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/sched/mm.h>
36 #include <linux/sched/coredump.h>
37 #include <linux/sched/signal.h>
38 #include <linux/sched/numa_balancing.h>
39 #include <linux/sched/task.h>
40 #include <linux/pagemap.h>
41 #include <linux/perf_event.h>
42 #include <linux/highmem.h>
43 #include <linux/spinlock.h>
44 #include <linux/key.h>
45 #include <linux/personality.h>
46 #include <linux/binfmts.h>
47 #include <linux/utsname.h>
48 #include <linux/pid_namespace.h>
49 #include <linux/module.h>
50 #include <linux/namei.h>
51 #include <linux/mount.h>
52 #include <linux/security.h>
53 #include <linux/syscalls.h>
54 #include <linux/tsacct_kern.h>
55 #include <linux/cn_proc.h>
56 #include <linux/audit.h>
57 #include <linux/tracehook.h>
58 #include <linux/kmod.h>
59 #include <linux/fsnotify.h>
60 #include <linux/fs_struct.h>
61 #include <linux/pipe_fs_i.h>
62 #include <linux/oom.h>
63 #include <linux/compat.h>
64 #include <linux/vmalloc.h>
66 #include <linux/uaccess.h>
67 #include <asm/mmu_context.h>
70 #include <trace/events/task.h>
73 #include <trace/events/sched.h>
75 int suid_dumpable = 0;
77 static LIST_HEAD(formats);
78 static DEFINE_RWLOCK(binfmt_lock);
80 void __register_binfmt(struct linux_binfmt * fmt, int insert)
83 if (WARN_ON(!fmt->load_binary))
85 write_lock(&binfmt_lock);
86 insert ? list_add(&fmt->lh, &formats) :
87 list_add_tail(&fmt->lh, &formats);
88 write_unlock(&binfmt_lock);
91 EXPORT_SYMBOL(__register_binfmt);
93 void unregister_binfmt(struct linux_binfmt * fmt)
95 write_lock(&binfmt_lock);
97 write_unlock(&binfmt_lock);
100 EXPORT_SYMBOL(unregister_binfmt);
102 static inline void put_binfmt(struct linux_binfmt * fmt)
104 module_put(fmt->module);
107 bool path_noexec(const struct path *path)
109 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
110 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
115 * Note that a shared library must be both readable and executable due to
118 * Also note that we take the address to load from from the file itself.
120 SYSCALL_DEFINE1(uselib, const char __user *, library)
122 struct linux_binfmt *fmt;
124 struct filename *tmp = getname(library);
125 int error = PTR_ERR(tmp);
126 static const struct open_flags uselib_flags = {
127 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
128 .acc_mode = MAY_READ | MAY_EXEC,
129 .intent = LOOKUP_OPEN,
130 .lookup_flags = LOOKUP_FOLLOW,
136 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
138 error = PTR_ERR(file);
143 if (!S_ISREG(file_inode(file)->i_mode))
147 if (path_noexec(&file->f_path))
154 read_lock(&binfmt_lock);
155 list_for_each_entry(fmt, &formats, lh) {
156 if (!fmt->load_shlib)
158 if (!try_module_get(fmt->module))
160 read_unlock(&binfmt_lock);
161 error = fmt->load_shlib(file);
162 read_lock(&binfmt_lock);
164 if (error != -ENOEXEC)
167 read_unlock(&binfmt_lock);
173 #endif /* #ifdef CONFIG_USELIB */
177 * The nascent bprm->mm is not visible until exec_mmap() but it can
178 * use a lot of memory, account these pages in current->mm temporary
179 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
180 * change the counter back via acct_arg_size(0).
182 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
184 struct mm_struct *mm = current->mm;
185 long diff = (long)(pages - bprm->vma_pages);
190 bprm->vma_pages = pages;
191 add_mm_counter(mm, MM_ANONPAGES, diff);
194 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
199 unsigned int gup_flags = FOLL_FORCE;
201 #ifdef CONFIG_STACK_GROWSUP
203 ret = expand_downwards(bprm->vma, pos);
210 gup_flags |= FOLL_WRITE;
213 * We are doing an exec(). 'current' is the process
214 * doing the exec and bprm->mm is the new process's mm.
216 ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
222 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
223 unsigned long ptr_size;
227 * Since the stack will hold pointers to the strings, we
228 * must account for them as well.
230 * The size calculation is the entire vma while each arg page is
231 * built, so each time we get here it's calculating how far it
232 * is currently (rather than each call being just the newly
233 * added size from the arg page). As a result, we need to
234 * always add the entire size of the pointers, so that on the
235 * last call to get_arg_page() we'll actually have the entire
238 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
239 if (ptr_size > ULONG_MAX - size)
243 acct_arg_size(bprm, size / PAGE_SIZE);
246 * We've historically supported up to 32 pages (ARG_MAX)
247 * of argument strings even with small stacks
253 * Limit to 1/4-th the stack size for the argv+env strings.
255 * - the remaining binfmt code will not run out of stack space,
256 * - the program will have a reasonable amount of stack left
259 rlim = current->signal->rlim;
260 if (size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4)
271 static void put_arg_page(struct page *page)
276 static void free_arg_pages(struct linux_binprm *bprm)
280 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
283 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
286 static int __bprm_mm_init(struct linux_binprm *bprm)
289 struct vm_area_struct *vma = NULL;
290 struct mm_struct *mm = bprm->mm;
292 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
296 if (down_write_killable(&mm->mmap_sem)) {
303 * Place the stack at the largest stack address the architecture
304 * supports. Later, we'll move this to an appropriate place. We don't
305 * use STACK_TOP because that can depend on attributes which aren't
308 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
309 vma->vm_end = STACK_TOP_MAX;
310 vma->vm_start = vma->vm_end - PAGE_SIZE;
311 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
312 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
313 INIT_LIST_HEAD(&vma->anon_vma_chain);
315 err = insert_vm_struct(mm, vma);
319 mm->stack_vm = mm->total_vm = 1;
320 arch_bprm_mm_init(mm, vma);
321 up_write(&mm->mmap_sem);
322 bprm->p = vma->vm_end - sizeof(void *);
325 up_write(&mm->mmap_sem);
328 kmem_cache_free(vm_area_cachep, vma);
332 static bool valid_arg_len(struct linux_binprm *bprm, long len)
334 return len <= MAX_ARG_STRLEN;
339 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
343 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
348 page = bprm->page[pos / PAGE_SIZE];
349 if (!page && write) {
350 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
353 bprm->page[pos / PAGE_SIZE] = page;
359 static void put_arg_page(struct page *page)
363 static void free_arg_page(struct linux_binprm *bprm, int i)
366 __free_page(bprm->page[i]);
367 bprm->page[i] = NULL;
371 static void free_arg_pages(struct linux_binprm *bprm)
375 for (i = 0; i < MAX_ARG_PAGES; i++)
376 free_arg_page(bprm, i);
379 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
384 static int __bprm_mm_init(struct linux_binprm *bprm)
386 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
390 static bool valid_arg_len(struct linux_binprm *bprm, long len)
392 return len <= bprm->p;
395 #endif /* CONFIG_MMU */
398 * Create a new mm_struct and populate it with a temporary stack
399 * vm_area_struct. We don't have enough context at this point to set the stack
400 * flags, permissions, and offset, so we use temporary values. We'll update
401 * them later in setup_arg_pages().
403 static int bprm_mm_init(struct linux_binprm *bprm)
406 struct mm_struct *mm = NULL;
408 bprm->mm = mm = mm_alloc();
413 err = __bprm_mm_init(bprm);
428 struct user_arg_ptr {
433 const char __user *const __user *native;
435 const compat_uptr_t __user *compat;
440 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
442 const char __user *native;
445 if (unlikely(argv.is_compat)) {
446 compat_uptr_t compat;
448 if (get_user(compat, argv.ptr.compat + nr))
449 return ERR_PTR(-EFAULT);
451 return compat_ptr(compat);
455 if (get_user(native, argv.ptr.native + nr))
456 return ERR_PTR(-EFAULT);
462 * count() counts the number of strings in array ARGV.
464 static int count(struct user_arg_ptr argv, int max)
468 if (argv.ptr.native != NULL) {
470 const char __user *p = get_user_arg_ptr(argv, i);
482 if (fatal_signal_pending(current))
483 return -ERESTARTNOHAND;
491 * 'copy_strings()' copies argument/environment strings from the old
492 * processes's memory to the new process's stack. The call to get_user_pages()
493 * ensures the destination page is created and not swapped out.
495 static int copy_strings(int argc, struct user_arg_ptr argv,
496 struct linux_binprm *bprm)
498 struct page *kmapped_page = NULL;
500 unsigned long kpos = 0;
504 const char __user *str;
509 str = get_user_arg_ptr(argv, argc);
513 len = strnlen_user(str, MAX_ARG_STRLEN);
518 if (!valid_arg_len(bprm, len))
521 /* We're going to work our way backwords. */
527 int offset, bytes_to_copy;
529 if (fatal_signal_pending(current)) {
530 ret = -ERESTARTNOHAND;
535 offset = pos % PAGE_SIZE;
539 bytes_to_copy = offset;
540 if (bytes_to_copy > len)
543 offset -= bytes_to_copy;
544 pos -= bytes_to_copy;
545 str -= bytes_to_copy;
546 len -= bytes_to_copy;
548 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
551 page = get_arg_page(bprm, pos, 1);
558 flush_kernel_dcache_page(kmapped_page);
559 kunmap(kmapped_page);
560 put_arg_page(kmapped_page);
563 kaddr = kmap(kmapped_page);
564 kpos = pos & PAGE_MASK;
565 flush_arg_page(bprm, kpos, kmapped_page);
567 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
576 flush_kernel_dcache_page(kmapped_page);
577 kunmap(kmapped_page);
578 put_arg_page(kmapped_page);
584 * Like copy_strings, but get argv and its values from kernel memory.
586 int copy_strings_kernel(int argc, const char *const *__argv,
587 struct linux_binprm *bprm)
590 mm_segment_t oldfs = get_fs();
591 struct user_arg_ptr argv = {
592 .ptr.native = (const char __user *const __user *)__argv,
596 r = copy_strings(argc, argv, bprm);
601 EXPORT_SYMBOL(copy_strings_kernel);
606 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
607 * the binfmt code determines where the new stack should reside, we shift it to
608 * its final location. The process proceeds as follows:
610 * 1) Use shift to calculate the new vma endpoints.
611 * 2) Extend vma to cover both the old and new ranges. This ensures the
612 * arguments passed to subsequent functions are consistent.
613 * 3) Move vma's page tables to the new range.
614 * 4) Free up any cleared pgd range.
615 * 5) Shrink the vma to cover only the new range.
617 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
619 struct mm_struct *mm = vma->vm_mm;
620 unsigned long old_start = vma->vm_start;
621 unsigned long old_end = vma->vm_end;
622 unsigned long length = old_end - old_start;
623 unsigned long new_start = old_start - shift;
624 unsigned long new_end = old_end - shift;
625 struct mmu_gather tlb;
627 BUG_ON(new_start > new_end);
630 * ensure there are no vmas between where we want to go
633 if (vma != find_vma(mm, new_start))
637 * cover the whole range: [new_start, old_end)
639 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
643 * move the page tables downwards, on failure we rely on
644 * process cleanup to remove whatever mess we made.
646 if (length != move_page_tables(vma, old_start,
647 vma, new_start, length, false))
651 tlb_gather_mmu(&tlb, mm, old_start, old_end);
652 if (new_end > old_start) {
654 * when the old and new regions overlap clear from new_end.
656 free_pgd_range(&tlb, new_end, old_end, new_end,
657 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
660 * otherwise, clean from old_start; this is done to not touch
661 * the address space in [new_end, old_start) some architectures
662 * have constraints on va-space that make this illegal (IA64) -
663 * for the others its just a little faster.
665 free_pgd_range(&tlb, old_start, old_end, new_end,
666 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
668 tlb_finish_mmu(&tlb, old_start, old_end);
671 * Shrink the vma to just the new range. Always succeeds.
673 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
679 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
680 * the stack is optionally relocated, and some extra space is added.
682 int setup_arg_pages(struct linux_binprm *bprm,
683 unsigned long stack_top,
684 int executable_stack)
687 unsigned long stack_shift;
688 struct mm_struct *mm = current->mm;
689 struct vm_area_struct *vma = bprm->vma;
690 struct vm_area_struct *prev = NULL;
691 unsigned long vm_flags;
692 unsigned long stack_base;
693 unsigned long stack_size;
694 unsigned long stack_expand;
695 unsigned long rlim_stack;
697 #ifdef CONFIG_STACK_GROWSUP
698 /* Limit stack size */
699 stack_base = rlimit_max(RLIMIT_STACK);
700 if (stack_base > STACK_SIZE_MAX)
701 stack_base = STACK_SIZE_MAX;
703 /* Add space for stack randomization. */
704 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
706 /* Make sure we didn't let the argument array grow too large. */
707 if (vma->vm_end - vma->vm_start > stack_base)
710 stack_base = PAGE_ALIGN(stack_top - stack_base);
712 stack_shift = vma->vm_start - stack_base;
713 mm->arg_start = bprm->p - stack_shift;
714 bprm->p = vma->vm_end - stack_shift;
716 stack_top = arch_align_stack(stack_top);
717 stack_top = PAGE_ALIGN(stack_top);
719 if (unlikely(stack_top < mmap_min_addr) ||
720 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
723 stack_shift = vma->vm_end - stack_top;
725 bprm->p -= stack_shift;
726 mm->arg_start = bprm->p;
730 bprm->loader -= stack_shift;
731 bprm->exec -= stack_shift;
733 if (down_write_killable(&mm->mmap_sem))
736 vm_flags = VM_STACK_FLAGS;
739 * Adjust stack execute permissions; explicitly enable for
740 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
741 * (arch default) otherwise.
743 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
745 else if (executable_stack == EXSTACK_DISABLE_X)
746 vm_flags &= ~VM_EXEC;
747 vm_flags |= mm->def_flags;
748 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
750 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
756 /* Move stack pages down in memory. */
758 ret = shift_arg_pages(vma, stack_shift);
763 /* mprotect_fixup is overkill to remove the temporary stack flags */
764 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
766 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
767 stack_size = vma->vm_end - vma->vm_start;
769 * Align this down to a page boundary as expand_stack
772 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
773 #ifdef CONFIG_STACK_GROWSUP
774 if (stack_size + stack_expand > rlim_stack)
775 stack_base = vma->vm_start + rlim_stack;
777 stack_base = vma->vm_end + stack_expand;
779 if (stack_size + stack_expand > rlim_stack)
780 stack_base = vma->vm_end - rlim_stack;
782 stack_base = vma->vm_start - stack_expand;
784 current->mm->start_stack = bprm->p;
785 ret = expand_stack(vma, stack_base);
790 up_write(&mm->mmap_sem);
793 EXPORT_SYMBOL(setup_arg_pages);
798 * Transfer the program arguments and environment from the holding pages
799 * onto the stack. The provided stack pointer is adjusted accordingly.
801 int transfer_args_to_stack(struct linux_binprm *bprm,
802 unsigned long *sp_location)
804 unsigned long index, stop, sp;
807 stop = bprm->p >> PAGE_SHIFT;
810 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
811 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
812 char *src = kmap(bprm->page[index]) + offset;
813 sp -= PAGE_SIZE - offset;
814 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
816 kunmap(bprm->page[index]);
826 EXPORT_SYMBOL(transfer_args_to_stack);
828 #endif /* CONFIG_MMU */
830 static struct file *do_open_execat(int fd, struct filename *name, int flags)
834 struct open_flags open_exec_flags = {
835 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
836 .acc_mode = MAY_EXEC,
837 .intent = LOOKUP_OPEN,
838 .lookup_flags = LOOKUP_FOLLOW,
841 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
842 return ERR_PTR(-EINVAL);
843 if (flags & AT_SYMLINK_NOFOLLOW)
844 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
845 if (flags & AT_EMPTY_PATH)
846 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
848 file = do_filp_open(fd, name, &open_exec_flags);
853 if (!S_ISREG(file_inode(file)->i_mode))
856 if (path_noexec(&file->f_path))
859 err = deny_write_access(file);
863 if (name->name[0] != '\0')
874 struct file *open_exec(const char *name)
876 struct filename *filename = getname_kernel(name);
877 struct file *f = ERR_CAST(filename);
879 if (!IS_ERR(filename)) {
880 f = do_open_execat(AT_FDCWD, filename, 0);
885 EXPORT_SYMBOL(open_exec);
887 int kernel_read(struct file *file, loff_t offset,
888 char *addr, unsigned long count)
896 /* The cast to a user pointer is valid due to the set_fs() */
897 result = vfs_read(file, (void __user *)addr, count, &pos);
902 EXPORT_SYMBOL(kernel_read);
904 int kernel_read_file(struct file *file, void **buf, loff_t *size,
905 loff_t max_size, enum kernel_read_file_id id)
911 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
914 ret = security_kernel_read_file(file, id);
918 ret = deny_write_access(file);
922 i_size = i_size_read(file_inode(file));
923 if (max_size > 0 && i_size > max_size) {
932 if (id != READING_FIRMWARE_PREALLOC_BUFFER)
933 *buf = vmalloc(i_size);
940 while (pos < i_size) {
941 bytes = kernel_read(file, pos, (char *)(*buf) + pos,
958 ret = security_kernel_post_read_file(file, *buf, i_size, id);
964 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
971 allow_write_access(file);
974 EXPORT_SYMBOL_GPL(kernel_read_file);
976 int kernel_read_file_from_path(char *path, void **buf, loff_t *size,
977 loff_t max_size, enum kernel_read_file_id id)
985 file = filp_open(path, O_RDONLY, 0);
987 return PTR_ERR(file);
989 ret = kernel_read_file(file, buf, size, max_size, id);
993 EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
995 int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
996 enum kernel_read_file_id id)
998 struct fd f = fdget(fd);
1004 ret = kernel_read_file(f.file, buf, size, max_size, id);
1009 EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
1011 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
1013 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
1015 flush_icache_range(addr, addr + len);
1018 EXPORT_SYMBOL(read_code);
1020 static int exec_mmap(struct mm_struct *mm)
1022 struct task_struct *tsk;
1023 struct mm_struct *old_mm, *active_mm;
1025 /* Notify parent that we're no longer interested in the old VM */
1027 old_mm = current->mm;
1028 mm_release(tsk, old_mm);
1031 sync_mm_rss(old_mm);
1033 * Make sure that if there is a core dump in progress
1034 * for the old mm, we get out and die instead of going
1035 * through with the exec. We must hold mmap_sem around
1036 * checking core_state and changing tsk->mm.
1038 down_read(&old_mm->mmap_sem);
1039 if (unlikely(old_mm->core_state)) {
1040 up_read(&old_mm->mmap_sem);
1045 active_mm = tsk->active_mm;
1047 tsk->active_mm = mm;
1048 activate_mm(active_mm, mm);
1049 tsk->mm->vmacache_seqnum = 0;
1050 vmacache_flush(tsk);
1053 up_read(&old_mm->mmap_sem);
1054 BUG_ON(active_mm != old_mm);
1055 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1056 mm_update_next_owner(old_mm);
1065 * This function makes sure the current process has its own signal table,
1066 * so that flush_signal_handlers can later reset the handlers without
1067 * disturbing other processes. (Other processes might share the signal
1068 * table via the CLONE_SIGHAND option to clone().)
1070 static int de_thread(struct task_struct *tsk)
1072 struct signal_struct *sig = tsk->signal;
1073 struct sighand_struct *oldsighand = tsk->sighand;
1074 spinlock_t *lock = &oldsighand->siglock;
1076 if (thread_group_empty(tsk))
1077 goto no_thread_group;
1080 * Kill all other threads in the thread group.
1082 spin_lock_irq(lock);
1083 if (signal_group_exit(sig)) {
1085 * Another group action in progress, just
1086 * return so that the signal is processed.
1088 spin_unlock_irq(lock);
1092 sig->group_exit_task = tsk;
1093 sig->notify_count = zap_other_threads(tsk);
1094 if (!thread_group_leader(tsk))
1095 sig->notify_count--;
1097 while (sig->notify_count) {
1098 __set_current_state(TASK_KILLABLE);
1099 spin_unlock_irq(lock);
1101 if (unlikely(__fatal_signal_pending(tsk)))
1103 spin_lock_irq(lock);
1105 spin_unlock_irq(lock);
1108 * At this point all other threads have exited, all we have to
1109 * do is to wait for the thread group leader to become inactive,
1110 * and to assume its PID:
1112 if (!thread_group_leader(tsk)) {
1113 struct task_struct *leader = tsk->group_leader;
1116 cgroup_threadgroup_change_begin(tsk);
1117 write_lock_irq(&tasklist_lock);
1119 * Do this under tasklist_lock to ensure that
1120 * exit_notify() can't miss ->group_exit_task
1122 sig->notify_count = -1;
1123 if (likely(leader->exit_state))
1125 __set_current_state(TASK_KILLABLE);
1126 write_unlock_irq(&tasklist_lock);
1127 cgroup_threadgroup_change_end(tsk);
1129 if (unlikely(__fatal_signal_pending(tsk)))
1134 * The only record we have of the real-time age of a
1135 * process, regardless of execs it's done, is start_time.
1136 * All the past CPU time is accumulated in signal_struct
1137 * from sister threads now dead. But in this non-leader
1138 * exec, nothing survives from the original leader thread,
1139 * whose birth marks the true age of this process now.
1140 * When we take on its identity by switching to its PID, we
1141 * also take its birthdate (always earlier than our own).
1143 tsk->start_time = leader->start_time;
1144 tsk->real_start_time = leader->real_start_time;
1146 BUG_ON(!same_thread_group(leader, tsk));
1147 BUG_ON(has_group_leader_pid(tsk));
1149 * An exec() starts a new thread group with the
1150 * TGID of the previous thread group. Rehash the
1151 * two threads with a switched PID, and release
1152 * the former thread group leader:
1155 /* Become a process group leader with the old leader's pid.
1156 * The old leader becomes a thread of the this thread group.
1157 * Note: The old leader also uses this pid until release_task
1158 * is called. Odd but simple and correct.
1160 tsk->pid = leader->pid;
1161 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
1162 transfer_pid(leader, tsk, PIDTYPE_PGID);
1163 transfer_pid(leader, tsk, PIDTYPE_SID);
1165 list_replace_rcu(&leader->tasks, &tsk->tasks);
1166 list_replace_init(&leader->sibling, &tsk->sibling);
1168 tsk->group_leader = tsk;
1169 leader->group_leader = tsk;
1171 tsk->exit_signal = SIGCHLD;
1172 leader->exit_signal = -1;
1174 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1175 leader->exit_state = EXIT_DEAD;
1178 * We are going to release_task()->ptrace_unlink() silently,
1179 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1180 * the tracer wont't block again waiting for this thread.
1182 if (unlikely(leader->ptrace))
1183 __wake_up_parent(leader, leader->parent);
1184 write_unlock_irq(&tasklist_lock);
1185 cgroup_threadgroup_change_end(tsk);
1187 release_task(leader);
1190 sig->group_exit_task = NULL;
1191 sig->notify_count = 0;
1194 /* we have changed execution domain */
1195 tsk->exit_signal = SIGCHLD;
1197 #ifdef CONFIG_POSIX_TIMERS
1199 flush_itimer_signals();
1202 if (atomic_read(&oldsighand->count) != 1) {
1203 struct sighand_struct *newsighand;
1205 * This ->sighand is shared with the CLONE_SIGHAND
1206 * but not CLONE_THREAD task, switch to the new one.
1208 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1212 atomic_set(&newsighand->count, 1);
1213 memcpy(newsighand->action, oldsighand->action,
1214 sizeof(newsighand->action));
1216 write_lock_irq(&tasklist_lock);
1217 spin_lock(&oldsighand->siglock);
1218 rcu_assign_pointer(tsk->sighand, newsighand);
1219 spin_unlock(&oldsighand->siglock);
1220 write_unlock_irq(&tasklist_lock);
1222 __cleanup_sighand(oldsighand);
1225 BUG_ON(!thread_group_leader(tsk));
1229 /* protects against exit_notify() and __exit_signal() */
1230 read_lock(&tasklist_lock);
1231 sig->group_exit_task = NULL;
1232 sig->notify_count = 0;
1233 read_unlock(&tasklist_lock);
1237 char *get_task_comm(char *buf, struct task_struct *tsk)
1239 /* buf must be at least sizeof(tsk->comm) in size */
1241 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1245 EXPORT_SYMBOL_GPL(get_task_comm);
1248 * These functions flushes out all traces of the currently running executable
1249 * so that a new one can be started
1252 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1255 trace_task_rename(tsk, buf);
1256 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1258 perf_event_comm(tsk, exec);
1261 int flush_old_exec(struct linux_binprm * bprm)
1266 * Make sure we have a private signal table and that
1267 * we are unassociated from the previous thread group.
1269 retval = de_thread(current);
1274 * Must be called _before_ exec_mmap() as bprm->mm is
1275 * not visibile until then. This also enables the update
1278 set_mm_exe_file(bprm->mm, bprm->file);
1281 * Release all of the old mmap stuff
1283 acct_arg_size(bprm, 0);
1284 retval = exec_mmap(bprm->mm);
1288 bprm->mm = NULL; /* We're using it now */
1291 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1292 PF_NOFREEZE | PF_NO_SETAFFINITY);
1294 current->personality &= ~bprm->per_clear;
1297 * We have to apply CLOEXEC before we change whether the process is
1298 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1299 * trying to access the should-be-closed file descriptors of a process
1300 * undergoing exec(2).
1302 do_close_on_exec(current->files);
1308 EXPORT_SYMBOL(flush_old_exec);
1310 void would_dump(struct linux_binprm *bprm, struct file *file)
1312 struct inode *inode = file_inode(file);
1313 if (inode_permission(inode, MAY_READ) < 0) {
1314 struct user_namespace *old, *user_ns;
1315 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1317 /* Ensure mm->user_ns contains the executable */
1318 user_ns = old = bprm->mm->user_ns;
1319 while ((user_ns != &init_user_ns) &&
1320 !privileged_wrt_inode_uidgid(user_ns, inode))
1321 user_ns = user_ns->parent;
1323 if (old != user_ns) {
1324 bprm->mm->user_ns = get_user_ns(user_ns);
1329 EXPORT_SYMBOL(would_dump);
1331 void setup_new_exec(struct linux_binprm * bprm)
1333 arch_pick_mmap_layout(current->mm);
1335 /* This is the point of no return */
1336 current->sas_ss_sp = current->sas_ss_size = 0;
1338 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1339 set_dumpable(current->mm, SUID_DUMP_USER);
1341 set_dumpable(current->mm, suid_dumpable);
1343 arch_setup_new_exec();
1345 __set_task_comm(current, kbasename(bprm->filename), true);
1347 /* Set the new mm task size. We have to do that late because it may
1348 * depend on TIF_32BIT which is only updated in flush_thread() on
1349 * some architectures like powerpc
1351 current->mm->task_size = TASK_SIZE;
1353 /* install the new credentials */
1354 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1355 !gid_eq(bprm->cred->gid, current_egid())) {
1356 current->pdeath_signal = 0;
1358 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1359 set_dumpable(current->mm, suid_dumpable);
1362 /* An exec changes our domain. We are no longer part of the thread
1364 current->self_exec_id++;
1365 flush_signal_handlers(current, 0);
1367 EXPORT_SYMBOL(setup_new_exec);
1370 * Prepare credentials and lock ->cred_guard_mutex.
1371 * install_exec_creds() commits the new creds and drops the lock.
1372 * Or, if exec fails before, free_bprm() should release ->cred and
1375 int prepare_bprm_creds(struct linux_binprm *bprm)
1377 if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1378 return -ERESTARTNOINTR;
1380 bprm->cred = prepare_exec_creds();
1381 if (likely(bprm->cred))
1384 mutex_unlock(¤t->signal->cred_guard_mutex);
1388 static void free_bprm(struct linux_binprm *bprm)
1390 free_arg_pages(bprm);
1392 mutex_unlock(¤t->signal->cred_guard_mutex);
1393 abort_creds(bprm->cred);
1396 allow_write_access(bprm->file);
1399 /* If a binfmt changed the interp, free it. */
1400 if (bprm->interp != bprm->filename)
1401 kfree(bprm->interp);
1405 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1407 /* If a binfmt changed the interp, free it first. */
1408 if (bprm->interp != bprm->filename)
1409 kfree(bprm->interp);
1410 bprm->interp = kstrdup(interp, GFP_KERNEL);
1415 EXPORT_SYMBOL(bprm_change_interp);
1418 * install the new credentials for this executable
1420 void install_exec_creds(struct linux_binprm *bprm)
1422 security_bprm_committing_creds(bprm);
1424 commit_creds(bprm->cred);
1428 * Disable monitoring for regular users
1429 * when executing setuid binaries. Must
1430 * wait until new credentials are committed
1431 * by commit_creds() above
1433 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1434 perf_event_exit_task(current);
1436 * cred_guard_mutex must be held at least to this point to prevent
1437 * ptrace_attach() from altering our determination of the task's
1438 * credentials; any time after this it may be unlocked.
1440 security_bprm_committed_creds(bprm);
1441 mutex_unlock(¤t->signal->cred_guard_mutex);
1443 EXPORT_SYMBOL(install_exec_creds);
1446 * determine how safe it is to execute the proposed program
1447 * - the caller must hold ->cred_guard_mutex to protect against
1448 * PTRACE_ATTACH or seccomp thread-sync
1450 static void check_unsafe_exec(struct linux_binprm *bprm)
1452 struct task_struct *p = current, *t;
1456 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1459 * This isn't strictly necessary, but it makes it harder for LSMs to
1462 if (task_no_new_privs(current))
1463 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1467 spin_lock(&p->fs->lock);
1469 while_each_thread(p, t) {
1475 if (p->fs->users > n_fs)
1476 bprm->unsafe |= LSM_UNSAFE_SHARE;
1479 spin_unlock(&p->fs->lock);
1482 static void bprm_fill_uid(struct linux_binprm *bprm)
1484 struct inode *inode;
1490 * Since this can be called multiple times (via prepare_binprm),
1491 * we must clear any previous work done when setting set[ug]id
1492 * bits from any earlier bprm->file uses (for example when run
1493 * first for a setuid script then again for its interpreter).
1495 bprm->cred->euid = current_euid();
1496 bprm->cred->egid = current_egid();
1498 if (!mnt_may_suid(bprm->file->f_path.mnt))
1501 if (task_no_new_privs(current))
1504 inode = bprm->file->f_path.dentry->d_inode;
1505 mode = READ_ONCE(inode->i_mode);
1506 if (!(mode & (S_ISUID|S_ISGID)))
1509 /* Be careful if suid/sgid is set */
1512 /* reload atomically mode/uid/gid now that lock held */
1513 mode = inode->i_mode;
1516 inode_unlock(inode);
1518 /* We ignore suid/sgid if there are no mappings for them in the ns */
1519 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1520 !kgid_has_mapping(bprm->cred->user_ns, gid))
1523 if (mode & S_ISUID) {
1524 bprm->per_clear |= PER_CLEAR_ON_SETID;
1525 bprm->cred->euid = uid;
1528 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1529 bprm->per_clear |= PER_CLEAR_ON_SETID;
1530 bprm->cred->egid = gid;
1535 * Fill the binprm structure from the inode.
1536 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1538 * This may be called multiple times for binary chains (scripts for example).
1540 int prepare_binprm(struct linux_binprm *bprm)
1544 bprm_fill_uid(bprm);
1546 /* fill in binprm security blob */
1547 retval = security_bprm_set_creds(bprm);
1550 bprm->cred_prepared = 1;
1552 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1553 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1556 EXPORT_SYMBOL(prepare_binprm);
1559 * Arguments are '\0' separated strings found at the location bprm->p
1560 * points to; chop off the first by relocating brpm->p to right after
1561 * the first '\0' encountered.
1563 int remove_arg_zero(struct linux_binprm *bprm)
1566 unsigned long offset;
1574 offset = bprm->p & ~PAGE_MASK;
1575 page = get_arg_page(bprm, bprm->p, 0);
1580 kaddr = kmap_atomic(page);
1582 for (; offset < PAGE_SIZE && kaddr[offset];
1583 offset++, bprm->p++)
1586 kunmap_atomic(kaddr);
1588 } while (offset == PAGE_SIZE);
1597 EXPORT_SYMBOL(remove_arg_zero);
1599 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1601 * cycle the list of binary formats handler, until one recognizes the image
1603 int search_binary_handler(struct linux_binprm *bprm)
1605 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1606 struct linux_binfmt *fmt;
1609 /* This allows 4 levels of binfmt rewrites before failing hard. */
1610 if (bprm->recursion_depth > 5)
1613 retval = security_bprm_check(bprm);
1619 read_lock(&binfmt_lock);
1620 list_for_each_entry(fmt, &formats, lh) {
1621 if (!try_module_get(fmt->module))
1623 read_unlock(&binfmt_lock);
1624 bprm->recursion_depth++;
1625 retval = fmt->load_binary(bprm);
1626 read_lock(&binfmt_lock);
1628 bprm->recursion_depth--;
1629 if (retval < 0 && !bprm->mm) {
1630 /* we got to flush_old_exec() and failed after it */
1631 read_unlock(&binfmt_lock);
1632 force_sigsegv(SIGSEGV, current);
1635 if (retval != -ENOEXEC || !bprm->file) {
1636 read_unlock(&binfmt_lock);
1640 read_unlock(&binfmt_lock);
1643 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1644 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1646 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1654 EXPORT_SYMBOL(search_binary_handler);
1656 static int exec_binprm(struct linux_binprm *bprm)
1658 pid_t old_pid, old_vpid;
1661 /* Need to fetch pid before load_binary changes it */
1662 old_pid = current->pid;
1664 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1667 ret = search_binary_handler(bprm);
1670 trace_sched_process_exec(current, old_pid, bprm);
1671 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1672 proc_exec_connector(current);
1679 * sys_execve() executes a new program.
1681 static int do_execveat_common(int fd, struct filename *filename,
1682 struct user_arg_ptr argv,
1683 struct user_arg_ptr envp,
1686 char *pathbuf = NULL;
1687 struct linux_binprm *bprm;
1689 struct files_struct *displaced;
1692 if (IS_ERR(filename))
1693 return PTR_ERR(filename);
1696 * We move the actual failure in case of RLIMIT_NPROC excess from
1697 * set*uid() to execve() because too many poorly written programs
1698 * don't check setuid() return code. Here we additionally recheck
1699 * whether NPROC limit is still exceeded.
1701 if ((current->flags & PF_NPROC_EXCEEDED) &&
1702 atomic_read(¤t_user()->processes) > rlimit(RLIMIT_NPROC)) {
1707 /* We're below the limit (still or again), so we don't want to make
1708 * further execve() calls fail. */
1709 current->flags &= ~PF_NPROC_EXCEEDED;
1711 retval = unshare_files(&displaced);
1716 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1720 retval = prepare_bprm_creds(bprm);
1724 check_unsafe_exec(bprm);
1725 current->in_execve = 1;
1727 file = do_open_execat(fd, filename, flags);
1728 retval = PTR_ERR(file);
1735 if (fd == AT_FDCWD || filename->name[0] == '/') {
1736 bprm->filename = filename->name;
1738 if (filename->name[0] == '\0')
1739 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1741 pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1742 fd, filename->name);
1748 * Record that a name derived from an O_CLOEXEC fd will be
1749 * inaccessible after exec. Relies on having exclusive access to
1750 * current->files (due to unshare_files above).
1752 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1753 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1754 bprm->filename = pathbuf;
1756 bprm->interp = bprm->filename;
1758 retval = bprm_mm_init(bprm);
1762 bprm->argc = count(argv, MAX_ARG_STRINGS);
1763 if ((retval = bprm->argc) < 0)
1766 bprm->envc = count(envp, MAX_ARG_STRINGS);
1767 if ((retval = bprm->envc) < 0)
1770 retval = prepare_binprm(bprm);
1774 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1778 bprm->exec = bprm->p;
1779 retval = copy_strings(bprm->envc, envp, bprm);
1783 retval = copy_strings(bprm->argc, argv, bprm);
1787 would_dump(bprm, bprm->file);
1789 retval = exec_binprm(bprm);
1793 /* execve succeeded */
1794 current->fs->in_exec = 0;
1795 current->in_execve = 0;
1796 acct_update_integrals(current);
1797 task_numa_free(current);
1802 put_files_struct(displaced);
1807 acct_arg_size(bprm, 0);
1812 current->fs->in_exec = 0;
1813 current->in_execve = 0;
1821 reset_files_struct(displaced);
1827 int do_execve(struct filename *filename,
1828 const char __user *const __user *__argv,
1829 const char __user *const __user *__envp)
1831 struct user_arg_ptr argv = { .ptr.native = __argv };
1832 struct user_arg_ptr envp = { .ptr.native = __envp };
1833 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1836 int do_execveat(int fd, struct filename *filename,
1837 const char __user *const __user *__argv,
1838 const char __user *const __user *__envp,
1841 struct user_arg_ptr argv = { .ptr.native = __argv };
1842 struct user_arg_ptr envp = { .ptr.native = __envp };
1844 return do_execveat_common(fd, filename, argv, envp, flags);
1847 #ifdef CONFIG_COMPAT
1848 static int compat_do_execve(struct filename *filename,
1849 const compat_uptr_t __user *__argv,
1850 const compat_uptr_t __user *__envp)
1852 struct user_arg_ptr argv = {
1854 .ptr.compat = __argv,
1856 struct user_arg_ptr envp = {
1858 .ptr.compat = __envp,
1860 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1863 static int compat_do_execveat(int fd, struct filename *filename,
1864 const compat_uptr_t __user *__argv,
1865 const compat_uptr_t __user *__envp,
1868 struct user_arg_ptr argv = {
1870 .ptr.compat = __argv,
1872 struct user_arg_ptr envp = {
1874 .ptr.compat = __envp,
1876 return do_execveat_common(fd, filename, argv, envp, flags);
1880 void set_binfmt(struct linux_binfmt *new)
1882 struct mm_struct *mm = current->mm;
1885 module_put(mm->binfmt->module);
1889 __module_get(new->module);
1891 EXPORT_SYMBOL(set_binfmt);
1894 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1896 void set_dumpable(struct mm_struct *mm, int value)
1898 unsigned long old, new;
1900 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1904 old = ACCESS_ONCE(mm->flags);
1905 new = (old & ~MMF_DUMPABLE_MASK) | value;
1906 } while (cmpxchg(&mm->flags, old, new) != old);
1909 SYSCALL_DEFINE3(execve,
1910 const char __user *, filename,
1911 const char __user *const __user *, argv,
1912 const char __user *const __user *, envp)
1914 return do_execve(getname(filename), argv, envp);
1917 SYSCALL_DEFINE5(execveat,
1918 int, fd, const char __user *, filename,
1919 const char __user *const __user *, argv,
1920 const char __user *const __user *, envp,
1923 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1925 return do_execveat(fd,
1926 getname_flags(filename, lookup_flags, NULL),
1930 #ifdef CONFIG_COMPAT
1931 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1932 const compat_uptr_t __user *, argv,
1933 const compat_uptr_t __user *, envp)
1935 return compat_do_execve(getname(filename), argv, envp);
1938 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1939 const char __user *, filename,
1940 const compat_uptr_t __user *, argv,
1941 const compat_uptr_t __user *, envp,
1944 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1946 return compat_do_execveat(fd,
1947 getname_flags(filename, lookup_flags, NULL),