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/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/perf_event.h>
37 #include <linux/highmem.h>
38 #include <linux/spinlock.h>
39 #include <linux/key.h>
40 #include <linux/personality.h>
41 #include <linux/binfmts.h>
42 #include <linux/utsname.h>
43 #include <linux/pid_namespace.h>
44 #include <linux/module.h>
45 #include <linux/namei.h>
46 #include <linux/proc_fs.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/ima.h>
50 #include <linux/syscalls.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54 #include <linux/tracehook.h>
55 #include <linux/kmod.h>
56 #include <linux/fsnotify.h>
57 #include <linux/fs_struct.h>
58 #include <linux/pipe_fs_i.h>
60 #include <asm/uaccess.h>
61 #include <asm/mmu_context.h>
66 char core_pattern[CORENAME_MAX_SIZE] = "core";
67 unsigned int core_pipe_limit;
68 int suid_dumpable = 0;
70 /* The maximal length of core_pattern is also specified in sysctl.c */
72 static LIST_HEAD(formats);
73 static DEFINE_RWLOCK(binfmt_lock);
75 int __register_binfmt(struct linux_binfmt * fmt, int insert)
79 write_lock(&binfmt_lock);
80 insert ? list_add(&fmt->lh, &formats) :
81 list_add_tail(&fmt->lh, &formats);
82 write_unlock(&binfmt_lock);
86 EXPORT_SYMBOL(__register_binfmt);
88 void unregister_binfmt(struct linux_binfmt * fmt)
90 write_lock(&binfmt_lock);
92 write_unlock(&binfmt_lock);
95 EXPORT_SYMBOL(unregister_binfmt);
97 static inline void put_binfmt(struct linux_binfmt * fmt)
99 module_put(fmt->module);
103 * Note that a shared library must be both readable and executable due to
106 * Also note that we take the address to load from from the file itself.
108 SYSCALL_DEFINE1(uselib, const char __user *, library)
111 char *tmp = getname(library);
112 int error = PTR_ERR(tmp);
117 file = do_filp_open(AT_FDCWD, tmp,
118 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
119 MAY_READ | MAY_EXEC | MAY_OPEN);
121 error = PTR_ERR(file);
126 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
130 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
133 fsnotify_open(file->f_path.dentry);
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 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
168 #ifdef CONFIG_STACK_GROWSUP
170 ret = expand_stack_downwards(bprm->vma, pos);
175 ret = get_user_pages(current, bprm->mm, pos,
176 1, write, 1, &page, NULL);
181 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
185 * We've historically supported up to 32 pages (ARG_MAX)
186 * of argument strings even with small stacks
192 * Limit to 1/4-th the stack size for the argv+env strings.
194 * - the remaining binfmt code will not run out of stack space,
195 * - the program will have a reasonable amount of stack left
198 rlim = current->signal->rlim;
199 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
208 static void put_arg_page(struct page *page)
213 static void free_arg_page(struct linux_binprm *bprm, int i)
217 static void free_arg_pages(struct linux_binprm *bprm)
221 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
224 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
227 static int __bprm_mm_init(struct linux_binprm *bprm)
230 struct vm_area_struct *vma = NULL;
231 struct mm_struct *mm = bprm->mm;
233 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
237 down_write(&mm->mmap_sem);
241 * Place the stack at the largest stack address the architecture
242 * supports. Later, we'll move this to an appropriate place. We don't
243 * use STACK_TOP because that can depend on attributes which aren't
246 vma->vm_end = STACK_TOP_MAX;
247 vma->vm_start = vma->vm_end - PAGE_SIZE;
248 vma->vm_flags = VM_STACK_FLAGS;
249 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
250 err = insert_vm_struct(mm, vma);
254 mm->stack_vm = mm->total_vm = 1;
255 up_write(&mm->mmap_sem);
256 bprm->p = vma->vm_end - sizeof(void *);
259 up_write(&mm->mmap_sem);
261 kmem_cache_free(vm_area_cachep, vma);
265 static bool valid_arg_len(struct linux_binprm *bprm, long len)
267 return len <= MAX_ARG_STRLEN;
272 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
277 page = bprm->page[pos / PAGE_SIZE];
278 if (!page && write) {
279 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
282 bprm->page[pos / PAGE_SIZE] = page;
288 static void put_arg_page(struct page *page)
292 static void free_arg_page(struct linux_binprm *bprm, int i)
295 __free_page(bprm->page[i]);
296 bprm->page[i] = NULL;
300 static void free_arg_pages(struct linux_binprm *bprm)
304 for (i = 0; i < MAX_ARG_PAGES; i++)
305 free_arg_page(bprm, i);
308 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
313 static int __bprm_mm_init(struct linux_binprm *bprm)
315 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
319 static bool valid_arg_len(struct linux_binprm *bprm, long len)
321 return len <= bprm->p;
324 #endif /* CONFIG_MMU */
327 * Create a new mm_struct and populate it with a temporary stack
328 * vm_area_struct. We don't have enough context at this point to set the stack
329 * flags, permissions, and offset, so we use temporary values. We'll update
330 * them later in setup_arg_pages().
332 int bprm_mm_init(struct linux_binprm *bprm)
335 struct mm_struct *mm = NULL;
337 bprm->mm = mm = mm_alloc();
342 err = init_new_context(current, mm);
346 err = __bprm_mm_init(bprm);
362 * count() counts the number of strings in array ARGV.
364 static int count(char __user * __user * argv, int max)
372 if (get_user(p, argv))
386 * 'copy_strings()' copies argument/environment strings from the old
387 * processes's memory to the new process's stack. The call to get_user_pages()
388 * ensures the destination page is created and not swapped out.
390 static int copy_strings(int argc, char __user * __user * argv,
391 struct linux_binprm *bprm)
393 struct page *kmapped_page = NULL;
395 unsigned long kpos = 0;
403 if (get_user(str, argv+argc) ||
404 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
409 if (!valid_arg_len(bprm, len)) {
414 /* We're going to work our way backwords. */
420 int offset, bytes_to_copy;
424 offset = pos % PAGE_SIZE;
428 bytes_to_copy = offset;
429 if (bytes_to_copy > len)
432 offset -= bytes_to_copy;
433 pos -= bytes_to_copy;
434 str -= bytes_to_copy;
435 len -= bytes_to_copy;
437 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
440 page = get_arg_page(bprm, pos, 1);
447 flush_kernel_dcache_page(kmapped_page);
448 kunmap(kmapped_page);
449 put_arg_page(kmapped_page);
452 kaddr = kmap(kmapped_page);
453 kpos = pos & PAGE_MASK;
454 flush_arg_page(bprm, kpos, kmapped_page);
456 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
465 flush_kernel_dcache_page(kmapped_page);
466 kunmap(kmapped_page);
467 put_arg_page(kmapped_page);
473 * Like copy_strings, but get argv and its values from kernel memory.
475 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
478 mm_segment_t oldfs = get_fs();
480 r = copy_strings(argc, (char __user * __user *)argv, bprm);
484 EXPORT_SYMBOL(copy_strings_kernel);
489 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
490 * the binfmt code determines where the new stack should reside, we shift it to
491 * its final location. The process proceeds as follows:
493 * 1) Use shift to calculate the new vma endpoints.
494 * 2) Extend vma to cover both the old and new ranges. This ensures the
495 * arguments passed to subsequent functions are consistent.
496 * 3) Move vma's page tables to the new range.
497 * 4) Free up any cleared pgd range.
498 * 5) Shrink the vma to cover only the new range.
500 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
502 struct mm_struct *mm = vma->vm_mm;
503 unsigned long old_start = vma->vm_start;
504 unsigned long old_end = vma->vm_end;
505 unsigned long length = old_end - old_start;
506 unsigned long new_start = old_start - shift;
507 unsigned long new_end = old_end - shift;
508 struct mmu_gather *tlb;
510 BUG_ON(new_start > new_end);
513 * ensure there are no vmas between where we want to go
516 if (vma != find_vma(mm, new_start))
520 * cover the whole range: [new_start, old_end)
522 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
525 * move the page tables downwards, on failure we rely on
526 * process cleanup to remove whatever mess we made.
528 if (length != move_page_tables(vma, old_start,
529 vma, new_start, length))
533 tlb = tlb_gather_mmu(mm, 0);
534 if (new_end > old_start) {
536 * when the old and new regions overlap clear from new_end.
538 free_pgd_range(tlb, new_end, old_end, new_end,
539 vma->vm_next ? vma->vm_next->vm_start : 0);
542 * otherwise, clean from old_start; this is done to not touch
543 * the address space in [new_end, old_start) some architectures
544 * have constraints on va-space that make this illegal (IA64) -
545 * for the others its just a little faster.
547 free_pgd_range(tlb, old_start, old_end, new_end,
548 vma->vm_next ? vma->vm_next->vm_start : 0);
550 tlb_finish_mmu(tlb, new_end, old_end);
553 * shrink the vma to just the new range.
555 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
560 #define EXTRA_STACK_VM_PAGES 20 /* random */
563 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
564 * the stack is optionally relocated, and some extra space is added.
566 int setup_arg_pages(struct linux_binprm *bprm,
567 unsigned long stack_top,
568 int executable_stack)
571 unsigned long stack_shift;
572 struct mm_struct *mm = current->mm;
573 struct vm_area_struct *vma = bprm->vma;
574 struct vm_area_struct *prev = NULL;
575 unsigned long vm_flags;
576 unsigned long stack_base;
577 unsigned long stack_size;
578 unsigned long stack_expand;
579 unsigned long rlim_stack;
581 #ifdef CONFIG_STACK_GROWSUP
582 /* Limit stack size to 1GB */
583 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
584 if (stack_base > (1 << 30))
585 stack_base = 1 << 30;
587 /* Make sure we didn't let the argument array grow too large. */
588 if (vma->vm_end - vma->vm_start > stack_base)
591 stack_base = PAGE_ALIGN(stack_top - stack_base);
593 stack_shift = vma->vm_start - stack_base;
594 mm->arg_start = bprm->p - stack_shift;
595 bprm->p = vma->vm_end - stack_shift;
597 stack_top = arch_align_stack(stack_top);
598 stack_top = PAGE_ALIGN(stack_top);
600 if (unlikely(stack_top < mmap_min_addr) ||
601 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
604 stack_shift = vma->vm_end - stack_top;
606 bprm->p -= stack_shift;
607 mm->arg_start = bprm->p;
611 bprm->loader -= stack_shift;
612 bprm->exec -= stack_shift;
614 down_write(&mm->mmap_sem);
615 vm_flags = VM_STACK_FLAGS;
618 * Adjust stack execute permissions; explicitly enable for
619 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
620 * (arch default) otherwise.
622 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
624 else if (executable_stack == EXSTACK_DISABLE_X)
625 vm_flags &= ~VM_EXEC;
626 vm_flags |= mm->def_flags;
628 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
634 /* Move stack pages down in memory. */
636 ret = shift_arg_pages(vma, stack_shift);
641 stack_expand = EXTRA_STACK_VM_PAGES * PAGE_SIZE;
642 stack_size = vma->vm_end - vma->vm_start;
644 * Align this down to a page boundary as expand_stack
647 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
648 #ifdef CONFIG_STACK_GROWSUP
649 if (stack_size + stack_expand > rlim_stack)
650 stack_base = vma->vm_start + rlim_stack;
652 stack_base = vma->vm_end + stack_expand;
654 if (stack_size + stack_expand > rlim_stack)
655 stack_base = vma->vm_end - rlim_stack;
657 stack_base = vma->vm_start - stack_expand;
659 ret = expand_stack(vma, stack_base);
664 up_write(&mm->mmap_sem);
667 EXPORT_SYMBOL(setup_arg_pages);
669 #endif /* CONFIG_MMU */
671 struct file *open_exec(const char *name)
676 file = do_filp_open(AT_FDCWD, name,
677 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
678 MAY_EXEC | MAY_OPEN);
683 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
686 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
689 fsnotify_open(file->f_path.dentry);
691 err = deny_write_access(file);
702 EXPORT_SYMBOL(open_exec);
704 int kernel_read(struct file *file, loff_t offset,
705 char *addr, unsigned long count)
713 /* The cast to a user pointer is valid due to the set_fs() */
714 result = vfs_read(file, (void __user *)addr, count, &pos);
719 EXPORT_SYMBOL(kernel_read);
721 static int exec_mmap(struct mm_struct *mm)
723 struct task_struct *tsk;
724 struct mm_struct * old_mm, *active_mm;
726 /* Notify parent that we're no longer interested in the old VM */
728 old_mm = current->mm;
729 mm_release(tsk, old_mm);
733 * Make sure that if there is a core dump in progress
734 * for the old mm, we get out and die instead of going
735 * through with the exec. We must hold mmap_sem around
736 * checking core_state and changing tsk->mm.
738 down_read(&old_mm->mmap_sem);
739 if (unlikely(old_mm->core_state)) {
740 up_read(&old_mm->mmap_sem);
745 active_mm = tsk->active_mm;
748 activate_mm(active_mm, mm);
750 arch_pick_mmap_layout(mm);
752 up_read(&old_mm->mmap_sem);
753 BUG_ON(active_mm != old_mm);
754 mm_update_next_owner(old_mm);
763 * This function makes sure the current process has its own signal table,
764 * so that flush_signal_handlers can later reset the handlers without
765 * disturbing other processes. (Other processes might share the signal
766 * table via the CLONE_SIGHAND option to clone().)
768 static int de_thread(struct task_struct *tsk)
770 struct signal_struct *sig = tsk->signal;
771 struct sighand_struct *oldsighand = tsk->sighand;
772 spinlock_t *lock = &oldsighand->siglock;
775 if (thread_group_empty(tsk))
776 goto no_thread_group;
779 * Kill all other threads in the thread group.
782 if (signal_group_exit(sig)) {
784 * Another group action in progress, just
785 * return so that the signal is processed.
787 spin_unlock_irq(lock);
790 sig->group_exit_task = tsk;
791 zap_other_threads(tsk);
793 /* Account for the thread group leader hanging around: */
794 count = thread_group_leader(tsk) ? 1 : 2;
795 sig->notify_count = count;
796 while (atomic_read(&sig->count) > count) {
797 __set_current_state(TASK_UNINTERRUPTIBLE);
798 spin_unlock_irq(lock);
802 spin_unlock_irq(lock);
805 * At this point all other threads have exited, all we have to
806 * do is to wait for the thread group leader to become inactive,
807 * and to assume its PID:
809 if (!thread_group_leader(tsk)) {
810 struct task_struct *leader = tsk->group_leader;
812 sig->notify_count = -1; /* for exit_notify() */
814 write_lock_irq(&tasklist_lock);
815 if (likely(leader->exit_state))
817 __set_current_state(TASK_UNINTERRUPTIBLE);
818 write_unlock_irq(&tasklist_lock);
823 * The only record we have of the real-time age of a
824 * process, regardless of execs it's done, is start_time.
825 * All the past CPU time is accumulated in signal_struct
826 * from sister threads now dead. But in this non-leader
827 * exec, nothing survives from the original leader thread,
828 * whose birth marks the true age of this process now.
829 * When we take on its identity by switching to its PID, we
830 * also take its birthdate (always earlier than our own).
832 tsk->start_time = leader->start_time;
834 BUG_ON(!same_thread_group(leader, tsk));
835 BUG_ON(has_group_leader_pid(tsk));
837 * An exec() starts a new thread group with the
838 * TGID of the previous thread group. Rehash the
839 * two threads with a switched PID, and release
840 * the former thread group leader:
843 /* Become a process group leader with the old leader's pid.
844 * The old leader becomes a thread of the this thread group.
845 * Note: The old leader also uses this pid until release_task
846 * is called. Odd but simple and correct.
848 detach_pid(tsk, PIDTYPE_PID);
849 tsk->pid = leader->pid;
850 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
851 transfer_pid(leader, tsk, PIDTYPE_PGID);
852 transfer_pid(leader, tsk, PIDTYPE_SID);
853 list_replace_rcu(&leader->tasks, &tsk->tasks);
855 tsk->group_leader = tsk;
856 leader->group_leader = tsk;
858 tsk->exit_signal = SIGCHLD;
860 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
861 leader->exit_state = EXIT_DEAD;
862 write_unlock_irq(&tasklist_lock);
864 release_task(leader);
867 sig->group_exit_task = NULL;
868 sig->notify_count = 0;
872 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
875 flush_itimer_signals();
877 if (atomic_read(&oldsighand->count) != 1) {
878 struct sighand_struct *newsighand;
880 * This ->sighand is shared with the CLONE_SIGHAND
881 * but not CLONE_THREAD task, switch to the new one.
883 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
887 atomic_set(&newsighand->count, 1);
888 memcpy(newsighand->action, oldsighand->action,
889 sizeof(newsighand->action));
891 write_lock_irq(&tasklist_lock);
892 spin_lock(&oldsighand->siglock);
893 rcu_assign_pointer(tsk->sighand, newsighand);
894 spin_unlock(&oldsighand->siglock);
895 write_unlock_irq(&tasklist_lock);
897 __cleanup_sighand(oldsighand);
900 BUG_ON(!thread_group_leader(tsk));
905 * These functions flushes out all traces of the currently running executable
906 * so that a new one can be started
908 static void flush_old_files(struct files_struct * files)
913 spin_lock(&files->file_lock);
915 unsigned long set, i;
919 fdt = files_fdtable(files);
920 if (i >= fdt->max_fds)
922 set = fdt->close_on_exec->fds_bits[j];
925 fdt->close_on_exec->fds_bits[j] = 0;
926 spin_unlock(&files->file_lock);
927 for ( ; set ; i++,set >>= 1) {
932 spin_lock(&files->file_lock);
935 spin_unlock(&files->file_lock);
938 char *get_task_comm(char *buf, struct task_struct *tsk)
940 /* buf must be at least sizeof(tsk->comm) in size */
942 strncpy(buf, tsk->comm, sizeof(tsk->comm));
947 void set_task_comm(struct task_struct *tsk, char *buf)
950 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
952 perf_event_comm(tsk);
955 int flush_old_exec(struct linux_binprm * bprm)
960 * Make sure we have a private signal table and that
961 * we are unassociated from the previous thread group.
963 retval = de_thread(current);
967 set_mm_exe_file(bprm->mm, bprm->file);
970 * Release all of the old mmap stuff
972 retval = exec_mmap(bprm->mm);
976 bprm->mm = NULL; /* We're using it now */
978 current->flags &= ~PF_RANDOMIZE;
980 current->personality &= ~bprm->per_clear;
987 EXPORT_SYMBOL(flush_old_exec);
989 void setup_new_exec(struct linux_binprm * bprm)
993 char tcomm[sizeof(current->comm)];
995 arch_pick_mmap_layout(current->mm);
997 /* This is the point of no return */
998 current->sas_ss_sp = current->sas_ss_size = 0;
1000 if (current_euid() == current_uid() && current_egid() == current_gid())
1001 set_dumpable(current->mm, 1);
1003 set_dumpable(current->mm, suid_dumpable);
1005 name = bprm->filename;
1007 /* Copies the binary name from after last slash */
1008 for (i=0; (ch = *(name++)) != '\0';) {
1010 i = 0; /* overwrite what we wrote */
1012 if (i < (sizeof(tcomm) - 1))
1016 set_task_comm(current, tcomm);
1018 /* Set the new mm task size. We have to do that late because it may
1019 * depend on TIF_32BIT which is only updated in flush_thread() on
1020 * some architectures like powerpc
1022 current->mm->task_size = TASK_SIZE;
1024 /* install the new credentials */
1025 if (bprm->cred->uid != current_euid() ||
1026 bprm->cred->gid != current_egid()) {
1027 current->pdeath_signal = 0;
1028 } else if (file_permission(bprm->file, MAY_READ) ||
1029 bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1030 set_dumpable(current->mm, suid_dumpable);
1034 * Flush performance counters when crossing a
1037 if (!get_dumpable(current->mm))
1038 perf_event_exit_task(current);
1040 /* An exec changes our domain. We are no longer part of the thread
1043 current->self_exec_id++;
1045 flush_signal_handlers(current, 0);
1046 flush_old_files(current->files);
1048 EXPORT_SYMBOL(setup_new_exec);
1051 * Prepare credentials and lock ->cred_guard_mutex.
1052 * install_exec_creds() commits the new creds and drops the lock.
1053 * Or, if exec fails before, free_bprm() should release ->cred and
1056 int prepare_bprm_creds(struct linux_binprm *bprm)
1058 if (mutex_lock_interruptible(¤t->cred_guard_mutex))
1059 return -ERESTARTNOINTR;
1061 bprm->cred = prepare_exec_creds();
1062 if (likely(bprm->cred))
1065 mutex_unlock(¤t->cred_guard_mutex);
1069 void free_bprm(struct linux_binprm *bprm)
1071 free_arg_pages(bprm);
1073 mutex_unlock(¤t->cred_guard_mutex);
1074 abort_creds(bprm->cred);
1080 * install the new credentials for this executable
1082 void install_exec_creds(struct linux_binprm *bprm)
1084 security_bprm_committing_creds(bprm);
1086 commit_creds(bprm->cred);
1089 * cred_guard_mutex must be held at least to this point to prevent
1090 * ptrace_attach() from altering our determination of the task's
1091 * credentials; any time after this it may be unlocked.
1093 security_bprm_committed_creds(bprm);
1094 mutex_unlock(¤t->cred_guard_mutex);
1096 EXPORT_SYMBOL(install_exec_creds);
1099 * determine how safe it is to execute the proposed program
1100 * - the caller must hold current->cred_guard_mutex to protect against
1103 int check_unsafe_exec(struct linux_binprm *bprm)
1105 struct task_struct *p = current, *t;
1109 bprm->unsafe = tracehook_unsafe_exec(p);
1112 write_lock(&p->fs->lock);
1114 for (t = next_thread(p); t != p; t = next_thread(t)) {
1120 if (p->fs->users > n_fs) {
1121 bprm->unsafe |= LSM_UNSAFE_SHARE;
1124 if (!p->fs->in_exec) {
1129 write_unlock(&p->fs->lock);
1135 * Fill the binprm structure from the inode.
1136 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1138 * This may be called multiple times for binary chains (scripts for example).
1140 int prepare_binprm(struct linux_binprm *bprm)
1143 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1146 mode = inode->i_mode;
1147 if (bprm->file->f_op == NULL)
1150 /* clear any previous set[ug]id data from a previous binary */
1151 bprm->cred->euid = current_euid();
1152 bprm->cred->egid = current_egid();
1154 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1156 if (mode & S_ISUID) {
1157 bprm->per_clear |= PER_CLEAR_ON_SETID;
1158 bprm->cred->euid = inode->i_uid;
1163 * If setgid is set but no group execute bit then this
1164 * is a candidate for mandatory locking, not a setgid
1167 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1168 bprm->per_clear |= PER_CLEAR_ON_SETID;
1169 bprm->cred->egid = inode->i_gid;
1173 /* fill in binprm security blob */
1174 retval = security_bprm_set_creds(bprm);
1177 bprm->cred_prepared = 1;
1179 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1180 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1183 EXPORT_SYMBOL(prepare_binprm);
1186 * Arguments are '\0' separated strings found at the location bprm->p
1187 * points to; chop off the first by relocating brpm->p to right after
1188 * the first '\0' encountered.
1190 int remove_arg_zero(struct linux_binprm *bprm)
1193 unsigned long offset;
1201 offset = bprm->p & ~PAGE_MASK;
1202 page = get_arg_page(bprm, bprm->p, 0);
1207 kaddr = kmap_atomic(page, KM_USER0);
1209 for (; offset < PAGE_SIZE && kaddr[offset];
1210 offset++, bprm->p++)
1213 kunmap_atomic(kaddr, KM_USER0);
1216 if (offset == PAGE_SIZE)
1217 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1218 } while (offset == PAGE_SIZE);
1227 EXPORT_SYMBOL(remove_arg_zero);
1230 * cycle the list of binary formats handler, until one recognizes the image
1232 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1234 unsigned int depth = bprm->recursion_depth;
1236 struct linux_binfmt *fmt;
1238 retval = security_bprm_check(bprm);
1241 retval = ima_bprm_check(bprm);
1245 /* kernel module loader fixup */
1246 /* so we don't try to load run modprobe in kernel space. */
1249 retval = audit_bprm(bprm);
1254 for (try=0; try<2; try++) {
1255 read_lock(&binfmt_lock);
1256 list_for_each_entry(fmt, &formats, lh) {
1257 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1260 if (!try_module_get(fmt->module))
1262 read_unlock(&binfmt_lock);
1263 retval = fn(bprm, regs);
1265 * Restore the depth counter to its starting value
1266 * in this call, so we don't have to rely on every
1267 * load_binary function to restore it on return.
1269 bprm->recursion_depth = depth;
1272 tracehook_report_exec(fmt, bprm, regs);
1274 allow_write_access(bprm->file);
1278 current->did_exec = 1;
1279 proc_exec_connector(current);
1282 read_lock(&binfmt_lock);
1284 if (retval != -ENOEXEC || bprm->mm == NULL)
1287 read_unlock(&binfmt_lock);
1291 read_unlock(&binfmt_lock);
1292 if (retval != -ENOEXEC || bprm->mm == NULL) {
1294 #ifdef CONFIG_MODULES
1296 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1297 if (printable(bprm->buf[0]) &&
1298 printable(bprm->buf[1]) &&
1299 printable(bprm->buf[2]) &&
1300 printable(bprm->buf[3]))
1301 break; /* -ENOEXEC */
1302 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1309 EXPORT_SYMBOL(search_binary_handler);
1312 * sys_execve() executes a new program.
1314 int do_execve(char * filename,
1315 char __user *__user *argv,
1316 char __user *__user *envp,
1317 struct pt_regs * regs)
1319 struct linux_binprm *bprm;
1321 struct files_struct *displaced;
1325 retval = unshare_files(&displaced);
1330 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1334 retval = prepare_bprm_creds(bprm);
1338 retval = check_unsafe_exec(bprm);
1341 clear_in_exec = retval;
1342 current->in_execve = 1;
1344 file = open_exec(filename);
1345 retval = PTR_ERR(file);
1352 bprm->filename = filename;
1353 bprm->interp = filename;
1355 retval = bprm_mm_init(bprm);
1359 bprm->argc = count(argv, MAX_ARG_STRINGS);
1360 if ((retval = bprm->argc) < 0)
1363 bprm->envc = count(envp, MAX_ARG_STRINGS);
1364 if ((retval = bprm->envc) < 0)
1367 retval = prepare_binprm(bprm);
1371 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1375 bprm->exec = bprm->p;
1376 retval = copy_strings(bprm->envc, envp, bprm);
1380 retval = copy_strings(bprm->argc, argv, bprm);
1384 current->flags &= ~PF_KTHREAD;
1385 retval = search_binary_handler(bprm,regs);
1389 /* execve succeeded */
1390 current->fs->in_exec = 0;
1391 current->in_execve = 0;
1392 acct_update_integrals(current);
1395 put_files_struct(displaced);
1404 allow_write_access(bprm->file);
1410 current->fs->in_exec = 0;
1411 current->in_execve = 0;
1418 reset_files_struct(displaced);
1423 void set_binfmt(struct linux_binfmt *new)
1425 struct mm_struct *mm = current->mm;
1428 module_put(mm->binfmt->module);
1432 __module_get(new->module);
1435 EXPORT_SYMBOL(set_binfmt);
1437 /* format_corename will inspect the pattern parameter, and output a
1438 * name into corename, which must have space for at least
1439 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1441 static int format_corename(char *corename, long signr)
1443 const struct cred *cred = current_cred();
1444 const char *pat_ptr = core_pattern;
1445 int ispipe = (*pat_ptr == '|');
1446 char *out_ptr = corename;
1447 char *const out_end = corename + CORENAME_MAX_SIZE;
1449 int pid_in_pattern = 0;
1451 /* Repeat as long as we have more pattern to process and more output
1454 if (*pat_ptr != '%') {
1455 if (out_ptr == out_end)
1457 *out_ptr++ = *pat_ptr++;
1459 switch (*++pat_ptr) {
1462 /* Double percent, output one percent */
1464 if (out_ptr == out_end)
1471 rc = snprintf(out_ptr, out_end - out_ptr,
1472 "%d", task_tgid_vnr(current));
1473 if (rc > out_end - out_ptr)
1479 rc = snprintf(out_ptr, out_end - out_ptr,
1481 if (rc > out_end - out_ptr)
1487 rc = snprintf(out_ptr, out_end - out_ptr,
1489 if (rc > out_end - out_ptr)
1493 /* signal that caused the coredump */
1495 rc = snprintf(out_ptr, out_end - out_ptr,
1497 if (rc > out_end - out_ptr)
1501 /* UNIX time of coredump */
1504 do_gettimeofday(&tv);
1505 rc = snprintf(out_ptr, out_end - out_ptr,
1507 if (rc > out_end - out_ptr)
1514 down_read(&uts_sem);
1515 rc = snprintf(out_ptr, out_end - out_ptr,
1516 "%s", utsname()->nodename);
1518 if (rc > out_end - out_ptr)
1524 rc = snprintf(out_ptr, out_end - out_ptr,
1525 "%s", current->comm);
1526 if (rc > out_end - out_ptr)
1530 /* core limit size */
1532 rc = snprintf(out_ptr, out_end - out_ptr,
1533 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1534 if (rc > out_end - out_ptr)
1544 /* Backward compatibility with core_uses_pid:
1546 * If core_pattern does not include a %p (as is the default)
1547 * and core_uses_pid is set, then .%pid will be appended to
1548 * the filename. Do not do this for piped commands. */
1549 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1550 rc = snprintf(out_ptr, out_end - out_ptr,
1551 ".%d", task_tgid_vnr(current));
1552 if (rc > out_end - out_ptr)
1561 static int zap_process(struct task_struct *start)
1563 struct task_struct *t;
1566 start->signal->flags = SIGNAL_GROUP_EXIT;
1567 start->signal->group_stop_count = 0;
1571 if (t != current && t->mm) {
1572 sigaddset(&t->pending.signal, SIGKILL);
1573 signal_wake_up(t, 1);
1576 } while_each_thread(start, t);
1581 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1582 struct core_state *core_state, int exit_code)
1584 struct task_struct *g, *p;
1585 unsigned long flags;
1588 spin_lock_irq(&tsk->sighand->siglock);
1589 if (!signal_group_exit(tsk->signal)) {
1590 mm->core_state = core_state;
1591 tsk->signal->group_exit_code = exit_code;
1592 nr = zap_process(tsk);
1594 spin_unlock_irq(&tsk->sighand->siglock);
1595 if (unlikely(nr < 0))
1598 if (atomic_read(&mm->mm_users) == nr + 1)
1601 * We should find and kill all tasks which use this mm, and we should
1602 * count them correctly into ->nr_threads. We don't take tasklist
1603 * lock, but this is safe wrt:
1606 * None of sub-threads can fork after zap_process(leader). All
1607 * processes which were created before this point should be
1608 * visible to zap_threads() because copy_process() adds the new
1609 * process to the tail of init_task.tasks list, and lock/unlock
1610 * of ->siglock provides a memory barrier.
1613 * The caller holds mm->mmap_sem. This means that the task which
1614 * uses this mm can't pass exit_mm(), so it can't exit or clear
1618 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1619 * we must see either old or new leader, this does not matter.
1620 * However, it can change p->sighand, so lock_task_sighand(p)
1621 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1624 * Note also that "g" can be the old leader with ->mm == NULL
1625 * and already unhashed and thus removed from ->thread_group.
1626 * This is OK, __unhash_process()->list_del_rcu() does not
1627 * clear the ->next pointer, we will find the new leader via
1631 for_each_process(g) {
1632 if (g == tsk->group_leader)
1634 if (g->flags & PF_KTHREAD)
1639 if (unlikely(p->mm == mm)) {
1640 lock_task_sighand(p, &flags);
1641 nr += zap_process(p);
1642 unlock_task_sighand(p, &flags);
1646 } while_each_thread(g, p);
1650 atomic_set(&core_state->nr_threads, nr);
1654 static int coredump_wait(int exit_code, struct core_state *core_state)
1656 struct task_struct *tsk = current;
1657 struct mm_struct *mm = tsk->mm;
1658 struct completion *vfork_done;
1661 init_completion(&core_state->startup);
1662 core_state->dumper.task = tsk;
1663 core_state->dumper.next = NULL;
1664 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1665 up_write(&mm->mmap_sem);
1667 if (unlikely(core_waiters < 0))
1671 * Make sure nobody is waiting for us to release the VM,
1672 * otherwise we can deadlock when we wait on each other
1674 vfork_done = tsk->vfork_done;
1676 tsk->vfork_done = NULL;
1677 complete(vfork_done);
1681 wait_for_completion(&core_state->startup);
1683 return core_waiters;
1686 static void coredump_finish(struct mm_struct *mm)
1688 struct core_thread *curr, *next;
1689 struct task_struct *task;
1691 next = mm->core_state->dumper.next;
1692 while ((curr = next) != NULL) {
1696 * see exit_mm(), curr->task must not see
1697 * ->task == NULL before we read ->next.
1701 wake_up_process(task);
1704 mm->core_state = NULL;
1708 * set_dumpable converts traditional three-value dumpable to two flags and
1709 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1710 * these bits are not changed atomically. So get_dumpable can observe the
1711 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1712 * return either old dumpable or new one by paying attention to the order of
1713 * modifying the bits.
1715 * dumpable | mm->flags (binary)
1716 * old new | initial interim final
1717 * ---------+-----------------------
1725 * (*) get_dumpable regards interim value of 10 as 11.
1727 void set_dumpable(struct mm_struct *mm, int value)
1731 clear_bit(MMF_DUMPABLE, &mm->flags);
1733 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1736 set_bit(MMF_DUMPABLE, &mm->flags);
1738 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1741 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1743 set_bit(MMF_DUMPABLE, &mm->flags);
1748 int get_dumpable(struct mm_struct *mm)
1752 ret = mm->flags & 0x3;
1753 return (ret >= 2) ? 2 : ret;
1756 static void wait_for_dump_helpers(struct file *file)
1758 struct pipe_inode_info *pipe;
1760 pipe = file->f_path.dentry->d_inode->i_pipe;
1766 while ((pipe->readers > 1) && (!signal_pending(current))) {
1767 wake_up_interruptible_sync(&pipe->wait);
1768 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1779 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1781 struct core_state core_state;
1782 char corename[CORENAME_MAX_SIZE + 1];
1783 struct mm_struct *mm = current->mm;
1784 struct linux_binfmt * binfmt;
1785 struct inode * inode;
1787 const struct cred *old_cred;
1792 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1793 char **helper_argv = NULL;
1794 int helper_argc = 0;
1796 static atomic_t core_dump_count = ATOMIC_INIT(0);
1798 audit_core_dumps(signr);
1800 binfmt = mm->binfmt;
1801 if (!binfmt || !binfmt->core_dump)
1804 cred = prepare_creds();
1810 down_write(&mm->mmap_sem);
1812 * If another thread got here first, or we are not dumpable, bail out.
1814 if (mm->core_state || !get_dumpable(mm)) {
1815 up_write(&mm->mmap_sem);
1821 * We cannot trust fsuid as being the "true" uid of the
1822 * process nor do we know its entire history. We only know it
1823 * was tainted so we dump it as root in mode 2.
1825 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1826 flag = O_EXCL; /* Stop rewrite attacks */
1827 cred->fsuid = 0; /* Dump root private */
1830 retval = coredump_wait(exit_code, &core_state);
1836 old_cred = override_creds(cred);
1839 * Clear any false indication of pending signals that might
1840 * be seen by the filesystem code called to write the core file.
1842 clear_thread_flag(TIF_SIGPENDING);
1845 * lock_kernel() because format_corename() is controlled by sysctl, which
1846 * uses lock_kernel()
1849 ispipe = format_corename(corename, signr);
1852 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1856 if (core_limit == 0) {
1858 * Normally core limits are irrelevant to pipes, since
1859 * we're not writing to the file system, but we use
1860 * core_limit of 0 here as a speacial value. Any
1861 * non-zero limit gets set to RLIM_INFINITY below, but
1862 * a limit of 0 skips the dump. This is a consistent
1863 * way to catch recursive crashes. We can still crash
1864 * if the core_pattern binary sets RLIM_CORE = !0
1865 * but it runs as root, and can do lots of stupid things
1866 * Note that we use task_tgid_vnr here to grab the pid
1867 * of the process group leader. That way we get the
1868 * right pid if a thread in a multi-threaded
1869 * core_pattern process dies.
1872 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1873 task_tgid_vnr(current), current->comm);
1874 printk(KERN_WARNING "Aborting core\n");
1878 dump_count = atomic_inc_return(&core_dump_count);
1879 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1880 printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1881 task_tgid_vnr(current), current->comm);
1882 printk(KERN_WARNING "Skipping core dump\n");
1883 goto fail_dropcount;
1886 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1888 printk(KERN_WARNING "%s failed to allocate memory\n",
1890 goto fail_dropcount;
1893 core_limit = RLIM_INFINITY;
1895 /* SIGPIPE can happen, but it's just never processed */
1896 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1898 printk(KERN_INFO "Core dump to %s pipe failed\n",
1900 goto fail_dropcount;
1903 file = filp_open(corename,
1904 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1907 goto fail_dropcount;
1908 inode = file->f_path.dentry->d_inode;
1909 if (inode->i_nlink > 1)
1910 goto close_fail; /* multiple links - don't dump */
1911 if (!ispipe && d_unhashed(file->f_path.dentry))
1914 /* AK: actually i see no reason to not allow this for named pipes etc.,
1915 but keep the previous behaviour for now. */
1916 if (!ispipe && !S_ISREG(inode->i_mode))
1919 * Dont allow local users get cute and trick others to coredump
1920 * into their pre-created files:
1921 * Note, this is not relevant for pipes
1923 if (!ispipe && (inode->i_uid != current_fsuid()))
1927 if (!file->f_op->write)
1929 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1932 retval = binfmt->core_dump(signr, regs, file, core_limit);
1935 current->signal->group_exit_code |= 0x80;
1937 if (ispipe && core_pipe_limit)
1938 wait_for_dump_helpers(file);
1939 filp_close(file, NULL);
1942 atomic_dec(&core_dump_count);
1945 argv_free(helper_argv);
1947 revert_creds(old_cred);
1949 coredump_finish(mm);