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);
101 bool path_noexec(const struct path *path)
103 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
104 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
109 * Note that a shared library must be both readable and executable due to
112 * Also note that we take the address to load from from the file itself.
114 SYSCALL_DEFINE1(uselib, const char __user *, library)
116 struct linux_binfmt *fmt;
118 struct filename *tmp = getname(library);
119 int error = PTR_ERR(tmp);
120 static const struct open_flags uselib_flags = {
121 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
122 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
123 .intent = LOOKUP_OPEN,
124 .lookup_flags = LOOKUP_FOLLOW,
130 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
132 error = PTR_ERR(file);
137 if (!S_ISREG(file_inode(file)->i_mode))
141 if (path_noexec(&file->f_path))
148 read_lock(&binfmt_lock);
149 list_for_each_entry(fmt, &formats, lh) {
150 if (!fmt->load_shlib)
152 if (!try_module_get(fmt->module))
154 read_unlock(&binfmt_lock);
155 error = fmt->load_shlib(file);
156 read_lock(&binfmt_lock);
158 if (error != -ENOEXEC)
161 read_unlock(&binfmt_lock);
167 #endif /* #ifdef CONFIG_USELIB */
171 * The nascent bprm->mm is not visible until exec_mmap() but it can
172 * use a lot of memory, account these pages in current->mm temporary
173 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
174 * change the counter back via acct_arg_size(0).
176 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
178 struct mm_struct *mm = current->mm;
179 long diff = (long)(pages - bprm->vma_pages);
184 bprm->vma_pages = pages;
185 add_mm_counter(mm, MM_ANONPAGES, diff);
188 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
194 #ifdef CONFIG_STACK_GROWSUP
196 ret = expand_downwards(bprm->vma, pos);
201 ret = get_user_pages(current, bprm->mm, pos,
202 1, write, 1, &page, NULL);
207 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
210 acct_arg_size(bprm, size / PAGE_SIZE);
213 * We've historically supported up to 32 pages (ARG_MAX)
214 * of argument strings even with small stacks
220 * Limit to 1/4-th the stack size for the argv+env strings.
222 * - the remaining binfmt code will not run out of stack space,
223 * - the program will have a reasonable amount of stack left
226 rlim = current->signal->rlim;
227 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
236 static void put_arg_page(struct page *page)
241 static void free_arg_page(struct linux_binprm *bprm, int i)
245 static void free_arg_pages(struct linux_binprm *bprm)
249 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
252 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
255 static int __bprm_mm_init(struct linux_binprm *bprm)
258 struct vm_area_struct *vma = NULL;
259 struct mm_struct *mm = bprm->mm;
261 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
265 down_write(&mm->mmap_sem);
269 * Place the stack at the largest stack address the architecture
270 * supports. Later, we'll move this to an appropriate place. We don't
271 * use STACK_TOP because that can depend on attributes which aren't
274 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
275 vma->vm_end = STACK_TOP_MAX;
276 vma->vm_start = vma->vm_end - PAGE_SIZE;
277 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
278 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
279 INIT_LIST_HEAD(&vma->anon_vma_chain);
281 err = insert_vm_struct(mm, vma);
285 mm->stack_vm = mm->total_vm = 1;
286 arch_bprm_mm_init(mm, vma);
287 up_write(&mm->mmap_sem);
288 bprm->p = vma->vm_end - sizeof(void *);
291 up_write(&mm->mmap_sem);
293 kmem_cache_free(vm_area_cachep, vma);
297 static bool valid_arg_len(struct linux_binprm *bprm, long len)
299 return len <= MAX_ARG_STRLEN;
304 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
308 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
313 page = bprm->page[pos / PAGE_SIZE];
314 if (!page && write) {
315 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
318 bprm->page[pos / PAGE_SIZE] = page;
324 static void put_arg_page(struct page *page)
328 static void free_arg_page(struct linux_binprm *bprm, int i)
331 __free_page(bprm->page[i]);
332 bprm->page[i] = NULL;
336 static void free_arg_pages(struct linux_binprm *bprm)
340 for (i = 0; i < MAX_ARG_PAGES; i++)
341 free_arg_page(bprm, i);
344 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
349 static int __bprm_mm_init(struct linux_binprm *bprm)
351 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
355 static bool valid_arg_len(struct linux_binprm *bprm, long len)
357 return len <= bprm->p;
360 #endif /* CONFIG_MMU */
363 * Create a new mm_struct and populate it with a temporary stack
364 * vm_area_struct. We don't have enough context at this point to set the stack
365 * flags, permissions, and offset, so we use temporary values. We'll update
366 * them later in setup_arg_pages().
368 static int bprm_mm_init(struct linux_binprm *bprm)
371 struct mm_struct *mm = NULL;
373 bprm->mm = mm = mm_alloc();
378 err = __bprm_mm_init(bprm);
393 struct user_arg_ptr {
398 const char __user *const __user *native;
400 const compat_uptr_t __user *compat;
405 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
407 const char __user *native;
410 if (unlikely(argv.is_compat)) {
411 compat_uptr_t compat;
413 if (get_user(compat, argv.ptr.compat + nr))
414 return ERR_PTR(-EFAULT);
416 return compat_ptr(compat);
420 if (get_user(native, argv.ptr.native + nr))
421 return ERR_PTR(-EFAULT);
427 * count() counts the number of strings in array ARGV.
429 static int count(struct user_arg_ptr argv, int max)
433 if (argv.ptr.native != NULL) {
435 const char __user *p = get_user_arg_ptr(argv, i);
447 if (fatal_signal_pending(current))
448 return -ERESTARTNOHAND;
456 * 'copy_strings()' copies argument/environment strings from the old
457 * processes's memory to the new process's stack. The call to get_user_pages()
458 * ensures the destination page is created and not swapped out.
460 static int copy_strings(int argc, struct user_arg_ptr argv,
461 struct linux_binprm *bprm)
463 struct page *kmapped_page = NULL;
465 unsigned long kpos = 0;
469 const char __user *str;
474 str = get_user_arg_ptr(argv, argc);
478 len = strnlen_user(str, MAX_ARG_STRLEN);
483 if (!valid_arg_len(bprm, len))
486 /* We're going to work our way backwords. */
492 int offset, bytes_to_copy;
494 if (fatal_signal_pending(current)) {
495 ret = -ERESTARTNOHAND;
500 offset = pos % PAGE_SIZE;
504 bytes_to_copy = offset;
505 if (bytes_to_copy > len)
508 offset -= bytes_to_copy;
509 pos -= bytes_to_copy;
510 str -= bytes_to_copy;
511 len -= bytes_to_copy;
513 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
516 page = get_arg_page(bprm, pos, 1);
523 flush_kernel_dcache_page(kmapped_page);
524 kunmap(kmapped_page);
525 put_arg_page(kmapped_page);
528 kaddr = kmap(kmapped_page);
529 kpos = pos & PAGE_MASK;
530 flush_arg_page(bprm, kpos, kmapped_page);
532 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
541 flush_kernel_dcache_page(kmapped_page);
542 kunmap(kmapped_page);
543 put_arg_page(kmapped_page);
549 * Like copy_strings, but get argv and its values from kernel memory.
551 int copy_strings_kernel(int argc, const char *const *__argv,
552 struct linux_binprm *bprm)
555 mm_segment_t oldfs = get_fs();
556 struct user_arg_ptr argv = {
557 .ptr.native = (const char __user *const __user *)__argv,
561 r = copy_strings(argc, argv, bprm);
566 EXPORT_SYMBOL(copy_strings_kernel);
571 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
572 * the binfmt code determines where the new stack should reside, we shift it to
573 * its final location. The process proceeds as follows:
575 * 1) Use shift to calculate the new vma endpoints.
576 * 2) Extend vma to cover both the old and new ranges. This ensures the
577 * arguments passed to subsequent functions are consistent.
578 * 3) Move vma's page tables to the new range.
579 * 4) Free up any cleared pgd range.
580 * 5) Shrink the vma to cover only the new range.
582 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
584 struct mm_struct *mm = vma->vm_mm;
585 unsigned long old_start = vma->vm_start;
586 unsigned long old_end = vma->vm_end;
587 unsigned long length = old_end - old_start;
588 unsigned long new_start = old_start - shift;
589 unsigned long new_end = old_end - shift;
590 struct mmu_gather tlb;
592 BUG_ON(new_start > new_end);
595 * ensure there are no vmas between where we want to go
598 if (vma != find_vma(mm, new_start))
602 * cover the whole range: [new_start, old_end)
604 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
608 * move the page tables downwards, on failure we rely on
609 * process cleanup to remove whatever mess we made.
611 if (length != move_page_tables(vma, old_start,
612 vma, new_start, length, false))
616 tlb_gather_mmu(&tlb, mm, old_start, old_end);
617 if (new_end > old_start) {
619 * when the old and new regions overlap clear from new_end.
621 free_pgd_range(&tlb, new_end, old_end, new_end,
622 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
625 * otherwise, clean from old_start; this is done to not touch
626 * the address space in [new_end, old_start) some architectures
627 * have constraints on va-space that make this illegal (IA64) -
628 * for the others its just a little faster.
630 free_pgd_range(&tlb, old_start, old_end, new_end,
631 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
633 tlb_finish_mmu(&tlb, old_start, old_end);
636 * Shrink the vma to just the new range. Always succeeds.
638 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
644 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
645 * the stack is optionally relocated, and some extra space is added.
647 int setup_arg_pages(struct linux_binprm *bprm,
648 unsigned long stack_top,
649 int executable_stack)
652 unsigned long stack_shift;
653 struct mm_struct *mm = current->mm;
654 struct vm_area_struct *vma = bprm->vma;
655 struct vm_area_struct *prev = NULL;
656 unsigned long vm_flags;
657 unsigned long stack_base;
658 unsigned long stack_size;
659 unsigned long stack_expand;
660 unsigned long rlim_stack;
662 #ifdef CONFIG_STACK_GROWSUP
663 /* Limit stack size */
664 stack_base = rlimit_max(RLIMIT_STACK);
665 if (stack_base > STACK_SIZE_MAX)
666 stack_base = STACK_SIZE_MAX;
668 /* Add space for stack randomization. */
669 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
671 /* Make sure we didn't let the argument array grow too large. */
672 if (vma->vm_end - vma->vm_start > stack_base)
675 stack_base = PAGE_ALIGN(stack_top - stack_base);
677 stack_shift = vma->vm_start - stack_base;
678 mm->arg_start = bprm->p - stack_shift;
679 bprm->p = vma->vm_end - stack_shift;
681 stack_top = arch_align_stack(stack_top);
682 stack_top = PAGE_ALIGN(stack_top);
684 if (unlikely(stack_top < mmap_min_addr) ||
685 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
688 stack_shift = vma->vm_end - stack_top;
690 bprm->p -= stack_shift;
691 mm->arg_start = bprm->p;
695 bprm->loader -= stack_shift;
696 bprm->exec -= stack_shift;
698 down_write(&mm->mmap_sem);
699 vm_flags = VM_STACK_FLAGS;
702 * Adjust stack execute permissions; explicitly enable for
703 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
704 * (arch default) otherwise.
706 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
708 else if (executable_stack == EXSTACK_DISABLE_X)
709 vm_flags &= ~VM_EXEC;
710 vm_flags |= mm->def_flags;
711 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
713 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
719 /* Move stack pages down in memory. */
721 ret = shift_arg_pages(vma, stack_shift);
726 /* mprotect_fixup is overkill to remove the temporary stack flags */
727 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
729 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
730 stack_size = vma->vm_end - vma->vm_start;
732 * Align this down to a page boundary as expand_stack
735 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
736 #ifdef CONFIG_STACK_GROWSUP
737 if (stack_size + stack_expand > rlim_stack)
738 stack_base = vma->vm_start + rlim_stack;
740 stack_base = vma->vm_end + stack_expand;
742 if (stack_size + stack_expand > rlim_stack)
743 stack_base = vma->vm_end - rlim_stack;
745 stack_base = vma->vm_start - stack_expand;
747 current->mm->start_stack = bprm->p;
748 ret = expand_stack(vma, stack_base);
753 up_write(&mm->mmap_sem);
756 EXPORT_SYMBOL(setup_arg_pages);
758 #endif /* CONFIG_MMU */
760 static struct file *do_open_execat(int fd, struct filename *name, int flags)
764 struct open_flags open_exec_flags = {
765 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
766 .acc_mode = MAY_EXEC | MAY_OPEN,
767 .intent = LOOKUP_OPEN,
768 .lookup_flags = LOOKUP_FOLLOW,
771 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
772 return ERR_PTR(-EINVAL);
773 if (flags & AT_SYMLINK_NOFOLLOW)
774 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
775 if (flags & AT_EMPTY_PATH)
776 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
778 file = do_filp_open(fd, name, &open_exec_flags);
783 if (!S_ISREG(file_inode(file)->i_mode))
786 if (path_noexec(&file->f_path))
789 err = deny_write_access(file);
793 if (name->name[0] != '\0')
804 struct file *open_exec(const char *name)
806 struct filename *filename = getname_kernel(name);
807 struct file *f = ERR_CAST(filename);
809 if (!IS_ERR(filename)) {
810 f = do_open_execat(AT_FDCWD, filename, 0);
815 EXPORT_SYMBOL(open_exec);
817 int kernel_read(struct file *file, loff_t offset,
818 char *addr, unsigned long count)
826 /* The cast to a user pointer is valid due to the set_fs() */
827 result = vfs_read(file, (void __user *)addr, count, &pos);
832 EXPORT_SYMBOL(kernel_read);
834 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
836 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
838 flush_icache_range(addr, addr + len);
841 EXPORT_SYMBOL(read_code);
843 static int exec_mmap(struct mm_struct *mm)
845 struct task_struct *tsk;
846 struct mm_struct *old_mm, *active_mm;
848 /* Notify parent that we're no longer interested in the old VM */
850 old_mm = current->mm;
851 mm_release(tsk, old_mm);
856 * Make sure that if there is a core dump in progress
857 * for the old mm, we get out and die instead of going
858 * through with the exec. We must hold mmap_sem around
859 * checking core_state and changing tsk->mm.
861 down_read(&old_mm->mmap_sem);
862 if (unlikely(old_mm->core_state)) {
863 up_read(&old_mm->mmap_sem);
868 preempt_disable_rt();
869 active_mm = tsk->active_mm;
872 activate_mm(active_mm, mm);
873 tsk->mm->vmacache_seqnum = 0;
878 up_read(&old_mm->mmap_sem);
879 BUG_ON(active_mm != old_mm);
880 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
881 mm_update_next_owner(old_mm);
890 * This function makes sure the current process has its own signal table,
891 * so that flush_signal_handlers can later reset the handlers without
892 * disturbing other processes. (Other processes might share the signal
893 * table via the CLONE_SIGHAND option to clone().)
895 static int de_thread(struct task_struct *tsk)
897 struct signal_struct *sig = tsk->signal;
898 struct sighand_struct *oldsighand = tsk->sighand;
899 spinlock_t *lock = &oldsighand->siglock;
901 if (thread_group_empty(tsk))
902 goto no_thread_group;
905 * Kill all other threads in the thread group.
908 if (signal_group_exit(sig)) {
910 * Another group action in progress, just
911 * return so that the signal is processed.
913 spin_unlock_irq(lock);
917 sig->group_exit_task = tsk;
918 sig->notify_count = zap_other_threads(tsk);
919 if (!thread_group_leader(tsk))
922 while (sig->notify_count) {
923 __set_current_state(TASK_KILLABLE);
924 spin_unlock_irq(lock);
926 if (unlikely(__fatal_signal_pending(tsk)))
930 spin_unlock_irq(lock);
933 * At this point all other threads have exited, all we have to
934 * do is to wait for the thread group leader to become inactive,
935 * and to assume its PID:
937 if (!thread_group_leader(tsk)) {
938 struct task_struct *leader = tsk->group_leader;
941 threadgroup_change_begin(tsk);
942 write_lock_irq(&tasklist_lock);
944 * Do this under tasklist_lock to ensure that
945 * exit_notify() can't miss ->group_exit_task
947 sig->notify_count = -1;
948 if (likely(leader->exit_state))
950 __set_current_state(TASK_KILLABLE);
951 write_unlock_irq(&tasklist_lock);
952 threadgroup_change_end(tsk);
954 if (unlikely(__fatal_signal_pending(tsk)))
959 * The only record we have of the real-time age of a
960 * process, regardless of execs it's done, is start_time.
961 * All the past CPU time is accumulated in signal_struct
962 * from sister threads now dead. But in this non-leader
963 * exec, nothing survives from the original leader thread,
964 * whose birth marks the true age of this process now.
965 * When we take on its identity by switching to its PID, we
966 * also take its birthdate (always earlier than our own).
968 tsk->start_time = leader->start_time;
969 tsk->real_start_time = leader->real_start_time;
971 BUG_ON(!same_thread_group(leader, tsk));
972 BUG_ON(has_group_leader_pid(tsk));
974 * An exec() starts a new thread group with the
975 * TGID of the previous thread group. Rehash the
976 * two threads with a switched PID, and release
977 * the former thread group leader:
980 /* Become a process group leader with the old leader's pid.
981 * The old leader becomes a thread of the this thread group.
982 * Note: The old leader also uses this pid until release_task
983 * is called. Odd but simple and correct.
985 tsk->pid = leader->pid;
986 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
987 transfer_pid(leader, tsk, PIDTYPE_PGID);
988 transfer_pid(leader, tsk, PIDTYPE_SID);
990 list_replace_rcu(&leader->tasks, &tsk->tasks);
991 list_replace_init(&leader->sibling, &tsk->sibling);
993 tsk->group_leader = tsk;
994 leader->group_leader = tsk;
996 tsk->exit_signal = SIGCHLD;
997 leader->exit_signal = -1;
999 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1000 leader->exit_state = EXIT_DEAD;
1003 * We are going to release_task()->ptrace_unlink() silently,
1004 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1005 * the tracer wont't block again waiting for this thread.
1007 if (unlikely(leader->ptrace))
1008 __wake_up_parent(leader, leader->parent);
1009 write_unlock_irq(&tasklist_lock);
1010 threadgroup_change_end(tsk);
1012 release_task(leader);
1015 sig->group_exit_task = NULL;
1016 sig->notify_count = 0;
1019 /* we have changed execution domain */
1020 tsk->exit_signal = SIGCHLD;
1023 flush_itimer_signals();
1025 if (atomic_read(&oldsighand->count) != 1) {
1026 struct sighand_struct *newsighand;
1028 * This ->sighand is shared with the CLONE_SIGHAND
1029 * but not CLONE_THREAD task, switch to the new one.
1031 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1035 atomic_set(&newsighand->count, 1);
1036 memcpy(newsighand->action, oldsighand->action,
1037 sizeof(newsighand->action));
1039 write_lock_irq(&tasklist_lock);
1040 spin_lock(&oldsighand->siglock);
1041 rcu_assign_pointer(tsk->sighand, newsighand);
1042 spin_unlock(&oldsighand->siglock);
1043 write_unlock_irq(&tasklist_lock);
1045 __cleanup_sighand(oldsighand);
1048 BUG_ON(!thread_group_leader(tsk));
1052 /* protects against exit_notify() and __exit_signal() */
1053 read_lock(&tasklist_lock);
1054 sig->group_exit_task = NULL;
1055 sig->notify_count = 0;
1056 read_unlock(&tasklist_lock);
1060 char *get_task_comm(char *buf, struct task_struct *tsk)
1062 /* buf must be at least sizeof(tsk->comm) in size */
1064 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1068 EXPORT_SYMBOL_GPL(get_task_comm);
1071 * These functions flushes out all traces of the currently running executable
1072 * so that a new one can be started
1075 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1078 trace_task_rename(tsk, buf);
1079 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1081 perf_event_comm(tsk, exec);
1084 int flush_old_exec(struct linux_binprm * bprm)
1089 * Make sure we have a private signal table and that
1090 * we are unassociated from the previous thread group.
1092 retval = de_thread(current);
1097 * Must be called _before_ exec_mmap() as bprm->mm is
1098 * not visibile until then. This also enables the update
1101 set_mm_exe_file(bprm->mm, bprm->file);
1104 * Release all of the old mmap stuff
1106 acct_arg_size(bprm, 0);
1107 retval = exec_mmap(bprm->mm);
1111 bprm->mm = NULL; /* We're using it now */
1114 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1115 PF_NOFREEZE | PF_NO_SETAFFINITY);
1117 current->personality &= ~bprm->per_clear;
1124 EXPORT_SYMBOL(flush_old_exec);
1126 void would_dump(struct linux_binprm *bprm, struct file *file)
1128 if (inode_permission(file_inode(file), MAY_READ) < 0)
1129 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1131 EXPORT_SYMBOL(would_dump);
1133 void setup_new_exec(struct linux_binprm * bprm)
1135 arch_pick_mmap_layout(current->mm);
1137 /* This is the point of no return */
1138 current->sas_ss_sp = current->sas_ss_size = 0;
1140 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1141 set_dumpable(current->mm, SUID_DUMP_USER);
1143 set_dumpable(current->mm, suid_dumpable);
1146 __set_task_comm(current, kbasename(bprm->filename), true);
1148 /* Set the new mm task size. We have to do that late because it may
1149 * depend on TIF_32BIT which is only updated in flush_thread() on
1150 * some architectures like powerpc
1152 current->mm->task_size = TASK_SIZE;
1154 /* install the new credentials */
1155 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1156 !gid_eq(bprm->cred->gid, current_egid())) {
1157 current->pdeath_signal = 0;
1159 would_dump(bprm, bprm->file);
1160 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1161 set_dumpable(current->mm, suid_dumpable);
1164 /* An exec changes our domain. We are no longer part of the thread
1166 current->self_exec_id++;
1167 flush_signal_handlers(current, 0);
1168 do_close_on_exec(current->files);
1170 EXPORT_SYMBOL(setup_new_exec);
1173 * Prepare credentials and lock ->cred_guard_mutex.
1174 * install_exec_creds() commits the new creds and drops the lock.
1175 * Or, if exec fails before, free_bprm() should release ->cred and
1178 int prepare_bprm_creds(struct linux_binprm *bprm)
1180 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1181 return -ERESTARTNOINTR;
1183 bprm->cred = prepare_exec_creds();
1184 if (likely(bprm->cred))
1187 mutex_unlock(¤t->signal->cred_guard_mutex);
1191 static void free_bprm(struct linux_binprm *bprm)
1193 free_arg_pages(bprm);
1195 mutex_unlock(¤t->signal->cred_guard_mutex);
1196 abort_creds(bprm->cred);
1199 allow_write_access(bprm->file);
1202 /* If a binfmt changed the interp, free it. */
1203 if (bprm->interp != bprm->filename)
1204 kfree(bprm->interp);
1208 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1210 /* If a binfmt changed the interp, free it first. */
1211 if (bprm->interp != bprm->filename)
1212 kfree(bprm->interp);
1213 bprm->interp = kstrdup(interp, GFP_KERNEL);
1218 EXPORT_SYMBOL(bprm_change_interp);
1221 * install the new credentials for this executable
1223 void install_exec_creds(struct linux_binprm *bprm)
1225 security_bprm_committing_creds(bprm);
1227 commit_creds(bprm->cred);
1231 * Disable monitoring for regular users
1232 * when executing setuid binaries. Must
1233 * wait until new credentials are committed
1234 * by commit_creds() above
1236 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1237 perf_event_exit_task(current);
1239 * cred_guard_mutex must be held at least to this point to prevent
1240 * ptrace_attach() from altering our determination of the task's
1241 * credentials; any time after this it may be unlocked.
1243 security_bprm_committed_creds(bprm);
1244 mutex_unlock(¤t->signal->cred_guard_mutex);
1246 EXPORT_SYMBOL(install_exec_creds);
1249 * determine how safe it is to execute the proposed program
1250 * - the caller must hold ->cred_guard_mutex to protect against
1251 * PTRACE_ATTACH or seccomp thread-sync
1253 static void check_unsafe_exec(struct linux_binprm *bprm)
1255 struct task_struct *p = current, *t;
1259 if (p->ptrace & PT_PTRACE_CAP)
1260 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1262 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1266 * This isn't strictly necessary, but it makes it harder for LSMs to
1269 if (task_no_new_privs(current))
1270 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1274 spin_lock(&p->fs->lock);
1276 while_each_thread(p, t) {
1282 if (p->fs->users > n_fs)
1283 bprm->unsafe |= LSM_UNSAFE_SHARE;
1286 spin_unlock(&p->fs->lock);
1289 static void bprm_fill_uid(struct linux_binprm *bprm)
1291 struct inode *inode;
1296 /* clear any previous set[ug]id data from a previous binary */
1297 bprm->cred->euid = current_euid();
1298 bprm->cred->egid = current_egid();
1300 if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1303 if (task_no_new_privs(current))
1306 inode = file_inode(bprm->file);
1307 mode = READ_ONCE(inode->i_mode);
1308 if (!(mode & (S_ISUID|S_ISGID)))
1311 /* Be careful if suid/sgid is set */
1312 mutex_lock(&inode->i_mutex);
1314 /* reload atomically mode/uid/gid now that lock held */
1315 mode = inode->i_mode;
1318 mutex_unlock(&inode->i_mutex);
1320 /* We ignore suid/sgid if there are no mappings for them in the ns */
1321 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1322 !kgid_has_mapping(bprm->cred->user_ns, gid))
1325 if (mode & S_ISUID) {
1326 bprm->per_clear |= PER_CLEAR_ON_SETID;
1327 bprm->cred->euid = uid;
1330 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1331 bprm->per_clear |= PER_CLEAR_ON_SETID;
1332 bprm->cred->egid = gid;
1337 * Fill the binprm structure from the inode.
1338 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1340 * This may be called multiple times for binary chains (scripts for example).
1342 int prepare_binprm(struct linux_binprm *bprm)
1346 bprm_fill_uid(bprm);
1348 /* fill in binprm security blob */
1349 retval = security_bprm_set_creds(bprm);
1352 bprm->cred_prepared = 1;
1354 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1355 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1358 EXPORT_SYMBOL(prepare_binprm);
1361 * Arguments are '\0' separated strings found at the location bprm->p
1362 * points to; chop off the first by relocating brpm->p to right after
1363 * the first '\0' encountered.
1365 int remove_arg_zero(struct linux_binprm *bprm)
1368 unsigned long offset;
1376 offset = bprm->p & ~PAGE_MASK;
1377 page = get_arg_page(bprm, bprm->p, 0);
1382 kaddr = kmap_atomic(page);
1384 for (; offset < PAGE_SIZE && kaddr[offset];
1385 offset++, bprm->p++)
1388 kunmap_atomic(kaddr);
1391 if (offset == PAGE_SIZE)
1392 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1393 } while (offset == PAGE_SIZE);
1402 EXPORT_SYMBOL(remove_arg_zero);
1404 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1406 * cycle the list of binary formats handler, until one recognizes the image
1408 int search_binary_handler(struct linux_binprm *bprm)
1410 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1411 struct linux_binfmt *fmt;
1414 /* This allows 4 levels of binfmt rewrites before failing hard. */
1415 if (bprm->recursion_depth > 5)
1418 retval = security_bprm_check(bprm);
1424 read_lock(&binfmt_lock);
1425 list_for_each_entry(fmt, &formats, lh) {
1426 if (!try_module_get(fmt->module))
1428 read_unlock(&binfmt_lock);
1429 bprm->recursion_depth++;
1430 retval = fmt->load_binary(bprm);
1431 read_lock(&binfmt_lock);
1433 bprm->recursion_depth--;
1434 if (retval < 0 && !bprm->mm) {
1435 /* we got to flush_old_exec() and failed after it */
1436 read_unlock(&binfmt_lock);
1437 force_sigsegv(SIGSEGV, current);
1440 if (retval != -ENOEXEC || !bprm->file) {
1441 read_unlock(&binfmt_lock);
1445 read_unlock(&binfmt_lock);
1448 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1449 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1451 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1459 EXPORT_SYMBOL(search_binary_handler);
1461 static int exec_binprm(struct linux_binprm *bprm)
1463 pid_t old_pid, old_vpid;
1466 /* Need to fetch pid before load_binary changes it */
1467 old_pid = current->pid;
1469 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1472 ret = search_binary_handler(bprm);
1475 trace_sched_process_exec(current, old_pid, bprm);
1476 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1477 proc_exec_connector(current);
1484 * sys_execve() executes a new program.
1486 static int do_execveat_common(int fd, struct filename *filename,
1487 struct user_arg_ptr argv,
1488 struct user_arg_ptr envp,
1491 char *pathbuf = NULL;
1492 struct linux_binprm *bprm;
1494 struct files_struct *displaced;
1497 if (IS_ERR(filename))
1498 return PTR_ERR(filename);
1501 * We move the actual failure in case of RLIMIT_NPROC excess from
1502 * set*uid() to execve() because too many poorly written programs
1503 * don't check setuid() return code. Here we additionally recheck
1504 * whether NPROC limit is still exceeded.
1506 if ((current->flags & PF_NPROC_EXCEEDED) &&
1507 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1512 /* We're below the limit (still or again), so we don't want to make
1513 * further execve() calls fail. */
1514 current->flags &= ~PF_NPROC_EXCEEDED;
1516 retval = unshare_files(&displaced);
1521 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1525 retval = prepare_bprm_creds(bprm);
1529 check_unsafe_exec(bprm);
1530 current->in_execve = 1;
1532 file = do_open_execat(fd, filename, flags);
1533 retval = PTR_ERR(file);
1540 if (fd == AT_FDCWD || filename->name[0] == '/') {
1541 bprm->filename = filename->name;
1543 if (filename->name[0] == '\0')
1544 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1546 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1547 fd, filename->name);
1553 * Record that a name derived from an O_CLOEXEC fd will be
1554 * inaccessible after exec. Relies on having exclusive access to
1555 * current->files (due to unshare_files above).
1557 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1558 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1559 bprm->filename = pathbuf;
1561 bprm->interp = bprm->filename;
1563 retval = bprm_mm_init(bprm);
1567 bprm->argc = count(argv, MAX_ARG_STRINGS);
1568 if ((retval = bprm->argc) < 0)
1571 bprm->envc = count(envp, MAX_ARG_STRINGS);
1572 if ((retval = bprm->envc) < 0)
1575 retval = prepare_binprm(bprm);
1579 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1583 bprm->exec = bprm->p;
1584 retval = copy_strings(bprm->envc, envp, bprm);
1588 retval = copy_strings(bprm->argc, argv, bprm);
1592 retval = exec_binprm(bprm);
1596 /* execve succeeded */
1597 current->fs->in_exec = 0;
1598 current->in_execve = 0;
1599 acct_update_integrals(current);
1600 task_numa_free(current);
1605 put_files_struct(displaced);
1610 acct_arg_size(bprm, 0);
1615 current->fs->in_exec = 0;
1616 current->in_execve = 0;
1624 reset_files_struct(displaced);
1630 int do_execve(struct filename *filename,
1631 const char __user *const __user *__argv,
1632 const char __user *const __user *__envp)
1634 struct user_arg_ptr argv = { .ptr.native = __argv };
1635 struct user_arg_ptr envp = { .ptr.native = __envp };
1636 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1639 int do_execveat(int fd, struct filename *filename,
1640 const char __user *const __user *__argv,
1641 const char __user *const __user *__envp,
1644 struct user_arg_ptr argv = { .ptr.native = __argv };
1645 struct user_arg_ptr envp = { .ptr.native = __envp };
1647 return do_execveat_common(fd, filename, argv, envp, flags);
1650 #ifdef CONFIG_COMPAT
1651 static int compat_do_execve(struct filename *filename,
1652 const compat_uptr_t __user *__argv,
1653 const compat_uptr_t __user *__envp)
1655 struct user_arg_ptr argv = {
1657 .ptr.compat = __argv,
1659 struct user_arg_ptr envp = {
1661 .ptr.compat = __envp,
1663 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1666 static int compat_do_execveat(int fd, struct filename *filename,
1667 const compat_uptr_t __user *__argv,
1668 const compat_uptr_t __user *__envp,
1671 struct user_arg_ptr argv = {
1673 .ptr.compat = __argv,
1675 struct user_arg_ptr envp = {
1677 .ptr.compat = __envp,
1679 return do_execveat_common(fd, filename, argv, envp, flags);
1683 void set_binfmt(struct linux_binfmt *new)
1685 struct mm_struct *mm = current->mm;
1688 module_put(mm->binfmt->module);
1692 __module_get(new->module);
1694 EXPORT_SYMBOL(set_binfmt);
1697 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1699 void set_dumpable(struct mm_struct *mm, int value)
1701 unsigned long old, new;
1703 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1707 old = ACCESS_ONCE(mm->flags);
1708 new = (old & ~MMF_DUMPABLE_MASK) | value;
1709 } while (cmpxchg(&mm->flags, old, new) != old);
1712 SYSCALL_DEFINE3(execve,
1713 const char __user *, filename,
1714 const char __user *const __user *, argv,
1715 const char __user *const __user *, envp)
1717 return do_execve(getname(filename), argv, envp);
1720 SYSCALL_DEFINE5(execveat,
1721 int, fd, const char __user *, filename,
1722 const char __user *const __user *, argv,
1723 const char __user *const __user *, envp,
1726 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1728 return do_execveat(fd,
1729 getname_flags(filename, lookup_flags, NULL),
1733 #ifdef CONFIG_COMPAT
1734 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1735 const compat_uptr_t __user *, argv,
1736 const compat_uptr_t __user *, envp)
1738 return compat_do_execve(getname(filename), argv, envp);
1741 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1742 const char __user *, filename,
1743 const compat_uptr_t __user *, argv,
1744 const compat_uptr_t __user *, envp,
1747 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1749 return compat_do_execveat(fd,
1750 getname_flags(filename, lookup_flags, NULL),