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/pagemap.h>
36 #include <linux/perf_event.h>
37 #include <linux/highmem.h>
38 #include <linux/spinlock.h>
39 #include <linux/key.h>
40 #include <linux/personality.h>
41 #include <linux/binfmts.h>
42 #include <linux/utsname.h>
43 #include <linux/pid_namespace.h>
44 #include <linux/module.h>
45 #include <linux/namei.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
53 #include <linux/kmod.h>
54 #include <linux/fsnotify.h>
55 #include <linux/fs_struct.h>
56 #include <linux/pipe_fs_i.h>
57 #include <linux/oom.h>
58 #include <linux/compat.h>
59 #include <linux/vmalloc.h>
61 #include <asm/uaccess.h>
62 #include <asm/mmu_context.h>
65 #include <trace/events/task.h>
68 #include <trace/events/sched.h>
70 int suid_dumpable = 0;
72 static LIST_HEAD(formats);
73 static DEFINE_RWLOCK(binfmt_lock);
75 void __register_binfmt(struct linux_binfmt * fmt, int insert)
78 if (WARN_ON(!fmt->load_binary))
80 write_lock(&binfmt_lock);
81 insert ? list_add(&fmt->lh, &formats) :
82 list_add_tail(&fmt->lh, &formats);
83 write_unlock(&binfmt_lock);
86 EXPORT_SYMBOL(__register_binfmt);
88 void unregister_binfmt(struct linux_binfmt * fmt)
90 write_lock(&binfmt_lock);
92 write_unlock(&binfmt_lock);
95 EXPORT_SYMBOL(unregister_binfmt);
97 static inline void put_binfmt(struct linux_binfmt * fmt)
99 module_put(fmt->module);
102 bool path_noexec(const struct path *path)
104 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
105 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
110 * Note that a shared library must be both readable and executable due to
113 * Also note that we take the address to load from from the file itself.
115 SYSCALL_DEFINE1(uselib, const char __user *, library)
117 struct linux_binfmt *fmt;
119 struct filename *tmp = getname(library);
120 int error = PTR_ERR(tmp);
121 static const struct open_flags uselib_flags = {
122 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
123 .acc_mode = MAY_READ | MAY_EXEC,
124 .intent = LOOKUP_OPEN,
125 .lookup_flags = LOOKUP_FOLLOW,
131 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
133 error = PTR_ERR(file);
138 if (!S_ISREG(file_inode(file)->i_mode))
142 if (path_noexec(&file->f_path))
149 read_lock(&binfmt_lock);
150 list_for_each_entry(fmt, &formats, lh) {
151 if (!fmt->load_shlib)
153 if (!try_module_get(fmt->module))
155 read_unlock(&binfmt_lock);
156 error = fmt->load_shlib(file);
157 read_lock(&binfmt_lock);
159 if (error != -ENOEXEC)
162 read_unlock(&binfmt_lock);
168 #endif /* #ifdef CONFIG_USELIB */
172 * The nascent bprm->mm is not visible until exec_mmap() but it can
173 * use a lot of memory, account these pages in current->mm temporary
174 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
175 * change the counter back via acct_arg_size(0).
177 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
179 struct mm_struct *mm = current->mm;
180 long diff = (long)(pages - bprm->vma_pages);
185 bprm->vma_pages = pages;
186 add_mm_counter(mm, MM_ANONPAGES, diff);
189 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
195 #ifdef CONFIG_STACK_GROWSUP
197 ret = expand_downwards(bprm->vma, pos);
202 ret = get_user_pages(current, bprm->mm, pos,
203 1, write, 1, &page, NULL);
208 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
211 acct_arg_size(bprm, size / PAGE_SIZE);
214 * We've historically supported up to 32 pages (ARG_MAX)
215 * of argument strings even with small stacks
221 * Limit to 1/4-th the stack size for the argv+env strings.
223 * - the remaining binfmt code will not run out of stack space,
224 * - the program will have a reasonable amount of stack left
227 rlim = current->signal->rlim;
228 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
237 static void put_arg_page(struct page *page)
242 static void free_arg_page(struct linux_binprm *bprm, int i)
246 static void free_arg_pages(struct linux_binprm *bprm)
250 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
253 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
256 static int __bprm_mm_init(struct linux_binprm *bprm)
259 struct vm_area_struct *vma = NULL;
260 struct mm_struct *mm = bprm->mm;
262 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
266 down_write(&mm->mmap_sem);
270 * Place the stack at the largest stack address the architecture
271 * supports. Later, we'll move this to an appropriate place. We don't
272 * use STACK_TOP because that can depend on attributes which aren't
275 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
276 vma->vm_end = STACK_TOP_MAX;
277 vma->vm_start = vma->vm_end - PAGE_SIZE;
278 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
279 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
280 INIT_LIST_HEAD(&vma->anon_vma_chain);
282 err = insert_vm_struct(mm, vma);
286 mm->stack_vm = mm->total_vm = 1;
287 arch_bprm_mm_init(mm, vma);
288 up_write(&mm->mmap_sem);
289 bprm->p = vma->vm_end - sizeof(void *);
292 up_write(&mm->mmap_sem);
294 kmem_cache_free(vm_area_cachep, vma);
298 static bool valid_arg_len(struct linux_binprm *bprm, long len)
300 return len <= MAX_ARG_STRLEN;
305 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
309 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
314 page = bprm->page[pos / PAGE_SIZE];
315 if (!page && write) {
316 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
319 bprm->page[pos / PAGE_SIZE] = page;
325 static void put_arg_page(struct page *page)
329 static void free_arg_page(struct linux_binprm *bprm, int i)
332 __free_page(bprm->page[i]);
333 bprm->page[i] = NULL;
337 static void free_arg_pages(struct linux_binprm *bprm)
341 for (i = 0; i < MAX_ARG_PAGES; i++)
342 free_arg_page(bprm, i);
345 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
350 static int __bprm_mm_init(struct linux_binprm *bprm)
352 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
356 static bool valid_arg_len(struct linux_binprm *bprm, long len)
358 return len <= bprm->p;
361 #endif /* CONFIG_MMU */
364 * Create a new mm_struct and populate it with a temporary stack
365 * vm_area_struct. We don't have enough context at this point to set the stack
366 * flags, permissions, and offset, so we use temporary values. We'll update
367 * them later in setup_arg_pages().
369 static int bprm_mm_init(struct linux_binprm *bprm)
372 struct mm_struct *mm = NULL;
374 bprm->mm = mm = mm_alloc();
379 err = __bprm_mm_init(bprm);
394 struct user_arg_ptr {
399 const char __user *const __user *native;
401 const compat_uptr_t __user *compat;
406 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
408 const char __user *native;
411 if (unlikely(argv.is_compat)) {
412 compat_uptr_t compat;
414 if (get_user(compat, argv.ptr.compat + nr))
415 return ERR_PTR(-EFAULT);
417 return compat_ptr(compat);
421 if (get_user(native, argv.ptr.native + nr))
422 return ERR_PTR(-EFAULT);
428 * count() counts the number of strings in array ARGV.
430 static int count(struct user_arg_ptr argv, int max)
434 if (argv.ptr.native != NULL) {
436 const char __user *p = get_user_arg_ptr(argv, i);
448 if (fatal_signal_pending(current))
449 return -ERESTARTNOHAND;
457 * 'copy_strings()' copies argument/environment strings from the old
458 * processes's memory to the new process's stack. The call to get_user_pages()
459 * ensures the destination page is created and not swapped out.
461 static int copy_strings(int argc, struct user_arg_ptr argv,
462 struct linux_binprm *bprm)
464 struct page *kmapped_page = NULL;
466 unsigned long kpos = 0;
470 const char __user *str;
475 str = get_user_arg_ptr(argv, argc);
479 len = strnlen_user(str, MAX_ARG_STRLEN);
484 if (!valid_arg_len(bprm, len))
487 /* We're going to work our way backwords. */
493 int offset, bytes_to_copy;
495 if (fatal_signal_pending(current)) {
496 ret = -ERESTARTNOHAND;
501 offset = pos % PAGE_SIZE;
505 bytes_to_copy = offset;
506 if (bytes_to_copy > len)
509 offset -= bytes_to_copy;
510 pos -= bytes_to_copy;
511 str -= bytes_to_copy;
512 len -= bytes_to_copy;
514 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
517 page = get_arg_page(bprm, pos, 1);
524 flush_kernel_dcache_page(kmapped_page);
525 kunmap(kmapped_page);
526 put_arg_page(kmapped_page);
529 kaddr = kmap(kmapped_page);
530 kpos = pos & PAGE_MASK;
531 flush_arg_page(bprm, kpos, kmapped_page);
533 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
542 flush_kernel_dcache_page(kmapped_page);
543 kunmap(kmapped_page);
544 put_arg_page(kmapped_page);
550 * Like copy_strings, but get argv and its values from kernel memory.
552 int copy_strings_kernel(int argc, const char *const *__argv,
553 struct linux_binprm *bprm)
556 mm_segment_t oldfs = get_fs();
557 struct user_arg_ptr argv = {
558 .ptr.native = (const char __user *const __user *)__argv,
562 r = copy_strings(argc, argv, bprm);
567 EXPORT_SYMBOL(copy_strings_kernel);
572 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
573 * the binfmt code determines where the new stack should reside, we shift it to
574 * its final location. The process proceeds as follows:
576 * 1) Use shift to calculate the new vma endpoints.
577 * 2) Extend vma to cover both the old and new ranges. This ensures the
578 * arguments passed to subsequent functions are consistent.
579 * 3) Move vma's page tables to the new range.
580 * 4) Free up any cleared pgd range.
581 * 5) Shrink the vma to cover only the new range.
583 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
585 struct mm_struct *mm = vma->vm_mm;
586 unsigned long old_start = vma->vm_start;
587 unsigned long old_end = vma->vm_end;
588 unsigned long length = old_end - old_start;
589 unsigned long new_start = old_start - shift;
590 unsigned long new_end = old_end - shift;
591 struct mmu_gather tlb;
593 BUG_ON(new_start > new_end);
596 * ensure there are no vmas between where we want to go
599 if (vma != find_vma(mm, new_start))
603 * cover the whole range: [new_start, old_end)
605 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
609 * move the page tables downwards, on failure we rely on
610 * process cleanup to remove whatever mess we made.
612 if (length != move_page_tables(vma, old_start,
613 vma, new_start, length, false))
617 tlb_gather_mmu(&tlb, mm, old_start, old_end);
618 if (new_end > old_start) {
620 * when the old and new regions overlap clear from new_end.
622 free_pgd_range(&tlb, new_end, old_end, new_end,
623 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
626 * otherwise, clean from old_start; this is done to not touch
627 * the address space in [new_end, old_start) some architectures
628 * have constraints on va-space that make this illegal (IA64) -
629 * for the others its just a little faster.
631 free_pgd_range(&tlb, old_start, old_end, new_end,
632 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
634 tlb_finish_mmu(&tlb, old_start, old_end);
637 * Shrink the vma to just the new range. Always succeeds.
639 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
645 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
646 * the stack is optionally relocated, and some extra space is added.
648 int setup_arg_pages(struct linux_binprm *bprm,
649 unsigned long stack_top,
650 int executable_stack)
653 unsigned long stack_shift;
654 struct mm_struct *mm = current->mm;
655 struct vm_area_struct *vma = bprm->vma;
656 struct vm_area_struct *prev = NULL;
657 unsigned long vm_flags;
658 unsigned long stack_base;
659 unsigned long stack_size;
660 unsigned long stack_expand;
661 unsigned long rlim_stack;
663 #ifdef CONFIG_STACK_GROWSUP
664 /* Limit stack size */
665 stack_base = rlimit_max(RLIMIT_STACK);
666 if (stack_base > STACK_SIZE_MAX)
667 stack_base = STACK_SIZE_MAX;
669 /* Add space for stack randomization. */
670 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
672 /* Make sure we didn't let the argument array grow too large. */
673 if (vma->vm_end - vma->vm_start > stack_base)
676 stack_base = PAGE_ALIGN(stack_top - stack_base);
678 stack_shift = vma->vm_start - stack_base;
679 mm->arg_start = bprm->p - stack_shift;
680 bprm->p = vma->vm_end - stack_shift;
682 stack_top = arch_align_stack(stack_top);
683 stack_top = PAGE_ALIGN(stack_top);
685 if (unlikely(stack_top < mmap_min_addr) ||
686 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
689 stack_shift = vma->vm_end - stack_top;
691 bprm->p -= stack_shift;
692 mm->arg_start = bprm->p;
696 bprm->loader -= stack_shift;
697 bprm->exec -= stack_shift;
699 down_write(&mm->mmap_sem);
700 vm_flags = VM_STACK_FLAGS;
703 * Adjust stack execute permissions; explicitly enable for
704 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
705 * (arch default) otherwise.
707 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
709 else if (executable_stack == EXSTACK_DISABLE_X)
710 vm_flags &= ~VM_EXEC;
711 vm_flags |= mm->def_flags;
712 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
714 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
720 /* Move stack pages down in memory. */
722 ret = shift_arg_pages(vma, stack_shift);
727 /* mprotect_fixup is overkill to remove the temporary stack flags */
728 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
730 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
731 stack_size = vma->vm_end - vma->vm_start;
733 * Align this down to a page boundary as expand_stack
736 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
737 #ifdef CONFIG_STACK_GROWSUP
738 if (stack_size + stack_expand > rlim_stack)
739 stack_base = vma->vm_start + rlim_stack;
741 stack_base = vma->vm_end + stack_expand;
743 if (stack_size + stack_expand > rlim_stack)
744 stack_base = vma->vm_end - rlim_stack;
746 stack_base = vma->vm_start - stack_expand;
748 current->mm->start_stack = bprm->p;
749 ret = expand_stack(vma, stack_base);
754 up_write(&mm->mmap_sem);
757 EXPORT_SYMBOL(setup_arg_pages);
759 #endif /* CONFIG_MMU */
761 static struct file *do_open_execat(int fd, struct filename *name, int flags)
765 struct open_flags open_exec_flags = {
766 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
767 .acc_mode = MAY_EXEC,
768 .intent = LOOKUP_OPEN,
769 .lookup_flags = LOOKUP_FOLLOW,
772 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
773 return ERR_PTR(-EINVAL);
774 if (flags & AT_SYMLINK_NOFOLLOW)
775 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
776 if (flags & AT_EMPTY_PATH)
777 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
779 file = do_filp_open(fd, name, &open_exec_flags);
784 if (!S_ISREG(file_inode(file)->i_mode))
787 if (path_noexec(&file->f_path))
790 err = deny_write_access(file);
794 if (name->name[0] != '\0')
805 struct file *open_exec(const char *name)
807 struct filename *filename = getname_kernel(name);
808 struct file *f = ERR_CAST(filename);
810 if (!IS_ERR(filename)) {
811 f = do_open_execat(AT_FDCWD, filename, 0);
816 EXPORT_SYMBOL(open_exec);
818 int kernel_read(struct file *file, loff_t offset,
819 char *addr, unsigned long count)
827 /* The cast to a user pointer is valid due to the set_fs() */
828 result = vfs_read(file, (void __user *)addr, count, &pos);
833 EXPORT_SYMBOL(kernel_read);
835 int kernel_read_file(struct file *file, void **buf, loff_t *size,
836 loff_t max_size, enum kernel_read_file_id id)
842 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
845 i_size = i_size_read(file_inode(file));
846 if (max_size > 0 && i_size > max_size)
851 *buf = vmalloc(i_size);
856 while (pos < i_size) {
857 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
874 ret = security_kernel_post_read_file(file, *buf, i_size, id);
885 EXPORT_SYMBOL_GPL(kernel_read_file);
887 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
888 loff_t max_size, enum kernel_read_file_id id)
896 file = filp_open(path, O_RDONLY, 0);
898 return PTR_ERR(file);
900 ret = kernel_read_file(file, buf, size, max_size, id);
904 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
906 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
908 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
910 flush_icache_range(addr, addr + len);
913 EXPORT_SYMBOL(read_code);
915 static int exec_mmap(struct mm_struct *mm)
917 struct task_struct *tsk;
918 struct mm_struct *old_mm, *active_mm;
920 /* Notify parent that we're no longer interested in the old VM */
922 old_mm = current->mm;
923 mm_release(tsk, old_mm);
928 * Make sure that if there is a core dump in progress
929 * for the old mm, we get out and die instead of going
930 * through with the exec. We must hold mmap_sem around
931 * checking core_state and changing tsk->mm.
933 down_read(&old_mm->mmap_sem);
934 if (unlikely(old_mm->core_state)) {
935 up_read(&old_mm->mmap_sem);
940 active_mm = tsk->active_mm;
943 activate_mm(active_mm, mm);
944 tsk->mm->vmacache_seqnum = 0;
948 up_read(&old_mm->mmap_sem);
949 BUG_ON(active_mm != old_mm);
950 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
951 mm_update_next_owner(old_mm);
960 * This function makes sure the current process has its own signal table,
961 * so that flush_signal_handlers can later reset the handlers without
962 * disturbing other processes. (Other processes might share the signal
963 * table via the CLONE_SIGHAND option to clone().)
965 static int de_thread(struct task_struct *tsk)
967 struct signal_struct *sig = tsk->signal;
968 struct sighand_struct *oldsighand = tsk->sighand;
969 spinlock_t *lock = &oldsighand->siglock;
971 if (thread_group_empty(tsk))
972 goto no_thread_group;
975 * Kill all other threads in the thread group.
978 if (signal_group_exit(sig)) {
980 * Another group action in progress, just
981 * return so that the signal is processed.
983 spin_unlock_irq(lock);
987 sig->group_exit_task = tsk;
988 sig->notify_count = zap_other_threads(tsk);
989 if (!thread_group_leader(tsk))
992 while (sig->notify_count) {
993 __set_current_state(TASK_KILLABLE);
994 spin_unlock_irq(lock);
996 if (unlikely(__fatal_signal_pending(tsk)))
1000 spin_unlock_irq(lock);
1003 * At this point all other threads have exited, all we have to
1004 * do is to wait for the thread group leader to become inactive,
1005 * and to assume its PID:
1007 if (!thread_group_leader(tsk)) {
1008 struct task_struct *leader = tsk->group_leader;
1011 threadgroup_change_begin(tsk);
1012 write_lock_irq(&tasklist_lock);
1014 * Do this under tasklist_lock to ensure that
1015 * exit_notify() can't miss ->group_exit_task
1017 sig->notify_count = -1;
1018 if (likely(leader->exit_state))
1020 __set_current_state(TASK_KILLABLE);
1021 write_unlock_irq(&tasklist_lock);
1022 threadgroup_change_end(tsk);
1024 if (unlikely(__fatal_signal_pending(tsk)))
1029 * The only record we have of the real-time age of a
1030 * process, regardless of execs it's done, is start_time.
1031 * All the past CPU time is accumulated in signal_struct
1032 * from sister threads now dead. But in this non-leader
1033 * exec, nothing survives from the original leader thread,
1034 * whose birth marks the true age of this process now.
1035 * When we take on its identity by switching to its PID, we
1036 * also take its birthdate (always earlier than our own).
1038 tsk->start_time = leader->start_time;
1039 tsk->real_start_time = leader->real_start_time;
1041 BUG_ON(!same_thread_group(leader, tsk));
1042 BUG_ON(has_group_leader_pid(tsk));
1044 * An exec() starts a new thread group with the
1045 * TGID of the previous thread group. Rehash the
1046 * two threads with a switched PID, and release
1047 * the former thread group leader:
1050 /* Become a process group leader with the old leader's pid.
1051 * The old leader becomes a thread of the this thread group.
1052 * Note: The old leader also uses this pid until release_task
1053 * is called. Odd but simple and correct.
1055 tsk->pid = leader->pid;
1056 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1057 transfer_pid(leader, tsk, PIDTYPE_PGID);
1058 transfer_pid(leader, tsk, PIDTYPE_SID);
1060 list_replace_rcu(&leader->tasks, &tsk->tasks);
1061 list_replace_init(&leader->sibling, &tsk->sibling);
1063 tsk->group_leader = tsk;
1064 leader->group_leader = tsk;
1066 tsk->exit_signal = SIGCHLD;
1067 leader->exit_signal = -1;
1069 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1070 leader->exit_state = EXIT_DEAD;
1073 * We are going to release_task()->ptrace_unlink() silently,
1074 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1075 * the tracer wont't block again waiting for this thread.
1077 if (unlikely(leader->ptrace))
1078 __wake_up_parent(leader, leader->parent);
1079 write_unlock_irq(&tasklist_lock);
1080 threadgroup_change_end(tsk);
1082 release_task(leader);
1085 sig->group_exit_task = NULL;
1086 sig->notify_count = 0;
1089 /* we have changed execution domain */
1090 tsk->exit_signal = SIGCHLD;
1093 flush_itimer_signals();
1095 if (atomic_read(&oldsighand->count) != 1) {
1096 struct sighand_struct *newsighand;
1098 * This ->sighand is shared with the CLONE_SIGHAND
1099 * but not CLONE_THREAD task, switch to the new one.
1101 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1105 atomic_set(&newsighand->count, 1);
1106 memcpy(newsighand->action, oldsighand->action,
1107 sizeof(newsighand->action));
1109 write_lock_irq(&tasklist_lock);
1110 spin_lock(&oldsighand->siglock);
1111 rcu_assign_pointer(tsk->sighand, newsighand);
1112 spin_unlock(&oldsighand->siglock);
1113 write_unlock_irq(&tasklist_lock);
1115 __cleanup_sighand(oldsighand);
1118 BUG_ON(!thread_group_leader(tsk));
1122 /* protects against exit_notify() and __exit_signal() */
1123 read_lock(&tasklist_lock);
1124 sig->group_exit_task = NULL;
1125 sig->notify_count = 0;
1126 read_unlock(&tasklist_lock);
1130 char *get_task_comm(char *buf, struct task_struct *tsk)
1132 /* buf must be at least sizeof(tsk->comm) in size */
1134 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1138 EXPORT_SYMBOL_GPL(get_task_comm);
1141 * These functions flushes out all traces of the currently running executable
1142 * so that a new one can be started
1145 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1148 trace_task_rename(tsk, buf);
1149 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1151 perf_event_comm(tsk, exec);
1154 int flush_old_exec(struct linux_binprm * bprm)
1159 * Make sure we have a private signal table and that
1160 * we are unassociated from the previous thread group.
1162 retval = de_thread(current);
1167 * Must be called _before_ exec_mmap() as bprm->mm is
1168 * not visibile until then. This also enables the update
1171 set_mm_exe_file(bprm->mm, bprm->file);
1174 * Release all of the old mmap stuff
1176 acct_arg_size(bprm, 0);
1177 retval = exec_mmap(bprm->mm);
1181 bprm->mm = NULL; /* We're using it now */
1184 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1185 PF_NOFREEZE | PF_NO_SETAFFINITY);
1187 current->personality &= ~bprm->per_clear;
1194 EXPORT_SYMBOL(flush_old_exec);
1196 void would_dump(struct linux_binprm *bprm, struct file *file)
1198 if (inode_permission(file_inode(file), MAY_READ) < 0)
1199 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1201 EXPORT_SYMBOL(would_dump);
1203 void setup_new_exec(struct linux_binprm * bprm)
1205 arch_pick_mmap_layout(current->mm);
1207 /* This is the point of no return */
1208 current->sas_ss_sp = current->sas_ss_size = 0;
1210 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1211 set_dumpable(current->mm, SUID_DUMP_USER);
1213 set_dumpable(current->mm, suid_dumpable);
1216 __set_task_comm(current, kbasename(bprm->filename), true);
1218 /* Set the new mm task size. We have to do that late because it may
1219 * depend on TIF_32BIT which is only updated in flush_thread() on
1220 * some architectures like powerpc
1222 current->mm->task_size = TASK_SIZE;
1224 /* install the new credentials */
1225 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1226 !gid_eq(bprm->cred->gid, current_egid())) {
1227 current->pdeath_signal = 0;
1229 would_dump(bprm, bprm->file);
1230 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1231 set_dumpable(current->mm, suid_dumpable);
1234 /* An exec changes our domain. We are no longer part of the thread
1236 current->self_exec_id++;
1237 flush_signal_handlers(current, 0);
1238 do_close_on_exec(current->files);
1240 EXPORT_SYMBOL(setup_new_exec);
1243 * Prepare credentials and lock ->cred_guard_mutex.
1244 * install_exec_creds() commits the new creds and drops the lock.
1245 * Or, if exec fails before, free_bprm() should release ->cred and
1248 int prepare_bprm_creds(struct linux_binprm *bprm)
1250 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1251 return -ERESTARTNOINTR;
1253 bprm->cred = prepare_exec_creds();
1254 if (likely(bprm->cred))
1257 mutex_unlock(¤t->signal->cred_guard_mutex);
1261 static void free_bprm(struct linux_binprm *bprm)
1263 free_arg_pages(bprm);
1265 mutex_unlock(¤t->signal->cred_guard_mutex);
1266 abort_creds(bprm->cred);
1269 allow_write_access(bprm->file);
1272 /* If a binfmt changed the interp, free it. */
1273 if (bprm->interp != bprm->filename)
1274 kfree(bprm->interp);
1278 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1280 /* If a binfmt changed the interp, free it first. */
1281 if (bprm->interp != bprm->filename)
1282 kfree(bprm->interp);
1283 bprm->interp = kstrdup(interp, GFP_KERNEL);
1288 EXPORT_SYMBOL(bprm_change_interp);
1291 * install the new credentials for this executable
1293 void install_exec_creds(struct linux_binprm *bprm)
1295 security_bprm_committing_creds(bprm);
1297 commit_creds(bprm->cred);
1301 * Disable monitoring for regular users
1302 * when executing setuid binaries. Must
1303 * wait until new credentials are committed
1304 * by commit_creds() above
1306 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1307 perf_event_exit_task(current);
1309 * cred_guard_mutex must be held at least to this point to prevent
1310 * ptrace_attach() from altering our determination of the task's
1311 * credentials; any time after this it may be unlocked.
1313 security_bprm_committed_creds(bprm);
1314 mutex_unlock(¤t->signal->cred_guard_mutex);
1316 EXPORT_SYMBOL(install_exec_creds);
1319 * determine how safe it is to execute the proposed program
1320 * - the caller must hold ->cred_guard_mutex to protect against
1321 * PTRACE_ATTACH or seccomp thread-sync
1323 static void check_unsafe_exec(struct linux_binprm *bprm)
1325 struct task_struct *p = current, *t;
1329 if (p->ptrace & PT_PTRACE_CAP)
1330 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1332 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1336 * This isn't strictly necessary, but it makes it harder for LSMs to
1339 if (task_no_new_privs(current))
1340 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1344 spin_lock(&p->fs->lock);
1346 while_each_thread(p, t) {
1352 if (p->fs->users > n_fs)
1353 bprm->unsafe |= LSM_UNSAFE_SHARE;
1356 spin_unlock(&p->fs->lock);
1359 static void bprm_fill_uid(struct linux_binprm *bprm)
1361 struct inode *inode;
1366 /* clear any previous set[ug]id data from a previous binary */
1367 bprm->cred->euid = current_euid();
1368 bprm->cred->egid = current_egid();
1370 if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1373 if (task_no_new_privs(current))
1376 inode = file_inode(bprm->file);
1377 mode = READ_ONCE(inode->i_mode);
1378 if (!(mode & (S_ISUID|S_ISGID)))
1381 /* Be careful if suid/sgid is set */
1384 /* reload atomically mode/uid/gid now that lock held */
1385 mode = inode->i_mode;
1388 inode_unlock(inode);
1390 /* We ignore suid/sgid if there are no mappings for them in the ns */
1391 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1392 !kgid_has_mapping(bprm->cred->user_ns, gid))
1395 if (mode & S_ISUID) {
1396 bprm->per_clear |= PER_CLEAR_ON_SETID;
1397 bprm->cred->euid = uid;
1400 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1401 bprm->per_clear |= PER_CLEAR_ON_SETID;
1402 bprm->cred->egid = gid;
1407 * Fill the binprm structure from the inode.
1408 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1410 * This may be called multiple times for binary chains (scripts for example).
1412 int prepare_binprm(struct linux_binprm *bprm)
1416 bprm_fill_uid(bprm);
1418 /* fill in binprm security blob */
1419 retval = security_bprm_set_creds(bprm);
1422 bprm->cred_prepared = 1;
1424 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1425 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1428 EXPORT_SYMBOL(prepare_binprm);
1431 * Arguments are '\0' separated strings found at the location bprm->p
1432 * points to; chop off the first by relocating brpm->p to right after
1433 * the first '\0' encountered.
1435 int remove_arg_zero(struct linux_binprm *bprm)
1438 unsigned long offset;
1446 offset = bprm->p & ~PAGE_MASK;
1447 page = get_arg_page(bprm, bprm->p, 0);
1452 kaddr = kmap_atomic(page);
1454 for (; offset < PAGE_SIZE && kaddr[offset];
1455 offset++, bprm->p++)
1458 kunmap_atomic(kaddr);
1461 if (offset == PAGE_SIZE)
1462 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1463 } while (offset == PAGE_SIZE);
1472 EXPORT_SYMBOL(remove_arg_zero);
1474 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1476 * cycle the list of binary formats handler, until one recognizes the image
1478 int search_binary_handler(struct linux_binprm *bprm)
1480 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1481 struct linux_binfmt *fmt;
1484 /* This allows 4 levels of binfmt rewrites before failing hard. */
1485 if (bprm->recursion_depth > 5)
1488 retval = security_bprm_check(bprm);
1494 read_lock(&binfmt_lock);
1495 list_for_each_entry(fmt, &formats, lh) {
1496 if (!try_module_get(fmt->module))
1498 read_unlock(&binfmt_lock);
1499 bprm->recursion_depth++;
1500 retval = fmt->load_binary(bprm);
1501 read_lock(&binfmt_lock);
1503 bprm->recursion_depth--;
1504 if (retval < 0 && !bprm->mm) {
1505 /* we got to flush_old_exec() and failed after it */
1506 read_unlock(&binfmt_lock);
1507 force_sigsegv(SIGSEGV, current);
1510 if (retval != -ENOEXEC || !bprm->file) {
1511 read_unlock(&binfmt_lock);
1515 read_unlock(&binfmt_lock);
1518 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1519 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1521 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1529 EXPORT_SYMBOL(search_binary_handler);
1531 static int exec_binprm(struct linux_binprm *bprm)
1533 pid_t old_pid, old_vpid;
1536 /* Need to fetch pid before load_binary changes it */
1537 old_pid = current->pid;
1539 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1542 ret = search_binary_handler(bprm);
1545 trace_sched_process_exec(current, old_pid, bprm);
1546 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1547 proc_exec_connector(current);
1554 * sys_execve() executes a new program.
1556 static int do_execveat_common(int fd, struct filename *filename,
1557 struct user_arg_ptr argv,
1558 struct user_arg_ptr envp,
1561 char *pathbuf = NULL;
1562 struct linux_binprm *bprm;
1564 struct files_struct *displaced;
1567 if (IS_ERR(filename))
1568 return PTR_ERR(filename);
1571 * We move the actual failure in case of RLIMIT_NPROC excess from
1572 * set*uid() to execve() because too many poorly written programs
1573 * don't check setuid() return code. Here we additionally recheck
1574 * whether NPROC limit is still exceeded.
1576 if ((current->flags & PF_NPROC_EXCEEDED) &&
1577 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1582 /* We're below the limit (still or again), so we don't want to make
1583 * further execve() calls fail. */
1584 current->flags &= ~PF_NPROC_EXCEEDED;
1586 retval = unshare_files(&displaced);
1591 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1595 retval = prepare_bprm_creds(bprm);
1599 check_unsafe_exec(bprm);
1600 current->in_execve = 1;
1602 file = do_open_execat(fd, filename, flags);
1603 retval = PTR_ERR(file);
1610 if (fd == AT_FDCWD || filename->name[0] == '/') {
1611 bprm->filename = filename->name;
1613 if (filename->name[0] == '\0')
1614 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1616 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1617 fd, filename->name);
1623 * Record that a name derived from an O_CLOEXEC fd will be
1624 * inaccessible after exec. Relies on having exclusive access to
1625 * current->files (due to unshare_files above).
1627 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1628 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1629 bprm->filename = pathbuf;
1631 bprm->interp = bprm->filename;
1633 retval = bprm_mm_init(bprm);
1637 bprm->argc = count(argv, MAX_ARG_STRINGS);
1638 if ((retval = bprm->argc) < 0)
1641 bprm->envc = count(envp, MAX_ARG_STRINGS);
1642 if ((retval = bprm->envc) < 0)
1645 retval = prepare_binprm(bprm);
1649 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1653 bprm->exec = bprm->p;
1654 retval = copy_strings(bprm->envc, envp, bprm);
1658 retval = copy_strings(bprm->argc, argv, bprm);
1662 retval = exec_binprm(bprm);
1666 /* execve succeeded */
1667 current->fs->in_exec = 0;
1668 current->in_execve = 0;
1669 acct_update_integrals(current);
1670 task_numa_free(current);
1675 put_files_struct(displaced);
1680 acct_arg_size(bprm, 0);
1685 current->fs->in_exec = 0;
1686 current->in_execve = 0;
1694 reset_files_struct(displaced);
1700 int do_execve(struct filename *filename,
1701 const char __user *const __user *__argv,
1702 const char __user *const __user *__envp)
1704 struct user_arg_ptr argv = { .ptr.native = __argv };
1705 struct user_arg_ptr envp = { .ptr.native = __envp };
1706 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1709 int do_execveat(int fd, struct filename *filename,
1710 const char __user *const __user *__argv,
1711 const char __user *const __user *__envp,
1714 struct user_arg_ptr argv = { .ptr.native = __argv };
1715 struct user_arg_ptr envp = { .ptr.native = __envp };
1717 return do_execveat_common(fd, filename, argv, envp, flags);
1720 #ifdef CONFIG_COMPAT
1721 static int compat_do_execve(struct filename *filename,
1722 const compat_uptr_t __user *__argv,
1723 const compat_uptr_t __user *__envp)
1725 struct user_arg_ptr argv = {
1727 .ptr.compat = __argv,
1729 struct user_arg_ptr envp = {
1731 .ptr.compat = __envp,
1733 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1736 static int compat_do_execveat(int fd, struct filename *filename,
1737 const compat_uptr_t __user *__argv,
1738 const compat_uptr_t __user *__envp,
1741 struct user_arg_ptr argv = {
1743 .ptr.compat = __argv,
1745 struct user_arg_ptr envp = {
1747 .ptr.compat = __envp,
1749 return do_execveat_common(fd, filename, argv, envp, flags);
1753 void set_binfmt(struct linux_binfmt *new)
1755 struct mm_struct *mm = current->mm;
1758 module_put(mm->binfmt->module);
1762 __module_get(new->module);
1764 EXPORT_SYMBOL(set_binfmt);
1767 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1769 void set_dumpable(struct mm_struct *mm, int value)
1771 unsigned long old, new;
1773 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1777 old = ACCESS_ONCE(mm->flags);
1778 new = (old & ~MMF_DUMPABLE_MASK) | value;
1779 } while (cmpxchg(&mm->flags, old, new) != old);
1782 SYSCALL_DEFINE3(execve,
1783 const char __user *, filename,
1784 const char __user *const __user *, argv,
1785 const char __user *const __user *, envp)
1787 return do_execve(getname(filename), argv, envp);
1790 SYSCALL_DEFINE5(execveat,
1791 int, fd, const char __user *, filename,
1792 const char __user *const __user *, argv,
1793 const char __user *const __user *, envp,
1796 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1798 return do_execveat(fd,
1799 getname_flags(filename, lookup_flags, NULL),
1803 #ifdef CONFIG_COMPAT
1804 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1805 const compat_uptr_t __user *, argv,
1806 const compat_uptr_t __user *, envp)
1808 return compat_do_execve(getname(filename), argv, envp);
1811 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1812 const char __user *, filename,
1813 const compat_uptr_t __user *, argv,
1814 const compat_uptr_t __user *, envp,
1817 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1819 return compat_do_execveat(fd,
1820 getname_flags(filename, lookup_flags, NULL),