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/user_namespace.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 | MAY_OPEN,
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;
209 unsigned long ptr_size;
213 * Since the stack will hold pointers to the strings, we
214 * must account for them as well.
216 * The size calculation is the entire vma while each arg page is
217 * built, so each time we get here it's calculating how far it
218 * is currently (rather than each call being just the newly
219 * added size from the arg page). As a result, we need to
220 * always add the entire size of the pointers, so that on the
221 * last call to get_arg_page() we'll actually have the entire
224 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
225 if (ptr_size > ULONG_MAX - size)
229 acct_arg_size(bprm, size / PAGE_SIZE);
232 * We've historically supported up to 32 pages (ARG_MAX)
233 * of argument strings even with small stacks
239 * Limit to 1/4-th the stack size for the argv+env strings.
241 * - the remaining binfmt code will not run out of stack space,
242 * - the program will have a reasonable amount of stack left
245 rlim = current->signal->rlim;
246 if (size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4)
257 static void put_arg_page(struct page *page)
262 static void free_arg_page(struct linux_binprm *bprm, int i)
266 static void free_arg_pages(struct linux_binprm *bprm)
270 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
273 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
276 static int __bprm_mm_init(struct linux_binprm *bprm)
279 struct vm_area_struct *vma = NULL;
280 struct mm_struct *mm = bprm->mm;
282 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
286 down_write(&mm->mmap_sem);
290 * Place the stack at the largest stack address the architecture
291 * supports. Later, we'll move this to an appropriate place. We don't
292 * use STACK_TOP because that can depend on attributes which aren't
295 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
296 vma->vm_end = STACK_TOP_MAX;
297 vma->vm_start = vma->vm_end - PAGE_SIZE;
298 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
299 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
300 INIT_LIST_HEAD(&vma->anon_vma_chain);
302 err = insert_vm_struct(mm, vma);
306 mm->stack_vm = mm->total_vm = 1;
307 arch_bprm_mm_init(mm, vma);
308 up_write(&mm->mmap_sem);
309 bprm->p = vma->vm_end - sizeof(void *);
312 up_write(&mm->mmap_sem);
314 kmem_cache_free(vm_area_cachep, vma);
318 static bool valid_arg_len(struct linux_binprm *bprm, long len)
320 return len <= MAX_ARG_STRLEN;
325 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
329 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
334 page = bprm->page[pos / PAGE_SIZE];
335 if (!page && write) {
336 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
339 bprm->page[pos / PAGE_SIZE] = page;
345 static void put_arg_page(struct page *page)
349 static void free_arg_page(struct linux_binprm *bprm, int i)
352 __free_page(bprm->page[i]);
353 bprm->page[i] = NULL;
357 static void free_arg_pages(struct linux_binprm *bprm)
361 for (i = 0; i < MAX_ARG_PAGES; i++)
362 free_arg_page(bprm, i);
365 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
370 static int __bprm_mm_init(struct linux_binprm *bprm)
372 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
376 static bool valid_arg_len(struct linux_binprm *bprm, long len)
378 return len <= bprm->p;
381 #endif /* CONFIG_MMU */
384 * Create a new mm_struct and populate it with a temporary stack
385 * vm_area_struct. We don't have enough context at this point to set the stack
386 * flags, permissions, and offset, so we use temporary values. We'll update
387 * them later in setup_arg_pages().
389 static int bprm_mm_init(struct linux_binprm *bprm)
392 struct mm_struct *mm = NULL;
394 bprm->mm = mm = mm_alloc();
399 err = __bprm_mm_init(bprm);
414 struct user_arg_ptr {
419 const char __user *const __user *native;
421 const compat_uptr_t __user *compat;
426 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
428 const char __user *native;
431 if (unlikely(argv.is_compat)) {
432 compat_uptr_t compat;
434 if (get_user(compat, argv.ptr.compat + nr))
435 return ERR_PTR(-EFAULT);
437 return compat_ptr(compat);
441 if (get_user(native, argv.ptr.native + nr))
442 return ERR_PTR(-EFAULT);
448 * count() counts the number of strings in array ARGV.
450 static int count(struct user_arg_ptr argv, int max)
454 if (argv.ptr.native != NULL) {
456 const char __user *p = get_user_arg_ptr(argv, i);
468 if (fatal_signal_pending(current))
469 return -ERESTARTNOHAND;
477 * 'copy_strings()' copies argument/environment strings from the old
478 * processes's memory to the new process's stack. The call to get_user_pages()
479 * ensures the destination page is created and not swapped out.
481 static int copy_strings(int argc, struct user_arg_ptr argv,
482 struct linux_binprm *bprm)
484 struct page *kmapped_page = NULL;
486 unsigned long kpos = 0;
490 const char __user *str;
495 str = get_user_arg_ptr(argv, argc);
499 len = strnlen_user(str, MAX_ARG_STRLEN);
504 if (!valid_arg_len(bprm, len))
507 /* We're going to work our way backwords. */
513 int offset, bytes_to_copy;
515 if (fatal_signal_pending(current)) {
516 ret = -ERESTARTNOHAND;
521 offset = pos % PAGE_SIZE;
525 bytes_to_copy = offset;
526 if (bytes_to_copy > len)
529 offset -= bytes_to_copy;
530 pos -= bytes_to_copy;
531 str -= bytes_to_copy;
532 len -= bytes_to_copy;
534 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
537 page = get_arg_page(bprm, pos, 1);
544 flush_kernel_dcache_page(kmapped_page);
545 kunmap(kmapped_page);
546 put_arg_page(kmapped_page);
549 kaddr = kmap(kmapped_page);
550 kpos = pos & PAGE_MASK;
551 flush_arg_page(bprm, kpos, kmapped_page);
553 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
562 flush_kernel_dcache_page(kmapped_page);
563 kunmap(kmapped_page);
564 put_arg_page(kmapped_page);
570 * Like copy_strings, but get argv and its values from kernel memory.
572 int copy_strings_kernel(int argc, const char *const *__argv,
573 struct linux_binprm *bprm)
576 mm_segment_t oldfs = get_fs();
577 struct user_arg_ptr argv = {
578 .ptr.native = (const char __user *const __user *)__argv,
582 r = copy_strings(argc, argv, bprm);
587 EXPORT_SYMBOL(copy_strings_kernel);
592 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
593 * the binfmt code determines where the new stack should reside, we shift it to
594 * its final location. The process proceeds as follows:
596 * 1) Use shift to calculate the new vma endpoints.
597 * 2) Extend vma to cover both the old and new ranges. This ensures the
598 * arguments passed to subsequent functions are consistent.
599 * 3) Move vma's page tables to the new range.
600 * 4) Free up any cleared pgd range.
601 * 5) Shrink the vma to cover only the new range.
603 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
605 struct mm_struct *mm = vma->vm_mm;
606 unsigned long old_start = vma->vm_start;
607 unsigned long old_end = vma->vm_end;
608 unsigned long length = old_end - old_start;
609 unsigned long new_start = old_start - shift;
610 unsigned long new_end = old_end - shift;
611 struct mmu_gather tlb;
613 BUG_ON(new_start > new_end);
616 * ensure there are no vmas between where we want to go
619 if (vma != find_vma(mm, new_start))
623 * cover the whole range: [new_start, old_end)
625 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
629 * move the page tables downwards, on failure we rely on
630 * process cleanup to remove whatever mess we made.
632 if (length != move_page_tables(vma, old_start,
633 vma, new_start, length, false))
637 tlb_gather_mmu(&tlb, mm, old_start, old_end);
638 if (new_end > old_start) {
640 * when the old and new regions overlap clear from new_end.
642 free_pgd_range(&tlb, new_end, old_end, new_end,
643 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
646 * otherwise, clean from old_start; this is done to not touch
647 * the address space in [new_end, old_start) some architectures
648 * have constraints on va-space that make this illegal (IA64) -
649 * for the others its just a little faster.
651 free_pgd_range(&tlb, old_start, old_end, new_end,
652 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
654 tlb_finish_mmu(&tlb, old_start, old_end);
657 * Shrink the vma to just the new range. Always succeeds.
659 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
665 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
666 * the stack is optionally relocated, and some extra space is added.
668 int setup_arg_pages(struct linux_binprm *bprm,
669 unsigned long stack_top,
670 int executable_stack)
673 unsigned long stack_shift;
674 struct mm_struct *mm = current->mm;
675 struct vm_area_struct *vma = bprm->vma;
676 struct vm_area_struct *prev = NULL;
677 unsigned long vm_flags;
678 unsigned long stack_base;
679 unsigned long stack_size;
680 unsigned long stack_expand;
681 unsigned long rlim_stack;
683 #ifdef CONFIG_STACK_GROWSUP
684 /* Limit stack size */
685 stack_base = rlimit_max(RLIMIT_STACK);
686 if (stack_base > STACK_SIZE_MAX)
687 stack_base = STACK_SIZE_MAX;
689 /* Add space for stack randomization. */
690 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
692 /* Make sure we didn't let the argument array grow too large. */
693 if (vma->vm_end - vma->vm_start > stack_base)
696 stack_base = PAGE_ALIGN(stack_top - stack_base);
698 stack_shift = vma->vm_start - stack_base;
699 mm->arg_start = bprm->p - stack_shift;
700 bprm->p = vma->vm_end - stack_shift;
702 stack_top = arch_align_stack(stack_top);
703 stack_top = PAGE_ALIGN(stack_top);
705 if (unlikely(stack_top < mmap_min_addr) ||
706 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
709 stack_shift = vma->vm_end - stack_top;
711 bprm->p -= stack_shift;
712 mm->arg_start = bprm->p;
716 bprm->loader -= stack_shift;
717 bprm->exec -= stack_shift;
719 down_write(&mm->mmap_sem);
720 vm_flags = VM_STACK_FLAGS;
723 * Adjust stack execute permissions; explicitly enable for
724 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
725 * (arch default) otherwise.
727 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
729 else if (executable_stack == EXSTACK_DISABLE_X)
730 vm_flags &= ~VM_EXEC;
731 vm_flags |= mm->def_flags;
732 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
734 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
740 /* Move stack pages down in memory. */
742 ret = shift_arg_pages(vma, stack_shift);
747 /* mprotect_fixup is overkill to remove the temporary stack flags */
748 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
750 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
751 stack_size = vma->vm_end - vma->vm_start;
753 * Align this down to a page boundary as expand_stack
756 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
757 #ifdef CONFIG_STACK_GROWSUP
758 if (stack_size + stack_expand > rlim_stack)
759 stack_base = vma->vm_start + rlim_stack;
761 stack_base = vma->vm_end + stack_expand;
763 if (stack_size + stack_expand > rlim_stack)
764 stack_base = vma->vm_end - rlim_stack;
766 stack_base = vma->vm_start - stack_expand;
768 current->mm->start_stack = bprm->p;
769 ret = expand_stack(vma, stack_base);
774 up_write(&mm->mmap_sem);
777 EXPORT_SYMBOL(setup_arg_pages);
779 #endif /* CONFIG_MMU */
781 static struct file *do_open_execat(int fd, struct filename *name, int flags)
785 struct open_flags open_exec_flags = {
786 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
787 .acc_mode = MAY_EXEC | MAY_OPEN,
788 .intent = LOOKUP_OPEN,
789 .lookup_flags = LOOKUP_FOLLOW,
792 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
793 return ERR_PTR(-EINVAL);
794 if (flags & AT_SYMLINK_NOFOLLOW)
795 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
796 if (flags & AT_EMPTY_PATH)
797 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
799 file = do_filp_open(fd, name, &open_exec_flags);
804 if (!S_ISREG(file_inode(file)->i_mode))
807 if (path_noexec(&file->f_path))
810 err = deny_write_access(file);
814 if (name->name[0] != '\0')
825 struct file *open_exec(const char *name)
827 struct filename *filename = getname_kernel(name);
828 struct file *f = ERR_CAST(filename);
830 if (!IS_ERR(filename)) {
831 f = do_open_execat(AT_FDCWD, filename, 0);
836 EXPORT_SYMBOL(open_exec);
838 int kernel_read(struct file *file, loff_t offset,
839 char *addr, unsigned long count)
847 /* The cast to a user pointer is valid due to the set_fs() */
848 result = vfs_read(file, (void __user *)addr, count, &pos);
853 EXPORT_SYMBOL(kernel_read);
855 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
857 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
859 flush_icache_range(addr, addr + len);
862 EXPORT_SYMBOL(read_code);
864 static int exec_mmap(struct mm_struct *mm)
866 struct task_struct *tsk;
867 struct mm_struct *old_mm, *active_mm;
869 /* Notify parent that we're no longer interested in the old VM */
871 old_mm = current->mm;
872 mm_release(tsk, old_mm);
877 * Make sure that if there is a core dump in progress
878 * for the old mm, we get out and die instead of going
879 * through with the exec. We must hold mmap_sem around
880 * checking core_state and changing tsk->mm.
882 down_read(&old_mm->mmap_sem);
883 if (unlikely(old_mm->core_state)) {
884 up_read(&old_mm->mmap_sem);
889 active_mm = tsk->active_mm;
892 activate_mm(active_mm, mm);
893 tsk->mm->vmacache_seqnum = 0;
897 up_read(&old_mm->mmap_sem);
898 BUG_ON(active_mm != old_mm);
899 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
900 mm_update_next_owner(old_mm);
909 * This function makes sure the current process has its own signal table,
910 * so that flush_signal_handlers can later reset the handlers without
911 * disturbing other processes. (Other processes might share the signal
912 * table via the CLONE_SIGHAND option to clone().)
914 static int de_thread(struct task_struct *tsk)
916 struct signal_struct *sig = tsk->signal;
917 struct sighand_struct *oldsighand = tsk->sighand;
918 spinlock_t *lock = &oldsighand->siglock;
920 if (thread_group_empty(tsk))
921 goto no_thread_group;
924 * Kill all other threads in the thread group.
927 if (signal_group_exit(sig)) {
929 * Another group action in progress, just
930 * return so that the signal is processed.
932 spin_unlock_irq(lock);
936 sig->group_exit_task = tsk;
937 sig->notify_count = zap_other_threads(tsk);
938 if (!thread_group_leader(tsk))
941 while (sig->notify_count) {
942 __set_current_state(TASK_KILLABLE);
943 spin_unlock_irq(lock);
945 if (unlikely(__fatal_signal_pending(tsk)))
949 spin_unlock_irq(lock);
952 * At this point all other threads have exited, all we have to
953 * do is to wait for the thread group leader to become inactive,
954 * and to assume its PID:
956 if (!thread_group_leader(tsk)) {
957 struct task_struct *leader = tsk->group_leader;
960 threadgroup_change_begin(tsk);
961 write_lock_irq(&tasklist_lock);
963 * Do this under tasklist_lock to ensure that
964 * exit_notify() can't miss ->group_exit_task
966 sig->notify_count = -1;
967 if (likely(leader->exit_state))
969 __set_current_state(TASK_KILLABLE);
970 write_unlock_irq(&tasklist_lock);
971 threadgroup_change_end(tsk);
973 if (unlikely(__fatal_signal_pending(tsk)))
978 * The only record we have of the real-time age of a
979 * process, regardless of execs it's done, is start_time.
980 * All the past CPU time is accumulated in signal_struct
981 * from sister threads now dead. But in this non-leader
982 * exec, nothing survives from the original leader thread,
983 * whose birth marks the true age of this process now.
984 * When we take on its identity by switching to its PID, we
985 * also take its birthdate (always earlier than our own).
987 tsk->start_time = leader->start_time;
988 tsk->real_start_time = leader->real_start_time;
990 BUG_ON(!same_thread_group(leader, tsk));
991 BUG_ON(has_group_leader_pid(tsk));
993 * An exec() starts a new thread group with the
994 * TGID of the previous thread group. Rehash the
995 * two threads with a switched PID, and release
996 * the former thread group leader:
999 /* Become a process group leader with the old leader's pid.
1000 * The old leader becomes a thread of the this thread group.
1001 * Note: The old leader also uses this pid until release_task
1002 * is called. Odd but simple and correct.
1004 tsk->pid = leader->pid;
1005 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1006 transfer_pid(leader, tsk, PIDTYPE_PGID);
1007 transfer_pid(leader, tsk, PIDTYPE_SID);
1009 list_replace_rcu(&leader->tasks, &tsk->tasks);
1010 list_replace_init(&leader->sibling, &tsk->sibling);
1012 tsk->group_leader = tsk;
1013 leader->group_leader = tsk;
1015 tsk->exit_signal = SIGCHLD;
1016 leader->exit_signal = -1;
1018 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1019 leader->exit_state = EXIT_DEAD;
1022 * We are going to release_task()->ptrace_unlink() silently,
1023 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1024 * the tracer wont't block again waiting for this thread.
1026 if (unlikely(leader->ptrace))
1027 __wake_up_parent(leader, leader->parent);
1028 write_unlock_irq(&tasklist_lock);
1029 threadgroup_change_end(tsk);
1031 release_task(leader);
1034 sig->group_exit_task = NULL;
1035 sig->notify_count = 0;
1038 /* we have changed execution domain */
1039 tsk->exit_signal = SIGCHLD;
1042 flush_itimer_signals();
1044 if (atomic_read(&oldsighand->count) != 1) {
1045 struct sighand_struct *newsighand;
1047 * This ->sighand is shared with the CLONE_SIGHAND
1048 * but not CLONE_THREAD task, switch to the new one.
1050 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1054 atomic_set(&newsighand->count, 1);
1055 memcpy(newsighand->action, oldsighand->action,
1056 sizeof(newsighand->action));
1058 write_lock_irq(&tasklist_lock);
1059 spin_lock(&oldsighand->siglock);
1060 rcu_assign_pointer(tsk->sighand, newsighand);
1061 spin_unlock(&oldsighand->siglock);
1062 write_unlock_irq(&tasklist_lock);
1064 __cleanup_sighand(oldsighand);
1067 BUG_ON(!thread_group_leader(tsk));
1071 /* protects against exit_notify() and __exit_signal() */
1072 read_lock(&tasklist_lock);
1073 sig->group_exit_task = NULL;
1074 sig->notify_count = 0;
1075 read_unlock(&tasklist_lock);
1079 char *get_task_comm(char *buf, struct task_struct *tsk)
1081 /* buf must be at least sizeof(tsk->comm) in size */
1083 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1087 EXPORT_SYMBOL_GPL(get_task_comm);
1090 * These functions flushes out all traces of the currently running executable
1091 * so that a new one can be started
1094 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1097 trace_task_rename(tsk, buf);
1098 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1100 perf_event_comm(tsk, exec);
1103 int flush_old_exec(struct linux_binprm * bprm)
1108 * Make sure we have a private signal table and that
1109 * we are unassociated from the previous thread group.
1111 retval = de_thread(current);
1116 * Must be called _before_ exec_mmap() as bprm->mm is
1117 * not visibile until then. This also enables the update
1120 set_mm_exe_file(bprm->mm, bprm->file);
1123 * Release all of the old mmap stuff
1125 acct_arg_size(bprm, 0);
1126 retval = exec_mmap(bprm->mm);
1130 bprm->mm = NULL; /* We're using it now */
1133 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1134 PF_NOFREEZE | PF_NO_SETAFFINITY);
1136 current->personality &= ~bprm->per_clear;
1139 * We have to apply CLOEXEC before we change whether the process is
1140 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1141 * trying to access the should-be-closed file descriptors of a process
1142 * undergoing exec(2).
1144 do_close_on_exec(current->files);
1150 EXPORT_SYMBOL(flush_old_exec);
1152 void would_dump(struct linux_binprm *bprm, struct file *file)
1154 struct inode *inode = file_inode(file);
1155 if (inode_permission(inode, MAY_READ) < 0) {
1156 struct user_namespace *old, *user_ns;
1157 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1159 /* Ensure mm->user_ns contains the executable */
1160 user_ns = old = bprm->mm->user_ns;
1161 while ((user_ns != &init_user_ns) &&
1162 !privileged_wrt_inode_uidgid(user_ns, inode))
1163 user_ns = user_ns->parent;
1165 if (old != user_ns) {
1166 bprm->mm->user_ns = get_user_ns(user_ns);
1171 EXPORT_SYMBOL(would_dump);
1173 void setup_new_exec(struct linux_binprm * bprm)
1175 arch_pick_mmap_layout(current->mm);
1177 /* This is the point of no return */
1178 current->sas_ss_sp = current->sas_ss_size = 0;
1180 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1181 set_dumpable(current->mm, SUID_DUMP_USER);
1183 set_dumpable(current->mm, suid_dumpable);
1186 __set_task_comm(current, kbasename(bprm->filename), true);
1188 /* Set the new mm task size. We have to do that late because it may
1189 * depend on TIF_32BIT which is only updated in flush_thread() on
1190 * some architectures like powerpc
1192 current->mm->task_size = TASK_SIZE;
1194 /* install the new credentials */
1195 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1196 !gid_eq(bprm->cred->gid, current_egid())) {
1197 current->pdeath_signal = 0;
1199 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1200 set_dumpable(current->mm, suid_dumpable);
1203 /* An exec changes our domain. We are no longer part of the thread
1205 current->self_exec_id++;
1206 flush_signal_handlers(current, 0);
1208 EXPORT_SYMBOL(setup_new_exec);
1211 * Prepare credentials and lock ->cred_guard_mutex.
1212 * install_exec_creds() commits the new creds and drops the lock.
1213 * Or, if exec fails before, free_bprm() should release ->cred and
1216 int prepare_bprm_creds(struct linux_binprm *bprm)
1218 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1219 return -ERESTARTNOINTR;
1221 bprm->cred = prepare_exec_creds();
1222 if (likely(bprm->cred))
1225 mutex_unlock(¤t->signal->cred_guard_mutex);
1229 static void free_bprm(struct linux_binprm *bprm)
1231 free_arg_pages(bprm);
1233 mutex_unlock(¤t->signal->cred_guard_mutex);
1234 abort_creds(bprm->cred);
1237 allow_write_access(bprm->file);
1240 /* If a binfmt changed the interp, free it. */
1241 if (bprm->interp != bprm->filename)
1242 kfree(bprm->interp);
1246 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1248 /* If a binfmt changed the interp, free it first. */
1249 if (bprm->interp != bprm->filename)
1250 kfree(bprm->interp);
1251 bprm->interp = kstrdup(interp, GFP_KERNEL);
1256 EXPORT_SYMBOL(bprm_change_interp);
1259 * install the new credentials for this executable
1261 void install_exec_creds(struct linux_binprm *bprm)
1263 security_bprm_committing_creds(bprm);
1265 commit_creds(bprm->cred);
1269 * Disable monitoring for regular users
1270 * when executing setuid binaries. Must
1271 * wait until new credentials are committed
1272 * by commit_creds() above
1274 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1275 perf_event_exit_task(current);
1277 * cred_guard_mutex must be held at least to this point to prevent
1278 * ptrace_attach() from altering our determination of the task's
1279 * credentials; any time after this it may be unlocked.
1281 security_bprm_committed_creds(bprm);
1282 mutex_unlock(¤t->signal->cred_guard_mutex);
1284 EXPORT_SYMBOL(install_exec_creds);
1287 * determine how safe it is to execute the proposed program
1288 * - the caller must hold ->cred_guard_mutex to protect against
1289 * PTRACE_ATTACH or seccomp thread-sync
1291 static void check_unsafe_exec(struct linux_binprm *bprm)
1293 struct task_struct *p = current, *t;
1297 if (ptracer_capable(p, current_user_ns()))
1298 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1300 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1304 * This isn't strictly necessary, but it makes it harder for LSMs to
1307 if (task_no_new_privs(current))
1308 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1312 spin_lock(&p->fs->lock);
1314 while_each_thread(p, t) {
1320 if (p->fs->users > n_fs)
1321 bprm->unsafe |= LSM_UNSAFE_SHARE;
1324 spin_unlock(&p->fs->lock);
1327 static void bprm_fill_uid(struct linux_binprm *bprm)
1329 struct inode *inode;
1334 /* clear any previous set[ug]id data from a previous binary */
1335 bprm->cred->euid = current_euid();
1336 bprm->cred->egid = current_egid();
1338 if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1341 if (task_no_new_privs(current))
1344 inode = file_inode(bprm->file);
1345 mode = READ_ONCE(inode->i_mode);
1346 if (!(mode & (S_ISUID|S_ISGID)))
1349 /* Be careful if suid/sgid is set */
1350 mutex_lock(&inode->i_mutex);
1352 /* reload atomically mode/uid/gid now that lock held */
1353 mode = inode->i_mode;
1356 mutex_unlock(&inode->i_mutex);
1358 /* We ignore suid/sgid if there are no mappings for them in the ns */
1359 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1360 !kgid_has_mapping(bprm->cred->user_ns, gid))
1363 if (mode & S_ISUID) {
1364 bprm->per_clear |= PER_CLEAR_ON_SETID;
1365 bprm->cred->euid = uid;
1368 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1369 bprm->per_clear |= PER_CLEAR_ON_SETID;
1370 bprm->cred->egid = gid;
1375 * Fill the binprm structure from the inode.
1376 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1378 * This may be called multiple times for binary chains (scripts for example).
1380 int prepare_binprm(struct linux_binprm *bprm)
1384 bprm_fill_uid(bprm);
1386 /* fill in binprm security blob */
1387 retval = security_bprm_set_creds(bprm);
1390 bprm->cred_prepared = 1;
1392 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1393 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1396 EXPORT_SYMBOL(prepare_binprm);
1399 * Arguments are '\0' separated strings found at the location bprm->p
1400 * points to; chop off the first by relocating brpm->p to right after
1401 * the first '\0' encountered.
1403 int remove_arg_zero(struct linux_binprm *bprm)
1406 unsigned long offset;
1414 offset = bprm->p & ~PAGE_MASK;
1415 page = get_arg_page(bprm, bprm->p, 0);
1420 kaddr = kmap_atomic(page);
1422 for (; offset < PAGE_SIZE && kaddr[offset];
1423 offset++, bprm->p++)
1426 kunmap_atomic(kaddr);
1429 if (offset == PAGE_SIZE)
1430 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1431 } while (offset == PAGE_SIZE);
1440 EXPORT_SYMBOL(remove_arg_zero);
1442 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1444 * cycle the list of binary formats handler, until one recognizes the image
1446 int search_binary_handler(struct linux_binprm *bprm)
1448 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1449 struct linux_binfmt *fmt;
1452 /* This allows 4 levels of binfmt rewrites before failing hard. */
1453 if (bprm->recursion_depth > 5)
1456 retval = security_bprm_check(bprm);
1462 read_lock(&binfmt_lock);
1463 list_for_each_entry(fmt, &formats, lh) {
1464 if (!try_module_get(fmt->module))
1466 read_unlock(&binfmt_lock);
1467 bprm->recursion_depth++;
1468 retval = fmt->load_binary(bprm);
1469 read_lock(&binfmt_lock);
1471 bprm->recursion_depth--;
1472 if (retval < 0 && !bprm->mm) {
1473 /* we got to flush_old_exec() and failed after it */
1474 read_unlock(&binfmt_lock);
1475 force_sigsegv(SIGSEGV, current);
1478 if (retval != -ENOEXEC || !bprm->file) {
1479 read_unlock(&binfmt_lock);
1483 read_unlock(&binfmt_lock);
1486 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1487 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1489 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1497 EXPORT_SYMBOL(search_binary_handler);
1499 static int exec_binprm(struct linux_binprm *bprm)
1501 pid_t old_pid, old_vpid;
1504 /* Need to fetch pid before load_binary changes it */
1505 old_pid = current->pid;
1507 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1510 ret = search_binary_handler(bprm);
1513 trace_sched_process_exec(current, old_pid, bprm);
1514 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1515 proc_exec_connector(current);
1522 * sys_execve() executes a new program.
1524 static int do_execveat_common(int fd, struct filename *filename,
1525 struct user_arg_ptr argv,
1526 struct user_arg_ptr envp,
1529 char *pathbuf = NULL;
1530 struct linux_binprm *bprm;
1532 struct files_struct *displaced;
1535 if (IS_ERR(filename))
1536 return PTR_ERR(filename);
1539 * We move the actual failure in case of RLIMIT_NPROC excess from
1540 * set*uid() to execve() because too many poorly written programs
1541 * don't check setuid() return code. Here we additionally recheck
1542 * whether NPROC limit is still exceeded.
1544 if ((current->flags & PF_NPROC_EXCEEDED) &&
1545 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1550 /* We're below the limit (still or again), so we don't want to make
1551 * further execve() calls fail. */
1552 current->flags &= ~PF_NPROC_EXCEEDED;
1554 retval = unshare_files(&displaced);
1559 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1563 retval = prepare_bprm_creds(bprm);
1567 check_unsafe_exec(bprm);
1568 current->in_execve = 1;
1570 file = do_open_execat(fd, filename, flags);
1571 retval = PTR_ERR(file);
1578 if (fd == AT_FDCWD || filename->name[0] == '/') {
1579 bprm->filename = filename->name;
1581 if (filename->name[0] == '\0')
1582 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1584 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1585 fd, filename->name);
1591 * Record that a name derived from an O_CLOEXEC fd will be
1592 * inaccessible after exec. Relies on having exclusive access to
1593 * current->files (due to unshare_files above).
1595 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1596 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1597 bprm->filename = pathbuf;
1599 bprm->interp = bprm->filename;
1601 retval = bprm_mm_init(bprm);
1605 bprm->argc = count(argv, MAX_ARG_STRINGS);
1606 if ((retval = bprm->argc) < 0)
1609 bprm->envc = count(envp, MAX_ARG_STRINGS);
1610 if ((retval = bprm->envc) < 0)
1613 retval = prepare_binprm(bprm);
1617 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1621 bprm->exec = bprm->p;
1622 retval = copy_strings(bprm->envc, envp, bprm);
1626 retval = copy_strings(bprm->argc, argv, bprm);
1630 would_dump(bprm, bprm->file);
1632 retval = exec_binprm(bprm);
1636 /* execve succeeded */
1637 current->fs->in_exec = 0;
1638 current->in_execve = 0;
1639 acct_update_integrals(current);
1640 task_numa_free(current);
1645 put_files_struct(displaced);
1650 acct_arg_size(bprm, 0);
1655 current->fs->in_exec = 0;
1656 current->in_execve = 0;
1664 reset_files_struct(displaced);
1670 int do_execve(struct filename *filename,
1671 const char __user *const __user *__argv,
1672 const char __user *const __user *__envp)
1674 struct user_arg_ptr argv = { .ptr.native = __argv };
1675 struct user_arg_ptr envp = { .ptr.native = __envp };
1676 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1679 int do_execveat(int fd, struct filename *filename,
1680 const char __user *const __user *__argv,
1681 const char __user *const __user *__envp,
1684 struct user_arg_ptr argv = { .ptr.native = __argv };
1685 struct user_arg_ptr envp = { .ptr.native = __envp };
1687 return do_execveat_common(fd, filename, argv, envp, flags);
1690 #ifdef CONFIG_COMPAT
1691 static int compat_do_execve(struct filename *filename,
1692 const compat_uptr_t __user *__argv,
1693 const compat_uptr_t __user *__envp)
1695 struct user_arg_ptr argv = {
1697 .ptr.compat = __argv,
1699 struct user_arg_ptr envp = {
1701 .ptr.compat = __envp,
1703 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1706 static int compat_do_execveat(int fd, struct filename *filename,
1707 const compat_uptr_t __user *__argv,
1708 const compat_uptr_t __user *__envp,
1711 struct user_arg_ptr argv = {
1713 .ptr.compat = __argv,
1715 struct user_arg_ptr envp = {
1717 .ptr.compat = __envp,
1719 return do_execveat_common(fd, filename, argv, envp, flags);
1723 void set_binfmt(struct linux_binfmt *new)
1725 struct mm_struct *mm = current->mm;
1728 module_put(mm->binfmt->module);
1732 __module_get(new->module);
1734 EXPORT_SYMBOL(set_binfmt);
1737 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1739 void set_dumpable(struct mm_struct *mm, int value)
1741 unsigned long old, new;
1743 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1747 old = ACCESS_ONCE(mm->flags);
1748 new = (old & ~MMF_DUMPABLE_MASK) | value;
1749 } while (cmpxchg(&mm->flags, old, new) != old);
1752 SYSCALL_DEFINE3(execve,
1753 const char __user *, filename,
1754 const char __user *const __user *, argv,
1755 const char __user *const __user *, envp)
1757 return do_execve(getname(filename), argv, envp);
1760 SYSCALL_DEFINE5(execveat,
1761 int, fd, const char __user *, filename,
1762 const char __user *const __user *, argv,
1763 const char __user *const __user *, envp,
1766 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1768 return do_execveat(fd,
1769 getname_flags(filename, lookup_flags, NULL),
1773 #ifdef CONFIG_COMPAT
1774 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1775 const compat_uptr_t __user *, argv,
1776 const compat_uptr_t __user *, envp)
1778 return compat_do_execve(getname(filename), argv, envp);
1781 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1782 const char __user *, filename,
1783 const compat_uptr_t __user *, argv,
1784 const compat_uptr_t __user *, envp,
1787 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1789 return compat_do_execveat(fd,
1790 getname_flags(filename, lookup_flags, NULL),