256 static int __bprm_mm_init(struct linux_binprm *bprm)
259 struct vm_area_struct *vma = NULL;
260 struct mm_struct *mm = bprm->mm;
262 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
266 if (down_write_killable(&mm->mmap_sem)) {
273 * Place the stack at the largest stack address the architecture
274 * supports. Later, we'll move this to an appropriate place. We don't
275 * use STACK_TOP because that can depend on attributes which aren't
278 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
279 vma->vm_end = STACK_TOP_MAX;
280 vma->vm_start = vma->vm_end - PAGE_SIZE;
281 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
282 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
283 INIT_LIST_HEAD(&vma->anon_vma_chain);
285 err = insert_vm_struct(mm, vma);
289 mm->stack_vm = mm->total_vm = 1;
290 arch_bprm_mm_init(mm, vma);
291 up_write(&mm->mmap_sem);
292 bprm->p = vma->vm_end - sizeof(void *);
295 up_write(&mm->mmap_sem);
298 kmem_cache_free(vm_area_cachep, vma);
302 static bool valid_arg_len(struct linux_binprm *bprm, long len)
304 return len <= MAX_ARG_STRLEN;
309 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
313 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
318 page = bprm->page[pos / PAGE_SIZE];
319 if (!page && write) {
320 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
323 bprm->page[pos / PAGE_SIZE] = page;
329 static void put_arg_page(struct page *page)
333 static void free_arg_page(struct linux_binprm *bprm, int i)
336 __free_page(bprm->page[i]);
337 bprm->page[i] = NULL;
341 static void free_arg_pages(struct linux_binprm *bprm)
345 for (i = 0; i < MAX_ARG_PAGES; i++)
346 free_arg_page(bprm, i);
349 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
354 static int __bprm_mm_init(struct linux_binprm *bprm)
356 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
360 static bool valid_arg_len(struct linux_binprm *bprm, long len)
362 return len <= bprm->p;
365 #endif /* CONFIG_MMU */
368 * Create a new mm_struct and populate it with a temporary stack
369 * vm_area_struct. We don't have enough context at this point to set the stack
370 * flags, permissions, and offset, so we use temporary values. We'll update
371 * them later in setup_arg_pages().
373 static int bprm_mm_init(struct linux_binprm *bprm)
376 struct mm_struct *mm = NULL;
378 bprm->mm = mm = mm_alloc();
383 err = __bprm_mm_init(bprm);
398 struct user_arg_ptr {
403 const char __user *const __user *native;
405 const compat_uptr_t __user *compat;
410 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
412 const char __user *native;
415 if (unlikely(argv.is_compat)) {
416 compat_uptr_t compat;
418 if (get_user(compat, argv.ptr.compat + nr))
419 return ERR_PTR(-EFAULT);
421 return compat_ptr(compat);
425 if (get_user(native, argv.ptr.native + nr))
426 return ERR_PTR(-EFAULT);
432 * count() counts the number of strings in array ARGV.
434 static int count(struct user_arg_ptr argv, int max)
438 if (argv.ptr.native != NULL) {
440 const char __user *p = get_user_arg_ptr(argv, i);
452 if (fatal_signal_pending(current))
453 return -ERESTARTNOHAND;
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;
470 unsigned long kpos = 0;
474 const char __user *str;
479 str = get_user_arg_ptr(argv, argc);
483 len = strnlen_user(str, MAX_ARG_STRLEN);
488 if (!valid_arg_len(bprm, len))
491 /* We're going to work our way backwords. */
497 int offset, bytes_to_copy;
499 if (fatal_signal_pending(current)) {
500 ret = -ERESTARTNOHAND;
505 offset = pos % PAGE_SIZE;
509 bytes_to_copy = offset;
510 if (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)) {
521 page = get_arg_page(bprm, pos, 1);
528 flush_kernel_dcache_page(kmapped_page);
529 kunmap(kmapped_page);
530 put_arg_page(kmapped_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)) {
546 flush_kernel_dcache_page(kmapped_page);
547 kunmap(kmapped_page);
548 put_arg_page(kmapped_page);
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)
560 mm_segment_t oldfs = get_fs();
561 struct user_arg_ptr argv = {
562 .ptr.native = (const char __user *const __user *)__argv,
566 r = copy_strings(argc, argv, bprm);
571 EXPORT_SYMBOL(copy_strings_kernel);
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
603 if (vma != find_vma(mm, new_start))
607 * cover the whole range: [new_start, old_end)
609 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
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, false))
621 tlb_gather_mmu(&tlb, mm, old_start, old_end);
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 : USER_PGTABLES_CEILING);
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 : USER_PGTABLES_CEILING);
638 tlb_finish_mmu(&tlb, old_start, 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);
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)
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 */
669 stack_base = rlimit_max(RLIMIT_STACK);
670 if (stack_base > STACK_SIZE_MAX)
671 stack_base = STACK_SIZE_MAX;
673 /* Add space for stack randomization. */
674 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
676 /* Make sure we didn't let the argument array grow too large. */
677 if (vma->vm_end - vma->vm_start > stack_base)
680 stack_base = PAGE_ALIGN(stack_top - stack_base);
682 stack_shift = vma->vm_start - stack_base;
683 mm->arg_start = bprm->p - stack_shift;
684 bprm->p = vma->vm_end - stack_shift;
686 stack_top = arch_align_stack(stack_top);
687 stack_top = PAGE_ALIGN(stack_top);
689 if (unlikely(stack_top < mmap_min_addr) ||
690 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
693 stack_shift = vma->vm_end - stack_top;
695 bprm->p -= stack_shift;
696 mm->arg_start = bprm->p;
700 bprm->loader -= stack_shift;
701 bprm->exec -= stack_shift;
703 if (down_write_killable(&mm->mmap_sem))
706 vm_flags = VM_STACK_FLAGS;
709 * Adjust stack execute permissions; explicitly enable for
710 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
711 * (arch default) otherwise.
713 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
715 else if (executable_stack == EXSTACK_DISABLE_X)
716 vm_flags &= ~VM_EXEC;
717 vm_flags |= mm->def_flags;
718 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
720 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
726 /* Move stack pages down in memory. */
728 ret = shift_arg_pages(vma, stack_shift);
733 /* mprotect_fixup is overkill to remove the temporary stack flags */
734 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
736 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
737 stack_size = vma->vm_end - vma->vm_start;
739 * Align this down to a page boundary as expand_stack
742 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
743 #ifdef CONFIG_STACK_GROWSUP
744 if (stack_size + stack_expand > rlim_stack)
745 stack_base = vma->vm_start + rlim_stack;
747 stack_base = vma->vm_end + stack_expand;
749 if (stack_size + stack_expand > rlim_stack)
750 stack_base = vma->vm_end - rlim_stack;
752 stack_base = vma->vm_start - stack_expand;
754 current->mm->start_stack = bprm->p;
755 ret = expand_stack(vma, stack_base);
760 up_write(&mm->mmap_sem);
763 EXPORT_SYMBOL(setup_arg_pages);
765 #endif /* CONFIG_MMU */
767 static struct file *do_open_execat(int fd, struct filename *name, int flags)
771 struct open_flags open_exec_flags = {
772 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
773 .acc_mode = MAY_EXEC,
774 .intent = LOOKUP_OPEN,
775 .lookup_flags = LOOKUP_FOLLOW,
778 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
779 return ERR_PTR(-EINVAL);
780 if (flags & AT_SYMLINK_NOFOLLOW)
781 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
782 if (flags & AT_EMPTY_PATH)
783 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
785 file = do_filp_open(fd, name, &open_exec_flags);
790 if (!S_ISREG(file_inode(file)->i_mode))
793 if (path_noexec(&file->f_path))
796 err = deny_write_access(file);
800 if (name->name[0] != '\0')
811 struct file *open_exec(const char *name)
813 struct filename *filename = getname_kernel(name);
814 struct file *f = ERR_CAST(filename);
816 if (!IS_ERR(filename)) {
817 f = do_open_execat(AT_FDCWD, filename, 0);
822 EXPORT_SYMBOL(open_exec);
824 int kernel_read(struct file *file, loff_t offset,
825 char *addr, unsigned long count)
833 /* The cast to a user pointer is valid due to the set_fs() */
834 result = vfs_read(file, (void __user *)addr, count, &pos);
839 EXPORT_SYMBOL(kernel_read);
841 int kernel_read_file(struct file *file, void **buf, loff_t *size,
842 loff_t max_size, enum kernel_read_file_id id)
848 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
851 ret = security_kernel_read_file(file, id);
855 ret = deny_write_access(file);
859 i_size = i_size_read(file_inode(file));
860 if (max_size > 0 && i_size > max_size) {
869 *buf = vmalloc(i_size);
876 while (pos < i_size) {
877 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
894 ret = security_kernel_post_read_file(file, *buf, i_size, id);
905 allow_write_access(file);
908 EXPORT_SYMBOL_GPL(kernel_read_file);
910 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
911 loff_t max_size, enum kernel_read_file_id id)
919 file = filp_open(path, O_RDONLY, 0);
921 return PTR_ERR(file);
923 ret = kernel_read_file(file, buf, size, max_size, id);
927 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
929 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
930 enum kernel_read_file_id id)
932 struct fd f = fdget(fd);
938 ret = kernel_read_file(f.file, buf, size, max_size, id);
943 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
945 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
947 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
949 flush_icache_range(addr, addr + len);
952 EXPORT_SYMBOL(read_code);
954 static int exec_mmap(struct mm_struct *mm)
956 struct task_struct *tsk;
957 struct mm_struct *old_mm, *active_mm;
959 /* Notify parent that we're no longer interested in the old VM */
961 old_mm = current->mm;
962 mm_release(tsk, old_mm);
967 * Make sure that if there is a core dump in progress
968 * for the old mm, we get out and die instead of going
969 * through with the exec. We must hold mmap_sem around
970 * checking core_state and changing tsk->mm.
972 down_read(&old_mm->mmap_sem);
973 if (unlikely(old_mm->core_state)) {
974 up_read(&old_mm->mmap_sem);
979 active_mm = tsk->active_mm;
982 activate_mm(active_mm, mm);
983 tsk->mm->vmacache_seqnum = 0;
987 up_read(&old_mm->mmap_sem);
988 BUG_ON(active_mm != old_mm);
989 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
990 mm_update_next_owner(old_mm);
999 * This function makes sure the current process has its own signal table,
1000 * so that flush_signal_handlers can later reset the handlers without
1001 * disturbing other processes. (Other processes might share the signal
1002 * table via the CLONE_SIGHAND option to clone().)
1004 static int de_thread(struct task_struct *tsk)
1006 struct signal_struct *sig = tsk->signal;
1007 struct sighand_struct *oldsighand = tsk->sighand;
1008 spinlock_t *lock = &oldsighand->siglock;
1010 if (thread_group_empty(tsk))
1011 goto no_thread_group;
1014 * Kill all other threads in the thread group.
1016 spin_lock_irq(lock);
1017 if (signal_group_exit(sig)) {
1019 * Another group action in progress, just
1020 * return so that the signal is processed.
1022 spin_unlock_irq(lock);
1026 sig->group_exit_task = tsk;
1027 sig->notify_count = zap_other_threads(tsk);
1028 if (!thread_group_leader(tsk))
1029 sig->notify_count--;
1031 while (sig->notify_count) {
1032 __set_current_state(TASK_KILLABLE);
1033 spin_unlock_irq(lock);
1035 if (unlikely(__fatal_signal_pending(tsk)))
1037 spin_lock_irq(lock);
1039 spin_unlock_irq(lock);
1042 * At this point all other threads have exited, all we have to
1043 * do is to wait for the thread group leader to become inactive,
1044 * and to assume its PID:
1046 if (!thread_group_leader(tsk)) {
1047 struct task_struct *leader = tsk->group_leader;
1050 threadgroup_change_begin(tsk);
1051 write_lock_irq(&tasklist_lock);
1053 * Do this under tasklist_lock to ensure that
1054 * exit_notify() can't miss ->group_exit_task
1056 sig->notify_count = -1;
1057 if (likely(leader->exit_state))
1059 __set_current_state(TASK_KILLABLE);
1060 write_unlock_irq(&tasklist_lock);
1061 threadgroup_change_end(tsk);
1063 if (unlikely(__fatal_signal_pending(tsk)))
1068 * The only record we have of the real-time age of a
1069 * process, regardless of execs it's done, is start_time.
1070 * All the past CPU time is accumulated in signal_struct
1071 * from sister threads now dead. But in this non-leader
1072 * exec, nothing survives from the original leader thread,
1073 * whose birth marks the true age of this process now.
1074 * When we take on its identity by switching to its PID, we
1075 * also take its birthdate (always earlier than our own).
1077 tsk->start_time = leader->start_time;
1078 tsk->real_start_time = leader->real_start_time;
1080 BUG_ON(!same_thread_group(leader, tsk));
1081 BUG_ON(has_group_leader_pid(tsk));
1083 * An exec() starts a new thread group with the
1084 * TGID of the previous thread group. Rehash the
1085 * two threads with a switched PID, and release
1086 * the former thread group leader:
1089 /* Become a process group leader with the old leader's pid.
1090 * The old leader becomes a thread of the this thread group.
1091 * Note: The old leader also uses this pid until release_task
1092 * is called. Odd but simple and correct.
1094 tsk->pid = leader->pid;
1095 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1096 transfer_pid(leader, tsk, PIDTYPE_PGID);
1097 transfer_pid(leader, tsk, PIDTYPE_SID);
1099 list_replace_rcu(&leader->tasks, &tsk->tasks);
1100 list_replace_init(&leader->sibling, &tsk->sibling);
1102 tsk->group_leader = tsk;
1103 leader->group_leader = tsk;
1105 tsk->exit_signal = SIGCHLD;
1106 leader->exit_signal = -1;
1108 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1109 leader->exit_state = EXIT_DEAD;
1112 * We are going to release_task()->ptrace_unlink() silently,
1113 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1114 * the tracer wont't block again waiting for this thread.
1116 if (unlikely(leader->ptrace))
1117 __wake_up_parent(leader, leader->parent);
1118 write_unlock_irq(&tasklist_lock);
1119 threadgroup_change_end(tsk);
1121 release_task(leader);
1124 sig->group_exit_task = NULL;
1125 sig->notify_count = 0;
1128 /* we have changed execution domain */
1129 tsk->exit_signal = SIGCHLD;
1132 flush_itimer_signals();
1134 if (atomic_read(&oldsighand->count) != 1) {
1135 struct sighand_struct *newsighand;
1137 * This ->sighand is shared with the CLONE_SIGHAND
1138 * but not CLONE_THREAD task, switch to the new one.
1140 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1144 atomic_set(&newsighand->count, 1);
1145 memcpy(newsighand->action, oldsighand->action,
1146 sizeof(newsighand->action));
1148 write_lock_irq(&tasklist_lock);
1149 spin_lock(&oldsighand->siglock);
1150 rcu_assign_pointer(tsk->sighand, newsighand);
1151 spin_unlock(&oldsighand->siglock);
1152 write_unlock_irq(&tasklist_lock);
1154 __cleanup_sighand(oldsighand);
1157 BUG_ON(!thread_group_leader(tsk));
1161 /* protects against exit_notify() and __exit_signal() */
1162 read_lock(&tasklist_lock);
1163 sig->group_exit_task = NULL;
1164 sig->notify_count = 0;
1165 read_unlock(&tasklist_lock);
1169 char *get_task_comm(char *buf, struct task_struct *tsk)
1171 /* buf must be at least sizeof(tsk->comm) in size */
1173 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1177 EXPORT_SYMBOL_GPL(get_task_comm);
1180 * These functions flushes out all traces of the currently running executable
1181 * so that a new one can be started
1184 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1187 trace_task_rename(tsk, buf);
1188 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1190 perf_event_comm(tsk, exec);
1193 int flush_old_exec(struct linux_binprm * bprm)
1198 * Make sure we have a private signal table and that
1199 * we are unassociated from the previous thread group.
1201 retval = de_thread(current);
1206 * Must be called _before_ exec_mmap() as bprm->mm is
1207 * not visibile until then. This also enables the update
1210 set_mm_exe_file(bprm->mm, bprm->file);
1213 * Release all of the old mmap stuff
1215 acct_arg_size(bprm, 0);
1216 retval = exec_mmap(bprm->mm);
1220 bprm->mm = NULL; /* We're using it now */
1223 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1224 PF_NOFREEZE | PF_NO_SETAFFINITY);
1226 current->personality &= ~bprm->per_clear;
1233 EXPORT_SYMBOL(flush_old_exec);
1235 void would_dump(struct linux_binprm *bprm, struct file *file)
1237 if (inode_permission(file_inode(file), MAY_READ) < 0)
1238 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1240 EXPORT_SYMBOL(would_dump);
1242 void setup_new_exec(struct linux_binprm * bprm)
1244 arch_pick_mmap_layout(current->mm);
1246 /* This is the point of no return */
1247 current->sas_ss_sp = current->sas_ss_size = 0;
1249 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1250 set_dumpable(current->mm, SUID_DUMP_USER);
1252 set_dumpable(current->mm, suid_dumpable);
1255 __set_task_comm(current, kbasename(bprm->filename), true);
1257 /* Set the new mm task size. We have to do that late because it may
1258 * depend on TIF_32BIT which is only updated in flush_thread() on
1259 * some architectures like powerpc
1261 current->mm->task_size = TASK_SIZE;
1263 /* install the new credentials */
1264 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1265 !gid_eq(bprm->cred->gid, current_egid())) {
1266 current->pdeath_signal = 0;
1268 would_dump(bprm, bprm->file);
1269 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1270 set_dumpable(current->mm, suid_dumpable);
1273 /* An exec changes our domain. We are no longer part of the thread
1275 current->self_exec_id++;
1276 flush_signal_handlers(current, 0);
1277 do_close_on_exec(current->files);
1279 EXPORT_SYMBOL(setup_new_exec);
1282 * Prepare credentials and lock ->cred_guard_mutex.
1283 * install_exec_creds() commits the new creds and drops the lock.
1284 * Or, if exec fails before, free_bprm() should release ->cred and
1287 int prepare_bprm_creds(struct linux_binprm *bprm)
1289 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1290 return -ERESTARTNOINTR;
1292 bprm->cred = prepare_exec_creds();
1293 if (likely(bprm->cred))
1296 mutex_unlock(¤t->signal->cred_guard_mutex);
1300 static void free_bprm(struct linux_binprm *bprm)
1302 free_arg_pages(bprm);
1304 mutex_unlock(¤t->signal->cred_guard_mutex);
1305 abort_creds(bprm->cred);
1308 allow_write_access(bprm->file);
1311 /* If a binfmt changed the interp, free it. */
1312 if (bprm->interp != bprm->filename)
1313 kfree(bprm->interp);
1317 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1319 /* If a binfmt changed the interp, free it first. */
1320 if (bprm->interp != bprm->filename)
1321 kfree(bprm->interp);
1322 bprm->interp = kstrdup(interp, GFP_KERNEL);
1327 EXPORT_SYMBOL(bprm_change_interp);
1330 * install the new credentials for this executable
1332 void install_exec_creds(struct linux_binprm *bprm)
1334 security_bprm_committing_creds(bprm);
1336 commit_creds(bprm->cred);
1340 * Disable monitoring for regular users
1341 * when executing setuid binaries. Must
1342 * wait until new credentials are committed
1343 * by commit_creds() above
1345 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1346 perf_event_exit_task(current);
1348 * cred_guard_mutex must be held at least to this point to prevent
1349 * ptrace_attach() from altering our determination of the task's
1350 * credentials; any time after this it may be unlocked.
1352 security_bprm_committed_creds(bprm);
1353 mutex_unlock(¤t->signal->cred_guard_mutex);
1355 EXPORT_SYMBOL(install_exec_creds);
1358 * determine how safe it is to execute the proposed program
1359 * - the caller must hold ->cred_guard_mutex to protect against
1360 * PTRACE_ATTACH or seccomp thread-sync
1362 static void check_unsafe_exec(struct linux_binprm *bprm)
1364 struct task_struct *p = current, *t;
1368 if (p->ptrace & PT_PTRACE_CAP)
1369 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1371 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1375 * This isn't strictly necessary, but it makes it harder for LSMs to
1378 if (task_no_new_privs(current))
1379 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1383 spin_lock(&p->fs->lock);
1385 while_each_thread(p, t) {
1391 if (p->fs->users > n_fs)
1392 bprm->unsafe |= LSM_UNSAFE_SHARE;
1395 spin_unlock(&p->fs->lock);
1398 static void bprm_fill_uid(struct linux_binprm *bprm)
1400 struct inode *inode;
1406 * Since this can be called multiple times (via prepare_binprm),
1407 * we must clear any previous work done when setting set[ug]id
1408 * bits from any earlier bprm->file uses (for example when run
1409 * first for a setuid script then again for its interpreter).
1411 bprm->cred->euid = current_euid();
1412 bprm->cred->egid = current_egid();
1414 if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1417 if (task_no_new_privs(current))
1420 inode = file_inode(bprm->file);
1421 mode = READ_ONCE(inode->i_mode);
1422 if (!(mode & (S_ISUID|S_ISGID)))
1425 /* Be careful if suid/sgid is set */
1428 /* reload atomically mode/uid/gid now that lock held */
1429 mode = inode->i_mode;
1432 inode_unlock(inode);
1434 /* We ignore suid/sgid if there are no mappings for them in the ns */
1435 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1436 !kgid_has_mapping(bprm->cred->user_ns, gid))
1439 if (mode & S_ISUID) {
1440 bprm->per_clear |= PER_CLEAR_ON_SETID;
1441 bprm->cred->euid = uid;
1444 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1445 bprm->per_clear |= PER_CLEAR_ON_SETID;
1446 bprm->cred->egid = gid;
1451 * Fill the binprm structure from the inode.
1452 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1454 * This may be called multiple times for binary chains (scripts for example).
1456 int prepare_binprm(struct linux_binprm *bprm)
1460 bprm_fill_uid(bprm);
1462 /* fill in binprm security blob */
1463 retval = security_bprm_set_creds(bprm);
1466 bprm->cred_prepared = 1;
1468 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1469 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1472 EXPORT_SYMBOL(prepare_binprm);
1475 * Arguments are '\0' separated strings found at the location bprm->p
1476 * points to; chop off the first by relocating brpm->p to right after
1477 * the first '\0' encountered.
1479 int remove_arg_zero(struct linux_binprm *bprm)
1482 unsigned long offset;
1490 offset = bprm->p & ~PAGE_MASK;
1491 page = get_arg_page(bprm, bprm->p, 0);
1496 kaddr = kmap_atomic(page);
1498 for (; offset < PAGE_SIZE && kaddr[offset];
1499 offset++, bprm->p++)
1502 kunmap_atomic(kaddr);
1504 } while (offset == PAGE_SIZE);
1513 EXPORT_SYMBOL(remove_arg_zero);
1515 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1517 * cycle the list of binary formats handler, until one recognizes the image
1519 int search_binary_handler(struct linux_binprm *bprm)
1521 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1522 struct linux_binfmt *fmt;
1525 /* This allows 4 levels of binfmt rewrites before failing hard. */
1526 if (bprm->recursion_depth > 5)
1529 retval = security_bprm_check(bprm);
1535 read_lock(&binfmt_lock);
1536 list_for_each_entry(fmt, &formats, lh) {
1537 if (!try_module_get(fmt->module))
1539 read_unlock(&binfmt_lock);
1540 bprm->recursion_depth++;
1541 retval = fmt->load_binary(bprm);
1542 read_lock(&binfmt_lock);
1544 bprm->recursion_depth--;
1545 if (retval < 0 && !bprm->mm) {
1546 /* we got to flush_old_exec() and failed after it */
1547 read_unlock(&binfmt_lock);
1548 force_sigsegv(SIGSEGV, current);
1551 if (retval != -ENOEXEC || !bprm->file) {
1552 read_unlock(&binfmt_lock);
1556 read_unlock(&binfmt_lock);
1559 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1560 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1562 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1570 EXPORT_SYMBOL(search_binary_handler);
1572 static int exec_binprm(struct linux_binprm *bprm)
1574 pid_t old_pid, old_vpid;
1577 /* Need to fetch pid before load_binary changes it */
1578 old_pid = current->pid;
1580 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1583 ret = search_binary_handler(bprm);
1586 trace_sched_process_exec(current, old_pid, bprm);
1587 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1588 proc_exec_connector(current);
1595 * sys_execve() executes a new program.
1597 static int do_execveat_common(int fd, struct filename *filename,
1598 struct user_arg_ptr argv,
1599 struct user_arg_ptr envp,
1602 char *pathbuf = NULL;
1603 struct linux_binprm *bprm;
1605 struct files_struct *displaced;
1608 if (IS_ERR(filename))
1609 return PTR_ERR(filename);
1612 * We move the actual failure in case of RLIMIT_NPROC excess from
1613 * set*uid() to execve() because too many poorly written programs
1614 * don't check setuid() return code. Here we additionally recheck
1615 * whether NPROC limit is still exceeded.
1617 if ((current->flags & PF_NPROC_EXCEEDED) &&
1618 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1623 /* We're below the limit (still or again), so we don't want to make
1624 * further execve() calls fail. */
1625 current->flags &= ~PF_NPROC_EXCEEDED;
1627 retval = unshare_files(&displaced);
1632 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1636 retval = prepare_bprm_creds(bprm);
1640 check_unsafe_exec(bprm);
1641 current->in_execve = 1;
1643 file = do_open_execat(fd, filename, flags);
1644 retval = PTR_ERR(file);
1651 if (fd == AT_FDCWD || filename->name[0] == '/') {
1652 bprm->filename = filename->name;
1654 if (filename->name[0] == '\0')
1655 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1657 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1658 fd, filename->name);
1664 * Record that a name derived from an O_CLOEXEC fd will be
1665 * inaccessible after exec. Relies on having exclusive access to
1666 * current->files (due to unshare_files above).
1668 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1669 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1670 bprm->filename = pathbuf;
1672 bprm->interp = bprm->filename;
1674 retval = bprm_mm_init(bprm);
1678 bprm->argc = count(argv, MAX_ARG_STRINGS);
1679 if ((retval = bprm->argc) < 0)
1682 bprm->envc = count(envp, MAX_ARG_STRINGS);
1683 if ((retval = bprm->envc) < 0)
1686 retval = prepare_binprm(bprm);
1690 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1694 bprm->exec = bprm->p;
1695 retval = copy_strings(bprm->envc, envp, bprm);
1699 retval = copy_strings(bprm->argc, argv, bprm);
1703 retval = exec_binprm(bprm);
1707 /* execve succeeded */
1708 current->fs->in_exec = 0;
1709 current->in_execve = 0;
1710 acct_update_integrals(current);
1711 task_numa_free(current);
1716 put_files_struct(displaced);
1721 acct_arg_size(bprm, 0);
1726 current->fs->in_exec = 0;
1727 current->in_execve = 0;
1735 reset_files_struct(displaced);
1741 int do_execve(struct filename *filename,
1742 const char __user *const __user *__argv,
1743 const char __user *const __user *__envp)
1745 struct user_arg_ptr argv = { .ptr.native = __argv };
1746 struct user_arg_ptr envp = { .ptr.native = __envp };
1747 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1750 int do_execveat(int fd, struct filename *filename,
1751 const char __user *const __user *__argv,
1752 const char __user *const __user *__envp,
1755 struct user_arg_ptr argv = { .ptr.native = __argv };
1756 struct user_arg_ptr envp = { .ptr.native = __envp };
1758 return do_execveat_common(fd, filename, argv, envp, flags);
1761 #ifdef CONFIG_COMPAT
1762 static int compat_do_execve(struct filename *filename,
1763 const compat_uptr_t __user *__argv,
1764 const compat_uptr_t __user *__envp)
1766 struct user_arg_ptr argv = {
1768 .ptr.compat = __argv,
1770 struct user_arg_ptr envp = {
1772 .ptr.compat = __envp,
1774 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1777 static int compat_do_execveat(int fd, struct filename *filename,
1778 const compat_uptr_t __user *__argv,
1779 const compat_uptr_t __user *__envp,
1782 struct user_arg_ptr argv = {
1784 .ptr.compat = __argv,
1786 struct user_arg_ptr envp = {
1788 .ptr.compat = __envp,
1790 return do_execveat_common(fd, filename, argv, envp, flags);
1794 void set_binfmt(struct linux_binfmt *new)
1796 struct mm_struct *mm = current->mm;
1799 module_put(mm->binfmt->module);
1803 __module_get(new->module);
1805 EXPORT_SYMBOL(set_binfmt);
1808 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1810 void set_dumpable(struct mm_struct *mm, int value)
1812 unsigned long old, new;
1814 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1818 old = ACCESS_ONCE(mm->flags);
1819 new = (old & ~MMF_DUMPABLE_MASK) | value;
1820 } while (cmpxchg(&mm->flags, old, new) != old);
1823 SYSCALL_DEFINE3(execve,
1824 const char __user *, filename,
1825 const char __user *const __user *, argv,
1826 const char __user *const __user *, envp)
1828 return do_execve(getname(filename), argv, envp);
1831 SYSCALL_DEFINE5(execveat,
1832 int, fd, const char __user *, filename,
1833 const char __user *const __user *, argv,
1834 const char __user *const __user *, envp,
1837 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1839 return do_execveat(fd,
1840 getname_flags(filename, lookup_flags, NULL),
1844 #ifdef CONFIG_COMPAT
1845 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1846 const compat_uptr_t __user *, argv,
1847 const compat_uptr_t __user *, envp)
1849 return compat_do_execve(getname(filename), argv, envp);
1852 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1853 const char __user *, filename,
1854 const compat_uptr_t __user *, argv,
1855 const compat_uptr_t __user *, envp,
1858 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1860 return compat_do_execveat(fd,
1861 getname_flags(filename, lookup_flags, NULL),