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/vmacache.h>
32 #include <linux/stat.h>
33 #include <linux/fcntl.h>
34 #include <linux/swap.h>
35 #include <linux/string.h>
36 #include <linux/init.h>
37 #include <linux/sched/mm.h>
38 #include <linux/sched/coredump.h>
39 #include <linux/sched/signal.h>
40 #include <linux/sched/numa_balancing.h>
41 #include <linux/sched/task.h>
42 #include <linux/pagemap.h>
43 #include <linux/perf_event.h>
44 #include <linux/highmem.h>
45 #include <linux/spinlock.h>
46 #include <linux/key.h>
47 #include <linux/personality.h>
48 #include <linux/binfmts.h>
49 #include <linux/utsname.h>
50 #include <linux/pid_namespace.h>
51 #include <linux/module.h>
52 #include <linux/namei.h>
53 #include <linux/mount.h>
54 #include <linux/security.h>
55 #include <linux/syscalls.h>
56 #include <linux/tsacct_kern.h>
57 #include <linux/cn_proc.h>
58 #include <linux/audit.h>
59 #include <linux/tracehook.h>
60 #include <linux/kmod.h>
61 #include <linux/fsnotify.h>
62 #include <linux/fs_struct.h>
63 #include <linux/oom.h>
64 #include <linux/compat.h>
65 #include <linux/vmalloc.h>
67 #include <linux/uaccess.h>
68 #include <asm/mmu_context.h>
71 #include <trace/events/task.h>
74 #include <trace/events/sched.h>
76 static int bprm_creds_from_file(struct linux_binprm *bprm);
78 int suid_dumpable = 0;
80 static LIST_HEAD(formats);
81 static DEFINE_RWLOCK(binfmt_lock);
83 void __register_binfmt(struct linux_binfmt * fmt, int insert)
86 if (WARN_ON(!fmt->load_binary))
88 write_lock(&binfmt_lock);
89 insert ? list_add(&fmt->lh, &formats) :
90 list_add_tail(&fmt->lh, &formats);
91 write_unlock(&binfmt_lock);
94 EXPORT_SYMBOL(__register_binfmt);
96 void unregister_binfmt(struct linux_binfmt * fmt)
98 write_lock(&binfmt_lock);
100 write_unlock(&binfmt_lock);
103 EXPORT_SYMBOL(unregister_binfmt);
105 static inline void put_binfmt(struct linux_binfmt * fmt)
107 module_put(fmt->module);
110 bool path_noexec(const struct path *path)
112 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
113 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
118 * Note that a shared library must be both readable and executable due to
121 * Also note that we take the address to load from from the file itself.
123 SYSCALL_DEFINE1(uselib, const char __user *, library)
125 struct linux_binfmt *fmt;
127 struct filename *tmp = getname(library);
128 int error = PTR_ERR(tmp);
129 static const struct open_flags uselib_flags = {
130 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
131 .acc_mode = MAY_READ | MAY_EXEC,
132 .intent = LOOKUP_OPEN,
133 .lookup_flags = LOOKUP_FOLLOW,
139 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
141 error = PTR_ERR(file);
146 * may_open() has already checked for this, so it should be
147 * impossible to trip now. But we need to be extra cautious
148 * and check again at the very end too.
151 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
152 path_noexec(&file->f_path)))
159 read_lock(&binfmt_lock);
160 list_for_each_entry(fmt, &formats, lh) {
161 if (!fmt->load_shlib)
163 if (!try_module_get(fmt->module))
165 read_unlock(&binfmt_lock);
166 error = fmt->load_shlib(file);
167 read_lock(&binfmt_lock);
169 if (error != -ENOEXEC)
172 read_unlock(&binfmt_lock);
178 #endif /* #ifdef CONFIG_USELIB */
182 * The nascent bprm->mm is not visible until exec_mmap() but it can
183 * use a lot of memory, account these pages in current->mm temporary
184 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
185 * change the counter back via acct_arg_size(0).
187 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
189 struct mm_struct *mm = current->mm;
190 long diff = (long)(pages - bprm->vma_pages);
195 bprm->vma_pages = pages;
196 add_mm_counter(mm, MM_ANONPAGES, diff);
199 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
204 unsigned int gup_flags = FOLL_FORCE;
206 #ifdef CONFIG_STACK_GROWSUP
208 ret = expand_downwards(bprm->vma, pos);
215 gup_flags |= FOLL_WRITE;
218 * We are doing an exec(). 'current' is the process
219 * doing the exec and bprm->mm is the new process's mm.
221 ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
227 acct_arg_size(bprm, vma_pages(bprm->vma));
232 static void put_arg_page(struct page *page)
237 static void free_arg_pages(struct linux_binprm *bprm)
241 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
244 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
247 static int __bprm_mm_init(struct linux_binprm *bprm)
250 struct vm_area_struct *vma = NULL;
251 struct mm_struct *mm = bprm->mm;
253 bprm->vma = vma = vm_area_alloc(mm);
256 vma_set_anonymous(vma);
258 if (mmap_write_lock_killable(mm)) {
264 * Place the stack at the largest stack address the architecture
265 * supports. Later, we'll move this to an appropriate place. We don't
266 * use STACK_TOP because that can depend on attributes which aren't
269 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
270 vma->vm_end = STACK_TOP_MAX;
271 vma->vm_start = vma->vm_end - PAGE_SIZE;
272 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
273 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
275 err = insert_vm_struct(mm, vma);
279 mm->stack_vm = mm->total_vm = 1;
280 mmap_write_unlock(mm);
281 bprm->p = vma->vm_end - sizeof(void *);
284 mmap_write_unlock(mm);
291 static bool valid_arg_len(struct linux_binprm *bprm, long len)
293 return len <= MAX_ARG_STRLEN;
298 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
302 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
307 page = bprm->page[pos / PAGE_SIZE];
308 if (!page && write) {
309 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
312 bprm->page[pos / PAGE_SIZE] = page;
318 static void put_arg_page(struct page *page)
322 static void free_arg_page(struct linux_binprm *bprm, int i)
325 __free_page(bprm->page[i]);
326 bprm->page[i] = NULL;
330 static void free_arg_pages(struct linux_binprm *bprm)
334 for (i = 0; i < MAX_ARG_PAGES; i++)
335 free_arg_page(bprm, i);
338 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
343 static int __bprm_mm_init(struct linux_binprm *bprm)
345 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
349 static bool valid_arg_len(struct linux_binprm *bprm, long len)
351 return len <= bprm->p;
354 #endif /* CONFIG_MMU */
357 * Create a new mm_struct and populate it with a temporary stack
358 * vm_area_struct. We don't have enough context at this point to set the stack
359 * flags, permissions, and offset, so we use temporary values. We'll update
360 * them later in setup_arg_pages().
362 static int bprm_mm_init(struct linux_binprm *bprm)
365 struct mm_struct *mm = NULL;
367 bprm->mm = mm = mm_alloc();
372 /* Save current stack limit for all calculations made during exec. */
373 task_lock(current->group_leader);
374 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
375 task_unlock(current->group_leader);
377 err = __bprm_mm_init(bprm);
392 struct user_arg_ptr {
397 const char __user *const __user *native;
399 const compat_uptr_t __user *compat;
404 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
406 const char __user *native;
409 if (unlikely(argv.is_compat)) {
410 compat_uptr_t compat;
412 if (get_user(compat, argv.ptr.compat + nr))
413 return ERR_PTR(-EFAULT);
415 return compat_ptr(compat);
419 if (get_user(native, argv.ptr.native + nr))
420 return ERR_PTR(-EFAULT);
426 * count() counts the number of strings in array ARGV.
428 static int count(struct user_arg_ptr argv, int max)
432 if (argv.ptr.native != NULL) {
434 const char __user *p = get_user_arg_ptr(argv, i);
446 if (fatal_signal_pending(current))
447 return -ERESTARTNOHAND;
454 static int count_strings_kernel(const char *const *argv)
461 for (i = 0; argv[i]; ++i) {
462 if (i >= MAX_ARG_STRINGS)
464 if (fatal_signal_pending(current))
465 return -ERESTARTNOHAND;
471 static int bprm_stack_limits(struct linux_binprm *bprm)
473 unsigned long limit, ptr_size;
476 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
477 * (whichever is smaller) for the argv+env strings.
479 * - the remaining binfmt code will not run out of stack space,
480 * - the program will have a reasonable amount of stack left
483 limit = _STK_LIM / 4 * 3;
484 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
486 * We've historically supported up to 32 pages (ARG_MAX)
487 * of argument strings even with small stacks
489 limit = max_t(unsigned long, limit, ARG_MAX);
491 * We must account for the size of all the argv and envp pointers to
492 * the argv and envp strings, since they will also take up space in
493 * the stack. They aren't stored until much later when we can't
494 * signal to the parent that the child has run out of stack space.
495 * Instead, calculate it here so it's possible to fail gracefully.
497 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
498 if (limit <= ptr_size)
502 bprm->argmin = bprm->p - limit;
507 * 'copy_strings()' copies argument/environment strings from the old
508 * processes's memory to the new process's stack. The call to get_user_pages()
509 * ensures the destination page is created and not swapped out.
511 static int copy_strings(int argc, struct user_arg_ptr argv,
512 struct linux_binprm *bprm)
514 struct page *kmapped_page = NULL;
516 unsigned long kpos = 0;
520 const char __user *str;
525 str = get_user_arg_ptr(argv, argc);
529 len = strnlen_user(str, MAX_ARG_STRLEN);
534 if (!valid_arg_len(bprm, len))
537 /* We're going to work our way backwords. */
542 if (bprm->p < bprm->argmin)
547 int offset, bytes_to_copy;
549 if (fatal_signal_pending(current)) {
550 ret = -ERESTARTNOHAND;
555 offset = pos % PAGE_SIZE;
559 bytes_to_copy = offset;
560 if (bytes_to_copy > len)
563 offset -= bytes_to_copy;
564 pos -= bytes_to_copy;
565 str -= bytes_to_copy;
566 len -= bytes_to_copy;
568 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
571 page = get_arg_page(bprm, pos, 1);
578 flush_kernel_dcache_page(kmapped_page);
579 kunmap(kmapped_page);
580 put_arg_page(kmapped_page);
583 kaddr = kmap(kmapped_page);
584 kpos = pos & PAGE_MASK;
585 flush_arg_page(bprm, kpos, kmapped_page);
587 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
596 flush_kernel_dcache_page(kmapped_page);
597 kunmap(kmapped_page);
598 put_arg_page(kmapped_page);
604 * Copy and argument/environment string from the kernel to the processes stack.
606 int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
608 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
609 unsigned long pos = bprm->p;
613 if (!valid_arg_len(bprm, len))
616 /* We're going to work our way backwards. */
619 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
623 unsigned int bytes_to_copy = min_t(unsigned int, len,
624 min_not_zero(offset_in_page(pos), PAGE_SIZE));
628 pos -= bytes_to_copy;
629 arg -= bytes_to_copy;
630 len -= bytes_to_copy;
632 page = get_arg_page(bprm, pos, 1);
635 kaddr = kmap_atomic(page);
636 flush_arg_page(bprm, pos & PAGE_MASK, page);
637 memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
638 flush_kernel_dcache_page(page);
639 kunmap_atomic(kaddr);
645 EXPORT_SYMBOL(copy_string_kernel);
647 static int copy_strings_kernel(int argc, const char *const *argv,
648 struct linux_binprm *bprm)
651 int ret = copy_string_kernel(argv[argc], bprm);
654 if (fatal_signal_pending(current))
655 return -ERESTARTNOHAND;
664 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
665 * the binfmt code determines where the new stack should reside, we shift it to
666 * its final location. The process proceeds as follows:
668 * 1) Use shift to calculate the new vma endpoints.
669 * 2) Extend vma to cover both the old and new ranges. This ensures the
670 * arguments passed to subsequent functions are consistent.
671 * 3) Move vma's page tables to the new range.
672 * 4) Free up any cleared pgd range.
673 * 5) Shrink the vma to cover only the new range.
675 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
677 struct mm_struct *mm = vma->vm_mm;
678 unsigned long old_start = vma->vm_start;
679 unsigned long old_end = vma->vm_end;
680 unsigned long length = old_end - old_start;
681 unsigned long new_start = old_start - shift;
682 unsigned long new_end = old_end - shift;
683 struct mmu_gather tlb;
685 BUG_ON(new_start > new_end);
688 * ensure there are no vmas between where we want to go
691 if (vma != find_vma(mm, new_start))
695 * cover the whole range: [new_start, old_end)
697 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
701 * move the page tables downwards, on failure we rely on
702 * process cleanup to remove whatever mess we made.
704 if (length != move_page_tables(vma, old_start,
705 vma, new_start, length, false))
709 tlb_gather_mmu(&tlb, mm, old_start, old_end);
710 if (new_end > old_start) {
712 * when the old and new regions overlap clear from new_end.
714 free_pgd_range(&tlb, new_end, old_end, new_end,
715 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
718 * otherwise, clean from old_start; this is done to not touch
719 * the address space in [new_end, old_start) some architectures
720 * have constraints on va-space that make this illegal (IA64) -
721 * for the others its just a little faster.
723 free_pgd_range(&tlb, old_start, old_end, new_end,
724 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
726 tlb_finish_mmu(&tlb, old_start, old_end);
729 * Shrink the vma to just the new range. Always succeeds.
731 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
737 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
738 * the stack is optionally relocated, and some extra space is added.
740 int setup_arg_pages(struct linux_binprm *bprm,
741 unsigned long stack_top,
742 int executable_stack)
745 unsigned long stack_shift;
746 struct mm_struct *mm = current->mm;
747 struct vm_area_struct *vma = bprm->vma;
748 struct vm_area_struct *prev = NULL;
749 unsigned long vm_flags;
750 unsigned long stack_base;
751 unsigned long stack_size;
752 unsigned long stack_expand;
753 unsigned long rlim_stack;
755 #ifdef CONFIG_STACK_GROWSUP
756 /* Limit stack size */
757 stack_base = bprm->rlim_stack.rlim_max;
758 if (stack_base > STACK_SIZE_MAX)
759 stack_base = STACK_SIZE_MAX;
761 /* Add space for stack randomization. */
762 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
764 /* Make sure we didn't let the argument array grow too large. */
765 if (vma->vm_end - vma->vm_start > stack_base)
768 stack_base = PAGE_ALIGN(stack_top - stack_base);
770 stack_shift = vma->vm_start - stack_base;
771 mm->arg_start = bprm->p - stack_shift;
772 bprm->p = vma->vm_end - stack_shift;
774 stack_top = arch_align_stack(stack_top);
775 stack_top = PAGE_ALIGN(stack_top);
777 if (unlikely(stack_top < mmap_min_addr) ||
778 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
781 stack_shift = vma->vm_end - stack_top;
783 bprm->p -= stack_shift;
784 mm->arg_start = bprm->p;
788 bprm->loader -= stack_shift;
789 bprm->exec -= stack_shift;
791 if (mmap_write_lock_killable(mm))
794 vm_flags = VM_STACK_FLAGS;
797 * Adjust stack execute permissions; explicitly enable for
798 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
799 * (arch default) otherwise.
801 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
803 else if (executable_stack == EXSTACK_DISABLE_X)
804 vm_flags &= ~VM_EXEC;
805 vm_flags |= mm->def_flags;
806 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
808 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
814 if (unlikely(vm_flags & VM_EXEC)) {
815 pr_warn_once("process '%pD4' started with executable stack\n",
819 /* Move stack pages down in memory. */
821 ret = shift_arg_pages(vma, stack_shift);
826 /* mprotect_fixup is overkill to remove the temporary stack flags */
827 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
829 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
830 stack_size = vma->vm_end - vma->vm_start;
832 * Align this down to a page boundary as expand_stack
835 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
836 #ifdef CONFIG_STACK_GROWSUP
837 if (stack_size + stack_expand > rlim_stack)
838 stack_base = vma->vm_start + rlim_stack;
840 stack_base = vma->vm_end + stack_expand;
842 if (stack_size + stack_expand > rlim_stack)
843 stack_base = vma->vm_end - rlim_stack;
845 stack_base = vma->vm_start - stack_expand;
847 current->mm->start_stack = bprm->p;
848 ret = expand_stack(vma, stack_base);
853 mmap_write_unlock(mm);
856 EXPORT_SYMBOL(setup_arg_pages);
861 * Transfer the program arguments and environment from the holding pages
862 * onto the stack. The provided stack pointer is adjusted accordingly.
864 int transfer_args_to_stack(struct linux_binprm *bprm,
865 unsigned long *sp_location)
867 unsigned long index, stop, sp;
870 stop = bprm->p >> PAGE_SHIFT;
873 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
874 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
875 char *src = kmap(bprm->page[index]) + offset;
876 sp -= PAGE_SIZE - offset;
877 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
879 kunmap(bprm->page[index]);
889 EXPORT_SYMBOL(transfer_args_to_stack);
891 #endif /* CONFIG_MMU */
893 static struct file *do_open_execat(int fd, struct filename *name, int flags)
897 struct open_flags open_exec_flags = {
898 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
899 .acc_mode = MAY_EXEC,
900 .intent = LOOKUP_OPEN,
901 .lookup_flags = LOOKUP_FOLLOW,
904 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
905 return ERR_PTR(-EINVAL);
906 if (flags & AT_SYMLINK_NOFOLLOW)
907 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
908 if (flags & AT_EMPTY_PATH)
909 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
911 file = do_filp_open(fd, name, &open_exec_flags);
916 * may_open() has already checked for this, so it should be
917 * impossible to trip now. But we need to be extra cautious
918 * and check again at the very end too.
921 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
922 path_noexec(&file->f_path)))
925 err = deny_write_access(file);
929 if (name->name[0] != '\0')
940 struct file *open_exec(const char *name)
942 struct filename *filename = getname_kernel(name);
943 struct file *f = ERR_CAST(filename);
945 if (!IS_ERR(filename)) {
946 f = do_open_execat(AT_FDCWD, filename, 0);
951 EXPORT_SYMBOL(open_exec);
953 int kernel_read_file(struct file *file, void **buf, loff_t *size,
954 loff_t max_size, enum kernel_read_file_id id)
958 void *allocated = NULL;
961 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
964 ret = deny_write_access(file);
968 ret = security_kernel_read_file(file, id);
972 i_size = i_size_read(file_inode(file));
977 if (i_size > SIZE_MAX || (max_size > 0 && i_size > max_size)) {
983 *buf = allocated = vmalloc(i_size);
990 while (pos < i_size) {
991 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
1001 if (pos != i_size) {
1006 ret = security_kernel_post_read_file(file, *buf, i_size, id);
1019 allow_write_access(file);
1022 EXPORT_SYMBOL_GPL(kernel_read_file);
1024 int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
1025 loff_t max_size, enum kernel_read_file_id id)
1030 if (!path || !*path)
1033 file = filp_open(path, O_RDONLY, 0);
1035 return PTR_ERR(file);
1037 ret = kernel_read_file(file, buf, size, max_size, id);
1041 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
1043 int kernel_read_file_from_path_initns(const char *path, void **buf,
1044 loff_t *size, loff_t max_size,
1045 enum kernel_read_file_id id)
1051 if (!path || !*path)
1054 task_lock(&init_task);
1055 get_fs_root(init_task.fs, &root);
1056 task_unlock(&init_task);
1058 file = file_open_root(root.dentry, root.mnt, path, O_RDONLY, 0);
1061 return PTR_ERR(file);
1063 ret = kernel_read_file(file, buf, size, max_size, id);
1067 EXPORT_SYMBOL_GPL(kernel_read_file_from_path_initns);
1069 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
1070 enum kernel_read_file_id id)
1072 struct fd f = fdget(fd);
1078 ret = kernel_read_file(f.file, buf, size, max_size, id);
1083 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
1085 #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
1086 defined(CONFIG_BINFMT_ELF_FDPIC)
1087 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1089 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1091 flush_icache_user_range(addr, addr + len);
1094 EXPORT_SYMBOL(read_code);
1098 * Maps the mm_struct mm into the current task struct.
1099 * On success, this function returns with the mutex
1100 * exec_update_mutex locked.
1102 static int exec_mmap(struct mm_struct *mm)
1104 struct task_struct *tsk;
1105 struct mm_struct *old_mm, *active_mm;
1108 /* Notify parent that we're no longer interested in the old VM */
1110 old_mm = current->mm;
1111 exec_mm_release(tsk, old_mm);
1113 sync_mm_rss(old_mm);
1115 ret = mutex_lock_killable(&tsk->signal->exec_update_mutex);
1121 * Make sure that if there is a core dump in progress
1122 * for the old mm, we get out and die instead of going
1123 * through with the exec. We must hold mmap_lock around
1124 * checking core_state and changing tsk->mm.
1126 mmap_read_lock(old_mm);
1127 if (unlikely(old_mm->core_state)) {
1128 mmap_read_unlock(old_mm);
1129 mutex_unlock(&tsk->signal->exec_update_mutex);
1135 active_mm = tsk->active_mm;
1136 membarrier_exec_mmap(mm);
1138 tsk->active_mm = mm;
1139 activate_mm(active_mm, mm);
1140 tsk->mm->vmacache_seqnum = 0;
1141 vmacache_flush(tsk);
1144 mmap_read_unlock(old_mm);
1145 BUG_ON(active_mm != old_mm);
1146 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1147 mm_update_next_owner(old_mm);
1155 static int de_thread(struct task_struct *tsk)
1157 struct signal_struct *sig = tsk->signal;
1158 struct sighand_struct *oldsighand = tsk->sighand;
1159 spinlock_t *lock = &oldsighand->siglock;
1161 if (thread_group_empty(tsk))
1162 goto no_thread_group;
1165 * Kill all other threads in the thread group.
1167 spin_lock_irq(lock);
1168 if (signal_group_exit(sig)) {
1170 * Another group action in progress, just
1171 * return so that the signal is processed.
1173 spin_unlock_irq(lock);
1177 sig->group_exit_task = tsk;
1178 sig->notify_count = zap_other_threads(tsk);
1179 if (!thread_group_leader(tsk))
1180 sig->notify_count--;
1182 while (sig->notify_count) {
1183 __set_current_state(TASK_KILLABLE);
1184 spin_unlock_irq(lock);
1186 if (__fatal_signal_pending(tsk))
1188 spin_lock_irq(lock);
1190 spin_unlock_irq(lock);
1193 * At this point all other threads have exited, all we have to
1194 * do is to wait for the thread group leader to become inactive,
1195 * and to assume its PID:
1197 if (!thread_group_leader(tsk)) {
1198 struct task_struct *leader = tsk->group_leader;
1201 cgroup_threadgroup_change_begin(tsk);
1202 write_lock_irq(&tasklist_lock);
1204 * Do this under tasklist_lock to ensure that
1205 * exit_notify() can't miss ->group_exit_task
1207 sig->notify_count = -1;
1208 if (likely(leader->exit_state))
1210 __set_current_state(TASK_KILLABLE);
1211 write_unlock_irq(&tasklist_lock);
1212 cgroup_threadgroup_change_end(tsk);
1214 if (__fatal_signal_pending(tsk))
1219 * The only record we have of the real-time age of a
1220 * process, regardless of execs it's done, is start_time.
1221 * All the past CPU time is accumulated in signal_struct
1222 * from sister threads now dead. But in this non-leader
1223 * exec, nothing survives from the original leader thread,
1224 * whose birth marks the true age of this process now.
1225 * When we take on its identity by switching to its PID, we
1226 * also take its birthdate (always earlier than our own).
1228 tsk->start_time = leader->start_time;
1229 tsk->start_boottime = leader->start_boottime;
1231 BUG_ON(!same_thread_group(leader, tsk));
1233 * An exec() starts a new thread group with the
1234 * TGID of the previous thread group. Rehash the
1235 * two threads with a switched PID, and release
1236 * the former thread group leader:
1239 /* Become a process group leader with the old leader's pid.
1240 * The old leader becomes a thread of the this thread group.
1242 exchange_tids(tsk, leader);
1243 transfer_pid(leader, tsk, PIDTYPE_TGID);
1244 transfer_pid(leader, tsk, PIDTYPE_PGID);
1245 transfer_pid(leader, tsk, PIDTYPE_SID);
1247 list_replace_rcu(&leader->tasks, &tsk->tasks);
1248 list_replace_init(&leader->sibling, &tsk->sibling);
1250 tsk->group_leader = tsk;
1251 leader->group_leader = tsk;
1253 tsk->exit_signal = SIGCHLD;
1254 leader->exit_signal = -1;
1256 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1257 leader->exit_state = EXIT_DEAD;
1260 * We are going to release_task()->ptrace_unlink() silently,
1261 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1262 * the tracer wont't block again waiting for this thread.
1264 if (unlikely(leader->ptrace))
1265 __wake_up_parent(leader, leader->parent);
1266 write_unlock_irq(&tasklist_lock);
1267 cgroup_threadgroup_change_end(tsk);
1269 release_task(leader);
1272 sig->group_exit_task = NULL;
1273 sig->notify_count = 0;
1276 /* we have changed execution domain */
1277 tsk->exit_signal = SIGCHLD;
1279 BUG_ON(!thread_group_leader(tsk));
1283 /* protects against exit_notify() and __exit_signal() */
1284 read_lock(&tasklist_lock);
1285 sig->group_exit_task = NULL;
1286 sig->notify_count = 0;
1287 read_unlock(&tasklist_lock);
1293 * This function makes sure the current process has its own signal table,
1294 * so that flush_signal_handlers can later reset the handlers without
1295 * disturbing other processes. (Other processes might share the signal
1296 * table via the CLONE_SIGHAND option to clone().)
1298 static int unshare_sighand(struct task_struct *me)
1300 struct sighand_struct *oldsighand = me->sighand;
1302 if (refcount_read(&oldsighand->count) != 1) {
1303 struct sighand_struct *newsighand;
1305 * This ->sighand is shared with the CLONE_SIGHAND
1306 * but not CLONE_THREAD task, switch to the new one.
1308 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1312 refcount_set(&newsighand->count, 1);
1313 memcpy(newsighand->action, oldsighand->action,
1314 sizeof(newsighand->action));
1316 write_lock_irq(&tasklist_lock);
1317 spin_lock(&oldsighand->siglock);
1318 rcu_assign_pointer(me->sighand, newsighand);
1319 spin_unlock(&oldsighand->siglock);
1320 write_unlock_irq(&tasklist_lock);
1322 __cleanup_sighand(oldsighand);
1327 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1330 strncpy(buf, tsk->comm, buf_size);
1334 EXPORT_SYMBOL_GPL(__get_task_comm);
1337 * These functions flushes out all traces of the currently running executable
1338 * so that a new one can be started
1341 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1344 trace_task_rename(tsk, buf);
1345 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1347 perf_event_comm(tsk, exec);
1351 * Calling this is the point of no return. None of the failures will be
1352 * seen by userspace since either the process is already taking a fatal
1353 * signal (via de_thread() or coredump), or will have SEGV raised
1354 * (after exec_mmap()) by search_binary_handler (see below).
1356 int begin_new_exec(struct linux_binprm * bprm)
1358 struct task_struct *me = current;
1361 /* Once we are committed compute the creds */
1362 retval = bprm_creds_from_file(bprm);
1367 * Ensure all future errors are fatal.
1369 bprm->point_of_no_return = true;
1372 * Make this the only thread in the thread group.
1374 retval = de_thread(me);
1379 * Must be called _before_ exec_mmap() as bprm->mm is
1380 * not visibile until then. This also enables the update
1383 set_mm_exe_file(bprm->mm, bprm->file);
1385 /* If the binary is not readable then enforce mm->dumpable=0 */
1386 would_dump(bprm, bprm->file);
1387 if (bprm->have_execfd)
1388 would_dump(bprm, bprm->executable);
1391 * Release all of the old mmap stuff
1393 acct_arg_size(bprm, 0);
1394 retval = exec_mmap(bprm->mm);
1400 #ifdef CONFIG_POSIX_TIMERS
1401 exit_itimers(me->signal);
1402 flush_itimer_signals();
1406 * Make the signal table private.
1408 retval = unshare_sighand(me);
1413 * Ensure that the uaccess routines can actually operate on userspace
1416 force_uaccess_begin();
1418 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1419 PF_NOFREEZE | PF_NO_SETAFFINITY);
1421 me->personality &= ~bprm->per_clear;
1424 * We have to apply CLOEXEC before we change whether the process is
1425 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1426 * trying to access the should-be-closed file descriptors of a process
1427 * undergoing exec(2).
1429 do_close_on_exec(me->files);
1431 if (bprm->secureexec) {
1432 /* Make sure parent cannot signal privileged process. */
1433 me->pdeath_signal = 0;
1436 * For secureexec, reset the stack limit to sane default to
1437 * avoid bad behavior from the prior rlimits. This has to
1438 * happen before arch_pick_mmap_layout(), which examines
1439 * RLIMIT_STACK, but after the point of no return to avoid
1440 * needing to clean up the change on failure.
1442 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1443 bprm->rlim_stack.rlim_cur = _STK_LIM;
1446 me->sas_ss_sp = me->sas_ss_size = 0;
1449 * Figure out dumpability. Note that this checking only of current
1450 * is wrong, but userspace depends on it. This should be testing
1451 * bprm->secureexec instead.
1453 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1454 !(uid_eq(current_euid(), current_uid()) &&
1455 gid_eq(current_egid(), current_gid())))
1456 set_dumpable(current->mm, suid_dumpable);
1458 set_dumpable(current->mm, SUID_DUMP_USER);
1461 __set_task_comm(me, kbasename(bprm->filename), true);
1463 /* An exec changes our domain. We are no longer part of the thread
1465 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1466 flush_signal_handlers(me, 0);
1469 * install the new credentials for this executable
1471 security_bprm_committing_creds(bprm);
1473 commit_creds(bprm->cred);
1477 * Disable monitoring for regular users
1478 * when executing setuid binaries. Must
1479 * wait until new credentials are committed
1480 * by commit_creds() above
1482 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1483 perf_event_exit_task(me);
1485 * cred_guard_mutex must be held at least to this point to prevent
1486 * ptrace_attach() from altering our determination of the task's
1487 * credentials; any time after this it may be unlocked.
1489 security_bprm_committed_creds(bprm);
1491 /* Pass the opened binary to the interpreter. */
1492 if (bprm->have_execfd) {
1493 retval = get_unused_fd_flags(0);
1496 fd_install(retval, bprm->executable);
1497 bprm->executable = NULL;
1498 bprm->execfd = retval;
1503 mutex_unlock(&me->signal->exec_update_mutex);
1507 EXPORT_SYMBOL(begin_new_exec);
1509 void would_dump(struct linux_binprm *bprm, struct file *file)
1511 struct inode *inode = file_inode(file);
1512 if (inode_permission(inode, MAY_READ) < 0) {
1513 struct user_namespace *old, *user_ns;
1514 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1516 /* Ensure mm->user_ns contains the executable */
1517 user_ns = old = bprm->mm->user_ns;
1518 while ((user_ns != &init_user_ns) &&
1519 !privileged_wrt_inode_uidgid(user_ns, inode))
1520 user_ns = user_ns->parent;
1522 if (old != user_ns) {
1523 bprm->mm->user_ns = get_user_ns(user_ns);
1528 EXPORT_SYMBOL(would_dump);
1530 void setup_new_exec(struct linux_binprm * bprm)
1532 /* Setup things that can depend upon the personality */
1533 struct task_struct *me = current;
1535 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1537 arch_setup_new_exec();
1539 /* Set the new mm task size. We have to do that late because it may
1540 * depend on TIF_32BIT which is only updated in flush_thread() on
1541 * some architectures like powerpc
1543 me->mm->task_size = TASK_SIZE;
1544 mutex_unlock(&me->signal->exec_update_mutex);
1545 mutex_unlock(&me->signal->cred_guard_mutex);
1547 EXPORT_SYMBOL(setup_new_exec);
1549 /* Runs immediately before start_thread() takes over. */
1550 void finalize_exec(struct linux_binprm *bprm)
1552 /* Store any stack rlimit changes before starting thread. */
1553 task_lock(current->group_leader);
1554 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1555 task_unlock(current->group_leader);
1557 EXPORT_SYMBOL(finalize_exec);
1560 * Prepare credentials and lock ->cred_guard_mutex.
1561 * setup_new_exec() commits the new creds and drops the lock.
1562 * Or, if exec fails before, free_bprm() should release ->cred and
1565 static int prepare_bprm_creds(struct linux_binprm *bprm)
1567 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1568 return -ERESTARTNOINTR;
1570 bprm->cred = prepare_exec_creds();
1571 if (likely(bprm->cred))
1574 mutex_unlock(¤t->signal->cred_guard_mutex);
1578 static void free_bprm(struct linux_binprm *bprm)
1581 acct_arg_size(bprm, 0);
1584 free_arg_pages(bprm);
1586 mutex_unlock(¤t->signal->cred_guard_mutex);
1587 abort_creds(bprm->cred);
1590 allow_write_access(bprm->file);
1593 if (bprm->executable)
1594 fput(bprm->executable);
1595 /* If a binfmt changed the interp, free it. */
1596 if (bprm->interp != bprm->filename)
1597 kfree(bprm->interp);
1598 kfree(bprm->fdpath);
1602 static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1604 struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1605 int retval = -ENOMEM;
1609 if (fd == AT_FDCWD || filename->name[0] == '/') {
1610 bprm->filename = filename->name;
1612 if (filename->name[0] == '\0')
1613 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1615 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1616 fd, filename->name);
1620 bprm->filename = bprm->fdpath;
1622 bprm->interp = bprm->filename;
1624 retval = bprm_mm_init(bprm);
1632 return ERR_PTR(retval);
1635 int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1637 /* If a binfmt changed the interp, free it first. */
1638 if (bprm->interp != bprm->filename)
1639 kfree(bprm->interp);
1640 bprm->interp = kstrdup(interp, GFP_KERNEL);
1645 EXPORT_SYMBOL(bprm_change_interp);
1648 * determine how safe it is to execute the proposed program
1649 * - the caller must hold ->cred_guard_mutex to protect against
1650 * PTRACE_ATTACH or seccomp thread-sync
1652 static void check_unsafe_exec(struct linux_binprm *bprm)
1654 struct task_struct *p = current, *t;
1658 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1661 * This isn't strictly necessary, but it makes it harder for LSMs to
1664 if (task_no_new_privs(current))
1665 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1669 spin_lock(&p->fs->lock);
1671 while_each_thread(p, t) {
1677 if (p->fs->users > n_fs)
1678 bprm->unsafe |= LSM_UNSAFE_SHARE;
1681 spin_unlock(&p->fs->lock);
1684 static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1686 /* Handle suid and sgid on files */
1687 struct inode *inode;
1692 if (!mnt_may_suid(file->f_path.mnt))
1695 if (task_no_new_privs(current))
1698 inode = file->f_path.dentry->d_inode;
1699 mode = READ_ONCE(inode->i_mode);
1700 if (!(mode & (S_ISUID|S_ISGID)))
1703 /* Be careful if suid/sgid is set */
1706 /* reload atomically mode/uid/gid now that lock held */
1707 mode = inode->i_mode;
1710 inode_unlock(inode);
1712 /* We ignore suid/sgid if there are no mappings for them in the ns */
1713 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1714 !kgid_has_mapping(bprm->cred->user_ns, gid))
1717 if (mode & S_ISUID) {
1718 bprm->per_clear |= PER_CLEAR_ON_SETID;
1719 bprm->cred->euid = uid;
1722 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1723 bprm->per_clear |= PER_CLEAR_ON_SETID;
1724 bprm->cred->egid = gid;
1729 * Compute brpm->cred based upon the final binary.
1731 static int bprm_creds_from_file(struct linux_binprm *bprm)
1733 /* Compute creds based on which file? */
1734 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1736 bprm_fill_uid(bprm, file);
1737 return security_bprm_creds_from_file(bprm, file);
1741 * Fill the binprm structure from the inode.
1742 * Read the first BINPRM_BUF_SIZE bytes
1744 * This may be called multiple times for binary chains (scripts for example).
1746 static int prepare_binprm(struct linux_binprm *bprm)
1750 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1751 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1755 * Arguments are '\0' separated strings found at the location bprm->p
1756 * points to; chop off the first by relocating brpm->p to right after
1757 * the first '\0' encountered.
1759 int remove_arg_zero(struct linux_binprm *bprm)
1762 unsigned long offset;
1770 offset = bprm->p & ~PAGE_MASK;
1771 page = get_arg_page(bprm, bprm->p, 0);
1776 kaddr = kmap_atomic(page);
1778 for (; offset < PAGE_SIZE && kaddr[offset];
1779 offset++, bprm->p++)
1782 kunmap_atomic(kaddr);
1784 } while (offset == PAGE_SIZE);
1793 EXPORT_SYMBOL(remove_arg_zero);
1795 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1797 * cycle the list of binary formats handler, until one recognizes the image
1799 static int search_binary_handler(struct linux_binprm *bprm)
1801 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1802 struct linux_binfmt *fmt;
1805 retval = prepare_binprm(bprm);
1809 retval = security_bprm_check(bprm);
1815 read_lock(&binfmt_lock);
1816 list_for_each_entry(fmt, &formats, lh) {
1817 if (!try_module_get(fmt->module))
1819 read_unlock(&binfmt_lock);
1821 retval = fmt->load_binary(bprm);
1823 read_lock(&binfmt_lock);
1825 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1826 read_unlock(&binfmt_lock);
1830 read_unlock(&binfmt_lock);
1833 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1834 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1836 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1845 static int exec_binprm(struct linux_binprm *bprm)
1847 pid_t old_pid, old_vpid;
1850 /* Need to fetch pid before load_binary changes it */
1851 old_pid = current->pid;
1853 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1856 /* This allows 4 levels of binfmt rewrites before failing hard. */
1857 for (depth = 0;; depth++) {
1862 ret = search_binary_handler(bprm);
1865 if (!bprm->interpreter)
1869 bprm->file = bprm->interpreter;
1870 bprm->interpreter = NULL;
1872 allow_write_access(exec);
1873 if (unlikely(bprm->have_execfd)) {
1874 if (bprm->executable) {
1878 bprm->executable = exec;
1884 trace_sched_process_exec(current, old_pid, bprm);
1885 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1886 proc_exec_connector(current);
1891 * sys_execve() executes a new program.
1893 static int bprm_execve(struct linux_binprm *bprm,
1894 int fd, struct filename *filename, int flags)
1897 struct files_struct *displaced;
1900 retval = unshare_files(&displaced);
1904 retval = prepare_bprm_creds(bprm);
1908 check_unsafe_exec(bprm);
1909 current->in_execve = 1;
1911 file = do_open_execat(fd, filename, flags);
1912 retval = PTR_ERR(file);
1920 * Record that a name derived from an O_CLOEXEC fd will be
1921 * inaccessible after exec. Relies on having exclusive access to
1922 * current->files (due to unshare_files above).
1925 close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1926 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1928 /* Set the unchanging part of bprm->cred */
1929 retval = security_bprm_creds_for_exec(bprm);
1933 retval = exec_binprm(bprm);
1937 /* execve succeeded */
1938 current->fs->in_exec = 0;
1939 current->in_execve = 0;
1940 rseq_execve(current);
1941 acct_update_integrals(current);
1942 task_numa_free(current, false);
1944 put_files_struct(displaced);
1949 * If past the point of no return ensure the the code never
1950 * returns to the userspace process. Use an existing fatal
1951 * signal if present otherwise terminate the process with
1954 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1955 force_sigsegv(SIGSEGV);
1958 current->fs->in_exec = 0;
1959 current->in_execve = 0;
1963 reset_files_struct(displaced);
1968 static int do_execveat_common(int fd, struct filename *filename,
1969 struct user_arg_ptr argv,
1970 struct user_arg_ptr envp,
1973 struct linux_binprm *bprm;
1976 if (IS_ERR(filename))
1977 return PTR_ERR(filename);
1980 * We move the actual failure in case of RLIMIT_NPROC excess from
1981 * set*uid() to execve() because too many poorly written programs
1982 * don't check setuid() return code. Here we additionally recheck
1983 * whether NPROC limit is still exceeded.
1985 if ((current->flags & PF_NPROC_EXCEEDED) &&
1986 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1991 /* We're below the limit (still or again), so we don't want to make
1992 * further execve() calls fail. */
1993 current->flags &= ~PF_NPROC_EXCEEDED;
1995 bprm = alloc_bprm(fd, filename);
1997 retval = PTR_ERR(bprm);
2001 retval = count(argv, MAX_ARG_STRINGS);
2004 bprm->argc = retval;
2006 retval = count(envp, MAX_ARG_STRINGS);
2009 bprm->envc = retval;
2011 retval = bprm_stack_limits(bprm);
2015 retval = copy_string_kernel(bprm->filename, bprm);
2018 bprm->exec = bprm->p;
2020 retval = copy_strings(bprm->envc, envp, bprm);
2024 retval = copy_strings(bprm->argc, argv, bprm);
2028 retval = bprm_execve(bprm, fd, filename, flags);
2037 int kernel_execve(const char *kernel_filename,
2038 const char *const *argv, const char *const *envp)
2040 struct filename *filename;
2041 struct linux_binprm *bprm;
2045 filename = getname_kernel(kernel_filename);
2046 if (IS_ERR(filename))
2047 return PTR_ERR(filename);
2049 bprm = alloc_bprm(fd, filename);
2051 retval = PTR_ERR(bprm);
2055 retval = count_strings_kernel(argv);
2058 bprm->argc = retval;
2060 retval = count_strings_kernel(envp);
2063 bprm->envc = retval;
2065 retval = bprm_stack_limits(bprm);
2069 retval = copy_string_kernel(bprm->filename, bprm);
2072 bprm->exec = bprm->p;
2074 retval = copy_strings_kernel(bprm->envc, envp, bprm);
2078 retval = copy_strings_kernel(bprm->argc, argv, bprm);
2082 retval = bprm_execve(bprm, fd, filename, 0);
2090 static int do_execve(struct filename *filename,
2091 const char __user *const __user *__argv,
2092 const char __user *const __user *__envp)
2094 struct user_arg_ptr argv = { .ptr.native = __argv };
2095 struct user_arg_ptr envp = { .ptr.native = __envp };
2096 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2099 static int do_execveat(int fd, struct filename *filename,
2100 const char __user *const __user *__argv,
2101 const char __user *const __user *__envp,
2104 struct user_arg_ptr argv = { .ptr.native = __argv };
2105 struct user_arg_ptr envp = { .ptr.native = __envp };
2107 return do_execveat_common(fd, filename, argv, envp, flags);
2110 #ifdef CONFIG_COMPAT
2111 static int compat_do_execve(struct filename *filename,
2112 const compat_uptr_t __user *__argv,
2113 const compat_uptr_t __user *__envp)
2115 struct user_arg_ptr argv = {
2117 .ptr.compat = __argv,
2119 struct user_arg_ptr envp = {
2121 .ptr.compat = __envp,
2123 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2126 static int compat_do_execveat(int fd, struct filename *filename,
2127 const compat_uptr_t __user *__argv,
2128 const compat_uptr_t __user *__envp,
2131 struct user_arg_ptr argv = {
2133 .ptr.compat = __argv,
2135 struct user_arg_ptr envp = {
2137 .ptr.compat = __envp,
2139 return do_execveat_common(fd, filename, argv, envp, flags);
2143 void set_binfmt(struct linux_binfmt *new)
2145 struct mm_struct *mm = current->mm;
2148 module_put(mm->binfmt->module);
2152 __module_get(new->module);
2154 EXPORT_SYMBOL(set_binfmt);
2157 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2159 void set_dumpable(struct mm_struct *mm, int value)
2161 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2164 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2167 SYSCALL_DEFINE3(execve,
2168 const char __user *, filename,
2169 const char __user *const __user *, argv,
2170 const char __user *const __user *, envp)
2172 return do_execve(getname(filename), argv, envp);
2175 SYSCALL_DEFINE5(execveat,
2176 int, fd, const char __user *, filename,
2177 const char __user *const __user *, argv,
2178 const char __user *const __user *, envp,
2181 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2183 return do_execveat(fd,
2184 getname_flags(filename, lookup_flags, NULL),
2188 #ifdef CONFIG_COMPAT
2189 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2190 const compat_uptr_t __user *, argv,
2191 const compat_uptr_t __user *, envp)
2193 return compat_do_execve(getname(filename), argv, envp);
2196 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2197 const char __user *, filename,
2198 const compat_uptr_t __user *, argv,
2199 const compat_uptr_t __user *, envp,
2202 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2204 return compat_do_execveat(fd,
2205 getname_flags(filename, lookup_flags, NULL),