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>
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 write_lock(&binfmt_lock);
78 insert ? list_add(&fmt->lh, &formats) :
79 list_add_tail(&fmt->lh, &formats);
80 write_unlock(&binfmt_lock);
83 EXPORT_SYMBOL(__register_binfmt);
85 void unregister_binfmt(struct linux_binfmt * fmt)
87 write_lock(&binfmt_lock);
89 write_unlock(&binfmt_lock);
92 EXPORT_SYMBOL(unregister_binfmt);
94 static inline void put_binfmt(struct linux_binfmt * fmt)
96 module_put(fmt->module);
100 * Note that a shared library must be both readable and executable due to
103 * Also note that we take the address to load from from the file itself.
105 SYSCALL_DEFINE1(uselib, const char __user *, library)
108 struct filename *tmp = getname(library);
109 int error = PTR_ERR(tmp);
110 static const struct open_flags uselib_flags = {
111 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
112 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
113 .intent = LOOKUP_OPEN
119 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
121 error = PTR_ERR(file);
126 if (!S_ISREG(file_inode(file)->i_mode))
130 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
137 struct linux_binfmt * fmt;
139 read_lock(&binfmt_lock);
140 list_for_each_entry(fmt, &formats, lh) {
141 if (!fmt->load_shlib)
143 if (!try_module_get(fmt->module))
145 read_unlock(&binfmt_lock);
146 error = fmt->load_shlib(file);
147 read_lock(&binfmt_lock);
149 if (error != -ENOEXEC)
152 read_unlock(&binfmt_lock);
162 * The nascent bprm->mm is not visible until exec_mmap() but it can
163 * use a lot of memory, account these pages in current->mm temporary
164 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
165 * change the counter back via acct_arg_size(0).
167 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
169 struct mm_struct *mm = current->mm;
170 long diff = (long)(pages - bprm->vma_pages);
175 bprm->vma_pages = pages;
176 add_mm_counter(mm, MM_ANONPAGES, diff);
179 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
185 #ifdef CONFIG_STACK_GROWSUP
187 ret = expand_downwards(bprm->vma, pos);
192 ret = get_user_pages(current, bprm->mm, pos,
193 1, write, 1, &page, NULL);
198 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
201 acct_arg_size(bprm, size / PAGE_SIZE);
204 * We've historically supported up to 32 pages (ARG_MAX)
205 * of argument strings even with small stacks
211 * Limit to 1/4-th the stack size for the argv+env strings.
213 * - the remaining binfmt code will not run out of stack space,
214 * - the program will have a reasonable amount of stack left
217 rlim = current->signal->rlim;
218 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
227 static void put_arg_page(struct page *page)
232 static void free_arg_page(struct linux_binprm *bprm, int i)
236 static void free_arg_pages(struct linux_binprm *bprm)
240 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
243 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
246 static int __bprm_mm_init(struct linux_binprm *bprm)
249 struct vm_area_struct *vma = NULL;
250 struct mm_struct *mm = bprm->mm;
252 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
256 down_write(&mm->mmap_sem);
260 * Place the stack at the largest stack address the architecture
261 * supports. Later, we'll move this to an appropriate place. We don't
262 * use STACK_TOP because that can depend on attributes which aren't
265 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
266 vma->vm_end = STACK_TOP_MAX;
267 vma->vm_start = vma->vm_end - PAGE_SIZE;
268 vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
269 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
270 INIT_LIST_HEAD(&vma->anon_vma_chain);
272 err = insert_vm_struct(mm, vma);
276 mm->stack_vm = mm->total_vm = 1;
277 up_write(&mm->mmap_sem);
278 bprm->p = vma->vm_end - sizeof(void *);
281 up_write(&mm->mmap_sem);
283 kmem_cache_free(vm_area_cachep, vma);
287 static bool valid_arg_len(struct linux_binprm *bprm, long len)
289 return len <= MAX_ARG_STRLEN;
294 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
298 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
303 page = bprm->page[pos / PAGE_SIZE];
304 if (!page && write) {
305 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
308 bprm->page[pos / PAGE_SIZE] = page;
314 static void put_arg_page(struct page *page)
318 static void free_arg_page(struct linux_binprm *bprm, int i)
321 __free_page(bprm->page[i]);
322 bprm->page[i] = NULL;
326 static void free_arg_pages(struct linux_binprm *bprm)
330 for (i = 0; i < MAX_ARG_PAGES; i++)
331 free_arg_page(bprm, i);
334 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
339 static int __bprm_mm_init(struct linux_binprm *bprm)
341 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
345 static bool valid_arg_len(struct linux_binprm *bprm, long len)
347 return len <= bprm->p;
350 #endif /* CONFIG_MMU */
353 * Create a new mm_struct and populate it with a temporary stack
354 * vm_area_struct. We don't have enough context at this point to set the stack
355 * flags, permissions, and offset, so we use temporary values. We'll update
356 * them later in setup_arg_pages().
358 static int bprm_mm_init(struct linux_binprm *bprm)
361 struct mm_struct *mm = NULL;
363 bprm->mm = mm = mm_alloc();
368 err = init_new_context(current, mm);
372 err = __bprm_mm_init(bprm);
387 struct user_arg_ptr {
392 const char __user *const __user *native;
394 const compat_uptr_t __user *compat;
399 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
401 const char __user *native;
404 if (unlikely(argv.is_compat)) {
405 compat_uptr_t compat;
407 if (get_user(compat, argv.ptr.compat + nr))
408 return ERR_PTR(-EFAULT);
410 return compat_ptr(compat);
414 if (get_user(native, argv.ptr.native + nr))
415 return ERR_PTR(-EFAULT);
421 * count() counts the number of strings in array ARGV.
423 static int count(struct user_arg_ptr argv, int max)
427 if (argv.ptr.native != NULL) {
429 const char __user *p = get_user_arg_ptr(argv, i);
441 if (fatal_signal_pending(current))
442 return -ERESTARTNOHAND;
450 * 'copy_strings()' copies argument/environment strings from the old
451 * processes's memory to the new process's stack. The call to get_user_pages()
452 * ensures the destination page is created and not swapped out.
454 static int copy_strings(int argc, struct user_arg_ptr argv,
455 struct linux_binprm *bprm)
457 struct page *kmapped_page = NULL;
459 unsigned long kpos = 0;
463 const char __user *str;
468 str = get_user_arg_ptr(argv, argc);
472 len = strnlen_user(str, MAX_ARG_STRLEN);
477 if (!valid_arg_len(bprm, len))
480 /* We're going to work our way backwords. */
486 int offset, bytes_to_copy;
488 if (fatal_signal_pending(current)) {
489 ret = -ERESTARTNOHAND;
494 offset = pos % PAGE_SIZE;
498 bytes_to_copy = offset;
499 if (bytes_to_copy > len)
502 offset -= bytes_to_copy;
503 pos -= bytes_to_copy;
504 str -= bytes_to_copy;
505 len -= bytes_to_copy;
507 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
510 page = get_arg_page(bprm, pos, 1);
517 flush_kernel_dcache_page(kmapped_page);
518 kunmap(kmapped_page);
519 put_arg_page(kmapped_page);
522 kaddr = kmap(kmapped_page);
523 kpos = pos & PAGE_MASK;
524 flush_arg_page(bprm, kpos, kmapped_page);
526 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
535 flush_kernel_dcache_page(kmapped_page);
536 kunmap(kmapped_page);
537 put_arg_page(kmapped_page);
543 * Like copy_strings, but get argv and its values from kernel memory.
545 int copy_strings_kernel(int argc, const char *const *__argv,
546 struct linux_binprm *bprm)
549 mm_segment_t oldfs = get_fs();
550 struct user_arg_ptr argv = {
551 .ptr.native = (const char __user *const __user *)__argv,
555 r = copy_strings(argc, argv, bprm);
560 EXPORT_SYMBOL(copy_strings_kernel);
565 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
566 * the binfmt code determines where the new stack should reside, we shift it to
567 * its final location. The process proceeds as follows:
569 * 1) Use shift to calculate the new vma endpoints.
570 * 2) Extend vma to cover both the old and new ranges. This ensures the
571 * arguments passed to subsequent functions are consistent.
572 * 3) Move vma's page tables to the new range.
573 * 4) Free up any cleared pgd range.
574 * 5) Shrink the vma to cover only the new range.
576 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
578 struct mm_struct *mm = vma->vm_mm;
579 unsigned long old_start = vma->vm_start;
580 unsigned long old_end = vma->vm_end;
581 unsigned long length = old_end - old_start;
582 unsigned long new_start = old_start - shift;
583 unsigned long new_end = old_end - shift;
584 struct mmu_gather tlb;
586 BUG_ON(new_start > new_end);
589 * ensure there are no vmas between where we want to go
592 if (vma != find_vma(mm, new_start))
596 * cover the whole range: [new_start, old_end)
598 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
602 * move the page tables downwards, on failure we rely on
603 * process cleanup to remove whatever mess we made.
605 if (length != move_page_tables(vma, old_start,
606 vma, new_start, length, false))
610 tlb_gather_mmu(&tlb, mm, 0);
611 if (new_end > old_start) {
613 * when the old and new regions overlap clear from new_end.
615 free_pgd_range(&tlb, new_end, old_end, new_end,
616 vma->vm_next ? vma->vm_next->vm_start : 0);
619 * otherwise, clean from old_start; this is done to not touch
620 * the address space in [new_end, old_start) some architectures
621 * have constraints on va-space that make this illegal (IA64) -
622 * for the others its just a little faster.
624 free_pgd_range(&tlb, old_start, old_end, new_end,
625 vma->vm_next ? vma->vm_next->vm_start : 0);
627 tlb_finish_mmu(&tlb, new_end, old_end);
630 * Shrink the vma to just the new range. Always succeeds.
632 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
638 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
639 * the stack is optionally relocated, and some extra space is added.
641 int setup_arg_pages(struct linux_binprm *bprm,
642 unsigned long stack_top,
643 int executable_stack)
646 unsigned long stack_shift;
647 struct mm_struct *mm = current->mm;
648 struct vm_area_struct *vma = bprm->vma;
649 struct vm_area_struct *prev = NULL;
650 unsigned long vm_flags;
651 unsigned long stack_base;
652 unsigned long stack_size;
653 unsigned long stack_expand;
654 unsigned long rlim_stack;
656 #ifdef CONFIG_STACK_GROWSUP
657 /* Limit stack size to 1GB */
658 stack_base = rlimit_max(RLIMIT_STACK);
659 if (stack_base > (1 << 30))
660 stack_base = 1 << 30;
662 /* Make sure we didn't let the argument array grow too large. */
663 if (vma->vm_end - vma->vm_start > stack_base)
666 stack_base = PAGE_ALIGN(stack_top - stack_base);
668 stack_shift = vma->vm_start - stack_base;
669 mm->arg_start = bprm->p - stack_shift;
670 bprm->p = vma->vm_end - stack_shift;
672 stack_top = arch_align_stack(stack_top);
673 stack_top = PAGE_ALIGN(stack_top);
675 if (unlikely(stack_top < mmap_min_addr) ||
676 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
679 stack_shift = vma->vm_end - stack_top;
681 bprm->p -= stack_shift;
682 mm->arg_start = bprm->p;
686 bprm->loader -= stack_shift;
687 bprm->exec -= stack_shift;
689 down_write(&mm->mmap_sem);
690 vm_flags = VM_STACK_FLAGS;
693 * Adjust stack execute permissions; explicitly enable for
694 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
695 * (arch default) otherwise.
697 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
699 else if (executable_stack == EXSTACK_DISABLE_X)
700 vm_flags &= ~VM_EXEC;
701 vm_flags |= mm->def_flags;
702 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
704 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
710 /* Move stack pages down in memory. */
712 ret = shift_arg_pages(vma, stack_shift);
717 /* mprotect_fixup is overkill to remove the temporary stack flags */
718 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
720 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
721 stack_size = vma->vm_end - vma->vm_start;
723 * Align this down to a page boundary as expand_stack
726 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
727 #ifdef CONFIG_STACK_GROWSUP
728 if (stack_size + stack_expand > rlim_stack)
729 stack_base = vma->vm_start + rlim_stack;
731 stack_base = vma->vm_end + stack_expand;
733 if (stack_size + stack_expand > rlim_stack)
734 stack_base = vma->vm_end - rlim_stack;
736 stack_base = vma->vm_start - stack_expand;
738 current->mm->start_stack = bprm->p;
739 ret = expand_stack(vma, stack_base);
744 up_write(&mm->mmap_sem);
747 EXPORT_SYMBOL(setup_arg_pages);
749 #endif /* CONFIG_MMU */
751 struct file *open_exec(const char *name)
755 struct filename tmp = { .name = 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
762 file = do_filp_open(AT_FDCWD, &tmp, &open_exec_flags, LOOKUP_FOLLOW);
767 if (!S_ISREG(file_inode(file)->i_mode))
770 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
775 err = deny_write_access(file);
786 EXPORT_SYMBOL(open_exec);
788 int kernel_read(struct file *file, loff_t offset,
789 char *addr, unsigned long count)
797 /* The cast to a user pointer is valid due to the set_fs() */
798 result = vfs_read(file, (void __user *)addr, count, &pos);
803 EXPORT_SYMBOL(kernel_read);
805 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
807 ssize_t res = file->f_op->read(file, (void __user *)addr, len, &pos);
809 flush_icache_range(addr, addr + len);
812 EXPORT_SYMBOL(read_code);
814 static int exec_mmap(struct mm_struct *mm)
816 struct task_struct *tsk;
817 struct mm_struct * old_mm, *active_mm;
819 /* Notify parent that we're no longer interested in the old VM */
821 old_mm = current->mm;
822 mm_release(tsk, old_mm);
827 * Make sure that if there is a core dump in progress
828 * for the old mm, we get out and die instead of going
829 * through with the exec. We must hold mmap_sem around
830 * checking core_state and changing tsk->mm.
832 down_read(&old_mm->mmap_sem);
833 if (unlikely(old_mm->core_state)) {
834 up_read(&old_mm->mmap_sem);
839 active_mm = tsk->active_mm;
842 activate_mm(active_mm, mm);
844 arch_pick_mmap_layout(mm);
846 up_read(&old_mm->mmap_sem);
847 BUG_ON(active_mm != old_mm);
848 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
849 mm_update_next_owner(old_mm);
858 * This function makes sure the current process has its own signal table,
859 * so that flush_signal_handlers can later reset the handlers without
860 * disturbing other processes. (Other processes might share the signal
861 * table via the CLONE_SIGHAND option to clone().)
863 static int de_thread(struct task_struct *tsk)
865 struct signal_struct *sig = tsk->signal;
866 struct sighand_struct *oldsighand = tsk->sighand;
867 spinlock_t *lock = &oldsighand->siglock;
869 if (thread_group_empty(tsk))
870 goto no_thread_group;
873 * Kill all other threads in the thread group.
876 if (signal_group_exit(sig)) {
878 * Another group action in progress, just
879 * return so that the signal is processed.
881 spin_unlock_irq(lock);
885 sig->group_exit_task = tsk;
886 sig->notify_count = zap_other_threads(tsk);
887 if (!thread_group_leader(tsk))
890 while (sig->notify_count) {
891 __set_current_state(TASK_KILLABLE);
892 spin_unlock_irq(lock);
894 if (unlikely(__fatal_signal_pending(tsk)))
898 spin_unlock_irq(lock);
901 * At this point all other threads have exited, all we have to
902 * do is to wait for the thread group leader to become inactive,
903 * and to assume its PID:
905 if (!thread_group_leader(tsk)) {
906 struct task_struct *leader = tsk->group_leader;
908 sig->notify_count = -1; /* for exit_notify() */
910 write_lock_irq(&tasklist_lock);
911 if (likely(leader->exit_state))
913 __set_current_state(TASK_KILLABLE);
914 write_unlock_irq(&tasklist_lock);
916 if (unlikely(__fatal_signal_pending(tsk)))
921 * The only record we have of the real-time age of a
922 * process, regardless of execs it's done, is start_time.
923 * All the past CPU time is accumulated in signal_struct
924 * from sister threads now dead. But in this non-leader
925 * exec, nothing survives from the original leader thread,
926 * whose birth marks the true age of this process now.
927 * When we take on its identity by switching to its PID, we
928 * also take its birthdate (always earlier than our own).
930 tsk->start_time = leader->start_time;
932 BUG_ON(!same_thread_group(leader, tsk));
933 BUG_ON(has_group_leader_pid(tsk));
935 * An exec() starts a new thread group with the
936 * TGID of the previous thread group. Rehash the
937 * two threads with a switched PID, and release
938 * the former thread group leader:
941 /* Become a process group leader with the old leader's pid.
942 * The old leader becomes a thread of the this thread group.
943 * Note: The old leader also uses this pid until release_task
944 * is called. Odd but simple and correct.
946 detach_pid(tsk, PIDTYPE_PID);
947 tsk->pid = leader->pid;
948 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
949 transfer_pid(leader, tsk, PIDTYPE_PGID);
950 transfer_pid(leader, tsk, PIDTYPE_SID);
952 list_replace_rcu(&leader->tasks, &tsk->tasks);
953 list_replace_init(&leader->sibling, &tsk->sibling);
955 tsk->group_leader = tsk;
956 leader->group_leader = tsk;
958 tsk->exit_signal = SIGCHLD;
959 leader->exit_signal = -1;
961 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
962 leader->exit_state = EXIT_DEAD;
965 * We are going to release_task()->ptrace_unlink() silently,
966 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
967 * the tracer wont't block again waiting for this thread.
969 if (unlikely(leader->ptrace))
970 __wake_up_parent(leader, leader->parent);
971 write_unlock_irq(&tasklist_lock);
973 release_task(leader);
976 sig->group_exit_task = NULL;
977 sig->notify_count = 0;
980 /* we have changed execution domain */
981 tsk->exit_signal = SIGCHLD;
984 flush_itimer_signals();
986 if (atomic_read(&oldsighand->count) != 1) {
987 struct sighand_struct *newsighand;
989 * This ->sighand is shared with the CLONE_SIGHAND
990 * but not CLONE_THREAD task, switch to the new one.
992 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
996 atomic_set(&newsighand->count, 1);
997 memcpy(newsighand->action, oldsighand->action,
998 sizeof(newsighand->action));
1000 write_lock_irq(&tasklist_lock);
1001 spin_lock(&oldsighand->siglock);
1002 rcu_assign_pointer(tsk->sighand, newsighand);
1003 spin_unlock(&oldsighand->siglock);
1004 write_unlock_irq(&tasklist_lock);
1006 __cleanup_sighand(oldsighand);
1009 BUG_ON(!thread_group_leader(tsk));
1013 /* protects against exit_notify() and __exit_signal() */
1014 read_lock(&tasklist_lock);
1015 sig->group_exit_task = NULL;
1016 sig->notify_count = 0;
1017 read_unlock(&tasklist_lock);
1021 char *get_task_comm(char *buf, struct task_struct *tsk)
1023 /* buf must be at least sizeof(tsk->comm) in size */
1025 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1029 EXPORT_SYMBOL_GPL(get_task_comm);
1032 * These functions flushes out all traces of the currently running executable
1033 * so that a new one can be started
1036 void set_task_comm(struct task_struct *tsk, char *buf)
1040 trace_task_rename(tsk, buf);
1043 * Threads may access current->comm without holding
1044 * the task lock, so write the string carefully.
1045 * Readers without a lock may see incomplete new
1046 * names but are safe from non-terminating string reads.
1048 memset(tsk->comm, 0, TASK_COMM_LEN);
1050 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1052 perf_event_comm(tsk);
1055 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1059 /* Copies the binary name from after last slash */
1060 for (i = 0; (ch = *(fn++)) != '\0';) {
1062 i = 0; /* overwrite what we wrote */
1070 int flush_old_exec(struct linux_binprm * bprm)
1075 * Make sure we have a private signal table and that
1076 * we are unassociated from the previous thread group.
1078 retval = de_thread(current);
1082 set_mm_exe_file(bprm->mm, bprm->file);
1084 filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1086 * Release all of the old mmap stuff
1088 acct_arg_size(bprm, 0);
1089 retval = exec_mmap(bprm->mm);
1093 bprm->mm = NULL; /* We're using it now */
1097 ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD | PF_NOFREEZE);
1099 current->personality &= ~bprm->per_clear;
1106 EXPORT_SYMBOL(flush_old_exec);
1108 void would_dump(struct linux_binprm *bprm, struct file *file)
1110 if (inode_permission(file_inode(file), MAY_READ) < 0)
1111 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1113 EXPORT_SYMBOL(would_dump);
1115 void setup_new_exec(struct linux_binprm * bprm)
1117 arch_pick_mmap_layout(current->mm);
1119 /* This is the point of no return */
1120 current->sas_ss_sp = current->sas_ss_size = 0;
1122 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1123 set_dumpable(current->mm, SUID_DUMP_USER);
1125 set_dumpable(current->mm, suid_dumpable);
1127 set_task_comm(current, bprm->tcomm);
1129 /* Set the new mm task size. We have to do that late because it may
1130 * depend on TIF_32BIT which is only updated in flush_thread() on
1131 * some architectures like powerpc
1133 current->mm->task_size = TASK_SIZE;
1135 /* install the new credentials */
1136 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1137 !gid_eq(bprm->cred->gid, current_egid())) {
1138 current->pdeath_signal = 0;
1140 would_dump(bprm, bprm->file);
1141 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1142 set_dumpable(current->mm, suid_dumpable);
1146 * Flush performance counters when crossing a
1149 if (!get_dumpable(current->mm))
1150 perf_event_exit_task(current);
1152 /* An exec changes our domain. We are no longer part of the thread
1155 current->self_exec_id++;
1157 flush_signal_handlers(current, 0);
1158 do_close_on_exec(current->files);
1160 EXPORT_SYMBOL(setup_new_exec);
1163 * Prepare credentials and lock ->cred_guard_mutex.
1164 * install_exec_creds() commits the new creds and drops the lock.
1165 * Or, if exec fails before, free_bprm() should release ->cred and
1168 int prepare_bprm_creds(struct linux_binprm *bprm)
1170 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1171 return -ERESTARTNOINTR;
1173 bprm->cred = prepare_exec_creds();
1174 if (likely(bprm->cred))
1177 mutex_unlock(¤t->signal->cred_guard_mutex);
1181 void free_bprm(struct linux_binprm *bprm)
1183 free_arg_pages(bprm);
1185 mutex_unlock(¤t->signal->cred_guard_mutex);
1186 abort_creds(bprm->cred);
1188 /* If a binfmt changed the interp, free it. */
1189 if (bprm->interp != bprm->filename)
1190 kfree(bprm->interp);
1194 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1196 /* If a binfmt changed the interp, free it first. */
1197 if (bprm->interp != bprm->filename)
1198 kfree(bprm->interp);
1199 bprm->interp = kstrdup(interp, GFP_KERNEL);
1204 EXPORT_SYMBOL(bprm_change_interp);
1207 * install the new credentials for this executable
1209 void install_exec_creds(struct linux_binprm *bprm)
1211 security_bprm_committing_creds(bprm);
1213 commit_creds(bprm->cred);
1216 * cred_guard_mutex must be held at least to this point to prevent
1217 * ptrace_attach() from altering our determination of the task's
1218 * credentials; any time after this it may be unlocked.
1220 security_bprm_committed_creds(bprm);
1221 mutex_unlock(¤t->signal->cred_guard_mutex);
1223 EXPORT_SYMBOL(install_exec_creds);
1226 * determine how safe it is to execute the proposed program
1227 * - the caller must hold ->cred_guard_mutex to protect against
1230 static int check_unsafe_exec(struct linux_binprm *bprm)
1232 struct task_struct *p = current, *t;
1237 if (p->ptrace & PT_PTRACE_CAP)
1238 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1240 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1244 * This isn't strictly necessary, but it makes it harder for LSMs to
1247 if (current->no_new_privs)
1248 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1251 spin_lock(&p->fs->lock);
1253 for (t = next_thread(p); t != p; t = next_thread(t)) {
1259 if (p->fs->users > n_fs) {
1260 bprm->unsafe |= LSM_UNSAFE_SHARE;
1263 if (!p->fs->in_exec) {
1268 spin_unlock(&p->fs->lock);
1274 * Fill the binprm structure from the inode.
1275 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1277 * This may be called multiple times for binary chains (scripts for example).
1279 int prepare_binprm(struct linux_binprm *bprm)
1282 struct inode * inode = file_inode(bprm->file);
1285 mode = inode->i_mode;
1286 if (bprm->file->f_op == NULL)
1289 /* clear any previous set[ug]id data from a previous binary */
1290 bprm->cred->euid = current_euid();
1291 bprm->cred->egid = current_egid();
1293 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1294 !current->no_new_privs &&
1295 kuid_has_mapping(bprm->cred->user_ns, inode->i_uid) &&
1296 kgid_has_mapping(bprm->cred->user_ns, inode->i_gid)) {
1298 if (mode & S_ISUID) {
1299 bprm->per_clear |= PER_CLEAR_ON_SETID;
1300 bprm->cred->euid = inode->i_uid;
1305 * If setgid is set but no group execute bit then this
1306 * is a candidate for mandatory locking, not a setgid
1309 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1310 bprm->per_clear |= PER_CLEAR_ON_SETID;
1311 bprm->cred->egid = inode->i_gid;
1315 /* fill in binprm security blob */
1316 retval = security_bprm_set_creds(bprm);
1319 bprm->cred_prepared = 1;
1321 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1322 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1325 EXPORT_SYMBOL(prepare_binprm);
1328 * Arguments are '\0' separated strings found at the location bprm->p
1329 * points to; chop off the first by relocating brpm->p to right after
1330 * the first '\0' encountered.
1332 int remove_arg_zero(struct linux_binprm *bprm)
1335 unsigned long offset;
1343 offset = bprm->p & ~PAGE_MASK;
1344 page = get_arg_page(bprm, bprm->p, 0);
1349 kaddr = kmap_atomic(page);
1351 for (; offset < PAGE_SIZE && kaddr[offset];
1352 offset++, bprm->p++)
1355 kunmap_atomic(kaddr);
1358 if (offset == PAGE_SIZE)
1359 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1360 } while (offset == PAGE_SIZE);
1369 EXPORT_SYMBOL(remove_arg_zero);
1372 * cycle the list of binary formats handler, until one recognizes the image
1374 int search_binary_handler(struct linux_binprm *bprm)
1376 unsigned int depth = bprm->recursion_depth;
1378 struct linux_binfmt *fmt;
1379 pid_t old_pid, old_vpid;
1381 /* This allows 4 levels of binfmt rewrites before failing hard. */
1385 retval = security_bprm_check(bprm);
1389 retval = audit_bprm(bprm);
1393 /* Need to fetch pid before load_binary changes it */
1394 old_pid = current->pid;
1396 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1400 for (try=0; try<2; try++) {
1401 read_lock(&binfmt_lock);
1402 list_for_each_entry(fmt, &formats, lh) {
1403 int (*fn)(struct linux_binprm *) = fmt->load_binary;
1406 if (!try_module_get(fmt->module))
1408 read_unlock(&binfmt_lock);
1409 bprm->recursion_depth = depth + 1;
1411 bprm->recursion_depth = depth;
1414 trace_sched_process_exec(current, old_pid, bprm);
1415 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1418 allow_write_access(bprm->file);
1422 current->did_exec = 1;
1423 proc_exec_connector(current);
1426 read_lock(&binfmt_lock);
1428 if (retval != -ENOEXEC || bprm->mm == NULL)
1431 read_unlock(&binfmt_lock);
1435 read_unlock(&binfmt_lock);
1436 #ifdef CONFIG_MODULES
1437 if (retval != -ENOEXEC || bprm->mm == NULL) {
1440 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1441 if (printable(bprm->buf[0]) &&
1442 printable(bprm->buf[1]) &&
1443 printable(bprm->buf[2]) &&
1444 printable(bprm->buf[3]))
1445 break; /* -ENOEXEC */
1447 break; /* -ENOEXEC */
1448 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1457 EXPORT_SYMBOL(search_binary_handler);
1460 * sys_execve() executes a new program.
1462 static int do_execve_common(const char *filename,
1463 struct user_arg_ptr argv,
1464 struct user_arg_ptr envp)
1466 struct linux_binprm *bprm;
1468 struct files_struct *displaced;
1471 const struct cred *cred = current_cred();
1474 * We move the actual failure in case of RLIMIT_NPROC excess from
1475 * set*uid() to execve() because too many poorly written programs
1476 * don't check setuid() return code. Here we additionally recheck
1477 * whether NPROC limit is still exceeded.
1479 if ((current->flags & PF_NPROC_EXCEEDED) &&
1480 atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1485 /* We're below the limit (still or again), so we don't want to make
1486 * further execve() calls fail. */
1487 current->flags &= ~PF_NPROC_EXCEEDED;
1489 retval = unshare_files(&displaced);
1494 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1498 retval = prepare_bprm_creds(bprm);
1502 retval = check_unsafe_exec(bprm);
1505 clear_in_exec = retval;
1506 current->in_execve = 1;
1508 file = open_exec(filename);
1509 retval = PTR_ERR(file);
1516 bprm->filename = filename;
1517 bprm->interp = filename;
1519 retval = bprm_mm_init(bprm);
1523 bprm->argc = count(argv, MAX_ARG_STRINGS);
1524 if ((retval = bprm->argc) < 0)
1527 bprm->envc = count(envp, MAX_ARG_STRINGS);
1528 if ((retval = bprm->envc) < 0)
1531 retval = prepare_binprm(bprm);
1535 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1539 bprm->exec = bprm->p;
1540 retval = copy_strings(bprm->envc, envp, bprm);
1544 retval = copy_strings(bprm->argc, argv, bprm);
1548 retval = search_binary_handler(bprm);
1552 /* execve succeeded */
1553 current->fs->in_exec = 0;
1554 current->in_execve = 0;
1555 acct_update_integrals(current);
1558 put_files_struct(displaced);
1563 acct_arg_size(bprm, 0);
1569 allow_write_access(bprm->file);
1575 current->fs->in_exec = 0;
1576 current->in_execve = 0;
1583 reset_files_struct(displaced);
1588 int do_execve(const char *filename,
1589 const char __user *const __user *__argv,
1590 const char __user *const __user *__envp)
1592 struct user_arg_ptr argv = { .ptr.native = __argv };
1593 struct user_arg_ptr envp = { .ptr.native = __envp };
1594 return do_execve_common(filename, argv, envp);
1597 #ifdef CONFIG_COMPAT
1598 static int compat_do_execve(const char *filename,
1599 const compat_uptr_t __user *__argv,
1600 const compat_uptr_t __user *__envp)
1602 struct user_arg_ptr argv = {
1604 .ptr.compat = __argv,
1606 struct user_arg_ptr envp = {
1608 .ptr.compat = __envp,
1610 return do_execve_common(filename, argv, envp);
1614 void set_binfmt(struct linux_binfmt *new)
1616 struct mm_struct *mm = current->mm;
1619 module_put(mm->binfmt->module);
1623 __module_get(new->module);
1626 EXPORT_SYMBOL(set_binfmt);
1629 * set_dumpable converts traditional three-value dumpable to two flags and
1630 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1631 * these bits are not changed atomically. So get_dumpable can observe the
1632 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1633 * return either old dumpable or new one by paying attention to the order of
1634 * modifying the bits.
1636 * dumpable | mm->flags (binary)
1637 * old new | initial interim final
1638 * ---------+-----------------------
1646 * (*) get_dumpable regards interim value of 10 as 11.
1648 void set_dumpable(struct mm_struct *mm, int value)
1651 case SUID_DUMP_DISABLE:
1652 clear_bit(MMF_DUMPABLE, &mm->flags);
1654 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1656 case SUID_DUMP_USER:
1657 set_bit(MMF_DUMPABLE, &mm->flags);
1659 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1661 case SUID_DUMP_ROOT:
1662 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1664 set_bit(MMF_DUMPABLE, &mm->flags);
1669 int __get_dumpable(unsigned long mm_flags)
1673 ret = mm_flags & MMF_DUMPABLE_MASK;
1674 return (ret > SUID_DUMP_USER) ? SUID_DUMP_ROOT : ret;
1677 int get_dumpable(struct mm_struct *mm)
1679 return __get_dumpable(mm->flags);
1682 SYSCALL_DEFINE3(execve,
1683 const char __user *, filename,
1684 const char __user *const __user *, argv,
1685 const char __user *const __user *, envp)
1687 struct filename *path = getname(filename);
1688 int error = PTR_ERR(path);
1689 if (!IS_ERR(path)) {
1690 error = do_execve(path->name, argv, envp);
1695 #ifdef CONFIG_COMPAT
1696 asmlinkage long compat_sys_execve(const char __user * filename,
1697 const compat_uptr_t __user * argv,
1698 const compat_uptr_t __user * envp)
1700 struct filename *path = getname(filename);
1701 int error = PTR_ERR(path);
1702 if (!IS_ERR(path)) {
1703 error = compat_do_execve(path->name, argv, envp);