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>
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);
104 * Note that a shared library must be both readable and executable due to
107 * Also note that we take the address to load from from the file itself.
109 SYSCALL_DEFINE1(uselib, const char __user *, library)
111 struct linux_binfmt *fmt;
113 struct filename *tmp = getname(library);
114 int error = PTR_ERR(tmp);
115 static const struct open_flags uselib_flags = {
116 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
117 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
118 .intent = LOOKUP_OPEN,
119 .lookup_flags = LOOKUP_FOLLOW,
125 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
127 error = PTR_ERR(file);
132 if (!S_ISREG(file_inode(file)->i_mode))
136 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
143 read_lock(&binfmt_lock);
144 list_for_each_entry(fmt, &formats, lh) {
145 if (!fmt->load_shlib)
147 if (!try_module_get(fmt->module))
149 read_unlock(&binfmt_lock);
150 error = fmt->load_shlib(file);
151 read_lock(&binfmt_lock);
153 if (error != -ENOEXEC)
156 read_unlock(&binfmt_lock);
162 #endif /* #ifdef CONFIG_USELIB */
166 * The nascent bprm->mm is not visible until exec_mmap() but it can
167 * use a lot of memory, account these pages in current->mm temporary
168 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
169 * change the counter back via acct_arg_size(0).
171 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
173 struct mm_struct *mm = current->mm;
174 long diff = (long)(pages - bprm->vma_pages);
179 bprm->vma_pages = pages;
180 add_mm_counter(mm, MM_ANONPAGES, diff);
183 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
189 #ifdef CONFIG_STACK_GROWSUP
191 ret = expand_downwards(bprm->vma, pos);
196 ret = get_user_pages(current, bprm->mm, pos,
197 1, write, 1, &page, NULL);
202 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
205 acct_arg_size(bprm, size / PAGE_SIZE);
208 * We've historically supported up to 32 pages (ARG_MAX)
209 * of argument strings even with small stacks
215 * Limit to 1/4-th the stack size for the argv+env strings.
217 * - the remaining binfmt code will not run out of stack space,
218 * - the program will have a reasonable amount of stack left
221 rlim = current->signal->rlim;
222 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
231 static void put_arg_page(struct page *page)
236 static void free_arg_page(struct linux_binprm *bprm, int i)
240 static void free_arg_pages(struct linux_binprm *bprm)
244 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
247 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
250 static int __bprm_mm_init(struct linux_binprm *bprm)
253 struct vm_area_struct *vma = NULL;
254 struct mm_struct *mm = bprm->mm;
256 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
260 down_write(&mm->mmap_sem);
264 * Place the stack at the largest stack address the architecture
265 * supports. Later, we'll move this to an appropriate place. We don't
266 * use STACK_TOP because that can depend on attributes which aren't
269 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
270 vma->vm_end = STACK_TOP_MAX;
271 vma->vm_start = vma->vm_end - PAGE_SIZE;
272 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
273 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
274 INIT_LIST_HEAD(&vma->anon_vma_chain);
276 err = insert_vm_struct(mm, vma);
280 mm->stack_vm = mm->total_vm = 1;
281 arch_bprm_mm_init(mm, vma);
282 up_write(&mm->mmap_sem);
283 bprm->p = vma->vm_end - sizeof(void *);
286 up_write(&mm->mmap_sem);
288 kmem_cache_free(vm_area_cachep, vma);
292 static bool valid_arg_len(struct linux_binprm *bprm, long len)
294 return len <= MAX_ARG_STRLEN;
299 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
303 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
308 page = bprm->page[pos / PAGE_SIZE];
309 if (!page && write) {
310 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
313 bprm->page[pos / PAGE_SIZE] = page;
319 static void put_arg_page(struct page *page)
323 static void free_arg_page(struct linux_binprm *bprm, int i)
326 __free_page(bprm->page[i]);
327 bprm->page[i] = NULL;
331 static void free_arg_pages(struct linux_binprm *bprm)
335 for (i = 0; i < MAX_ARG_PAGES; i++)
336 free_arg_page(bprm, i);
339 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
344 static int __bprm_mm_init(struct linux_binprm *bprm)
346 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
350 static bool valid_arg_len(struct linux_binprm *bprm, long len)
352 return len <= bprm->p;
355 #endif /* CONFIG_MMU */
358 * Create a new mm_struct and populate it with a temporary stack
359 * vm_area_struct. We don't have enough context at this point to set the stack
360 * flags, permissions, and offset, so we use temporary values. We'll update
361 * them later in setup_arg_pages().
363 static int bprm_mm_init(struct linux_binprm *bprm)
366 struct mm_struct *mm = NULL;
368 bprm->mm = mm = mm_alloc();
373 err = __bprm_mm_init(bprm);
388 struct user_arg_ptr {
393 const char __user *const __user *native;
395 const compat_uptr_t __user *compat;
400 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
402 const char __user *native;
405 if (unlikely(argv.is_compat)) {
406 compat_uptr_t compat;
408 if (get_user(compat, argv.ptr.compat + nr))
409 return ERR_PTR(-EFAULT);
411 return compat_ptr(compat);
415 if (get_user(native, argv.ptr.native + nr))
416 return ERR_PTR(-EFAULT);
422 * count() counts the number of strings in array ARGV.
424 static int count(struct user_arg_ptr argv, int max)
428 if (argv.ptr.native != NULL) {
430 const char __user *p = get_user_arg_ptr(argv, i);
442 if (fatal_signal_pending(current))
443 return -ERESTARTNOHAND;
451 * 'copy_strings()' copies argument/environment strings from the old
452 * processes's memory to the new process's stack. The call to get_user_pages()
453 * ensures the destination page is created and not swapped out.
455 static int copy_strings(int argc, struct user_arg_ptr argv,
456 struct linux_binprm *bprm)
458 struct page *kmapped_page = NULL;
460 unsigned long kpos = 0;
464 const char __user *str;
469 str = get_user_arg_ptr(argv, argc);
473 len = strnlen_user(str, MAX_ARG_STRLEN);
478 if (!valid_arg_len(bprm, len))
481 /* We're going to work our way backwords. */
487 int offset, bytes_to_copy;
489 if (fatal_signal_pending(current)) {
490 ret = -ERESTARTNOHAND;
495 offset = pos % PAGE_SIZE;
499 bytes_to_copy = offset;
500 if (bytes_to_copy > len)
503 offset -= bytes_to_copy;
504 pos -= bytes_to_copy;
505 str -= bytes_to_copy;
506 len -= bytes_to_copy;
508 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
511 page = get_arg_page(bprm, pos, 1);
518 flush_kernel_dcache_page(kmapped_page);
519 kunmap(kmapped_page);
520 put_arg_page(kmapped_page);
523 kaddr = kmap(kmapped_page);
524 kpos = pos & PAGE_MASK;
525 flush_arg_page(bprm, kpos, kmapped_page);
527 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
536 flush_kernel_dcache_page(kmapped_page);
537 kunmap(kmapped_page);
538 put_arg_page(kmapped_page);
544 * Like copy_strings, but get argv and its values from kernel memory.
546 int copy_strings_kernel(int argc, const char *const *__argv,
547 struct linux_binprm *bprm)
550 mm_segment_t oldfs = get_fs();
551 struct user_arg_ptr argv = {
552 .ptr.native = (const char __user *const __user *)__argv,
556 r = copy_strings(argc, argv, bprm);
561 EXPORT_SYMBOL(copy_strings_kernel);
566 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
567 * the binfmt code determines where the new stack should reside, we shift it to
568 * its final location. The process proceeds as follows:
570 * 1) Use shift to calculate the new vma endpoints.
571 * 2) Extend vma to cover both the old and new ranges. This ensures the
572 * arguments passed to subsequent functions are consistent.
573 * 3) Move vma's page tables to the new range.
574 * 4) Free up any cleared pgd range.
575 * 5) Shrink the vma to cover only the new range.
577 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
579 struct mm_struct *mm = vma->vm_mm;
580 unsigned long old_start = vma->vm_start;
581 unsigned long old_end = vma->vm_end;
582 unsigned long length = old_end - old_start;
583 unsigned long new_start = old_start - shift;
584 unsigned long new_end = old_end - shift;
585 struct mmu_gather tlb;
587 BUG_ON(new_start > new_end);
590 * ensure there are no vmas between where we want to go
593 if (vma != find_vma(mm, new_start))
597 * cover the whole range: [new_start, old_end)
599 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
603 * move the page tables downwards, on failure we rely on
604 * process cleanup to remove whatever mess we made.
606 if (length != move_page_tables(vma, old_start,
607 vma, new_start, length, false))
611 tlb_gather_mmu(&tlb, mm, old_start, old_end);
612 if (new_end > old_start) {
614 * when the old and new regions overlap clear from new_end.
616 free_pgd_range(&tlb, new_end, old_end, new_end,
617 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
620 * otherwise, clean from old_start; this is done to not touch
621 * the address space in [new_end, old_start) some architectures
622 * have constraints on va-space that make this illegal (IA64) -
623 * for the others its just a little faster.
625 free_pgd_range(&tlb, old_start, old_end, new_end,
626 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
628 tlb_finish_mmu(&tlb, old_start, old_end);
631 * Shrink the vma to just the new range. Always succeeds.
633 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
639 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
640 * the stack is optionally relocated, and some extra space is added.
642 int setup_arg_pages(struct linux_binprm *bprm,
643 unsigned long stack_top,
644 int executable_stack)
647 unsigned long stack_shift;
648 struct mm_struct *mm = current->mm;
649 struct vm_area_struct *vma = bprm->vma;
650 struct vm_area_struct *prev = NULL;
651 unsigned long vm_flags;
652 unsigned long stack_base;
653 unsigned long stack_size;
654 unsigned long stack_expand;
655 unsigned long rlim_stack;
657 #ifdef CONFIG_STACK_GROWSUP
658 /* Limit stack size */
659 stack_base = rlimit_max(RLIMIT_STACK);
660 if (stack_base > STACK_SIZE_MAX)
661 stack_base = STACK_SIZE_MAX;
663 /* Make sure we didn't let the argument array grow too large. */
664 if (vma->vm_end - vma->vm_start > stack_base)
667 stack_base = PAGE_ALIGN(stack_top - stack_base);
669 stack_shift = vma->vm_start - stack_base;
670 mm->arg_start = bprm->p - stack_shift;
671 bprm->p = vma->vm_end - stack_shift;
673 stack_top = arch_align_stack(stack_top);
674 stack_top = PAGE_ALIGN(stack_top);
676 if (unlikely(stack_top < mmap_min_addr) ||
677 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
680 stack_shift = vma->vm_end - stack_top;
682 bprm->p -= stack_shift;
683 mm->arg_start = bprm->p;
687 bprm->loader -= stack_shift;
688 bprm->exec -= stack_shift;
690 down_write(&mm->mmap_sem);
691 vm_flags = VM_STACK_FLAGS;
694 * Adjust stack execute permissions; explicitly enable for
695 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
696 * (arch default) otherwise.
698 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
700 else if (executable_stack == EXSTACK_DISABLE_X)
701 vm_flags &= ~VM_EXEC;
702 vm_flags |= mm->def_flags;
703 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
705 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
711 /* Move stack pages down in memory. */
713 ret = shift_arg_pages(vma, stack_shift);
718 /* mprotect_fixup is overkill to remove the temporary stack flags */
719 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
721 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
722 stack_size = vma->vm_end - vma->vm_start;
724 * Align this down to a page boundary as expand_stack
727 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
728 #ifdef CONFIG_STACK_GROWSUP
729 if (stack_size + stack_expand > rlim_stack)
730 stack_base = vma->vm_start + rlim_stack;
732 stack_base = vma->vm_end + stack_expand;
734 if (stack_size + stack_expand > rlim_stack)
735 stack_base = vma->vm_end - rlim_stack;
737 stack_base = vma->vm_start - stack_expand;
739 current->mm->start_stack = bprm->p;
740 ret = expand_stack(vma, stack_base);
745 up_write(&mm->mmap_sem);
748 EXPORT_SYMBOL(setup_arg_pages);
750 #endif /* CONFIG_MMU */
752 static struct file *do_open_exec(struct filename *name)
756 static const struct open_flags open_exec_flags = {
757 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
758 .acc_mode = MAY_EXEC | MAY_OPEN,
759 .intent = LOOKUP_OPEN,
760 .lookup_flags = LOOKUP_FOLLOW,
763 file = do_filp_open(AT_FDCWD, name, &open_exec_flags);
768 if (!S_ISREG(file_inode(file)->i_mode))
771 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
776 err = deny_write_access(file);
788 struct file *open_exec(const char *name)
790 struct filename tmp = { .name = name };
791 return do_open_exec(&tmp);
793 EXPORT_SYMBOL(open_exec);
795 int kernel_read(struct file *file, loff_t offset,
796 char *addr, unsigned long count)
804 /* The cast to a user pointer is valid due to the set_fs() */
805 result = vfs_read(file, (void __user *)addr, count, &pos);
810 EXPORT_SYMBOL(kernel_read);
812 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
814 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
816 flush_icache_range(addr, addr + len);
819 EXPORT_SYMBOL(read_code);
821 static int exec_mmap(struct mm_struct *mm)
823 struct task_struct *tsk;
824 struct mm_struct *old_mm, *active_mm;
826 /* Notify parent that we're no longer interested in the old VM */
828 old_mm = current->mm;
829 mm_release(tsk, old_mm);
834 * Make sure that if there is a core dump in progress
835 * for the old mm, we get out and die instead of going
836 * through with the exec. We must hold mmap_sem around
837 * checking core_state and changing tsk->mm.
839 down_read(&old_mm->mmap_sem);
840 if (unlikely(old_mm->core_state)) {
841 up_read(&old_mm->mmap_sem);
846 active_mm = tsk->active_mm;
849 activate_mm(active_mm, mm);
850 tsk->mm->vmacache_seqnum = 0;
854 up_read(&old_mm->mmap_sem);
855 BUG_ON(active_mm != old_mm);
856 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
857 mm_update_next_owner(old_mm);
866 * This function makes sure the current process has its own signal table,
867 * so that flush_signal_handlers can later reset the handlers without
868 * disturbing other processes. (Other processes might share the signal
869 * table via the CLONE_SIGHAND option to clone().)
871 static int de_thread(struct task_struct *tsk)
873 struct signal_struct *sig = tsk->signal;
874 struct sighand_struct *oldsighand = tsk->sighand;
875 spinlock_t *lock = &oldsighand->siglock;
877 if (thread_group_empty(tsk))
878 goto no_thread_group;
881 * Kill all other threads in the thread group.
884 if (signal_group_exit(sig)) {
886 * Another group action in progress, just
887 * return so that the signal is processed.
889 spin_unlock_irq(lock);
893 sig->group_exit_task = tsk;
894 sig->notify_count = zap_other_threads(tsk);
895 if (!thread_group_leader(tsk))
898 while (sig->notify_count) {
899 __set_current_state(TASK_KILLABLE);
900 spin_unlock_irq(lock);
902 if (unlikely(__fatal_signal_pending(tsk)))
906 spin_unlock_irq(lock);
909 * At this point all other threads have exited, all we have to
910 * do is to wait for the thread group leader to become inactive,
911 * and to assume its PID:
913 if (!thread_group_leader(tsk)) {
914 struct task_struct *leader = tsk->group_leader;
916 sig->notify_count = -1; /* for exit_notify() */
918 threadgroup_change_begin(tsk);
919 write_lock_irq(&tasklist_lock);
920 if (likely(leader->exit_state))
922 __set_current_state(TASK_KILLABLE);
923 write_unlock_irq(&tasklist_lock);
924 threadgroup_change_end(tsk);
926 if (unlikely(__fatal_signal_pending(tsk)))
931 * The only record we have of the real-time age of a
932 * process, regardless of execs it's done, is start_time.
933 * All the past CPU time is accumulated in signal_struct
934 * from sister threads now dead. But in this non-leader
935 * exec, nothing survives from the original leader thread,
936 * whose birth marks the true age of this process now.
937 * When we take on its identity by switching to its PID, we
938 * also take its birthdate (always earlier than our own).
940 tsk->start_time = leader->start_time;
941 tsk->real_start_time = leader->real_start_time;
943 BUG_ON(!same_thread_group(leader, tsk));
944 BUG_ON(has_group_leader_pid(tsk));
946 * An exec() starts a new thread group with the
947 * TGID of the previous thread group. Rehash the
948 * two threads with a switched PID, and release
949 * the former thread group leader:
952 /* Become a process group leader with the old leader's pid.
953 * The old leader becomes a thread of the this thread group.
954 * Note: The old leader also uses this pid until release_task
955 * is called. Odd but simple and correct.
957 tsk->pid = leader->pid;
958 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
959 transfer_pid(leader, tsk, PIDTYPE_PGID);
960 transfer_pid(leader, tsk, PIDTYPE_SID);
962 list_replace_rcu(&leader->tasks, &tsk->tasks);
963 list_replace_init(&leader->sibling, &tsk->sibling);
965 tsk->group_leader = tsk;
966 leader->group_leader = tsk;
968 tsk->exit_signal = SIGCHLD;
969 leader->exit_signal = -1;
971 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
972 leader->exit_state = EXIT_DEAD;
975 * We are going to release_task()->ptrace_unlink() silently,
976 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
977 * the tracer wont't block again waiting for this thread.
979 if (unlikely(leader->ptrace))
980 __wake_up_parent(leader, leader->parent);
981 write_unlock_irq(&tasklist_lock);
982 threadgroup_change_end(tsk);
984 release_task(leader);
987 sig->group_exit_task = NULL;
988 sig->notify_count = 0;
991 /* we have changed execution domain */
992 tsk->exit_signal = SIGCHLD;
995 flush_itimer_signals();
997 if (atomic_read(&oldsighand->count) != 1) {
998 struct sighand_struct *newsighand;
1000 * This ->sighand is shared with the CLONE_SIGHAND
1001 * but not CLONE_THREAD task, switch to the new one.
1003 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1007 atomic_set(&newsighand->count, 1);
1008 memcpy(newsighand->action, oldsighand->action,
1009 sizeof(newsighand->action));
1011 write_lock_irq(&tasklist_lock);
1012 spin_lock(&oldsighand->siglock);
1013 rcu_assign_pointer(tsk->sighand, newsighand);
1014 spin_unlock(&oldsighand->siglock);
1015 write_unlock_irq(&tasklist_lock);
1017 __cleanup_sighand(oldsighand);
1020 BUG_ON(!thread_group_leader(tsk));
1024 /* protects against exit_notify() and __exit_signal() */
1025 read_lock(&tasklist_lock);
1026 sig->group_exit_task = NULL;
1027 sig->notify_count = 0;
1028 read_unlock(&tasklist_lock);
1032 char *get_task_comm(char *buf, struct task_struct *tsk)
1034 /* buf must be at least sizeof(tsk->comm) in size */
1036 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1040 EXPORT_SYMBOL_GPL(get_task_comm);
1043 * These functions flushes out all traces of the currently running executable
1044 * so that a new one can be started
1047 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1050 trace_task_rename(tsk, buf);
1051 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1053 perf_event_comm(tsk, exec);
1056 int flush_old_exec(struct linux_binprm * bprm)
1061 * Make sure we have a private signal table and that
1062 * we are unassociated from the previous thread group.
1064 retval = de_thread(current);
1068 set_mm_exe_file(bprm->mm, bprm->file);
1070 * Release all of the old mmap stuff
1072 acct_arg_size(bprm, 0);
1073 retval = exec_mmap(bprm->mm);
1077 bprm->mm = NULL; /* We're using it now */
1080 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1081 PF_NOFREEZE | PF_NO_SETAFFINITY);
1083 current->personality &= ~bprm->per_clear;
1090 EXPORT_SYMBOL(flush_old_exec);
1092 void would_dump(struct linux_binprm *bprm, struct file *file)
1094 if (inode_permission(file_inode(file), MAY_READ) < 0)
1095 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1097 EXPORT_SYMBOL(would_dump);
1099 void setup_new_exec(struct linux_binprm * bprm)
1101 arch_pick_mmap_layout(current->mm);
1103 /* This is the point of no return */
1104 current->sas_ss_sp = current->sas_ss_size = 0;
1106 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1107 set_dumpable(current->mm, SUID_DUMP_USER);
1109 set_dumpable(current->mm, suid_dumpable);
1112 __set_task_comm(current, kbasename(bprm->filename), true);
1114 /* Set the new mm task size. We have to do that late because it may
1115 * depend on TIF_32BIT which is only updated in flush_thread() on
1116 * some architectures like powerpc
1118 current->mm->task_size = TASK_SIZE;
1120 /* install the new credentials */
1121 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1122 !gid_eq(bprm->cred->gid, current_egid())) {
1123 current->pdeath_signal = 0;
1125 would_dump(bprm, bprm->file);
1126 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1127 set_dumpable(current->mm, suid_dumpable);
1130 /* An exec changes our domain. We are no longer part of the thread
1132 current->self_exec_id++;
1133 flush_signal_handlers(current, 0);
1134 do_close_on_exec(current->files);
1136 EXPORT_SYMBOL(setup_new_exec);
1139 * Prepare credentials and lock ->cred_guard_mutex.
1140 * install_exec_creds() commits the new creds and drops the lock.
1141 * Or, if exec fails before, free_bprm() should release ->cred and
1144 int prepare_bprm_creds(struct linux_binprm *bprm)
1146 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1147 return -ERESTARTNOINTR;
1149 bprm->cred = prepare_exec_creds();
1150 if (likely(bprm->cred))
1153 mutex_unlock(¤t->signal->cred_guard_mutex);
1157 static void free_bprm(struct linux_binprm *bprm)
1159 free_arg_pages(bprm);
1161 mutex_unlock(¤t->signal->cred_guard_mutex);
1162 abort_creds(bprm->cred);
1165 allow_write_access(bprm->file);
1168 /* If a binfmt changed the interp, free it. */
1169 if (bprm->interp != bprm->filename)
1170 kfree(bprm->interp);
1174 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1176 /* If a binfmt changed the interp, free it first. */
1177 if (bprm->interp != bprm->filename)
1178 kfree(bprm->interp);
1179 bprm->interp = kstrdup(interp, GFP_KERNEL);
1184 EXPORT_SYMBOL(bprm_change_interp);
1187 * install the new credentials for this executable
1189 void install_exec_creds(struct linux_binprm *bprm)
1191 security_bprm_committing_creds(bprm);
1193 commit_creds(bprm->cred);
1197 * Disable monitoring for regular users
1198 * when executing setuid binaries. Must
1199 * wait until new credentials are committed
1200 * by commit_creds() above
1202 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1203 perf_event_exit_task(current);
1205 * cred_guard_mutex must be held at least to this point to prevent
1206 * ptrace_attach() from altering our determination of the task's
1207 * credentials; any time after this it may be unlocked.
1209 security_bprm_committed_creds(bprm);
1210 mutex_unlock(¤t->signal->cred_guard_mutex);
1212 EXPORT_SYMBOL(install_exec_creds);
1215 * determine how safe it is to execute the proposed program
1216 * - the caller must hold ->cred_guard_mutex to protect against
1217 * PTRACE_ATTACH or seccomp thread-sync
1219 static void check_unsafe_exec(struct linux_binprm *bprm)
1221 struct task_struct *p = current, *t;
1225 if (p->ptrace & PT_PTRACE_CAP)
1226 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1228 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1232 * This isn't strictly necessary, but it makes it harder for LSMs to
1235 if (task_no_new_privs(current))
1236 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1240 spin_lock(&p->fs->lock);
1242 while_each_thread(p, t) {
1248 if (p->fs->users > n_fs)
1249 bprm->unsafe |= LSM_UNSAFE_SHARE;
1252 spin_unlock(&p->fs->lock);
1256 * Fill the binprm structure from the inode.
1257 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1259 * This may be called multiple times for binary chains (scripts for example).
1261 int prepare_binprm(struct linux_binprm *bprm)
1263 struct inode *inode = file_inode(bprm->file);
1264 umode_t mode = inode->i_mode;
1268 /* clear any previous set[ug]id data from a previous binary */
1269 bprm->cred->euid = current_euid();
1270 bprm->cred->egid = current_egid();
1272 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1273 !task_no_new_privs(current) &&
1274 kuid_has_mapping(bprm->cred->user_ns, inode->i_uid) &&
1275 kgid_has_mapping(bprm->cred->user_ns, inode->i_gid)) {
1277 if (mode & S_ISUID) {
1278 bprm->per_clear |= PER_CLEAR_ON_SETID;
1279 bprm->cred->euid = inode->i_uid;
1284 * If setgid is set but no group execute bit then this
1285 * is a candidate for mandatory locking, not a setgid
1288 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1289 bprm->per_clear |= PER_CLEAR_ON_SETID;
1290 bprm->cred->egid = inode->i_gid;
1294 /* fill in binprm security blob */
1295 retval = security_bprm_set_creds(bprm);
1298 bprm->cred_prepared = 1;
1300 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1301 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1304 EXPORT_SYMBOL(prepare_binprm);
1307 * Arguments are '\0' separated strings found at the location bprm->p
1308 * points to; chop off the first by relocating brpm->p to right after
1309 * the first '\0' encountered.
1311 int remove_arg_zero(struct linux_binprm *bprm)
1314 unsigned long offset;
1322 offset = bprm->p & ~PAGE_MASK;
1323 page = get_arg_page(bprm, bprm->p, 0);
1328 kaddr = kmap_atomic(page);
1330 for (; offset < PAGE_SIZE && kaddr[offset];
1331 offset++, bprm->p++)
1334 kunmap_atomic(kaddr);
1337 if (offset == PAGE_SIZE)
1338 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1339 } while (offset == PAGE_SIZE);
1348 EXPORT_SYMBOL(remove_arg_zero);
1350 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1352 * cycle the list of binary formats handler, until one recognizes the image
1354 int search_binary_handler(struct linux_binprm *bprm)
1356 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1357 struct linux_binfmt *fmt;
1360 /* This allows 4 levels of binfmt rewrites before failing hard. */
1361 if (bprm->recursion_depth > 5)
1364 retval = security_bprm_check(bprm);
1370 read_lock(&binfmt_lock);
1371 list_for_each_entry(fmt, &formats, lh) {
1372 if (!try_module_get(fmt->module))
1374 read_unlock(&binfmt_lock);
1375 bprm->recursion_depth++;
1376 retval = fmt->load_binary(bprm);
1377 read_lock(&binfmt_lock);
1379 bprm->recursion_depth--;
1380 if (retval < 0 && !bprm->mm) {
1381 /* we got to flush_old_exec() and failed after it */
1382 read_unlock(&binfmt_lock);
1383 force_sigsegv(SIGSEGV, current);
1386 if (retval != -ENOEXEC || !bprm->file) {
1387 read_unlock(&binfmt_lock);
1391 read_unlock(&binfmt_lock);
1394 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1395 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1397 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1405 EXPORT_SYMBOL(search_binary_handler);
1407 static int exec_binprm(struct linux_binprm *bprm)
1409 pid_t old_pid, old_vpid;
1412 /* Need to fetch pid before load_binary changes it */
1413 old_pid = current->pid;
1415 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1418 ret = search_binary_handler(bprm);
1421 trace_sched_process_exec(current, old_pid, bprm);
1422 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1423 proc_exec_connector(current);
1430 * sys_execve() executes a new program.
1432 static int do_execve_common(struct filename *filename,
1433 struct user_arg_ptr argv,
1434 struct user_arg_ptr envp)
1436 struct linux_binprm *bprm;
1438 struct files_struct *displaced;
1441 if (IS_ERR(filename))
1442 return PTR_ERR(filename);
1445 * We move the actual failure in case of RLIMIT_NPROC excess from
1446 * set*uid() to execve() because too many poorly written programs
1447 * don't check setuid() return code. Here we additionally recheck
1448 * whether NPROC limit is still exceeded.
1450 if ((current->flags & PF_NPROC_EXCEEDED) &&
1451 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1456 /* We're below the limit (still or again), so we don't want to make
1457 * further execve() calls fail. */
1458 current->flags &= ~PF_NPROC_EXCEEDED;
1460 retval = unshare_files(&displaced);
1465 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1469 retval = prepare_bprm_creds(bprm);
1473 check_unsafe_exec(bprm);
1474 current->in_execve = 1;
1476 file = do_open_exec(filename);
1477 retval = PTR_ERR(file);
1484 bprm->filename = bprm->interp = filename->name;
1486 retval = bprm_mm_init(bprm);
1490 bprm->argc = count(argv, MAX_ARG_STRINGS);
1491 if ((retval = bprm->argc) < 0)
1494 bprm->envc = count(envp, MAX_ARG_STRINGS);
1495 if ((retval = bprm->envc) < 0)
1498 retval = prepare_binprm(bprm);
1502 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1506 bprm->exec = bprm->p;
1507 retval = copy_strings(bprm->envc, envp, bprm);
1511 retval = copy_strings(bprm->argc, argv, bprm);
1515 retval = exec_binprm(bprm);
1519 /* execve succeeded */
1520 current->fs->in_exec = 0;
1521 current->in_execve = 0;
1522 acct_update_integrals(current);
1523 task_numa_free(current);
1527 put_files_struct(displaced);
1532 acct_arg_size(bprm, 0);
1537 current->fs->in_exec = 0;
1538 current->in_execve = 0;
1545 reset_files_struct(displaced);
1551 int do_execve(struct filename *filename,
1552 const char __user *const __user *__argv,
1553 const char __user *const __user *__envp)
1555 struct user_arg_ptr argv = { .ptr.native = __argv };
1556 struct user_arg_ptr envp = { .ptr.native = __envp };
1557 return do_execve_common(filename, argv, envp);
1560 #ifdef CONFIG_COMPAT
1561 static int compat_do_execve(struct filename *filename,
1562 const compat_uptr_t __user *__argv,
1563 const compat_uptr_t __user *__envp)
1565 struct user_arg_ptr argv = {
1567 .ptr.compat = __argv,
1569 struct user_arg_ptr envp = {
1571 .ptr.compat = __envp,
1573 return do_execve_common(filename, argv, envp);
1577 void set_binfmt(struct linux_binfmt *new)
1579 struct mm_struct *mm = current->mm;
1582 module_put(mm->binfmt->module);
1586 __module_get(new->module);
1588 EXPORT_SYMBOL(set_binfmt);
1591 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1593 void set_dumpable(struct mm_struct *mm, int value)
1595 unsigned long old, new;
1597 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1601 old = ACCESS_ONCE(mm->flags);
1602 new = (old & ~MMF_DUMPABLE_MASK) | value;
1603 } while (cmpxchg(&mm->flags, old, new) != old);
1606 SYSCALL_DEFINE3(execve,
1607 const char __user *, filename,
1608 const char __user *const __user *, argv,
1609 const char __user *const __user *, envp)
1611 return do_execve(getname(filename), argv, envp);
1613 #ifdef CONFIG_COMPAT
1614 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1615 const compat_uptr_t __user *, argv,
1616 const compat_uptr_t __user *, envp)
1618 return compat_do_execve(getname(filename), argv, envp);