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>
60 #include <asm/uaccess.h>
61 #include <asm/mmu_context.h>
64 #include <trace/events/task.h>
67 #include <trace/events/sched.h>
69 int suid_dumpable = 0;
71 static LIST_HEAD(formats);
72 static DEFINE_RWLOCK(binfmt_lock);
74 void __register_binfmt(struct linux_binfmt * fmt, int insert)
77 if (WARN_ON(!fmt->load_binary))
79 write_lock(&binfmt_lock);
80 insert ? list_add(&fmt->lh, &formats) :
81 list_add_tail(&fmt->lh, &formats);
82 write_unlock(&binfmt_lock);
85 EXPORT_SYMBOL(__register_binfmt);
87 void unregister_binfmt(struct linux_binfmt * fmt)
89 write_lock(&binfmt_lock);
91 write_unlock(&binfmt_lock);
94 EXPORT_SYMBOL(unregister_binfmt);
96 static inline void put_binfmt(struct linux_binfmt * fmt)
98 module_put(fmt->module);
103 * Note that a shared library must be both readable and executable due to
106 * Also note that we take the address to load from from the file itself.
108 SYSCALL_DEFINE1(uselib, const char __user *, library)
110 struct linux_binfmt *fmt;
112 struct filename *tmp = getname(library);
113 int error = PTR_ERR(tmp);
114 static const struct open_flags uselib_flags = {
115 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
116 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
117 .intent = LOOKUP_OPEN,
118 .lookup_flags = LOOKUP_FOLLOW,
124 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
126 error = PTR_ERR(file);
131 if (!S_ISREG(file_inode(file)->i_mode))
135 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
142 read_lock(&binfmt_lock);
143 list_for_each_entry(fmt, &formats, lh) {
144 if (!fmt->load_shlib)
146 if (!try_module_get(fmt->module))
148 read_unlock(&binfmt_lock);
149 error = fmt->load_shlib(file);
150 read_lock(&binfmt_lock);
152 if (error != -ENOEXEC)
155 read_unlock(&binfmt_lock);
161 #endif /* #ifdef CONFIG_USELIB */
165 * The nascent bprm->mm is not visible until exec_mmap() but it can
166 * use a lot of memory, account these pages in current->mm temporary
167 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
168 * change the counter back via acct_arg_size(0).
170 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
172 struct mm_struct *mm = current->mm;
173 long diff = (long)(pages - bprm->vma_pages);
178 bprm->vma_pages = pages;
179 add_mm_counter(mm, MM_ANONPAGES, diff);
182 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
188 #ifdef CONFIG_STACK_GROWSUP
190 ret = expand_downwards(bprm->vma, pos);
195 ret = get_user_pages(current, bprm->mm, pos,
196 1, write, 1, &page, NULL);
201 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
204 acct_arg_size(bprm, size / PAGE_SIZE);
207 * We've historically supported up to 32 pages (ARG_MAX)
208 * of argument strings even with small stacks
214 * Limit to 1/4-th the stack size for the argv+env strings.
216 * - the remaining binfmt code will not run out of stack space,
217 * - the program will have a reasonable amount of stack left
220 rlim = current->signal->rlim;
221 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
230 static void put_arg_page(struct page *page)
235 static void free_arg_page(struct linux_binprm *bprm, int i)
239 static void free_arg_pages(struct linux_binprm *bprm)
243 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
246 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
249 static int __bprm_mm_init(struct linux_binprm *bprm)
252 struct vm_area_struct *vma = NULL;
253 struct mm_struct *mm = bprm->mm;
255 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
259 down_write(&mm->mmap_sem);
263 * Place the stack at the largest stack address the architecture
264 * supports. Later, we'll move this to an appropriate place. We don't
265 * use STACK_TOP because that can depend on attributes which aren't
268 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
269 vma->vm_end = STACK_TOP_MAX;
270 vma->vm_start = vma->vm_end - PAGE_SIZE;
271 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
272 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
273 INIT_LIST_HEAD(&vma->anon_vma_chain);
275 err = insert_vm_struct(mm, vma);
279 mm->stack_vm = mm->total_vm = 1;
280 arch_bprm_mm_init(mm, vma);
281 up_write(&mm->mmap_sem);
282 bprm->p = vma->vm_end - sizeof(void *);
285 up_write(&mm->mmap_sem);
287 kmem_cache_free(vm_area_cachep, vma);
291 static bool valid_arg_len(struct linux_binprm *bprm, long len)
293 return len <= MAX_ARG_STRLEN;
298 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
302 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
307 page = bprm->page[pos / PAGE_SIZE];
308 if (!page && write) {
309 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
312 bprm->page[pos / PAGE_SIZE] = page;
318 static void put_arg_page(struct page *page)
322 static void free_arg_page(struct linux_binprm *bprm, int i)
325 __free_page(bprm->page[i]);
326 bprm->page[i] = NULL;
330 static void free_arg_pages(struct linux_binprm *bprm)
334 for (i = 0; i < MAX_ARG_PAGES; i++)
335 free_arg_page(bprm, i);
338 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
343 static int __bprm_mm_init(struct linux_binprm *bprm)
345 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
349 static bool valid_arg_len(struct linux_binprm *bprm, long len)
351 return len <= bprm->p;
354 #endif /* CONFIG_MMU */
357 * Create a new mm_struct and populate it with a temporary stack
358 * vm_area_struct. We don't have enough context at this point to set the stack
359 * flags, permissions, and offset, so we use temporary values. We'll update
360 * them later in setup_arg_pages().
362 static int bprm_mm_init(struct linux_binprm *bprm)
365 struct mm_struct *mm = NULL;
367 bprm->mm = mm = mm_alloc();
372 err = __bprm_mm_init(bprm);
387 struct user_arg_ptr {
392 const char __user *const __user *native;
394 const compat_uptr_t __user *compat;
399 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
401 const char __user *native;
404 if (unlikely(argv.is_compat)) {
405 compat_uptr_t compat;
407 if (get_user(compat, argv.ptr.compat + nr))
408 return ERR_PTR(-EFAULT);
410 return compat_ptr(compat);
414 if (get_user(native, argv.ptr.native + nr))
415 return ERR_PTR(-EFAULT);
421 * count() counts the number of strings in array ARGV.
423 static int count(struct user_arg_ptr argv, int max)
427 if (argv.ptr.native != NULL) {
429 const char __user *p = get_user_arg_ptr(argv, i);
441 if (fatal_signal_pending(current))
442 return -ERESTARTNOHAND;
450 * 'copy_strings()' copies argument/environment strings from the old
451 * processes's memory to the new process's stack. The call to get_user_pages()
452 * ensures the destination page is created and not swapped out.
454 static int copy_strings(int argc, struct user_arg_ptr argv,
455 struct linux_binprm *bprm)
457 struct page *kmapped_page = NULL;
459 unsigned long kpos = 0;
463 const char __user *str;
468 str = get_user_arg_ptr(argv, argc);
472 len = strnlen_user(str, MAX_ARG_STRLEN);
477 if (!valid_arg_len(bprm, len))
480 /* We're going to work our way backwords. */
486 int offset, bytes_to_copy;
488 if (fatal_signal_pending(current)) {
489 ret = -ERESTARTNOHAND;
494 offset = pos % PAGE_SIZE;
498 bytes_to_copy = offset;
499 if (bytes_to_copy > len)
502 offset -= bytes_to_copy;
503 pos -= bytes_to_copy;
504 str -= bytes_to_copy;
505 len -= bytes_to_copy;
507 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
510 page = get_arg_page(bprm, pos, 1);
517 flush_kernel_dcache_page(kmapped_page);
518 kunmap(kmapped_page);
519 put_arg_page(kmapped_page);
522 kaddr = kmap(kmapped_page);
523 kpos = pos & PAGE_MASK;
524 flush_arg_page(bprm, kpos, kmapped_page);
526 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
535 flush_kernel_dcache_page(kmapped_page);
536 kunmap(kmapped_page);
537 put_arg_page(kmapped_page);
543 * Like copy_strings, but get argv and its values from kernel memory.
545 int copy_strings_kernel(int argc, const char *const *__argv,
546 struct linux_binprm *bprm)
549 mm_segment_t oldfs = get_fs();
550 struct user_arg_ptr argv = {
551 .ptr.native = (const char __user *const __user *)__argv,
555 r = copy_strings(argc, argv, bprm);
560 EXPORT_SYMBOL(copy_strings_kernel);
565 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
566 * the binfmt code determines where the new stack should reside, we shift it to
567 * its final location. The process proceeds as follows:
569 * 1) Use shift to calculate the new vma endpoints.
570 * 2) Extend vma to cover both the old and new ranges. This ensures the
571 * arguments passed to subsequent functions are consistent.
572 * 3) Move vma's page tables to the new range.
573 * 4) Free up any cleared pgd range.
574 * 5) Shrink the vma to cover only the new range.
576 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
578 struct mm_struct *mm = vma->vm_mm;
579 unsigned long old_start = vma->vm_start;
580 unsigned long old_end = vma->vm_end;
581 unsigned long length = old_end - old_start;
582 unsigned long new_start = old_start - shift;
583 unsigned long new_end = old_end - shift;
584 struct mmu_gather tlb;
586 BUG_ON(new_start > new_end);
589 * ensure there are no vmas between where we want to go
592 if (vma != find_vma(mm, new_start))
596 * cover the whole range: [new_start, old_end)
598 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
602 * move the page tables downwards, on failure we rely on
603 * process cleanup to remove whatever mess we made.
605 if (length != move_page_tables(vma, old_start,
606 vma, new_start, length, false))
610 tlb_gather_mmu(&tlb, mm, old_start, old_end);
611 if (new_end > old_start) {
613 * when the old and new regions overlap clear from new_end.
615 free_pgd_range(&tlb, new_end, old_end, new_end,
616 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
619 * otherwise, clean from old_start; this is done to not touch
620 * the address space in [new_end, old_start) some architectures
621 * have constraints on va-space that make this illegal (IA64) -
622 * for the others its just a little faster.
624 free_pgd_range(&tlb, old_start, old_end, new_end,
625 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
627 tlb_finish_mmu(&tlb, old_start, old_end);
630 * Shrink the vma to just the new range. Always succeeds.
632 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
638 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
639 * the stack is optionally relocated, and some extra space is added.
641 int setup_arg_pages(struct linux_binprm *bprm,
642 unsigned long stack_top,
643 int executable_stack)
646 unsigned long stack_shift;
647 struct mm_struct *mm = current->mm;
648 struct vm_area_struct *vma = bprm->vma;
649 struct vm_area_struct *prev = NULL;
650 unsigned long vm_flags;
651 unsigned long stack_base;
652 unsigned long stack_size;
653 unsigned long stack_expand;
654 unsigned long rlim_stack;
656 #ifdef CONFIG_STACK_GROWSUP
657 /* Limit stack size */
658 stack_base = rlimit_max(RLIMIT_STACK);
659 if (stack_base > STACK_SIZE_MAX)
660 stack_base = STACK_SIZE_MAX;
662 /* Add space for stack randomization. */
663 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
665 /* Make sure we didn't let the argument array grow too large. */
666 if (vma->vm_end - vma->vm_start > stack_base)
669 stack_base = PAGE_ALIGN(stack_top - stack_base);
671 stack_shift = vma->vm_start - stack_base;
672 mm->arg_start = bprm->p - stack_shift;
673 bprm->p = vma->vm_end - stack_shift;
675 stack_top = arch_align_stack(stack_top);
676 stack_top = PAGE_ALIGN(stack_top);
678 if (unlikely(stack_top < mmap_min_addr) ||
679 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
682 stack_shift = vma->vm_end - stack_top;
684 bprm->p -= stack_shift;
685 mm->arg_start = bprm->p;
689 bprm->loader -= stack_shift;
690 bprm->exec -= stack_shift;
692 down_write(&mm->mmap_sem);
693 vm_flags = VM_STACK_FLAGS;
696 * Adjust stack execute permissions; explicitly enable for
697 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
698 * (arch default) otherwise.
700 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
702 else if (executable_stack == EXSTACK_DISABLE_X)
703 vm_flags &= ~VM_EXEC;
704 vm_flags |= mm->def_flags;
705 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
707 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
713 /* Move stack pages down in memory. */
715 ret = shift_arg_pages(vma, stack_shift);
720 /* mprotect_fixup is overkill to remove the temporary stack flags */
721 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
723 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
724 stack_size = vma->vm_end - vma->vm_start;
726 * Align this down to a page boundary as expand_stack
729 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
730 #ifdef CONFIG_STACK_GROWSUP
731 if (stack_size + stack_expand > rlim_stack)
732 stack_base = vma->vm_start + rlim_stack;
734 stack_base = vma->vm_end + stack_expand;
736 if (stack_size + stack_expand > rlim_stack)
737 stack_base = vma->vm_end - rlim_stack;
739 stack_base = vma->vm_start - stack_expand;
741 current->mm->start_stack = bprm->p;
742 ret = expand_stack(vma, stack_base);
747 up_write(&mm->mmap_sem);
750 EXPORT_SYMBOL(setup_arg_pages);
752 #endif /* CONFIG_MMU */
754 static struct file *do_open_execat(int fd, struct filename *name, int flags)
758 struct open_flags open_exec_flags = {
759 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
760 .acc_mode = MAY_EXEC | MAY_OPEN,
761 .intent = LOOKUP_OPEN,
762 .lookup_flags = LOOKUP_FOLLOW,
765 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
766 return ERR_PTR(-EINVAL);
767 if (flags & AT_SYMLINK_NOFOLLOW)
768 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
769 if (flags & AT_EMPTY_PATH)
770 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
772 file = do_filp_open(fd, name, &open_exec_flags);
777 if (!S_ISREG(file_inode(file)->i_mode))
780 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
783 err = deny_write_access(file);
787 if (name->name[0] != '\0')
798 struct file *open_exec(const char *name)
800 struct filename *filename = getname_kernel(name);
801 struct file *f = ERR_CAST(filename);
803 if (!IS_ERR(filename)) {
804 f = do_open_execat(AT_FDCWD, filename, 0);
809 EXPORT_SYMBOL(open_exec);
811 int kernel_read(struct file *file, loff_t offset,
812 char *addr, unsigned long count)
820 /* The cast to a user pointer is valid due to the set_fs() */
821 result = vfs_read(file, (void __user *)addr, count, &pos);
826 EXPORT_SYMBOL(kernel_read);
828 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
830 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
832 flush_icache_range(addr, addr + len);
835 EXPORT_SYMBOL(read_code);
837 static int exec_mmap(struct mm_struct *mm)
839 struct task_struct *tsk;
840 struct mm_struct *old_mm, *active_mm;
842 /* Notify parent that we're no longer interested in the old VM */
844 old_mm = current->mm;
845 mm_release(tsk, old_mm);
850 * Make sure that if there is a core dump in progress
851 * for the old mm, we get out and die instead of going
852 * through with the exec. We must hold mmap_sem around
853 * checking core_state and changing tsk->mm.
855 down_read(&old_mm->mmap_sem);
856 if (unlikely(old_mm->core_state)) {
857 up_read(&old_mm->mmap_sem);
862 active_mm = tsk->active_mm;
865 activate_mm(active_mm, mm);
866 tsk->mm->vmacache_seqnum = 0;
870 up_read(&old_mm->mmap_sem);
871 BUG_ON(active_mm != old_mm);
872 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
873 mm_update_next_owner(old_mm);
882 * This function makes sure the current process has its own signal table,
883 * so that flush_signal_handlers can later reset the handlers without
884 * disturbing other processes. (Other processes might share the signal
885 * table via the CLONE_SIGHAND option to clone().)
887 static int de_thread(struct task_struct *tsk)
889 struct signal_struct *sig = tsk->signal;
890 struct sighand_struct *oldsighand = tsk->sighand;
891 spinlock_t *lock = &oldsighand->siglock;
893 if (thread_group_empty(tsk))
894 goto no_thread_group;
897 * Kill all other threads in the thread group.
900 if (signal_group_exit(sig)) {
902 * Another group action in progress, just
903 * return so that the signal is processed.
905 spin_unlock_irq(lock);
909 sig->group_exit_task = tsk;
910 sig->notify_count = zap_other_threads(tsk);
911 if (!thread_group_leader(tsk))
914 while (sig->notify_count) {
915 __set_current_state(TASK_KILLABLE);
916 spin_unlock_irq(lock);
918 if (unlikely(__fatal_signal_pending(tsk)))
922 spin_unlock_irq(lock);
925 * At this point all other threads have exited, all we have to
926 * do is to wait for the thread group leader to become inactive,
927 * and to assume its PID:
929 if (!thread_group_leader(tsk)) {
930 struct task_struct *leader = tsk->group_leader;
932 sig->notify_count = -1; /* for exit_notify() */
934 threadgroup_change_begin(tsk);
935 write_lock_irq(&tasklist_lock);
936 if (likely(leader->exit_state))
938 __set_current_state(TASK_KILLABLE);
939 write_unlock_irq(&tasklist_lock);
940 threadgroup_change_end(tsk);
942 if (unlikely(__fatal_signal_pending(tsk)))
947 * The only record we have of the real-time age of a
948 * process, regardless of execs it's done, is start_time.
949 * All the past CPU time is accumulated in signal_struct
950 * from sister threads now dead. But in this non-leader
951 * exec, nothing survives from the original leader thread,
952 * whose birth marks the true age of this process now.
953 * When we take on its identity by switching to its PID, we
954 * also take its birthdate (always earlier than our own).
956 tsk->start_time = leader->start_time;
957 tsk->real_start_time = leader->real_start_time;
959 BUG_ON(!same_thread_group(leader, tsk));
960 BUG_ON(has_group_leader_pid(tsk));
962 * An exec() starts a new thread group with the
963 * TGID of the previous thread group. Rehash the
964 * two threads with a switched PID, and release
965 * the former thread group leader:
968 /* Become a process group leader with the old leader's pid.
969 * The old leader becomes a thread of the this thread group.
970 * Note: The old leader also uses this pid until release_task
971 * is called. Odd but simple and correct.
973 tsk->pid = leader->pid;
974 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
975 transfer_pid(leader, tsk, PIDTYPE_PGID);
976 transfer_pid(leader, tsk, PIDTYPE_SID);
978 list_replace_rcu(&leader->tasks, &tsk->tasks);
979 list_replace_init(&leader->sibling, &tsk->sibling);
981 tsk->group_leader = tsk;
982 leader->group_leader = tsk;
984 tsk->exit_signal = SIGCHLD;
985 leader->exit_signal = -1;
987 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
988 leader->exit_state = EXIT_DEAD;
991 * We are going to release_task()->ptrace_unlink() silently,
992 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
993 * the tracer wont't block again waiting for this thread.
995 if (unlikely(leader->ptrace))
996 __wake_up_parent(leader, leader->parent);
997 write_unlock_irq(&tasklist_lock);
998 threadgroup_change_end(tsk);
1000 release_task(leader);
1003 sig->group_exit_task = NULL;
1004 sig->notify_count = 0;
1007 /* we have changed execution domain */
1008 tsk->exit_signal = SIGCHLD;
1011 flush_itimer_signals();
1013 if (atomic_read(&oldsighand->count) != 1) {
1014 struct sighand_struct *newsighand;
1016 * This ->sighand is shared with the CLONE_SIGHAND
1017 * but not CLONE_THREAD task, switch to the new one.
1019 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1023 atomic_set(&newsighand->count, 1);
1024 memcpy(newsighand->action, oldsighand->action,
1025 sizeof(newsighand->action));
1027 write_lock_irq(&tasklist_lock);
1028 spin_lock(&oldsighand->siglock);
1029 rcu_assign_pointer(tsk->sighand, newsighand);
1030 spin_unlock(&oldsighand->siglock);
1031 write_unlock_irq(&tasklist_lock);
1033 __cleanup_sighand(oldsighand);
1036 BUG_ON(!thread_group_leader(tsk));
1040 /* protects against exit_notify() and __exit_signal() */
1041 read_lock(&tasklist_lock);
1042 sig->group_exit_task = NULL;
1043 sig->notify_count = 0;
1044 read_unlock(&tasklist_lock);
1048 char *get_task_comm(char *buf, struct task_struct *tsk)
1050 /* buf must be at least sizeof(tsk->comm) in size */
1052 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1056 EXPORT_SYMBOL_GPL(get_task_comm);
1059 * These functions flushes out all traces of the currently running executable
1060 * so that a new one can be started
1063 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1066 trace_task_rename(tsk, buf);
1067 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1069 perf_event_comm(tsk, exec);
1072 int flush_old_exec(struct linux_binprm * bprm)
1077 * Make sure we have a private signal table and that
1078 * we are unassociated from the previous thread group.
1080 retval = de_thread(current);
1084 set_mm_exe_file(bprm->mm, bprm->file);
1086 * Release all of the old mmap stuff
1088 acct_arg_size(bprm, 0);
1089 retval = exec_mmap(bprm->mm);
1093 bprm->mm = NULL; /* We're using it now */
1096 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1097 PF_NOFREEZE | PF_NO_SETAFFINITY);
1099 current->personality &= ~bprm->per_clear;
1106 EXPORT_SYMBOL(flush_old_exec);
1108 void would_dump(struct linux_binprm *bprm, struct file *file)
1110 if (inode_permission(file_inode(file), MAY_READ) < 0)
1111 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1113 EXPORT_SYMBOL(would_dump);
1115 void setup_new_exec(struct linux_binprm * bprm)
1117 arch_pick_mmap_layout(current->mm);
1119 /* This is the point of no return */
1120 current->sas_ss_sp = current->sas_ss_size = 0;
1122 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1123 set_dumpable(current->mm, SUID_DUMP_USER);
1125 set_dumpable(current->mm, suid_dumpable);
1128 __set_task_comm(current, kbasename(bprm->filename), true);
1130 /* Set the new mm task size. We have to do that late because it may
1131 * depend on TIF_32BIT which is only updated in flush_thread() on
1132 * some architectures like powerpc
1134 current->mm->task_size = TASK_SIZE;
1136 /* install the new credentials */
1137 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1138 !gid_eq(bprm->cred->gid, current_egid())) {
1139 current->pdeath_signal = 0;
1141 would_dump(bprm, bprm->file);
1142 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1143 set_dumpable(current->mm, suid_dumpable);
1146 /* An exec changes our domain. We are no longer part of the thread
1148 current->self_exec_id++;
1149 flush_signal_handlers(current, 0);
1150 do_close_on_exec(current->files);
1152 EXPORT_SYMBOL(setup_new_exec);
1155 * Prepare credentials and lock ->cred_guard_mutex.
1156 * install_exec_creds() commits the new creds and drops the lock.
1157 * Or, if exec fails before, free_bprm() should release ->cred and
1160 int prepare_bprm_creds(struct linux_binprm *bprm)
1162 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1163 return -ERESTARTNOINTR;
1165 bprm->cred = prepare_exec_creds();
1166 if (likely(bprm->cred))
1169 mutex_unlock(¤t->signal->cred_guard_mutex);
1173 static void free_bprm(struct linux_binprm *bprm)
1175 free_arg_pages(bprm);
1177 mutex_unlock(¤t->signal->cred_guard_mutex);
1178 abort_creds(bprm->cred);
1181 allow_write_access(bprm->file);
1184 /* If a binfmt changed the interp, free it. */
1185 if (bprm->interp != bprm->filename)
1186 kfree(bprm->interp);
1190 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1192 /* If a binfmt changed the interp, free it first. */
1193 if (bprm->interp != bprm->filename)
1194 kfree(bprm->interp);
1195 bprm->interp = kstrdup(interp, GFP_KERNEL);
1200 EXPORT_SYMBOL(bprm_change_interp);
1203 * install the new credentials for this executable
1205 void install_exec_creds(struct linux_binprm *bprm)
1207 security_bprm_committing_creds(bprm);
1209 commit_creds(bprm->cred);
1213 * Disable monitoring for regular users
1214 * when executing setuid binaries. Must
1215 * wait until new credentials are committed
1216 * by commit_creds() above
1218 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1219 perf_event_exit_task(current);
1221 * cred_guard_mutex must be held at least to this point to prevent
1222 * ptrace_attach() from altering our determination of the task's
1223 * credentials; any time after this it may be unlocked.
1225 security_bprm_committed_creds(bprm);
1226 mutex_unlock(¤t->signal->cred_guard_mutex);
1228 EXPORT_SYMBOL(install_exec_creds);
1231 * determine how safe it is to execute the proposed program
1232 * - the caller must hold ->cred_guard_mutex to protect against
1233 * PTRACE_ATTACH or seccomp thread-sync
1235 static void check_unsafe_exec(struct linux_binprm *bprm)
1237 struct task_struct *p = current, *t;
1241 if (p->ptrace & PT_PTRACE_CAP)
1242 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1244 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1248 * This isn't strictly necessary, but it makes it harder for LSMs to
1251 if (task_no_new_privs(current))
1252 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1256 spin_lock(&p->fs->lock);
1258 while_each_thread(p, t) {
1264 if (p->fs->users > n_fs)
1265 bprm->unsafe |= LSM_UNSAFE_SHARE;
1268 spin_unlock(&p->fs->lock);
1271 static void bprm_fill_uid(struct linux_binprm *bprm)
1273 struct inode *inode;
1278 /* clear any previous set[ug]id data from a previous binary */
1279 bprm->cred->euid = current_euid();
1280 bprm->cred->egid = current_egid();
1282 if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1285 if (task_no_new_privs(current))
1288 inode = file_inode(bprm->file);
1289 mode = READ_ONCE(inode->i_mode);
1290 if (!(mode & (S_ISUID|S_ISGID)))
1293 /* Be careful if suid/sgid is set */
1294 mutex_lock(&inode->i_mutex);
1296 /* reload atomically mode/uid/gid now that lock held */
1297 mode = inode->i_mode;
1300 mutex_unlock(&inode->i_mutex);
1302 /* We ignore suid/sgid if there are no mappings for them in the ns */
1303 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1304 !kgid_has_mapping(bprm->cred->user_ns, gid))
1307 if (mode & S_ISUID) {
1308 bprm->per_clear |= PER_CLEAR_ON_SETID;
1309 bprm->cred->euid = uid;
1312 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1313 bprm->per_clear |= PER_CLEAR_ON_SETID;
1314 bprm->cred->egid = gid;
1319 * Fill the binprm structure from the inode.
1320 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1322 * This may be called multiple times for binary chains (scripts for example).
1324 int prepare_binprm(struct linux_binprm *bprm)
1328 bprm_fill_uid(bprm);
1330 /* fill in binprm security blob */
1331 retval = security_bprm_set_creds(bprm);
1334 bprm->cred_prepared = 1;
1336 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1337 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1340 EXPORT_SYMBOL(prepare_binprm);
1343 * Arguments are '\0' separated strings found at the location bprm->p
1344 * points to; chop off the first by relocating brpm->p to right after
1345 * the first '\0' encountered.
1347 int remove_arg_zero(struct linux_binprm *bprm)
1350 unsigned long offset;
1358 offset = bprm->p & ~PAGE_MASK;
1359 page = get_arg_page(bprm, bprm->p, 0);
1364 kaddr = kmap_atomic(page);
1366 for (; offset < PAGE_SIZE && kaddr[offset];
1367 offset++, bprm->p++)
1370 kunmap_atomic(kaddr);
1373 if (offset == PAGE_SIZE)
1374 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1375 } while (offset == PAGE_SIZE);
1384 EXPORT_SYMBOL(remove_arg_zero);
1386 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1388 * cycle the list of binary formats handler, until one recognizes the image
1390 int search_binary_handler(struct linux_binprm *bprm)
1392 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1393 struct linux_binfmt *fmt;
1396 /* This allows 4 levels of binfmt rewrites before failing hard. */
1397 if (bprm->recursion_depth > 5)
1400 retval = security_bprm_check(bprm);
1406 read_lock(&binfmt_lock);
1407 list_for_each_entry(fmt, &formats, lh) {
1408 if (!try_module_get(fmt->module))
1410 read_unlock(&binfmt_lock);
1411 bprm->recursion_depth++;
1412 retval = fmt->load_binary(bprm);
1413 read_lock(&binfmt_lock);
1415 bprm->recursion_depth--;
1416 if (retval < 0 && !bprm->mm) {
1417 /* we got to flush_old_exec() and failed after it */
1418 read_unlock(&binfmt_lock);
1419 force_sigsegv(SIGSEGV, current);
1422 if (retval != -ENOEXEC || !bprm->file) {
1423 read_unlock(&binfmt_lock);
1427 read_unlock(&binfmt_lock);
1430 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1431 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1433 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1441 EXPORT_SYMBOL(search_binary_handler);
1443 static int exec_binprm(struct linux_binprm *bprm)
1445 pid_t old_pid, old_vpid;
1448 /* Need to fetch pid before load_binary changes it */
1449 old_pid = current->pid;
1451 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1454 ret = search_binary_handler(bprm);
1457 trace_sched_process_exec(current, old_pid, bprm);
1458 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1459 proc_exec_connector(current);
1466 * sys_execve() executes a new program.
1468 static int do_execveat_common(int fd, struct filename *filename,
1469 struct user_arg_ptr argv,
1470 struct user_arg_ptr envp,
1473 char *pathbuf = NULL;
1474 struct linux_binprm *bprm;
1476 struct files_struct *displaced;
1479 if (IS_ERR(filename))
1480 return PTR_ERR(filename);
1483 * We move the actual failure in case of RLIMIT_NPROC excess from
1484 * set*uid() to execve() because too many poorly written programs
1485 * don't check setuid() return code. Here we additionally recheck
1486 * whether NPROC limit is still exceeded.
1488 if ((current->flags & PF_NPROC_EXCEEDED) &&
1489 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1494 /* We're below the limit (still or again), so we don't want to make
1495 * further execve() calls fail. */
1496 current->flags &= ~PF_NPROC_EXCEEDED;
1498 retval = unshare_files(&displaced);
1503 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1507 retval = prepare_bprm_creds(bprm);
1511 check_unsafe_exec(bprm);
1512 current->in_execve = 1;
1514 file = do_open_execat(fd, filename, flags);
1515 retval = PTR_ERR(file);
1522 if (fd == AT_FDCWD || filename->name[0] == '/') {
1523 bprm->filename = filename->name;
1525 if (filename->name[0] == '\0')
1526 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1528 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1529 fd, filename->name);
1535 * Record that a name derived from an O_CLOEXEC fd will be
1536 * inaccessible after exec. Relies on having exclusive access to
1537 * current->files (due to unshare_files above).
1539 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1540 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1541 bprm->filename = pathbuf;
1543 bprm->interp = bprm->filename;
1545 retval = bprm_mm_init(bprm);
1549 bprm->argc = count(argv, MAX_ARG_STRINGS);
1550 if ((retval = bprm->argc) < 0)
1553 bprm->envc = count(envp, MAX_ARG_STRINGS);
1554 if ((retval = bprm->envc) < 0)
1557 retval = prepare_binprm(bprm);
1561 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1565 bprm->exec = bprm->p;
1566 retval = copy_strings(bprm->envc, envp, bprm);
1570 retval = copy_strings(bprm->argc, argv, bprm);
1574 retval = exec_binprm(bprm);
1578 /* execve succeeded */
1579 current->fs->in_exec = 0;
1580 current->in_execve = 0;
1581 acct_update_integrals(current);
1582 task_numa_free(current);
1587 put_files_struct(displaced);
1592 acct_arg_size(bprm, 0);
1597 current->fs->in_exec = 0;
1598 current->in_execve = 0;
1606 reset_files_struct(displaced);
1612 int do_execve(struct filename *filename,
1613 const char __user *const __user *__argv,
1614 const char __user *const __user *__envp)
1616 struct user_arg_ptr argv = { .ptr.native = __argv };
1617 struct user_arg_ptr envp = { .ptr.native = __envp };
1618 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1621 int do_execveat(int fd, struct filename *filename,
1622 const char __user *const __user *__argv,
1623 const char __user *const __user *__envp,
1626 struct user_arg_ptr argv = { .ptr.native = __argv };
1627 struct user_arg_ptr envp = { .ptr.native = __envp };
1629 return do_execveat_common(fd, filename, argv, envp, flags);
1632 #ifdef CONFIG_COMPAT
1633 static int compat_do_execve(struct filename *filename,
1634 const compat_uptr_t __user *__argv,
1635 const compat_uptr_t __user *__envp)
1637 struct user_arg_ptr argv = {
1639 .ptr.compat = __argv,
1641 struct user_arg_ptr envp = {
1643 .ptr.compat = __envp,
1645 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1648 static int compat_do_execveat(int fd, struct filename *filename,
1649 const compat_uptr_t __user *__argv,
1650 const compat_uptr_t __user *__envp,
1653 struct user_arg_ptr argv = {
1655 .ptr.compat = __argv,
1657 struct user_arg_ptr envp = {
1659 .ptr.compat = __envp,
1661 return do_execveat_common(fd, filename, argv, envp, flags);
1665 void set_binfmt(struct linux_binfmt *new)
1667 struct mm_struct *mm = current->mm;
1670 module_put(mm->binfmt->module);
1674 __module_get(new->module);
1676 EXPORT_SYMBOL(set_binfmt);
1679 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1681 void set_dumpable(struct mm_struct *mm, int value)
1683 unsigned long old, new;
1685 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1689 old = ACCESS_ONCE(mm->flags);
1690 new = (old & ~MMF_DUMPABLE_MASK) | value;
1691 } while (cmpxchg(&mm->flags, old, new) != old);
1694 SYSCALL_DEFINE3(execve,
1695 const char __user *, filename,
1696 const char __user *const __user *, argv,
1697 const char __user *const __user *, envp)
1699 return do_execve(getname(filename), argv, envp);
1702 SYSCALL_DEFINE5(execveat,
1703 int, fd, const char __user *, filename,
1704 const char __user *const __user *, argv,
1705 const char __user *const __user *, envp,
1708 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1710 return do_execveat(fd,
1711 getname_flags(filename, lookup_flags, NULL),
1715 #ifdef CONFIG_COMPAT
1716 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1717 const compat_uptr_t __user *, argv,
1718 const compat_uptr_t __user *, envp)
1720 return compat_do_execve(getname(filename), argv, envp);
1723 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1724 const char __user *, filename,
1725 const compat_uptr_t __user *, argv,
1726 const compat_uptr_t __user *, envp,
1729 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1731 return compat_do_execveat(fd,
1732 getname_flags(filename, lookup_flags, NULL),