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/kernel_read_file.h>
27 #include <linux/slab.h>
28 #include <linux/file.h>
29 #include <linux/fdtable.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/kmod.h>
59 #include <linux/fsnotify.h>
60 #include <linux/fs_struct.h>
61 #include <linux/oom.h>
62 #include <linux/compat.h>
63 #include <linux/vmalloc.h>
64 #include <linux/io_uring.h>
65 #include <linux/syscall_user_dispatch.h>
66 #include <linux/coredump.h>
67 #include <linux/time_namespace.h>
69 #include <linux/uaccess.h>
70 #include <asm/mmu_context.h>
73 #include <trace/events/task.h>
76 #include <trace/events/sched.h>
78 static int bprm_creds_from_file(struct linux_binprm *bprm);
80 int suid_dumpable = 0;
82 static LIST_HEAD(formats);
83 static DEFINE_RWLOCK(binfmt_lock);
85 void __register_binfmt(struct linux_binfmt * fmt, int insert)
87 write_lock(&binfmt_lock);
88 insert ? list_add(&fmt->lh, &formats) :
89 list_add_tail(&fmt->lh, &formats);
90 write_unlock(&binfmt_lock);
93 EXPORT_SYMBOL(__register_binfmt);
95 void unregister_binfmt(struct linux_binfmt * fmt)
97 write_lock(&binfmt_lock);
99 write_unlock(&binfmt_lock);
102 EXPORT_SYMBOL(unregister_binfmt);
104 static inline void put_binfmt(struct linux_binfmt * fmt)
106 module_put(fmt->module);
109 bool path_noexec(const struct path *path)
111 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
112 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
117 * Note that a shared library must be both readable and executable due to
120 * Also note that we take the address to load from the file itself.
122 SYSCALL_DEFINE1(uselib, const char __user *, library)
124 struct linux_binfmt *fmt;
126 struct filename *tmp = getname(library);
127 int error = PTR_ERR(tmp);
128 static const struct open_flags uselib_flags = {
129 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
130 .acc_mode = MAY_READ | MAY_EXEC,
131 .intent = LOOKUP_OPEN,
132 .lookup_flags = LOOKUP_FOLLOW,
138 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
140 error = PTR_ERR(file);
145 * may_open() has already checked for this, so it should be
146 * impossible to trip now. But we need to be extra cautious
147 * and check again at the very end too.
150 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
151 path_noexec(&file->f_path)))
158 read_lock(&binfmt_lock);
159 list_for_each_entry(fmt, &formats, lh) {
160 if (!fmt->load_shlib)
162 if (!try_module_get(fmt->module))
164 read_unlock(&binfmt_lock);
165 error = fmt->load_shlib(file);
166 read_lock(&binfmt_lock);
168 if (error != -ENOEXEC)
171 read_unlock(&binfmt_lock);
177 #endif /* #ifdef CONFIG_USELIB */
181 * The nascent bprm->mm is not visible until exec_mmap() but it can
182 * use a lot of memory, account these pages in current->mm temporary
183 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
184 * change the counter back via acct_arg_size(0).
186 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
188 struct mm_struct *mm = current->mm;
189 long diff = (long)(pages - bprm->vma_pages);
194 bprm->vma_pages = pages;
195 add_mm_counter(mm, MM_ANONPAGES, diff);
198 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
203 unsigned int gup_flags = FOLL_FORCE;
205 #ifdef CONFIG_STACK_GROWSUP
207 ret = expand_downwards(bprm->vma, pos);
214 gup_flags |= FOLL_WRITE;
217 * We are doing an exec(). 'current' is the process
218 * doing the exec and bprm->mm is the new process's mm.
220 mmap_read_lock(bprm->mm);
221 ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
223 mmap_read_unlock(bprm->mm);
228 acct_arg_size(bprm, vma_pages(bprm->vma));
233 static void put_arg_page(struct page *page)
238 static void free_arg_pages(struct linux_binprm *bprm)
242 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
245 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
248 static int __bprm_mm_init(struct linux_binprm *bprm)
251 struct vm_area_struct *vma = NULL;
252 struct mm_struct *mm = bprm->mm;
254 bprm->vma = vma = vm_area_alloc(mm);
257 vma_set_anonymous(vma);
259 if (mmap_write_lock_killable(mm)) {
265 * Place the stack at the largest stack address the architecture
266 * supports. Later, we'll move this to an appropriate place. We don't
267 * use STACK_TOP because that can depend on attributes which aren't
270 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
271 vma->vm_end = STACK_TOP_MAX;
272 vma->vm_start = vma->vm_end - PAGE_SIZE;
273 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
274 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
276 err = insert_vm_struct(mm, vma);
280 mm->stack_vm = mm->total_vm = 1;
281 mmap_write_unlock(mm);
282 bprm->p = vma->vm_end - sizeof(void *);
285 mmap_write_unlock(mm);
292 static bool valid_arg_len(struct linux_binprm *bprm, long len)
294 return len <= MAX_ARG_STRLEN;
299 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
303 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
308 page = bprm->page[pos / PAGE_SIZE];
309 if (!page && write) {
310 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
313 bprm->page[pos / PAGE_SIZE] = page;
319 static void put_arg_page(struct page *page)
323 static void free_arg_page(struct linux_binprm *bprm, int i)
326 __free_page(bprm->page[i]);
327 bprm->page[i] = NULL;
331 static void free_arg_pages(struct linux_binprm *bprm)
335 for (i = 0; i < MAX_ARG_PAGES; i++)
336 free_arg_page(bprm, i);
339 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
344 static int __bprm_mm_init(struct linux_binprm *bprm)
346 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
350 static bool valid_arg_len(struct linux_binprm *bprm, long len)
352 return len <= bprm->p;
355 #endif /* CONFIG_MMU */
358 * Create a new mm_struct and populate it with a temporary stack
359 * vm_area_struct. We don't have enough context at this point to set the stack
360 * flags, permissions, and offset, so we use temporary values. We'll update
361 * them later in setup_arg_pages().
363 static int bprm_mm_init(struct linux_binprm *bprm)
366 struct mm_struct *mm = NULL;
368 bprm->mm = mm = mm_alloc();
373 /* Save current stack limit for all calculations made during exec. */
374 task_lock(current->group_leader);
375 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
376 task_unlock(current->group_leader);
378 err = __bprm_mm_init(bprm);
393 struct user_arg_ptr {
398 const char __user *const __user *native;
400 const compat_uptr_t __user *compat;
405 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
407 const char __user *native;
410 if (unlikely(argv.is_compat)) {
411 compat_uptr_t compat;
413 if (get_user(compat, argv.ptr.compat + nr))
414 return ERR_PTR(-EFAULT);
416 return compat_ptr(compat);
420 if (get_user(native, argv.ptr.native + nr))
421 return ERR_PTR(-EFAULT);
427 * count() counts the number of strings in array ARGV.
429 static int count(struct user_arg_ptr argv, int max)
433 if (argv.ptr.native != NULL) {
435 const char __user *p = get_user_arg_ptr(argv, i);
447 if (fatal_signal_pending(current))
448 return -ERESTARTNOHAND;
455 static int count_strings_kernel(const char *const *argv)
462 for (i = 0; argv[i]; ++i) {
463 if (i >= MAX_ARG_STRINGS)
465 if (fatal_signal_pending(current))
466 return -ERESTARTNOHAND;
472 static int bprm_stack_limits(struct linux_binprm *bprm)
474 unsigned long limit, ptr_size;
477 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
478 * (whichever is smaller) for the argv+env strings.
480 * - the remaining binfmt code will not run out of stack space,
481 * - the program will have a reasonable amount of stack left
484 limit = _STK_LIM / 4 * 3;
485 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
487 * We've historically supported up to 32 pages (ARG_MAX)
488 * of argument strings even with small stacks
490 limit = max_t(unsigned long, limit, ARG_MAX);
492 * We must account for the size of all the argv and envp pointers to
493 * the argv and envp strings, since they will also take up space in
494 * the stack. They aren't stored until much later when we can't
495 * signal to the parent that the child has run out of stack space.
496 * Instead, calculate it here so it's possible to fail gracefully.
498 * In the case of argc = 0, make sure there is space for adding a
499 * empty string (which will bump argc to 1), to ensure confused
500 * userspace programs don't start processing from argv[1], thinking
501 * argc can never be 0, to keep them from walking envp by accident.
502 * See do_execveat_common().
504 ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
505 if (limit <= ptr_size)
509 bprm->argmin = bprm->p - limit;
514 * 'copy_strings()' copies argument/environment strings from the old
515 * processes's memory to the new process's stack. The call to get_user_pages()
516 * ensures the destination page is created and not swapped out.
518 static int copy_strings(int argc, struct user_arg_ptr argv,
519 struct linux_binprm *bprm)
521 struct page *kmapped_page = NULL;
523 unsigned long kpos = 0;
527 const char __user *str;
532 str = get_user_arg_ptr(argv, argc);
536 len = strnlen_user(str, MAX_ARG_STRLEN);
541 if (!valid_arg_len(bprm, len))
544 /* We're going to work our way backwards. */
549 if (bprm->p < bprm->argmin)
554 int offset, bytes_to_copy;
556 if (fatal_signal_pending(current)) {
557 ret = -ERESTARTNOHAND;
562 offset = pos % PAGE_SIZE;
566 bytes_to_copy = offset;
567 if (bytes_to_copy > len)
570 offset -= bytes_to_copy;
571 pos -= bytes_to_copy;
572 str -= bytes_to_copy;
573 len -= bytes_to_copy;
575 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
578 page = get_arg_page(bprm, pos, 1);
585 flush_dcache_page(kmapped_page);
587 put_arg_page(kmapped_page);
590 kaddr = kmap_local_page(kmapped_page);
591 kpos = pos & PAGE_MASK;
592 flush_arg_page(bprm, kpos, kmapped_page);
594 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
603 flush_dcache_page(kmapped_page);
605 put_arg_page(kmapped_page);
611 * Copy and argument/environment string from the kernel to the processes stack.
613 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
615 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
616 unsigned long pos = bprm->p;
620 if (!valid_arg_len(bprm, len))
623 /* We're going to work our way backwards. */
626 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
630 unsigned int bytes_to_copy = min_t(unsigned int, len,
631 min_not_zero(offset_in_page(pos), PAGE_SIZE));
634 pos -= bytes_to_copy;
635 arg -= bytes_to_copy;
636 len -= bytes_to_copy;
638 page = get_arg_page(bprm, pos, 1);
641 flush_arg_page(bprm, pos & PAGE_MASK, page);
642 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
648 EXPORT_SYMBOL(copy_string_kernel);
650 static int copy_strings_kernel(int argc, const char *const *argv,
651 struct linux_binprm *bprm)
654 int ret = copy_string_kernel(argv[argc], bprm);
657 if (fatal_signal_pending(current))
658 return -ERESTARTNOHAND;
667 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
668 * the binfmt code determines where the new stack should reside, we shift it to
669 * its final location. The process proceeds as follows:
671 * 1) Use shift to calculate the new vma endpoints.
672 * 2) Extend vma to cover both the old and new ranges. This ensures the
673 * arguments passed to subsequent functions are consistent.
674 * 3) Move vma's page tables to the new range.
675 * 4) Free up any cleared pgd range.
676 * 5) Shrink the vma to cover only the new range.
678 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
680 struct mm_struct *mm = vma->vm_mm;
681 unsigned long old_start = vma->vm_start;
682 unsigned long old_end = vma->vm_end;
683 unsigned long length = old_end - old_start;
684 unsigned long new_start = old_start - shift;
685 unsigned long new_end = old_end - shift;
686 VMA_ITERATOR(vmi, mm, new_start);
687 struct vm_area_struct *next;
688 struct mmu_gather tlb;
690 BUG_ON(new_start > new_end);
693 * ensure there are no vmas between where we want to go
696 if (vma != vma_next(&vmi))
700 * cover the whole range: [new_start, old_end)
702 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
706 * move the page tables downwards, on failure we rely on
707 * process cleanup to remove whatever mess we made.
709 if (length != move_page_tables(vma, old_start,
710 vma, new_start, length, false))
714 tlb_gather_mmu(&tlb, mm);
715 next = vma_next(&vmi);
716 if (new_end > old_start) {
718 * when the old and new regions overlap clear from new_end.
720 free_pgd_range(&tlb, new_end, old_end, new_end,
721 next ? next->vm_start : USER_PGTABLES_CEILING);
724 * otherwise, clean from old_start; this is done to not touch
725 * the address space in [new_end, old_start) some architectures
726 * have constraints on va-space that make this illegal (IA64) -
727 * for the others its just a little faster.
729 free_pgd_range(&tlb, old_start, old_end, new_end,
730 next ? next->vm_start : USER_PGTABLES_CEILING);
732 tlb_finish_mmu(&tlb);
735 * Shrink the vma to just the new range. Always succeeds.
737 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
743 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
744 * the stack is optionally relocated, and some extra space is added.
746 int setup_arg_pages(struct linux_binprm *bprm,
747 unsigned long stack_top,
748 int executable_stack)
751 unsigned long stack_shift;
752 struct mm_struct *mm = current->mm;
753 struct vm_area_struct *vma = bprm->vma;
754 struct vm_area_struct *prev = NULL;
755 unsigned long vm_flags;
756 unsigned long stack_base;
757 unsigned long stack_size;
758 unsigned long stack_expand;
759 unsigned long rlim_stack;
760 struct mmu_gather tlb;
762 #ifdef CONFIG_STACK_GROWSUP
763 /* Limit stack size */
764 stack_base = bprm->rlim_stack.rlim_max;
766 stack_base = calc_max_stack_size(stack_base);
768 /* Add space for stack randomization. */
769 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
771 /* Make sure we didn't let the argument array grow too large. */
772 if (vma->vm_end - vma->vm_start > stack_base)
775 stack_base = PAGE_ALIGN(stack_top - stack_base);
777 stack_shift = vma->vm_start - stack_base;
778 mm->arg_start = bprm->p - stack_shift;
779 bprm->p = vma->vm_end - stack_shift;
781 stack_top = arch_align_stack(stack_top);
782 stack_top = PAGE_ALIGN(stack_top);
784 if (unlikely(stack_top < mmap_min_addr) ||
785 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
788 stack_shift = vma->vm_end - stack_top;
790 bprm->p -= stack_shift;
791 mm->arg_start = bprm->p;
795 bprm->loader -= stack_shift;
796 bprm->exec -= stack_shift;
798 if (mmap_write_lock_killable(mm))
801 vm_flags = VM_STACK_FLAGS;
804 * Adjust stack execute permissions; explicitly enable for
805 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
806 * (arch default) otherwise.
808 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
810 else if (executable_stack == EXSTACK_DISABLE_X)
811 vm_flags &= ~VM_EXEC;
812 vm_flags |= mm->def_flags;
813 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
815 tlb_gather_mmu(&tlb, mm);
816 ret = mprotect_fixup(&tlb, vma, &prev, vma->vm_start, vma->vm_end,
818 tlb_finish_mmu(&tlb);
824 if (unlikely(vm_flags & VM_EXEC)) {
825 pr_warn_once("process '%pD4' started with executable stack\n",
829 /* Move stack pages down in memory. */
831 ret = shift_arg_pages(vma, stack_shift);
836 /* mprotect_fixup is overkill to remove the temporary stack flags */
837 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
839 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
840 stack_size = vma->vm_end - vma->vm_start;
842 * Align this down to a page boundary as expand_stack
845 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
846 #ifdef CONFIG_STACK_GROWSUP
847 if (stack_size + stack_expand > rlim_stack)
848 stack_base = vma->vm_start + rlim_stack;
850 stack_base = vma->vm_end + stack_expand;
852 if (stack_size + stack_expand > rlim_stack)
853 stack_base = vma->vm_end - rlim_stack;
855 stack_base = vma->vm_start - stack_expand;
857 current->mm->start_stack = bprm->p;
858 ret = expand_stack(vma, stack_base);
863 mmap_write_unlock(mm);
866 EXPORT_SYMBOL(setup_arg_pages);
871 * Transfer the program arguments and environment from the holding pages
872 * onto the stack. The provided stack pointer is adjusted accordingly.
874 int transfer_args_to_stack(struct linux_binprm *bprm,
875 unsigned long *sp_location)
877 unsigned long index, stop, sp;
880 stop = bprm->p >> PAGE_SHIFT;
883 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
884 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
885 char *src = kmap_local_page(bprm->page[index]) + offset;
886 sp -= PAGE_SIZE - offset;
887 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
899 EXPORT_SYMBOL(transfer_args_to_stack);
901 #endif /* CONFIG_MMU */
903 static struct file *do_open_execat(int fd, struct filename *name, int flags)
907 struct open_flags open_exec_flags = {
908 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
909 .acc_mode = MAY_EXEC,
910 .intent = LOOKUP_OPEN,
911 .lookup_flags = LOOKUP_FOLLOW,
914 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
915 return ERR_PTR(-EINVAL);
916 if (flags & AT_SYMLINK_NOFOLLOW)
917 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
918 if (flags & AT_EMPTY_PATH)
919 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
921 file = do_filp_open(fd, name, &open_exec_flags);
926 * may_open() has already checked for this, so it should be
927 * impossible to trip now. But we need to be extra cautious
928 * and check again at the very end too.
931 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
932 path_noexec(&file->f_path)))
935 err = deny_write_access(file);
939 if (name->name[0] != '\0')
950 struct file *open_exec(const char *name)
952 struct filename *filename = getname_kernel(name);
953 struct file *f = ERR_CAST(filename);
955 if (!IS_ERR(filename)) {
956 f = do_open_execat(AT_FDCWD, filename, 0);
961 EXPORT_SYMBOL(open_exec);
963 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
964 defined(CONFIG_BINFMT_ELF_FDPIC)
965 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
967 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
969 flush_icache_user_range(addr, addr + len);
972 EXPORT_SYMBOL(read_code);
976 * Maps the mm_struct mm into the current task struct.
977 * On success, this function returns with exec_update_lock
980 static int exec_mmap(struct mm_struct *mm)
982 struct task_struct *tsk;
983 struct mm_struct *old_mm, *active_mm;
987 /* Notify parent that we're no longer interested in the old VM */
989 vfork = !!tsk->vfork_done;
990 old_mm = current->mm;
991 exec_mm_release(tsk, old_mm);
995 ret = down_write_killable(&tsk->signal->exec_update_lock);
1001 * If there is a pending fatal signal perhaps a signal
1002 * whose default action is to create a coredump get
1003 * out and die instead of going through with the exec.
1005 ret = mmap_read_lock_killable(old_mm);
1007 up_write(&tsk->signal->exec_update_lock);
1013 membarrier_exec_mmap(mm);
1015 local_irq_disable();
1016 active_mm = tsk->active_mm;
1017 tsk->active_mm = mm;
1021 * This prevents preemption while active_mm is being loaded and
1022 * it and mm are being updated, which could cause problems for
1023 * lazy tlb mm refcounting when these are updated by context
1024 * switches. Not all architectures can handle irqs off over
1027 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1029 activate_mm(active_mm, mm);
1030 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1036 timens_on_fork(tsk->nsproxy, tsk);
1039 mmap_read_unlock(old_mm);
1040 BUG_ON(active_mm != old_mm);
1041 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1042 mm_update_next_owner(old_mm);
1050 static int de_thread(struct task_struct *tsk)
1052 struct signal_struct *sig = tsk->signal;
1053 struct sighand_struct *oldsighand = tsk->sighand;
1054 spinlock_t *lock = &oldsighand->siglock;
1056 if (thread_group_empty(tsk))
1057 goto no_thread_group;
1060 * Kill all other threads in the thread group.
1062 spin_lock_irq(lock);
1063 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1065 * Another group action in progress, just
1066 * return so that the signal is processed.
1068 spin_unlock_irq(lock);
1072 sig->group_exec_task = tsk;
1073 sig->notify_count = zap_other_threads(tsk);
1074 if (!thread_group_leader(tsk))
1075 sig->notify_count--;
1077 while (sig->notify_count) {
1078 __set_current_state(TASK_KILLABLE);
1079 spin_unlock_irq(lock);
1081 if (__fatal_signal_pending(tsk))
1083 spin_lock_irq(lock);
1085 spin_unlock_irq(lock);
1088 * At this point all other threads have exited, all we have to
1089 * do is to wait for the thread group leader to become inactive,
1090 * and to assume its PID:
1092 if (!thread_group_leader(tsk)) {
1093 struct task_struct *leader = tsk->group_leader;
1096 cgroup_threadgroup_change_begin(tsk);
1097 write_lock_irq(&tasklist_lock);
1099 * Do this under tasklist_lock to ensure that
1100 * exit_notify() can't miss ->group_exec_task
1102 sig->notify_count = -1;
1103 if (likely(leader->exit_state))
1105 __set_current_state(TASK_KILLABLE);
1106 write_unlock_irq(&tasklist_lock);
1107 cgroup_threadgroup_change_end(tsk);
1109 if (__fatal_signal_pending(tsk))
1114 * The only record we have of the real-time age of a
1115 * process, regardless of execs it's done, is start_time.
1116 * All the past CPU time is accumulated in signal_struct
1117 * from sister threads now dead. But in this non-leader
1118 * exec, nothing survives from the original leader thread,
1119 * whose birth marks the true age of this process now.
1120 * When we take on its identity by switching to its PID, we
1121 * also take its birthdate (always earlier than our own).
1123 tsk->start_time = leader->start_time;
1124 tsk->start_boottime = leader->start_boottime;
1126 BUG_ON(!same_thread_group(leader, tsk));
1128 * An exec() starts a new thread group with the
1129 * TGID of the previous thread group. Rehash the
1130 * two threads with a switched PID, and release
1131 * the former thread group leader:
1134 /* Become a process group leader with the old leader's pid.
1135 * The old leader becomes a thread of the this thread group.
1137 exchange_tids(tsk, leader);
1138 transfer_pid(leader, tsk, PIDTYPE_TGID);
1139 transfer_pid(leader, tsk, PIDTYPE_PGID);
1140 transfer_pid(leader, tsk, PIDTYPE_SID);
1142 list_replace_rcu(&leader->tasks, &tsk->tasks);
1143 list_replace_init(&leader->sibling, &tsk->sibling);
1145 tsk->group_leader = tsk;
1146 leader->group_leader = tsk;
1148 tsk->exit_signal = SIGCHLD;
1149 leader->exit_signal = -1;
1151 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1152 leader->exit_state = EXIT_DEAD;
1155 * We are going to release_task()->ptrace_unlink() silently,
1156 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1157 * the tracer won't block again waiting for this thread.
1159 if (unlikely(leader->ptrace))
1160 __wake_up_parent(leader, leader->parent);
1161 write_unlock_irq(&tasklist_lock);
1162 cgroup_threadgroup_change_end(tsk);
1164 release_task(leader);
1167 sig->group_exec_task = NULL;
1168 sig->notify_count = 0;
1171 /* we have changed execution domain */
1172 tsk->exit_signal = SIGCHLD;
1174 BUG_ON(!thread_group_leader(tsk));
1178 /* protects against exit_notify() and __exit_signal() */
1179 read_lock(&tasklist_lock);
1180 sig->group_exec_task = NULL;
1181 sig->notify_count = 0;
1182 read_unlock(&tasklist_lock);
1188 * This function makes sure the current process has its own signal table,
1189 * so that flush_signal_handlers can later reset the handlers without
1190 * disturbing other processes. (Other processes might share the signal
1191 * table via the CLONE_SIGHAND option to clone().)
1193 static int unshare_sighand(struct task_struct *me)
1195 struct sighand_struct *oldsighand = me->sighand;
1197 if (refcount_read(&oldsighand->count) != 1) {
1198 struct sighand_struct *newsighand;
1200 * This ->sighand is shared with the CLONE_SIGHAND
1201 * but not CLONE_THREAD task, switch to the new one.
1203 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1207 refcount_set(&newsighand->count, 1);
1208 memcpy(newsighand->action, oldsighand->action,
1209 sizeof(newsighand->action));
1211 write_lock_irq(&tasklist_lock);
1212 spin_lock(&oldsighand->siglock);
1213 rcu_assign_pointer(me->sighand, newsighand);
1214 spin_unlock(&oldsighand->siglock);
1215 write_unlock_irq(&tasklist_lock);
1217 __cleanup_sighand(oldsighand);
1222 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1225 /* Always NUL terminated and zero-padded */
1226 strscpy_pad(buf, tsk->comm, buf_size);
1230 EXPORT_SYMBOL_GPL(__get_task_comm);
1233 * These functions flushes out all traces of the currently running executable
1234 * so that a new one can be started
1237 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1240 trace_task_rename(tsk, buf);
1241 strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1243 perf_event_comm(tsk, exec);
1247 * Calling this is the point of no return. None of the failures will be
1248 * seen by userspace since either the process is already taking a fatal
1249 * signal (via de_thread() or coredump), or will have SEGV raised
1250 * (after exec_mmap()) by search_binary_handler (see below).
1252 int begin_new_exec(struct linux_binprm * bprm)
1254 struct task_struct *me = current;
1257 /* Once we are committed compute the creds */
1258 retval = bprm_creds_from_file(bprm);
1263 * Ensure all future errors are fatal.
1265 bprm->point_of_no_return = true;
1268 * Make this the only thread in the thread group.
1270 retval = de_thread(me);
1275 * Cancel any io_uring activity across execve
1277 io_uring_task_cancel();
1279 /* Ensure the files table is not shared. */
1280 retval = unshare_files();
1285 * Must be called _before_ exec_mmap() as bprm->mm is
1286 * not visible until then. This also enables the update
1289 retval = set_mm_exe_file(bprm->mm, bprm->file);
1293 /* If the binary is not readable then enforce mm->dumpable=0 */
1294 would_dump(bprm, bprm->file);
1295 if (bprm->have_execfd)
1296 would_dump(bprm, bprm->executable);
1299 * Release all of the old mmap stuff
1301 acct_arg_size(bprm, 0);
1302 retval = exec_mmap(bprm->mm);
1308 #ifdef CONFIG_POSIX_TIMERS
1309 spin_lock_irq(&me->sighand->siglock);
1310 posix_cpu_timers_exit(me);
1311 spin_unlock_irq(&me->sighand->siglock);
1313 flush_itimer_signals();
1317 * Make the signal table private.
1319 retval = unshare_sighand(me);
1323 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1324 PF_NOFREEZE | PF_NO_SETAFFINITY);
1326 me->personality &= ~bprm->per_clear;
1328 clear_syscall_work_syscall_user_dispatch(me);
1331 * We have to apply CLOEXEC before we change whether the process is
1332 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1333 * trying to access the should-be-closed file descriptors of a process
1334 * undergoing exec(2).
1336 do_close_on_exec(me->files);
1338 if (bprm->secureexec) {
1339 /* Make sure parent cannot signal privileged process. */
1340 me->pdeath_signal = 0;
1343 * For secureexec, reset the stack limit to sane default to
1344 * avoid bad behavior from the prior rlimits. This has to
1345 * happen before arch_pick_mmap_layout(), which examines
1346 * RLIMIT_STACK, but after the point of no return to avoid
1347 * needing to clean up the change on failure.
1349 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1350 bprm->rlim_stack.rlim_cur = _STK_LIM;
1353 me->sas_ss_sp = me->sas_ss_size = 0;
1356 * Figure out dumpability. Note that this checking only of current
1357 * is wrong, but userspace depends on it. This should be testing
1358 * bprm->secureexec instead.
1360 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1361 !(uid_eq(current_euid(), current_uid()) &&
1362 gid_eq(current_egid(), current_gid())))
1363 set_dumpable(current->mm, suid_dumpable);
1365 set_dumpable(current->mm, SUID_DUMP_USER);
1368 __set_task_comm(me, kbasename(bprm->filename), true);
1370 /* An exec changes our domain. We are no longer part of the thread
1372 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1373 flush_signal_handlers(me, 0);
1375 retval = set_cred_ucounts(bprm->cred);
1380 * install the new credentials for this executable
1382 security_bprm_committing_creds(bprm);
1384 commit_creds(bprm->cred);
1388 * Disable monitoring for regular users
1389 * when executing setuid binaries. Must
1390 * wait until new credentials are committed
1391 * by commit_creds() above
1393 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1394 perf_event_exit_task(me);
1396 * cred_guard_mutex must be held at least to this point to prevent
1397 * ptrace_attach() from altering our determination of the task's
1398 * credentials; any time after this it may be unlocked.
1400 security_bprm_committed_creds(bprm);
1402 /* Pass the opened binary to the interpreter. */
1403 if (bprm->have_execfd) {
1404 retval = get_unused_fd_flags(0);
1407 fd_install(retval, bprm->executable);
1408 bprm->executable = NULL;
1409 bprm->execfd = retval;
1414 up_write(&me->signal->exec_update_lock);
1418 EXPORT_SYMBOL(begin_new_exec);
1420 void would_dump(struct linux_binprm *bprm, struct file *file)
1422 struct inode *inode = file_inode(file);
1423 struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1424 if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1425 struct user_namespace *old, *user_ns;
1426 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1428 /* Ensure mm->user_ns contains the executable */
1429 user_ns = old = bprm->mm->user_ns;
1430 while ((user_ns != &init_user_ns) &&
1431 !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1432 user_ns = user_ns->parent;
1434 if (old != user_ns) {
1435 bprm->mm->user_ns = get_user_ns(user_ns);
1440 EXPORT_SYMBOL(would_dump);
1442 void setup_new_exec(struct linux_binprm * bprm)
1444 /* Setup things that can depend upon the personality */
1445 struct task_struct *me = current;
1447 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1449 arch_setup_new_exec();
1451 /* Set the new mm task size. We have to do that late because it may
1452 * depend on TIF_32BIT which is only updated in flush_thread() on
1453 * some architectures like powerpc
1455 me->mm->task_size = TASK_SIZE;
1456 up_write(&me->signal->exec_update_lock);
1457 mutex_unlock(&me->signal->cred_guard_mutex);
1459 EXPORT_SYMBOL(setup_new_exec);
1461 /* Runs immediately before start_thread() takes over. */
1462 void finalize_exec(struct linux_binprm *bprm)
1464 /* Store any stack rlimit changes before starting thread. */
1465 task_lock(current->group_leader);
1466 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1467 task_unlock(current->group_leader);
1469 EXPORT_SYMBOL(finalize_exec);
1472 * Prepare credentials and lock ->cred_guard_mutex.
1473 * setup_new_exec() commits the new creds and drops the lock.
1474 * Or, if exec fails before, free_bprm() should release ->cred
1477 static int prepare_bprm_creds(struct linux_binprm *bprm)
1479 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1480 return -ERESTARTNOINTR;
1482 bprm->cred = prepare_exec_creds();
1483 if (likely(bprm->cred))
1486 mutex_unlock(¤t->signal->cred_guard_mutex);
1490 static void free_bprm(struct linux_binprm *bprm)
1493 acct_arg_size(bprm, 0);
1496 free_arg_pages(bprm);
1498 mutex_unlock(¤t->signal->cred_guard_mutex);
1499 abort_creds(bprm->cred);
1502 allow_write_access(bprm->file);
1505 if (bprm->executable)
1506 fput(bprm->executable);
1507 /* If a binfmt changed the interp, free it. */
1508 if (bprm->interp != bprm->filename)
1509 kfree(bprm->interp);
1510 kfree(bprm->fdpath);
1514 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1516 struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1517 int retval = -ENOMEM;
1521 if (fd == AT_FDCWD || filename->name[0] == '/') {
1522 bprm->filename = filename->name;
1524 if (filename->name[0] == '\0')
1525 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1527 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1528 fd, filename->name);
1532 bprm->filename = bprm->fdpath;
1534 bprm->interp = bprm->filename;
1536 retval = bprm_mm_init(bprm);
1544 return ERR_PTR(retval);
1547 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1549 /* If a binfmt changed the interp, free it first. */
1550 if (bprm->interp != bprm->filename)
1551 kfree(bprm->interp);
1552 bprm->interp = kstrdup(interp, GFP_KERNEL);
1557 EXPORT_SYMBOL(bprm_change_interp);
1560 * determine how safe it is to execute the proposed program
1561 * - the caller must hold ->cred_guard_mutex to protect against
1562 * PTRACE_ATTACH or seccomp thread-sync
1564 static void check_unsafe_exec(struct linux_binprm *bprm)
1566 struct task_struct *p = current, *t;
1570 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1573 * This isn't strictly necessary, but it makes it harder for LSMs to
1576 if (task_no_new_privs(current))
1577 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1581 spin_lock(&p->fs->lock);
1583 while_each_thread(p, t) {
1589 if (p->fs->users > n_fs)
1590 bprm->unsafe |= LSM_UNSAFE_SHARE;
1593 spin_unlock(&p->fs->lock);
1596 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1598 /* Handle suid and sgid on files */
1599 struct user_namespace *mnt_userns;
1600 struct inode *inode;
1605 if (!mnt_may_suid(file->f_path.mnt))
1608 if (task_no_new_privs(current))
1611 inode = file->f_path.dentry->d_inode;
1612 mode = READ_ONCE(inode->i_mode);
1613 if (!(mode & (S_ISUID|S_ISGID)))
1616 mnt_userns = file_mnt_user_ns(file);
1618 /* Be careful if suid/sgid is set */
1621 /* reload atomically mode/uid/gid now that lock held */
1622 mode = inode->i_mode;
1623 uid = i_uid_into_mnt(mnt_userns, inode);
1624 gid = i_gid_into_mnt(mnt_userns, inode);
1625 inode_unlock(inode);
1627 /* We ignore suid/sgid if there are no mappings for them in the ns */
1628 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1629 !kgid_has_mapping(bprm->cred->user_ns, gid))
1632 if (mode & S_ISUID) {
1633 bprm->per_clear |= PER_CLEAR_ON_SETID;
1634 bprm->cred->euid = uid;
1637 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1638 bprm->per_clear |= PER_CLEAR_ON_SETID;
1639 bprm->cred->egid = gid;
1644 * Compute brpm->cred based upon the final binary.
1646 static int bprm_creds_from_file(struct linux_binprm *bprm)
1648 /* Compute creds based on which file? */
1649 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1651 bprm_fill_uid(bprm, file);
1652 return security_bprm_creds_from_file(bprm, file);
1656 * Fill the binprm structure from the inode.
1657 * Read the first BINPRM_BUF_SIZE bytes
1659 * This may be called multiple times for binary chains (scripts for example).
1661 static int prepare_binprm(struct linux_binprm *bprm)
1665 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1666 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1670 * Arguments are '\0' separated strings found at the location bprm->p
1671 * points to; chop off the first by relocating brpm->p to right after
1672 * the first '\0' encountered.
1674 int remove_arg_zero(struct linux_binprm *bprm)
1677 unsigned long offset;
1685 offset = bprm->p & ~PAGE_MASK;
1686 page = get_arg_page(bprm, bprm->p, 0);
1691 kaddr = kmap_local_page(page);
1693 for (; offset < PAGE_SIZE && kaddr[offset];
1694 offset++, bprm->p++)
1697 kunmap_local(kaddr);
1699 } while (offset == PAGE_SIZE);
1708 EXPORT_SYMBOL(remove_arg_zero);
1710 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1712 * cycle the list of binary formats handler, until one recognizes the image
1714 static int search_binary_handler(struct linux_binprm *bprm)
1716 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1717 struct linux_binfmt *fmt;
1720 retval = prepare_binprm(bprm);
1724 retval = security_bprm_check(bprm);
1730 read_lock(&binfmt_lock);
1731 list_for_each_entry(fmt, &formats, lh) {
1732 if (!try_module_get(fmt->module))
1734 read_unlock(&binfmt_lock);
1736 retval = fmt->load_binary(bprm);
1738 read_lock(&binfmt_lock);
1740 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1741 read_unlock(&binfmt_lock);
1745 read_unlock(&binfmt_lock);
1748 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1749 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1751 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1760 static int exec_binprm(struct linux_binprm *bprm)
1762 pid_t old_pid, old_vpid;
1765 /* Need to fetch pid before load_binary changes it */
1766 old_pid = current->pid;
1768 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1771 /* This allows 4 levels of binfmt rewrites before failing hard. */
1772 for (depth = 0;; depth++) {
1777 ret = search_binary_handler(bprm);
1780 if (!bprm->interpreter)
1784 bprm->file = bprm->interpreter;
1785 bprm->interpreter = NULL;
1787 allow_write_access(exec);
1788 if (unlikely(bprm->have_execfd)) {
1789 if (bprm->executable) {
1793 bprm->executable = exec;
1799 trace_sched_process_exec(current, old_pid, bprm);
1800 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1801 proc_exec_connector(current);
1806 * sys_execve() executes a new program.
1808 static int bprm_execve(struct linux_binprm *bprm,
1809 int fd, struct filename *filename, int flags)
1814 retval = prepare_bprm_creds(bprm);
1818 check_unsafe_exec(bprm);
1819 current->in_execve = 1;
1821 file = do_open_execat(fd, filename, flags);
1822 retval = PTR_ERR(file);
1830 * Record that a name derived from an O_CLOEXEC fd will be
1831 * inaccessible after exec. This allows the code in exec to
1832 * choose to fail when the executable is not mmaped into the
1833 * interpreter and an open file descriptor is not passed to
1834 * the interpreter. This makes for a better user experience
1835 * than having the interpreter start and then immediately fail
1836 * when it finds the executable is inaccessible.
1838 if (bprm->fdpath && get_close_on_exec(fd))
1839 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1841 /* Set the unchanging part of bprm->cred */
1842 retval = security_bprm_creds_for_exec(bprm);
1846 retval = exec_binprm(bprm);
1850 /* execve succeeded */
1851 current->fs->in_exec = 0;
1852 current->in_execve = 0;
1853 rseq_execve(current);
1854 acct_update_integrals(current);
1855 task_numa_free(current, false);
1860 * If past the point of no return ensure the code never
1861 * returns to the userspace process. Use an existing fatal
1862 * signal if present otherwise terminate the process with
1865 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1866 force_fatal_sig(SIGSEGV);
1869 current->fs->in_exec = 0;
1870 current->in_execve = 0;
1875 static int do_execveat_common(int fd, struct filename *filename,
1876 struct user_arg_ptr argv,
1877 struct user_arg_ptr envp,
1880 struct linux_binprm *bprm;
1883 if (IS_ERR(filename))
1884 return PTR_ERR(filename);
1887 * We move the actual failure in case of RLIMIT_NPROC excess from
1888 * set*uid() to execve() because too many poorly written programs
1889 * don't check setuid() return code. Here we additionally recheck
1890 * whether NPROC limit is still exceeded.
1892 if ((current->flags & PF_NPROC_EXCEEDED) &&
1893 is_ucounts_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1898 /* We're below the limit (still or again), so we don't want to make
1899 * further execve() calls fail. */
1900 current->flags &= ~PF_NPROC_EXCEEDED;
1902 bprm = alloc_bprm(fd, filename);
1904 retval = PTR_ERR(bprm);
1908 retval = count(argv, MAX_ARG_STRINGS);
1910 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1911 current->comm, bprm->filename);
1914 bprm->argc = retval;
1916 retval = count(envp, MAX_ARG_STRINGS);
1919 bprm->envc = retval;
1921 retval = bprm_stack_limits(bprm);
1925 retval = copy_string_kernel(bprm->filename, bprm);
1928 bprm->exec = bprm->p;
1930 retval = copy_strings(bprm->envc, envp, bprm);
1934 retval = copy_strings(bprm->argc, argv, bprm);
1939 * When argv is empty, add an empty string ("") as argv[0] to
1940 * ensure confused userspace programs that start processing
1941 * from argv[1] won't end up walking envp. See also
1942 * bprm_stack_limits().
1944 if (bprm->argc == 0) {
1945 retval = copy_string_kernel("", bprm);
1951 retval = bprm_execve(bprm, fd, filename, flags);
1960 int kernel_execve(const char *kernel_filename,
1961 const char *const *argv, const char *const *envp)
1963 struct filename *filename;
1964 struct linux_binprm *bprm;
1968 /* It is non-sense for kernel threads to call execve */
1969 if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1972 filename = getname_kernel(kernel_filename);
1973 if (IS_ERR(filename))
1974 return PTR_ERR(filename);
1976 bprm = alloc_bprm(fd, filename);
1978 retval = PTR_ERR(bprm);
1982 retval = count_strings_kernel(argv);
1983 if (WARN_ON_ONCE(retval == 0))
1987 bprm->argc = retval;
1989 retval = count_strings_kernel(envp);
1992 bprm->envc = retval;
1994 retval = bprm_stack_limits(bprm);
1998 retval = copy_string_kernel(bprm->filename, bprm);
2001 bprm->exec = bprm->p;
2003 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2007 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2011 retval = bprm_execve(bprm, fd, filename, 0);
2019 static int do_execve(struct filename *filename,
2020 const char __user *const __user *__argv,
2021 const char __user *const __user *__envp)
2023 struct user_arg_ptr argv = { .ptr.native = __argv };
2024 struct user_arg_ptr envp = { .ptr.native = __envp };
2025 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2028 static int do_execveat(int fd, struct filename *filename,
2029 const char __user *const __user *__argv,
2030 const char __user *const __user *__envp,
2033 struct user_arg_ptr argv = { .ptr.native = __argv };
2034 struct user_arg_ptr envp = { .ptr.native = __envp };
2036 return do_execveat_common(fd, filename, argv, envp, flags);
2039 #ifdef CONFIG_COMPAT
2040 static int compat_do_execve(struct filename *filename,
2041 const compat_uptr_t __user *__argv,
2042 const compat_uptr_t __user *__envp)
2044 struct user_arg_ptr argv = {
2046 .ptr.compat = __argv,
2048 struct user_arg_ptr envp = {
2050 .ptr.compat = __envp,
2052 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2055 static int compat_do_execveat(int fd, struct filename *filename,
2056 const compat_uptr_t __user *__argv,
2057 const compat_uptr_t __user *__envp,
2060 struct user_arg_ptr argv = {
2062 .ptr.compat = __argv,
2064 struct user_arg_ptr envp = {
2066 .ptr.compat = __envp,
2068 return do_execveat_common(fd, filename, argv, envp, flags);
2072 void set_binfmt(struct linux_binfmt *new)
2074 struct mm_struct *mm = current->mm;
2077 module_put(mm->binfmt->module);
2081 __module_get(new->module);
2083 EXPORT_SYMBOL(set_binfmt);
2086 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2088 void set_dumpable(struct mm_struct *mm, int value)
2090 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2093 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2096 SYSCALL_DEFINE3(execve,
2097 const char __user *, filename,
2098 const char __user *const __user *, argv,
2099 const char __user *const __user *, envp)
2101 return do_execve(getname(filename), argv, envp);
2104 SYSCALL_DEFINE5(execveat,
2105 int, fd, const char __user *, filename,
2106 const char __user *const __user *, argv,
2107 const char __user *const __user *, envp,
2110 return do_execveat(fd,
2111 getname_uflags(filename, flags),
2115 #ifdef CONFIG_COMPAT
2116 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2117 const compat_uptr_t __user *, argv,
2118 const compat_uptr_t __user *, envp)
2120 return compat_do_execve(getname(filename), argv, envp);
2123 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2124 const char __user *, filename,
2125 const compat_uptr_t __user *, argv,
2126 const compat_uptr_t __user *, envp,
2129 return compat_do_execveat(fd,
2130 getname_uflags(filename, flags),
2135 #ifdef CONFIG_SYSCTL
2137 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2138 void *buffer, size_t *lenp, loff_t *ppos)
2140 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2143 validate_coredump_safety();
2147 static struct ctl_table fs_exec_sysctls[] = {
2149 .procname = "suid_dumpable",
2150 .data = &suid_dumpable,
2151 .maxlen = sizeof(int),
2153 .proc_handler = proc_dointvec_minmax_coredump,
2154 .extra1 = SYSCTL_ZERO,
2155 .extra2 = SYSCTL_TWO,
2160 static int __init init_fs_exec_sysctls(void)
2162 register_sysctl_init("fs", fs_exec_sysctls);
2166 fs_initcall(init_fs_exec_sysctls);
2167 #endif /* CONFIG_SYSCTL */