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/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/highmem.h>
36 #include <linux/spinlock.h>
37 #include <linux/key.h>
38 #include <linux/personality.h>
39 #include <linux/binfmts.h>
40 #include <linux/swap.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/rmap.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
60 #include <linux/kmod.h>
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 int suid_dumpable = 0;
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 void unregister_binfmt(struct linux_binfmt * fmt)
86 write_lock(&binfmt_lock);
88 write_unlock(&binfmt_lock);
91 EXPORT_SYMBOL(unregister_binfmt);
93 static inline void put_binfmt(struct linux_binfmt * fmt)
95 module_put(fmt->module);
99 * Note that a shared library must be both readable and executable due to
102 * Also note that we take the address to load from from the file itself.
104 asmlinkage long sys_uselib(const char __user * library)
110 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
115 if (!S_ISREG(nd.dentry->d_inode->i_mode))
118 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
122 file = nameidata_to_filp(&nd, O_RDONLY);
123 error = PTR_ERR(file);
129 struct linux_binfmt * fmt;
131 read_lock(&binfmt_lock);
132 list_for_each_entry(fmt, &formats, lh) {
133 if (!fmt->load_shlib)
135 if (!try_module_get(fmt->module))
137 read_unlock(&binfmt_lock);
138 error = fmt->load_shlib(file);
139 read_lock(&binfmt_lock);
141 if (error != -ENOEXEC)
144 read_unlock(&binfmt_lock);
150 release_open_intent(&nd);
157 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
163 #ifdef CONFIG_STACK_GROWSUP
165 ret = expand_stack_downwards(bprm->vma, pos);
170 ret = get_user_pages(current, bprm->mm, pos,
171 1, write, 1, &page, NULL);
176 struct rlimit *rlim = current->signal->rlim;
177 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
180 * Limit to 1/4-th the stack size for the argv+env strings.
182 * - the remaining binfmt code will not run out of stack space,
183 * - the program will have a reasonable amount of stack left
186 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
195 static void put_arg_page(struct page *page)
200 static void free_arg_page(struct linux_binprm *bprm, int i)
204 static void free_arg_pages(struct linux_binprm *bprm)
208 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
211 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
214 static int __bprm_mm_init(struct linux_binprm *bprm)
217 struct vm_area_struct *vma = NULL;
218 struct mm_struct *mm = bprm->mm;
220 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
224 down_write(&mm->mmap_sem);
228 * Place the stack at the largest stack address the architecture
229 * supports. Later, we'll move this to an appropriate place. We don't
230 * use STACK_TOP because that can depend on attributes which aren't
233 vma->vm_end = STACK_TOP_MAX;
234 vma->vm_start = vma->vm_end - PAGE_SIZE;
236 vma->vm_flags = VM_STACK_FLAGS;
237 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
238 err = insert_vm_struct(mm, vma);
240 up_write(&mm->mmap_sem);
244 mm->stack_vm = mm->total_vm = 1;
245 up_write(&mm->mmap_sem);
247 bprm->p = vma->vm_end - sizeof(void *);
254 kmem_cache_free(vm_area_cachep, vma);
260 static bool valid_arg_len(struct linux_binprm *bprm, long len)
262 return len <= MAX_ARG_STRLEN;
267 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
272 page = bprm->page[pos / PAGE_SIZE];
273 if (!page && write) {
274 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
277 bprm->page[pos / PAGE_SIZE] = page;
283 static void put_arg_page(struct page *page)
287 static void free_arg_page(struct linux_binprm *bprm, int i)
290 __free_page(bprm->page[i]);
291 bprm->page[i] = NULL;
295 static void free_arg_pages(struct linux_binprm *bprm)
299 for (i = 0; i < MAX_ARG_PAGES; i++)
300 free_arg_page(bprm, i);
303 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
308 static int __bprm_mm_init(struct linux_binprm *bprm)
310 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
314 static bool valid_arg_len(struct linux_binprm *bprm, long len)
316 return len <= bprm->p;
319 #endif /* CONFIG_MMU */
322 * Create a new mm_struct and populate it with a temporary stack
323 * vm_area_struct. We don't have enough context at this point to set the stack
324 * flags, permissions, and offset, so we use temporary values. We'll update
325 * them later in setup_arg_pages().
327 int bprm_mm_init(struct linux_binprm *bprm)
330 struct mm_struct *mm = NULL;
332 bprm->mm = mm = mm_alloc();
337 err = init_new_context(current, mm);
341 err = __bprm_mm_init(bprm);
357 * count() counts the number of strings in array ARGV.
359 static int count(char __user * __user * argv, int max)
367 if (get_user(p, argv))
381 * 'copy_strings()' copies argument/environment strings from the old
382 * processes's memory to the new process's stack. The call to get_user_pages()
383 * ensures the destination page is created and not swapped out.
385 static int copy_strings(int argc, char __user * __user * argv,
386 struct linux_binprm *bprm)
388 struct page *kmapped_page = NULL;
390 unsigned long kpos = 0;
398 if (get_user(str, argv+argc) ||
399 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
404 if (!valid_arg_len(bprm, len)) {
409 /* We're going to work our way backwords. */
415 int offset, bytes_to_copy;
417 offset = pos % PAGE_SIZE;
421 bytes_to_copy = offset;
422 if (bytes_to_copy > len)
425 offset -= bytes_to_copy;
426 pos -= bytes_to_copy;
427 str -= bytes_to_copy;
428 len -= bytes_to_copy;
430 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
433 page = get_arg_page(bprm, pos, 1);
440 flush_kernel_dcache_page(kmapped_page);
441 kunmap(kmapped_page);
442 put_arg_page(kmapped_page);
445 kaddr = kmap(kmapped_page);
446 kpos = pos & PAGE_MASK;
447 flush_arg_page(bprm, kpos, kmapped_page);
449 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
458 flush_kernel_dcache_page(kmapped_page);
459 kunmap(kmapped_page);
460 put_arg_page(kmapped_page);
466 * Like copy_strings, but get argv and its values from kernel memory.
468 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
471 mm_segment_t oldfs = get_fs();
473 r = copy_strings(argc, (char __user * __user *)argv, bprm);
477 EXPORT_SYMBOL(copy_strings_kernel);
482 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
483 * the binfmt code determines where the new stack should reside, we shift it to
484 * its final location. The process proceeds as follows:
486 * 1) Use shift to calculate the new vma endpoints.
487 * 2) Extend vma to cover both the old and new ranges. This ensures the
488 * arguments passed to subsequent functions are consistent.
489 * 3) Move vma's page tables to the new range.
490 * 4) Free up any cleared pgd range.
491 * 5) Shrink the vma to cover only the new range.
493 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
495 struct mm_struct *mm = vma->vm_mm;
496 unsigned long old_start = vma->vm_start;
497 unsigned long old_end = vma->vm_end;
498 unsigned long length = old_end - old_start;
499 unsigned long new_start = old_start - shift;
500 unsigned long new_end = old_end - shift;
501 struct mmu_gather *tlb;
503 BUG_ON(new_start > new_end);
506 * ensure there are no vmas between where we want to go
509 if (vma != find_vma(mm, new_start))
513 * cover the whole range: [new_start, old_end)
515 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
518 * move the page tables downwards, on failure we rely on
519 * process cleanup to remove whatever mess we made.
521 if (length != move_page_tables(vma, old_start,
522 vma, new_start, length))
526 tlb = tlb_gather_mmu(mm, 0);
527 if (new_end > old_start) {
529 * when the old and new regions overlap clear from new_end.
531 free_pgd_range(&tlb, new_end, old_end, new_end,
532 vma->vm_next ? vma->vm_next->vm_start : 0);
535 * otherwise, clean from old_start; this is done to not touch
536 * the address space in [new_end, old_start) some architectures
537 * have constraints on va-space that make this illegal (IA64) -
538 * for the others its just a little faster.
540 free_pgd_range(&tlb, old_start, old_end, new_end,
541 vma->vm_next ? vma->vm_next->vm_start : 0);
543 tlb_finish_mmu(tlb, new_end, old_end);
546 * shrink the vma to just the new range.
548 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
553 #define EXTRA_STACK_VM_PAGES 20 /* random */
556 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
557 * the stack is optionally relocated, and some extra space is added.
559 int setup_arg_pages(struct linux_binprm *bprm,
560 unsigned long stack_top,
561 int executable_stack)
564 unsigned long stack_shift;
565 struct mm_struct *mm = current->mm;
566 struct vm_area_struct *vma = bprm->vma;
567 struct vm_area_struct *prev = NULL;
568 unsigned long vm_flags;
569 unsigned long stack_base;
571 #ifdef CONFIG_STACK_GROWSUP
572 /* Limit stack size to 1GB */
573 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
574 if (stack_base > (1 << 30))
575 stack_base = 1 << 30;
577 /* Make sure we didn't let the argument array grow too large. */
578 if (vma->vm_end - vma->vm_start > stack_base)
581 stack_base = PAGE_ALIGN(stack_top - stack_base);
583 stack_shift = vma->vm_start - stack_base;
584 mm->arg_start = bprm->p - stack_shift;
585 bprm->p = vma->vm_end - stack_shift;
587 stack_top = arch_align_stack(stack_top);
588 stack_top = PAGE_ALIGN(stack_top);
589 stack_shift = vma->vm_end - stack_top;
591 bprm->p -= stack_shift;
592 mm->arg_start = bprm->p;
596 bprm->loader -= stack_shift;
597 bprm->exec -= stack_shift;
599 down_write(&mm->mmap_sem);
600 vm_flags = vma->vm_flags;
603 * Adjust stack execute permissions; explicitly enable for
604 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
605 * (arch default) otherwise.
607 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
609 else if (executable_stack == EXSTACK_DISABLE_X)
610 vm_flags &= ~VM_EXEC;
611 vm_flags |= mm->def_flags;
613 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
619 /* Move stack pages down in memory. */
621 ret = shift_arg_pages(vma, stack_shift);
623 up_write(&mm->mmap_sem);
628 #ifdef CONFIG_STACK_GROWSUP
629 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
631 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
633 ret = expand_stack(vma, stack_base);
638 up_write(&mm->mmap_sem);
641 EXPORT_SYMBOL(setup_arg_pages);
643 #endif /* CONFIG_MMU */
645 struct file *open_exec(const char *name)
651 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
655 struct inode *inode = nd.dentry->d_inode;
656 file = ERR_PTR(-EACCES);
657 if (S_ISREG(inode->i_mode)) {
658 int err = vfs_permission(&nd, MAY_EXEC);
661 file = nameidata_to_filp(&nd, O_RDONLY);
663 err = deny_write_access(file);
673 release_open_intent(&nd);
679 EXPORT_SYMBOL(open_exec);
681 int kernel_read(struct file *file, unsigned long offset,
682 char *addr, unsigned long count)
690 /* The cast to a user pointer is valid due to the set_fs() */
691 result = vfs_read(file, (void __user *)addr, count, &pos);
696 EXPORT_SYMBOL(kernel_read);
698 static int exec_mmap(struct mm_struct *mm)
700 struct task_struct *tsk;
701 struct mm_struct * old_mm, *active_mm;
703 /* Notify parent that we're no longer interested in the old VM */
705 old_mm = current->mm;
706 mm_release(tsk, old_mm);
710 * Make sure that if there is a core dump in progress
711 * for the old mm, we get out and die instead of going
712 * through with the exec. We must hold mmap_sem around
713 * checking core_waiters and changing tsk->mm. The
714 * core-inducing thread will increment core_waiters for
715 * each thread whose ->mm == old_mm.
717 down_read(&old_mm->mmap_sem);
718 if (unlikely(old_mm->core_waiters)) {
719 up_read(&old_mm->mmap_sem);
724 active_mm = tsk->active_mm;
727 activate_mm(active_mm, mm);
729 arch_pick_mmap_layout(mm);
731 up_read(&old_mm->mmap_sem);
732 BUG_ON(active_mm != old_mm);
741 * This function makes sure the current process has its own signal table,
742 * so that flush_signal_handlers can later reset the handlers without
743 * disturbing other processes. (Other processes might share the signal
744 * table via the CLONE_SIGHAND option to clone().)
746 static int de_thread(struct task_struct *tsk)
748 struct signal_struct *sig = tsk->signal;
749 struct sighand_struct *oldsighand = tsk->sighand;
750 spinlock_t *lock = &oldsighand->siglock;
751 struct task_struct *leader = NULL;
754 if (thread_group_empty(tsk))
755 goto no_thread_group;
758 * Kill all other threads in the thread group.
759 * We must hold tasklist_lock to call zap_other_threads.
761 read_lock(&tasklist_lock);
763 if (sig->flags & SIGNAL_GROUP_EXIT) {
765 * Another group action in progress, just
766 * return so that the signal is processed.
768 spin_unlock_irq(lock);
769 read_unlock(&tasklist_lock);
774 * child_reaper ignores SIGKILL, change it now.
775 * Reparenting needs write_lock on tasklist_lock,
776 * so it is safe to do it under read_lock.
778 if (unlikely(tsk->group_leader == task_child_reaper(tsk)))
779 task_active_pid_ns(tsk)->child_reaper = tsk;
781 zap_other_threads(tsk);
782 read_unlock(&tasklist_lock);
785 * Account for the thread group leader hanging around:
788 if (!thread_group_leader(tsk)) {
791 * The SIGALRM timer survives the exec, but needs to point
792 * at us as the new group leader now. We have a race with
793 * a timer firing now getting the old leader, so we need to
794 * synchronize with any firing (by calling del_timer_sync)
795 * before we can safely let the old group leader die.
798 spin_unlock_irq(lock);
799 if (hrtimer_cancel(&sig->real_timer))
800 hrtimer_restart(&sig->real_timer);
804 sig->notify_count = count;
805 sig->group_exit_task = tsk;
806 while (atomic_read(&sig->count) > count) {
807 __set_current_state(TASK_UNINTERRUPTIBLE);
808 spin_unlock_irq(lock);
812 spin_unlock_irq(lock);
815 * At this point all other threads have exited, all we have to
816 * do is to wait for the thread group leader to become inactive,
817 * and to assume its PID:
819 if (!thread_group_leader(tsk)) {
820 leader = tsk->group_leader;
822 sig->notify_count = -1;
824 write_lock_irq(&tasklist_lock);
825 if (likely(leader->exit_state))
827 __set_current_state(TASK_UNINTERRUPTIBLE);
828 write_unlock_irq(&tasklist_lock);
833 * The only record we have of the real-time age of a
834 * process, regardless of execs it's done, is start_time.
835 * All the past CPU time is accumulated in signal_struct
836 * from sister threads now dead. But in this non-leader
837 * exec, nothing survives from the original leader thread,
838 * whose birth marks the true age of this process now.
839 * When we take on its identity by switching to its PID, we
840 * also take its birthdate (always earlier than our own).
842 tsk->start_time = leader->start_time;
844 BUG_ON(!same_thread_group(leader, tsk));
845 BUG_ON(has_group_leader_pid(tsk));
847 * An exec() starts a new thread group with the
848 * TGID of the previous thread group. Rehash the
849 * two threads with a switched PID, and release
850 * the former thread group leader:
853 /* Become a process group leader with the old leader's pid.
854 * The old leader becomes a thread of the this thread group.
855 * Note: The old leader also uses this pid until release_task
856 * is called. Odd but simple and correct.
858 detach_pid(tsk, PIDTYPE_PID);
859 tsk->pid = leader->pid;
860 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
861 transfer_pid(leader, tsk, PIDTYPE_PGID);
862 transfer_pid(leader, tsk, PIDTYPE_SID);
863 list_replace_rcu(&leader->tasks, &tsk->tasks);
865 tsk->group_leader = tsk;
866 leader->group_leader = tsk;
868 tsk->exit_signal = SIGCHLD;
870 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
871 leader->exit_state = EXIT_DEAD;
873 write_unlock_irq(&tasklist_lock);
876 sig->group_exit_task = NULL;
877 sig->notify_count = 0;
879 * There may be one thread left which is just exiting,
880 * but it's safe to stop telling the group to kill themselves.
887 release_task(leader);
889 if (atomic_read(&oldsighand->count) != 1) {
890 struct sighand_struct *newsighand;
892 * This ->sighand is shared with the CLONE_SIGHAND
893 * but not CLONE_THREAD task, switch to the new one.
895 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
899 atomic_set(&newsighand->count, 1);
900 memcpy(newsighand->action, oldsighand->action,
901 sizeof(newsighand->action));
903 write_lock_irq(&tasklist_lock);
904 spin_lock(&oldsighand->siglock);
905 rcu_assign_pointer(tsk->sighand, newsighand);
906 spin_unlock(&oldsighand->siglock);
907 write_unlock_irq(&tasklist_lock);
909 __cleanup_sighand(oldsighand);
912 BUG_ON(!thread_group_leader(tsk));
917 * These functions flushes out all traces of the currently running executable
918 * so that a new one can be started
920 static void flush_old_files(struct files_struct * files)
925 spin_lock(&files->file_lock);
927 unsigned long set, i;
931 fdt = files_fdtable(files);
932 if (i >= fdt->max_fds)
934 set = fdt->close_on_exec->fds_bits[j];
937 fdt->close_on_exec->fds_bits[j] = 0;
938 spin_unlock(&files->file_lock);
939 for ( ; set ; i++,set >>= 1) {
944 spin_lock(&files->file_lock);
947 spin_unlock(&files->file_lock);
950 char *get_task_comm(char *buf, struct task_struct *tsk)
952 /* buf must be at least sizeof(tsk->comm) in size */
954 strncpy(buf, tsk->comm, sizeof(tsk->comm));
959 void set_task_comm(struct task_struct *tsk, char *buf)
962 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
966 int flush_old_exec(struct linux_binprm * bprm)
970 struct files_struct *files;
971 char tcomm[sizeof(current->comm)];
974 * Make sure we have a private signal table and that
975 * we are unassociated from the previous thread group.
977 retval = de_thread(current);
982 * Make sure we have private file handles. Ask the
983 * fork helper to do the work for us and the exit
984 * helper to do the cleanup of the old one.
986 files = current->files; /* refcounted so safe to hold */
987 retval = unshare_files();
991 * Release all of the old mmap stuff
993 retval = exec_mmap(bprm->mm);
997 bprm->mm = NULL; /* We're using it now */
999 /* This is the point of no return */
1000 put_files_struct(files);
1002 current->sas_ss_sp = current->sas_ss_size = 0;
1004 if (current->euid == current->uid && current->egid == current->gid)
1005 set_dumpable(current->mm, 1);
1007 set_dumpable(current->mm, suid_dumpable);
1009 name = bprm->filename;
1011 /* Copies the binary name from after last slash */
1012 for (i=0; (ch = *(name++)) != '\0';) {
1014 i = 0; /* overwrite what we wrote */
1016 if (i < (sizeof(tcomm) - 1))
1020 set_task_comm(current, tcomm);
1022 current->flags &= ~PF_RANDOMIZE;
1025 /* Set the new mm task size. We have to do that late because it may
1026 * depend on TIF_32BIT which is only updated in flush_thread() on
1027 * some architectures like powerpc
1029 current->mm->task_size = TASK_SIZE;
1031 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1033 set_dumpable(current->mm, suid_dumpable);
1034 current->pdeath_signal = 0;
1035 } else if (file_permission(bprm->file, MAY_READ) ||
1036 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1038 set_dumpable(current->mm, suid_dumpable);
1041 /* An exec changes our domain. We are no longer part of the thread
1044 current->self_exec_id++;
1046 flush_signal_handlers(current, 0);
1047 flush_old_files(current->files);
1052 reset_files_struct(current, files);
1057 EXPORT_SYMBOL(flush_old_exec);
1060 * Fill the binprm structure from the inode.
1061 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1063 int prepare_binprm(struct linux_binprm *bprm)
1066 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1069 mode = inode->i_mode;
1070 if (bprm->file->f_op == NULL)
1073 bprm->e_uid = current->euid;
1074 bprm->e_gid = current->egid;
1076 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1078 if (mode & S_ISUID) {
1079 current->personality &= ~PER_CLEAR_ON_SETID;
1080 bprm->e_uid = inode->i_uid;
1085 * If setgid is set but no group execute bit then this
1086 * is a candidate for mandatory locking, not a setgid
1089 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1090 current->personality &= ~PER_CLEAR_ON_SETID;
1091 bprm->e_gid = inode->i_gid;
1095 /* fill in binprm security blob */
1096 retval = security_bprm_set(bprm);
1100 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1101 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1104 EXPORT_SYMBOL(prepare_binprm);
1106 static int unsafe_exec(struct task_struct *p)
1109 if (p->ptrace & PT_PTRACED) {
1110 if (p->ptrace & PT_PTRACE_CAP)
1111 unsafe |= LSM_UNSAFE_PTRACE_CAP;
1113 unsafe |= LSM_UNSAFE_PTRACE;
1115 if (atomic_read(&p->fs->count) > 1 ||
1116 atomic_read(&p->files->count) > 1 ||
1117 atomic_read(&p->sighand->count) > 1)
1118 unsafe |= LSM_UNSAFE_SHARE;
1123 void compute_creds(struct linux_binprm *bprm)
1127 if (bprm->e_uid != current->uid) {
1129 current->pdeath_signal = 0;
1134 unsafe = unsafe_exec(current);
1135 security_bprm_apply_creds(bprm, unsafe);
1136 task_unlock(current);
1137 security_bprm_post_apply_creds(bprm);
1139 EXPORT_SYMBOL(compute_creds);
1142 * Arguments are '\0' separated strings found at the location bprm->p
1143 * points to; chop off the first by relocating brpm->p to right after
1144 * the first '\0' encountered.
1146 int remove_arg_zero(struct linux_binprm *bprm)
1149 unsigned long offset;
1157 offset = bprm->p & ~PAGE_MASK;
1158 page = get_arg_page(bprm, bprm->p, 0);
1163 kaddr = kmap_atomic(page, KM_USER0);
1165 for (; offset < PAGE_SIZE && kaddr[offset];
1166 offset++, bprm->p++)
1169 kunmap_atomic(kaddr, KM_USER0);
1172 if (offset == PAGE_SIZE)
1173 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1174 } while (offset == PAGE_SIZE);
1183 EXPORT_SYMBOL(remove_arg_zero);
1186 * cycle the list of binary formats handler, until one recognizes the image
1188 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1191 struct linux_binfmt *fmt;
1193 /* handle /sbin/loader.. */
1195 struct exec * eh = (struct exec *) bprm->buf;
1197 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1198 (eh->fh.f_flags & 0x3000) == 0x3000)
1201 unsigned long loader;
1203 allow_write_access(bprm->file);
1207 loader = bprm->vma->vm_end - sizeof(void *);
1209 file = open_exec("/sbin/loader");
1210 retval = PTR_ERR(file);
1214 /* Remember if the application is TASO. */
1215 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1218 bprm->loader = loader;
1219 retval = prepare_binprm(bprm);
1222 /* should call search_binary_handler recursively here,
1223 but it does not matter */
1227 retval = security_bprm_check(bprm);
1231 /* kernel module loader fixup */
1232 /* so we don't try to load run modprobe in kernel space. */
1235 retval = audit_bprm(bprm);
1240 for (try=0; try<2; try++) {
1241 read_lock(&binfmt_lock);
1242 list_for_each_entry(fmt, &formats, lh) {
1243 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1246 if (!try_module_get(fmt->module))
1248 read_unlock(&binfmt_lock);
1249 retval = fn(bprm, regs);
1252 allow_write_access(bprm->file);
1256 current->did_exec = 1;
1257 proc_exec_connector(current);
1260 read_lock(&binfmt_lock);
1262 if (retval != -ENOEXEC || bprm->mm == NULL)
1265 read_unlock(&binfmt_lock);
1269 read_unlock(&binfmt_lock);
1270 if (retval != -ENOEXEC || bprm->mm == NULL) {
1274 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1275 if (printable(bprm->buf[0]) &&
1276 printable(bprm->buf[1]) &&
1277 printable(bprm->buf[2]) &&
1278 printable(bprm->buf[3]))
1279 break; /* -ENOEXEC */
1280 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1287 EXPORT_SYMBOL(search_binary_handler);
1290 * sys_execve() executes a new program.
1292 int do_execve(char * filename,
1293 char __user *__user *argv,
1294 char __user *__user *envp,
1295 struct pt_regs * regs)
1297 struct linux_binprm *bprm;
1299 unsigned long env_p;
1303 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1307 file = open_exec(filename);
1308 retval = PTR_ERR(file);
1315 bprm->filename = filename;
1316 bprm->interp = filename;
1318 retval = bprm_mm_init(bprm);
1322 bprm->argc = count(argv, MAX_ARG_STRINGS);
1323 if ((retval = bprm->argc) < 0)
1326 bprm->envc = count(envp, MAX_ARG_STRINGS);
1327 if ((retval = bprm->envc) < 0)
1330 retval = security_bprm_alloc(bprm);
1334 retval = prepare_binprm(bprm);
1338 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1342 bprm->exec = bprm->p;
1343 retval = copy_strings(bprm->envc, envp, bprm);
1348 retval = copy_strings(bprm->argc, argv, bprm);
1351 bprm->argv_len = env_p - bprm->p;
1353 retval = search_binary_handler(bprm,regs);
1355 /* execve success */
1356 free_arg_pages(bprm);
1357 security_bprm_free(bprm);
1358 acct_update_integrals(current);
1364 free_arg_pages(bprm);
1366 security_bprm_free(bprm);
1374 allow_write_access(bprm->file);
1384 int set_binfmt(struct linux_binfmt *new)
1386 struct linux_binfmt *old = current->binfmt;
1389 if (!try_module_get(new->module))
1392 current->binfmt = new;
1394 module_put(old->module);
1398 EXPORT_SYMBOL(set_binfmt);
1400 /* format_corename will inspect the pattern parameter, and output a
1401 * name into corename, which must have space for at least
1402 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1404 static int format_corename(char *corename, const char *pattern, long signr)
1406 const char *pat_ptr = pattern;
1407 char *out_ptr = corename;
1408 char *const out_end = corename + CORENAME_MAX_SIZE;
1410 int pid_in_pattern = 0;
1413 if (*pattern == '|')
1416 /* Repeat as long as we have more pattern to process and more output
1419 if (*pat_ptr != '%') {
1420 if (out_ptr == out_end)
1422 *out_ptr++ = *pat_ptr++;
1424 switch (*++pat_ptr) {
1427 /* Double percent, output one percent */
1429 if (out_ptr == out_end)
1436 rc = snprintf(out_ptr, out_end - out_ptr,
1437 "%d", task_tgid_vnr(current));
1438 if (rc > out_end - out_ptr)
1444 rc = snprintf(out_ptr, out_end - out_ptr,
1445 "%d", current->uid);
1446 if (rc > out_end - out_ptr)
1452 rc = snprintf(out_ptr, out_end - out_ptr,
1453 "%d", current->gid);
1454 if (rc > out_end - out_ptr)
1458 /* signal that caused the coredump */
1460 rc = snprintf(out_ptr, out_end - out_ptr,
1462 if (rc > out_end - out_ptr)
1466 /* UNIX time of coredump */
1469 do_gettimeofday(&tv);
1470 rc = snprintf(out_ptr, out_end - out_ptr,
1472 if (rc > out_end - out_ptr)
1479 down_read(&uts_sem);
1480 rc = snprintf(out_ptr, out_end - out_ptr,
1481 "%s", utsname()->nodename);
1483 if (rc > out_end - out_ptr)
1489 rc = snprintf(out_ptr, out_end - out_ptr,
1490 "%s", current->comm);
1491 if (rc > out_end - out_ptr)
1495 /* core limit size */
1497 rc = snprintf(out_ptr, out_end - out_ptr,
1498 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1499 if (rc > out_end - out_ptr)
1509 /* Backward compatibility with core_uses_pid:
1511 * If core_pattern does not include a %p (as is the default)
1512 * and core_uses_pid is set, then .%pid will be appended to
1513 * the filename. Do not do this for piped commands. */
1514 if (!ispipe && !pid_in_pattern
1515 && (core_uses_pid || atomic_read(¤t->mm->mm_users) != 1)) {
1516 rc = snprintf(out_ptr, out_end - out_ptr,
1517 ".%d", task_tgid_vnr(current));
1518 if (rc > out_end - out_ptr)
1527 static void zap_process(struct task_struct *start)
1529 struct task_struct *t;
1531 start->signal->flags = SIGNAL_GROUP_EXIT;
1532 start->signal->group_stop_count = 0;
1536 if (t != current && t->mm) {
1537 t->mm->core_waiters++;
1538 sigaddset(&t->pending.signal, SIGKILL);
1539 signal_wake_up(t, 1);
1541 } while ((t = next_thread(t)) != start);
1544 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1547 struct task_struct *g, *p;
1548 unsigned long flags;
1551 spin_lock_irq(&tsk->sighand->siglock);
1552 if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1553 tsk->signal->group_exit_code = exit_code;
1557 spin_unlock_irq(&tsk->sighand->siglock);
1561 if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1565 for_each_process(g) {
1566 if (g == tsk->group_leader)
1574 * p->sighand can't disappear, but
1575 * may be changed by de_thread()
1577 lock_task_sighand(p, &flags);
1579 unlock_task_sighand(p, &flags);
1583 } while ((p = next_thread(p)) != g);
1587 return mm->core_waiters;
1590 static int coredump_wait(int exit_code)
1592 struct task_struct *tsk = current;
1593 struct mm_struct *mm = tsk->mm;
1594 struct completion startup_done;
1595 struct completion *vfork_done;
1598 init_completion(&mm->core_done);
1599 init_completion(&startup_done);
1600 mm->core_startup_done = &startup_done;
1602 core_waiters = zap_threads(tsk, mm, exit_code);
1603 up_write(&mm->mmap_sem);
1605 if (unlikely(core_waiters < 0))
1609 * Make sure nobody is waiting for us to release the VM,
1610 * otherwise we can deadlock when we wait on each other
1612 vfork_done = tsk->vfork_done;
1614 tsk->vfork_done = NULL;
1615 complete(vfork_done);
1619 wait_for_completion(&startup_done);
1621 BUG_ON(mm->core_waiters);
1622 return core_waiters;
1626 * set_dumpable converts traditional three-value dumpable to two flags and
1627 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1628 * these bits are not changed atomically. So get_dumpable can observe the
1629 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1630 * return either old dumpable or new one by paying attention to the order of
1631 * modifying the bits.
1633 * dumpable | mm->flags (binary)
1634 * old new | initial interim final
1635 * ---------+-----------------------
1643 * (*) get_dumpable regards interim value of 10 as 11.
1645 void set_dumpable(struct mm_struct *mm, int value)
1649 clear_bit(MMF_DUMPABLE, &mm->flags);
1651 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1654 set_bit(MMF_DUMPABLE, &mm->flags);
1656 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1659 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1661 set_bit(MMF_DUMPABLE, &mm->flags);
1666 int get_dumpable(struct mm_struct *mm)
1670 ret = mm->flags & 0x3;
1671 return (ret >= 2) ? 2 : ret;
1674 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1676 char corename[CORENAME_MAX_SIZE + 1];
1677 struct mm_struct *mm = current->mm;
1678 struct linux_binfmt * binfmt;
1679 struct inode * inode;
1682 int fsuid = current->fsuid;
1685 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1686 char **helper_argv = NULL;
1687 int helper_argc = 0;
1690 audit_core_dumps(signr);
1692 binfmt = current->binfmt;
1693 if (!binfmt || !binfmt->core_dump)
1695 down_write(&mm->mmap_sem);
1697 * If another thread got here first, or we are not dumpable, bail out.
1699 if (mm->core_waiters || !get_dumpable(mm)) {
1700 up_write(&mm->mmap_sem);
1705 * We cannot trust fsuid as being the "true" uid of the
1706 * process nor do we know its entire history. We only know it
1707 * was tainted so we dump it as root in mode 2.
1709 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1710 flag = O_EXCL; /* Stop rewrite attacks */
1711 current->fsuid = 0; /* Dump root private */
1714 retval = coredump_wait(exit_code);
1719 * Clear any false indication of pending signals that might
1720 * be seen by the filesystem code called to write the core file.
1722 clear_thread_flag(TIF_SIGPENDING);
1725 * lock_kernel() because format_corename() is controlled by sysctl, which
1726 * uses lock_kernel()
1729 ispipe = format_corename(corename, core_pattern, signr);
1732 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1733 * to a pipe. Since we're not writing directly to the filesystem
1734 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1735 * created unless the pipe reader choses to write out the core file
1736 * at which point file size limits and permissions will be imposed
1737 * as it does with any other process
1739 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1743 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1744 /* Terminate the string before the first option */
1745 delimit = strchr(corename, ' ');
1748 delimit = strrchr(helper_argv[0], '/');
1752 delimit = helper_argv[0];
1753 if (!strcmp(delimit, current->comm)) {
1754 printk(KERN_NOTICE "Recursive core dump detected, "
1759 core_limit = RLIM_INFINITY;
1761 /* SIGPIPE can happen, but it's just never processed */
1762 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1764 printk(KERN_INFO "Core dump to %s pipe failed\n",
1769 file = filp_open(corename,
1770 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1774 inode = file->f_path.dentry->d_inode;
1775 if (inode->i_nlink > 1)
1776 goto close_fail; /* multiple links - don't dump */
1777 if (!ispipe && d_unhashed(file->f_path.dentry))
1780 /* AK: actually i see no reason to not allow this for named pipes etc.,
1781 but keep the previous behaviour for now. */
1782 if (!ispipe && !S_ISREG(inode->i_mode))
1785 * Dont allow local users get cute and trick others to coredump
1786 * into their pre-created files:
1788 if (inode->i_uid != current->fsuid)
1792 if (!file->f_op->write)
1794 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1797 retval = binfmt->core_dump(signr, regs, file, core_limit);
1800 current->signal->group_exit_code |= 0x80;
1802 filp_close(file, NULL);
1805 argv_free(helper_argv);
1807 current->fsuid = fsuid;
1808 complete_all(&mm->core_done);