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/pagemap.h>
37 #include <linux/perf_event.h>
38 #include <linux/highmem.h>
39 #include <linux/spinlock.h>
40 #include <linux/key.h>
41 #include <linux/personality.h>
42 #include <linux/binfmts.h>
43 #include <linux/utsname.h>
44 #include <linux/pid_namespace.h>
45 #include <linux/module.h>
46 #include <linux/namei.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/tsacct_kern.h>
51 #include <linux/cn_proc.h>
52 #include <linux/audit.h>
53 #include <linux/tracehook.h>
54 #include <linux/kmod.h>
55 #include <linux/fsnotify.h>
56 #include <linux/fs_struct.h>
57 #include <linux/pipe_fs_i.h>
58 #include <linux/oom.h>
59 #include <linux/compat.h>
60 #include <linux/vmalloc.h>
62 #include <linux/uaccess.h>
63 #include <asm/mmu_context.h>
66 #include <trace/events/task.h>
69 #include <trace/events/sched.h>
71 int suid_dumpable = 0;
73 static LIST_HEAD(formats);
74 static DEFINE_RWLOCK(binfmt_lock);
76 void __register_binfmt(struct linux_binfmt * fmt, int insert)
79 if (WARN_ON(!fmt->load_binary))
81 write_lock(&binfmt_lock);
82 insert ? list_add(&fmt->lh, &formats) :
83 list_add_tail(&fmt->lh, &formats);
84 write_unlock(&binfmt_lock);
87 EXPORT_SYMBOL(__register_binfmt);
89 void unregister_binfmt(struct linux_binfmt * fmt)
91 write_lock(&binfmt_lock);
93 write_unlock(&binfmt_lock);
96 EXPORT_SYMBOL(unregister_binfmt);
98 static inline void put_binfmt(struct linux_binfmt * fmt)
100 module_put(fmt->module);
103 bool path_noexec(const struct path *path)
105 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
106 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
111 * Note that a shared library must be both readable and executable due to
114 * Also note that we take the address to load from from the file itself.
116 SYSCALL_DEFINE1(uselib, const char __user *, library)
118 struct linux_binfmt *fmt;
120 struct filename *tmp = getname(library);
121 int error = PTR_ERR(tmp);
122 static const struct open_flags uselib_flags = {
123 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
124 .acc_mode = MAY_READ | MAY_EXEC,
125 .intent = LOOKUP_OPEN,
126 .lookup_flags = LOOKUP_FOLLOW,
132 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
134 error = PTR_ERR(file);
139 if (!S_ISREG(file_inode(file)->i_mode))
143 if (path_noexec(&file->f_path))
150 read_lock(&binfmt_lock);
151 list_for_each_entry(fmt, &formats, lh) {
152 if (!fmt->load_shlib)
154 if (!try_module_get(fmt->module))
156 read_unlock(&binfmt_lock);
157 error = fmt->load_shlib(file);
158 read_lock(&binfmt_lock);
160 if (error != -ENOEXEC)
163 read_unlock(&binfmt_lock);
169 #endif /* #ifdef CONFIG_USELIB */
173 * The nascent bprm->mm is not visible until exec_mmap() but it can
174 * use a lot of memory, account these pages in current->mm temporary
175 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
176 * change the counter back via acct_arg_size(0).
178 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
180 struct mm_struct *mm = current->mm;
181 long diff = (long)(pages - bprm->vma_pages);
186 bprm->vma_pages = pages;
187 add_mm_counter(mm, MM_ANONPAGES, diff);
190 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
195 unsigned int gup_flags = FOLL_FORCE;
197 #ifdef CONFIG_STACK_GROWSUP
199 ret = expand_downwards(bprm->vma, pos);
206 gup_flags |= FOLL_WRITE;
209 * We are doing an exec(). 'current' is the process
210 * doing the exec and bprm->mm is the new process's mm.
212 ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
218 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
221 acct_arg_size(bprm, size / PAGE_SIZE);
224 * We've historically supported up to 32 pages (ARG_MAX)
225 * of argument strings even with small stacks
231 * Limit to 1/4-th the stack size for the argv+env strings.
233 * - the remaining binfmt code will not run out of stack space,
234 * - the program will have a reasonable amount of stack left
237 rlim = current->signal->rlim;
238 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
247 static void put_arg_page(struct page *page)
252 static void free_arg_pages(struct linux_binprm *bprm)
256 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
259 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
262 static int __bprm_mm_init(struct linux_binprm *bprm)
265 struct vm_area_struct *vma = NULL;
266 struct mm_struct *mm = bprm->mm;
268 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
272 if (down_write_killable(&mm->mmap_sem)) {
279 * Place the stack at the largest stack address the architecture
280 * supports. Later, we'll move this to an appropriate place. We don't
281 * use STACK_TOP because that can depend on attributes which aren't
284 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
285 vma->vm_end = STACK_TOP_MAX;
286 vma->vm_start = vma->vm_end - PAGE_SIZE;
287 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
288 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
289 INIT_LIST_HEAD(&vma->anon_vma_chain);
291 err = insert_vm_struct(mm, vma);
295 mm->stack_vm = mm->total_vm = 1;
296 arch_bprm_mm_init(mm, vma);
297 up_write(&mm->mmap_sem);
298 bprm->p = vma->vm_end - sizeof(void *);
301 up_write(&mm->mmap_sem);
304 kmem_cache_free(vm_area_cachep, vma);
308 static bool valid_arg_len(struct linux_binprm *bprm, long len)
310 return len <= MAX_ARG_STRLEN;
315 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
319 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
324 page = bprm->page[pos / PAGE_SIZE];
325 if (!page && write) {
326 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
329 bprm->page[pos / PAGE_SIZE] = page;
335 static void put_arg_page(struct page *page)
339 static void free_arg_page(struct linux_binprm *bprm, int i)
342 __free_page(bprm->page[i]);
343 bprm->page[i] = NULL;
347 static void free_arg_pages(struct linux_binprm *bprm)
351 for (i = 0; i < MAX_ARG_PAGES; i++)
352 free_arg_page(bprm, i);
355 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
360 static int __bprm_mm_init(struct linux_binprm *bprm)
362 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
366 static bool valid_arg_len(struct linux_binprm *bprm, long len)
368 return len <= bprm->p;
371 #endif /* CONFIG_MMU */
374 * Create a new mm_struct and populate it with a temporary stack
375 * vm_area_struct. We don't have enough context at this point to set the stack
376 * flags, permissions, and offset, so we use temporary values. We'll update
377 * them later in setup_arg_pages().
379 static int bprm_mm_init(struct linux_binprm *bprm)
382 struct mm_struct *mm = NULL;
384 bprm->mm = mm = mm_alloc();
389 err = __bprm_mm_init(bprm);
404 struct user_arg_ptr {
409 const char __user *const __user *native;
411 const compat_uptr_t __user *compat;
416 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
418 const char __user *native;
421 if (unlikely(argv.is_compat)) {
422 compat_uptr_t compat;
424 if (get_user(compat, argv.ptr.compat + nr))
425 return ERR_PTR(-EFAULT);
427 return compat_ptr(compat);
431 if (get_user(native, argv.ptr.native + nr))
432 return ERR_PTR(-EFAULT);
438 * count() counts the number of strings in array ARGV.
440 static int count(struct user_arg_ptr argv, int max)
444 if (argv.ptr.native != NULL) {
446 const char __user *p = get_user_arg_ptr(argv, i);
458 if (fatal_signal_pending(current))
459 return -ERESTARTNOHAND;
467 * 'copy_strings()' copies argument/environment strings from the old
468 * processes's memory to the new process's stack. The call to get_user_pages()
469 * ensures the destination page is created and not swapped out.
471 static int copy_strings(int argc, struct user_arg_ptr argv,
472 struct linux_binprm *bprm)
474 struct page *kmapped_page = NULL;
476 unsigned long kpos = 0;
480 const char __user *str;
485 str = get_user_arg_ptr(argv, argc);
489 len = strnlen_user(str, MAX_ARG_STRLEN);
494 if (!valid_arg_len(bprm, len))
497 /* We're going to work our way backwords. */
503 int offset, bytes_to_copy;
505 if (fatal_signal_pending(current)) {
506 ret = -ERESTARTNOHAND;
511 offset = pos % PAGE_SIZE;
515 bytes_to_copy = offset;
516 if (bytes_to_copy > len)
519 offset -= bytes_to_copy;
520 pos -= bytes_to_copy;
521 str -= bytes_to_copy;
522 len -= bytes_to_copy;
524 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
527 page = get_arg_page(bprm, pos, 1);
534 flush_kernel_dcache_page(kmapped_page);
535 kunmap(kmapped_page);
536 put_arg_page(kmapped_page);
539 kaddr = kmap(kmapped_page);
540 kpos = pos & PAGE_MASK;
541 flush_arg_page(bprm, kpos, kmapped_page);
543 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
552 flush_kernel_dcache_page(kmapped_page);
553 kunmap(kmapped_page);
554 put_arg_page(kmapped_page);
560 * Like copy_strings, but get argv and its values from kernel memory.
562 int copy_strings_kernel(int argc, const char *const *__argv,
563 struct linux_binprm *bprm)
566 mm_segment_t oldfs = get_fs();
567 struct user_arg_ptr argv = {
568 .ptr.native = (const char __user *const __user *)__argv,
572 r = copy_strings(argc, argv, bprm);
577 EXPORT_SYMBOL(copy_strings_kernel);
582 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
583 * the binfmt code determines where the new stack should reside, we shift it to
584 * its final location. The process proceeds as follows:
586 * 1) Use shift to calculate the new vma endpoints.
587 * 2) Extend vma to cover both the old and new ranges. This ensures the
588 * arguments passed to subsequent functions are consistent.
589 * 3) Move vma's page tables to the new range.
590 * 4) Free up any cleared pgd range.
591 * 5) Shrink the vma to cover only the new range.
593 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
595 struct mm_struct *mm = vma->vm_mm;
596 unsigned long old_start = vma->vm_start;
597 unsigned long old_end = vma->vm_end;
598 unsigned long length = old_end - old_start;
599 unsigned long new_start = old_start - shift;
600 unsigned long new_end = old_end - shift;
601 struct mmu_gather tlb;
603 BUG_ON(new_start > new_end);
606 * ensure there are no vmas between where we want to go
609 if (vma != find_vma(mm, new_start))
613 * cover the whole range: [new_start, old_end)
615 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
619 * move the page tables downwards, on failure we rely on
620 * process cleanup to remove whatever mess we made.
622 if (length != move_page_tables(vma, old_start,
623 vma, new_start, length, false))
627 tlb_gather_mmu(&tlb, mm, old_start, old_end);
628 if (new_end > old_start) {
630 * when the old and new regions overlap clear from new_end.
632 free_pgd_range(&tlb, new_end, old_end, new_end,
633 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
636 * otherwise, clean from old_start; this is done to not touch
637 * the address space in [new_end, old_start) some architectures
638 * have constraints on va-space that make this illegal (IA64) -
639 * for the others its just a little faster.
641 free_pgd_range(&tlb, old_start, old_end, new_end,
642 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
644 tlb_finish_mmu(&tlb, old_start, old_end);
647 * Shrink the vma to just the new range. Always succeeds.
649 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
655 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
656 * the stack is optionally relocated, and some extra space is added.
658 int setup_arg_pages(struct linux_binprm *bprm,
659 unsigned long stack_top,
660 int executable_stack)
663 unsigned long stack_shift;
664 struct mm_struct *mm = current->mm;
665 struct vm_area_struct *vma = bprm->vma;
666 struct vm_area_struct *prev = NULL;
667 unsigned long vm_flags;
668 unsigned long stack_base;
669 unsigned long stack_size;
670 unsigned long stack_expand;
671 unsigned long rlim_stack;
673 #ifdef CONFIG_STACK_GROWSUP
674 /* Limit stack size */
675 stack_base = rlimit_max(RLIMIT_STACK);
676 if (stack_base > STACK_SIZE_MAX)
677 stack_base = STACK_SIZE_MAX;
679 /* Add space for stack randomization. */
680 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
682 /* Make sure we didn't let the argument array grow too large. */
683 if (vma->vm_end - vma->vm_start > stack_base)
686 stack_base = PAGE_ALIGN(stack_top - stack_base);
688 stack_shift = vma->vm_start - stack_base;
689 mm->arg_start = bprm->p - stack_shift;
690 bprm->p = vma->vm_end - stack_shift;
692 stack_top = arch_align_stack(stack_top);
693 stack_top = PAGE_ALIGN(stack_top);
695 if (unlikely(stack_top < mmap_min_addr) ||
696 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
699 stack_shift = vma->vm_end - stack_top;
701 bprm->p -= stack_shift;
702 mm->arg_start = bprm->p;
706 bprm->loader -= stack_shift;
707 bprm->exec -= stack_shift;
709 if (down_write_killable(&mm->mmap_sem))
712 vm_flags = VM_STACK_FLAGS;
715 * Adjust stack execute permissions; explicitly enable for
716 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
717 * (arch default) otherwise.
719 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
721 else if (executable_stack == EXSTACK_DISABLE_X)
722 vm_flags &= ~VM_EXEC;
723 vm_flags |= mm->def_flags;
724 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
726 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
732 /* Move stack pages down in memory. */
734 ret = shift_arg_pages(vma, stack_shift);
739 /* mprotect_fixup is overkill to remove the temporary stack flags */
740 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
742 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
743 stack_size = vma->vm_end - vma->vm_start;
745 * Align this down to a page boundary as expand_stack
748 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
749 #ifdef CONFIG_STACK_GROWSUP
750 if (stack_size + stack_expand > rlim_stack)
751 stack_base = vma->vm_start + rlim_stack;
753 stack_base = vma->vm_end + stack_expand;
755 if (stack_size + stack_expand > rlim_stack)
756 stack_base = vma->vm_end - rlim_stack;
758 stack_base = vma->vm_start - stack_expand;
760 current->mm->start_stack = bprm->p;
761 ret = expand_stack(vma, stack_base);
766 up_write(&mm->mmap_sem);
769 EXPORT_SYMBOL(setup_arg_pages);
774 * Transfer the program arguments and environment from the holding pages
775 * onto the stack. The provided stack pointer is adjusted accordingly.
777 int transfer_args_to_stack(struct linux_binprm *bprm,
778 unsigned long *sp_location)
780 unsigned long index, stop, sp;
783 stop = bprm->p >> PAGE_SHIFT;
786 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
787 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
788 char *src = kmap(bprm->page[index]) + offset;
789 sp -= PAGE_SIZE - offset;
790 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
792 kunmap(bprm->page[index]);
802 EXPORT_SYMBOL(transfer_args_to_stack);
804 #endif /* CONFIG_MMU */
806 static struct file *do_open_execat(int fd, struct filename *name, int flags)
810 struct open_flags open_exec_flags = {
811 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
812 .acc_mode = MAY_EXEC,
813 .intent = LOOKUP_OPEN,
814 .lookup_flags = LOOKUP_FOLLOW,
817 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
818 return ERR_PTR(-EINVAL);
819 if (flags & AT_SYMLINK_NOFOLLOW)
820 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
821 if (flags & AT_EMPTY_PATH)
822 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
824 file = do_filp_open(fd, name, &open_exec_flags);
829 if (!S_ISREG(file_inode(file)->i_mode))
832 if (path_noexec(&file->f_path))
835 err = deny_write_access(file);
839 if (name->name[0] != '\0')
850 struct file *open_exec(const char *name)
852 struct filename *filename = getname_kernel(name);
853 struct file *f = ERR_CAST(filename);
855 if (!IS_ERR(filename)) {
856 f = do_open_execat(AT_FDCWD, filename, 0);
861 EXPORT_SYMBOL(open_exec);
863 int kernel_read(struct file *file, loff_t offset,
864 char *addr, unsigned long count)
872 /* The cast to a user pointer is valid due to the set_fs() */
873 result = vfs_read(file, (void __user *)addr, count, &pos);
878 EXPORT_SYMBOL(kernel_read);
880 int kernel_read_file(struct file *file, void **buf, loff_t *size,
881 loff_t max_size, enum kernel_read_file_id id)
887 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
890 ret = security_kernel_read_file(file, id);
894 ret = deny_write_access(file);
898 i_size = i_size_read(file_inode(file));
899 if (max_size > 0 && i_size > max_size) {
908 if (id != READING_FIRMWARE_PREALLOC_BUFFER)
909 *buf = vmalloc(i_size);
916 while (pos < i_size) {
917 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
934 ret = security_kernel_post_read_file(file, *buf, i_size, id);
940 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
947 allow_write_access(file);
950 EXPORT_SYMBOL_GPL(kernel_read_file);
952 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
953 loff_t max_size, enum kernel_read_file_id id)
961 file = filp_open(path, O_RDONLY, 0);
963 return PTR_ERR(file);
965 ret = kernel_read_file(file, buf, size, max_size, id);
969 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
971 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
972 enum kernel_read_file_id id)
974 struct fd f = fdget(fd);
980 ret = kernel_read_file(f.file, buf, size, max_size, id);
985 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
987 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
989 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
991 flush_icache_range(addr, addr + len);
994 EXPORT_SYMBOL(read_code);
996 static int exec_mmap(struct mm_struct *mm)
998 struct task_struct *tsk;
999 struct mm_struct *old_mm, *active_mm;
1001 /* Notify parent that we're no longer interested in the old VM */
1003 old_mm = current->mm;
1004 mm_release(tsk, old_mm);
1007 sync_mm_rss(old_mm);
1009 * Make sure that if there is a core dump in progress
1010 * for the old mm, we get out and die instead of going
1011 * through with the exec. We must hold mmap_sem around
1012 * checking core_state and changing tsk->mm.
1014 down_read(&old_mm->mmap_sem);
1015 if (unlikely(old_mm->core_state)) {
1016 up_read(&old_mm->mmap_sem);
1021 active_mm = tsk->active_mm;
1023 tsk->active_mm = mm;
1024 activate_mm(active_mm, mm);
1025 tsk->mm->vmacache_seqnum = 0;
1026 vmacache_flush(tsk);
1029 up_read(&old_mm->mmap_sem);
1030 BUG_ON(active_mm != old_mm);
1031 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1032 mm_update_next_owner(old_mm);
1041 * This function makes sure the current process has its own signal table,
1042 * so that flush_signal_handlers can later reset the handlers without
1043 * disturbing other processes. (Other processes might share the signal
1044 * table via the CLONE_SIGHAND option to clone().)
1046 static int de_thread(struct task_struct *tsk)
1048 struct signal_struct *sig = tsk->signal;
1049 struct sighand_struct *oldsighand = tsk->sighand;
1050 spinlock_t *lock = &oldsighand->siglock;
1052 if (thread_group_empty(tsk))
1053 goto no_thread_group;
1056 * Kill all other threads in the thread group.
1058 spin_lock_irq(lock);
1059 if (signal_group_exit(sig)) {
1061 * Another group action in progress, just
1062 * return so that the signal is processed.
1064 spin_unlock_irq(lock);
1068 sig->group_exit_task = tsk;
1069 sig->notify_count = zap_other_threads(tsk);
1070 if (!thread_group_leader(tsk))
1071 sig->notify_count--;
1073 while (sig->notify_count) {
1074 __set_current_state(TASK_KILLABLE);
1075 spin_unlock_irq(lock);
1077 if (unlikely(__fatal_signal_pending(tsk)))
1079 spin_lock_irq(lock);
1081 spin_unlock_irq(lock);
1084 * At this point all other threads have exited, all we have to
1085 * do is to wait for the thread group leader to become inactive,
1086 * and to assume its PID:
1088 if (!thread_group_leader(tsk)) {
1089 struct task_struct *leader = tsk->group_leader;
1092 cgroup_threadgroup_change_begin(tsk);
1093 write_lock_irq(&tasklist_lock);
1095 * Do this under tasklist_lock to ensure that
1096 * exit_notify() can't miss ->group_exit_task
1098 sig->notify_count = -1;
1099 if (likely(leader->exit_state))
1101 __set_current_state(TASK_KILLABLE);
1102 write_unlock_irq(&tasklist_lock);
1103 cgroup_threadgroup_change_end(tsk);
1105 if (unlikely(__fatal_signal_pending(tsk)))
1110 * The only record we have of the real-time age of a
1111 * process, regardless of execs it's done, is start_time.
1112 * All the past CPU time is accumulated in signal_struct
1113 * from sister threads now dead. But in this non-leader
1114 * exec, nothing survives from the original leader thread,
1115 * whose birth marks the true age of this process now.
1116 * When we take on its identity by switching to its PID, we
1117 * also take its birthdate (always earlier than our own).
1119 tsk->start_time = leader->start_time;
1120 tsk->real_start_time = leader->real_start_time;
1122 BUG_ON(!same_thread_group(leader, tsk));
1123 BUG_ON(has_group_leader_pid(tsk));
1125 * An exec() starts a new thread group with the
1126 * TGID of the previous thread group. Rehash the
1127 * two threads with a switched PID, and release
1128 * the former thread group leader:
1131 /* Become a process group leader with the old leader's pid.
1132 * The old leader becomes a thread of the this thread group.
1133 * Note: The old leader also uses this pid until release_task
1134 * is called. Odd but simple and correct.
1136 tsk->pid = leader->pid;
1137 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1138 transfer_pid(leader, tsk, PIDTYPE_PGID);
1139 transfer_pid(leader, tsk, PIDTYPE_SID);
1141 list_replace_rcu(&leader->tasks, &tsk->tasks);
1142 list_replace_init(&leader->sibling, &tsk->sibling);
1144 tsk->group_leader = tsk;
1145 leader->group_leader = tsk;
1147 tsk->exit_signal = SIGCHLD;
1148 leader->exit_signal = -1;
1150 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1151 leader->exit_state = EXIT_DEAD;
1154 * We are going to release_task()->ptrace_unlink() silently,
1155 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1156 * the tracer wont't block again waiting for this thread.
1158 if (unlikely(leader->ptrace))
1159 __wake_up_parent(leader, leader->parent);
1160 write_unlock_irq(&tasklist_lock);
1161 cgroup_threadgroup_change_end(tsk);
1163 release_task(leader);
1166 sig->group_exit_task = NULL;
1167 sig->notify_count = 0;
1170 /* we have changed execution domain */
1171 tsk->exit_signal = SIGCHLD;
1173 #ifdef CONFIG_POSIX_TIMERS
1175 flush_itimer_signals();
1178 if (atomic_read(&oldsighand->count) != 1) {
1179 struct sighand_struct *newsighand;
1181 * This ->sighand is shared with the CLONE_SIGHAND
1182 * but not CLONE_THREAD task, switch to the new one.
1184 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1188 atomic_set(&newsighand->count, 1);
1189 memcpy(newsighand->action, oldsighand->action,
1190 sizeof(newsighand->action));
1192 write_lock_irq(&tasklist_lock);
1193 spin_lock(&oldsighand->siglock);
1194 rcu_assign_pointer(tsk->sighand, newsighand);
1195 spin_unlock(&oldsighand->siglock);
1196 write_unlock_irq(&tasklist_lock);
1198 __cleanup_sighand(oldsighand);
1201 BUG_ON(!thread_group_leader(tsk));
1205 /* protects against exit_notify() and __exit_signal() */
1206 read_lock(&tasklist_lock);
1207 sig->group_exit_task = NULL;
1208 sig->notify_count = 0;
1209 read_unlock(&tasklist_lock);
1213 char *get_task_comm(char *buf, struct task_struct *tsk)
1215 /* buf must be at least sizeof(tsk->comm) in size */
1217 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1221 EXPORT_SYMBOL_GPL(get_task_comm);
1224 * These functions flushes out all traces of the currently running executable
1225 * so that a new one can be started
1228 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1231 trace_task_rename(tsk, buf);
1232 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1234 perf_event_comm(tsk, exec);
1237 int flush_old_exec(struct linux_binprm * bprm)
1242 * Make sure we have a private signal table and that
1243 * we are unassociated from the previous thread group.
1245 retval = de_thread(current);
1250 * Must be called _before_ exec_mmap() as bprm->mm is
1251 * not visibile until then. This also enables the update
1254 set_mm_exe_file(bprm->mm, bprm->file);
1257 * Release all of the old mmap stuff
1259 acct_arg_size(bprm, 0);
1260 retval = exec_mmap(bprm->mm);
1264 bprm->mm = NULL; /* We're using it now */
1267 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1268 PF_NOFREEZE | PF_NO_SETAFFINITY);
1270 current->personality &= ~bprm->per_clear;
1273 * We have to apply CLOEXEC before we change whether the process is
1274 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1275 * trying to access the should-be-closed file descriptors of a process
1276 * undergoing exec(2).
1278 do_close_on_exec(current->files);
1284 EXPORT_SYMBOL(flush_old_exec);
1286 void would_dump(struct linux_binprm *bprm, struct file *file)
1288 struct inode *inode = file_inode(file);
1289 if (inode_permission(inode, MAY_READ) < 0) {
1290 struct user_namespace *old, *user_ns;
1291 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1293 /* Ensure mm->user_ns contains the executable */
1294 user_ns = old = bprm->mm->user_ns;
1295 while ((user_ns != &init_user_ns) &&
1296 !privileged_wrt_inode_uidgid(user_ns, inode))
1297 user_ns = user_ns->parent;
1299 if (old != user_ns) {
1300 bprm->mm->user_ns = get_user_ns(user_ns);
1305 EXPORT_SYMBOL(would_dump);
1307 void setup_new_exec(struct linux_binprm * bprm)
1309 arch_pick_mmap_layout(current->mm);
1311 /* This is the point of no return */
1312 current->sas_ss_sp = current->sas_ss_size = 0;
1314 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1315 set_dumpable(current->mm, SUID_DUMP_USER);
1317 set_dumpable(current->mm, suid_dumpable);
1320 __set_task_comm(current, kbasename(bprm->filename), true);
1322 /* Set the new mm task size. We have to do that late because it may
1323 * depend on TIF_32BIT which is only updated in flush_thread() on
1324 * some architectures like powerpc
1326 current->mm->task_size = TASK_SIZE;
1328 /* install the new credentials */
1329 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1330 !gid_eq(bprm->cred->gid, current_egid())) {
1331 current->pdeath_signal = 0;
1333 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1334 set_dumpable(current->mm, suid_dumpable);
1337 /* An exec changes our domain. We are no longer part of the thread
1339 current->self_exec_id++;
1340 flush_signal_handlers(current, 0);
1342 EXPORT_SYMBOL(setup_new_exec);
1345 * Prepare credentials and lock ->cred_guard_mutex.
1346 * install_exec_creds() commits the new creds and drops the lock.
1347 * Or, if exec fails before, free_bprm() should release ->cred and
1350 int prepare_bprm_creds(struct linux_binprm *bprm)
1352 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1353 return -ERESTARTNOINTR;
1355 bprm->cred = prepare_exec_creds();
1356 if (likely(bprm->cred))
1359 mutex_unlock(¤t->signal->cred_guard_mutex);
1363 static void free_bprm(struct linux_binprm *bprm)
1365 free_arg_pages(bprm);
1367 mutex_unlock(¤t->signal->cred_guard_mutex);
1368 abort_creds(bprm->cred);
1371 allow_write_access(bprm->file);
1374 /* If a binfmt changed the interp, free it. */
1375 if (bprm->interp != bprm->filename)
1376 kfree(bprm->interp);
1380 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1382 /* If a binfmt changed the interp, free it first. */
1383 if (bprm->interp != bprm->filename)
1384 kfree(bprm->interp);
1385 bprm->interp = kstrdup(interp, GFP_KERNEL);
1390 EXPORT_SYMBOL(bprm_change_interp);
1393 * install the new credentials for this executable
1395 void install_exec_creds(struct linux_binprm *bprm)
1397 security_bprm_committing_creds(bprm);
1399 commit_creds(bprm->cred);
1403 * Disable monitoring for regular users
1404 * when executing setuid binaries. Must
1405 * wait until new credentials are committed
1406 * by commit_creds() above
1408 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1409 perf_event_exit_task(current);
1411 * cred_guard_mutex must be held at least to this point to prevent
1412 * ptrace_attach() from altering our determination of the task's
1413 * credentials; any time after this it may be unlocked.
1415 security_bprm_committed_creds(bprm);
1416 mutex_unlock(¤t->signal->cred_guard_mutex);
1418 EXPORT_SYMBOL(install_exec_creds);
1421 * determine how safe it is to execute the proposed program
1422 * - the caller must hold ->cred_guard_mutex to protect against
1423 * PTRACE_ATTACH or seccomp thread-sync
1425 static void check_unsafe_exec(struct linux_binprm *bprm)
1427 struct task_struct *p = current, *t;
1431 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1434 * This isn't strictly necessary, but it makes it harder for LSMs to
1437 if (task_no_new_privs(current))
1438 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1442 spin_lock(&p->fs->lock);
1444 while_each_thread(p, t) {
1450 if (p->fs->users > n_fs)
1451 bprm->unsafe |= LSM_UNSAFE_SHARE;
1454 spin_unlock(&p->fs->lock);
1457 static void bprm_fill_uid(struct linux_binprm *bprm)
1459 struct inode *inode;
1465 * Since this can be called multiple times (via prepare_binprm),
1466 * we must clear any previous work done when setting set[ug]id
1467 * bits from any earlier bprm->file uses (for example when run
1468 * first for a setuid script then again for its interpreter).
1470 bprm->cred->euid = current_euid();
1471 bprm->cred->egid = current_egid();
1473 if (!mnt_may_suid(bprm->file->f_path.mnt))
1476 if (task_no_new_privs(current))
1479 inode = bprm->file->f_path.dentry->d_inode;
1480 mode = READ_ONCE(inode->i_mode);
1481 if (!(mode & (S_ISUID|S_ISGID)))
1484 /* Be careful if suid/sgid is set */
1487 /* reload atomically mode/uid/gid now that lock held */
1488 mode = inode->i_mode;
1491 inode_unlock(inode);
1493 /* We ignore suid/sgid if there are no mappings for them in the ns */
1494 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1495 !kgid_has_mapping(bprm->cred->user_ns, gid))
1498 if (mode & S_ISUID) {
1499 bprm->per_clear |= PER_CLEAR_ON_SETID;
1500 bprm->cred->euid = uid;
1503 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1504 bprm->per_clear |= PER_CLEAR_ON_SETID;
1505 bprm->cred->egid = gid;
1510 * Fill the binprm structure from the inode.
1511 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1513 * This may be called multiple times for binary chains (scripts for example).
1515 int prepare_binprm(struct linux_binprm *bprm)
1519 bprm_fill_uid(bprm);
1521 /* fill in binprm security blob */
1522 retval = security_bprm_set_creds(bprm);
1525 bprm->cred_prepared = 1;
1527 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1528 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1531 EXPORT_SYMBOL(prepare_binprm);
1534 * Arguments are '\0' separated strings found at the location bprm->p
1535 * points to; chop off the first by relocating brpm->p to right after
1536 * the first '\0' encountered.
1538 int remove_arg_zero(struct linux_binprm *bprm)
1541 unsigned long offset;
1549 offset = bprm->p & ~PAGE_MASK;
1550 page = get_arg_page(bprm, bprm->p, 0);
1555 kaddr = kmap_atomic(page);
1557 for (; offset < PAGE_SIZE && kaddr[offset];
1558 offset++, bprm->p++)
1561 kunmap_atomic(kaddr);
1563 } while (offset == PAGE_SIZE);
1572 EXPORT_SYMBOL(remove_arg_zero);
1574 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1576 * cycle the list of binary formats handler, until one recognizes the image
1578 int search_binary_handler(struct linux_binprm *bprm)
1580 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1581 struct linux_binfmt *fmt;
1584 /* This allows 4 levels of binfmt rewrites before failing hard. */
1585 if (bprm->recursion_depth > 5)
1588 retval = security_bprm_check(bprm);
1594 read_lock(&binfmt_lock);
1595 list_for_each_entry(fmt, &formats, lh) {
1596 if (!try_module_get(fmt->module))
1598 read_unlock(&binfmt_lock);
1599 bprm->recursion_depth++;
1600 retval = fmt->load_binary(bprm);
1601 read_lock(&binfmt_lock);
1603 bprm->recursion_depth--;
1604 if (retval < 0 && !bprm->mm) {
1605 /* we got to flush_old_exec() and failed after it */
1606 read_unlock(&binfmt_lock);
1607 force_sigsegv(SIGSEGV, current);
1610 if (retval != -ENOEXEC || !bprm->file) {
1611 read_unlock(&binfmt_lock);
1615 read_unlock(&binfmt_lock);
1618 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1619 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1621 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1629 EXPORT_SYMBOL(search_binary_handler);
1631 static int exec_binprm(struct linux_binprm *bprm)
1633 pid_t old_pid, old_vpid;
1636 /* Need to fetch pid before load_binary changes it */
1637 old_pid = current->pid;
1639 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1642 ret = search_binary_handler(bprm);
1645 trace_sched_process_exec(current, old_pid, bprm);
1646 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1647 proc_exec_connector(current);
1654 * sys_execve() executes a new program.
1656 static int do_execveat_common(int fd, struct filename *filename,
1657 struct user_arg_ptr argv,
1658 struct user_arg_ptr envp,
1661 char *pathbuf = NULL;
1662 struct linux_binprm *bprm;
1664 struct files_struct *displaced;
1667 if (IS_ERR(filename))
1668 return PTR_ERR(filename);
1671 * We move the actual failure in case of RLIMIT_NPROC excess from
1672 * set*uid() to execve() because too many poorly written programs
1673 * don't check setuid() return code. Here we additionally recheck
1674 * whether NPROC limit is still exceeded.
1676 if ((current->flags & PF_NPROC_EXCEEDED) &&
1677 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1682 /* We're below the limit (still or again), so we don't want to make
1683 * further execve() calls fail. */
1684 current->flags &= ~PF_NPROC_EXCEEDED;
1686 retval = unshare_files(&displaced);
1691 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1695 retval = prepare_bprm_creds(bprm);
1699 check_unsafe_exec(bprm);
1700 current->in_execve = 1;
1702 file = do_open_execat(fd, filename, flags);
1703 retval = PTR_ERR(file);
1710 if (fd == AT_FDCWD || filename->name[0] == '/') {
1711 bprm->filename = filename->name;
1713 if (filename->name[0] == '\0')
1714 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1716 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1717 fd, filename->name);
1723 * Record that a name derived from an O_CLOEXEC fd will be
1724 * inaccessible after exec. Relies on having exclusive access to
1725 * current->files (due to unshare_files above).
1727 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1728 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1729 bprm->filename = pathbuf;
1731 bprm->interp = bprm->filename;
1733 retval = bprm_mm_init(bprm);
1737 bprm->argc = count(argv, MAX_ARG_STRINGS);
1738 if ((retval = bprm->argc) < 0)
1741 bprm->envc = count(envp, MAX_ARG_STRINGS);
1742 if ((retval = bprm->envc) < 0)
1745 retval = prepare_binprm(bprm);
1749 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1753 bprm->exec = bprm->p;
1754 retval = copy_strings(bprm->envc, envp, bprm);
1758 retval = copy_strings(bprm->argc, argv, bprm);
1762 would_dump(bprm, bprm->file);
1764 retval = exec_binprm(bprm);
1768 /* execve succeeded */
1769 current->fs->in_exec = 0;
1770 current->in_execve = 0;
1771 acct_update_integrals(current);
1772 task_numa_free(current);
1777 put_files_struct(displaced);
1782 acct_arg_size(bprm, 0);
1787 current->fs->in_exec = 0;
1788 current->in_execve = 0;
1796 reset_files_struct(displaced);
1802 int do_execve(struct filename *filename,
1803 const char __user *const __user *__argv,
1804 const char __user *const __user *__envp)
1806 struct user_arg_ptr argv = { .ptr.native = __argv };
1807 struct user_arg_ptr envp = { .ptr.native = __envp };
1808 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1811 int do_execveat(int fd, struct filename *filename,
1812 const char __user *const __user *__argv,
1813 const char __user *const __user *__envp,
1816 struct user_arg_ptr argv = { .ptr.native = __argv };
1817 struct user_arg_ptr envp = { .ptr.native = __envp };
1819 return do_execveat_common(fd, filename, argv, envp, flags);
1822 #ifdef CONFIG_COMPAT
1823 static int compat_do_execve(struct filename *filename,
1824 const compat_uptr_t __user *__argv,
1825 const compat_uptr_t __user *__envp)
1827 struct user_arg_ptr argv = {
1829 .ptr.compat = __argv,
1831 struct user_arg_ptr envp = {
1833 .ptr.compat = __envp,
1835 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1838 static int compat_do_execveat(int fd, struct filename *filename,
1839 const compat_uptr_t __user *__argv,
1840 const compat_uptr_t __user *__envp,
1843 struct user_arg_ptr argv = {
1845 .ptr.compat = __argv,
1847 struct user_arg_ptr envp = {
1849 .ptr.compat = __envp,
1851 return do_execveat_common(fd, filename, argv, envp, flags);
1855 void set_binfmt(struct linux_binfmt *new)
1857 struct mm_struct *mm = current->mm;
1860 module_put(mm->binfmt->module);
1864 __module_get(new->module);
1866 EXPORT_SYMBOL(set_binfmt);
1869 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1871 void set_dumpable(struct mm_struct *mm, int value)
1873 unsigned long old, new;
1875 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1879 old = ACCESS_ONCE(mm->flags);
1880 new = (old & ~MMF_DUMPABLE_MASK) | value;
1881 } while (cmpxchg(&mm->flags, old, new) != old);
1884 SYSCALL_DEFINE3(execve,
1885 const char __user *, filename,
1886 const char __user *const __user *, argv,
1887 const char __user *const __user *, envp)
1889 return do_execve(getname(filename), argv, envp);
1892 SYSCALL_DEFINE5(execveat,
1893 int, fd, const char __user *, filename,
1894 const char __user *const __user *, argv,
1895 const char __user *const __user *, envp,
1898 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1900 return do_execveat(fd,
1901 getname_flags(filename, lookup_flags, NULL),
1905 #ifdef CONFIG_COMPAT
1906 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1907 const compat_uptr_t __user *, argv,
1908 const compat_uptr_t __user *, envp)
1910 return compat_do_execve(getname(filename), argv, envp);
1913 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1914 const char __user *, filename,
1915 const compat_uptr_t __user *, argv,
1916 const compat_uptr_t __user *, envp,
1919 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1921 return compat_do_execveat(fd,
1922 getname_flags(filename, lookup_flags, NULL),