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/sched/mm.h>
36 #include <linux/sched/coredump.h>
37 #include <linux/sched/signal.h>
38 #include <linux/sched/numa_balancing.h>
39 #include <linux/sched/task.h>
40 #include <linux/pagemap.h>
41 #include <linux/perf_event.h>
42 #include <linux/highmem.h>
43 #include <linux/spinlock.h>
44 #include <linux/key.h>
45 #include <linux/personality.h>
46 #include <linux/binfmts.h>
47 #include <linux/utsname.h>
48 #include <linux/pid_namespace.h>
49 #include <linux/module.h>
50 #include <linux/namei.h>
51 #include <linux/mount.h>
52 #include <linux/security.h>
53 #include <linux/syscalls.h>
54 #include <linux/tsacct_kern.h>
55 #include <linux/cn_proc.h>
56 #include <linux/audit.h>
57 #include <linux/tracehook.h>
58 #include <linux/kmod.h>
59 #include <linux/fsnotify.h>
60 #include <linux/fs_struct.h>
61 #include <linux/pipe_fs_i.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 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
223 unsigned long ptr_size, limit;
226 * Since the stack will hold pointers to the strings, we
227 * must account for them as well.
229 * The size calculation is the entire vma while each arg page is
230 * built, so each time we get here it's calculating how far it
231 * is currently (rather than each call being just the newly
232 * added size from the arg page). As a result, we need to
233 * always add the entire size of the pointers, so that on the
234 * last call to get_arg_page() we'll actually have the entire
237 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
238 if (ptr_size > ULONG_MAX - size)
242 acct_arg_size(bprm, size / PAGE_SIZE);
245 * We've historically supported up to 32 pages (ARG_MAX)
246 * of argument strings even with small stacks
252 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
253 * (whichever is smaller) for the argv+env strings.
255 * - the remaining binfmt code will not run out of stack space,
256 * - the program will have a reasonable amount of stack left
259 limit = _STK_LIM / 4 * 3;
260 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
272 static void put_arg_page(struct page *page)
277 static void free_arg_pages(struct linux_binprm *bprm)
281 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
284 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
287 static int __bprm_mm_init(struct linux_binprm *bprm)
290 struct vm_area_struct *vma = NULL;
291 struct mm_struct *mm = bprm->mm;
293 bprm->vma = vma = vm_area_alloc(mm);
296 vma_set_anonymous(vma);
298 if (down_write_killable(&mm->mmap_sem)) {
304 * Place the stack at the largest stack address the architecture
305 * supports. Later, we'll move this to an appropriate place. We don't
306 * use STACK_TOP because that can depend on attributes which aren't
309 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
310 vma->vm_end = STACK_TOP_MAX;
311 vma->vm_start = vma->vm_end - PAGE_SIZE;
312 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
313 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
315 err = insert_vm_struct(mm, vma);
319 mm->stack_vm = mm->total_vm = 1;
320 arch_bprm_mm_init(mm, vma);
321 up_write(&mm->mmap_sem);
322 bprm->p = vma->vm_end - sizeof(void *);
325 up_write(&mm->mmap_sem);
332 static bool valid_arg_len(struct linux_binprm *bprm, long len)
334 return len <= MAX_ARG_STRLEN;
339 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
343 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
348 page = bprm->page[pos / PAGE_SIZE];
349 if (!page && write) {
350 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
353 bprm->page[pos / PAGE_SIZE] = page;
359 static void put_arg_page(struct page *page)
363 static void free_arg_page(struct linux_binprm *bprm, int i)
366 __free_page(bprm->page[i]);
367 bprm->page[i] = NULL;
371 static void free_arg_pages(struct linux_binprm *bprm)
375 for (i = 0; i < MAX_ARG_PAGES; i++)
376 free_arg_page(bprm, i);
379 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
384 static int __bprm_mm_init(struct linux_binprm *bprm)
386 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
390 static bool valid_arg_len(struct linux_binprm *bprm, long len)
392 return len <= bprm->p;
395 #endif /* CONFIG_MMU */
398 * Create a new mm_struct and populate it with a temporary stack
399 * vm_area_struct. We don't have enough context at this point to set the stack
400 * flags, permissions, and offset, so we use temporary values. We'll update
401 * them later in setup_arg_pages().
403 static int bprm_mm_init(struct linux_binprm *bprm)
406 struct mm_struct *mm = NULL;
408 bprm->mm = mm = mm_alloc();
413 /* Save current stack limit for all calculations made during exec. */
414 task_lock(current->group_leader);
415 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
416 task_unlock(current->group_leader);
418 err = __bprm_mm_init(bprm);
433 struct user_arg_ptr {
438 const char __user *const __user *native;
440 const compat_uptr_t __user *compat;
445 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
447 const char __user *native;
450 if (unlikely(argv.is_compat)) {
451 compat_uptr_t compat;
453 if (get_user(compat, argv.ptr.compat + nr))
454 return ERR_PTR(-EFAULT);
456 return compat_ptr(compat);
460 if (get_user(native, argv.ptr.native + nr))
461 return ERR_PTR(-EFAULT);
467 * count() counts the number of strings in array ARGV.
469 static int count(struct user_arg_ptr argv, int max)
473 if (argv.ptr.native != NULL) {
475 const char __user *p = get_user_arg_ptr(argv, i);
487 if (fatal_signal_pending(current))
488 return -ERESTARTNOHAND;
496 * 'copy_strings()' copies argument/environment strings from the old
497 * processes's memory to the new process's stack. The call to get_user_pages()
498 * ensures the destination page is created and not swapped out.
500 static int copy_strings(int argc, struct user_arg_ptr argv,
501 struct linux_binprm *bprm)
503 struct page *kmapped_page = NULL;
505 unsigned long kpos = 0;
509 const char __user *str;
514 str = get_user_arg_ptr(argv, argc);
518 len = strnlen_user(str, MAX_ARG_STRLEN);
523 if (!valid_arg_len(bprm, len))
526 /* We're going to work our way backwords. */
532 int offset, bytes_to_copy;
534 if (fatal_signal_pending(current)) {
535 ret = -ERESTARTNOHAND;
540 offset = pos % PAGE_SIZE;
544 bytes_to_copy = offset;
545 if (bytes_to_copy > len)
548 offset -= bytes_to_copy;
549 pos -= bytes_to_copy;
550 str -= bytes_to_copy;
551 len -= bytes_to_copy;
553 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
556 page = get_arg_page(bprm, pos, 1);
563 flush_kernel_dcache_page(kmapped_page);
564 kunmap(kmapped_page);
565 put_arg_page(kmapped_page);
568 kaddr = kmap(kmapped_page);
569 kpos = pos & PAGE_MASK;
570 flush_arg_page(bprm, kpos, kmapped_page);
572 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
581 flush_kernel_dcache_page(kmapped_page);
582 kunmap(kmapped_page);
583 put_arg_page(kmapped_page);
589 * Like copy_strings, but get argv and its values from kernel memory.
591 int copy_strings_kernel(int argc, const char *const *__argv,
592 struct linux_binprm *bprm)
595 mm_segment_t oldfs = get_fs();
596 struct user_arg_ptr argv = {
597 .ptr.native = (const char __user *const __user *)__argv,
601 r = copy_strings(argc, argv, bprm);
606 EXPORT_SYMBOL(copy_strings_kernel);
611 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
612 * the binfmt code determines where the new stack should reside, we shift it to
613 * its final location. The process proceeds as follows:
615 * 1) Use shift to calculate the new vma endpoints.
616 * 2) Extend vma to cover both the old and new ranges. This ensures the
617 * arguments passed to subsequent functions are consistent.
618 * 3) Move vma's page tables to the new range.
619 * 4) Free up any cleared pgd range.
620 * 5) Shrink the vma to cover only the new range.
622 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
624 struct mm_struct *mm = vma->vm_mm;
625 unsigned long old_start = vma->vm_start;
626 unsigned long old_end = vma->vm_end;
627 unsigned long length = old_end - old_start;
628 unsigned long new_start = old_start - shift;
629 unsigned long new_end = old_end - shift;
630 struct mmu_gather tlb;
632 BUG_ON(new_start > new_end);
635 * ensure there are no vmas between where we want to go
638 if (vma != find_vma(mm, new_start))
642 * cover the whole range: [new_start, old_end)
644 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
648 * move the page tables downwards, on failure we rely on
649 * process cleanup to remove whatever mess we made.
651 if (length != move_page_tables(vma, old_start,
652 vma, new_start, length, false))
656 tlb_gather_mmu(&tlb, mm, old_start, old_end);
657 if (new_end > old_start) {
659 * when the old and new regions overlap clear from new_end.
661 free_pgd_range(&tlb, new_end, old_end, new_end,
662 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
665 * otherwise, clean from old_start; this is done to not touch
666 * the address space in [new_end, old_start) some architectures
667 * have constraints on va-space that make this illegal (IA64) -
668 * for the others its just a little faster.
670 free_pgd_range(&tlb, old_start, old_end, new_end,
671 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
673 tlb_finish_mmu(&tlb, old_start, old_end);
676 * Shrink the vma to just the new range. Always succeeds.
678 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
684 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
685 * the stack is optionally relocated, and some extra space is added.
687 int setup_arg_pages(struct linux_binprm *bprm,
688 unsigned long stack_top,
689 int executable_stack)
692 unsigned long stack_shift;
693 struct mm_struct *mm = current->mm;
694 struct vm_area_struct *vma = bprm->vma;
695 struct vm_area_struct *prev = NULL;
696 unsigned long vm_flags;
697 unsigned long stack_base;
698 unsigned long stack_size;
699 unsigned long stack_expand;
700 unsigned long rlim_stack;
702 #ifdef CONFIG_STACK_GROWSUP
703 /* Limit stack size */
704 stack_base = bprm->rlim_stack.rlim_max;
705 if (stack_base > STACK_SIZE_MAX)
706 stack_base = STACK_SIZE_MAX;
708 /* Add space for stack randomization. */
709 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
711 /* Make sure we didn't let the argument array grow too large. */
712 if (vma->vm_end - vma->vm_start > stack_base)
715 stack_base = PAGE_ALIGN(stack_top - stack_base);
717 stack_shift = vma->vm_start - stack_base;
718 mm->arg_start = bprm->p - stack_shift;
719 bprm->p = vma->vm_end - stack_shift;
721 stack_top = arch_align_stack(stack_top);
722 stack_top = PAGE_ALIGN(stack_top);
724 if (unlikely(stack_top < mmap_min_addr) ||
725 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
728 stack_shift = vma->vm_end - stack_top;
730 bprm->p -= stack_shift;
731 mm->arg_start = bprm->p;
735 bprm->loader -= stack_shift;
736 bprm->exec -= stack_shift;
738 if (down_write_killable(&mm->mmap_sem))
741 vm_flags = VM_STACK_FLAGS;
744 * Adjust stack execute permissions; explicitly enable for
745 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
746 * (arch default) otherwise.
748 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
750 else if (executable_stack == EXSTACK_DISABLE_X)
751 vm_flags &= ~VM_EXEC;
752 vm_flags |= mm->def_flags;
753 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
755 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
761 /* Move stack pages down in memory. */
763 ret = shift_arg_pages(vma, stack_shift);
768 /* mprotect_fixup is overkill to remove the temporary stack flags */
769 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
771 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
772 stack_size = vma->vm_end - vma->vm_start;
774 * Align this down to a page boundary as expand_stack
777 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
778 #ifdef CONFIG_STACK_GROWSUP
779 if (stack_size + stack_expand > rlim_stack)
780 stack_base = vma->vm_start + rlim_stack;
782 stack_base = vma->vm_end + stack_expand;
784 if (stack_size + stack_expand > rlim_stack)
785 stack_base = vma->vm_end - rlim_stack;
787 stack_base = vma->vm_start - stack_expand;
789 current->mm->start_stack = bprm->p;
790 ret = expand_stack(vma, stack_base);
795 up_write(&mm->mmap_sem);
798 EXPORT_SYMBOL(setup_arg_pages);
803 * Transfer the program arguments and environment from the holding pages
804 * onto the stack. The provided stack pointer is adjusted accordingly.
806 int transfer_args_to_stack(struct linux_binprm *bprm,
807 unsigned long *sp_location)
809 unsigned long index, stop, sp;
812 stop = bprm->p >> PAGE_SHIFT;
815 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
816 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
817 char *src = kmap(bprm->page[index]) + offset;
818 sp -= PAGE_SIZE - offset;
819 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
821 kunmap(bprm->page[index]);
826 bprm->exec += *sp_location - MAX_ARG_PAGES * PAGE_SIZE;
832 EXPORT_SYMBOL(transfer_args_to_stack);
834 #endif /* CONFIG_MMU */
836 static struct file *do_open_execat(int fd, struct filename *name, int flags)
840 struct open_flags open_exec_flags = {
841 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
842 .acc_mode = MAY_EXEC,
843 .intent = LOOKUP_OPEN,
844 .lookup_flags = LOOKUP_FOLLOW,
847 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
848 return ERR_PTR(-EINVAL);
849 if (flags & AT_SYMLINK_NOFOLLOW)
850 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
851 if (flags & AT_EMPTY_PATH)
852 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
854 file = do_filp_open(fd, name, &open_exec_flags);
859 if (!S_ISREG(file_inode(file)->i_mode))
862 if (path_noexec(&file->f_path))
865 err = deny_write_access(file);
869 if (name->name[0] != '\0')
880 struct file *open_exec(const char *name)
882 struct filename *filename = getname_kernel(name);
883 struct file *f = ERR_CAST(filename);
885 if (!IS_ERR(filename)) {
886 f = do_open_execat(AT_FDCWD, filename, 0);
891 EXPORT_SYMBOL(open_exec);
893 int kernel_read_file(struct file *file, void **buf, loff_t *size,
894 loff_t max_size, enum kernel_read_file_id id)
900 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
903 ret = deny_write_access(file);
907 ret = security_kernel_read_file(file, id);
911 i_size = i_size_read(file_inode(file));
912 if (max_size > 0 && i_size > max_size) {
921 if (id != READING_FIRMWARE_PREALLOC_BUFFER)
922 *buf = vmalloc(i_size);
929 while (pos < i_size) {
930 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
945 ret = security_kernel_post_read_file(file, *buf, i_size, id);
951 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
958 allow_write_access(file);
961 EXPORT_SYMBOL_GPL(kernel_read_file);
963 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
964 loff_t max_size, enum kernel_read_file_id id)
972 file = filp_open(path, O_RDONLY, 0);
974 return PTR_ERR(file);
976 ret = kernel_read_file(file, buf, size, max_size, id);
980 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
982 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
983 enum kernel_read_file_id id)
985 struct fd f = fdget(fd);
988 if (!f.file || !(f.file->f_mode & FMODE_READ))
991 ret = kernel_read_file(f.file, buf, size, max_size, id);
996 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
998 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1000 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1002 flush_icache_range(addr, addr + len);
1005 EXPORT_SYMBOL(read_code);
1007 static int exec_mmap(struct mm_struct *mm)
1009 struct task_struct *tsk;
1010 struct mm_struct *old_mm, *active_mm;
1012 /* Notify parent that we're no longer interested in the old VM */
1014 old_mm = current->mm;
1015 exec_mm_release(tsk, old_mm);
1018 sync_mm_rss(old_mm);
1020 * Make sure that if there is a core dump in progress
1021 * for the old mm, we get out and die instead of going
1022 * through with the exec. We must hold mmap_sem around
1023 * checking core_state and changing tsk->mm.
1025 down_read(&old_mm->mmap_sem);
1026 if (unlikely(old_mm->core_state)) {
1027 up_read(&old_mm->mmap_sem);
1033 local_irq_disable();
1034 active_mm = tsk->active_mm;
1035 tsk->active_mm = mm;
1038 * This prevents preemption while active_mm is being loaded and
1039 * it and mm are being updated, which could cause problems for
1040 * lazy tlb mm refcounting when these are updated by context
1041 * switches. Not all architectures can handle irqs off over
1044 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1046 activate_mm(active_mm, mm);
1047 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1049 tsk->mm->vmacache_seqnum = 0;
1050 vmacache_flush(tsk);
1053 up_read(&old_mm->mmap_sem);
1054 BUG_ON(active_mm != old_mm);
1055 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1056 mm_update_next_owner(old_mm);
1065 * This function makes sure the current process has its own signal table,
1066 * so that flush_signal_handlers can later reset the handlers without
1067 * disturbing other processes. (Other processes might share the signal
1068 * table via the CLONE_SIGHAND option to clone().)
1070 static int de_thread(struct task_struct *tsk)
1072 struct signal_struct *sig = tsk->signal;
1073 struct sighand_struct *oldsighand = tsk->sighand;
1074 spinlock_t *lock = &oldsighand->siglock;
1076 if (thread_group_empty(tsk))
1077 goto no_thread_group;
1080 * Kill all other threads in the thread group.
1082 spin_lock_irq(lock);
1083 if (signal_group_exit(sig)) {
1085 * Another group action in progress, just
1086 * return so that the signal is processed.
1088 spin_unlock_irq(lock);
1092 sig->group_exit_task = tsk;
1093 sig->notify_count = zap_other_threads(tsk);
1094 if (!thread_group_leader(tsk))
1095 sig->notify_count--;
1097 while (sig->notify_count) {
1098 __set_current_state(TASK_KILLABLE);
1099 spin_unlock_irq(lock);
1101 if (unlikely(__fatal_signal_pending(tsk)))
1103 spin_lock_irq(lock);
1105 spin_unlock_irq(lock);
1108 * At this point all other threads have exited, all we have to
1109 * do is to wait for the thread group leader to become inactive,
1110 * and to assume its PID:
1112 if (!thread_group_leader(tsk)) {
1113 struct task_struct *leader = tsk->group_leader;
1116 cgroup_threadgroup_change_begin(tsk);
1117 write_lock_irq(&tasklist_lock);
1119 * Do this under tasklist_lock to ensure that
1120 * exit_notify() can't miss ->group_exit_task
1122 sig->notify_count = -1;
1123 if (likely(leader->exit_state))
1125 __set_current_state(TASK_KILLABLE);
1126 write_unlock_irq(&tasklist_lock);
1127 cgroup_threadgroup_change_end(tsk);
1129 if (unlikely(__fatal_signal_pending(tsk)))
1134 * The only record we have of the real-time age of a
1135 * process, regardless of execs it's done, is start_time.
1136 * All the past CPU time is accumulated in signal_struct
1137 * from sister threads now dead. But in this non-leader
1138 * exec, nothing survives from the original leader thread,
1139 * whose birth marks the true age of this process now.
1140 * When we take on its identity by switching to its PID, we
1141 * also take its birthdate (always earlier than our own).
1143 tsk->start_time = leader->start_time;
1144 tsk->real_start_time = leader->real_start_time;
1146 BUG_ON(!same_thread_group(leader, tsk));
1147 BUG_ON(has_group_leader_pid(tsk));
1149 * An exec() starts a new thread group with the
1150 * TGID of the previous thread group. Rehash the
1151 * two threads with a switched PID, and release
1152 * the former thread group leader:
1155 /* Become a process group leader with the old leader's pid.
1156 * The old leader becomes a thread of the this thread group.
1157 * Note: The old leader also uses this pid until release_task
1158 * is called. Odd but simple and correct.
1160 tsk->pid = leader->pid;
1161 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1162 transfer_pid(leader, tsk, PIDTYPE_TGID);
1163 transfer_pid(leader, tsk, PIDTYPE_PGID);
1164 transfer_pid(leader, tsk, PIDTYPE_SID);
1166 list_replace_rcu(&leader->tasks, &tsk->tasks);
1167 list_replace_init(&leader->sibling, &tsk->sibling);
1169 tsk->group_leader = tsk;
1170 leader->group_leader = tsk;
1172 tsk->exit_signal = SIGCHLD;
1173 leader->exit_signal = -1;
1175 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1176 leader->exit_state = EXIT_DEAD;
1179 * We are going to release_task()->ptrace_unlink() silently,
1180 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1181 * the tracer wont't block again waiting for this thread.
1183 if (unlikely(leader->ptrace))
1184 __wake_up_parent(leader, leader->parent);
1185 write_unlock_irq(&tasklist_lock);
1186 cgroup_threadgroup_change_end(tsk);
1188 release_task(leader);
1191 sig->group_exit_task = NULL;
1192 sig->notify_count = 0;
1195 /* we have changed execution domain */
1196 tsk->exit_signal = SIGCHLD;
1198 #ifdef CONFIG_POSIX_TIMERS
1200 flush_itimer_signals();
1203 if (atomic_read(&oldsighand->count) != 1) {
1204 struct sighand_struct *newsighand;
1206 * This ->sighand is shared with the CLONE_SIGHAND
1207 * but not CLONE_THREAD task, switch to the new one.
1209 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1213 atomic_set(&newsighand->count, 1);
1214 memcpy(newsighand->action, oldsighand->action,
1215 sizeof(newsighand->action));
1217 write_lock_irq(&tasklist_lock);
1218 spin_lock(&oldsighand->siglock);
1219 rcu_assign_pointer(tsk->sighand, newsighand);
1220 spin_unlock(&oldsighand->siglock);
1221 write_unlock_irq(&tasklist_lock);
1223 __cleanup_sighand(oldsighand);
1226 BUG_ON(!thread_group_leader(tsk));
1230 /* protects against exit_notify() and __exit_signal() */
1231 read_lock(&tasklist_lock);
1232 sig->group_exit_task = NULL;
1233 sig->notify_count = 0;
1234 read_unlock(&tasklist_lock);
1238 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1241 strncpy(buf, tsk->comm, buf_size);
1245 EXPORT_SYMBOL_GPL(__get_task_comm);
1248 * These functions flushes out all traces of the currently running executable
1249 * so that a new one can be started
1252 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1255 trace_task_rename(tsk, buf);
1256 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1258 perf_event_comm(tsk, exec);
1262 * Calling this is the point of no return. None of the failures will be
1263 * seen by userspace since either the process is already taking a fatal
1264 * signal (via de_thread() or coredump), or will have SEGV raised
1265 * (after exec_mmap()) by search_binary_handlers (see below).
1267 int flush_old_exec(struct linux_binprm * bprm)
1272 * Make sure we have a private signal table and that
1273 * we are unassociated from the previous thread group.
1275 retval = de_thread(current);
1280 * Must be called _before_ exec_mmap() as bprm->mm is
1281 * not visibile until then. This also enables the update
1284 set_mm_exe_file(bprm->mm, bprm->file);
1286 would_dump(bprm, bprm->file);
1289 * Release all of the old mmap stuff
1291 acct_arg_size(bprm, 0);
1292 retval = exec_mmap(bprm->mm);
1297 * After clearing bprm->mm (to mark that current is using the
1298 * prepared mm now), we have nothing left of the original
1299 * process. If anything from here on returns an error, the check
1300 * in search_binary_handler() will SEGV current.
1305 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1306 PF_NOFREEZE | PF_NO_SETAFFINITY);
1308 current->personality &= ~bprm->per_clear;
1311 * We have to apply CLOEXEC before we change whether the process is
1312 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1313 * trying to access the should-be-closed file descriptors of a process
1314 * undergoing exec(2).
1316 do_close_on_exec(current->files);
1322 EXPORT_SYMBOL(flush_old_exec);
1324 void would_dump(struct linux_binprm *bprm, struct file *file)
1326 struct inode *inode = file_inode(file);
1327 if (inode_permission(inode, MAY_READ) < 0) {
1328 struct user_namespace *old, *user_ns;
1329 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1331 /* Ensure mm->user_ns contains the executable */
1332 user_ns = old = bprm->mm->user_ns;
1333 while ((user_ns != &init_user_ns) &&
1334 !privileged_wrt_inode_uidgid(user_ns, inode))
1335 user_ns = user_ns->parent;
1337 if (old != user_ns) {
1338 bprm->mm->user_ns = get_user_ns(user_ns);
1343 EXPORT_SYMBOL(would_dump);
1345 void setup_new_exec(struct linux_binprm * bprm)
1348 * Once here, prepare_binrpm() will not be called any more, so
1349 * the final state of setuid/setgid/fscaps can be merged into the
1352 bprm->secureexec |= bprm->cap_elevated;
1354 if (bprm->secureexec) {
1355 /* Make sure parent cannot signal privileged process. */
1356 current->pdeath_signal = 0;
1359 * For secureexec, reset the stack limit to sane default to
1360 * avoid bad behavior from the prior rlimits. This has to
1361 * happen before arch_pick_mmap_layout(), which examines
1362 * RLIMIT_STACK, but after the point of no return to avoid
1363 * needing to clean up the change on failure.
1365 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1366 bprm->rlim_stack.rlim_cur = _STK_LIM;
1369 arch_pick_mmap_layout(current->mm, &bprm->rlim_stack);
1371 current->sas_ss_sp = current->sas_ss_size = 0;
1374 * Figure out dumpability. Note that this checking only of current
1375 * is wrong, but userspace depends on it. This should be testing
1376 * bprm->secureexec instead.
1378 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1379 !(uid_eq(current_euid(), current_uid()) &&
1380 gid_eq(current_egid(), current_gid())))
1381 set_dumpable(current->mm, suid_dumpable);
1383 set_dumpable(current->mm, SUID_DUMP_USER);
1385 arch_setup_new_exec();
1387 __set_task_comm(current, kbasename(bprm->filename), true);
1389 /* Set the new mm task size. We have to do that late because it may
1390 * depend on TIF_32BIT which is only updated in flush_thread() on
1391 * some architectures like powerpc
1393 current->mm->task_size = TASK_SIZE;
1395 /* An exec changes our domain. We are no longer part of the thread
1397 WRITE_ONCE(current->self_exec_id, current->self_exec_id + 1);
1398 flush_signal_handlers(current, 0);
1400 EXPORT_SYMBOL(setup_new_exec);
1402 /* Runs immediately before start_thread() takes over. */
1403 void finalize_exec(struct linux_binprm *bprm)
1405 /* Store any stack rlimit changes before starting thread. */
1406 task_lock(current->group_leader);
1407 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1408 task_unlock(current->group_leader);
1410 EXPORT_SYMBOL(finalize_exec);
1413 * Prepare credentials and lock ->cred_guard_mutex.
1414 * install_exec_creds() commits the new creds and drops the lock.
1415 * Or, if exec fails before, free_bprm() should release ->cred and
1418 int prepare_bprm_creds(struct linux_binprm *bprm)
1420 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1421 return -ERESTARTNOINTR;
1423 bprm->cred = prepare_exec_creds();
1424 if (likely(bprm->cred))
1427 mutex_unlock(¤t->signal->cred_guard_mutex);
1431 static void free_bprm(struct linux_binprm *bprm)
1433 free_arg_pages(bprm);
1435 mutex_unlock(¤t->signal->cred_guard_mutex);
1436 abort_creds(bprm->cred);
1439 allow_write_access(bprm->file);
1442 /* If a binfmt changed the interp, free it. */
1443 if (bprm->interp != bprm->filename)
1444 kfree(bprm->interp);
1448 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1450 /* If a binfmt changed the interp, free it first. */
1451 if (bprm->interp != bprm->filename)
1452 kfree(bprm->interp);
1453 bprm->interp = kstrdup(interp, GFP_KERNEL);
1458 EXPORT_SYMBOL(bprm_change_interp);
1461 * install the new credentials for this executable
1463 void install_exec_creds(struct linux_binprm *bprm)
1465 security_bprm_committing_creds(bprm);
1467 commit_creds(bprm->cred);
1471 * Disable monitoring for regular users
1472 * when executing setuid binaries. Must
1473 * wait until new credentials are committed
1474 * by commit_creds() above
1476 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1477 perf_event_exit_task(current);
1479 * cred_guard_mutex must be held at least to this point to prevent
1480 * ptrace_attach() from altering our determination of the task's
1481 * credentials; any time after this it may be unlocked.
1483 security_bprm_committed_creds(bprm);
1484 mutex_unlock(¤t->signal->cred_guard_mutex);
1486 EXPORT_SYMBOL(install_exec_creds);
1489 * determine how safe it is to execute the proposed program
1490 * - the caller must hold ->cred_guard_mutex to protect against
1491 * PTRACE_ATTACH or seccomp thread-sync
1493 static void check_unsafe_exec(struct linux_binprm *bprm)
1495 struct task_struct *p = current, *t;
1499 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1502 * This isn't strictly necessary, but it makes it harder for LSMs to
1505 if (task_no_new_privs(current))
1506 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1510 spin_lock(&p->fs->lock);
1512 while_each_thread(p, t) {
1518 if (p->fs->users > n_fs)
1519 bprm->unsafe |= LSM_UNSAFE_SHARE;
1522 spin_unlock(&p->fs->lock);
1525 static void bprm_fill_uid(struct linux_binprm *bprm)
1527 struct inode *inode;
1533 * Since this can be called multiple times (via prepare_binprm),
1534 * we must clear any previous work done when setting set[ug]id
1535 * bits from any earlier bprm->file uses (for example when run
1536 * first for a setuid script then again for its interpreter).
1538 bprm->cred->euid = current_euid();
1539 bprm->cred->egid = current_egid();
1541 if (!mnt_may_suid(bprm->file->f_path.mnt))
1544 if (task_no_new_privs(current))
1547 inode = bprm->file->f_path.dentry->d_inode;
1548 mode = READ_ONCE(inode->i_mode);
1549 if (!(mode & (S_ISUID|S_ISGID)))
1552 /* Be careful if suid/sgid is set */
1555 /* reload atomically mode/uid/gid now that lock held */
1556 mode = inode->i_mode;
1559 inode_unlock(inode);
1561 /* We ignore suid/sgid if there are no mappings for them in the ns */
1562 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1563 !kgid_has_mapping(bprm->cred->user_ns, gid))
1566 if (mode & S_ISUID) {
1567 bprm->per_clear |= PER_CLEAR_ON_SETID;
1568 bprm->cred->euid = uid;
1571 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1572 bprm->per_clear |= PER_CLEAR_ON_SETID;
1573 bprm->cred->egid = gid;
1578 * Fill the binprm structure from the inode.
1579 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1581 * This may be called multiple times for binary chains (scripts for example).
1583 int prepare_binprm(struct linux_binprm *bprm)
1588 bprm_fill_uid(bprm);
1590 /* fill in binprm security blob */
1591 retval = security_bprm_set_creds(bprm);
1594 bprm->called_set_creds = 1;
1596 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1597 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1600 EXPORT_SYMBOL(prepare_binprm);
1603 * Arguments are '\0' separated strings found at the location bprm->p
1604 * points to; chop off the first by relocating brpm->p to right after
1605 * the first '\0' encountered.
1607 int remove_arg_zero(struct linux_binprm *bprm)
1610 unsigned long offset;
1618 offset = bprm->p & ~PAGE_MASK;
1619 page = get_arg_page(bprm, bprm->p, 0);
1624 kaddr = kmap_atomic(page);
1626 for (; offset < PAGE_SIZE && kaddr[offset];
1627 offset++, bprm->p++)
1630 kunmap_atomic(kaddr);
1632 } while (offset == PAGE_SIZE);
1641 EXPORT_SYMBOL(remove_arg_zero);
1643 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1645 * cycle the list of binary formats handler, until one recognizes the image
1647 int search_binary_handler(struct linux_binprm *bprm)
1649 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1650 struct linux_binfmt *fmt;
1653 /* This allows 4 levels of binfmt rewrites before failing hard. */
1654 if (bprm->recursion_depth > 5)
1657 retval = security_bprm_check(bprm);
1663 read_lock(&binfmt_lock);
1664 list_for_each_entry(fmt, &formats, lh) {
1665 if (!try_module_get(fmt->module))
1667 read_unlock(&binfmt_lock);
1668 bprm->recursion_depth++;
1669 retval = fmt->load_binary(bprm);
1670 read_lock(&binfmt_lock);
1672 bprm->recursion_depth--;
1673 if (retval < 0 && !bprm->mm) {
1674 /* we got to flush_old_exec() and failed after it */
1675 read_unlock(&binfmt_lock);
1676 force_sigsegv(SIGSEGV, current);
1679 if (retval != -ENOEXEC || !bprm->file) {
1680 read_unlock(&binfmt_lock);
1684 read_unlock(&binfmt_lock);
1687 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1688 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1690 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1698 EXPORT_SYMBOL(search_binary_handler);
1700 static int exec_binprm(struct linux_binprm *bprm)
1702 pid_t old_pid, old_vpid;
1705 /* Need to fetch pid before load_binary changes it */
1706 old_pid = current->pid;
1708 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1711 ret = search_binary_handler(bprm);
1714 trace_sched_process_exec(current, old_pid, bprm);
1715 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1716 proc_exec_connector(current);
1723 * sys_execve() executes a new program.
1725 static int __do_execve_file(int fd, struct filename *filename,
1726 struct user_arg_ptr argv,
1727 struct user_arg_ptr envp,
1728 int flags, struct file *file)
1730 char *pathbuf = NULL;
1731 struct linux_binprm *bprm;
1732 struct files_struct *displaced;
1735 if (IS_ERR(filename))
1736 return PTR_ERR(filename);
1739 * We move the actual failure in case of RLIMIT_NPROC excess from
1740 * set*uid() to execve() because too many poorly written programs
1741 * don't check setuid() return code. Here we additionally recheck
1742 * whether NPROC limit is still exceeded.
1744 if ((current->flags & PF_NPROC_EXCEEDED) &&
1745 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1750 /* We're below the limit (still or again), so we don't want to make
1751 * further execve() calls fail. */
1752 current->flags &= ~PF_NPROC_EXCEEDED;
1754 retval = unshare_files(&displaced);
1759 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1763 retval = prepare_bprm_creds(bprm);
1767 check_unsafe_exec(bprm);
1768 current->in_execve = 1;
1771 file = do_open_execat(fd, filename, flags);
1772 retval = PTR_ERR(file);
1780 bprm->filename = "none";
1781 } else if (fd == AT_FDCWD || filename->name[0] == '/') {
1782 bprm->filename = filename->name;
1784 if (filename->name[0] == '\0')
1785 pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1787 pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1788 fd, filename->name);
1794 * Record that a name derived from an O_CLOEXEC fd will be
1795 * inaccessible after exec. Relies on having exclusive access to
1796 * current->files (due to unshare_files above).
1798 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1799 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1800 bprm->filename = pathbuf;
1802 bprm->interp = bprm->filename;
1804 retval = bprm_mm_init(bprm);
1808 bprm->argc = count(argv, MAX_ARG_STRINGS);
1809 if (bprm->argc == 0)
1810 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1811 current->comm, bprm->filename);
1812 if ((retval = bprm->argc) < 0)
1815 bprm->envc = count(envp, MAX_ARG_STRINGS);
1816 if ((retval = bprm->envc) < 0)
1819 retval = prepare_binprm(bprm);
1823 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1827 bprm->exec = bprm->p;
1828 retval = copy_strings(bprm->envc, envp, bprm);
1832 retval = copy_strings(bprm->argc, argv, bprm);
1837 * When argv is empty, add an empty string ("") as argv[0] to
1838 * ensure confused userspace programs that start processing
1839 * from argv[1] won't end up walking envp. See also
1840 * bprm_stack_limits().
1842 if (bprm->argc == 0) {
1843 const char *argv[] = { "", NULL };
1844 retval = copy_strings_kernel(1, argv, bprm);
1850 retval = exec_binprm(bprm);
1854 /* execve succeeded */
1855 current->fs->in_exec = 0;
1856 current->in_execve = 0;
1857 membarrier_execve(current);
1858 rseq_execve(current);
1859 acct_update_integrals(current);
1860 task_numa_free(current, false);
1866 put_files_struct(displaced);
1871 acct_arg_size(bprm, 0);
1876 current->fs->in_exec = 0;
1877 current->in_execve = 0;
1885 reset_files_struct(displaced);
1892 static int do_execveat_common(int fd, struct filename *filename,
1893 struct user_arg_ptr argv,
1894 struct user_arg_ptr envp,
1897 return __do_execve_file(fd, filename, argv, envp, flags, NULL);
1900 int do_execve_file(struct file *file, void *__argv, void *__envp)
1902 struct user_arg_ptr argv = { .ptr.native = __argv };
1903 struct user_arg_ptr envp = { .ptr.native = __envp };
1905 return __do_execve_file(AT_FDCWD, NULL, argv, envp, 0, file);
1908 int do_execve(struct filename *filename,
1909 const char __user *const __user *__argv,
1910 const char __user *const __user *__envp)
1912 struct user_arg_ptr argv = { .ptr.native = __argv };
1913 struct user_arg_ptr envp = { .ptr.native = __envp };
1914 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1917 int do_execveat(int fd, struct filename *filename,
1918 const char __user *const __user *__argv,
1919 const char __user *const __user *__envp,
1922 struct user_arg_ptr argv = { .ptr.native = __argv };
1923 struct user_arg_ptr envp = { .ptr.native = __envp };
1925 return do_execveat_common(fd, filename, argv, envp, flags);
1928 #ifdef CONFIG_COMPAT
1929 static int compat_do_execve(struct filename *filename,
1930 const compat_uptr_t __user *__argv,
1931 const compat_uptr_t __user *__envp)
1933 struct user_arg_ptr argv = {
1935 .ptr.compat = __argv,
1937 struct user_arg_ptr envp = {
1939 .ptr.compat = __envp,
1941 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1944 static int compat_do_execveat(int fd, struct filename *filename,
1945 const compat_uptr_t __user *__argv,
1946 const compat_uptr_t __user *__envp,
1949 struct user_arg_ptr argv = {
1951 .ptr.compat = __argv,
1953 struct user_arg_ptr envp = {
1955 .ptr.compat = __envp,
1957 return do_execveat_common(fd, filename, argv, envp, flags);
1961 void set_binfmt(struct linux_binfmt *new)
1963 struct mm_struct *mm = current->mm;
1966 module_put(mm->binfmt->module);
1970 __module_get(new->module);
1972 EXPORT_SYMBOL(set_binfmt);
1975 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1977 void set_dumpable(struct mm_struct *mm, int value)
1979 unsigned long old, new;
1981 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1985 old = READ_ONCE(mm->flags);
1986 new = (old & ~MMF_DUMPABLE_MASK) | value;
1987 } while (cmpxchg(&mm->flags, old, new) != old);
1990 SYSCALL_DEFINE3(execve,
1991 const char __user *, filename,
1992 const char __user *const __user *, argv,
1993 const char __user *const __user *, envp)
1995 return do_execve(getname(filename), argv, envp);
1998 SYSCALL_DEFINE5(execveat,
1999 int, fd, const char __user *, filename,
2000 const char __user *const __user *, argv,
2001 const char __user *const __user *, envp,
2004 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2006 return do_execveat(fd,
2007 getname_flags(filename, lookup_flags, NULL),
2011 #ifdef CONFIG_COMPAT
2012 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2013 const compat_uptr_t __user *, argv,
2014 const compat_uptr_t __user *, envp)
2016 return compat_do_execve(getname(filename), argv, envp);
2019 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2020 const char __user *, filename,
2021 const compat_uptr_t __user *, argv,
2022 const compat_uptr_t __user *, envp,
2025 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2027 return compat_do_execveat(fd,
2028 getname_flags(filename, lookup_flags, NULL),