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/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/swap.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/perf_event.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/mount.h>
46 #include <linux/security.h>
47 #include <linux/syscalls.h>
48 #include <linux/tsacct_kern.h>
49 #include <linux/cn_proc.h>
50 #include <linux/audit.h>
51 #include <linux/tracehook.h>
52 #include <linux/kmod.h>
53 #include <linux/fsnotify.h>
54 #include <linux/fs_struct.h>
55 #include <linux/pipe_fs_i.h>
56 #include <linux/oom.h>
57 #include <linux/compat.h>
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
63 #include <trace/events/task.h>
66 #include <trace/events/sched.h>
68 int suid_dumpable = 0;
70 static LIST_HEAD(formats);
71 static DEFINE_RWLOCK(binfmt_lock);
73 void __register_binfmt(struct linux_binfmt * fmt, int insert)
76 write_lock(&binfmt_lock);
77 insert ? list_add(&fmt->lh, &formats) :
78 list_add_tail(&fmt->lh, &formats);
79 write_unlock(&binfmt_lock);
82 EXPORT_SYMBOL(__register_binfmt);
84 void unregister_binfmt(struct linux_binfmt * fmt)
86 write_lock(&binfmt_lock);
88 write_unlock(&binfmt_lock);
91 EXPORT_SYMBOL(unregister_binfmt);
93 static inline void put_binfmt(struct linux_binfmt * fmt)
95 module_put(fmt->module);
99 * Note that a shared library must be both readable and executable due to
102 * Also note that we take the address to load from from the file itself.
104 SYSCALL_DEFINE1(uselib, const char __user *, library)
107 char *tmp = getname(library);
108 int error = PTR_ERR(tmp);
109 static const struct open_flags uselib_flags = {
110 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
111 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
112 .intent = LOOKUP_OPEN
118 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
120 error = PTR_ERR(file);
125 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
129 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
136 struct linux_binfmt * fmt;
138 read_lock(&binfmt_lock);
139 list_for_each_entry(fmt, &formats, lh) {
140 if (!fmt->load_shlib)
142 if (!try_module_get(fmt->module))
144 read_unlock(&binfmt_lock);
145 error = fmt->load_shlib(file);
146 read_lock(&binfmt_lock);
148 if (error != -ENOEXEC)
151 read_unlock(&binfmt_lock);
161 * The nascent bprm->mm is not visible until exec_mmap() but it can
162 * use a lot of memory, account these pages in current->mm temporary
163 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
164 * change the counter back via acct_arg_size(0).
166 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
168 struct mm_struct *mm = current->mm;
169 long diff = (long)(pages - bprm->vma_pages);
174 bprm->vma_pages = pages;
175 add_mm_counter(mm, MM_ANONPAGES, diff);
178 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
184 #ifdef CONFIG_STACK_GROWSUP
186 ret = expand_downwards(bprm->vma, pos);
191 ret = get_user_pages(current, bprm->mm, pos,
192 1, write, 1, &page, NULL);
197 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
200 acct_arg_size(bprm, size / PAGE_SIZE);
203 * We've historically supported up to 32 pages (ARG_MAX)
204 * of argument strings even with small stacks
210 * Limit to 1/4-th the stack size for the argv+env strings.
212 * - the remaining binfmt code will not run out of stack space,
213 * - the program will have a reasonable amount of stack left
216 rlim = current->signal->rlim;
217 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
226 static void put_arg_page(struct page *page)
231 static void free_arg_page(struct linux_binprm *bprm, int i)
235 static void free_arg_pages(struct linux_binprm *bprm)
239 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
242 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
245 static int __bprm_mm_init(struct linux_binprm *bprm)
248 struct vm_area_struct *vma = NULL;
249 struct mm_struct *mm = bprm->mm;
251 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
255 down_write(&mm->mmap_sem);
259 * Place the stack at the largest stack address the architecture
260 * supports. Later, we'll move this to an appropriate place. We don't
261 * use STACK_TOP because that can depend on attributes which aren't
264 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
265 vma->vm_end = STACK_TOP_MAX;
266 vma->vm_start = vma->vm_end - PAGE_SIZE;
267 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
268 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
269 INIT_LIST_HEAD(&vma->anon_vma_chain);
271 err = insert_vm_struct(mm, vma);
275 mm->stack_vm = mm->total_vm = 1;
276 up_write(&mm->mmap_sem);
277 bprm->p = vma->vm_end - sizeof(void *);
280 up_write(&mm->mmap_sem);
282 kmem_cache_free(vm_area_cachep, vma);
286 static bool valid_arg_len(struct linux_binprm *bprm, long len)
288 return len <= MAX_ARG_STRLEN;
293 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
297 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
302 page = bprm->page[pos / PAGE_SIZE];
303 if (!page && write) {
304 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
307 bprm->page[pos / PAGE_SIZE] = page;
313 static void put_arg_page(struct page *page)
317 static void free_arg_page(struct linux_binprm *bprm, int i)
320 __free_page(bprm->page[i]);
321 bprm->page[i] = NULL;
325 static void free_arg_pages(struct linux_binprm *bprm)
329 for (i = 0; i < MAX_ARG_PAGES; i++)
330 free_arg_page(bprm, i);
333 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
338 static int __bprm_mm_init(struct linux_binprm *bprm)
340 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
344 static bool valid_arg_len(struct linux_binprm *bprm, long len)
346 return len <= bprm->p;
349 #endif /* CONFIG_MMU */
352 * Create a new mm_struct and populate it with a temporary stack
353 * vm_area_struct. We don't have enough context at this point to set the stack
354 * flags, permissions, and offset, so we use temporary values. We'll update
355 * them later in setup_arg_pages().
357 int bprm_mm_init(struct linux_binprm *bprm)
360 struct mm_struct *mm = NULL;
362 bprm->mm = mm = mm_alloc();
367 err = init_new_context(current, mm);
371 err = __bprm_mm_init(bprm);
386 struct user_arg_ptr {
391 const char __user *const __user *native;
393 const compat_uptr_t __user *compat;
398 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
400 const char __user *native;
403 if (unlikely(argv.is_compat)) {
404 compat_uptr_t compat;
406 if (get_user(compat, argv.ptr.compat + nr))
407 return ERR_PTR(-EFAULT);
409 return compat_ptr(compat);
413 if (get_user(native, argv.ptr.native + nr))
414 return ERR_PTR(-EFAULT);
420 * count() counts the number of strings in array ARGV.
422 static int count(struct user_arg_ptr argv, int max)
426 if (argv.ptr.native != NULL) {
428 const char __user *p = get_user_arg_ptr(argv, i);
439 if (fatal_signal_pending(current))
440 return -ERESTARTNOHAND;
448 * 'copy_strings()' copies argument/environment strings from the old
449 * processes's memory to the new process's stack. The call to get_user_pages()
450 * ensures the destination page is created and not swapped out.
452 static int copy_strings(int argc, struct user_arg_ptr argv,
453 struct linux_binprm *bprm)
455 struct page *kmapped_page = NULL;
457 unsigned long kpos = 0;
461 const char __user *str;
466 str = get_user_arg_ptr(argv, argc);
470 len = strnlen_user(str, MAX_ARG_STRLEN);
475 if (!valid_arg_len(bprm, len))
478 /* We're going to work our way backwords. */
484 int offset, bytes_to_copy;
486 if (fatal_signal_pending(current)) {
487 ret = -ERESTARTNOHAND;
492 offset = pos % PAGE_SIZE;
496 bytes_to_copy = offset;
497 if (bytes_to_copy > len)
500 offset -= bytes_to_copy;
501 pos -= bytes_to_copy;
502 str -= bytes_to_copy;
503 len -= bytes_to_copy;
505 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
508 page = get_arg_page(bprm, pos, 1);
515 flush_kernel_dcache_page(kmapped_page);
516 kunmap(kmapped_page);
517 put_arg_page(kmapped_page);
520 kaddr = kmap(kmapped_page);
521 kpos = pos & PAGE_MASK;
522 flush_arg_page(bprm, kpos, kmapped_page);
524 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
533 flush_kernel_dcache_page(kmapped_page);
534 kunmap(kmapped_page);
535 put_arg_page(kmapped_page);
541 * Like copy_strings, but get argv and its values from kernel memory.
543 int copy_strings_kernel(int argc, const char *const *__argv,
544 struct linux_binprm *bprm)
547 mm_segment_t oldfs = get_fs();
548 struct user_arg_ptr argv = {
549 .ptr.native = (const char __user *const __user *)__argv,
553 r = copy_strings(argc, argv, bprm);
558 EXPORT_SYMBOL(copy_strings_kernel);
563 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
564 * the binfmt code determines where the new stack should reside, we shift it to
565 * its final location. The process proceeds as follows:
567 * 1) Use shift to calculate the new vma endpoints.
568 * 2) Extend vma to cover both the old and new ranges. This ensures the
569 * arguments passed to subsequent functions are consistent.
570 * 3) Move vma's page tables to the new range.
571 * 4) Free up any cleared pgd range.
572 * 5) Shrink the vma to cover only the new range.
574 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
576 struct mm_struct *mm = vma->vm_mm;
577 unsigned long old_start = vma->vm_start;
578 unsigned long old_end = vma->vm_end;
579 unsigned long length = old_end - old_start;
580 unsigned long new_start = old_start - shift;
581 unsigned long new_end = old_end - shift;
582 struct mmu_gather tlb;
584 BUG_ON(new_start > new_end);
587 * ensure there are no vmas between where we want to go
590 if (vma != find_vma(mm, new_start))
594 * cover the whole range: [new_start, old_end)
596 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
600 * move the page tables downwards, on failure we rely on
601 * process cleanup to remove whatever mess we made.
603 if (length != move_page_tables(vma, old_start,
604 vma, new_start, length, false))
608 tlb_gather_mmu(&tlb, mm, 0);
609 if (new_end > old_start) {
611 * when the old and new regions overlap clear from new_end.
613 free_pgd_range(&tlb, new_end, old_end, new_end,
614 vma->vm_next ? vma->vm_next->vm_start : 0);
617 * otherwise, clean from old_start; this is done to not touch
618 * the address space in [new_end, old_start) some architectures
619 * have constraints on va-space that make this illegal (IA64) -
620 * for the others its just a little faster.
622 free_pgd_range(&tlb, old_start, old_end, new_end,
623 vma->vm_next ? vma->vm_next->vm_start : 0);
625 tlb_finish_mmu(&tlb, new_end, old_end);
628 * Shrink the vma to just the new range. Always succeeds.
630 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
636 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
637 * the stack is optionally relocated, and some extra space is added.
639 int setup_arg_pages(struct linux_binprm *bprm,
640 unsigned long stack_top,
641 int executable_stack)
644 unsigned long stack_shift;
645 struct mm_struct *mm = current->mm;
646 struct vm_area_struct *vma = bprm->vma;
647 struct vm_area_struct *prev = NULL;
648 unsigned long vm_flags;
649 unsigned long stack_base;
650 unsigned long stack_size;
651 unsigned long stack_expand;
652 unsigned long rlim_stack;
654 #ifdef CONFIG_STACK_GROWSUP
655 /* Limit stack size to 1GB */
656 stack_base = rlimit_max(RLIMIT_STACK);
657 if (stack_base > (1 << 30))
658 stack_base = 1 << 30;
660 /* Make sure we didn't let the argument array grow too large. */
661 if (vma->vm_end - vma->vm_start > stack_base)
664 stack_base = PAGE_ALIGN(stack_top - stack_base);
666 stack_shift = vma->vm_start - stack_base;
667 mm->arg_start = bprm->p - stack_shift;
668 bprm->p = vma->vm_end - stack_shift;
670 stack_top = arch_align_stack(stack_top);
671 stack_top = PAGE_ALIGN(stack_top);
673 if (unlikely(stack_top < mmap_min_addr) ||
674 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
677 stack_shift = vma->vm_end - stack_top;
679 bprm->p -= stack_shift;
680 mm->arg_start = bprm->p;
684 bprm->loader -= stack_shift;
685 bprm->exec -= stack_shift;
687 down_write(&mm->mmap_sem);
688 vm_flags = VM_STACK_FLAGS;
691 * Adjust stack execute permissions; explicitly enable for
692 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
693 * (arch default) otherwise.
695 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
697 else if (executable_stack == EXSTACK_DISABLE_X)
698 vm_flags &= ~VM_EXEC;
699 vm_flags |= mm->def_flags;
700 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
702 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
708 /* Move stack pages down in memory. */
710 ret = shift_arg_pages(vma, stack_shift);
715 /* mprotect_fixup is overkill to remove the temporary stack flags */
716 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
718 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
719 stack_size = vma->vm_end - vma->vm_start;
721 * Align this down to a page boundary as expand_stack
724 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
725 #ifdef CONFIG_STACK_GROWSUP
726 if (stack_size + stack_expand > rlim_stack)
727 stack_base = vma->vm_start + rlim_stack;
729 stack_base = vma->vm_end + stack_expand;
731 if (stack_size + stack_expand > rlim_stack)
732 stack_base = vma->vm_end - rlim_stack;
734 stack_base = vma->vm_start - stack_expand;
736 current->mm->start_stack = bprm->p;
737 ret = expand_stack(vma, stack_base);
742 up_write(&mm->mmap_sem);
745 EXPORT_SYMBOL(setup_arg_pages);
747 #endif /* CONFIG_MMU */
749 struct file *open_exec(const char *name)
753 static const struct open_flags open_exec_flags = {
754 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
755 .acc_mode = MAY_EXEC | MAY_OPEN,
756 .intent = LOOKUP_OPEN
759 file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
764 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
767 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
772 err = deny_write_access(file);
783 EXPORT_SYMBOL(open_exec);
785 int kernel_read(struct file *file, loff_t offset,
786 char *addr, unsigned long count)
794 /* The cast to a user pointer is valid due to the set_fs() */
795 result = vfs_read(file, (void __user *)addr, count, &pos);
800 EXPORT_SYMBOL(kernel_read);
802 static int exec_mmap(struct mm_struct *mm)
804 struct task_struct *tsk;
805 struct mm_struct * old_mm, *active_mm;
807 /* Notify parent that we're no longer interested in the old VM */
809 old_mm = current->mm;
810 mm_release(tsk, old_mm);
815 * Make sure that if there is a core dump in progress
816 * for the old mm, we get out and die instead of going
817 * through with the exec. We must hold mmap_sem around
818 * checking core_state and changing tsk->mm.
820 down_read(&old_mm->mmap_sem);
821 if (unlikely(old_mm->core_state)) {
822 up_read(&old_mm->mmap_sem);
827 active_mm = tsk->active_mm;
830 activate_mm(active_mm, mm);
832 arch_pick_mmap_layout(mm);
834 up_read(&old_mm->mmap_sem);
835 BUG_ON(active_mm != old_mm);
836 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
837 mm_update_next_owner(old_mm);
846 * This function makes sure the current process has its own signal table,
847 * so that flush_signal_handlers can later reset the handlers without
848 * disturbing other processes. (Other processes might share the signal
849 * table via the CLONE_SIGHAND option to clone().)
851 static int de_thread(struct task_struct *tsk)
853 struct signal_struct *sig = tsk->signal;
854 struct sighand_struct *oldsighand = tsk->sighand;
855 spinlock_t *lock = &oldsighand->siglock;
857 if (thread_group_empty(tsk))
858 goto no_thread_group;
861 * Kill all other threads in the thread group.
864 if (signal_group_exit(sig)) {
866 * Another group action in progress, just
867 * return so that the signal is processed.
869 spin_unlock_irq(lock);
873 sig->group_exit_task = tsk;
874 sig->notify_count = zap_other_threads(tsk);
875 if (!thread_group_leader(tsk))
878 while (sig->notify_count) {
879 __set_current_state(TASK_UNINTERRUPTIBLE);
880 spin_unlock_irq(lock);
884 spin_unlock_irq(lock);
887 * At this point all other threads have exited, all we have to
888 * do is to wait for the thread group leader to become inactive,
889 * and to assume its PID:
891 if (!thread_group_leader(tsk)) {
892 struct task_struct *leader = tsk->group_leader;
894 sig->notify_count = -1; /* for exit_notify() */
896 write_lock_irq(&tasklist_lock);
897 if (likely(leader->exit_state))
899 __set_current_state(TASK_UNINTERRUPTIBLE);
900 write_unlock_irq(&tasklist_lock);
905 * The only record we have of the real-time age of a
906 * process, regardless of execs it's done, is start_time.
907 * All the past CPU time is accumulated in signal_struct
908 * from sister threads now dead. But in this non-leader
909 * exec, nothing survives from the original leader thread,
910 * whose birth marks the true age of this process now.
911 * When we take on its identity by switching to its PID, we
912 * also take its birthdate (always earlier than our own).
914 tsk->start_time = leader->start_time;
916 BUG_ON(!same_thread_group(leader, tsk));
917 BUG_ON(has_group_leader_pid(tsk));
919 * An exec() starts a new thread group with the
920 * TGID of the previous thread group. Rehash the
921 * two threads with a switched PID, and release
922 * the former thread group leader:
925 /* Become a process group leader with the old leader's pid.
926 * The old leader becomes a thread of the this thread group.
927 * Note: The old leader also uses this pid until release_task
928 * is called. Odd but simple and correct.
930 detach_pid(tsk, PIDTYPE_PID);
931 tsk->pid = leader->pid;
932 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
933 transfer_pid(leader, tsk, PIDTYPE_PGID);
934 transfer_pid(leader, tsk, PIDTYPE_SID);
936 list_replace_rcu(&leader->tasks, &tsk->tasks);
937 list_replace_init(&leader->sibling, &tsk->sibling);
939 tsk->group_leader = tsk;
940 leader->group_leader = tsk;
942 tsk->exit_signal = SIGCHLD;
943 leader->exit_signal = -1;
945 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
946 leader->exit_state = EXIT_DEAD;
949 * We are going to release_task()->ptrace_unlink() silently,
950 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
951 * the tracer wont't block again waiting for this thread.
953 if (unlikely(leader->ptrace))
954 __wake_up_parent(leader, leader->parent);
955 write_unlock_irq(&tasklist_lock);
957 release_task(leader);
960 sig->group_exit_task = NULL;
961 sig->notify_count = 0;
964 /* we have changed execution domain */
965 tsk->exit_signal = SIGCHLD;
968 flush_itimer_signals();
970 if (atomic_read(&oldsighand->count) != 1) {
971 struct sighand_struct *newsighand;
973 * This ->sighand is shared with the CLONE_SIGHAND
974 * but not CLONE_THREAD task, switch to the new one.
976 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
980 atomic_set(&newsighand->count, 1);
981 memcpy(newsighand->action, oldsighand->action,
982 sizeof(newsighand->action));
984 write_lock_irq(&tasklist_lock);
985 spin_lock(&oldsighand->siglock);
986 rcu_assign_pointer(tsk->sighand, newsighand);
987 spin_unlock(&oldsighand->siglock);
988 write_unlock_irq(&tasklist_lock);
990 __cleanup_sighand(oldsighand);
993 BUG_ON(!thread_group_leader(tsk));
998 * These functions flushes out all traces of the currently running executable
999 * so that a new one can be started
1001 static void flush_old_files(struct files_struct * files)
1004 struct fdtable *fdt;
1006 spin_lock(&files->file_lock);
1008 unsigned long set, i;
1011 i = j * BITS_PER_LONG;
1012 fdt = files_fdtable(files);
1013 if (i >= fdt->max_fds)
1015 set = fdt->close_on_exec[j];
1018 fdt->close_on_exec[j] = 0;
1019 spin_unlock(&files->file_lock);
1020 for ( ; set ; i++,set >>= 1) {
1025 spin_lock(&files->file_lock);
1028 spin_unlock(&files->file_lock);
1031 char *get_task_comm(char *buf, struct task_struct *tsk)
1033 /* buf must be at least sizeof(tsk->comm) in size */
1035 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1039 EXPORT_SYMBOL_GPL(get_task_comm);
1041 void set_task_comm(struct task_struct *tsk, char *buf)
1045 trace_task_rename(tsk, buf);
1048 * Threads may access current->comm without holding
1049 * the task lock, so write the string carefully.
1050 * Readers without a lock may see incomplete new
1051 * names but are safe from non-terminating string reads.
1053 memset(tsk->comm, 0, TASK_COMM_LEN);
1055 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1057 perf_event_comm(tsk);
1060 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1064 /* Copies the binary name from after last slash */
1065 for (i = 0; (ch = *(fn++)) != '\0';) {
1067 i = 0; /* overwrite what we wrote */
1075 int flush_old_exec(struct linux_binprm * bprm)
1080 * Make sure we have a private signal table and that
1081 * we are unassociated from the previous thread group.
1083 retval = de_thread(current);
1087 set_mm_exe_file(bprm->mm, bprm->file);
1089 filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1091 * Release all of the old mmap stuff
1093 acct_arg_size(bprm, 0);
1094 retval = exec_mmap(bprm->mm);
1098 bprm->mm = NULL; /* We're using it now */
1101 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD);
1103 current->personality &= ~bprm->per_clear;
1110 EXPORT_SYMBOL(flush_old_exec);
1112 void would_dump(struct linux_binprm *bprm, struct file *file)
1114 if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1115 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1117 EXPORT_SYMBOL(would_dump);
1119 void setup_new_exec(struct linux_binprm * bprm)
1121 arch_pick_mmap_layout(current->mm);
1123 /* This is the point of no return */
1124 current->sas_ss_sp = current->sas_ss_size = 0;
1126 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1127 set_dumpable(current->mm, 1);
1129 set_dumpable(current->mm, suid_dumpable);
1131 set_task_comm(current, bprm->tcomm);
1133 /* Set the new mm task size. We have to do that late because it may
1134 * depend on TIF_32BIT which is only updated in flush_thread() on
1135 * some architectures like powerpc
1137 current->mm->task_size = TASK_SIZE;
1139 /* install the new credentials */
1140 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1141 !gid_eq(bprm->cred->gid, current_egid())) {
1142 current->pdeath_signal = 0;
1144 would_dump(bprm, bprm->file);
1145 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1146 set_dumpable(current->mm, suid_dumpable);
1150 * Flush performance counters when crossing a
1153 if (!get_dumpable(current->mm))
1154 perf_event_exit_task(current);
1156 /* An exec changes our domain. We are no longer part of the thread
1159 current->self_exec_id++;
1161 flush_signal_handlers(current, 0);
1162 flush_old_files(current->files);
1164 EXPORT_SYMBOL(setup_new_exec);
1167 * Prepare credentials and lock ->cred_guard_mutex.
1168 * install_exec_creds() commits the new creds and drops the lock.
1169 * Or, if exec fails before, free_bprm() should release ->cred and
1172 int prepare_bprm_creds(struct linux_binprm *bprm)
1174 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1175 return -ERESTARTNOINTR;
1177 bprm->cred = prepare_exec_creds();
1178 if (likely(bprm->cred))
1181 mutex_unlock(¤t->signal->cred_guard_mutex);
1185 void free_bprm(struct linux_binprm *bprm)
1187 free_arg_pages(bprm);
1189 mutex_unlock(¤t->signal->cred_guard_mutex);
1190 abort_creds(bprm->cred);
1196 * install the new credentials for this executable
1198 void install_exec_creds(struct linux_binprm *bprm)
1200 security_bprm_committing_creds(bprm);
1202 commit_creds(bprm->cred);
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
1219 static int check_unsafe_exec(struct linux_binprm *bprm)
1221 struct task_struct *p = current, *t;
1226 if (p->ptrace & PT_PTRACE_CAP)
1227 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1229 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1233 * This isn't strictly necessary, but it makes it harder for LSMs to
1236 if (current->no_new_privs)
1237 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1240 spin_lock(&p->fs->lock);
1242 for (t = next_thread(p); t != p; t = next_thread(t)) {
1248 if (p->fs->users > n_fs) {
1249 bprm->unsafe |= LSM_UNSAFE_SHARE;
1252 if (!p->fs->in_exec) {
1257 spin_unlock(&p->fs->lock);
1263 * Fill the binprm structure from the inode.
1264 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1266 * This may be called multiple times for binary chains (scripts for example).
1268 int prepare_binprm(struct linux_binprm *bprm)
1271 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1274 mode = inode->i_mode;
1275 if (bprm->file->f_op == NULL)
1278 /* clear any previous set[ug]id data from a previous binary */
1279 bprm->cred->euid = current_euid();
1280 bprm->cred->egid = current_egid();
1282 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1283 !current->no_new_privs) {
1285 if (mode & S_ISUID) {
1286 if (!kuid_has_mapping(bprm->cred->user_ns, inode->i_uid))
1288 bprm->per_clear |= PER_CLEAR_ON_SETID;
1289 bprm->cred->euid = inode->i_uid;
1295 * If setgid is set but no group execute bit then this
1296 * is a candidate for mandatory locking, not a setgid
1299 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1300 if (!kgid_has_mapping(bprm->cred->user_ns, inode->i_gid))
1302 bprm->per_clear |= PER_CLEAR_ON_SETID;
1303 bprm->cred->egid = inode->i_gid;
1307 /* fill in binprm security blob */
1308 retval = security_bprm_set_creds(bprm);
1311 bprm->cred_prepared = 1;
1313 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1314 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1317 EXPORT_SYMBOL(prepare_binprm);
1320 * Arguments are '\0' separated strings found at the location bprm->p
1321 * points to; chop off the first by relocating brpm->p to right after
1322 * the first '\0' encountered.
1324 int remove_arg_zero(struct linux_binprm *bprm)
1327 unsigned long offset;
1335 offset = bprm->p & ~PAGE_MASK;
1336 page = get_arg_page(bprm, bprm->p, 0);
1341 kaddr = kmap_atomic(page);
1343 for (; offset < PAGE_SIZE && kaddr[offset];
1344 offset++, bprm->p++)
1347 kunmap_atomic(kaddr);
1350 if (offset == PAGE_SIZE)
1351 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1352 } while (offset == PAGE_SIZE);
1361 EXPORT_SYMBOL(remove_arg_zero);
1364 * cycle the list of binary formats handler, until one recognizes the image
1366 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1368 unsigned int depth = bprm->recursion_depth;
1370 struct linux_binfmt *fmt;
1371 pid_t old_pid, old_vpid;
1373 retval = security_bprm_check(bprm);
1377 retval = audit_bprm(bprm);
1381 /* Need to fetch pid before load_binary changes it */
1382 old_pid = current->pid;
1384 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1388 for (try=0; try<2; try++) {
1389 read_lock(&binfmt_lock);
1390 list_for_each_entry(fmt, &formats, lh) {
1391 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1394 if (!try_module_get(fmt->module))
1396 read_unlock(&binfmt_lock);
1397 retval = fn(bprm, regs);
1399 * Restore the depth counter to its starting value
1400 * in this call, so we don't have to rely on every
1401 * load_binary function to restore it on return.
1403 bprm->recursion_depth = depth;
1406 trace_sched_process_exec(current, old_pid, bprm);
1407 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1410 allow_write_access(bprm->file);
1414 current->did_exec = 1;
1415 proc_exec_connector(current);
1418 read_lock(&binfmt_lock);
1420 if (retval != -ENOEXEC || bprm->mm == NULL)
1423 read_unlock(&binfmt_lock);
1427 read_unlock(&binfmt_lock);
1428 #ifdef CONFIG_MODULES
1429 if (retval != -ENOEXEC || bprm->mm == NULL) {
1432 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1433 if (printable(bprm->buf[0]) &&
1434 printable(bprm->buf[1]) &&
1435 printable(bprm->buf[2]) &&
1436 printable(bprm->buf[3]))
1437 break; /* -ENOEXEC */
1439 break; /* -ENOEXEC */
1440 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1449 EXPORT_SYMBOL(search_binary_handler);
1452 * sys_execve() executes a new program.
1454 static int do_execve_common(const char *filename,
1455 struct user_arg_ptr argv,
1456 struct user_arg_ptr envp,
1457 struct pt_regs *regs)
1459 struct linux_binprm *bprm;
1461 struct files_struct *displaced;
1464 const struct cred *cred = current_cred();
1467 * We move the actual failure in case of RLIMIT_NPROC excess from
1468 * set*uid() to execve() because too many poorly written programs
1469 * don't check setuid() return code. Here we additionally recheck
1470 * whether NPROC limit is still exceeded.
1472 if ((current->flags & PF_NPROC_EXCEEDED) &&
1473 atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1478 /* We're below the limit (still or again), so we don't want to make
1479 * further execve() calls fail. */
1480 current->flags &= ~PF_NPROC_EXCEEDED;
1482 retval = unshare_files(&displaced);
1487 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1491 retval = prepare_bprm_creds(bprm);
1495 retval = check_unsafe_exec(bprm);
1498 clear_in_exec = retval;
1499 current->in_execve = 1;
1501 file = open_exec(filename);
1502 retval = PTR_ERR(file);
1509 bprm->filename = filename;
1510 bprm->interp = filename;
1512 retval = bprm_mm_init(bprm);
1516 bprm->argc = count(argv, MAX_ARG_STRINGS);
1517 if ((retval = bprm->argc) < 0)
1520 bprm->envc = count(envp, MAX_ARG_STRINGS);
1521 if ((retval = bprm->envc) < 0)
1524 retval = prepare_binprm(bprm);
1528 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1532 bprm->exec = bprm->p;
1533 retval = copy_strings(bprm->envc, envp, bprm);
1537 retval = copy_strings(bprm->argc, argv, bprm);
1541 retval = search_binary_handler(bprm,regs);
1545 /* execve succeeded */
1546 current->fs->in_exec = 0;
1547 current->in_execve = 0;
1548 acct_update_integrals(current);
1551 put_files_struct(displaced);
1556 acct_arg_size(bprm, 0);
1562 allow_write_access(bprm->file);
1568 current->fs->in_exec = 0;
1569 current->in_execve = 0;
1576 reset_files_struct(displaced);
1581 int do_execve(const char *filename,
1582 const char __user *const __user *__argv,
1583 const char __user *const __user *__envp,
1584 struct pt_regs *regs)
1586 struct user_arg_ptr argv = { .ptr.native = __argv };
1587 struct user_arg_ptr envp = { .ptr.native = __envp };
1588 return do_execve_common(filename, argv, envp, regs);
1591 #ifdef CONFIG_COMPAT
1592 int compat_do_execve(const char *filename,
1593 const compat_uptr_t __user *__argv,
1594 const compat_uptr_t __user *__envp,
1595 struct pt_regs *regs)
1597 struct user_arg_ptr argv = {
1599 .ptr.compat = __argv,
1601 struct user_arg_ptr envp = {
1603 .ptr.compat = __envp,
1605 return do_execve_common(filename, argv, envp, regs);
1609 void set_binfmt(struct linux_binfmt *new)
1611 struct mm_struct *mm = current->mm;
1614 module_put(mm->binfmt->module);
1618 __module_get(new->module);
1621 EXPORT_SYMBOL(set_binfmt);
1624 * set_dumpable converts traditional three-value dumpable to two flags and
1625 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1626 * these bits are not changed atomically. So get_dumpable can observe the
1627 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1628 * return either old dumpable or new one by paying attention to the order of
1629 * modifying the bits.
1631 * dumpable | mm->flags (binary)
1632 * old new | initial interim final
1633 * ---------+-----------------------
1641 * (*) get_dumpable regards interim value of 10 as 11.
1643 void set_dumpable(struct mm_struct *mm, int value)
1646 case SUID_DUMPABLE_DISABLED:
1647 clear_bit(MMF_DUMPABLE, &mm->flags);
1649 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1651 case SUID_DUMPABLE_ENABLED:
1652 set_bit(MMF_DUMPABLE, &mm->flags);
1654 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1656 case SUID_DUMPABLE_SAFE:
1657 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1659 set_bit(MMF_DUMPABLE, &mm->flags);
1664 int __get_dumpable(unsigned long mm_flags)
1668 ret = mm_flags & MMF_DUMPABLE_MASK;
1669 return (ret > SUID_DUMPABLE_ENABLED) ? SUID_DUMPABLE_SAFE : ret;
1672 int get_dumpable(struct mm_struct *mm)
1674 return __get_dumpable(mm->flags);
1677 #ifdef __ARCH_WANT_SYS_EXECVE
1678 SYSCALL_DEFINE3(execve,
1679 const char __user *, filename,
1680 const char __user *const __user *, argv,
1681 const char __user *const __user *, envp)
1683 const char *path = getname(filename);
1684 int error = PTR_ERR(path);
1685 if (!IS_ERR(path)) {
1686 error = do_execve(path, argv, envp, current_pt_regs());
1691 #ifdef CONFIG_COMPAT
1692 asmlinkage long compat_sys_execve(const char __user * filename,
1693 const compat_uptr_t __user * argv,
1694 const compat_uptr_t __user * envp)
1696 const char *path = getname(filename);
1697 int error = PTR_ERR(path);
1698 if (!IS_ERR(path)) {
1699 error = compat_do_execve(path, argv, envp, current_pt_regs());
1707 #ifdef __ARCH_WANT_KERNEL_EXECVE
1708 int kernel_execve(const char *filename,
1709 const char *const argv[],
1710 const char *const envp[])
1712 struct pt_regs regs;
1715 memset(®s, 0, sizeof(struct pt_regs));
1716 ret = do_execve(filename,
1717 (const char __user *const __user *)argv,
1718 (const char __user *const __user *)envp, ®s);
1723 * We were successful. We won't be returning to our caller, but
1724 * instead to user space by manipulating the kernel stack.
1726 ret_from_kernel_execve(current_pt_regs(), ®s);