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/mman.h>
28 #include <linux/a.out.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/init.h>
33 #include <linux/pagemap.h>
34 #include <linux/highmem.h>
35 #include <linux/spinlock.h>
36 #include <linux/key.h>
37 #include <linux/personality.h>
38 #include <linux/binfmts.h>
39 #include <linux/swap.h>
40 #include <linux/utsname.h>
41 #include <linux/pid_namespace.h>
42 #include <linux/module.h>
43 #include <linux/namei.h>
44 #include <linux/proc_fs.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/rmap.h>
50 #include <linux/tsacct_kern.h>
51 #include <linux/cn_proc.h>
52 #include <linux/audit.h>
54 #include <asm/uaccess.h>
55 #include <asm/mmu_context.h>
59 #include <linux/kmod.h>
63 char core_pattern[CORENAME_MAX_SIZE] = "core";
64 int suid_dumpable = 0;
66 EXPORT_SYMBOL(suid_dumpable);
67 /* The maximal length of core_pattern is also specified in sysctl.c */
69 static LIST_HEAD(formats);
70 static DEFINE_RWLOCK(binfmt_lock);
72 int register_binfmt(struct linux_binfmt * fmt)
76 write_lock(&binfmt_lock);
77 list_add(&fmt->lh, &formats);
78 write_unlock(&binfmt_lock);
82 EXPORT_SYMBOL(register_binfmt);
84 int unregister_binfmt(struct linux_binfmt * fmt)
86 write_lock(&binfmt_lock);
88 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 asmlinkage long sys_uselib(const char __user * library)
111 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
116 if (nd.mnt->mnt_flags & MNT_NOEXEC)
119 if (!S_ISREG(nd.dentry->d_inode->i_mode))
122 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
126 file = nameidata_to_filp(&nd, O_RDONLY);
127 error = PTR_ERR(file);
133 struct linux_binfmt * fmt;
135 read_lock(&binfmt_lock);
136 list_for_each_entry(fmt, &formats, lh) {
137 if (!fmt->load_shlib)
139 if (!try_module_get(fmt->module))
141 read_unlock(&binfmt_lock);
142 error = fmt->load_shlib(file);
143 read_lock(&binfmt_lock);
145 if (error != -ENOEXEC)
148 read_unlock(&binfmt_lock);
154 release_open_intent(&nd);
161 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
167 #ifdef CONFIG_STACK_GROWSUP
169 ret = expand_stack_downwards(bprm->vma, pos);
174 ret = get_user_pages(current, bprm->mm, pos,
175 1, write, 1, &page, NULL);
180 struct rlimit *rlim = current->signal->rlim;
181 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
184 * Limit to 1/4-th the stack size for the argv+env strings.
186 * - the remaining binfmt code will not run out of stack space,
187 * - the program will have a reasonable amount of stack left
190 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
199 static void put_arg_page(struct page *page)
204 static void free_arg_page(struct linux_binprm *bprm, int i)
208 static void free_arg_pages(struct linux_binprm *bprm)
212 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
215 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
218 static int __bprm_mm_init(struct linux_binprm *bprm)
221 struct vm_area_struct *vma = NULL;
222 struct mm_struct *mm = bprm->mm;
224 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
228 down_write(&mm->mmap_sem);
232 * Place the stack at the largest stack address the architecture
233 * supports. Later, we'll move this to an appropriate place. We don't
234 * use STACK_TOP because that can depend on attributes which aren't
237 vma->vm_end = STACK_TOP_MAX;
238 vma->vm_start = vma->vm_end - PAGE_SIZE;
240 vma->vm_flags = VM_STACK_FLAGS;
241 vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
242 err = insert_vm_struct(mm, vma);
244 up_write(&mm->mmap_sem);
248 mm->stack_vm = mm->total_vm = 1;
249 up_write(&mm->mmap_sem);
251 bprm->p = vma->vm_end - sizeof(void *);
258 kmem_cache_free(vm_area_cachep, vma);
264 static bool valid_arg_len(struct linux_binprm *bprm, long len)
266 return len <= MAX_ARG_STRLEN;
271 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
276 page = bprm->page[pos / PAGE_SIZE];
277 if (!page && write) {
278 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
281 bprm->page[pos / PAGE_SIZE] = page;
287 static void put_arg_page(struct page *page)
291 static void free_arg_page(struct linux_binprm *bprm, int i)
294 __free_page(bprm->page[i]);
295 bprm->page[i] = NULL;
299 static void free_arg_pages(struct linux_binprm *bprm)
303 for (i = 0; i < MAX_ARG_PAGES; i++)
304 free_arg_page(bprm, i);
307 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
312 static int __bprm_mm_init(struct linux_binprm *bprm)
314 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
318 static bool valid_arg_len(struct linux_binprm *bprm, long len)
320 return len <= bprm->p;
323 #endif /* CONFIG_MMU */
326 * Create a new mm_struct and populate it with a temporary stack
327 * vm_area_struct. We don't have enough context at this point to set the stack
328 * flags, permissions, and offset, so we use temporary values. We'll update
329 * them later in setup_arg_pages().
331 int bprm_mm_init(struct linux_binprm *bprm)
334 struct mm_struct *mm = NULL;
336 bprm->mm = mm = mm_alloc();
341 err = init_new_context(current, mm);
345 err = __bprm_mm_init(bprm);
361 * count() counts the number of strings in array ARGV.
363 static int count(char __user * __user * argv, int max)
371 if (get_user(p, argv))
385 * 'copy_strings()' copies argument/environment strings from the old
386 * processes's memory to the new process's stack. The call to get_user_pages()
387 * ensures the destination page is created and not swapped out.
389 static int copy_strings(int argc, char __user * __user * argv,
390 struct linux_binprm *bprm)
392 struct page *kmapped_page = NULL;
394 unsigned long kpos = 0;
402 if (get_user(str, argv+argc) ||
403 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
408 if (!valid_arg_len(bprm, len)) {
413 /* We're going to work our way backwords. */
419 int offset, bytes_to_copy;
421 offset = pos % PAGE_SIZE;
425 bytes_to_copy = offset;
426 if (bytes_to_copy > len)
429 offset -= bytes_to_copy;
430 pos -= bytes_to_copy;
431 str -= bytes_to_copy;
432 len -= bytes_to_copy;
434 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
437 page = get_arg_page(bprm, pos, 1);
444 flush_kernel_dcache_page(kmapped_page);
445 kunmap(kmapped_page);
446 put_arg_page(kmapped_page);
449 kaddr = kmap(kmapped_page);
450 kpos = pos & PAGE_MASK;
451 flush_arg_page(bprm, kpos, kmapped_page);
453 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
462 flush_kernel_dcache_page(kmapped_page);
463 kunmap(kmapped_page);
464 put_arg_page(kmapped_page);
470 * Like copy_strings, but get argv and its values from kernel memory.
472 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
475 mm_segment_t oldfs = get_fs();
477 r = copy_strings(argc, (char __user * __user *)argv, bprm);
481 EXPORT_SYMBOL(copy_strings_kernel);
486 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
487 * the binfmt code determines where the new stack should reside, we shift it to
488 * its final location. The process proceeds as follows:
490 * 1) Use shift to calculate the new vma endpoints.
491 * 2) Extend vma to cover both the old and new ranges. This ensures the
492 * arguments passed to subsequent functions are consistent.
493 * 3) Move vma's page tables to the new range.
494 * 4) Free up any cleared pgd range.
495 * 5) Shrink the vma to cover only the new range.
497 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
499 struct mm_struct *mm = vma->vm_mm;
500 unsigned long old_start = vma->vm_start;
501 unsigned long old_end = vma->vm_end;
502 unsigned long length = old_end - old_start;
503 unsigned long new_start = old_start - shift;
504 unsigned long new_end = old_end - shift;
505 struct mmu_gather *tlb;
507 BUG_ON(new_start > new_end);
510 * ensure there are no vmas between where we want to go
513 if (vma != find_vma(mm, new_start))
517 * cover the whole range: [new_start, old_end)
519 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
522 * move the page tables downwards, on failure we rely on
523 * process cleanup to remove whatever mess we made.
525 if (length != move_page_tables(vma, old_start,
526 vma, new_start, length))
530 tlb = tlb_gather_mmu(mm, 0);
531 if (new_end > old_start) {
533 * when the old and new regions overlap clear from new_end.
535 free_pgd_range(&tlb, new_end, old_end, new_end,
536 vma->vm_next ? vma->vm_next->vm_start : 0);
539 * otherwise, clean from old_start; this is done to not touch
540 * the address space in [new_end, old_start) some architectures
541 * have constraints on va-space that make this illegal (IA64) -
542 * for the others its just a little faster.
544 free_pgd_range(&tlb, old_start, old_end, new_end,
545 vma->vm_next ? vma->vm_next->vm_start : 0);
547 tlb_finish_mmu(tlb, new_end, old_end);
550 * shrink the vma to just the new range.
552 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
557 #define EXTRA_STACK_VM_PAGES 20 /* random */
560 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
561 * the stack is optionally relocated, and some extra space is added.
563 int setup_arg_pages(struct linux_binprm *bprm,
564 unsigned long stack_top,
565 int executable_stack)
568 unsigned long stack_shift;
569 struct mm_struct *mm = current->mm;
570 struct vm_area_struct *vma = bprm->vma;
571 struct vm_area_struct *prev = NULL;
572 unsigned long vm_flags;
573 unsigned long stack_base;
575 #ifdef CONFIG_STACK_GROWSUP
576 /* Limit stack size to 1GB */
577 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
578 if (stack_base > (1 << 30))
579 stack_base = 1 << 30;
581 /* Make sure we didn't let the argument array grow too large. */
582 if (vma->vm_end - vma->vm_start > stack_base)
585 stack_base = PAGE_ALIGN(stack_top - stack_base);
587 stack_shift = vma->vm_start - stack_base;
588 mm->arg_start = bprm->p - stack_shift;
589 bprm->p = vma->vm_end - stack_shift;
591 stack_top = arch_align_stack(stack_top);
592 stack_top = PAGE_ALIGN(stack_top);
593 stack_shift = vma->vm_end - stack_top;
595 bprm->p -= stack_shift;
596 mm->arg_start = bprm->p;
600 bprm->loader -= stack_shift;
601 bprm->exec -= stack_shift;
603 down_write(&mm->mmap_sem);
604 vm_flags = vma->vm_flags;
607 * Adjust stack execute permissions; explicitly enable for
608 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
609 * (arch default) otherwise.
611 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
613 else if (executable_stack == EXSTACK_DISABLE_X)
614 vm_flags &= ~VM_EXEC;
615 vm_flags |= mm->def_flags;
617 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
623 /* Move stack pages down in memory. */
625 ret = shift_arg_pages(vma, stack_shift);
627 up_write(&mm->mmap_sem);
632 #ifdef CONFIG_STACK_GROWSUP
633 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
635 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
637 ret = expand_stack(vma, stack_base);
642 up_write(&mm->mmap_sem);
645 EXPORT_SYMBOL(setup_arg_pages);
647 #endif /* CONFIG_MMU */
649 struct file *open_exec(const char *name)
655 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
659 struct inode *inode = nd.dentry->d_inode;
660 file = ERR_PTR(-EACCES);
661 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
662 S_ISREG(inode->i_mode)) {
663 int err = vfs_permission(&nd, MAY_EXEC);
666 file = nameidata_to_filp(&nd, O_RDONLY);
668 err = deny_write_access(file);
678 release_open_intent(&nd);
684 EXPORT_SYMBOL(open_exec);
686 int kernel_read(struct file *file, unsigned long offset,
687 char *addr, unsigned long count)
695 /* The cast to a user pointer is valid due to the set_fs() */
696 result = vfs_read(file, (void __user *)addr, count, &pos);
701 EXPORT_SYMBOL(kernel_read);
703 static int exec_mmap(struct mm_struct *mm)
705 struct task_struct *tsk;
706 struct mm_struct * old_mm, *active_mm;
708 /* Notify parent that we're no longer interested in the old VM */
710 old_mm = current->mm;
711 mm_release(tsk, old_mm);
715 * Make sure that if there is a core dump in progress
716 * for the old mm, we get out and die instead of going
717 * through with the exec. We must hold mmap_sem around
718 * checking core_waiters and changing tsk->mm. The
719 * core-inducing thread will increment core_waiters for
720 * each thread whose ->mm == old_mm.
722 down_read(&old_mm->mmap_sem);
723 if (unlikely(old_mm->core_waiters)) {
724 up_read(&old_mm->mmap_sem);
729 active_mm = tsk->active_mm;
732 activate_mm(active_mm, mm);
734 arch_pick_mmap_layout(mm);
736 up_read(&old_mm->mmap_sem);
737 BUG_ON(active_mm != old_mm);
746 * This function makes sure the current process has its own signal table,
747 * so that flush_signal_handlers can later reset the handlers without
748 * disturbing other processes. (Other processes might share the signal
749 * table via the CLONE_SIGHAND option to clone().)
751 static int de_thread(struct task_struct *tsk)
753 struct signal_struct *sig = tsk->signal;
754 struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
755 spinlock_t *lock = &oldsighand->siglock;
756 struct task_struct *leader = NULL;
760 * If we don't share sighandlers, then we aren't sharing anything
761 * and we can just re-use it all.
763 if (atomic_read(&oldsighand->count) <= 1) {
768 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
772 if (thread_group_empty(tsk))
773 goto no_thread_group;
776 * Kill all other threads in the thread group.
777 * We must hold tasklist_lock to call zap_other_threads.
779 read_lock(&tasklist_lock);
781 if (sig->flags & SIGNAL_GROUP_EXIT) {
783 * Another group action in progress, just
784 * return so that the signal is processed.
786 spin_unlock_irq(lock);
787 read_unlock(&tasklist_lock);
788 kmem_cache_free(sighand_cachep, newsighand);
793 * child_reaper ignores SIGKILL, change it now.
794 * Reparenting needs write_lock on tasklist_lock,
795 * so it is safe to do it under read_lock.
797 if (unlikely(tsk->group_leader == child_reaper(tsk)))
798 tsk->nsproxy->pid_ns->child_reaper = tsk;
800 zap_other_threads(tsk);
801 read_unlock(&tasklist_lock);
804 * Account for the thread group leader hanging around:
807 if (!thread_group_leader(tsk)) {
810 * The SIGALRM timer survives the exec, but needs to point
811 * at us as the new group leader now. We have a race with
812 * a timer firing now getting the old leader, so we need to
813 * synchronize with any firing (by calling del_timer_sync)
814 * before we can safely let the old group leader die.
817 spin_unlock_irq(lock);
818 if (hrtimer_cancel(&sig->real_timer))
819 hrtimer_restart(&sig->real_timer);
822 while (atomic_read(&sig->count) > count) {
823 sig->group_exit_task = tsk;
824 sig->notify_count = count;
825 __set_current_state(TASK_UNINTERRUPTIBLE);
826 spin_unlock_irq(lock);
830 sig->group_exit_task = NULL;
831 sig->notify_count = 0;
832 spin_unlock_irq(lock);
835 * At this point all other threads have exited, all we have to
836 * do is to wait for the thread group leader to become inactive,
837 * and to assume its PID:
839 if (!thread_group_leader(tsk)) {
841 * Wait for the thread group leader to be a zombie.
842 * It should already be zombie at this point, most
845 leader = tsk->group_leader;
846 while (leader->exit_state != EXIT_ZOMBIE)
850 * The only record we have of the real-time age of a
851 * process, regardless of execs it's done, is start_time.
852 * All the past CPU time is accumulated in signal_struct
853 * from sister threads now dead. But in this non-leader
854 * exec, nothing survives from the original leader thread,
855 * whose birth marks the true age of this process now.
856 * When we take on its identity by switching to its PID, we
857 * also take its birthdate (always earlier than our own).
859 tsk->start_time = leader->start_time;
861 write_lock_irq(&tasklist_lock);
863 BUG_ON(leader->tgid != tsk->tgid);
864 BUG_ON(tsk->pid == tsk->tgid);
866 * An exec() starts a new thread group with the
867 * TGID of the previous thread group. Rehash the
868 * two threads with a switched PID, and release
869 * the former thread group leader:
872 /* Become a process group leader with the old leader's pid.
873 * The old leader becomes a thread of the this thread group.
874 * Note: The old leader also uses this pid until release_task
875 * is called. Odd but simple and correct.
877 detach_pid(tsk, PIDTYPE_PID);
878 tsk->pid = leader->pid;
879 attach_pid(tsk, PIDTYPE_PID, find_pid(tsk->pid));
880 transfer_pid(leader, tsk, PIDTYPE_PGID);
881 transfer_pid(leader, tsk, PIDTYPE_SID);
882 list_replace_rcu(&leader->tasks, &tsk->tasks);
884 tsk->group_leader = tsk;
885 leader->group_leader = tsk;
887 tsk->exit_signal = SIGCHLD;
889 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
890 leader->exit_state = EXIT_DEAD;
892 write_unlock_irq(&tasklist_lock);
896 * There may be one thread left which is just exiting,
897 * but it's safe to stop telling the group to kill themselves.
904 release_task(leader);
906 if (atomic_read(&oldsighand->count) == 1) {
908 * Now that we nuked the rest of the thread group,
909 * it turns out we are not sharing sighand any more either.
910 * So we can just keep it.
912 kmem_cache_free(sighand_cachep, newsighand);
915 * Move our state over to newsighand and switch it in.
917 atomic_set(&newsighand->count, 1);
918 memcpy(newsighand->action, oldsighand->action,
919 sizeof(newsighand->action));
921 write_lock_irq(&tasklist_lock);
922 spin_lock(&oldsighand->siglock);
923 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
925 rcu_assign_pointer(tsk->sighand, newsighand);
928 spin_unlock(&newsighand->siglock);
929 spin_unlock(&oldsighand->siglock);
930 write_unlock_irq(&tasklist_lock);
932 __cleanup_sighand(oldsighand);
935 BUG_ON(!thread_group_leader(tsk));
940 * These functions flushes out all traces of the currently running executable
941 * so that a new one can be started
944 static void flush_old_files(struct files_struct * files)
949 spin_lock(&files->file_lock);
951 unsigned long set, i;
955 fdt = files_fdtable(files);
956 if (i >= fdt->max_fds)
958 set = fdt->close_on_exec->fds_bits[j];
961 fdt->close_on_exec->fds_bits[j] = 0;
962 spin_unlock(&files->file_lock);
963 for ( ; set ; i++,set >>= 1) {
968 spin_lock(&files->file_lock);
971 spin_unlock(&files->file_lock);
974 void get_task_comm(char *buf, struct task_struct *tsk)
976 /* buf must be at least sizeof(tsk->comm) in size */
978 strncpy(buf, tsk->comm, sizeof(tsk->comm));
982 void set_task_comm(struct task_struct *tsk, char *buf)
985 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
989 int flush_old_exec(struct linux_binprm * bprm)
993 struct files_struct *files;
994 char tcomm[sizeof(current->comm)];
997 * Make sure we have a private signal table and that
998 * we are unassociated from the previous thread group.
1000 retval = de_thread(current);
1005 * Make sure we have private file handles. Ask the
1006 * fork helper to do the work for us and the exit
1007 * helper to do the cleanup of the old one.
1009 files = current->files; /* refcounted so safe to hold */
1010 retval = unshare_files();
1014 * Release all of the old mmap stuff
1016 retval = exec_mmap(bprm->mm);
1020 bprm->mm = NULL; /* We're using it now */
1022 /* This is the point of no return */
1023 put_files_struct(files);
1025 current->sas_ss_sp = current->sas_ss_size = 0;
1027 if (current->euid == current->uid && current->egid == current->gid)
1028 set_dumpable(current->mm, 1);
1030 set_dumpable(current->mm, suid_dumpable);
1032 name = bprm->filename;
1034 /* Copies the binary name from after last slash */
1035 for (i=0; (ch = *(name++)) != '\0';) {
1037 i = 0; /* overwrite what we wrote */
1039 if (i < (sizeof(tcomm) - 1))
1043 set_task_comm(current, tcomm);
1045 current->flags &= ~PF_RANDOMIZE;
1048 /* Set the new mm task size. We have to do that late because it may
1049 * depend on TIF_32BIT which is only updated in flush_thread() on
1050 * some architectures like powerpc
1052 current->mm->task_size = TASK_SIZE;
1054 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1056 set_dumpable(current->mm, suid_dumpable);
1057 current->pdeath_signal = 0;
1058 } else if (file_permission(bprm->file, MAY_READ) ||
1059 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1061 set_dumpable(current->mm, suid_dumpable);
1064 /* An exec changes our domain. We are no longer part of the thread
1067 current->self_exec_id++;
1069 flush_signal_handlers(current, 0);
1070 flush_old_files(current->files);
1075 reset_files_struct(current, files);
1080 EXPORT_SYMBOL(flush_old_exec);
1083 * Fill the binprm structure from the inode.
1084 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1086 int prepare_binprm(struct linux_binprm *bprm)
1089 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1092 mode = inode->i_mode;
1093 if (bprm->file->f_op == NULL)
1096 bprm->e_uid = current->euid;
1097 bprm->e_gid = current->egid;
1099 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1101 if (mode & S_ISUID) {
1102 current->personality &= ~PER_CLEAR_ON_SETID;
1103 bprm->e_uid = inode->i_uid;
1108 * If setgid is set but no group execute bit then this
1109 * is a candidate for mandatory locking, not a setgid
1112 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1113 current->personality &= ~PER_CLEAR_ON_SETID;
1114 bprm->e_gid = inode->i_gid;
1118 /* fill in binprm security blob */
1119 retval = security_bprm_set(bprm);
1123 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1124 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1127 EXPORT_SYMBOL(prepare_binprm);
1129 static int unsafe_exec(struct task_struct *p)
1132 if (p->ptrace & PT_PTRACED) {
1133 if (p->ptrace & PT_PTRACE_CAP)
1134 unsafe |= LSM_UNSAFE_PTRACE_CAP;
1136 unsafe |= LSM_UNSAFE_PTRACE;
1138 if (atomic_read(&p->fs->count) > 1 ||
1139 atomic_read(&p->files->count) > 1 ||
1140 atomic_read(&p->sighand->count) > 1)
1141 unsafe |= LSM_UNSAFE_SHARE;
1146 void compute_creds(struct linux_binprm *bprm)
1150 if (bprm->e_uid != current->uid) {
1152 current->pdeath_signal = 0;
1157 unsafe = unsafe_exec(current);
1158 security_bprm_apply_creds(bprm, unsafe);
1159 task_unlock(current);
1160 security_bprm_post_apply_creds(bprm);
1162 EXPORT_SYMBOL(compute_creds);
1165 * Arguments are '\0' separated strings found at the location bprm->p
1166 * points to; chop off the first by relocating brpm->p to right after
1167 * the first '\0' encountered.
1169 int remove_arg_zero(struct linux_binprm *bprm)
1172 unsigned long offset;
1180 offset = bprm->p & ~PAGE_MASK;
1181 page = get_arg_page(bprm, bprm->p, 0);
1186 kaddr = kmap_atomic(page, KM_USER0);
1188 for (; offset < PAGE_SIZE && kaddr[offset];
1189 offset++, bprm->p++)
1192 kunmap_atomic(kaddr, KM_USER0);
1195 if (offset == PAGE_SIZE)
1196 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1197 } while (offset == PAGE_SIZE);
1206 EXPORT_SYMBOL(remove_arg_zero);
1209 * cycle the list of binary formats handler, until one recognizes the image
1211 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1214 struct linux_binfmt *fmt;
1216 /* handle /sbin/loader.. */
1218 struct exec * eh = (struct exec *) bprm->buf;
1220 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1221 (eh->fh.f_flags & 0x3000) == 0x3000)
1224 unsigned long loader;
1226 allow_write_access(bprm->file);
1230 loader = bprm->vma->vm_end - sizeof(void *);
1232 file = open_exec("/sbin/loader");
1233 retval = PTR_ERR(file);
1237 /* Remember if the application is TASO. */
1238 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1241 bprm->loader = loader;
1242 retval = prepare_binprm(bprm);
1245 /* should call search_binary_handler recursively here,
1246 but it does not matter */
1250 retval = security_bprm_check(bprm);
1254 /* kernel module loader fixup */
1255 /* so we don't try to load run modprobe in kernel space. */
1258 retval = audit_bprm(bprm);
1263 for (try=0; try<2; try++) {
1264 read_lock(&binfmt_lock);
1265 list_for_each_entry(fmt, &formats, lh) {
1266 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1269 if (!try_module_get(fmt->module))
1271 read_unlock(&binfmt_lock);
1272 retval = fn(bprm, regs);
1275 allow_write_access(bprm->file);
1279 current->did_exec = 1;
1280 proc_exec_connector(current);
1283 read_lock(&binfmt_lock);
1285 if (retval != -ENOEXEC || bprm->mm == NULL)
1288 read_unlock(&binfmt_lock);
1292 read_unlock(&binfmt_lock);
1293 if (retval != -ENOEXEC || bprm->mm == NULL) {
1297 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1298 if (printable(bprm->buf[0]) &&
1299 printable(bprm->buf[1]) &&
1300 printable(bprm->buf[2]) &&
1301 printable(bprm->buf[3]))
1302 break; /* -ENOEXEC */
1303 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1310 EXPORT_SYMBOL(search_binary_handler);
1313 * sys_execve() executes a new program.
1315 int do_execve(char * filename,
1316 char __user *__user *argv,
1317 char __user *__user *envp,
1318 struct pt_regs * regs)
1320 struct linux_binprm *bprm;
1322 unsigned long env_p;
1326 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1330 file = open_exec(filename);
1331 retval = PTR_ERR(file);
1338 bprm->filename = filename;
1339 bprm->interp = filename;
1341 retval = bprm_mm_init(bprm);
1345 bprm->argc = count(argv, MAX_ARG_STRINGS);
1346 if ((retval = bprm->argc) < 0)
1349 bprm->envc = count(envp, MAX_ARG_STRINGS);
1350 if ((retval = bprm->envc) < 0)
1353 retval = security_bprm_alloc(bprm);
1357 retval = prepare_binprm(bprm);
1361 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1365 bprm->exec = bprm->p;
1366 retval = copy_strings(bprm->envc, envp, bprm);
1371 retval = copy_strings(bprm->argc, argv, bprm);
1374 bprm->argv_len = env_p - bprm->p;
1376 retval = search_binary_handler(bprm,regs);
1378 /* execve success */
1379 free_arg_pages(bprm);
1380 security_bprm_free(bprm);
1381 acct_update_integrals(current);
1387 free_arg_pages(bprm);
1389 security_bprm_free(bprm);
1397 allow_write_access(bprm->file);
1407 int set_binfmt(struct linux_binfmt *new)
1409 struct linux_binfmt *old = current->binfmt;
1412 if (!try_module_get(new->module))
1415 current->binfmt = new;
1417 module_put(old->module);
1421 EXPORT_SYMBOL(set_binfmt);
1423 /* format_corename will inspect the pattern parameter, and output a
1424 * name into corename, which must have space for at least
1425 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1427 static int format_corename(char *corename, const char *pattern, long signr)
1429 const char *pat_ptr = pattern;
1430 char *out_ptr = corename;
1431 char *const out_end = corename + CORENAME_MAX_SIZE;
1433 int pid_in_pattern = 0;
1436 if (*pattern == '|')
1439 /* Repeat as long as we have more pattern to process and more output
1442 if (*pat_ptr != '%') {
1443 if (out_ptr == out_end)
1445 *out_ptr++ = *pat_ptr++;
1447 switch (*++pat_ptr) {
1450 /* Double percent, output one percent */
1452 if (out_ptr == out_end)
1459 rc = snprintf(out_ptr, out_end - out_ptr,
1460 "%d", current->tgid);
1461 if (rc > out_end - out_ptr)
1467 rc = snprintf(out_ptr, out_end - out_ptr,
1468 "%d", current->uid);
1469 if (rc > out_end - out_ptr)
1475 rc = snprintf(out_ptr, out_end - out_ptr,
1476 "%d", current->gid);
1477 if (rc > out_end - out_ptr)
1481 /* signal that caused the coredump */
1483 rc = snprintf(out_ptr, out_end - out_ptr,
1485 if (rc > out_end - out_ptr)
1489 /* UNIX time of coredump */
1492 do_gettimeofday(&tv);
1493 rc = snprintf(out_ptr, out_end - out_ptr,
1495 if (rc > out_end - out_ptr)
1502 down_read(&uts_sem);
1503 rc = snprintf(out_ptr, out_end - out_ptr,
1504 "%s", utsname()->nodename);
1506 if (rc > out_end - out_ptr)
1512 rc = snprintf(out_ptr, out_end - out_ptr,
1513 "%s", current->comm);
1514 if (rc > out_end - out_ptr)
1524 /* Backward compatibility with core_uses_pid:
1526 * If core_pattern does not include a %p (as is the default)
1527 * and core_uses_pid is set, then .%pid will be appended to
1528 * the filename. Do not do this for piped commands. */
1529 if (!ispipe && !pid_in_pattern
1530 && (core_uses_pid || atomic_read(¤t->mm->mm_users) != 1)) {
1531 rc = snprintf(out_ptr, out_end - out_ptr,
1532 ".%d", current->tgid);
1533 if (rc > out_end - out_ptr)
1542 static void zap_process(struct task_struct *start)
1544 struct task_struct *t;
1546 start->signal->flags = SIGNAL_GROUP_EXIT;
1547 start->signal->group_stop_count = 0;
1551 if (t != current && t->mm) {
1552 t->mm->core_waiters++;
1553 sigaddset(&t->pending.signal, SIGKILL);
1554 signal_wake_up(t, 1);
1556 } while ((t = next_thread(t)) != start);
1559 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1562 struct task_struct *g, *p;
1563 unsigned long flags;
1566 spin_lock_irq(&tsk->sighand->siglock);
1567 if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1568 tsk->signal->group_exit_code = exit_code;
1572 spin_unlock_irq(&tsk->sighand->siglock);
1576 if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1580 for_each_process(g) {
1581 if (g == tsk->group_leader)
1589 * p->sighand can't disappear, but
1590 * may be changed by de_thread()
1592 lock_task_sighand(p, &flags);
1594 unlock_task_sighand(p, &flags);
1598 } while ((p = next_thread(p)) != g);
1602 return mm->core_waiters;
1605 static int coredump_wait(int exit_code)
1607 struct task_struct *tsk = current;
1608 struct mm_struct *mm = tsk->mm;
1609 struct completion startup_done;
1610 struct completion *vfork_done;
1613 init_completion(&mm->core_done);
1614 init_completion(&startup_done);
1615 mm->core_startup_done = &startup_done;
1617 core_waiters = zap_threads(tsk, mm, exit_code);
1618 up_write(&mm->mmap_sem);
1620 if (unlikely(core_waiters < 0))
1624 * Make sure nobody is waiting for us to release the VM,
1625 * otherwise we can deadlock when we wait on each other
1627 vfork_done = tsk->vfork_done;
1629 tsk->vfork_done = NULL;
1630 complete(vfork_done);
1634 wait_for_completion(&startup_done);
1636 BUG_ON(mm->core_waiters);
1637 return core_waiters;
1641 * set_dumpable converts traditional three-value dumpable to two flags and
1642 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1643 * these bits are not changed atomically. So get_dumpable can observe the
1644 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1645 * return either old dumpable or new one by paying attention to the order of
1646 * modifying the bits.
1648 * dumpable | mm->flags (binary)
1649 * old new | initial interim final
1650 * ---------+-----------------------
1658 * (*) get_dumpable regards interim value of 10 as 11.
1660 void set_dumpable(struct mm_struct *mm, int value)
1664 clear_bit(MMF_DUMPABLE, &mm->flags);
1666 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1669 set_bit(MMF_DUMPABLE, &mm->flags);
1671 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1674 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1676 set_bit(MMF_DUMPABLE, &mm->flags);
1680 EXPORT_SYMBOL_GPL(set_dumpable);
1682 int get_dumpable(struct mm_struct *mm)
1686 ret = mm->flags & 0x3;
1687 return (ret >= 2) ? 2 : ret;
1690 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1692 char corename[CORENAME_MAX_SIZE + 1];
1693 struct mm_struct *mm = current->mm;
1694 struct linux_binfmt * binfmt;
1695 struct inode * inode;
1698 int fsuid = current->fsuid;
1702 audit_core_dumps(signr);
1704 binfmt = current->binfmt;
1705 if (!binfmt || !binfmt->core_dump)
1707 down_write(&mm->mmap_sem);
1708 if (!get_dumpable(mm)) {
1709 up_write(&mm->mmap_sem);
1714 * We cannot trust fsuid as being the "true" uid of the
1715 * process nor do we know its entire history. We only know it
1716 * was tainted so we dump it as root in mode 2.
1718 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1719 flag = O_EXCL; /* Stop rewrite attacks */
1720 current->fsuid = 0; /* Dump root private */
1722 set_dumpable(mm, 0);
1724 retval = coredump_wait(exit_code);
1729 * Clear any false indication of pending signals that might
1730 * be seen by the filesystem code called to write the core file.
1732 clear_thread_flag(TIF_SIGPENDING);
1734 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1738 * lock_kernel() because format_corename() is controlled by sysctl, which
1739 * uses lock_kernel()
1742 ispipe = format_corename(corename, core_pattern, signr);
1745 /* SIGPIPE can happen, but it's just never processed */
1746 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1747 printk(KERN_INFO "Core dump to %s pipe failed\n",
1752 file = filp_open(corename,
1753 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1757 inode = file->f_path.dentry->d_inode;
1758 if (inode->i_nlink > 1)
1759 goto close_fail; /* multiple links - don't dump */
1760 if (!ispipe && d_unhashed(file->f_path.dentry))
1763 /* AK: actually i see no reason to not allow this for named pipes etc.,
1764 but keep the previous behaviour for now. */
1765 if (!ispipe && !S_ISREG(inode->i_mode))
1769 if (!file->f_op->write)
1771 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1774 retval = binfmt->core_dump(signr, regs, file);
1777 current->signal->group_exit_code |= 0x80;
1779 filp_close(file, NULL);
1781 current->fsuid = fsuid;
1782 complete_all(&mm->core_done);