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/sched/signal.h>
38 #include <linux/pagemap.h>
39 #include <linux/perf_event.h>
40 #include <linux/highmem.h>
41 #include <linux/spinlock.h>
42 #include <linux/key.h>
43 #include <linux/personality.h>
44 #include <linux/binfmts.h>
45 #include <linux/utsname.h>
46 #include <linux/pid_namespace.h>
47 #include <linux/module.h>
48 #include <linux/namei.h>
49 #include <linux/mount.h>
50 #include <linux/security.h>
51 #include <linux/syscalls.h>
52 #include <linux/tsacct_kern.h>
53 #include <linux/cn_proc.h>
54 #include <linux/audit.h>
55 #include <linux/tracehook.h>
56 #include <linux/kmod.h>
57 #include <linux/fsnotify.h>
58 #include <linux/fs_struct.h>
59 #include <linux/pipe_fs_i.h>
60 #include <linux/oom.h>
61 #include <linux/compat.h>
62 #include <linux/vmalloc.h>
64 #include <linux/uaccess.h>
65 #include <asm/mmu_context.h>
68 #include <trace/events/task.h>
71 #include <trace/events/sched.h>
73 int suid_dumpable = 0;
75 static LIST_HEAD(formats);
76 static DEFINE_RWLOCK(binfmt_lock);
78 void __register_binfmt(struct linux_binfmt * fmt, int insert)
81 if (WARN_ON(!fmt->load_binary))
83 write_lock(&binfmt_lock);
84 insert ? list_add(&fmt->lh, &formats) :
85 list_add_tail(&fmt->lh, &formats);
86 write_unlock(&binfmt_lock);
89 EXPORT_SYMBOL(__register_binfmt);
91 void unregister_binfmt(struct linux_binfmt * fmt)
93 write_lock(&binfmt_lock);
95 write_unlock(&binfmt_lock);
98 EXPORT_SYMBOL(unregister_binfmt);
100 static inline void put_binfmt(struct linux_binfmt * fmt)
102 module_put(fmt->module);
105 bool path_noexec(const struct path *path)
107 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
108 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
113 * Note that a shared library must be both readable and executable due to
116 * Also note that we take the address to load from from the file itself.
118 SYSCALL_DEFINE1(uselib, const char __user *, library)
120 struct linux_binfmt *fmt;
122 struct filename *tmp = getname(library);
123 int error = PTR_ERR(tmp);
124 static const struct open_flags uselib_flags = {
125 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
126 .acc_mode = MAY_READ | MAY_EXEC,
127 .intent = LOOKUP_OPEN,
128 .lookup_flags = LOOKUP_FOLLOW,
134 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
136 error = PTR_ERR(file);
141 if (!S_ISREG(file_inode(file)->i_mode))
145 if (path_noexec(&file->f_path))
152 read_lock(&binfmt_lock);
153 list_for_each_entry(fmt, &formats, lh) {
154 if (!fmt->load_shlib)
156 if (!try_module_get(fmt->module))
158 read_unlock(&binfmt_lock);
159 error = fmt->load_shlib(file);
160 read_lock(&binfmt_lock);
162 if (error != -ENOEXEC)
165 read_unlock(&binfmt_lock);
171 #endif /* #ifdef CONFIG_USELIB */
175 * The nascent bprm->mm is not visible until exec_mmap() but it can
176 * use a lot of memory, account these pages in current->mm temporary
177 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
178 * change the counter back via acct_arg_size(0).
180 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
182 struct mm_struct *mm = current->mm;
183 long diff = (long)(pages - bprm->vma_pages);
188 bprm->vma_pages = pages;
189 add_mm_counter(mm, MM_ANONPAGES, diff);
192 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
197 unsigned int gup_flags = FOLL_FORCE;
199 #ifdef CONFIG_STACK_GROWSUP
201 ret = expand_downwards(bprm->vma, pos);
208 gup_flags |= FOLL_WRITE;
211 * We are doing an exec(). 'current' is the process
212 * doing the exec and bprm->mm is the new process's mm.
214 ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
220 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
223 acct_arg_size(bprm, size / PAGE_SIZE);
226 * We've historically supported up to 32 pages (ARG_MAX)
227 * of argument strings even with small stacks
233 * Limit to 1/4-th the stack size for the argv+env strings.
235 * - the remaining binfmt code will not run out of stack space,
236 * - the program will have a reasonable amount of stack left
239 rlim = current->signal->rlim;
240 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
249 static void put_arg_page(struct page *page)
254 static void free_arg_pages(struct linux_binprm *bprm)
258 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
261 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
264 static int __bprm_mm_init(struct linux_binprm *bprm)
267 struct vm_area_struct *vma = NULL;
268 struct mm_struct *mm = bprm->mm;
270 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
274 if (down_write_killable(&mm->mmap_sem)) {
281 * Place the stack at the largest stack address the architecture
282 * supports. Later, we'll move this to an appropriate place. We don't
283 * use STACK_TOP because that can depend on attributes which aren't
286 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
287 vma->vm_end = STACK_TOP_MAX;
288 vma->vm_start = vma->vm_end - PAGE_SIZE;
289 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
290 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
291 INIT_LIST_HEAD(&vma->anon_vma_chain);
293 err = insert_vm_struct(mm, vma);
297 mm->stack_vm = mm->total_vm = 1;
298 arch_bprm_mm_init(mm, vma);
299 up_write(&mm->mmap_sem);
300 bprm->p = vma->vm_end - sizeof(void *);
303 up_write(&mm->mmap_sem);
306 kmem_cache_free(vm_area_cachep, vma);
310 static bool valid_arg_len(struct linux_binprm *bprm, long len)
312 return len <= MAX_ARG_STRLEN;
317 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
321 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
326 page = bprm->page[pos / PAGE_SIZE];
327 if (!page && write) {
328 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
331 bprm->page[pos / PAGE_SIZE] = page;
337 static void put_arg_page(struct page *page)
341 static void free_arg_page(struct linux_binprm *bprm, int i)
344 __free_page(bprm->page[i]);
345 bprm->page[i] = NULL;
349 static void free_arg_pages(struct linux_binprm *bprm)
353 for (i = 0; i < MAX_ARG_PAGES; i++)
354 free_arg_page(bprm, i);
357 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
362 static int __bprm_mm_init(struct linux_binprm *bprm)
364 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
368 static bool valid_arg_len(struct linux_binprm *bprm, long len)
370 return len <= bprm->p;
373 #endif /* CONFIG_MMU */
376 * Create a new mm_struct and populate it with a temporary stack
377 * vm_area_struct. We don't have enough context at this point to set the stack
378 * flags, permissions, and offset, so we use temporary values. We'll update
379 * them later in setup_arg_pages().
381 static int bprm_mm_init(struct linux_binprm *bprm)
384 struct mm_struct *mm = NULL;
386 bprm->mm = mm = mm_alloc();
391 err = __bprm_mm_init(bprm);
406 struct user_arg_ptr {
411 const char __user *const __user *native;
413 const compat_uptr_t __user *compat;
418 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
420 const char __user *native;
423 if (unlikely(argv.is_compat)) {
424 compat_uptr_t compat;
426 if (get_user(compat, argv.ptr.compat + nr))
427 return ERR_PTR(-EFAULT);
429 return compat_ptr(compat);
433 if (get_user(native, argv.ptr.native + nr))
434 return ERR_PTR(-EFAULT);
440 * count() counts the number of strings in array ARGV.
442 static int count(struct user_arg_ptr argv, int max)
446 if (argv.ptr.native != NULL) {
448 const char __user *p = get_user_arg_ptr(argv, i);
460 if (fatal_signal_pending(current))
461 return -ERESTARTNOHAND;
469 * 'copy_strings()' copies argument/environment strings from the old
470 * processes's memory to the new process's stack. The call to get_user_pages()
471 * ensures the destination page is created and not swapped out.
473 static int copy_strings(int argc, struct user_arg_ptr argv,
474 struct linux_binprm *bprm)
476 struct page *kmapped_page = NULL;
478 unsigned long kpos = 0;
482 const char __user *str;
487 str = get_user_arg_ptr(argv, argc);
491 len = strnlen_user(str, MAX_ARG_STRLEN);
496 if (!valid_arg_len(bprm, len))
499 /* We're going to work our way backwords. */
505 int offset, bytes_to_copy;
507 if (fatal_signal_pending(current)) {
508 ret = -ERESTARTNOHAND;
513 offset = pos % PAGE_SIZE;
517 bytes_to_copy = offset;
518 if (bytes_to_copy > len)
521 offset -= bytes_to_copy;
522 pos -= bytes_to_copy;
523 str -= bytes_to_copy;
524 len -= bytes_to_copy;
526 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
529 page = get_arg_page(bprm, pos, 1);
536 flush_kernel_dcache_page(kmapped_page);
537 kunmap(kmapped_page);
538 put_arg_page(kmapped_page);
541 kaddr = kmap(kmapped_page);
542 kpos = pos & PAGE_MASK;
543 flush_arg_page(bprm, kpos, kmapped_page);
545 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
554 flush_kernel_dcache_page(kmapped_page);
555 kunmap(kmapped_page);
556 put_arg_page(kmapped_page);
562 * Like copy_strings, but get argv and its values from kernel memory.
564 int copy_strings_kernel(int argc, const char *const *__argv,
565 struct linux_binprm *bprm)
568 mm_segment_t oldfs = get_fs();
569 struct user_arg_ptr argv = {
570 .ptr.native = (const char __user *const __user *)__argv,
574 r = copy_strings(argc, argv, bprm);
579 EXPORT_SYMBOL(copy_strings_kernel);
584 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
585 * the binfmt code determines where the new stack should reside, we shift it to
586 * its final location. The process proceeds as follows:
588 * 1) Use shift to calculate the new vma endpoints.
589 * 2) Extend vma to cover both the old and new ranges. This ensures the
590 * arguments passed to subsequent functions are consistent.
591 * 3) Move vma's page tables to the new range.
592 * 4) Free up any cleared pgd range.
593 * 5) Shrink the vma to cover only the new range.
595 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
597 struct mm_struct *mm = vma->vm_mm;
598 unsigned long old_start = vma->vm_start;
599 unsigned long old_end = vma->vm_end;
600 unsigned long length = old_end - old_start;
601 unsigned long new_start = old_start - shift;
602 unsigned long new_end = old_end - shift;
603 struct mmu_gather tlb;
605 BUG_ON(new_start > new_end);
608 * ensure there are no vmas between where we want to go
611 if (vma != find_vma(mm, new_start))
615 * cover the whole range: [new_start, old_end)
617 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
621 * move the page tables downwards, on failure we rely on
622 * process cleanup to remove whatever mess we made.
624 if (length != move_page_tables(vma, old_start,
625 vma, new_start, length, false))
629 tlb_gather_mmu(&tlb, mm, old_start, old_end);
630 if (new_end > old_start) {
632 * when the old and new regions overlap clear from new_end.
634 free_pgd_range(&tlb, new_end, old_end, new_end,
635 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
638 * otherwise, clean from old_start; this is done to not touch
639 * the address space in [new_end, old_start) some architectures
640 * have constraints on va-space that make this illegal (IA64) -
641 * for the others its just a little faster.
643 free_pgd_range(&tlb, old_start, old_end, new_end,
644 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
646 tlb_finish_mmu(&tlb, old_start, old_end);
649 * Shrink the vma to just the new range. Always succeeds.
651 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
657 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
658 * the stack is optionally relocated, and some extra space is added.
660 int setup_arg_pages(struct linux_binprm *bprm,
661 unsigned long stack_top,
662 int executable_stack)
665 unsigned long stack_shift;
666 struct mm_struct *mm = current->mm;
667 struct vm_area_struct *vma = bprm->vma;
668 struct vm_area_struct *prev = NULL;
669 unsigned long vm_flags;
670 unsigned long stack_base;
671 unsigned long stack_size;
672 unsigned long stack_expand;
673 unsigned long rlim_stack;
675 #ifdef CONFIG_STACK_GROWSUP
676 /* Limit stack size */
677 stack_base = rlimit_max(RLIMIT_STACK);
678 if (stack_base > STACK_SIZE_MAX)
679 stack_base = STACK_SIZE_MAX;
681 /* Add space for stack randomization. */
682 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
684 /* Make sure we didn't let the argument array grow too large. */
685 if (vma->vm_end - vma->vm_start > stack_base)
688 stack_base = PAGE_ALIGN(stack_top - stack_base);
690 stack_shift = vma->vm_start - stack_base;
691 mm->arg_start = bprm->p - stack_shift;
692 bprm->p = vma->vm_end - stack_shift;
694 stack_top = arch_align_stack(stack_top);
695 stack_top = PAGE_ALIGN(stack_top);
697 if (unlikely(stack_top < mmap_min_addr) ||
698 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
701 stack_shift = vma->vm_end - stack_top;
703 bprm->p -= stack_shift;
704 mm->arg_start = bprm->p;
708 bprm->loader -= stack_shift;
709 bprm->exec -= stack_shift;
711 if (down_write_killable(&mm->mmap_sem))
714 vm_flags = VM_STACK_FLAGS;
717 * Adjust stack execute permissions; explicitly enable for
718 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
719 * (arch default) otherwise.
721 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
723 else if (executable_stack == EXSTACK_DISABLE_X)
724 vm_flags &= ~VM_EXEC;
725 vm_flags |= mm->def_flags;
726 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
728 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
734 /* Move stack pages down in memory. */
736 ret = shift_arg_pages(vma, stack_shift);
741 /* mprotect_fixup is overkill to remove the temporary stack flags */
742 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
744 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
745 stack_size = vma->vm_end - vma->vm_start;
747 * Align this down to a page boundary as expand_stack
750 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
751 #ifdef CONFIG_STACK_GROWSUP
752 if (stack_size + stack_expand > rlim_stack)
753 stack_base = vma->vm_start + rlim_stack;
755 stack_base = vma->vm_end + stack_expand;
757 if (stack_size + stack_expand > rlim_stack)
758 stack_base = vma->vm_end - rlim_stack;
760 stack_base = vma->vm_start - stack_expand;
762 current->mm->start_stack = bprm->p;
763 ret = expand_stack(vma, stack_base);
768 up_write(&mm->mmap_sem);
771 EXPORT_SYMBOL(setup_arg_pages);
776 * Transfer the program arguments and environment from the holding pages
777 * onto the stack. The provided stack pointer is adjusted accordingly.
779 int transfer_args_to_stack(struct linux_binprm *bprm,
780 unsigned long *sp_location)
782 unsigned long index, stop, sp;
785 stop = bprm->p >> PAGE_SHIFT;
788 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
789 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
790 char *src = kmap(bprm->page[index]) + offset;
791 sp -= PAGE_SIZE - offset;
792 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
794 kunmap(bprm->page[index]);
804 EXPORT_SYMBOL(transfer_args_to_stack);
806 #endif /* CONFIG_MMU */
808 static struct file *do_open_execat(int fd, struct filename *name, int flags)
812 struct open_flags open_exec_flags = {
813 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
814 .acc_mode = MAY_EXEC,
815 .intent = LOOKUP_OPEN,
816 .lookup_flags = LOOKUP_FOLLOW,
819 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
820 return ERR_PTR(-EINVAL);
821 if (flags & AT_SYMLINK_NOFOLLOW)
822 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
823 if (flags & AT_EMPTY_PATH)
824 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
826 file = do_filp_open(fd, name, &open_exec_flags);
831 if (!S_ISREG(file_inode(file)->i_mode))
834 if (path_noexec(&file->f_path))
837 err = deny_write_access(file);
841 if (name->name[0] != '\0')
852 struct file *open_exec(const char *name)
854 struct filename *filename = getname_kernel(name);
855 struct file *f = ERR_CAST(filename);
857 if (!IS_ERR(filename)) {
858 f = do_open_execat(AT_FDCWD, filename, 0);
863 EXPORT_SYMBOL(open_exec);
865 int kernel_read(struct file *file, loff_t offset,
866 char *addr, unsigned long count)
874 /* The cast to a user pointer is valid due to the set_fs() */
875 result = vfs_read(file, (void __user *)addr, count, &pos);
880 EXPORT_SYMBOL(kernel_read);
882 int kernel_read_file(struct file *file, void **buf, loff_t *size,
883 loff_t max_size, enum kernel_read_file_id id)
889 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
892 ret = security_kernel_read_file(file, id);
896 ret = deny_write_access(file);
900 i_size = i_size_read(file_inode(file));
901 if (max_size > 0 && i_size > max_size) {
910 if (id != READING_FIRMWARE_PREALLOC_BUFFER)
911 *buf = vmalloc(i_size);
918 while (pos < i_size) {
919 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
936 ret = security_kernel_post_read_file(file, *buf, i_size, id);
942 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
949 allow_write_access(file);
952 EXPORT_SYMBOL_GPL(kernel_read_file);
954 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
955 loff_t max_size, enum kernel_read_file_id id)
963 file = filp_open(path, O_RDONLY, 0);
965 return PTR_ERR(file);
967 ret = kernel_read_file(file, buf, size, max_size, id);
971 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
973 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
974 enum kernel_read_file_id id)
976 struct fd f = fdget(fd);
982 ret = kernel_read_file(f.file, buf, size, max_size, id);
987 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
989 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
991 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
993 flush_icache_range(addr, addr + len);
996 EXPORT_SYMBOL(read_code);
998 static int exec_mmap(struct mm_struct *mm)
1000 struct task_struct *tsk;
1001 struct mm_struct *old_mm, *active_mm;
1003 /* Notify parent that we're no longer interested in the old VM */
1005 old_mm = current->mm;
1006 mm_release(tsk, old_mm);
1009 sync_mm_rss(old_mm);
1011 * Make sure that if there is a core dump in progress
1012 * for the old mm, we get out and die instead of going
1013 * through with the exec. We must hold mmap_sem around
1014 * checking core_state and changing tsk->mm.
1016 down_read(&old_mm->mmap_sem);
1017 if (unlikely(old_mm->core_state)) {
1018 up_read(&old_mm->mmap_sem);
1023 active_mm = tsk->active_mm;
1025 tsk->active_mm = mm;
1026 activate_mm(active_mm, mm);
1027 tsk->mm->vmacache_seqnum = 0;
1028 vmacache_flush(tsk);
1031 up_read(&old_mm->mmap_sem);
1032 BUG_ON(active_mm != old_mm);
1033 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1034 mm_update_next_owner(old_mm);
1043 * This function makes sure the current process has its own signal table,
1044 * so that flush_signal_handlers can later reset the handlers without
1045 * disturbing other processes. (Other processes might share the signal
1046 * table via the CLONE_SIGHAND option to clone().)
1048 static int de_thread(struct task_struct *tsk)
1050 struct signal_struct *sig = tsk->signal;
1051 struct sighand_struct *oldsighand = tsk->sighand;
1052 spinlock_t *lock = &oldsighand->siglock;
1054 if (thread_group_empty(tsk))
1055 goto no_thread_group;
1058 * Kill all other threads in the thread group.
1060 spin_lock_irq(lock);
1061 if (signal_group_exit(sig)) {
1063 * Another group action in progress, just
1064 * return so that the signal is processed.
1066 spin_unlock_irq(lock);
1070 sig->group_exit_task = tsk;
1071 sig->notify_count = zap_other_threads(tsk);
1072 if (!thread_group_leader(tsk))
1073 sig->notify_count--;
1075 while (sig->notify_count) {
1076 __set_current_state(TASK_KILLABLE);
1077 spin_unlock_irq(lock);
1079 if (unlikely(__fatal_signal_pending(tsk)))
1081 spin_lock_irq(lock);
1083 spin_unlock_irq(lock);
1086 * At this point all other threads have exited, all we have to
1087 * do is to wait for the thread group leader to become inactive,
1088 * and to assume its PID:
1090 if (!thread_group_leader(tsk)) {
1091 struct task_struct *leader = tsk->group_leader;
1094 cgroup_threadgroup_change_begin(tsk);
1095 write_lock_irq(&tasklist_lock);
1097 * Do this under tasklist_lock to ensure that
1098 * exit_notify() can't miss ->group_exit_task
1100 sig->notify_count = -1;
1101 if (likely(leader->exit_state))
1103 __set_current_state(TASK_KILLABLE);
1104 write_unlock_irq(&tasklist_lock);
1105 cgroup_threadgroup_change_end(tsk);
1107 if (unlikely(__fatal_signal_pending(tsk)))
1112 * The only record we have of the real-time age of a
1113 * process, regardless of execs it's done, is start_time.
1114 * All the past CPU time is accumulated in signal_struct
1115 * from sister threads now dead. But in this non-leader
1116 * exec, nothing survives from the original leader thread,
1117 * whose birth marks the true age of this process now.
1118 * When we take on its identity by switching to its PID, we
1119 * also take its birthdate (always earlier than our own).
1121 tsk->start_time = leader->start_time;
1122 tsk->real_start_time = leader->real_start_time;
1124 BUG_ON(!same_thread_group(leader, tsk));
1125 BUG_ON(has_group_leader_pid(tsk));
1127 * An exec() starts a new thread group with the
1128 * TGID of the previous thread group. Rehash the
1129 * two threads with a switched PID, and release
1130 * the former thread group leader:
1133 /* Become a process group leader with the old leader's pid.
1134 * The old leader becomes a thread of the this thread group.
1135 * Note: The old leader also uses this pid until release_task
1136 * is called. Odd but simple and correct.
1138 tsk->pid = leader->pid;
1139 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1140 transfer_pid(leader, tsk, PIDTYPE_PGID);
1141 transfer_pid(leader, tsk, PIDTYPE_SID);
1143 list_replace_rcu(&leader->tasks, &tsk->tasks);
1144 list_replace_init(&leader->sibling, &tsk->sibling);
1146 tsk->group_leader = tsk;
1147 leader->group_leader = tsk;
1149 tsk->exit_signal = SIGCHLD;
1150 leader->exit_signal = -1;
1152 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1153 leader->exit_state = EXIT_DEAD;
1156 * We are going to release_task()->ptrace_unlink() silently,
1157 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1158 * the tracer wont't block again waiting for this thread.
1160 if (unlikely(leader->ptrace))
1161 __wake_up_parent(leader, leader->parent);
1162 write_unlock_irq(&tasklist_lock);
1163 cgroup_threadgroup_change_end(tsk);
1165 release_task(leader);
1168 sig->group_exit_task = NULL;
1169 sig->notify_count = 0;
1172 /* we have changed execution domain */
1173 tsk->exit_signal = SIGCHLD;
1175 #ifdef CONFIG_POSIX_TIMERS
1177 flush_itimer_signals();
1180 if (atomic_read(&oldsighand->count) != 1) {
1181 struct sighand_struct *newsighand;
1183 * This ->sighand is shared with the CLONE_SIGHAND
1184 * but not CLONE_THREAD task, switch to the new one.
1186 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1190 atomic_set(&newsighand->count, 1);
1191 memcpy(newsighand->action, oldsighand->action,
1192 sizeof(newsighand->action));
1194 write_lock_irq(&tasklist_lock);
1195 spin_lock(&oldsighand->siglock);
1196 rcu_assign_pointer(tsk->sighand, newsighand);
1197 spin_unlock(&oldsighand->siglock);
1198 write_unlock_irq(&tasklist_lock);
1200 __cleanup_sighand(oldsighand);
1203 BUG_ON(!thread_group_leader(tsk));
1207 /* protects against exit_notify() and __exit_signal() */
1208 read_lock(&tasklist_lock);
1209 sig->group_exit_task = NULL;
1210 sig->notify_count = 0;
1211 read_unlock(&tasklist_lock);
1215 char *get_task_comm(char *buf, struct task_struct *tsk)
1217 /* buf must be at least sizeof(tsk->comm) in size */
1219 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1223 EXPORT_SYMBOL_GPL(get_task_comm);
1226 * These functions flushes out all traces of the currently running executable
1227 * so that a new one can be started
1230 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1233 trace_task_rename(tsk, buf);
1234 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1236 perf_event_comm(tsk, exec);
1239 int flush_old_exec(struct linux_binprm * bprm)
1244 * Make sure we have a private signal table and that
1245 * we are unassociated from the previous thread group.
1247 retval = de_thread(current);
1252 * Must be called _before_ exec_mmap() as bprm->mm is
1253 * not visibile until then. This also enables the update
1256 set_mm_exe_file(bprm->mm, bprm->file);
1259 * Release all of the old mmap stuff
1261 acct_arg_size(bprm, 0);
1262 retval = exec_mmap(bprm->mm);
1266 bprm->mm = NULL; /* We're using it now */
1269 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1270 PF_NOFREEZE | PF_NO_SETAFFINITY);
1272 current->personality &= ~bprm->per_clear;
1275 * We have to apply CLOEXEC before we change whether the process is
1276 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1277 * trying to access the should-be-closed file descriptors of a process
1278 * undergoing exec(2).
1280 do_close_on_exec(current->files);
1286 EXPORT_SYMBOL(flush_old_exec);
1288 void would_dump(struct linux_binprm *bprm, struct file *file)
1290 struct inode *inode = file_inode(file);
1291 if (inode_permission(inode, MAY_READ) < 0) {
1292 struct user_namespace *old, *user_ns;
1293 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1295 /* Ensure mm->user_ns contains the executable */
1296 user_ns = old = bprm->mm->user_ns;
1297 while ((user_ns != &init_user_ns) &&
1298 !privileged_wrt_inode_uidgid(user_ns, inode))
1299 user_ns = user_ns->parent;
1301 if (old != user_ns) {
1302 bprm->mm->user_ns = get_user_ns(user_ns);
1307 EXPORT_SYMBOL(would_dump);
1309 void setup_new_exec(struct linux_binprm * bprm)
1311 arch_pick_mmap_layout(current->mm);
1313 /* This is the point of no return */
1314 current->sas_ss_sp = current->sas_ss_size = 0;
1316 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1317 set_dumpable(current->mm, SUID_DUMP_USER);
1319 set_dumpable(current->mm, suid_dumpable);
1322 __set_task_comm(current, kbasename(bprm->filename), true);
1324 /* Set the new mm task size. We have to do that late because it may
1325 * depend on TIF_32BIT which is only updated in flush_thread() on
1326 * some architectures like powerpc
1328 current->mm->task_size = TASK_SIZE;
1330 /* install the new credentials */
1331 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1332 !gid_eq(bprm->cred->gid, current_egid())) {
1333 current->pdeath_signal = 0;
1335 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1336 set_dumpable(current->mm, suid_dumpable);
1339 /* An exec changes our domain. We are no longer part of the thread
1341 current->self_exec_id++;
1342 flush_signal_handlers(current, 0);
1344 EXPORT_SYMBOL(setup_new_exec);
1347 * Prepare credentials and lock ->cred_guard_mutex.
1348 * install_exec_creds() commits the new creds and drops the lock.
1349 * Or, if exec fails before, free_bprm() should release ->cred and
1352 int prepare_bprm_creds(struct linux_binprm *bprm)
1354 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1355 return -ERESTARTNOINTR;
1357 bprm->cred = prepare_exec_creds();
1358 if (likely(bprm->cred))
1361 mutex_unlock(¤t->signal->cred_guard_mutex);
1365 static void free_bprm(struct linux_binprm *bprm)
1367 free_arg_pages(bprm);
1369 mutex_unlock(¤t->signal->cred_guard_mutex);
1370 abort_creds(bprm->cred);
1373 allow_write_access(bprm->file);
1376 /* If a binfmt changed the interp, free it. */
1377 if (bprm->interp != bprm->filename)
1378 kfree(bprm->interp);
1382 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1384 /* If a binfmt changed the interp, free it first. */
1385 if (bprm->interp != bprm->filename)
1386 kfree(bprm->interp);
1387 bprm->interp = kstrdup(interp, GFP_KERNEL);
1392 EXPORT_SYMBOL(bprm_change_interp);
1395 * install the new credentials for this executable
1397 void install_exec_creds(struct linux_binprm *bprm)
1399 security_bprm_committing_creds(bprm);
1401 commit_creds(bprm->cred);
1405 * Disable monitoring for regular users
1406 * when executing setuid binaries. Must
1407 * wait until new credentials are committed
1408 * by commit_creds() above
1410 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1411 perf_event_exit_task(current);
1413 * cred_guard_mutex must be held at least to this point to prevent
1414 * ptrace_attach() from altering our determination of the task's
1415 * credentials; any time after this it may be unlocked.
1417 security_bprm_committed_creds(bprm);
1418 mutex_unlock(¤t->signal->cred_guard_mutex);
1420 EXPORT_SYMBOL(install_exec_creds);
1423 * determine how safe it is to execute the proposed program
1424 * - the caller must hold ->cred_guard_mutex to protect against
1425 * PTRACE_ATTACH or seccomp thread-sync
1427 static void check_unsafe_exec(struct linux_binprm *bprm)
1429 struct task_struct *p = current, *t;
1433 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1436 * This isn't strictly necessary, but it makes it harder for LSMs to
1439 if (task_no_new_privs(current))
1440 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1444 spin_lock(&p->fs->lock);
1446 while_each_thread(p, t) {
1452 if (p->fs->users > n_fs)
1453 bprm->unsafe |= LSM_UNSAFE_SHARE;
1456 spin_unlock(&p->fs->lock);
1459 static void bprm_fill_uid(struct linux_binprm *bprm)
1461 struct inode *inode;
1467 * Since this can be called multiple times (via prepare_binprm),
1468 * we must clear any previous work done when setting set[ug]id
1469 * bits from any earlier bprm->file uses (for example when run
1470 * first for a setuid script then again for its interpreter).
1472 bprm->cred->euid = current_euid();
1473 bprm->cred->egid = current_egid();
1475 if (!mnt_may_suid(bprm->file->f_path.mnt))
1478 if (task_no_new_privs(current))
1481 inode = bprm->file->f_path.dentry->d_inode;
1482 mode = READ_ONCE(inode->i_mode);
1483 if (!(mode & (S_ISUID|S_ISGID)))
1486 /* Be careful if suid/sgid is set */
1489 /* reload atomically mode/uid/gid now that lock held */
1490 mode = inode->i_mode;
1493 inode_unlock(inode);
1495 /* We ignore suid/sgid if there are no mappings for them in the ns */
1496 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1497 !kgid_has_mapping(bprm->cred->user_ns, gid))
1500 if (mode & S_ISUID) {
1501 bprm->per_clear |= PER_CLEAR_ON_SETID;
1502 bprm->cred->euid = uid;
1505 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1506 bprm->per_clear |= PER_CLEAR_ON_SETID;
1507 bprm->cred->egid = gid;
1512 * Fill the binprm structure from the inode.
1513 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1515 * This may be called multiple times for binary chains (scripts for example).
1517 int prepare_binprm(struct linux_binprm *bprm)
1521 bprm_fill_uid(bprm);
1523 /* fill in binprm security blob */
1524 retval = security_bprm_set_creds(bprm);
1527 bprm->cred_prepared = 1;
1529 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1530 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1533 EXPORT_SYMBOL(prepare_binprm);
1536 * Arguments are '\0' separated strings found at the location bprm->p
1537 * points to; chop off the first by relocating brpm->p to right after
1538 * the first '\0' encountered.
1540 int remove_arg_zero(struct linux_binprm *bprm)
1543 unsigned long offset;
1551 offset = bprm->p & ~PAGE_MASK;
1552 page = get_arg_page(bprm, bprm->p, 0);
1557 kaddr = kmap_atomic(page);
1559 for (; offset < PAGE_SIZE && kaddr[offset];
1560 offset++, bprm->p++)
1563 kunmap_atomic(kaddr);
1565 } while (offset == PAGE_SIZE);
1574 EXPORT_SYMBOL(remove_arg_zero);
1576 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1578 * cycle the list of binary formats handler, until one recognizes the image
1580 int search_binary_handler(struct linux_binprm *bprm)
1582 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1583 struct linux_binfmt *fmt;
1586 /* This allows 4 levels of binfmt rewrites before failing hard. */
1587 if (bprm->recursion_depth > 5)
1590 retval = security_bprm_check(bprm);
1596 read_lock(&binfmt_lock);
1597 list_for_each_entry(fmt, &formats, lh) {
1598 if (!try_module_get(fmt->module))
1600 read_unlock(&binfmt_lock);
1601 bprm->recursion_depth++;
1602 retval = fmt->load_binary(bprm);
1603 read_lock(&binfmt_lock);
1605 bprm->recursion_depth--;
1606 if (retval < 0 && !bprm->mm) {
1607 /* we got to flush_old_exec() and failed after it */
1608 read_unlock(&binfmt_lock);
1609 force_sigsegv(SIGSEGV, current);
1612 if (retval != -ENOEXEC || !bprm->file) {
1613 read_unlock(&binfmt_lock);
1617 read_unlock(&binfmt_lock);
1620 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1621 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1623 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1631 EXPORT_SYMBOL(search_binary_handler);
1633 static int exec_binprm(struct linux_binprm *bprm)
1635 pid_t old_pid, old_vpid;
1638 /* Need to fetch pid before load_binary changes it */
1639 old_pid = current->pid;
1641 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1644 ret = search_binary_handler(bprm);
1647 trace_sched_process_exec(current, old_pid, bprm);
1648 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1649 proc_exec_connector(current);
1656 * sys_execve() executes a new program.
1658 static int do_execveat_common(int fd, struct filename *filename,
1659 struct user_arg_ptr argv,
1660 struct user_arg_ptr envp,
1663 char *pathbuf = NULL;
1664 struct linux_binprm *bprm;
1666 struct files_struct *displaced;
1669 if (IS_ERR(filename))
1670 return PTR_ERR(filename);
1673 * We move the actual failure in case of RLIMIT_NPROC excess from
1674 * set*uid() to execve() because too many poorly written programs
1675 * don't check setuid() return code. Here we additionally recheck
1676 * whether NPROC limit is still exceeded.
1678 if ((current->flags & PF_NPROC_EXCEEDED) &&
1679 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1684 /* We're below the limit (still or again), so we don't want to make
1685 * further execve() calls fail. */
1686 current->flags &= ~PF_NPROC_EXCEEDED;
1688 retval = unshare_files(&displaced);
1693 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1697 retval = prepare_bprm_creds(bprm);
1701 check_unsafe_exec(bprm);
1702 current->in_execve = 1;
1704 file = do_open_execat(fd, filename, flags);
1705 retval = PTR_ERR(file);
1712 if (fd == AT_FDCWD || filename->name[0] == '/') {
1713 bprm->filename = filename->name;
1715 if (filename->name[0] == '\0')
1716 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1718 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1719 fd, filename->name);
1725 * Record that a name derived from an O_CLOEXEC fd will be
1726 * inaccessible after exec. Relies on having exclusive access to
1727 * current->files (due to unshare_files above).
1729 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1730 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1731 bprm->filename = pathbuf;
1733 bprm->interp = bprm->filename;
1735 retval = bprm_mm_init(bprm);
1739 bprm->argc = count(argv, MAX_ARG_STRINGS);
1740 if ((retval = bprm->argc) < 0)
1743 bprm->envc = count(envp, MAX_ARG_STRINGS);
1744 if ((retval = bprm->envc) < 0)
1747 retval = prepare_binprm(bprm);
1751 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1755 bprm->exec = bprm->p;
1756 retval = copy_strings(bprm->envc, envp, bprm);
1760 retval = copy_strings(bprm->argc, argv, bprm);
1764 would_dump(bprm, bprm->file);
1766 retval = exec_binprm(bprm);
1770 /* execve succeeded */
1771 current->fs->in_exec = 0;
1772 current->in_execve = 0;
1773 acct_update_integrals(current);
1774 task_numa_free(current);
1779 put_files_struct(displaced);
1784 acct_arg_size(bprm, 0);
1789 current->fs->in_exec = 0;
1790 current->in_execve = 0;
1798 reset_files_struct(displaced);
1804 int do_execve(struct filename *filename,
1805 const char __user *const __user *__argv,
1806 const char __user *const __user *__envp)
1808 struct user_arg_ptr argv = { .ptr.native = __argv };
1809 struct user_arg_ptr envp = { .ptr.native = __envp };
1810 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1813 int do_execveat(int fd, struct filename *filename,
1814 const char __user *const __user *__argv,
1815 const char __user *const __user *__envp,
1818 struct user_arg_ptr argv = { .ptr.native = __argv };
1819 struct user_arg_ptr envp = { .ptr.native = __envp };
1821 return do_execveat_common(fd, filename, argv, envp, flags);
1824 #ifdef CONFIG_COMPAT
1825 static int compat_do_execve(struct filename *filename,
1826 const compat_uptr_t __user *__argv,
1827 const compat_uptr_t __user *__envp)
1829 struct user_arg_ptr argv = {
1831 .ptr.compat = __argv,
1833 struct user_arg_ptr envp = {
1835 .ptr.compat = __envp,
1837 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1840 static int compat_do_execveat(int fd, struct filename *filename,
1841 const compat_uptr_t __user *__argv,
1842 const compat_uptr_t __user *__envp,
1845 struct user_arg_ptr argv = {
1847 .ptr.compat = __argv,
1849 struct user_arg_ptr envp = {
1851 .ptr.compat = __envp,
1853 return do_execveat_common(fd, filename, argv, envp, flags);
1857 void set_binfmt(struct linux_binfmt *new)
1859 struct mm_struct *mm = current->mm;
1862 module_put(mm->binfmt->module);
1866 __module_get(new->module);
1868 EXPORT_SYMBOL(set_binfmt);
1871 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1873 void set_dumpable(struct mm_struct *mm, int value)
1875 unsigned long old, new;
1877 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1881 old = ACCESS_ONCE(mm->flags);
1882 new = (old & ~MMF_DUMPABLE_MASK) | value;
1883 } while (cmpxchg(&mm->flags, old, new) != old);
1886 SYSCALL_DEFINE3(execve,
1887 const char __user *, filename,
1888 const char __user *const __user *, argv,
1889 const char __user *const __user *, envp)
1891 return do_execve(getname(filename), argv, envp);
1894 SYSCALL_DEFINE5(execveat,
1895 int, fd, const char __user *, filename,
1896 const char __user *const __user *, argv,
1897 const char __user *const __user *, envp,
1900 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1902 return do_execveat(fd,
1903 getname_flags(filename, lookup_flags, NULL),
1907 #ifdef CONFIG_COMPAT
1908 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1909 const compat_uptr_t __user *, argv,
1910 const compat_uptr_t __user *, envp)
1912 return compat_do_execve(getname(filename), argv, envp);
1915 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1916 const char __user *, filename,
1917 const compat_uptr_t __user *, argv,
1918 const compat_uptr_t __user *, envp,
1921 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1923 return compat_do_execveat(fd,
1924 getname_flags(filename, lookup_flags, NULL),