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>
68 #include <linux/user_events.h>
69 #include <linux/rseq.h>
71 #include <linux/uaccess.h>
72 #include <asm/mmu_context.h>
75 #include <trace/events/task.h>
78 #include <trace/events/sched.h>
80 static int bprm_creds_from_file(struct linux_binprm *bprm);
82 int suid_dumpable = 0;
84 static LIST_HEAD(formats);
85 static DEFINE_RWLOCK(binfmt_lock);
87 void __register_binfmt(struct linux_binfmt * fmt, int insert)
89 write_lock(&binfmt_lock);
90 insert ? list_add(&fmt->lh, &formats) :
91 list_add_tail(&fmt->lh, &formats);
92 write_unlock(&binfmt_lock);
95 EXPORT_SYMBOL(__register_binfmt);
97 void unregister_binfmt(struct linux_binfmt * fmt)
99 write_lock(&binfmt_lock);
101 write_unlock(&binfmt_lock);
104 EXPORT_SYMBOL(unregister_binfmt);
106 static inline void put_binfmt(struct linux_binfmt * fmt)
108 module_put(fmt->module);
111 bool path_noexec(const struct path *path)
113 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
114 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
119 * Note that a shared library must be both readable and executable due to
122 * Also note that we take the address to load from the file itself.
124 SYSCALL_DEFINE1(uselib, const char __user *, library)
126 struct linux_binfmt *fmt;
128 struct filename *tmp = getname(library);
129 int error = PTR_ERR(tmp);
130 static const struct open_flags uselib_flags = {
131 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
132 .acc_mode = MAY_READ | MAY_EXEC,
133 .intent = LOOKUP_OPEN,
134 .lookup_flags = LOOKUP_FOLLOW,
140 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
142 error = PTR_ERR(file);
147 * may_open() has already checked for this, so it should be
148 * impossible to trip now. But we need to be extra cautious
149 * and check again at the very end too.
152 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
153 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,
202 struct vm_area_struct *vma = bprm->vma;
203 struct mm_struct *mm = bprm->mm;
207 * Avoid relying on expanding the stack down in GUP (which
208 * does not work for STACK_GROWSUP anyway), and just do it
209 * by hand ahead of time.
211 if (write && pos < vma->vm_start) {
213 ret = expand_downwards(vma, pos);
214 if (unlikely(ret < 0)) {
215 mmap_write_unlock(mm);
218 mmap_write_downgrade(mm);
223 * We are doing an exec(). 'current' is the process
224 * doing the exec and 'mm' is the new process's mm.
226 ret = get_user_pages_remote(mm, pos, 1,
227 write ? FOLL_WRITE : 0,
229 mmap_read_unlock(mm);
234 acct_arg_size(bprm, vma_pages(vma));
239 static void put_arg_page(struct page *page)
244 static void free_arg_pages(struct linux_binprm *bprm)
248 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
251 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
254 static int __bprm_mm_init(struct linux_binprm *bprm)
257 struct vm_area_struct *vma = NULL;
258 struct mm_struct *mm = bprm->mm;
260 bprm->vma = vma = vm_area_alloc(mm);
263 vma_set_anonymous(vma);
265 if (mmap_write_lock_killable(mm)) {
271 * Place the stack at the largest stack address the architecture
272 * supports. Later, we'll move this to an appropriate place. We don't
273 * use STACK_TOP because that can depend on attributes which aren't
276 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
277 vma->vm_end = STACK_TOP_MAX;
278 vma->vm_start = vma->vm_end - PAGE_SIZE;
279 vm_flags_init(vma, VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP);
280 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
282 err = insert_vm_struct(mm, vma);
286 mm->stack_vm = mm->total_vm = 1;
287 mmap_write_unlock(mm);
288 bprm->p = vma->vm_end - sizeof(void *);
291 mmap_write_unlock(mm);
298 static bool valid_arg_len(struct linux_binprm *bprm, long len)
300 return len <= MAX_ARG_STRLEN;
305 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
309 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
314 page = bprm->page[pos / PAGE_SIZE];
315 if (!page && write) {
316 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
319 bprm->page[pos / PAGE_SIZE] = page;
325 static void put_arg_page(struct page *page)
329 static void free_arg_page(struct linux_binprm *bprm, int i)
332 __free_page(bprm->page[i]);
333 bprm->page[i] = NULL;
337 static void free_arg_pages(struct linux_binprm *bprm)
341 for (i = 0; i < MAX_ARG_PAGES; i++)
342 free_arg_page(bprm, i);
345 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
350 static int __bprm_mm_init(struct linux_binprm *bprm)
352 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
356 static bool valid_arg_len(struct linux_binprm *bprm, long len)
358 return len <= bprm->p;
361 #endif /* CONFIG_MMU */
364 * Create a new mm_struct and populate it with a temporary stack
365 * vm_area_struct. We don't have enough context at this point to set the stack
366 * flags, permissions, and offset, so we use temporary values. We'll update
367 * them later in setup_arg_pages().
369 static int bprm_mm_init(struct linux_binprm *bprm)
372 struct mm_struct *mm = NULL;
374 bprm->mm = mm = mm_alloc();
379 /* Save current stack limit for all calculations made during exec. */
380 task_lock(current->group_leader);
381 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
382 task_unlock(current->group_leader);
384 err = __bprm_mm_init(bprm);
399 struct user_arg_ptr {
404 const char __user *const __user *native;
406 const compat_uptr_t __user *compat;
411 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
413 const char __user *native;
416 if (unlikely(argv.is_compat)) {
417 compat_uptr_t compat;
419 if (get_user(compat, argv.ptr.compat + nr))
420 return ERR_PTR(-EFAULT);
422 return compat_ptr(compat);
426 if (get_user(native, argv.ptr.native + nr))
427 return ERR_PTR(-EFAULT);
433 * count() counts the number of strings in array ARGV.
435 static int count(struct user_arg_ptr argv, int max)
439 if (argv.ptr.native != NULL) {
441 const char __user *p = get_user_arg_ptr(argv, i);
453 if (fatal_signal_pending(current))
454 return -ERESTARTNOHAND;
461 static int count_strings_kernel(const char *const *argv)
468 for (i = 0; argv[i]; ++i) {
469 if (i >= MAX_ARG_STRINGS)
471 if (fatal_signal_pending(current))
472 return -ERESTARTNOHAND;
478 static int bprm_stack_limits(struct linux_binprm *bprm)
480 unsigned long limit, ptr_size;
483 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
484 * (whichever is smaller) for the argv+env strings.
486 * - the remaining binfmt code will not run out of stack space,
487 * - the program will have a reasonable amount of stack left
490 limit = _STK_LIM / 4 * 3;
491 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
493 * We've historically supported up to 32 pages (ARG_MAX)
494 * of argument strings even with small stacks
496 limit = max_t(unsigned long, limit, ARG_MAX);
498 * We must account for the size of all the argv and envp pointers to
499 * the argv and envp strings, since they will also take up space in
500 * the stack. They aren't stored until much later when we can't
501 * signal to the parent that the child has run out of stack space.
502 * Instead, calculate it here so it's possible to fail gracefully.
504 * In the case of argc = 0, make sure there is space for adding a
505 * empty string (which will bump argc to 1), to ensure confused
506 * userspace programs don't start processing from argv[1], thinking
507 * argc can never be 0, to keep them from walking envp by accident.
508 * See do_execveat_common().
510 ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
511 if (limit <= ptr_size)
515 bprm->argmin = bprm->p - limit;
520 * 'copy_strings()' copies argument/environment strings from the old
521 * processes's memory to the new process's stack. The call to get_user_pages()
522 * ensures the destination page is created and not swapped out.
524 static int copy_strings(int argc, struct user_arg_ptr argv,
525 struct linux_binprm *bprm)
527 struct page *kmapped_page = NULL;
529 unsigned long kpos = 0;
533 const char __user *str;
538 str = get_user_arg_ptr(argv, argc);
542 len = strnlen_user(str, MAX_ARG_STRLEN);
547 if (!valid_arg_len(bprm, len))
550 /* We're going to work our way backwards. */
555 if (bprm->p < bprm->argmin)
560 int offset, bytes_to_copy;
562 if (fatal_signal_pending(current)) {
563 ret = -ERESTARTNOHAND;
568 offset = pos % PAGE_SIZE;
572 bytes_to_copy = offset;
573 if (bytes_to_copy > len)
576 offset -= bytes_to_copy;
577 pos -= bytes_to_copy;
578 str -= bytes_to_copy;
579 len -= bytes_to_copy;
581 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
584 page = get_arg_page(bprm, pos, 1);
591 flush_dcache_page(kmapped_page);
593 put_arg_page(kmapped_page);
596 kaddr = kmap_local_page(kmapped_page);
597 kpos = pos & PAGE_MASK;
598 flush_arg_page(bprm, kpos, kmapped_page);
600 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
609 flush_dcache_page(kmapped_page);
611 put_arg_page(kmapped_page);
617 * Copy and argument/environment string from the kernel to the processes stack.
619 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
621 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
622 unsigned long pos = bprm->p;
626 if (!valid_arg_len(bprm, len))
629 /* We're going to work our way backwards. */
632 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
636 unsigned int bytes_to_copy = min_t(unsigned int, len,
637 min_not_zero(offset_in_page(pos), PAGE_SIZE));
640 pos -= bytes_to_copy;
641 arg -= bytes_to_copy;
642 len -= bytes_to_copy;
644 page = get_arg_page(bprm, pos, 1);
647 flush_arg_page(bprm, pos & PAGE_MASK, page);
648 memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
654 EXPORT_SYMBOL(copy_string_kernel);
656 static int copy_strings_kernel(int argc, const char *const *argv,
657 struct linux_binprm *bprm)
660 int ret = copy_string_kernel(argv[argc], bprm);
663 if (fatal_signal_pending(current))
664 return -ERESTARTNOHAND;
673 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
674 * the binfmt code determines where the new stack should reside, we shift it to
675 * its final location. The process proceeds as follows:
677 * 1) Use shift to calculate the new vma endpoints.
678 * 2) Extend vma to cover both the old and new ranges. This ensures the
679 * arguments passed to subsequent functions are consistent.
680 * 3) Move vma's page tables to the new range.
681 * 4) Free up any cleared pgd range.
682 * 5) Shrink the vma to cover only the new range.
684 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
686 struct mm_struct *mm = vma->vm_mm;
687 unsigned long old_start = vma->vm_start;
688 unsigned long old_end = vma->vm_end;
689 unsigned long length = old_end - old_start;
690 unsigned long new_start = old_start - shift;
691 unsigned long new_end = old_end - shift;
692 VMA_ITERATOR(vmi, mm, new_start);
693 struct vm_area_struct *next;
694 struct mmu_gather tlb;
696 BUG_ON(new_start > new_end);
699 * ensure there are no vmas between where we want to go
702 if (vma != vma_next(&vmi))
705 vma_iter_prev_range(&vmi);
707 * cover the whole range: [new_start, old_end)
709 if (vma_expand(&vmi, vma, new_start, old_end, vma->vm_pgoff, NULL))
713 * move the page tables downwards, on failure we rely on
714 * process cleanup to remove whatever mess we made.
716 if (length != move_page_tables(vma, old_start,
717 vma, new_start, length, false, true))
721 tlb_gather_mmu(&tlb, mm);
722 next = vma_next(&vmi);
723 if (new_end > old_start) {
725 * when the old and new regions overlap clear from new_end.
727 free_pgd_range(&tlb, new_end, old_end, new_end,
728 next ? next->vm_start : USER_PGTABLES_CEILING);
731 * otherwise, clean from old_start; this is done to not touch
732 * the address space in [new_end, old_start) some architectures
733 * have constraints on va-space that make this illegal (IA64) -
734 * for the others its just a little faster.
736 free_pgd_range(&tlb, old_start, old_end, new_end,
737 next ? next->vm_start : USER_PGTABLES_CEILING);
739 tlb_finish_mmu(&tlb);
742 /* Shrink the vma to just the new range */
743 return vma_shrink(&vmi, vma, new_start, new_end, vma->vm_pgoff);
747 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
748 * the stack is optionally relocated, and some extra space is added.
750 int setup_arg_pages(struct linux_binprm *bprm,
751 unsigned long stack_top,
752 int executable_stack)
755 unsigned long stack_shift;
756 struct mm_struct *mm = current->mm;
757 struct vm_area_struct *vma = bprm->vma;
758 struct vm_area_struct *prev = NULL;
759 unsigned long vm_flags;
760 unsigned long stack_base;
761 unsigned long stack_size;
762 unsigned long stack_expand;
763 unsigned long rlim_stack;
764 struct mmu_gather tlb;
765 struct vma_iterator vmi;
767 #ifdef CONFIG_STACK_GROWSUP
768 /* Limit stack size */
769 stack_base = bprm->rlim_stack.rlim_max;
771 stack_base = calc_max_stack_size(stack_base);
773 /* Add space for stack randomization. */
774 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
776 /* Make sure we didn't let the argument array grow too large. */
777 if (vma->vm_end - vma->vm_start > stack_base)
780 stack_base = PAGE_ALIGN(stack_top - stack_base);
782 stack_shift = vma->vm_start - stack_base;
783 mm->arg_start = bprm->p - stack_shift;
784 bprm->p = vma->vm_end - stack_shift;
786 stack_top = arch_align_stack(stack_top);
787 stack_top = PAGE_ALIGN(stack_top);
789 if (unlikely(stack_top < mmap_min_addr) ||
790 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
793 stack_shift = vma->vm_end - stack_top;
795 bprm->p -= stack_shift;
796 mm->arg_start = bprm->p;
800 bprm->loader -= stack_shift;
801 bprm->exec -= stack_shift;
803 if (mmap_write_lock_killable(mm))
806 vm_flags = VM_STACK_FLAGS;
809 * Adjust stack execute permissions; explicitly enable for
810 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
811 * (arch default) otherwise.
813 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
815 else if (executable_stack == EXSTACK_DISABLE_X)
816 vm_flags &= ~VM_EXEC;
817 vm_flags |= mm->def_flags;
818 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
820 vma_iter_init(&vmi, mm, vma->vm_start);
822 tlb_gather_mmu(&tlb, mm);
823 ret = mprotect_fixup(&vmi, &tlb, vma, &prev, vma->vm_start, vma->vm_end,
825 tlb_finish_mmu(&tlb);
831 if (unlikely(vm_flags & VM_EXEC)) {
832 pr_warn_once("process '%pD4' started with executable stack\n",
836 /* Move stack pages down in memory. */
838 ret = shift_arg_pages(vma, stack_shift);
843 /* mprotect_fixup is overkill to remove the temporary stack flags */
844 vm_flags_clear(vma, VM_STACK_INCOMPLETE_SETUP);
846 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
847 stack_size = vma->vm_end - vma->vm_start;
849 * Align this down to a page boundary as expand_stack
852 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
854 stack_expand = min(rlim_stack, stack_size + stack_expand);
856 #ifdef CONFIG_STACK_GROWSUP
857 stack_base = vma->vm_start + stack_expand;
859 stack_base = vma->vm_end - stack_expand;
861 current->mm->start_stack = bprm->p;
862 ret = expand_stack_locked(vma, stack_base);
867 mmap_write_unlock(mm);
870 EXPORT_SYMBOL(setup_arg_pages);
875 * Transfer the program arguments and environment from the holding pages
876 * onto the stack. The provided stack pointer is adjusted accordingly.
878 int transfer_args_to_stack(struct linux_binprm *bprm,
879 unsigned long *sp_location)
881 unsigned long index, stop, sp;
884 stop = bprm->p >> PAGE_SHIFT;
887 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
888 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
889 char *src = kmap_local_page(bprm->page[index]) + offset;
890 sp -= PAGE_SIZE - offset;
891 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
903 EXPORT_SYMBOL(transfer_args_to_stack);
905 #endif /* CONFIG_MMU */
907 static struct file *do_open_execat(int fd, struct filename *name, int flags)
911 struct open_flags open_exec_flags = {
912 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
913 .acc_mode = MAY_EXEC,
914 .intent = LOOKUP_OPEN,
915 .lookup_flags = LOOKUP_FOLLOW,
918 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
919 return ERR_PTR(-EINVAL);
920 if (flags & AT_SYMLINK_NOFOLLOW)
921 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
922 if (flags & AT_EMPTY_PATH)
923 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
925 file = do_filp_open(fd, name, &open_exec_flags);
930 * may_open() has already checked for this, so it should be
931 * impossible to trip now. But we need to be extra cautious
932 * and check again at the very end too.
935 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
936 path_noexec(&file->f_path)))
939 err = deny_write_access(file);
951 struct file *open_exec(const char *name)
953 struct filename *filename = getname_kernel(name);
954 struct file *f = ERR_CAST(filename);
956 if (!IS_ERR(filename)) {
957 f = do_open_execat(AT_FDCWD, filename, 0);
962 EXPORT_SYMBOL(open_exec);
964 #if defined(CONFIG_BINFMT_FLAT) || 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;
986 /* Notify parent that we're no longer interested in the old VM */
988 old_mm = current->mm;
989 exec_mm_release(tsk, old_mm);
991 ret = down_write_killable(&tsk->signal->exec_update_lock);
997 * If there is a pending fatal signal perhaps a signal
998 * whose default action is to create a coredump get
999 * out and die instead of going through with the exec.
1001 ret = mmap_read_lock_killable(old_mm);
1003 up_write(&tsk->signal->exec_update_lock);
1009 membarrier_exec_mmap(mm);
1011 local_irq_disable();
1012 active_mm = tsk->active_mm;
1013 tsk->active_mm = mm;
1017 * This prevents preemption while active_mm is being loaded and
1018 * it and mm are being updated, which could cause problems for
1019 * lazy tlb mm refcounting when these are updated by context
1020 * switches. Not all architectures can handle irqs off over
1023 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1025 activate_mm(active_mm, mm);
1026 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1032 mmap_read_unlock(old_mm);
1033 BUG_ON(active_mm != old_mm);
1034 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1035 mm_update_next_owner(old_mm);
1039 mmdrop_lazy_tlb(active_mm);
1043 static int de_thread(struct task_struct *tsk)
1045 struct signal_struct *sig = tsk->signal;
1046 struct sighand_struct *oldsighand = tsk->sighand;
1047 spinlock_t *lock = &oldsighand->siglock;
1049 if (thread_group_empty(tsk))
1050 goto no_thread_group;
1053 * Kill all other threads in the thread group.
1055 spin_lock_irq(lock);
1056 if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1058 * Another group action in progress, just
1059 * return so that the signal is processed.
1061 spin_unlock_irq(lock);
1065 sig->group_exec_task = tsk;
1066 sig->notify_count = zap_other_threads(tsk);
1067 if (!thread_group_leader(tsk))
1068 sig->notify_count--;
1070 while (sig->notify_count) {
1071 __set_current_state(TASK_KILLABLE);
1072 spin_unlock_irq(lock);
1074 if (__fatal_signal_pending(tsk))
1076 spin_lock_irq(lock);
1078 spin_unlock_irq(lock);
1081 * At this point all other threads have exited, all we have to
1082 * do is to wait for the thread group leader to become inactive,
1083 * and to assume its PID:
1085 if (!thread_group_leader(tsk)) {
1086 struct task_struct *leader = tsk->group_leader;
1089 cgroup_threadgroup_change_begin(tsk);
1090 write_lock_irq(&tasklist_lock);
1092 * Do this under tasklist_lock to ensure that
1093 * exit_notify() can't miss ->group_exec_task
1095 sig->notify_count = -1;
1096 if (likely(leader->exit_state))
1098 __set_current_state(TASK_KILLABLE);
1099 write_unlock_irq(&tasklist_lock);
1100 cgroup_threadgroup_change_end(tsk);
1102 if (__fatal_signal_pending(tsk))
1107 * The only record we have of the real-time age of a
1108 * process, regardless of execs it's done, is start_time.
1109 * All the past CPU time is accumulated in signal_struct
1110 * from sister threads now dead. But in this non-leader
1111 * exec, nothing survives from the original leader thread,
1112 * whose birth marks the true age of this process now.
1113 * When we take on its identity by switching to its PID, we
1114 * also take its birthdate (always earlier than our own).
1116 tsk->start_time = leader->start_time;
1117 tsk->start_boottime = leader->start_boottime;
1119 BUG_ON(!same_thread_group(leader, tsk));
1121 * An exec() starts a new thread group with the
1122 * TGID of the previous thread group. Rehash the
1123 * two threads with a switched PID, and release
1124 * the former thread group leader:
1127 /* Become a process group leader with the old leader's pid.
1128 * The old leader becomes a thread of the this thread group.
1130 exchange_tids(tsk, leader);
1131 transfer_pid(leader, tsk, PIDTYPE_TGID);
1132 transfer_pid(leader, tsk, PIDTYPE_PGID);
1133 transfer_pid(leader, tsk, PIDTYPE_SID);
1135 list_replace_rcu(&leader->tasks, &tsk->tasks);
1136 list_replace_init(&leader->sibling, &tsk->sibling);
1138 tsk->group_leader = tsk;
1139 leader->group_leader = tsk;
1141 tsk->exit_signal = SIGCHLD;
1142 leader->exit_signal = -1;
1144 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1145 leader->exit_state = EXIT_DEAD;
1148 * We are going to release_task()->ptrace_unlink() silently,
1149 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1150 * the tracer won't block again waiting for this thread.
1152 if (unlikely(leader->ptrace))
1153 __wake_up_parent(leader, leader->parent);
1154 write_unlock_irq(&tasklist_lock);
1155 cgroup_threadgroup_change_end(tsk);
1157 release_task(leader);
1160 sig->group_exec_task = NULL;
1161 sig->notify_count = 0;
1164 /* we have changed execution domain */
1165 tsk->exit_signal = SIGCHLD;
1167 BUG_ON(!thread_group_leader(tsk));
1171 /* protects against exit_notify() and __exit_signal() */
1172 read_lock(&tasklist_lock);
1173 sig->group_exec_task = NULL;
1174 sig->notify_count = 0;
1175 read_unlock(&tasklist_lock);
1181 * This function makes sure the current process has its own signal table,
1182 * so that flush_signal_handlers can later reset the handlers without
1183 * disturbing other processes. (Other processes might share the signal
1184 * table via the CLONE_SIGHAND option to clone().)
1186 static int unshare_sighand(struct task_struct *me)
1188 struct sighand_struct *oldsighand = me->sighand;
1190 if (refcount_read(&oldsighand->count) != 1) {
1191 struct sighand_struct *newsighand;
1193 * This ->sighand is shared with the CLONE_SIGHAND
1194 * but not CLONE_THREAD task, switch to the new one.
1196 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1200 refcount_set(&newsighand->count, 1);
1202 write_lock_irq(&tasklist_lock);
1203 spin_lock(&oldsighand->siglock);
1204 memcpy(newsighand->action, oldsighand->action,
1205 sizeof(newsighand->action));
1206 rcu_assign_pointer(me->sighand, newsighand);
1207 spin_unlock(&oldsighand->siglock);
1208 write_unlock_irq(&tasklist_lock);
1210 __cleanup_sighand(oldsighand);
1215 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1218 /* Always NUL terminated and zero-padded */
1219 strscpy_pad(buf, tsk->comm, buf_size);
1223 EXPORT_SYMBOL_GPL(__get_task_comm);
1226 * These functions flushes out all traces of the currently running executable
1227 * so that a new one can be started
1230 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1233 trace_task_rename(tsk, buf);
1234 strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1236 perf_event_comm(tsk, exec);
1240 * Calling this is the point of no return. None of the failures will be
1241 * seen by userspace since either the process is already taking a fatal
1242 * signal (via de_thread() or coredump), or will have SEGV raised
1243 * (after exec_mmap()) by search_binary_handler (see below).
1245 int begin_new_exec(struct linux_binprm * bprm)
1247 struct task_struct *me = current;
1250 /* Once we are committed compute the creds */
1251 retval = bprm_creds_from_file(bprm);
1256 * Ensure all future errors are fatal.
1258 bprm->point_of_no_return = true;
1261 * Make this the only thread in the thread group.
1263 retval = de_thread(me);
1268 * Cancel any io_uring activity across execve
1270 io_uring_task_cancel();
1272 /* Ensure the files table is not shared. */
1273 retval = unshare_files();
1278 * Must be called _before_ exec_mmap() as bprm->mm is
1279 * not visible until then. Doing it here also ensures
1280 * we don't race against replace_mm_exe_file().
1282 retval = set_mm_exe_file(bprm->mm, bprm->file);
1286 /* If the binary is not readable then enforce mm->dumpable=0 */
1287 would_dump(bprm, bprm->file);
1288 if (bprm->have_execfd)
1289 would_dump(bprm, bprm->executable);
1292 * Release all of the old mmap stuff
1294 acct_arg_size(bprm, 0);
1295 retval = exec_mmap(bprm->mm);
1301 retval = exec_task_namespaces();
1305 #ifdef CONFIG_POSIX_TIMERS
1306 spin_lock_irq(&me->sighand->siglock);
1307 posix_cpu_timers_exit(me);
1308 spin_unlock_irq(&me->sighand->siglock);
1310 flush_itimer_signals();
1314 * Make the signal table private.
1316 retval = unshare_sighand(me);
1320 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1321 PF_NOFREEZE | PF_NO_SETAFFINITY);
1323 me->personality &= ~bprm->per_clear;
1325 clear_syscall_work_syscall_user_dispatch(me);
1328 * We have to apply CLOEXEC before we change whether the process is
1329 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1330 * trying to access the should-be-closed file descriptors of a process
1331 * undergoing exec(2).
1333 do_close_on_exec(me->files);
1335 if (bprm->secureexec) {
1336 /* Make sure parent cannot signal privileged process. */
1337 me->pdeath_signal = 0;
1340 * For secureexec, reset the stack limit to sane default to
1341 * avoid bad behavior from the prior rlimits. This has to
1342 * happen before arch_pick_mmap_layout(), which examines
1343 * RLIMIT_STACK, but after the point of no return to avoid
1344 * needing to clean up the change on failure.
1346 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1347 bprm->rlim_stack.rlim_cur = _STK_LIM;
1350 me->sas_ss_sp = me->sas_ss_size = 0;
1353 * Figure out dumpability. Note that this checking only of current
1354 * is wrong, but userspace depends on it. This should be testing
1355 * bprm->secureexec instead.
1357 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1358 !(uid_eq(current_euid(), current_uid()) &&
1359 gid_eq(current_egid(), current_gid())))
1360 set_dumpable(current->mm, suid_dumpable);
1362 set_dumpable(current->mm, SUID_DUMP_USER);
1365 __set_task_comm(me, kbasename(bprm->filename), true);
1367 /* An exec changes our domain. We are no longer part of the thread
1369 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1370 flush_signal_handlers(me, 0);
1372 retval = set_cred_ucounts(bprm->cred);
1377 * install the new credentials for this executable
1379 security_bprm_committing_creds(bprm);
1381 commit_creds(bprm->cred);
1385 * Disable monitoring for regular users
1386 * when executing setuid binaries. Must
1387 * wait until new credentials are committed
1388 * by commit_creds() above
1390 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1391 perf_event_exit_task(me);
1393 * cred_guard_mutex must be held at least to this point to prevent
1394 * ptrace_attach() from altering our determination of the task's
1395 * credentials; any time after this it may be unlocked.
1397 security_bprm_committed_creds(bprm);
1399 /* Pass the opened binary to the interpreter. */
1400 if (bprm->have_execfd) {
1401 retval = get_unused_fd_flags(0);
1404 fd_install(retval, bprm->executable);
1405 bprm->executable = NULL;
1406 bprm->execfd = retval;
1411 up_write(&me->signal->exec_update_lock);
1415 EXPORT_SYMBOL(begin_new_exec);
1417 void would_dump(struct linux_binprm *bprm, struct file *file)
1419 struct inode *inode = file_inode(file);
1420 struct mnt_idmap *idmap = file_mnt_idmap(file);
1421 if (inode_permission(idmap, inode, MAY_READ) < 0) {
1422 struct user_namespace *old, *user_ns;
1423 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1425 /* Ensure mm->user_ns contains the executable */
1426 user_ns = old = bprm->mm->user_ns;
1427 while ((user_ns != &init_user_ns) &&
1428 !privileged_wrt_inode_uidgid(user_ns, idmap, inode))
1429 user_ns = user_ns->parent;
1431 if (old != user_ns) {
1432 bprm->mm->user_ns = get_user_ns(user_ns);
1437 EXPORT_SYMBOL(would_dump);
1439 void setup_new_exec(struct linux_binprm * bprm)
1441 /* Setup things that can depend upon the personality */
1442 struct task_struct *me = current;
1444 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1446 arch_setup_new_exec();
1448 /* Set the new mm task size. We have to do that late because it may
1449 * depend on TIF_32BIT which is only updated in flush_thread() on
1450 * some architectures like powerpc
1452 me->mm->task_size = TASK_SIZE;
1453 up_write(&me->signal->exec_update_lock);
1454 mutex_unlock(&me->signal->cred_guard_mutex);
1456 EXPORT_SYMBOL(setup_new_exec);
1458 /* Runs immediately before start_thread() takes over. */
1459 void finalize_exec(struct linux_binprm *bprm)
1461 /* Store any stack rlimit changes before starting thread. */
1462 task_lock(current->group_leader);
1463 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1464 task_unlock(current->group_leader);
1466 EXPORT_SYMBOL(finalize_exec);
1469 * Prepare credentials and lock ->cred_guard_mutex.
1470 * setup_new_exec() commits the new creds and drops the lock.
1471 * Or, if exec fails before, free_bprm() should release ->cred
1474 static int prepare_bprm_creds(struct linux_binprm *bprm)
1476 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1477 return -ERESTARTNOINTR;
1479 bprm->cred = prepare_exec_creds();
1480 if (likely(bprm->cred))
1483 mutex_unlock(¤t->signal->cred_guard_mutex);
1487 static void free_bprm(struct linux_binprm *bprm)
1490 acct_arg_size(bprm, 0);
1493 free_arg_pages(bprm);
1495 mutex_unlock(¤t->signal->cred_guard_mutex);
1496 abort_creds(bprm->cred);
1499 allow_write_access(bprm->file);
1502 if (bprm->executable)
1503 fput(bprm->executable);
1504 /* If a binfmt changed the interp, free it. */
1505 if (bprm->interp != bprm->filename)
1506 kfree(bprm->interp);
1507 kfree(bprm->fdpath);
1511 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1513 struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1514 int retval = -ENOMEM;
1518 if (fd == AT_FDCWD || filename->name[0] == '/') {
1519 bprm->filename = filename->name;
1521 if (filename->name[0] == '\0')
1522 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1524 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1525 fd, filename->name);
1529 bprm->filename = bprm->fdpath;
1531 bprm->interp = bprm->filename;
1533 retval = bprm_mm_init(bprm);
1541 return ERR_PTR(retval);
1544 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1546 /* If a binfmt changed the interp, free it first. */
1547 if (bprm->interp != bprm->filename)
1548 kfree(bprm->interp);
1549 bprm->interp = kstrdup(interp, GFP_KERNEL);
1554 EXPORT_SYMBOL(bprm_change_interp);
1557 * determine how safe it is to execute the proposed program
1558 * - the caller must hold ->cred_guard_mutex to protect against
1559 * PTRACE_ATTACH or seccomp thread-sync
1561 static void check_unsafe_exec(struct linux_binprm *bprm)
1563 struct task_struct *p = current, *t;
1567 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1570 * This isn't strictly necessary, but it makes it harder for LSMs to
1573 if (task_no_new_privs(current))
1574 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1577 * If another task is sharing our fs, we cannot safely
1578 * suid exec because the differently privileged task
1579 * will be able to manipulate the current directory, etc.
1580 * It would be nice to force an unshare instead...
1583 spin_lock(&p->fs->lock);
1585 for_other_threads(p, t) {
1591 if (p->fs->users > n_fs)
1592 bprm->unsafe |= LSM_UNSAFE_SHARE;
1595 spin_unlock(&p->fs->lock);
1598 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1600 /* Handle suid and sgid on files */
1601 struct mnt_idmap *idmap;
1602 struct inode *inode = file_inode(file);
1607 if (!mnt_may_suid(file->f_path.mnt))
1610 if (task_no_new_privs(current))
1613 mode = READ_ONCE(inode->i_mode);
1614 if (!(mode & (S_ISUID|S_ISGID)))
1617 idmap = file_mnt_idmap(file);
1619 /* Be careful if suid/sgid is set */
1622 /* reload atomically mode/uid/gid now that lock held */
1623 mode = inode->i_mode;
1624 vfsuid = i_uid_into_vfsuid(idmap, inode);
1625 vfsgid = i_gid_into_vfsgid(idmap, inode);
1626 inode_unlock(inode);
1628 /* We ignore suid/sgid if there are no mappings for them in the ns */
1629 if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) ||
1630 !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))
1633 if (mode & S_ISUID) {
1634 bprm->per_clear |= PER_CLEAR_ON_SETID;
1635 bprm->cred->euid = vfsuid_into_kuid(vfsuid);
1638 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1639 bprm->per_clear |= PER_CLEAR_ON_SETID;
1640 bprm->cred->egid = vfsgid_into_kgid(vfsgid);
1645 * Compute brpm->cred based upon the final binary.
1647 static int bprm_creds_from_file(struct linux_binprm *bprm)
1649 /* Compute creds based on which file? */
1650 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1652 bprm_fill_uid(bprm, file);
1653 return security_bprm_creds_from_file(bprm, file);
1657 * Fill the binprm structure from the inode.
1658 * Read the first BINPRM_BUF_SIZE bytes
1660 * This may be called multiple times for binary chains (scripts for example).
1662 static int prepare_binprm(struct linux_binprm *bprm)
1666 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1667 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1671 * Arguments are '\0' separated strings found at the location bprm->p
1672 * points to; chop off the first by relocating brpm->p to right after
1673 * the first '\0' encountered.
1675 int remove_arg_zero(struct linux_binprm *bprm)
1678 unsigned long offset;
1686 offset = bprm->p & ~PAGE_MASK;
1687 page = get_arg_page(bprm, bprm->p, 0);
1692 kaddr = kmap_local_page(page);
1694 for (; offset < PAGE_SIZE && kaddr[offset];
1695 offset++, bprm->p++)
1698 kunmap_local(kaddr);
1700 } while (offset == PAGE_SIZE);
1709 EXPORT_SYMBOL(remove_arg_zero);
1711 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1713 * cycle the list of binary formats handler, until one recognizes the image
1715 static int search_binary_handler(struct linux_binprm *bprm)
1717 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1718 struct linux_binfmt *fmt;
1721 retval = prepare_binprm(bprm);
1725 retval = security_bprm_check(bprm);
1731 read_lock(&binfmt_lock);
1732 list_for_each_entry(fmt, &formats, lh) {
1733 if (!try_module_get(fmt->module))
1735 read_unlock(&binfmt_lock);
1737 retval = fmt->load_binary(bprm);
1739 read_lock(&binfmt_lock);
1741 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1742 read_unlock(&binfmt_lock);
1746 read_unlock(&binfmt_lock);
1749 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1750 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1752 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1761 /* binfmt handlers will call back into begin_new_exec() on success. */
1762 static int exec_binprm(struct linux_binprm *bprm)
1764 pid_t old_pid, old_vpid;
1767 /* Need to fetch pid before load_binary changes it */
1768 old_pid = current->pid;
1770 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1773 /* This allows 4 levels of binfmt rewrites before failing hard. */
1774 for (depth = 0;; depth++) {
1779 ret = search_binary_handler(bprm);
1782 if (!bprm->interpreter)
1786 bprm->file = bprm->interpreter;
1787 bprm->interpreter = NULL;
1789 allow_write_access(exec);
1790 if (unlikely(bprm->have_execfd)) {
1791 if (bprm->executable) {
1795 bprm->executable = exec;
1801 trace_sched_process_exec(current, old_pid, bprm);
1802 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1803 proc_exec_connector(current);
1808 * sys_execve() executes a new program.
1810 static int bprm_execve(struct linux_binprm *bprm,
1811 int fd, struct filename *filename, int flags)
1816 retval = prepare_bprm_creds(bprm);
1821 * Check for unsafe execution states before exec_binprm(), which
1822 * will call back into begin_new_exec(), into bprm_creds_from_file(),
1823 * where setuid-ness is evaluated.
1825 check_unsafe_exec(bprm);
1826 current->in_execve = 1;
1827 sched_mm_cid_before_execve(current);
1829 file = do_open_execat(fd, filename, flags);
1830 retval = PTR_ERR(file);
1838 * Record that a name derived from an O_CLOEXEC fd will be
1839 * inaccessible after exec. This allows the code in exec to
1840 * choose to fail when the executable is not mmaped into the
1841 * interpreter and an open file descriptor is not passed to
1842 * the interpreter. This makes for a better user experience
1843 * than having the interpreter start and then immediately fail
1844 * when it finds the executable is inaccessible.
1846 if (bprm->fdpath && get_close_on_exec(fd))
1847 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1849 /* Set the unchanging part of bprm->cred */
1850 retval = security_bprm_creds_for_exec(bprm);
1854 retval = exec_binprm(bprm);
1858 sched_mm_cid_after_execve(current);
1859 /* execve succeeded */
1860 current->fs->in_exec = 0;
1861 current->in_execve = 0;
1862 rseq_execve(current);
1863 user_events_execve(current);
1864 acct_update_integrals(current);
1865 task_numa_free(current, false);
1870 * If past the point of no return ensure the code never
1871 * returns to the userspace process. Use an existing fatal
1872 * signal if present otherwise terminate the process with
1875 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1876 force_fatal_sig(SIGSEGV);
1879 sched_mm_cid_after_execve(current);
1880 current->fs->in_exec = 0;
1881 current->in_execve = 0;
1886 static int do_execveat_common(int fd, struct filename *filename,
1887 struct user_arg_ptr argv,
1888 struct user_arg_ptr envp,
1891 struct linux_binprm *bprm;
1894 if (IS_ERR(filename))
1895 return PTR_ERR(filename);
1898 * We move the actual failure in case of RLIMIT_NPROC excess from
1899 * set*uid() to execve() because too many poorly written programs
1900 * don't check setuid() return code. Here we additionally recheck
1901 * whether NPROC limit is still exceeded.
1903 if ((current->flags & PF_NPROC_EXCEEDED) &&
1904 is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1909 /* We're below the limit (still or again), so we don't want to make
1910 * further execve() calls fail. */
1911 current->flags &= ~PF_NPROC_EXCEEDED;
1913 bprm = alloc_bprm(fd, filename);
1915 retval = PTR_ERR(bprm);
1919 retval = count(argv, MAX_ARG_STRINGS);
1921 pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1922 current->comm, bprm->filename);
1925 bprm->argc = retval;
1927 retval = count(envp, MAX_ARG_STRINGS);
1930 bprm->envc = retval;
1932 retval = bprm_stack_limits(bprm);
1936 retval = copy_string_kernel(bprm->filename, bprm);
1939 bprm->exec = bprm->p;
1941 retval = copy_strings(bprm->envc, envp, bprm);
1945 retval = copy_strings(bprm->argc, argv, bprm);
1950 * When argv is empty, add an empty string ("") as argv[0] to
1951 * ensure confused userspace programs that start processing
1952 * from argv[1] won't end up walking envp. See also
1953 * bprm_stack_limits().
1955 if (bprm->argc == 0) {
1956 retval = copy_string_kernel("", bprm);
1962 retval = bprm_execve(bprm, fd, filename, flags);
1971 int kernel_execve(const char *kernel_filename,
1972 const char *const *argv, const char *const *envp)
1974 struct filename *filename;
1975 struct linux_binprm *bprm;
1979 /* It is non-sense for kernel threads to call execve */
1980 if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1983 filename = getname_kernel(kernel_filename);
1984 if (IS_ERR(filename))
1985 return PTR_ERR(filename);
1987 bprm = alloc_bprm(fd, filename);
1989 retval = PTR_ERR(bprm);
1993 retval = count_strings_kernel(argv);
1994 if (WARN_ON_ONCE(retval == 0))
1998 bprm->argc = retval;
2000 retval = count_strings_kernel(envp);
2003 bprm->envc = retval;
2005 retval = bprm_stack_limits(bprm);
2009 retval = copy_string_kernel(bprm->filename, bprm);
2012 bprm->exec = bprm->p;
2014 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2018 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2022 retval = bprm_execve(bprm, fd, filename, 0);
2030 static int do_execve(struct filename *filename,
2031 const char __user *const __user *__argv,
2032 const char __user *const __user *__envp)
2034 struct user_arg_ptr argv = { .ptr.native = __argv };
2035 struct user_arg_ptr envp = { .ptr.native = __envp };
2036 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2039 static int do_execveat(int fd, struct filename *filename,
2040 const char __user *const __user *__argv,
2041 const char __user *const __user *__envp,
2044 struct user_arg_ptr argv = { .ptr.native = __argv };
2045 struct user_arg_ptr envp = { .ptr.native = __envp };
2047 return do_execveat_common(fd, filename, argv, envp, flags);
2050 #ifdef CONFIG_COMPAT
2051 static int compat_do_execve(struct filename *filename,
2052 const compat_uptr_t __user *__argv,
2053 const compat_uptr_t __user *__envp)
2055 struct user_arg_ptr argv = {
2057 .ptr.compat = __argv,
2059 struct user_arg_ptr envp = {
2061 .ptr.compat = __envp,
2063 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2066 static int compat_do_execveat(int fd, struct filename *filename,
2067 const compat_uptr_t __user *__argv,
2068 const compat_uptr_t __user *__envp,
2071 struct user_arg_ptr argv = {
2073 .ptr.compat = __argv,
2075 struct user_arg_ptr envp = {
2077 .ptr.compat = __envp,
2079 return do_execveat_common(fd, filename, argv, envp, flags);
2083 void set_binfmt(struct linux_binfmt *new)
2085 struct mm_struct *mm = current->mm;
2088 module_put(mm->binfmt->module);
2092 __module_get(new->module);
2094 EXPORT_SYMBOL(set_binfmt);
2097 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2099 void set_dumpable(struct mm_struct *mm, int value)
2101 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2104 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2107 SYSCALL_DEFINE3(execve,
2108 const char __user *, filename,
2109 const char __user *const __user *, argv,
2110 const char __user *const __user *, envp)
2112 return do_execve(getname(filename), argv, envp);
2115 SYSCALL_DEFINE5(execveat,
2116 int, fd, const char __user *, filename,
2117 const char __user *const __user *, argv,
2118 const char __user *const __user *, envp,
2121 return do_execveat(fd,
2122 getname_uflags(filename, flags),
2126 #ifdef CONFIG_COMPAT
2127 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2128 const compat_uptr_t __user *, argv,
2129 const compat_uptr_t __user *, envp)
2131 return compat_do_execve(getname(filename), argv, envp);
2134 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2135 const char __user *, filename,
2136 const compat_uptr_t __user *, argv,
2137 const compat_uptr_t __user *, envp,
2140 return compat_do_execveat(fd,
2141 getname_uflags(filename, flags),
2146 #ifdef CONFIG_SYSCTL
2148 static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2149 void *buffer, size_t *lenp, loff_t *ppos)
2151 int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2154 validate_coredump_safety();
2158 static struct ctl_table fs_exec_sysctls[] = {
2160 .procname = "suid_dumpable",
2161 .data = &suid_dumpable,
2162 .maxlen = sizeof(int),
2164 .proc_handler = proc_dointvec_minmax_coredump,
2165 .extra1 = SYSCTL_ZERO,
2166 .extra2 = SYSCTL_TWO,
2170 static int __init init_fs_exec_sysctls(void)
2172 register_sysctl_init("fs", fs_exec_sysctls);
2176 fs_initcall(init_fs_exec_sysctls);
2177 #endif /* CONFIG_SYSCTL */