1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * #!-checking implemented by tytso.
12 * Demand-loading implemented 01.12.91 - no need to read anything but
13 * the header into memory. The inode of the executable is put into
14 * "current->executable", and page faults do the actual loading. Clean.
16 * Once more I can proudly say that linux stood up to being changed: it
17 * was less than 2 hours work to get demand-loading completely implemented.
19 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
20 * current->executable is only used by the procfs. This allows a dispatch
21 * table to check for several different types of binary formats. We keep
22 * trying until we recognize the file or we run out of supported binary
26 #include <linux/slab.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
30 #include <linux/vmacache.h>
31 #include <linux/stat.h>
32 #include <linux/fcntl.h>
33 #include <linux/swap.h>
34 #include <linux/string.h>
35 #include <linux/init.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/coredump.h>
38 #include <linux/sched/signal.h>
39 #include <linux/sched/numa_balancing.h>
40 #include <linux/sched/task.h>
41 #include <linux/pagemap.h>
42 #include <linux/perf_event.h>
43 #include <linux/highmem.h>
44 #include <linux/spinlock.h>
45 #include <linux/key.h>
46 #include <linux/personality.h>
47 #include <linux/binfmts.h>
48 #include <linux/utsname.h>
49 #include <linux/pid_namespace.h>
50 #include <linux/module.h>
51 #include <linux/namei.h>
52 #include <linux/mount.h>
53 #include <linux/security.h>
54 #include <linux/syscalls.h>
55 #include <linux/tsacct_kern.h>
56 #include <linux/cn_proc.h>
57 #include <linux/audit.h>
58 #include <linux/tracehook.h>
59 #include <linux/kmod.h>
60 #include <linux/fsnotify.h>
61 #include <linux/fs_struct.h>
62 #include <linux/oom.h>
63 #include <linux/compat.h>
64 #include <linux/vmalloc.h>
66 #include <linux/uaccess.h>
67 #include <asm/mmu_context.h>
70 #include <trace/events/task.h>
73 #include <trace/events/sched.h>
75 int suid_dumpable = 0;
77 static LIST_HEAD(formats);
78 static DEFINE_RWLOCK(binfmt_lock);
80 void __register_binfmt(struct linux_binfmt * fmt, int insert)
83 if (WARN_ON(!fmt->load_binary))
85 write_lock(&binfmt_lock);
86 insert ? list_add(&fmt->lh, &formats) :
87 list_add_tail(&fmt->lh, &formats);
88 write_unlock(&binfmt_lock);
91 EXPORT_SYMBOL(__register_binfmt);
93 void unregister_binfmt(struct linux_binfmt * fmt)
95 write_lock(&binfmt_lock);
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);
107 bool path_noexec(const struct path *path)
109 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
110 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
115 * Note that a shared library must be both readable and executable due to
118 * Also note that we take the address to load from from the file itself.
120 SYSCALL_DEFINE1(uselib, const char __user *, library)
122 struct linux_binfmt *fmt;
124 struct filename *tmp = getname(library);
125 int error = PTR_ERR(tmp);
126 static const struct open_flags uselib_flags = {
127 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
128 .acc_mode = MAY_READ | MAY_EXEC,
129 .intent = LOOKUP_OPEN,
130 .lookup_flags = LOOKUP_FOLLOW,
136 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
138 error = PTR_ERR(file);
143 if (!S_ISREG(file_inode(file)->i_mode))
147 if (path_noexec(&file->f_path))
154 read_lock(&binfmt_lock);
155 list_for_each_entry(fmt, &formats, lh) {
156 if (!fmt->load_shlib)
158 if (!try_module_get(fmt->module))
160 read_unlock(&binfmt_lock);
161 error = fmt->load_shlib(file);
162 read_lock(&binfmt_lock);
164 if (error != -ENOEXEC)
167 read_unlock(&binfmt_lock);
173 #endif /* #ifdef CONFIG_USELIB */
177 * The nascent bprm->mm is not visible until exec_mmap() but it can
178 * use a lot of memory, account these pages in current->mm temporary
179 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
180 * change the counter back via acct_arg_size(0).
182 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
184 struct mm_struct *mm = current->mm;
185 long diff = (long)(pages - bprm->vma_pages);
190 bprm->vma_pages = pages;
191 add_mm_counter(mm, MM_ANONPAGES, diff);
194 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
199 unsigned int gup_flags = FOLL_FORCE;
201 #ifdef CONFIG_STACK_GROWSUP
203 ret = expand_downwards(bprm->vma, pos);
210 gup_flags |= FOLL_WRITE;
213 * We are doing an exec(). 'current' is the process
214 * doing the exec and bprm->mm is the new process's mm.
216 ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
222 acct_arg_size(bprm, vma_pages(bprm->vma));
227 static void put_arg_page(struct page *page)
232 static void free_arg_pages(struct linux_binprm *bprm)
236 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
239 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
242 static int __bprm_mm_init(struct linux_binprm *bprm)
245 struct vm_area_struct *vma = NULL;
246 struct mm_struct *mm = bprm->mm;
248 bprm->vma = vma = vm_area_alloc(mm);
251 vma_set_anonymous(vma);
253 if (down_write_killable(&mm->mmap_sem)) {
259 * Place the stack at the largest stack address the architecture
260 * supports. Later, we'll move this to an appropriate place. We don't
261 * use STACK_TOP because that can depend on attributes which aren't
264 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
265 vma->vm_end = STACK_TOP_MAX;
266 vma->vm_start = vma->vm_end - PAGE_SIZE;
267 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
268 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
270 err = insert_vm_struct(mm, vma);
274 mm->stack_vm = mm->total_vm = 1;
275 arch_bprm_mm_init(mm, vma);
276 up_write(&mm->mmap_sem);
277 bprm->p = vma->vm_end - sizeof(void *);
280 up_write(&mm->mmap_sem);
287 static bool valid_arg_len(struct linux_binprm *bprm, long len)
289 return len <= MAX_ARG_STRLEN;
294 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
298 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
303 page = bprm->page[pos / PAGE_SIZE];
304 if (!page && write) {
305 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
308 bprm->page[pos / PAGE_SIZE] = page;
314 static void put_arg_page(struct page *page)
318 static void free_arg_page(struct linux_binprm *bprm, int i)
321 __free_page(bprm->page[i]);
322 bprm->page[i] = NULL;
326 static void free_arg_pages(struct linux_binprm *bprm)
330 for (i = 0; i < MAX_ARG_PAGES; i++)
331 free_arg_page(bprm, i);
334 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
339 static int __bprm_mm_init(struct linux_binprm *bprm)
341 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
345 static bool valid_arg_len(struct linux_binprm *bprm, long len)
347 return len <= bprm->p;
350 #endif /* CONFIG_MMU */
353 * Create a new mm_struct and populate it with a temporary stack
354 * vm_area_struct. We don't have enough context at this point to set the stack
355 * flags, permissions, and offset, so we use temporary values. We'll update
356 * them later in setup_arg_pages().
358 static int bprm_mm_init(struct linux_binprm *bprm)
361 struct mm_struct *mm = NULL;
363 bprm->mm = mm = mm_alloc();
368 /* Save current stack limit for all calculations made during exec. */
369 task_lock(current->group_leader);
370 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
371 task_unlock(current->group_leader);
373 err = __bprm_mm_init(bprm);
388 struct user_arg_ptr {
393 const char __user *const __user *native;
395 const compat_uptr_t __user *compat;
400 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
402 const char __user *native;
405 if (unlikely(argv.is_compat)) {
406 compat_uptr_t compat;
408 if (get_user(compat, argv.ptr.compat + nr))
409 return ERR_PTR(-EFAULT);
411 return compat_ptr(compat);
415 if (get_user(native, argv.ptr.native + nr))
416 return ERR_PTR(-EFAULT);
422 * count() counts the number of strings in array ARGV.
424 static int count(struct user_arg_ptr argv, int max)
428 if (argv.ptr.native != NULL) {
430 const char __user *p = get_user_arg_ptr(argv, i);
442 if (fatal_signal_pending(current))
443 return -ERESTARTNOHAND;
450 static int prepare_arg_pages(struct linux_binprm *bprm,
451 struct user_arg_ptr argv, struct user_arg_ptr envp)
453 unsigned long limit, ptr_size;
455 bprm->argc = count(argv, MAX_ARG_STRINGS);
459 bprm->envc = count(envp, MAX_ARG_STRINGS);
464 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
465 * (whichever is smaller) for the argv+env strings.
467 * - the remaining binfmt code will not run out of stack space,
468 * - the program will have a reasonable amount of stack left
471 limit = _STK_LIM / 4 * 3;
472 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
474 * We've historically supported up to 32 pages (ARG_MAX)
475 * of argument strings even with small stacks
477 limit = max_t(unsigned long, limit, ARG_MAX);
479 * We must account for the size of all the argv and envp pointers to
480 * the argv and envp strings, since they will also take up space in
481 * the stack. They aren't stored until much later when we can't
482 * signal to the parent that the child has run out of stack space.
483 * Instead, calculate it here so it's possible to fail gracefully.
485 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
486 if (limit <= ptr_size)
490 bprm->argmin = bprm->p - limit;
495 * 'copy_strings()' copies argument/environment strings from the old
496 * processes's memory to the new process's stack. The call to get_user_pages()
497 * ensures the destination page is created and not swapped out.
499 static int copy_strings(int argc, struct user_arg_ptr argv,
500 struct linux_binprm *bprm)
502 struct page *kmapped_page = NULL;
504 unsigned long kpos = 0;
508 const char __user *str;
513 str = get_user_arg_ptr(argv, argc);
517 len = strnlen_user(str, MAX_ARG_STRLEN);
522 if (!valid_arg_len(bprm, len))
525 /* We're going to work our way backwords. */
530 if (bprm->p < bprm->argmin)
535 int offset, bytes_to_copy;
537 if (fatal_signal_pending(current)) {
538 ret = -ERESTARTNOHAND;
543 offset = pos % PAGE_SIZE;
547 bytes_to_copy = offset;
548 if (bytes_to_copy > len)
551 offset -= bytes_to_copy;
552 pos -= bytes_to_copy;
553 str -= bytes_to_copy;
554 len -= bytes_to_copy;
556 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
559 page = get_arg_page(bprm, pos, 1);
566 flush_kernel_dcache_page(kmapped_page);
567 kunmap(kmapped_page);
568 put_arg_page(kmapped_page);
571 kaddr = kmap(kmapped_page);
572 kpos = pos & PAGE_MASK;
573 flush_arg_page(bprm, kpos, kmapped_page);
575 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
584 flush_kernel_dcache_page(kmapped_page);
585 kunmap(kmapped_page);
586 put_arg_page(kmapped_page);
592 * Like copy_strings, but get argv and its values from kernel memory.
594 int copy_strings_kernel(int argc, const char *const *__argv,
595 struct linux_binprm *bprm)
598 mm_segment_t oldfs = get_fs();
599 struct user_arg_ptr argv = {
600 .ptr.native = (const char __user *const __user *)__argv,
604 r = copy_strings(argc, argv, bprm);
609 EXPORT_SYMBOL(copy_strings_kernel);
614 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
615 * the binfmt code determines where the new stack should reside, we shift it to
616 * its final location. The process proceeds as follows:
618 * 1) Use shift to calculate the new vma endpoints.
619 * 2) Extend vma to cover both the old and new ranges. This ensures the
620 * arguments passed to subsequent functions are consistent.
621 * 3) Move vma's page tables to the new range.
622 * 4) Free up any cleared pgd range.
623 * 5) Shrink the vma to cover only the new range.
625 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
627 struct mm_struct *mm = vma->vm_mm;
628 unsigned long old_start = vma->vm_start;
629 unsigned long old_end = vma->vm_end;
630 unsigned long length = old_end - old_start;
631 unsigned long new_start = old_start - shift;
632 unsigned long new_end = old_end - shift;
633 struct mmu_gather tlb;
635 BUG_ON(new_start > new_end);
638 * ensure there are no vmas between where we want to go
641 if (vma != find_vma(mm, new_start))
645 * cover the whole range: [new_start, old_end)
647 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
651 * move the page tables downwards, on failure we rely on
652 * process cleanup to remove whatever mess we made.
654 if (length != move_page_tables(vma, old_start,
655 vma, new_start, length, false))
659 tlb_gather_mmu(&tlb, mm, old_start, old_end);
660 if (new_end > old_start) {
662 * when the old and new regions overlap clear from new_end.
664 free_pgd_range(&tlb, new_end, old_end, new_end,
665 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
668 * otherwise, clean from old_start; this is done to not touch
669 * the address space in [new_end, old_start) some architectures
670 * have constraints on va-space that make this illegal (IA64) -
671 * for the others its just a little faster.
673 free_pgd_range(&tlb, old_start, old_end, new_end,
674 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
676 tlb_finish_mmu(&tlb, old_start, old_end);
679 * Shrink the vma to just the new range. Always succeeds.
681 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
687 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
688 * the stack is optionally relocated, and some extra space is added.
690 int setup_arg_pages(struct linux_binprm *bprm,
691 unsigned long stack_top,
692 int executable_stack)
695 unsigned long stack_shift;
696 struct mm_struct *mm = current->mm;
697 struct vm_area_struct *vma = bprm->vma;
698 struct vm_area_struct *prev = NULL;
699 unsigned long vm_flags;
700 unsigned long stack_base;
701 unsigned long stack_size;
702 unsigned long stack_expand;
703 unsigned long rlim_stack;
705 #ifdef CONFIG_STACK_GROWSUP
706 /* Limit stack size */
707 stack_base = bprm->rlim_stack.rlim_max;
708 if (stack_base > STACK_SIZE_MAX)
709 stack_base = STACK_SIZE_MAX;
711 /* Add space for stack randomization. */
712 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
714 /* Make sure we didn't let the argument array grow too large. */
715 if (vma->vm_end - vma->vm_start > stack_base)
718 stack_base = PAGE_ALIGN(stack_top - stack_base);
720 stack_shift = vma->vm_start - stack_base;
721 mm->arg_start = bprm->p - stack_shift;
722 bprm->p = vma->vm_end - stack_shift;
724 stack_top = arch_align_stack(stack_top);
725 stack_top = PAGE_ALIGN(stack_top);
727 if (unlikely(stack_top < mmap_min_addr) ||
728 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
731 stack_shift = vma->vm_end - stack_top;
733 bprm->p -= stack_shift;
734 mm->arg_start = bprm->p;
738 bprm->loader -= stack_shift;
739 bprm->exec -= stack_shift;
741 if (down_write_killable(&mm->mmap_sem))
744 vm_flags = VM_STACK_FLAGS;
747 * Adjust stack execute permissions; explicitly enable for
748 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
749 * (arch default) otherwise.
751 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
753 else if (executable_stack == EXSTACK_DISABLE_X)
754 vm_flags &= ~VM_EXEC;
755 vm_flags |= mm->def_flags;
756 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
758 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
764 /* Move stack pages down in memory. */
766 ret = shift_arg_pages(vma, stack_shift);
771 /* mprotect_fixup is overkill to remove the temporary stack flags */
772 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
774 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
775 stack_size = vma->vm_end - vma->vm_start;
777 * Align this down to a page boundary as expand_stack
780 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
781 #ifdef CONFIG_STACK_GROWSUP
782 if (stack_size + stack_expand > rlim_stack)
783 stack_base = vma->vm_start + rlim_stack;
785 stack_base = vma->vm_end + stack_expand;
787 if (stack_size + stack_expand > rlim_stack)
788 stack_base = vma->vm_end - rlim_stack;
790 stack_base = vma->vm_start - stack_expand;
792 current->mm->start_stack = bprm->p;
793 ret = expand_stack(vma, stack_base);
798 up_write(&mm->mmap_sem);
801 EXPORT_SYMBOL(setup_arg_pages);
806 * Transfer the program arguments and environment from the holding pages
807 * onto the stack. The provided stack pointer is adjusted accordingly.
809 int transfer_args_to_stack(struct linux_binprm *bprm,
810 unsigned long *sp_location)
812 unsigned long index, stop, sp;
815 stop = bprm->p >> PAGE_SHIFT;
818 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
819 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
820 char *src = kmap(bprm->page[index]) + offset;
821 sp -= PAGE_SIZE - offset;
822 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
824 kunmap(bprm->page[index]);
834 EXPORT_SYMBOL(transfer_args_to_stack);
836 #endif /* CONFIG_MMU */
838 static struct file *do_open_execat(int fd, struct filename *name, int flags)
842 struct open_flags open_exec_flags = {
843 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
844 .acc_mode = MAY_EXEC,
845 .intent = LOOKUP_OPEN,
846 .lookup_flags = LOOKUP_FOLLOW,
849 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
850 return ERR_PTR(-EINVAL);
851 if (flags & AT_SYMLINK_NOFOLLOW)
852 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
853 if (flags & AT_EMPTY_PATH)
854 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
856 file = do_filp_open(fd, name, &open_exec_flags);
861 if (!S_ISREG(file_inode(file)->i_mode))
864 if (path_noexec(&file->f_path))
867 err = deny_write_access(file);
871 if (name->name[0] != '\0')
882 struct file *open_exec(const char *name)
884 struct filename *filename = getname_kernel(name);
885 struct file *f = ERR_CAST(filename);
887 if (!IS_ERR(filename)) {
888 f = do_open_execat(AT_FDCWD, filename, 0);
893 EXPORT_SYMBOL(open_exec);
895 int kernel_read_file(struct file *file, void **buf, loff_t *size,
896 loff_t max_size, enum kernel_read_file_id id)
902 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
905 ret = deny_write_access(file);
909 ret = security_kernel_read_file(file, id);
913 i_size = i_size_read(file_inode(file));
918 if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
923 if (id != READING_FIRMWARE_PREALLOC_BUFFER)
924 *buf = vmalloc(i_size);
931 while (pos < i_size) {
932 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
947 ret = security_kernel_post_read_file(file, *buf, i_size, id);
953 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
960 allow_write_access(file);
963 EXPORT_SYMBOL_GPL(kernel_read_file);
965 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
966 loff_t max_size, enum kernel_read_file_id id)
974 file = filp_open(path, O_RDONLY, 0);
976 return PTR_ERR(file);
978 ret = kernel_read_file(file, buf, size, max_size, id);
982 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
984 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
985 enum kernel_read_file_id id)
987 struct fd f = fdget(fd);
993 ret = kernel_read_file(f.file, buf, size, max_size, id);
998 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
1000 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1002 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1004 flush_icache_range(addr, addr + len);
1007 EXPORT_SYMBOL(read_code);
1009 static int exec_mmap(struct mm_struct *mm)
1011 struct task_struct *tsk;
1012 struct mm_struct *old_mm, *active_mm;
1014 /* Notify parent that we're no longer interested in the old VM */
1016 old_mm = current->mm;
1017 exec_mm_release(tsk, old_mm);
1020 sync_mm_rss(old_mm);
1022 * Make sure that if there is a core dump in progress
1023 * for the old mm, we get out and die instead of going
1024 * through with the exec. We must hold mmap_sem around
1025 * checking core_state and changing tsk->mm.
1027 down_read(&old_mm->mmap_sem);
1028 if (unlikely(old_mm->core_state)) {
1029 up_read(&old_mm->mmap_sem);
1034 active_mm = tsk->active_mm;
1035 membarrier_exec_mmap(mm);
1037 tsk->active_mm = mm;
1038 activate_mm(active_mm, mm);
1039 tsk->mm->vmacache_seqnum = 0;
1040 vmacache_flush(tsk);
1043 up_read(&old_mm->mmap_sem);
1044 BUG_ON(active_mm != old_mm);
1045 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1046 mm_update_next_owner(old_mm);
1055 * This function makes sure the current process has its own signal table,
1056 * so that flush_signal_handlers can later reset the handlers without
1057 * disturbing other processes. (Other processes might share the signal
1058 * table via the CLONE_SIGHAND option to clone().)
1060 static int de_thread(struct task_struct *tsk)
1062 struct signal_struct *sig = tsk->signal;
1063 struct sighand_struct *oldsighand = tsk->sighand;
1064 spinlock_t *lock = &oldsighand->siglock;
1066 if (thread_group_empty(tsk))
1067 goto no_thread_group;
1070 * Kill all other threads in the thread group.
1072 spin_lock_irq(lock);
1073 if (signal_group_exit(sig)) {
1075 * Another group action in progress, just
1076 * return so that the signal is processed.
1078 spin_unlock_irq(lock);
1082 sig->group_exit_task = tsk;
1083 sig->notify_count = zap_other_threads(tsk);
1084 if (!thread_group_leader(tsk))
1085 sig->notify_count--;
1087 while (sig->notify_count) {
1088 __set_current_state(TASK_KILLABLE);
1089 spin_unlock_irq(lock);
1091 if (__fatal_signal_pending(tsk))
1093 spin_lock_irq(lock);
1095 spin_unlock_irq(lock);
1098 * At this point all other threads have exited, all we have to
1099 * do is to wait for the thread group leader to become inactive,
1100 * and to assume its PID:
1102 if (!thread_group_leader(tsk)) {
1103 struct task_struct *leader = tsk->group_leader;
1106 cgroup_threadgroup_change_begin(tsk);
1107 write_lock_irq(&tasklist_lock);
1109 * Do this under tasklist_lock to ensure that
1110 * exit_notify() can't miss ->group_exit_task
1112 sig->notify_count = -1;
1113 if (likely(leader->exit_state))
1115 __set_current_state(TASK_KILLABLE);
1116 write_unlock_irq(&tasklist_lock);
1117 cgroup_threadgroup_change_end(tsk);
1119 if (__fatal_signal_pending(tsk))
1124 * The only record we have of the real-time age of a
1125 * process, regardless of execs it's done, is start_time.
1126 * All the past CPU time is accumulated in signal_struct
1127 * from sister threads now dead. But in this non-leader
1128 * exec, nothing survives from the original leader thread,
1129 * whose birth marks the true age of this process now.
1130 * When we take on its identity by switching to its PID, we
1131 * also take its birthdate (always earlier than our own).
1133 tsk->start_time = leader->start_time;
1134 tsk->start_boottime = leader->start_boottime;
1136 BUG_ON(!same_thread_group(leader, tsk));
1137 BUG_ON(has_group_leader_pid(tsk));
1139 * An exec() starts a new thread group with the
1140 * TGID of the previous thread group. Rehash the
1141 * two threads with a switched PID, and release
1142 * the former thread group leader:
1145 /* Become a process group leader with the old leader's pid.
1146 * The old leader becomes a thread of the this thread group.
1147 * Note: The old leader also uses this pid until release_task
1148 * is called. Odd but simple and correct.
1150 tsk->pid = leader->pid;
1151 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1152 transfer_pid(leader, tsk, PIDTYPE_TGID);
1153 transfer_pid(leader, tsk, PIDTYPE_PGID);
1154 transfer_pid(leader, tsk, PIDTYPE_SID);
1156 list_replace_rcu(&leader->tasks, &tsk->tasks);
1157 list_replace_init(&leader->sibling, &tsk->sibling);
1159 tsk->group_leader = tsk;
1160 leader->group_leader = tsk;
1162 tsk->exit_signal = SIGCHLD;
1163 leader->exit_signal = -1;
1165 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1166 leader->exit_state = EXIT_DEAD;
1169 * We are going to release_task()->ptrace_unlink() silently,
1170 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1171 * the tracer wont't block again waiting for this thread.
1173 if (unlikely(leader->ptrace))
1174 __wake_up_parent(leader, leader->parent);
1175 write_unlock_irq(&tasklist_lock);
1176 cgroup_threadgroup_change_end(tsk);
1178 release_task(leader);
1181 sig->group_exit_task = NULL;
1182 sig->notify_count = 0;
1185 /* we have changed execution domain */
1186 tsk->exit_signal = SIGCHLD;
1188 #ifdef CONFIG_POSIX_TIMERS
1190 flush_itimer_signals();
1193 if (refcount_read(&oldsighand->count) != 1) {
1194 struct sighand_struct *newsighand;
1196 * This ->sighand is shared with the CLONE_SIGHAND
1197 * but not CLONE_THREAD task, switch to the new one.
1199 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1203 refcount_set(&newsighand->count, 1);
1204 memcpy(newsighand->action, oldsighand->action,
1205 sizeof(newsighand->action));
1207 write_lock_irq(&tasklist_lock);
1208 spin_lock(&oldsighand->siglock);
1209 rcu_assign_pointer(tsk->sighand, newsighand);
1210 spin_unlock(&oldsighand->siglock);
1211 write_unlock_irq(&tasklist_lock);
1213 __cleanup_sighand(oldsighand);
1216 BUG_ON(!thread_group_leader(tsk));
1220 /* protects against exit_notify() and __exit_signal() */
1221 read_lock(&tasklist_lock);
1222 sig->group_exit_task = NULL;
1223 sig->notify_count = 0;
1224 read_unlock(&tasklist_lock);
1228 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1231 strncpy(buf, tsk->comm, buf_size);
1235 EXPORT_SYMBOL_GPL(__get_task_comm);
1238 * These functions flushes out all traces of the currently running executable
1239 * so that a new one can be started
1242 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1245 trace_task_rename(tsk, buf);
1246 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1248 perf_event_comm(tsk, exec);
1252 * Calling this is the point of no return. None of the failures will be
1253 * seen by userspace since either the process is already taking a fatal
1254 * signal (via de_thread() or coredump), or will have SEGV raised
1255 * (after exec_mmap()) by search_binary_handlers (see below).
1257 int flush_old_exec(struct linux_binprm * bprm)
1262 * Make sure we have a private signal table and that
1263 * we are unassociated from the previous thread group.
1265 retval = de_thread(current);
1270 * Must be called _before_ exec_mmap() as bprm->mm is
1271 * not visibile until then. This also enables the update
1274 set_mm_exe_file(bprm->mm, bprm->file);
1277 * Release all of the old mmap stuff
1279 acct_arg_size(bprm, 0);
1280 retval = exec_mmap(bprm->mm);
1285 * After clearing bprm->mm (to mark that current is using the
1286 * prepared mm now), we have nothing left of the original
1287 * process. If anything from here on returns an error, the check
1288 * in search_binary_handler() will SEGV current.
1293 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1294 PF_NOFREEZE | PF_NO_SETAFFINITY);
1296 current->personality &= ~bprm->per_clear;
1299 * We have to apply CLOEXEC before we change whether the process is
1300 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1301 * trying to access the should-be-closed file descriptors of a process
1302 * undergoing exec(2).
1304 do_close_on_exec(current->files);
1310 EXPORT_SYMBOL(flush_old_exec);
1312 void would_dump(struct linux_binprm *bprm, struct file *file)
1314 struct inode *inode = file_inode(file);
1315 if (inode_permission(inode, MAY_READ) < 0) {
1316 struct user_namespace *old, *user_ns;
1317 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1319 /* Ensure mm->user_ns contains the executable */
1320 user_ns = old = bprm->mm->user_ns;
1321 while ((user_ns != &init_user_ns) &&
1322 !privileged_wrt_inode_uidgid(user_ns, inode))
1323 user_ns = user_ns->parent;
1325 if (old != user_ns) {
1326 bprm->mm->user_ns = get_user_ns(user_ns);
1331 EXPORT_SYMBOL(would_dump);
1333 void setup_new_exec(struct linux_binprm * bprm)
1336 * Once here, prepare_binrpm() will not be called any more, so
1337 * the final state of setuid/setgid/fscaps can be merged into the
1340 bprm->secureexec |= bprm->cap_elevated;
1342 if (bprm->secureexec) {
1343 /* Make sure parent cannot signal privileged process. */
1344 current->pdeath_signal = 0;
1347 * For secureexec, reset the stack limit to sane default to
1348 * avoid bad behavior from the prior rlimits. This has to
1349 * happen before arch_pick_mmap_layout(), which examines
1350 * RLIMIT_STACK, but after the point of no return to avoid
1351 * needing to clean up the change on failure.
1353 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1354 bprm->rlim_stack.rlim_cur = _STK_LIM;
1357 arch_pick_mmap_layout(current->mm, &bprm->rlim_stack);
1359 current->sas_ss_sp = current->sas_ss_size = 0;
1362 * Figure out dumpability. Note that this checking only of current
1363 * is wrong, but userspace depends on it. This should be testing
1364 * bprm->secureexec instead.
1366 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1367 !(uid_eq(current_euid(), current_uid()) &&
1368 gid_eq(current_egid(), current_gid())))
1369 set_dumpable(current->mm, suid_dumpable);
1371 set_dumpable(current->mm, SUID_DUMP_USER);
1373 arch_setup_new_exec();
1375 __set_task_comm(current, kbasename(bprm->filename), true);
1377 /* Set the new mm task size. We have to do that late because it may
1378 * depend on TIF_32BIT which is only updated in flush_thread() on
1379 * some architectures like powerpc
1381 current->mm->task_size = TASK_SIZE;
1383 /* An exec changes our domain. We are no longer part of the thread
1385 current->self_exec_id++;
1386 flush_signal_handlers(current, 0);
1388 EXPORT_SYMBOL(setup_new_exec);
1390 /* Runs immediately before start_thread() takes over. */
1391 void finalize_exec(struct linux_binprm *bprm)
1393 /* Store any stack rlimit changes before starting thread. */
1394 task_lock(current->group_leader);
1395 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1396 task_unlock(current->group_leader);
1398 EXPORT_SYMBOL(finalize_exec);
1401 * Prepare credentials and lock ->cred_guard_mutex.
1402 * install_exec_creds() commits the new creds and drops the lock.
1403 * Or, if exec fails before, free_bprm() should release ->cred and
1406 static int prepare_bprm_creds(struct linux_binprm *bprm)
1408 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1409 return -ERESTARTNOINTR;
1411 bprm->cred = prepare_exec_creds();
1412 if (likely(bprm->cred))
1415 mutex_unlock(¤t->signal->cred_guard_mutex);
1419 static void free_bprm(struct linux_binprm *bprm)
1421 free_arg_pages(bprm);
1423 mutex_unlock(¤t->signal->cred_guard_mutex);
1424 abort_creds(bprm->cred);
1427 allow_write_access(bprm->file);
1430 /* If a binfmt changed the interp, free it. */
1431 if (bprm->interp != bprm->filename)
1432 kfree(bprm->interp);
1436 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1438 /* If a binfmt changed the interp, free it first. */
1439 if (bprm->interp != bprm->filename)
1440 kfree(bprm->interp);
1441 bprm->interp = kstrdup(interp, GFP_KERNEL);
1446 EXPORT_SYMBOL(bprm_change_interp);
1449 * install the new credentials for this executable
1451 void install_exec_creds(struct linux_binprm *bprm)
1453 security_bprm_committing_creds(bprm);
1455 commit_creds(bprm->cred);
1459 * Disable monitoring for regular users
1460 * when executing setuid binaries. Must
1461 * wait until new credentials are committed
1462 * by commit_creds() above
1464 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1465 perf_event_exit_task(current);
1467 * cred_guard_mutex must be held at least to this point to prevent
1468 * ptrace_attach() from altering our determination of the task's
1469 * credentials; any time after this it may be unlocked.
1471 security_bprm_committed_creds(bprm);
1472 mutex_unlock(¤t->signal->cred_guard_mutex);
1474 EXPORT_SYMBOL(install_exec_creds);
1477 * determine how safe it is to execute the proposed program
1478 * - the caller must hold ->cred_guard_mutex to protect against
1479 * PTRACE_ATTACH or seccomp thread-sync
1481 static void check_unsafe_exec(struct linux_binprm *bprm)
1483 struct task_struct *p = current, *t;
1487 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1490 * This isn't strictly necessary, but it makes it harder for LSMs to
1493 if (task_no_new_privs(current))
1494 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1498 spin_lock(&p->fs->lock);
1500 while_each_thread(p, t) {
1506 if (p->fs->users > n_fs)
1507 bprm->unsafe |= LSM_UNSAFE_SHARE;
1510 spin_unlock(&p->fs->lock);
1513 static void bprm_fill_uid(struct linux_binprm *bprm)
1515 struct inode *inode;
1521 * Since this can be called multiple times (via prepare_binprm),
1522 * we must clear any previous work done when setting set[ug]id
1523 * bits from any earlier bprm->file uses (for example when run
1524 * first for a setuid script then again for its interpreter).
1526 bprm->cred->euid = current_euid();
1527 bprm->cred->egid = current_egid();
1529 if (!mnt_may_suid(bprm->file->f_path.mnt))
1532 if (task_no_new_privs(current))
1535 inode = bprm->file->f_path.dentry->d_inode;
1536 mode = READ_ONCE(inode->i_mode);
1537 if (!(mode & (S_ISUID|S_ISGID)))
1540 /* Be careful if suid/sgid is set */
1543 /* reload atomically mode/uid/gid now that lock held */
1544 mode = inode->i_mode;
1547 inode_unlock(inode);
1549 /* We ignore suid/sgid if there are no mappings for them in the ns */
1550 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1551 !kgid_has_mapping(bprm->cred->user_ns, gid))
1554 if (mode & S_ISUID) {
1555 bprm->per_clear |= PER_CLEAR_ON_SETID;
1556 bprm->cred->euid = uid;
1559 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1560 bprm->per_clear |= PER_CLEAR_ON_SETID;
1561 bprm->cred->egid = gid;
1566 * Fill the binprm structure from the inode.
1567 * Check permissions, then read the first BINPRM_BUF_SIZE bytes
1569 * This may be called multiple times for binary chains (scripts for example).
1571 int prepare_binprm(struct linux_binprm *bprm)
1576 bprm_fill_uid(bprm);
1578 /* fill in binprm security blob */
1579 retval = security_bprm_set_creds(bprm);
1582 bprm->called_set_creds = 1;
1584 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1585 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1588 EXPORT_SYMBOL(prepare_binprm);
1591 * Arguments are '\0' separated strings found at the location bprm->p
1592 * points to; chop off the first by relocating brpm->p to right after
1593 * the first '\0' encountered.
1595 int remove_arg_zero(struct linux_binprm *bprm)
1598 unsigned long offset;
1606 offset = bprm->p & ~PAGE_MASK;
1607 page = get_arg_page(bprm, bprm->p, 0);
1612 kaddr = kmap_atomic(page);
1614 for (; offset < PAGE_SIZE && kaddr[offset];
1615 offset++, bprm->p++)
1618 kunmap_atomic(kaddr);
1620 } while (offset == PAGE_SIZE);
1629 EXPORT_SYMBOL(remove_arg_zero);
1631 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1633 * cycle the list of binary formats handler, until one recognizes the image
1635 int search_binary_handler(struct linux_binprm *bprm)
1637 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1638 struct linux_binfmt *fmt;
1641 /* This allows 4 levels of binfmt rewrites before failing hard. */
1642 if (bprm->recursion_depth > 5)
1645 retval = security_bprm_check(bprm);
1651 read_lock(&binfmt_lock);
1652 list_for_each_entry(fmt, &formats, lh) {
1653 if (!try_module_get(fmt->module))
1655 read_unlock(&binfmt_lock);
1657 bprm->recursion_depth++;
1658 retval = fmt->load_binary(bprm);
1659 bprm->recursion_depth--;
1661 read_lock(&binfmt_lock);
1663 if (retval < 0 && !bprm->mm) {
1664 /* we got to flush_old_exec() and failed after it */
1665 read_unlock(&binfmt_lock);
1666 force_sigsegv(SIGSEGV);
1669 if (retval != -ENOEXEC || !bprm->file) {
1670 read_unlock(&binfmt_lock);
1674 read_unlock(&binfmt_lock);
1677 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1678 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1680 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1688 EXPORT_SYMBOL(search_binary_handler);
1690 static int exec_binprm(struct linux_binprm *bprm)
1692 pid_t old_pid, old_vpid;
1695 /* Need to fetch pid before load_binary changes it */
1696 old_pid = current->pid;
1698 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1701 ret = search_binary_handler(bprm);
1704 trace_sched_process_exec(current, old_pid, bprm);
1705 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1706 proc_exec_connector(current);
1713 * sys_execve() executes a new program.
1715 static int __do_execve_file(int fd, struct filename *filename,
1716 struct user_arg_ptr argv,
1717 struct user_arg_ptr envp,
1718 int flags, struct file *file)
1720 char *pathbuf = NULL;
1721 struct linux_binprm *bprm;
1722 struct files_struct *displaced;
1725 if (IS_ERR(filename))
1726 return PTR_ERR(filename);
1729 * We move the actual failure in case of RLIMIT_NPROC excess from
1730 * set*uid() to execve() because too many poorly written programs
1731 * don't check setuid() return code. Here we additionally recheck
1732 * whether NPROC limit is still exceeded.
1734 if ((current->flags & PF_NPROC_EXCEEDED) &&
1735 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1740 /* We're below the limit (still or again), so we don't want to make
1741 * further execve() calls fail. */
1742 current->flags &= ~PF_NPROC_EXCEEDED;
1744 retval = unshare_files(&displaced);
1749 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1753 retval = prepare_bprm_creds(bprm);
1757 check_unsafe_exec(bprm);
1758 current->in_execve = 1;
1761 file = do_open_execat(fd, filename, flags);
1762 retval = PTR_ERR(file);
1770 bprm->filename = "none";
1771 } else if (fd == AT_FDCWD || filename->name[0] == '/') {
1772 bprm->filename = filename->name;
1774 if (filename->name[0] == '\0')
1775 pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1777 pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1778 fd, filename->name);
1784 * Record that a name derived from an O_CLOEXEC fd will be
1785 * inaccessible after exec. Relies on having exclusive access to
1786 * current->files (due to unshare_files above).
1788 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1789 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1790 bprm->filename = pathbuf;
1792 bprm->interp = bprm->filename;
1794 retval = bprm_mm_init(bprm);
1798 retval = prepare_arg_pages(bprm, argv, envp);
1802 retval = prepare_binprm(bprm);
1806 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1810 bprm->exec = bprm->p;
1811 retval = copy_strings(bprm->envc, envp, bprm);
1815 retval = copy_strings(bprm->argc, argv, bprm);
1819 would_dump(bprm, bprm->file);
1821 retval = exec_binprm(bprm);
1825 /* execve succeeded */
1826 current->fs->in_exec = 0;
1827 current->in_execve = 0;
1828 rseq_execve(current);
1829 acct_update_integrals(current);
1830 task_numa_free(current, false);
1836 put_files_struct(displaced);
1841 acct_arg_size(bprm, 0);
1846 current->fs->in_exec = 0;
1847 current->in_execve = 0;
1855 reset_files_struct(displaced);
1862 static int do_execveat_common(int fd, struct filename *filename,
1863 struct user_arg_ptr argv,
1864 struct user_arg_ptr envp,
1867 return __do_execve_file(fd, filename, argv, envp, flags, NULL);
1870 int do_execve_file(struct file *file, void *__argv, void *__envp)
1872 struct user_arg_ptr argv = { .ptr.native = __argv };
1873 struct user_arg_ptr envp = { .ptr.native = __envp };
1875 return __do_execve_file(AT_FDCWD, NULL, argv, envp, 0, file);
1878 int do_execve(struct filename *filename,
1879 const char __user *const __user *__argv,
1880 const char __user *const __user *__envp)
1882 struct user_arg_ptr argv = { .ptr.native = __argv };
1883 struct user_arg_ptr envp = { .ptr.native = __envp };
1884 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1887 int do_execveat(int fd, struct filename *filename,
1888 const char __user *const __user *__argv,
1889 const char __user *const __user *__envp,
1892 struct user_arg_ptr argv = { .ptr.native = __argv };
1893 struct user_arg_ptr envp = { .ptr.native = __envp };
1895 return do_execveat_common(fd, filename, argv, envp, flags);
1898 #ifdef CONFIG_COMPAT
1899 static int compat_do_execve(struct filename *filename,
1900 const compat_uptr_t __user *__argv,
1901 const compat_uptr_t __user *__envp)
1903 struct user_arg_ptr argv = {
1905 .ptr.compat = __argv,
1907 struct user_arg_ptr envp = {
1909 .ptr.compat = __envp,
1911 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1914 static int compat_do_execveat(int fd, struct filename *filename,
1915 const compat_uptr_t __user *__argv,
1916 const compat_uptr_t __user *__envp,
1919 struct user_arg_ptr argv = {
1921 .ptr.compat = __argv,
1923 struct user_arg_ptr envp = {
1925 .ptr.compat = __envp,
1927 return do_execveat_common(fd, filename, argv, envp, flags);
1931 void set_binfmt(struct linux_binfmt *new)
1933 struct mm_struct *mm = current->mm;
1936 module_put(mm->binfmt->module);
1940 __module_get(new->module);
1942 EXPORT_SYMBOL(set_binfmt);
1945 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1947 void set_dumpable(struct mm_struct *mm, int value)
1949 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1952 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
1955 SYSCALL_DEFINE3(execve,
1956 const char __user *, filename,
1957 const char __user *const __user *, argv,
1958 const char __user *const __user *, envp)
1960 return do_execve(getname(filename), argv, envp);
1963 SYSCALL_DEFINE5(execveat,
1964 int, fd, const char __user *, filename,
1965 const char __user *const __user *, argv,
1966 const char __user *const __user *, envp,
1969 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1971 return do_execveat(fd,
1972 getname_flags(filename, lookup_flags, NULL),
1976 #ifdef CONFIG_COMPAT
1977 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1978 const compat_uptr_t __user *, argv,
1979 const compat_uptr_t __user *, envp)
1981 return compat_do_execve(getname(filename), argv, envp);
1984 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1985 const char __user *, filename,
1986 const compat_uptr_t __user *, argv,
1987 const compat_uptr_t __user *, envp,
1990 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1992 return compat_do_execveat(fd,
1993 getname_flags(filename, lookup_flags, NULL),