4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * #!-checking implemented by tytso.
11 * Demand-loading implemented 01.12.91 - no need to read anything but
12 * the header into memory. The inode of the executable is put into
13 * "current->executable", and page faults do the actual loading. Clean.
15 * Once more I can proudly say that linux stood up to being changed: it
16 * was less than 2 hours work to get demand-loading completely implemented.
18 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
19 * current->executable is only used by the procfs. This allows a dispatch
20 * table to check for several different types of binary formats. We keep
21 * trying until we recognize the file or we run out of supported binary
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
29 #include <linux/vmacache.h>
30 #include <linux/stat.h>
31 #include <linux/fcntl.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/sched/mm.h>
36 #include <linux/sched/coredump.h>
37 #include <linux/pagemap.h>
38 #include <linux/perf_event.h>
39 #include <linux/highmem.h>
40 #include <linux/spinlock.h>
41 #include <linux/key.h>
42 #include <linux/personality.h>
43 #include <linux/binfmts.h>
44 #include <linux/utsname.h>
45 #include <linux/pid_namespace.h>
46 #include <linux/module.h>
47 #include <linux/namei.h>
48 #include <linux/mount.h>
49 #include <linux/security.h>
50 #include <linux/syscalls.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54 #include <linux/tracehook.h>
55 #include <linux/kmod.h>
56 #include <linux/fsnotify.h>
57 #include <linux/fs_struct.h>
58 #include <linux/pipe_fs_i.h>
59 #include <linux/oom.h>
60 #include <linux/compat.h>
61 #include <linux/vmalloc.h>
63 #include <linux/uaccess.h>
64 #include <asm/mmu_context.h>
67 #include <trace/events/task.h>
70 #include <trace/events/sched.h>
72 int suid_dumpable = 0;
74 static LIST_HEAD(formats);
75 static DEFINE_RWLOCK(binfmt_lock);
77 void __register_binfmt(struct linux_binfmt * fmt, int insert)
80 if (WARN_ON(!fmt->load_binary))
82 write_lock(&binfmt_lock);
83 insert ? list_add(&fmt->lh, &formats) :
84 list_add_tail(&fmt->lh, &formats);
85 write_unlock(&binfmt_lock);
88 EXPORT_SYMBOL(__register_binfmt);
90 void unregister_binfmt(struct linux_binfmt * fmt)
92 write_lock(&binfmt_lock);
94 write_unlock(&binfmt_lock);
97 EXPORT_SYMBOL(unregister_binfmt);
99 static inline void put_binfmt(struct linux_binfmt * fmt)
101 module_put(fmt->module);
104 bool path_noexec(const struct path *path)
106 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
107 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
112 * Note that a shared library must be both readable and executable due to
115 * Also note that we take the address to load from from the file itself.
117 SYSCALL_DEFINE1(uselib, const char __user *, library)
119 struct linux_binfmt *fmt;
121 struct filename *tmp = getname(library);
122 int error = PTR_ERR(tmp);
123 static const struct open_flags uselib_flags = {
124 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
125 .acc_mode = MAY_READ | MAY_EXEC,
126 .intent = LOOKUP_OPEN,
127 .lookup_flags = LOOKUP_FOLLOW,
133 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
135 error = PTR_ERR(file);
140 if (!S_ISREG(file_inode(file)->i_mode))
144 if (path_noexec(&file->f_path))
151 read_lock(&binfmt_lock);
152 list_for_each_entry(fmt, &formats, lh) {
153 if (!fmt->load_shlib)
155 if (!try_module_get(fmt->module))
157 read_unlock(&binfmt_lock);
158 error = fmt->load_shlib(file);
159 read_lock(&binfmt_lock);
161 if (error != -ENOEXEC)
164 read_unlock(&binfmt_lock);
170 #endif /* #ifdef CONFIG_USELIB */
174 * The nascent bprm->mm is not visible until exec_mmap() but it can
175 * use a lot of memory, account these pages in current->mm temporary
176 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
177 * change the counter back via acct_arg_size(0).
179 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
181 struct mm_struct *mm = current->mm;
182 long diff = (long)(pages - bprm->vma_pages);
187 bprm->vma_pages = pages;
188 add_mm_counter(mm, MM_ANONPAGES, diff);
191 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
196 unsigned int gup_flags = FOLL_FORCE;
198 #ifdef CONFIG_STACK_GROWSUP
200 ret = expand_downwards(bprm->vma, pos);
207 gup_flags |= FOLL_WRITE;
210 * We are doing an exec(). 'current' is the process
211 * doing the exec and bprm->mm is the new process's mm.
213 ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
219 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
222 acct_arg_size(bprm, size / PAGE_SIZE);
225 * We've historically supported up to 32 pages (ARG_MAX)
226 * of argument strings even with small stacks
232 * Limit to 1/4-th the stack size for the argv+env strings.
234 * - the remaining binfmt code will not run out of stack space,
235 * - the program will have a reasonable amount of stack left
238 rlim = current->signal->rlim;
239 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
248 static void put_arg_page(struct page *page)
253 static void free_arg_pages(struct linux_binprm *bprm)
257 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
260 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
263 static int __bprm_mm_init(struct linux_binprm *bprm)
266 struct vm_area_struct *vma = NULL;
267 struct mm_struct *mm = bprm->mm;
269 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
273 if (down_write_killable(&mm->mmap_sem)) {
280 * Place the stack at the largest stack address the architecture
281 * supports. Later, we'll move this to an appropriate place. We don't
282 * use STACK_TOP because that can depend on attributes which aren't
285 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
286 vma->vm_end = STACK_TOP_MAX;
287 vma->vm_start = vma->vm_end - PAGE_SIZE;
288 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
289 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
290 INIT_LIST_HEAD(&vma->anon_vma_chain);
292 err = insert_vm_struct(mm, vma);
296 mm->stack_vm = mm->total_vm = 1;
297 arch_bprm_mm_init(mm, vma);
298 up_write(&mm->mmap_sem);
299 bprm->p = vma->vm_end - sizeof(void *);
302 up_write(&mm->mmap_sem);
305 kmem_cache_free(vm_area_cachep, vma);
309 static bool valid_arg_len(struct linux_binprm *bprm, long len)
311 return len <= MAX_ARG_STRLEN;
316 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
320 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
325 page = bprm->page[pos / PAGE_SIZE];
326 if (!page && write) {
327 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
330 bprm->page[pos / PAGE_SIZE] = page;
336 static void put_arg_page(struct page *page)
340 static void free_arg_page(struct linux_binprm *bprm, int i)
343 __free_page(bprm->page[i]);
344 bprm->page[i] = NULL;
348 static void free_arg_pages(struct linux_binprm *bprm)
352 for (i = 0; i < MAX_ARG_PAGES; i++)
353 free_arg_page(bprm, i);
356 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
361 static int __bprm_mm_init(struct linux_binprm *bprm)
363 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
367 static bool valid_arg_len(struct linux_binprm *bprm, long len)
369 return len <= bprm->p;
372 #endif /* CONFIG_MMU */
375 * Create a new mm_struct and populate it with a temporary stack
376 * vm_area_struct. We don't have enough context at this point to set the stack
377 * flags, permissions, and offset, so we use temporary values. We'll update
378 * them later in setup_arg_pages().
380 static int bprm_mm_init(struct linux_binprm *bprm)
383 struct mm_struct *mm = NULL;
385 bprm->mm = mm = mm_alloc();
390 err = __bprm_mm_init(bprm);
405 struct user_arg_ptr {
410 const char __user *const __user *native;
412 const compat_uptr_t __user *compat;
417 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
419 const char __user *native;
422 if (unlikely(argv.is_compat)) {
423 compat_uptr_t compat;
425 if (get_user(compat, argv.ptr.compat + nr))
426 return ERR_PTR(-EFAULT);
428 return compat_ptr(compat);
432 if (get_user(native, argv.ptr.native + nr))
433 return ERR_PTR(-EFAULT);
439 * count() counts the number of strings in array ARGV.
441 static int count(struct user_arg_ptr argv, int max)
445 if (argv.ptr.native != NULL) {
447 const char __user *p = get_user_arg_ptr(argv, i);
459 if (fatal_signal_pending(current))
460 return -ERESTARTNOHAND;
468 * 'copy_strings()' copies argument/environment strings from the old
469 * processes's memory to the new process's stack. The call to get_user_pages()
470 * ensures the destination page is created and not swapped out.
472 static int copy_strings(int argc, struct user_arg_ptr argv,
473 struct linux_binprm *bprm)
475 struct page *kmapped_page = NULL;
477 unsigned long kpos = 0;
481 const char __user *str;
486 str = get_user_arg_ptr(argv, argc);
490 len = strnlen_user(str, MAX_ARG_STRLEN);
495 if (!valid_arg_len(bprm, len))
498 /* We're going to work our way backwords. */
504 int offset, bytes_to_copy;
506 if (fatal_signal_pending(current)) {
507 ret = -ERESTARTNOHAND;
512 offset = pos % PAGE_SIZE;
516 bytes_to_copy = offset;
517 if (bytes_to_copy > len)
520 offset -= bytes_to_copy;
521 pos -= bytes_to_copy;
522 str -= bytes_to_copy;
523 len -= bytes_to_copy;
525 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
528 page = get_arg_page(bprm, pos, 1);
535 flush_kernel_dcache_page(kmapped_page);
536 kunmap(kmapped_page);
537 put_arg_page(kmapped_page);
540 kaddr = kmap(kmapped_page);
541 kpos = pos & PAGE_MASK;
542 flush_arg_page(bprm, kpos, kmapped_page);
544 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
553 flush_kernel_dcache_page(kmapped_page);
554 kunmap(kmapped_page);
555 put_arg_page(kmapped_page);
561 * Like copy_strings, but get argv and its values from kernel memory.
563 int copy_strings_kernel(int argc, const char *const *__argv,
564 struct linux_binprm *bprm)
567 mm_segment_t oldfs = get_fs();
568 struct user_arg_ptr argv = {
569 .ptr.native = (const char __user *const __user *)__argv,
573 r = copy_strings(argc, argv, bprm);
578 EXPORT_SYMBOL(copy_strings_kernel);
583 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
584 * the binfmt code determines where the new stack should reside, we shift it to
585 * its final location. The process proceeds as follows:
587 * 1) Use shift to calculate the new vma endpoints.
588 * 2) Extend vma to cover both the old and new ranges. This ensures the
589 * arguments passed to subsequent functions are consistent.
590 * 3) Move vma's page tables to the new range.
591 * 4) Free up any cleared pgd range.
592 * 5) Shrink the vma to cover only the new range.
594 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
596 struct mm_struct *mm = vma->vm_mm;
597 unsigned long old_start = vma->vm_start;
598 unsigned long old_end = vma->vm_end;
599 unsigned long length = old_end - old_start;
600 unsigned long new_start = old_start - shift;
601 unsigned long new_end = old_end - shift;
602 struct mmu_gather tlb;
604 BUG_ON(new_start > new_end);
607 * ensure there are no vmas between where we want to go
610 if (vma != find_vma(mm, new_start))
614 * cover the whole range: [new_start, old_end)
616 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
620 * move the page tables downwards, on failure we rely on
621 * process cleanup to remove whatever mess we made.
623 if (length != move_page_tables(vma, old_start,
624 vma, new_start, length, false))
628 tlb_gather_mmu(&tlb, mm, old_start, old_end);
629 if (new_end > old_start) {
631 * when the old and new regions overlap clear from new_end.
633 free_pgd_range(&tlb, new_end, old_end, new_end,
634 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
637 * otherwise, clean from old_start; this is done to not touch
638 * the address space in [new_end, old_start) some architectures
639 * have constraints on va-space that make this illegal (IA64) -
640 * for the others its just a little faster.
642 free_pgd_range(&tlb, old_start, old_end, new_end,
643 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
645 tlb_finish_mmu(&tlb, old_start, old_end);
648 * Shrink the vma to just the new range. Always succeeds.
650 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
656 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
657 * the stack is optionally relocated, and some extra space is added.
659 int setup_arg_pages(struct linux_binprm *bprm,
660 unsigned long stack_top,
661 int executable_stack)
664 unsigned long stack_shift;
665 struct mm_struct *mm = current->mm;
666 struct vm_area_struct *vma = bprm->vma;
667 struct vm_area_struct *prev = NULL;
668 unsigned long vm_flags;
669 unsigned long stack_base;
670 unsigned long stack_size;
671 unsigned long stack_expand;
672 unsigned long rlim_stack;
674 #ifdef CONFIG_STACK_GROWSUP
675 /* Limit stack size */
676 stack_base = rlimit_max(RLIMIT_STACK);
677 if (stack_base > STACK_SIZE_MAX)
678 stack_base = STACK_SIZE_MAX;
680 /* Add space for stack randomization. */
681 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
683 /* Make sure we didn't let the argument array grow too large. */
684 if (vma->vm_end - vma->vm_start > stack_base)
687 stack_base = PAGE_ALIGN(stack_top - stack_base);
689 stack_shift = vma->vm_start - stack_base;
690 mm->arg_start = bprm->p - stack_shift;
691 bprm->p = vma->vm_end - stack_shift;
693 stack_top = arch_align_stack(stack_top);
694 stack_top = PAGE_ALIGN(stack_top);
696 if (unlikely(stack_top < mmap_min_addr) ||
697 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
700 stack_shift = vma->vm_end - stack_top;
702 bprm->p -= stack_shift;
703 mm->arg_start = bprm->p;
707 bprm->loader -= stack_shift;
708 bprm->exec -= stack_shift;
710 if (down_write_killable(&mm->mmap_sem))
713 vm_flags = VM_STACK_FLAGS;
716 * Adjust stack execute permissions; explicitly enable for
717 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
718 * (arch default) otherwise.
720 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
722 else if (executable_stack == EXSTACK_DISABLE_X)
723 vm_flags &= ~VM_EXEC;
724 vm_flags |= mm->def_flags;
725 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
727 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
733 /* Move stack pages down in memory. */
735 ret = shift_arg_pages(vma, stack_shift);
740 /* mprotect_fixup is overkill to remove the temporary stack flags */
741 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
743 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
744 stack_size = vma->vm_end - vma->vm_start;
746 * Align this down to a page boundary as expand_stack
749 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
750 #ifdef CONFIG_STACK_GROWSUP
751 if (stack_size + stack_expand > rlim_stack)
752 stack_base = vma->vm_start + rlim_stack;
754 stack_base = vma->vm_end + stack_expand;
756 if (stack_size + stack_expand > rlim_stack)
757 stack_base = vma->vm_end - rlim_stack;
759 stack_base = vma->vm_start - stack_expand;
761 current->mm->start_stack = bprm->p;
762 ret = expand_stack(vma, stack_base);
767 up_write(&mm->mmap_sem);
770 EXPORT_SYMBOL(setup_arg_pages);
775 * Transfer the program arguments and environment from the holding pages
776 * onto the stack. The provided stack pointer is adjusted accordingly.
778 int transfer_args_to_stack(struct linux_binprm *bprm,
779 unsigned long *sp_location)
781 unsigned long index, stop, sp;
784 stop = bprm->p >> PAGE_SHIFT;
787 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
788 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
789 char *src = kmap(bprm->page[index]) + offset;
790 sp -= PAGE_SIZE - offset;
791 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
793 kunmap(bprm->page[index]);
803 EXPORT_SYMBOL(transfer_args_to_stack);
805 #endif /* CONFIG_MMU */
807 static struct file *do_open_execat(int fd, struct filename *name, int flags)
811 struct open_flags open_exec_flags = {
812 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
813 .acc_mode = MAY_EXEC,
814 .intent = LOOKUP_OPEN,
815 .lookup_flags = LOOKUP_FOLLOW,
818 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
819 return ERR_PTR(-EINVAL);
820 if (flags & AT_SYMLINK_NOFOLLOW)
821 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
822 if (flags & AT_EMPTY_PATH)
823 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
825 file = do_filp_open(fd, name, &open_exec_flags);
830 if (!S_ISREG(file_inode(file)->i_mode))
833 if (path_noexec(&file->f_path))
836 err = deny_write_access(file);
840 if (name->name[0] != '\0')
851 struct file *open_exec(const char *name)
853 struct filename *filename = getname_kernel(name);
854 struct file *f = ERR_CAST(filename);
856 if (!IS_ERR(filename)) {
857 f = do_open_execat(AT_FDCWD, filename, 0);
862 EXPORT_SYMBOL(open_exec);
864 int kernel_read(struct file *file, loff_t offset,
865 char *addr, unsigned long count)
873 /* The cast to a user pointer is valid due to the set_fs() */
874 result = vfs_read(file, (void __user *)addr, count, &pos);
879 EXPORT_SYMBOL(kernel_read);
881 int kernel_read_file(struct file *file, void **buf, loff_t *size,
882 loff_t max_size, enum kernel_read_file_id id)
888 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
891 ret = security_kernel_read_file(file, id);
895 ret = deny_write_access(file);
899 i_size = i_size_read(file_inode(file));
900 if (max_size > 0 && i_size > max_size) {
909 if (id != READING_FIRMWARE_PREALLOC_BUFFER)
910 *buf = vmalloc(i_size);
917 while (pos < i_size) {
918 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
935 ret = security_kernel_post_read_file(file, *buf, i_size, id);
941 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
948 allow_write_access(file);
951 EXPORT_SYMBOL_GPL(kernel_read_file);
953 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
954 loff_t max_size, enum kernel_read_file_id id)
962 file = filp_open(path, O_RDONLY, 0);
964 return PTR_ERR(file);
966 ret = kernel_read_file(file, buf, size, max_size, id);
970 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
972 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
973 enum kernel_read_file_id id)
975 struct fd f = fdget(fd);
981 ret = kernel_read_file(f.file, buf, size, max_size, id);
986 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
988 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
990 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
992 flush_icache_range(addr, addr + len);
995 EXPORT_SYMBOL(read_code);
997 static int exec_mmap(struct mm_struct *mm)
999 struct task_struct *tsk;
1000 struct mm_struct *old_mm, *active_mm;
1002 /* Notify parent that we're no longer interested in the old VM */
1004 old_mm = current->mm;
1005 mm_release(tsk, old_mm);
1008 sync_mm_rss(old_mm);
1010 * Make sure that if there is a core dump in progress
1011 * for the old mm, we get out and die instead of going
1012 * through with the exec. We must hold mmap_sem around
1013 * checking core_state and changing tsk->mm.
1015 down_read(&old_mm->mmap_sem);
1016 if (unlikely(old_mm->core_state)) {
1017 up_read(&old_mm->mmap_sem);
1022 active_mm = tsk->active_mm;
1024 tsk->active_mm = mm;
1025 activate_mm(active_mm, mm);
1026 tsk->mm->vmacache_seqnum = 0;
1027 vmacache_flush(tsk);
1030 up_read(&old_mm->mmap_sem);
1031 BUG_ON(active_mm != old_mm);
1032 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1033 mm_update_next_owner(old_mm);
1042 * This function makes sure the current process has its own signal table,
1043 * so that flush_signal_handlers can later reset the handlers without
1044 * disturbing other processes. (Other processes might share the signal
1045 * table via the CLONE_SIGHAND option to clone().)
1047 static int de_thread(struct task_struct *tsk)
1049 struct signal_struct *sig = tsk->signal;
1050 struct sighand_struct *oldsighand = tsk->sighand;
1051 spinlock_t *lock = &oldsighand->siglock;
1053 if (thread_group_empty(tsk))
1054 goto no_thread_group;
1057 * Kill all other threads in the thread group.
1059 spin_lock_irq(lock);
1060 if (signal_group_exit(sig)) {
1062 * Another group action in progress, just
1063 * return so that the signal is processed.
1065 spin_unlock_irq(lock);
1069 sig->group_exit_task = tsk;
1070 sig->notify_count = zap_other_threads(tsk);
1071 if (!thread_group_leader(tsk))
1072 sig->notify_count--;
1074 while (sig->notify_count) {
1075 __set_current_state(TASK_KILLABLE);
1076 spin_unlock_irq(lock);
1078 if (unlikely(__fatal_signal_pending(tsk)))
1080 spin_lock_irq(lock);
1082 spin_unlock_irq(lock);
1085 * At this point all other threads have exited, all we have to
1086 * do is to wait for the thread group leader to become inactive,
1087 * and to assume its PID:
1089 if (!thread_group_leader(tsk)) {
1090 struct task_struct *leader = tsk->group_leader;
1093 cgroup_threadgroup_change_begin(tsk);
1094 write_lock_irq(&tasklist_lock);
1096 * Do this under tasklist_lock to ensure that
1097 * exit_notify() can't miss ->group_exit_task
1099 sig->notify_count = -1;
1100 if (likely(leader->exit_state))
1102 __set_current_state(TASK_KILLABLE);
1103 write_unlock_irq(&tasklist_lock);
1104 cgroup_threadgroup_change_end(tsk);
1106 if (unlikely(__fatal_signal_pending(tsk)))
1111 * The only record we have of the real-time age of a
1112 * process, regardless of execs it's done, is start_time.
1113 * All the past CPU time is accumulated in signal_struct
1114 * from sister threads now dead. But in this non-leader
1115 * exec, nothing survives from the original leader thread,
1116 * whose birth marks the true age of this process now.
1117 * When we take on its identity by switching to its PID, we
1118 * also take its birthdate (always earlier than our own).
1120 tsk->start_time = leader->start_time;
1121 tsk->real_start_time = leader->real_start_time;
1123 BUG_ON(!same_thread_group(leader, tsk));
1124 BUG_ON(has_group_leader_pid(tsk));
1126 * An exec() starts a new thread group with the
1127 * TGID of the previous thread group. Rehash the
1128 * two threads with a switched PID, and release
1129 * the former thread group leader:
1132 /* Become a process group leader with the old leader's pid.
1133 * The old leader becomes a thread of the this thread group.
1134 * Note: The old leader also uses this pid until release_task
1135 * is called. Odd but simple and correct.
1137 tsk->pid = leader->pid;
1138 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1139 transfer_pid(leader, tsk, PIDTYPE_PGID);
1140 transfer_pid(leader, tsk, PIDTYPE_SID);
1142 list_replace_rcu(&leader->tasks, &tsk->tasks);
1143 list_replace_init(&leader->sibling, &tsk->sibling);
1145 tsk->group_leader = tsk;
1146 leader->group_leader = tsk;
1148 tsk->exit_signal = SIGCHLD;
1149 leader->exit_signal = -1;
1151 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1152 leader->exit_state = EXIT_DEAD;
1155 * We are going to release_task()->ptrace_unlink() silently,
1156 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1157 * the tracer wont't block again waiting for this thread.
1159 if (unlikely(leader->ptrace))
1160 __wake_up_parent(leader, leader->parent);
1161 write_unlock_irq(&tasklist_lock);
1162 cgroup_threadgroup_change_end(tsk);
1164 release_task(leader);
1167 sig->group_exit_task = NULL;
1168 sig->notify_count = 0;
1171 /* we have changed execution domain */
1172 tsk->exit_signal = SIGCHLD;
1174 #ifdef CONFIG_POSIX_TIMERS
1176 flush_itimer_signals();
1179 if (atomic_read(&oldsighand->count) != 1) {
1180 struct sighand_struct *newsighand;
1182 * This ->sighand is shared with the CLONE_SIGHAND
1183 * but not CLONE_THREAD task, switch to the new one.
1185 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1189 atomic_set(&newsighand->count, 1);
1190 memcpy(newsighand->action, oldsighand->action,
1191 sizeof(newsighand->action));
1193 write_lock_irq(&tasklist_lock);
1194 spin_lock(&oldsighand->siglock);
1195 rcu_assign_pointer(tsk->sighand, newsighand);
1196 spin_unlock(&oldsighand->siglock);
1197 write_unlock_irq(&tasklist_lock);
1199 __cleanup_sighand(oldsighand);
1202 BUG_ON(!thread_group_leader(tsk));
1206 /* protects against exit_notify() and __exit_signal() */
1207 read_lock(&tasklist_lock);
1208 sig->group_exit_task = NULL;
1209 sig->notify_count = 0;
1210 read_unlock(&tasklist_lock);
1214 char *get_task_comm(char *buf, struct task_struct *tsk)
1216 /* buf must be at least sizeof(tsk->comm) in size */
1218 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1222 EXPORT_SYMBOL_GPL(get_task_comm);
1225 * These functions flushes out all traces of the currently running executable
1226 * so that a new one can be started
1229 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1232 trace_task_rename(tsk, buf);
1233 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1235 perf_event_comm(tsk, exec);
1238 int flush_old_exec(struct linux_binprm * bprm)
1243 * Make sure we have a private signal table and that
1244 * we are unassociated from the previous thread group.
1246 retval = de_thread(current);
1251 * Must be called _before_ exec_mmap() as bprm->mm is
1252 * not visibile until then. This also enables the update
1255 set_mm_exe_file(bprm->mm, bprm->file);
1258 * Release all of the old mmap stuff
1260 acct_arg_size(bprm, 0);
1261 retval = exec_mmap(bprm->mm);
1265 bprm->mm = NULL; /* We're using it now */
1268 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1269 PF_NOFREEZE | PF_NO_SETAFFINITY);
1271 current->personality &= ~bprm->per_clear;
1274 * We have to apply CLOEXEC before we change whether the process is
1275 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1276 * trying to access the should-be-closed file descriptors of a process
1277 * undergoing exec(2).
1279 do_close_on_exec(current->files);
1285 EXPORT_SYMBOL(flush_old_exec);
1287 void would_dump(struct linux_binprm *bprm, struct file *file)
1289 struct inode *inode = file_inode(file);
1290 if (inode_permission(inode, MAY_READ) < 0) {
1291 struct user_namespace *old, *user_ns;
1292 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1294 /* Ensure mm->user_ns contains the executable */
1295 user_ns = old = bprm->mm->user_ns;
1296 while ((user_ns != &init_user_ns) &&
1297 !privileged_wrt_inode_uidgid(user_ns, inode))
1298 user_ns = user_ns->parent;
1300 if (old != user_ns) {
1301 bprm->mm->user_ns = get_user_ns(user_ns);
1306 EXPORT_SYMBOL(would_dump);
1308 void setup_new_exec(struct linux_binprm * bprm)
1310 arch_pick_mmap_layout(current->mm);
1312 /* This is the point of no return */
1313 current->sas_ss_sp = current->sas_ss_size = 0;
1315 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1316 set_dumpable(current->mm, SUID_DUMP_USER);
1318 set_dumpable(current->mm, suid_dumpable);
1321 __set_task_comm(current, kbasename(bprm->filename), true);
1323 /* Set the new mm task size. We have to do that late because it may
1324 * depend on TIF_32BIT which is only updated in flush_thread() on
1325 * some architectures like powerpc
1327 current->mm->task_size = TASK_SIZE;
1329 /* install the new credentials */
1330 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1331 !gid_eq(bprm->cred->gid, current_egid())) {
1332 current->pdeath_signal = 0;
1334 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1335 set_dumpable(current->mm, suid_dumpable);
1338 /* An exec changes our domain. We are no longer part of the thread
1340 current->self_exec_id++;
1341 flush_signal_handlers(current, 0);
1343 EXPORT_SYMBOL(setup_new_exec);
1346 * Prepare credentials and lock ->cred_guard_mutex.
1347 * install_exec_creds() commits the new creds and drops the lock.
1348 * Or, if exec fails before, free_bprm() should release ->cred and
1351 int prepare_bprm_creds(struct linux_binprm *bprm)
1353 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1354 return -ERESTARTNOINTR;
1356 bprm->cred = prepare_exec_creds();
1357 if (likely(bprm->cred))
1360 mutex_unlock(¤t->signal->cred_guard_mutex);
1364 static void free_bprm(struct linux_binprm *bprm)
1366 free_arg_pages(bprm);
1368 mutex_unlock(¤t->signal->cred_guard_mutex);
1369 abort_creds(bprm->cred);
1372 allow_write_access(bprm->file);
1375 /* If a binfmt changed the interp, free it. */
1376 if (bprm->interp != bprm->filename)
1377 kfree(bprm->interp);
1381 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1383 /* If a binfmt changed the interp, free it first. */
1384 if (bprm->interp != bprm->filename)
1385 kfree(bprm->interp);
1386 bprm->interp = kstrdup(interp, GFP_KERNEL);
1391 EXPORT_SYMBOL(bprm_change_interp);
1394 * install the new credentials for this executable
1396 void install_exec_creds(struct linux_binprm *bprm)
1398 security_bprm_committing_creds(bprm);
1400 commit_creds(bprm->cred);
1404 * Disable monitoring for regular users
1405 * when executing setuid binaries. Must
1406 * wait until new credentials are committed
1407 * by commit_creds() above
1409 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1410 perf_event_exit_task(current);
1412 * cred_guard_mutex must be held at least to this point to prevent
1413 * ptrace_attach() from altering our determination of the task's
1414 * credentials; any time after this it may be unlocked.
1416 security_bprm_committed_creds(bprm);
1417 mutex_unlock(¤t->signal->cred_guard_mutex);
1419 EXPORT_SYMBOL(install_exec_creds);
1422 * determine how safe it is to execute the proposed program
1423 * - the caller must hold ->cred_guard_mutex to protect against
1424 * PTRACE_ATTACH or seccomp thread-sync
1426 static void check_unsafe_exec(struct linux_binprm *bprm)
1428 struct task_struct *p = current, *t;
1432 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1435 * This isn't strictly necessary, but it makes it harder for LSMs to
1438 if (task_no_new_privs(current))
1439 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1443 spin_lock(&p->fs->lock);
1445 while_each_thread(p, t) {
1451 if (p->fs->users > n_fs)
1452 bprm->unsafe |= LSM_UNSAFE_SHARE;
1455 spin_unlock(&p->fs->lock);
1458 static void bprm_fill_uid(struct linux_binprm *bprm)
1460 struct inode *inode;
1466 * Since this can be called multiple times (via prepare_binprm),
1467 * we must clear any previous work done when setting set[ug]id
1468 * bits from any earlier bprm->file uses (for example when run
1469 * first for a setuid script then again for its interpreter).
1471 bprm->cred->euid = current_euid();
1472 bprm->cred->egid = current_egid();
1474 if (!mnt_may_suid(bprm->file->f_path.mnt))
1477 if (task_no_new_privs(current))
1480 inode = bprm->file->f_path.dentry->d_inode;
1481 mode = READ_ONCE(inode->i_mode);
1482 if (!(mode & (S_ISUID|S_ISGID)))
1485 /* Be careful if suid/sgid is set */
1488 /* reload atomically mode/uid/gid now that lock held */
1489 mode = inode->i_mode;
1492 inode_unlock(inode);
1494 /* We ignore suid/sgid if there are no mappings for them in the ns */
1495 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1496 !kgid_has_mapping(bprm->cred->user_ns, gid))
1499 if (mode & S_ISUID) {
1500 bprm->per_clear |= PER_CLEAR_ON_SETID;
1501 bprm->cred->euid = uid;
1504 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1505 bprm->per_clear |= PER_CLEAR_ON_SETID;
1506 bprm->cred->egid = gid;
1511 * Fill the binprm structure from the inode.
1512 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1514 * This may be called multiple times for binary chains (scripts for example).
1516 int prepare_binprm(struct linux_binprm *bprm)
1520 bprm_fill_uid(bprm);
1522 /* fill in binprm security blob */
1523 retval = security_bprm_set_creds(bprm);
1526 bprm->cred_prepared = 1;
1528 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1529 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1532 EXPORT_SYMBOL(prepare_binprm);
1535 * Arguments are '\0' separated strings found at the location bprm->p
1536 * points to; chop off the first by relocating brpm->p to right after
1537 * the first '\0' encountered.
1539 int remove_arg_zero(struct linux_binprm *bprm)
1542 unsigned long offset;
1550 offset = bprm->p & ~PAGE_MASK;
1551 page = get_arg_page(bprm, bprm->p, 0);
1556 kaddr = kmap_atomic(page);
1558 for (; offset < PAGE_SIZE && kaddr[offset];
1559 offset++, bprm->p++)
1562 kunmap_atomic(kaddr);
1564 } while (offset == PAGE_SIZE);
1573 EXPORT_SYMBOL(remove_arg_zero);
1575 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1577 * cycle the list of binary formats handler, until one recognizes the image
1579 int search_binary_handler(struct linux_binprm *bprm)
1581 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1582 struct linux_binfmt *fmt;
1585 /* This allows 4 levels of binfmt rewrites before failing hard. */
1586 if (bprm->recursion_depth > 5)
1589 retval = security_bprm_check(bprm);
1595 read_lock(&binfmt_lock);
1596 list_for_each_entry(fmt, &formats, lh) {
1597 if (!try_module_get(fmt->module))
1599 read_unlock(&binfmt_lock);
1600 bprm->recursion_depth++;
1601 retval = fmt->load_binary(bprm);
1602 read_lock(&binfmt_lock);
1604 bprm->recursion_depth--;
1605 if (retval < 0 && !bprm->mm) {
1606 /* we got to flush_old_exec() and failed after it */
1607 read_unlock(&binfmt_lock);
1608 force_sigsegv(SIGSEGV, current);
1611 if (retval != -ENOEXEC || !bprm->file) {
1612 read_unlock(&binfmt_lock);
1616 read_unlock(&binfmt_lock);
1619 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1620 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1622 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1630 EXPORT_SYMBOL(search_binary_handler);
1632 static int exec_binprm(struct linux_binprm *bprm)
1634 pid_t old_pid, old_vpid;
1637 /* Need to fetch pid before load_binary changes it */
1638 old_pid = current->pid;
1640 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1643 ret = search_binary_handler(bprm);
1646 trace_sched_process_exec(current, old_pid, bprm);
1647 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1648 proc_exec_connector(current);
1655 * sys_execve() executes a new program.
1657 static int do_execveat_common(int fd, struct filename *filename,
1658 struct user_arg_ptr argv,
1659 struct user_arg_ptr envp,
1662 char *pathbuf = NULL;
1663 struct linux_binprm *bprm;
1665 struct files_struct *displaced;
1668 if (IS_ERR(filename))
1669 return PTR_ERR(filename);
1672 * We move the actual failure in case of RLIMIT_NPROC excess from
1673 * set*uid() to execve() because too many poorly written programs
1674 * don't check setuid() return code. Here we additionally recheck
1675 * whether NPROC limit is still exceeded.
1677 if ((current->flags & PF_NPROC_EXCEEDED) &&
1678 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1683 /* We're below the limit (still or again), so we don't want to make
1684 * further execve() calls fail. */
1685 current->flags &= ~PF_NPROC_EXCEEDED;
1687 retval = unshare_files(&displaced);
1692 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1696 retval = prepare_bprm_creds(bprm);
1700 check_unsafe_exec(bprm);
1701 current->in_execve = 1;
1703 file = do_open_execat(fd, filename, flags);
1704 retval = PTR_ERR(file);
1711 if (fd == AT_FDCWD || filename->name[0] == '/') {
1712 bprm->filename = filename->name;
1714 if (filename->name[0] == '\0')
1715 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1717 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1718 fd, filename->name);
1724 * Record that a name derived from an O_CLOEXEC fd will be
1725 * inaccessible after exec. Relies on having exclusive access to
1726 * current->files (due to unshare_files above).
1728 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1729 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1730 bprm->filename = pathbuf;
1732 bprm->interp = bprm->filename;
1734 retval = bprm_mm_init(bprm);
1738 bprm->argc = count(argv, MAX_ARG_STRINGS);
1739 if ((retval = bprm->argc) < 0)
1742 bprm->envc = count(envp, MAX_ARG_STRINGS);
1743 if ((retval = bprm->envc) < 0)
1746 retval = prepare_binprm(bprm);
1750 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1754 bprm->exec = bprm->p;
1755 retval = copy_strings(bprm->envc, envp, bprm);
1759 retval = copy_strings(bprm->argc, argv, bprm);
1763 would_dump(bprm, bprm->file);
1765 retval = exec_binprm(bprm);
1769 /* execve succeeded */
1770 current->fs->in_exec = 0;
1771 current->in_execve = 0;
1772 acct_update_integrals(current);
1773 task_numa_free(current);
1778 put_files_struct(displaced);
1783 acct_arg_size(bprm, 0);
1788 current->fs->in_exec = 0;
1789 current->in_execve = 0;
1797 reset_files_struct(displaced);
1803 int do_execve(struct filename *filename,
1804 const char __user *const __user *__argv,
1805 const char __user *const __user *__envp)
1807 struct user_arg_ptr argv = { .ptr.native = __argv };
1808 struct user_arg_ptr envp = { .ptr.native = __envp };
1809 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1812 int do_execveat(int fd, struct filename *filename,
1813 const char __user *const __user *__argv,
1814 const char __user *const __user *__envp,
1817 struct user_arg_ptr argv = { .ptr.native = __argv };
1818 struct user_arg_ptr envp = { .ptr.native = __envp };
1820 return do_execveat_common(fd, filename, argv, envp, flags);
1823 #ifdef CONFIG_COMPAT
1824 static int compat_do_execve(struct filename *filename,
1825 const compat_uptr_t __user *__argv,
1826 const compat_uptr_t __user *__envp)
1828 struct user_arg_ptr argv = {
1830 .ptr.compat = __argv,
1832 struct user_arg_ptr envp = {
1834 .ptr.compat = __envp,
1836 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1839 static int compat_do_execveat(int fd, struct filename *filename,
1840 const compat_uptr_t __user *__argv,
1841 const compat_uptr_t __user *__envp,
1844 struct user_arg_ptr argv = {
1846 .ptr.compat = __argv,
1848 struct user_arg_ptr envp = {
1850 .ptr.compat = __envp,
1852 return do_execveat_common(fd, filename, argv, envp, flags);
1856 void set_binfmt(struct linux_binfmt *new)
1858 struct mm_struct *mm = current->mm;
1861 module_put(mm->binfmt->module);
1865 __module_get(new->module);
1867 EXPORT_SYMBOL(set_binfmt);
1870 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1872 void set_dumpable(struct mm_struct *mm, int value)
1874 unsigned long old, new;
1876 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1880 old = ACCESS_ONCE(mm->flags);
1881 new = (old & ~MMF_DUMPABLE_MASK) | value;
1882 } while (cmpxchg(&mm->flags, old, new) != old);
1885 SYSCALL_DEFINE3(execve,
1886 const char __user *, filename,
1887 const char __user *const __user *, argv,
1888 const char __user *const __user *, envp)
1890 return do_execve(getname(filename), argv, envp);
1893 SYSCALL_DEFINE5(execveat,
1894 int, fd, const char __user *, filename,
1895 const char __user *const __user *, argv,
1896 const char __user *const __user *, envp,
1899 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1901 return do_execveat(fd,
1902 getname_flags(filename, lookup_flags, NULL),
1906 #ifdef CONFIG_COMPAT
1907 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1908 const compat_uptr_t __user *, argv,
1909 const compat_uptr_t __user *, envp)
1911 return compat_do_execve(getname(filename), argv, envp);
1914 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1915 const char __user *, filename,
1916 const compat_uptr_t __user *, argv,
1917 const compat_uptr_t __user *, envp,
1920 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1922 return compat_do_execveat(fd,
1923 getname_flags(filename, lookup_flags, NULL),