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,
194 unsigned int gup_flags = FOLL_FORCE;
196 #ifdef CONFIG_STACK_GROWSUP
198 ret = expand_downwards(bprm->vma, pos);
205 gup_flags |= FOLL_WRITE;
207 ret = get_user_pages(current, bprm->mm, pos, 1, gup_flags,
213 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
214 unsigned long ptr_size, limit;
217 * Since the stack will hold pointers to the strings, we
218 * must account for them as well.
220 * The size calculation is the entire vma while each arg page is
221 * built, so each time we get here it's calculating how far it
222 * is currently (rather than each call being just the newly
223 * added size from the arg page). As a result, we need to
224 * always add the entire size of the pointers, so that on the
225 * last call to get_arg_page() we'll actually have the entire
228 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
229 if (ptr_size > ULONG_MAX - size)
233 acct_arg_size(bprm, size / PAGE_SIZE);
236 * We've historically supported up to 32 pages (ARG_MAX)
237 * of argument strings even with small stacks
243 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
244 * (whichever is smaller) for the argv+env strings.
246 * - the remaining binfmt code will not run out of stack space,
247 * - the program will have a reasonable amount of stack left
250 limit = _STK_LIM / 4 * 3;
251 limit = min(limit, rlimit(RLIMIT_STACK) / 4);
263 static void put_arg_page(struct page *page)
268 static void free_arg_page(struct linux_binprm *bprm, int i)
272 static void free_arg_pages(struct linux_binprm *bprm)
276 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
279 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
282 static int __bprm_mm_init(struct linux_binprm *bprm)
285 struct vm_area_struct *vma = NULL;
286 struct mm_struct *mm = bprm->mm;
288 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
292 down_write(&mm->mmap_sem);
296 * Place the stack at the largest stack address the architecture
297 * supports. Later, we'll move this to an appropriate place. We don't
298 * use STACK_TOP because that can depend on attributes which aren't
301 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
302 vma->vm_end = STACK_TOP_MAX;
303 vma->vm_start = vma->vm_end - PAGE_SIZE;
304 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
305 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
306 INIT_LIST_HEAD(&vma->anon_vma_chain);
308 err = insert_vm_struct(mm, vma);
312 mm->stack_vm = mm->total_vm = 1;
313 arch_bprm_mm_init(mm, vma);
314 up_write(&mm->mmap_sem);
315 bprm->p = vma->vm_end - sizeof(void *);
318 up_write(&mm->mmap_sem);
320 kmem_cache_free(vm_area_cachep, vma);
324 static bool valid_arg_len(struct linux_binprm *bprm, long len)
326 return len <= MAX_ARG_STRLEN;
331 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
335 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
340 page = bprm->page[pos / PAGE_SIZE];
341 if (!page && write) {
342 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
345 bprm->page[pos / PAGE_SIZE] = page;
351 static void put_arg_page(struct page *page)
355 static void free_arg_page(struct linux_binprm *bprm, int i)
358 __free_page(bprm->page[i]);
359 bprm->page[i] = NULL;
363 static void free_arg_pages(struct linux_binprm *bprm)
367 for (i = 0; i < MAX_ARG_PAGES; i++)
368 free_arg_page(bprm, i);
371 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
376 static int __bprm_mm_init(struct linux_binprm *bprm)
378 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
382 static bool valid_arg_len(struct linux_binprm *bprm, long len)
384 return len <= bprm->p;
387 #endif /* CONFIG_MMU */
390 * Create a new mm_struct and populate it with a temporary stack
391 * vm_area_struct. We don't have enough context at this point to set the stack
392 * flags, permissions, and offset, so we use temporary values. We'll update
393 * them later in setup_arg_pages().
395 static int bprm_mm_init(struct linux_binprm *bprm)
398 struct mm_struct *mm = NULL;
400 bprm->mm = mm = mm_alloc();
405 err = __bprm_mm_init(bprm);
420 struct user_arg_ptr {
425 const char __user *const __user *native;
427 const compat_uptr_t __user *compat;
432 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
434 const char __user *native;
437 if (unlikely(argv.is_compat)) {
438 compat_uptr_t compat;
440 if (get_user(compat, argv.ptr.compat + nr))
441 return ERR_PTR(-EFAULT);
443 return compat_ptr(compat);
447 if (get_user(native, argv.ptr.native + nr))
448 return ERR_PTR(-EFAULT);
454 * count() counts the number of strings in array ARGV.
456 static int count(struct user_arg_ptr argv, int max)
460 if (argv.ptr.native != NULL) {
462 const char __user *p = get_user_arg_ptr(argv, i);
474 if (fatal_signal_pending(current))
475 return -ERESTARTNOHAND;
483 * 'copy_strings()' copies argument/environment strings from the old
484 * processes's memory to the new process's stack. The call to get_user_pages()
485 * ensures the destination page is created and not swapped out.
487 static int copy_strings(int argc, struct user_arg_ptr argv,
488 struct linux_binprm *bprm)
490 struct page *kmapped_page = NULL;
492 unsigned long kpos = 0;
496 const char __user *str;
501 str = get_user_arg_ptr(argv, argc);
505 len = strnlen_user(str, MAX_ARG_STRLEN);
510 if (!valid_arg_len(bprm, len))
513 /* We're going to work our way backwords. */
519 int offset, bytes_to_copy;
521 if (fatal_signal_pending(current)) {
522 ret = -ERESTARTNOHAND;
527 offset = pos % PAGE_SIZE;
531 bytes_to_copy = offset;
532 if (bytes_to_copy > len)
535 offset -= bytes_to_copy;
536 pos -= bytes_to_copy;
537 str -= bytes_to_copy;
538 len -= bytes_to_copy;
540 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
543 page = get_arg_page(bprm, pos, 1);
550 flush_kernel_dcache_page(kmapped_page);
551 kunmap(kmapped_page);
552 put_arg_page(kmapped_page);
555 kaddr = kmap(kmapped_page);
556 kpos = pos & PAGE_MASK;
557 flush_arg_page(bprm, kpos, kmapped_page);
559 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
568 flush_kernel_dcache_page(kmapped_page);
569 kunmap(kmapped_page);
570 put_arg_page(kmapped_page);
576 * Like copy_strings, but get argv and its values from kernel memory.
578 int copy_strings_kernel(int argc, const char *const *__argv,
579 struct linux_binprm *bprm)
582 mm_segment_t oldfs = get_fs();
583 struct user_arg_ptr argv = {
584 .ptr.native = (const char __user *const __user *)__argv,
588 r = copy_strings(argc, argv, bprm);
593 EXPORT_SYMBOL(copy_strings_kernel);
598 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
599 * the binfmt code determines where the new stack should reside, we shift it to
600 * its final location. The process proceeds as follows:
602 * 1) Use shift to calculate the new vma endpoints.
603 * 2) Extend vma to cover both the old and new ranges. This ensures the
604 * arguments passed to subsequent functions are consistent.
605 * 3) Move vma's page tables to the new range.
606 * 4) Free up any cleared pgd range.
607 * 5) Shrink the vma to cover only the new range.
609 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
611 struct mm_struct *mm = vma->vm_mm;
612 unsigned long old_start = vma->vm_start;
613 unsigned long old_end = vma->vm_end;
614 unsigned long length = old_end - old_start;
615 unsigned long new_start = old_start - shift;
616 unsigned long new_end = old_end - shift;
617 struct mmu_gather tlb;
619 BUG_ON(new_start > new_end);
622 * ensure there are no vmas between where we want to go
625 if (vma != find_vma(mm, new_start))
629 * cover the whole range: [new_start, old_end)
631 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
635 * move the page tables downwards, on failure we rely on
636 * process cleanup to remove whatever mess we made.
638 if (length != move_page_tables(vma, old_start,
639 vma, new_start, length, false))
643 tlb_gather_mmu(&tlb, mm, old_start, old_end);
644 if (new_end > old_start) {
646 * when the old and new regions overlap clear from new_end.
648 free_pgd_range(&tlb, new_end, old_end, new_end,
649 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
652 * otherwise, clean from old_start; this is done to not touch
653 * the address space in [new_end, old_start) some architectures
654 * have constraints on va-space that make this illegal (IA64) -
655 * for the others its just a little faster.
657 free_pgd_range(&tlb, old_start, old_end, new_end,
658 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
660 tlb_finish_mmu(&tlb, old_start, old_end);
663 * Shrink the vma to just the new range. Always succeeds.
665 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
671 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
672 * the stack is optionally relocated, and some extra space is added.
674 int setup_arg_pages(struct linux_binprm *bprm,
675 unsigned long stack_top,
676 int executable_stack)
679 unsigned long stack_shift;
680 struct mm_struct *mm = current->mm;
681 struct vm_area_struct *vma = bprm->vma;
682 struct vm_area_struct *prev = NULL;
683 unsigned long vm_flags;
684 unsigned long stack_base;
685 unsigned long stack_size;
686 unsigned long stack_expand;
687 unsigned long rlim_stack;
689 #ifdef CONFIG_STACK_GROWSUP
690 /* Limit stack size */
691 stack_base = rlimit_max(RLIMIT_STACK);
692 if (stack_base > STACK_SIZE_MAX)
693 stack_base = STACK_SIZE_MAX;
695 /* Add space for stack randomization. */
696 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
698 /* Make sure we didn't let the argument array grow too large. */
699 if (vma->vm_end - vma->vm_start > stack_base)
702 stack_base = PAGE_ALIGN(stack_top - stack_base);
704 stack_shift = vma->vm_start - stack_base;
705 mm->arg_start = bprm->p - stack_shift;
706 bprm->p = vma->vm_end - stack_shift;
708 stack_top = arch_align_stack(stack_top);
709 stack_top = PAGE_ALIGN(stack_top);
711 if (unlikely(stack_top < mmap_min_addr) ||
712 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
715 stack_shift = vma->vm_end - stack_top;
717 bprm->p -= stack_shift;
718 mm->arg_start = bprm->p;
722 bprm->loader -= stack_shift;
723 bprm->exec -= stack_shift;
725 down_write(&mm->mmap_sem);
726 vm_flags = VM_STACK_FLAGS;
729 * Adjust stack execute permissions; explicitly enable for
730 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
731 * (arch default) otherwise.
733 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
735 else if (executable_stack == EXSTACK_DISABLE_X)
736 vm_flags &= ~VM_EXEC;
737 vm_flags |= mm->def_flags;
738 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
740 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
746 /* Move stack pages down in memory. */
748 ret = shift_arg_pages(vma, stack_shift);
753 /* mprotect_fixup is overkill to remove the temporary stack flags */
754 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
756 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
757 stack_size = vma->vm_end - vma->vm_start;
759 * Align this down to a page boundary as expand_stack
762 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
763 #ifdef CONFIG_STACK_GROWSUP
764 if (stack_size + stack_expand > rlim_stack)
765 stack_base = vma->vm_start + rlim_stack;
767 stack_base = vma->vm_end + stack_expand;
769 if (stack_size + stack_expand > rlim_stack)
770 stack_base = vma->vm_end - rlim_stack;
772 stack_base = vma->vm_start - stack_expand;
774 current->mm->start_stack = bprm->p;
775 ret = expand_stack(vma, stack_base);
780 up_write(&mm->mmap_sem);
783 EXPORT_SYMBOL(setup_arg_pages);
785 #endif /* CONFIG_MMU */
787 static struct file *do_open_execat(int fd, struct filename *name, int flags)
791 struct open_flags open_exec_flags = {
792 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
793 .acc_mode = MAY_EXEC | MAY_OPEN,
794 .intent = LOOKUP_OPEN,
795 .lookup_flags = LOOKUP_FOLLOW,
798 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
799 return ERR_PTR(-EINVAL);
800 if (flags & AT_SYMLINK_NOFOLLOW)
801 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
802 if (flags & AT_EMPTY_PATH)
803 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
805 file = do_filp_open(fd, name, &open_exec_flags);
810 if (!S_ISREG(file_inode(file)->i_mode))
813 if (path_noexec(&file->f_path))
816 err = deny_write_access(file);
820 if (name->name[0] != '\0')
831 struct file *open_exec(const char *name)
833 struct filename *filename = getname_kernel(name);
834 struct file *f = ERR_CAST(filename);
836 if (!IS_ERR(filename)) {
837 f = do_open_execat(AT_FDCWD, filename, 0);
842 EXPORT_SYMBOL(open_exec);
844 int kernel_read(struct file *file, loff_t offset,
845 char *addr, unsigned long count)
853 /* The cast to a user pointer is valid due to the set_fs() */
854 result = vfs_read(file, (void __user *)addr, count, &pos);
859 EXPORT_SYMBOL(kernel_read);
861 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
863 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
865 flush_icache_range(addr, addr + len);
868 EXPORT_SYMBOL(read_code);
870 static int exec_mmap(struct mm_struct *mm)
872 struct task_struct *tsk;
873 struct mm_struct *old_mm, *active_mm;
875 /* Notify parent that we're no longer interested in the old VM */
877 old_mm = current->mm;
878 mm_release(tsk, old_mm);
883 * Make sure that if there is a core dump in progress
884 * for the old mm, we get out and die instead of going
885 * through with the exec. We must hold mmap_sem around
886 * checking core_state and changing tsk->mm.
888 down_read(&old_mm->mmap_sem);
889 if (unlikely(old_mm->core_state)) {
890 up_read(&old_mm->mmap_sem);
895 active_mm = tsk->active_mm;
898 activate_mm(active_mm, mm);
899 tsk->mm->vmacache_seqnum = 0;
903 up_read(&old_mm->mmap_sem);
904 BUG_ON(active_mm != old_mm);
905 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
906 mm_update_next_owner(old_mm);
915 * This function makes sure the current process has its own signal table,
916 * so that flush_signal_handlers can later reset the handlers without
917 * disturbing other processes. (Other processes might share the signal
918 * table via the CLONE_SIGHAND option to clone().)
920 static int de_thread(struct task_struct *tsk)
922 struct signal_struct *sig = tsk->signal;
923 struct sighand_struct *oldsighand = tsk->sighand;
924 spinlock_t *lock = &oldsighand->siglock;
926 if (thread_group_empty(tsk))
927 goto no_thread_group;
930 * Kill all other threads in the thread group.
933 if (signal_group_exit(sig)) {
935 * Another group action in progress, just
936 * return so that the signal is processed.
938 spin_unlock_irq(lock);
942 sig->group_exit_task = tsk;
943 sig->notify_count = zap_other_threads(tsk);
944 if (!thread_group_leader(tsk))
947 while (sig->notify_count) {
948 __set_current_state(TASK_KILLABLE);
949 spin_unlock_irq(lock);
951 if (unlikely(__fatal_signal_pending(tsk)))
955 spin_unlock_irq(lock);
958 * At this point all other threads have exited, all we have to
959 * do is to wait for the thread group leader to become inactive,
960 * and to assume its PID:
962 if (!thread_group_leader(tsk)) {
963 struct task_struct *leader = tsk->group_leader;
966 threadgroup_change_begin(tsk);
967 write_lock_irq(&tasklist_lock);
969 * Do this under tasklist_lock to ensure that
970 * exit_notify() can't miss ->group_exit_task
972 sig->notify_count = -1;
973 if (likely(leader->exit_state))
975 __set_current_state(TASK_KILLABLE);
976 write_unlock_irq(&tasklist_lock);
977 threadgroup_change_end(tsk);
979 if (unlikely(__fatal_signal_pending(tsk)))
984 * The only record we have of the real-time age of a
985 * process, regardless of execs it's done, is start_time.
986 * All the past CPU time is accumulated in signal_struct
987 * from sister threads now dead. But in this non-leader
988 * exec, nothing survives from the original leader thread,
989 * whose birth marks the true age of this process now.
990 * When we take on its identity by switching to its PID, we
991 * also take its birthdate (always earlier than our own).
993 tsk->start_time = leader->start_time;
994 tsk->real_start_time = leader->real_start_time;
996 BUG_ON(!same_thread_group(leader, tsk));
997 BUG_ON(has_group_leader_pid(tsk));
999 * An exec() starts a new thread group with the
1000 * TGID of the previous thread group. Rehash the
1001 * two threads with a switched PID, and release
1002 * the former thread group leader:
1005 /* Become a process group leader with the old leader's pid.
1006 * The old leader becomes a thread of the this thread group.
1007 * Note: The old leader also uses this pid until release_task
1008 * is called. Odd but simple and correct.
1010 tsk->pid = leader->pid;
1011 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1012 transfer_pid(leader, tsk, PIDTYPE_PGID);
1013 transfer_pid(leader, tsk, PIDTYPE_SID);
1015 list_replace_rcu(&leader->tasks, &tsk->tasks);
1016 list_replace_init(&leader->sibling, &tsk->sibling);
1018 tsk->group_leader = tsk;
1019 leader->group_leader = tsk;
1021 tsk->exit_signal = SIGCHLD;
1022 leader->exit_signal = -1;
1024 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1025 leader->exit_state = EXIT_DEAD;
1028 * We are going to release_task()->ptrace_unlink() silently,
1029 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1030 * the tracer wont't block again waiting for this thread.
1032 if (unlikely(leader->ptrace))
1033 __wake_up_parent(leader, leader->parent);
1034 write_unlock_irq(&tasklist_lock);
1035 threadgroup_change_end(tsk);
1037 release_task(leader);
1040 sig->group_exit_task = NULL;
1041 sig->notify_count = 0;
1044 /* we have changed execution domain */
1045 tsk->exit_signal = SIGCHLD;
1048 flush_itimer_signals();
1050 if (atomic_read(&oldsighand->count) != 1) {
1051 struct sighand_struct *newsighand;
1053 * This ->sighand is shared with the CLONE_SIGHAND
1054 * but not CLONE_THREAD task, switch to the new one.
1056 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1060 atomic_set(&newsighand->count, 1);
1061 memcpy(newsighand->action, oldsighand->action,
1062 sizeof(newsighand->action));
1064 write_lock_irq(&tasklist_lock);
1065 spin_lock(&oldsighand->siglock);
1066 rcu_assign_pointer(tsk->sighand, newsighand);
1067 spin_unlock(&oldsighand->siglock);
1068 write_unlock_irq(&tasklist_lock);
1070 __cleanup_sighand(oldsighand);
1073 BUG_ON(!thread_group_leader(tsk));
1077 /* protects against exit_notify() and __exit_signal() */
1078 read_lock(&tasklist_lock);
1079 sig->group_exit_task = NULL;
1080 sig->notify_count = 0;
1081 read_unlock(&tasklist_lock);
1085 char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1088 strncpy(buf, tsk->comm, buf_size);
1092 EXPORT_SYMBOL_GPL(__get_task_comm);
1095 * These functions flushes out all traces of the currently running executable
1096 * so that a new one can be started
1099 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1102 trace_task_rename(tsk, buf);
1103 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1105 perf_event_comm(tsk, exec);
1108 int flush_old_exec(struct linux_binprm * bprm)
1113 * Make sure we have a private signal table and that
1114 * we are unassociated from the previous thread group.
1116 retval = de_thread(current);
1121 * Must be called _before_ exec_mmap() as bprm->mm is
1122 * not visibile until then. This also enables the update
1125 set_mm_exe_file(bprm->mm, bprm->file);
1128 * Release all of the old mmap stuff
1130 acct_arg_size(bprm, 0);
1131 retval = exec_mmap(bprm->mm);
1135 bprm->mm = NULL; /* We're using it now */
1138 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1139 PF_NOFREEZE | PF_NO_SETAFFINITY);
1141 current->personality &= ~bprm->per_clear;
1144 * We have to apply CLOEXEC before we change whether the process is
1145 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1146 * trying to access the should-be-closed file descriptors of a process
1147 * undergoing exec(2).
1149 do_close_on_exec(current->files);
1155 EXPORT_SYMBOL(flush_old_exec);
1157 void would_dump(struct linux_binprm *bprm, struct file *file)
1159 struct inode *inode = file_inode(file);
1160 if (inode_permission(inode, MAY_READ) < 0) {
1161 struct user_namespace *old, *user_ns;
1162 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1164 /* Ensure mm->user_ns contains the executable */
1165 user_ns = old = bprm->mm->user_ns;
1166 while ((user_ns != &init_user_ns) &&
1167 !privileged_wrt_inode_uidgid(user_ns, inode))
1168 user_ns = user_ns->parent;
1170 if (old != user_ns) {
1171 bprm->mm->user_ns = get_user_ns(user_ns);
1176 EXPORT_SYMBOL(would_dump);
1178 void setup_new_exec(struct linux_binprm * bprm)
1180 arch_pick_mmap_layout(current->mm);
1182 /* This is the point of no return */
1183 current->sas_ss_sp = current->sas_ss_size = 0;
1185 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1186 set_dumpable(current->mm, SUID_DUMP_USER);
1188 set_dumpable(current->mm, suid_dumpable);
1191 __set_task_comm(current, kbasename(bprm->filename), true);
1193 /* Set the new mm task size. We have to do that late because it may
1194 * depend on TIF_32BIT which is only updated in flush_thread() on
1195 * some architectures like powerpc
1197 current->mm->task_size = TASK_SIZE;
1199 /* install the new credentials */
1200 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1201 !gid_eq(bprm->cred->gid, current_egid())) {
1202 current->pdeath_signal = 0;
1204 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1205 set_dumpable(current->mm, suid_dumpable);
1208 /* An exec changes our domain. We are no longer part of the thread
1210 current->self_exec_id++;
1211 flush_signal_handlers(current, 0);
1213 EXPORT_SYMBOL(setup_new_exec);
1216 * Prepare credentials and lock ->cred_guard_mutex.
1217 * install_exec_creds() commits the new creds and drops the lock.
1218 * Or, if exec fails before, free_bprm() should release ->cred and
1221 int prepare_bprm_creds(struct linux_binprm *bprm)
1223 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1224 return -ERESTARTNOINTR;
1226 bprm->cred = prepare_exec_creds();
1227 if (likely(bprm->cred))
1230 mutex_unlock(¤t->signal->cred_guard_mutex);
1234 static void free_bprm(struct linux_binprm *bprm)
1236 free_arg_pages(bprm);
1238 mutex_unlock(¤t->signal->cred_guard_mutex);
1239 abort_creds(bprm->cred);
1242 allow_write_access(bprm->file);
1245 /* If a binfmt changed the interp, free it. */
1246 if (bprm->interp != bprm->filename)
1247 kfree(bprm->interp);
1251 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1253 /* If a binfmt changed the interp, free it first. */
1254 if (bprm->interp != bprm->filename)
1255 kfree(bprm->interp);
1256 bprm->interp = kstrdup(interp, GFP_KERNEL);
1261 EXPORT_SYMBOL(bprm_change_interp);
1264 * install the new credentials for this executable
1266 void install_exec_creds(struct linux_binprm *bprm)
1268 security_bprm_committing_creds(bprm);
1270 commit_creds(bprm->cred);
1274 * Disable monitoring for regular users
1275 * when executing setuid binaries. Must
1276 * wait until new credentials are committed
1277 * by commit_creds() above
1279 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1280 perf_event_exit_task(current);
1282 * cred_guard_mutex must be held at least to this point to prevent
1283 * ptrace_attach() from altering our determination of the task's
1284 * credentials; any time after this it may be unlocked.
1286 security_bprm_committed_creds(bprm);
1287 mutex_unlock(¤t->signal->cred_guard_mutex);
1289 EXPORT_SYMBOL(install_exec_creds);
1292 * determine how safe it is to execute the proposed program
1293 * - the caller must hold ->cred_guard_mutex to protect against
1294 * PTRACE_ATTACH or seccomp thread-sync
1296 static void check_unsafe_exec(struct linux_binprm *bprm)
1298 struct task_struct *p = current, *t;
1302 if (ptracer_capable(p, current_user_ns()))
1303 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1305 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1309 * This isn't strictly necessary, but it makes it harder for LSMs to
1312 if (task_no_new_privs(current))
1313 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1317 spin_lock(&p->fs->lock);
1319 while_each_thread(p, t) {
1325 if (p->fs->users > n_fs)
1326 bprm->unsafe |= LSM_UNSAFE_SHARE;
1329 spin_unlock(&p->fs->lock);
1332 static void bprm_fill_uid(struct linux_binprm *bprm)
1334 struct inode *inode;
1339 /* clear any previous set[ug]id data from a previous binary */
1340 bprm->cred->euid = current_euid();
1341 bprm->cred->egid = current_egid();
1343 if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1346 if (task_no_new_privs(current))
1349 inode = file_inode(bprm->file);
1350 mode = READ_ONCE(inode->i_mode);
1351 if (!(mode & (S_ISUID|S_ISGID)))
1354 /* Be careful if suid/sgid is set */
1355 mutex_lock(&inode->i_mutex);
1357 /* reload atomically mode/uid/gid now that lock held */
1358 mode = inode->i_mode;
1361 mutex_unlock(&inode->i_mutex);
1363 /* We ignore suid/sgid if there are no mappings for them in the ns */
1364 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1365 !kgid_has_mapping(bprm->cred->user_ns, gid))
1368 if (mode & S_ISUID) {
1369 bprm->per_clear |= PER_CLEAR_ON_SETID;
1370 bprm->cred->euid = uid;
1373 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1374 bprm->per_clear |= PER_CLEAR_ON_SETID;
1375 bprm->cred->egid = gid;
1380 * Fill the binprm structure from the inode.
1381 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1383 * This may be called multiple times for binary chains (scripts for example).
1385 int prepare_binprm(struct linux_binprm *bprm)
1389 bprm_fill_uid(bprm);
1391 /* fill in binprm security blob */
1392 retval = security_bprm_set_creds(bprm);
1395 bprm->cred_prepared = 1;
1397 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1398 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1401 EXPORT_SYMBOL(prepare_binprm);
1404 * Arguments are '\0' separated strings found at the location bprm->p
1405 * points to; chop off the first by relocating brpm->p to right after
1406 * the first '\0' encountered.
1408 int remove_arg_zero(struct linux_binprm *bprm)
1411 unsigned long offset;
1419 offset = bprm->p & ~PAGE_MASK;
1420 page = get_arg_page(bprm, bprm->p, 0);
1425 kaddr = kmap_atomic(page);
1427 for (; offset < PAGE_SIZE && kaddr[offset];
1428 offset++, bprm->p++)
1431 kunmap_atomic(kaddr);
1434 if (offset == PAGE_SIZE)
1435 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1436 } while (offset == PAGE_SIZE);
1445 EXPORT_SYMBOL(remove_arg_zero);
1447 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1449 * cycle the list of binary formats handler, until one recognizes the image
1451 int search_binary_handler(struct linux_binprm *bprm)
1453 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1454 struct linux_binfmt *fmt;
1457 /* This allows 4 levels of binfmt rewrites before failing hard. */
1458 if (bprm->recursion_depth > 5)
1461 retval = security_bprm_check(bprm);
1467 read_lock(&binfmt_lock);
1468 list_for_each_entry(fmt, &formats, lh) {
1469 if (!try_module_get(fmt->module))
1471 read_unlock(&binfmt_lock);
1472 bprm->recursion_depth++;
1473 retval = fmt->load_binary(bprm);
1474 read_lock(&binfmt_lock);
1476 bprm->recursion_depth--;
1477 if (retval < 0 && !bprm->mm) {
1478 /* we got to flush_old_exec() and failed after it */
1479 read_unlock(&binfmt_lock);
1480 force_sigsegv(SIGSEGV, current);
1483 if (retval != -ENOEXEC || !bprm->file) {
1484 read_unlock(&binfmt_lock);
1488 read_unlock(&binfmt_lock);
1491 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1492 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1494 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1502 EXPORT_SYMBOL(search_binary_handler);
1504 static int exec_binprm(struct linux_binprm *bprm)
1506 pid_t old_pid, old_vpid;
1509 /* Need to fetch pid before load_binary changes it */
1510 old_pid = current->pid;
1512 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1515 ret = search_binary_handler(bprm);
1518 trace_sched_process_exec(current, old_pid, bprm);
1519 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1520 proc_exec_connector(current);
1527 * sys_execve() executes a new program.
1529 static int do_execveat_common(int fd, struct filename *filename,
1530 struct user_arg_ptr argv,
1531 struct user_arg_ptr envp,
1534 char *pathbuf = NULL;
1535 struct linux_binprm *bprm;
1537 struct files_struct *displaced;
1540 if (IS_ERR(filename))
1541 return PTR_ERR(filename);
1544 * We move the actual failure in case of RLIMIT_NPROC excess from
1545 * set*uid() to execve() because too many poorly written programs
1546 * don't check setuid() return code. Here we additionally recheck
1547 * whether NPROC limit is still exceeded.
1549 if ((current->flags & PF_NPROC_EXCEEDED) &&
1550 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1555 /* We're below the limit (still or again), so we don't want to make
1556 * further execve() calls fail. */
1557 current->flags &= ~PF_NPROC_EXCEEDED;
1559 retval = unshare_files(&displaced);
1564 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1568 retval = prepare_bprm_creds(bprm);
1572 check_unsafe_exec(bprm);
1573 current->in_execve = 1;
1575 file = do_open_execat(fd, filename, flags);
1576 retval = PTR_ERR(file);
1583 if (fd == AT_FDCWD || filename->name[0] == '/') {
1584 bprm->filename = filename->name;
1586 if (filename->name[0] == '\0')
1587 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1589 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1590 fd, filename->name);
1596 * Record that a name derived from an O_CLOEXEC fd will be
1597 * inaccessible after exec. Relies on having exclusive access to
1598 * current->files (due to unshare_files above).
1600 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1601 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1602 bprm->filename = pathbuf;
1604 bprm->interp = bprm->filename;
1606 retval = bprm_mm_init(bprm);
1610 bprm->argc = count(argv, MAX_ARG_STRINGS);
1611 if ((retval = bprm->argc) < 0)
1614 bprm->envc = count(envp, MAX_ARG_STRINGS);
1615 if ((retval = bprm->envc) < 0)
1618 retval = prepare_binprm(bprm);
1622 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1626 bprm->exec = bprm->p;
1627 retval = copy_strings(bprm->envc, envp, bprm);
1631 retval = copy_strings(bprm->argc, argv, bprm);
1635 would_dump(bprm, bprm->file);
1637 retval = exec_binprm(bprm);
1641 /* execve succeeded */
1642 current->fs->in_exec = 0;
1643 current->in_execve = 0;
1644 acct_update_integrals(current);
1645 task_numa_free(current, false);
1650 put_files_struct(displaced);
1655 acct_arg_size(bprm, 0);
1660 current->fs->in_exec = 0;
1661 current->in_execve = 0;
1669 reset_files_struct(displaced);
1675 int do_execve(struct filename *filename,
1676 const char __user *const __user *__argv,
1677 const char __user *const __user *__envp)
1679 struct user_arg_ptr argv = { .ptr.native = __argv };
1680 struct user_arg_ptr envp = { .ptr.native = __envp };
1681 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1684 int do_execveat(int fd, struct filename *filename,
1685 const char __user *const __user *__argv,
1686 const char __user *const __user *__envp,
1689 struct user_arg_ptr argv = { .ptr.native = __argv };
1690 struct user_arg_ptr envp = { .ptr.native = __envp };
1692 return do_execveat_common(fd, filename, argv, envp, flags);
1695 #ifdef CONFIG_COMPAT
1696 static int compat_do_execve(struct filename *filename,
1697 const compat_uptr_t __user *__argv,
1698 const compat_uptr_t __user *__envp)
1700 struct user_arg_ptr argv = {
1702 .ptr.compat = __argv,
1704 struct user_arg_ptr envp = {
1706 .ptr.compat = __envp,
1708 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1711 static int compat_do_execveat(int fd, struct filename *filename,
1712 const compat_uptr_t __user *__argv,
1713 const compat_uptr_t __user *__envp,
1716 struct user_arg_ptr argv = {
1718 .ptr.compat = __argv,
1720 struct user_arg_ptr envp = {
1722 .ptr.compat = __envp,
1724 return do_execveat_common(fd, filename, argv, envp, flags);
1728 void set_binfmt(struct linux_binfmt *new)
1730 struct mm_struct *mm = current->mm;
1733 module_put(mm->binfmt->module);
1737 __module_get(new->module);
1739 EXPORT_SYMBOL(set_binfmt);
1742 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1744 void set_dumpable(struct mm_struct *mm, int value)
1746 unsigned long old, new;
1748 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1752 old = ACCESS_ONCE(mm->flags);
1753 new = (old & ~MMF_DUMPABLE_MASK) | value;
1754 } while (cmpxchg(&mm->flags, old, new) != old);
1757 SYSCALL_DEFINE3(execve,
1758 const char __user *, filename,
1759 const char __user *const __user *, argv,
1760 const char __user *const __user *, envp)
1762 return do_execve(getname(filename), argv, envp);
1765 SYSCALL_DEFINE5(execveat,
1766 int, fd, const char __user *, filename,
1767 const char __user *const __user *, argv,
1768 const char __user *const __user *, envp,
1771 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1773 return do_execveat(fd,
1774 getname_flags(filename, lookup_flags, NULL),
1778 #ifdef CONFIG_COMPAT
1779 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1780 const compat_uptr_t __user *, argv,
1781 const compat_uptr_t __user *, envp)
1783 return compat_do_execve(getname(filename), argv, envp);
1786 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1787 const char __user *, filename,
1788 const compat_uptr_t __user *, argv,
1789 const compat_uptr_t __user *, envp,
1792 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1794 return compat_do_execveat(fd,
1795 getname_flags(filename, lookup_flags, NULL),