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 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 (nd.mnt->mnt_flags & MNT_NOEXEC)
118 if (!S_ISREG(nd.dentry->d_inode->i_mode))
121 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
125 file = nameidata_to_filp(&nd, O_RDONLY);
126 error = PTR_ERR(file);
132 struct linux_binfmt * fmt;
134 read_lock(&binfmt_lock);
135 list_for_each_entry(fmt, &formats, lh) {
136 if (!fmt->load_shlib)
138 if (!try_module_get(fmt->module))
140 read_unlock(&binfmt_lock);
141 error = fmt->load_shlib(file);
142 read_lock(&binfmt_lock);
144 if (error != -ENOEXEC)
147 read_unlock(&binfmt_lock);
153 release_open_intent(&nd);
160 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
166 #ifdef CONFIG_STACK_GROWSUP
168 ret = expand_stack_downwards(bprm->vma, pos);
173 ret = get_user_pages(current, bprm->mm, pos,
174 1, write, 1, &page, NULL);
179 struct rlimit *rlim = current->signal->rlim;
180 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
183 * Limit to 1/4-th the stack size for the argv+env strings.
185 * - the remaining binfmt code will not run out of stack space,
186 * - the program will have a reasonable amount of stack left
189 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
198 static void put_arg_page(struct page *page)
203 static void free_arg_page(struct linux_binprm *bprm, int i)
207 static void free_arg_pages(struct linux_binprm *bprm)
211 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
214 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
217 static int __bprm_mm_init(struct linux_binprm *bprm)
220 struct vm_area_struct *vma = NULL;
221 struct mm_struct *mm = bprm->mm;
223 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
227 down_write(&mm->mmap_sem);
231 * Place the stack at the largest stack address the architecture
232 * supports. Later, we'll move this to an appropriate place. We don't
233 * use STACK_TOP because that can depend on attributes which aren't
236 vma->vm_end = STACK_TOP_MAX;
237 vma->vm_start = vma->vm_end - PAGE_SIZE;
239 vma->vm_flags = VM_STACK_FLAGS;
240 vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
241 err = insert_vm_struct(mm, vma);
243 up_write(&mm->mmap_sem);
247 mm->stack_vm = mm->total_vm = 1;
248 up_write(&mm->mmap_sem);
250 bprm->p = vma->vm_end - sizeof(void *);
257 kmem_cache_free(vm_area_cachep, vma);
263 static bool valid_arg_len(struct linux_binprm *bprm, long len)
265 return len <= MAX_ARG_STRLEN;
270 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
275 page = bprm->page[pos / PAGE_SIZE];
276 if (!page && write) {
277 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
280 bprm->page[pos / PAGE_SIZE] = page;
286 static void put_arg_page(struct page *page)
290 static void free_arg_page(struct linux_binprm *bprm, int i)
293 __free_page(bprm->page[i]);
294 bprm->page[i] = NULL;
298 static void free_arg_pages(struct linux_binprm *bprm)
302 for (i = 0; i < MAX_ARG_PAGES; i++)
303 free_arg_page(bprm, i);
306 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
311 static int __bprm_mm_init(struct linux_binprm *bprm)
313 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
317 static bool valid_arg_len(struct linux_binprm *bprm, long len)
319 return len <= bprm->p;
322 #endif /* CONFIG_MMU */
325 * Create a new mm_struct and populate it with a temporary stack
326 * vm_area_struct. We don't have enough context at this point to set the stack
327 * flags, permissions, and offset, so we use temporary values. We'll update
328 * them later in setup_arg_pages().
330 int bprm_mm_init(struct linux_binprm *bprm)
333 struct mm_struct *mm = NULL;
335 bprm->mm = mm = mm_alloc();
340 err = init_new_context(current, mm);
344 err = __bprm_mm_init(bprm);
360 * count() counts the number of strings in array ARGV.
362 static int count(char __user * __user * argv, int max)
370 if (get_user(p, argv))
384 * 'copy_strings()' copies argument/environment strings from the old
385 * processes's memory to the new process's stack. The call to get_user_pages()
386 * ensures the destination page is created and not swapped out.
388 static int copy_strings(int argc, char __user * __user * argv,
389 struct linux_binprm *bprm)
391 struct page *kmapped_page = NULL;
393 unsigned long kpos = 0;
401 if (get_user(str, argv+argc) ||
402 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
407 if (!valid_arg_len(bprm, len)) {
412 /* We're going to work our way backwords. */
418 int offset, bytes_to_copy;
420 offset = pos % PAGE_SIZE;
424 bytes_to_copy = offset;
425 if (bytes_to_copy > len)
428 offset -= bytes_to_copy;
429 pos -= bytes_to_copy;
430 str -= bytes_to_copy;
431 len -= bytes_to_copy;
433 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
436 page = get_arg_page(bprm, pos, 1);
443 flush_kernel_dcache_page(kmapped_page);
444 kunmap(kmapped_page);
445 put_arg_page(kmapped_page);
448 kaddr = kmap(kmapped_page);
449 kpos = pos & PAGE_MASK;
450 flush_arg_page(bprm, kpos, kmapped_page);
452 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
461 flush_kernel_dcache_page(kmapped_page);
462 kunmap(kmapped_page);
463 put_arg_page(kmapped_page);
469 * Like copy_strings, but get argv and its values from kernel memory.
471 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
474 mm_segment_t oldfs = get_fs();
476 r = copy_strings(argc, (char __user * __user *)argv, bprm);
480 EXPORT_SYMBOL(copy_strings_kernel);
485 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
486 * the binfmt code determines where the new stack should reside, we shift it to
487 * its final location. The process proceeds as follows:
489 * 1) Use shift to calculate the new vma endpoints.
490 * 2) Extend vma to cover both the old and new ranges. This ensures the
491 * arguments passed to subsequent functions are consistent.
492 * 3) Move vma's page tables to the new range.
493 * 4) Free up any cleared pgd range.
494 * 5) Shrink the vma to cover only the new range.
496 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
498 struct mm_struct *mm = vma->vm_mm;
499 unsigned long old_start = vma->vm_start;
500 unsigned long old_end = vma->vm_end;
501 unsigned long length = old_end - old_start;
502 unsigned long new_start = old_start - shift;
503 unsigned long new_end = old_end - shift;
504 struct mmu_gather *tlb;
506 BUG_ON(new_start > new_end);
509 * ensure there are no vmas between where we want to go
512 if (vma != find_vma(mm, new_start))
516 * cover the whole range: [new_start, old_end)
518 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
521 * move the page tables downwards, on failure we rely on
522 * process cleanup to remove whatever mess we made.
524 if (length != move_page_tables(vma, old_start,
525 vma, new_start, length))
529 tlb = tlb_gather_mmu(mm, 0);
530 if (new_end > old_start) {
532 * when the old and new regions overlap clear from new_end.
534 free_pgd_range(&tlb, new_end, old_end, new_end,
535 vma->vm_next ? vma->vm_next->vm_start : 0);
538 * otherwise, clean from old_start; this is done to not touch
539 * the address space in [new_end, old_start) some architectures
540 * have constraints on va-space that make this illegal (IA64) -
541 * for the others its just a little faster.
543 free_pgd_range(&tlb, old_start, old_end, new_end,
544 vma->vm_next ? vma->vm_next->vm_start : 0);
546 tlb_finish_mmu(tlb, new_end, old_end);
549 * shrink the vma to just the new range.
551 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
556 #define EXTRA_STACK_VM_PAGES 20 /* random */
559 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
560 * the stack is optionally relocated, and some extra space is added.
562 int setup_arg_pages(struct linux_binprm *bprm,
563 unsigned long stack_top,
564 int executable_stack)
567 unsigned long stack_shift;
568 struct mm_struct *mm = current->mm;
569 struct vm_area_struct *vma = bprm->vma;
570 struct vm_area_struct *prev = NULL;
571 unsigned long vm_flags;
572 unsigned long stack_base;
574 #ifdef CONFIG_STACK_GROWSUP
575 /* Limit stack size to 1GB */
576 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
577 if (stack_base > (1 << 30))
578 stack_base = 1 << 30;
580 /* Make sure we didn't let the argument array grow too large. */
581 if (vma->vm_end - vma->vm_start > stack_base)
584 stack_base = PAGE_ALIGN(stack_top - stack_base);
586 stack_shift = vma->vm_start - stack_base;
587 mm->arg_start = bprm->p - stack_shift;
588 bprm->p = vma->vm_end - stack_shift;
590 stack_top = arch_align_stack(stack_top);
591 stack_top = PAGE_ALIGN(stack_top);
592 stack_shift = vma->vm_end - stack_top;
594 bprm->p -= stack_shift;
595 mm->arg_start = bprm->p;
599 bprm->loader -= stack_shift;
600 bprm->exec -= stack_shift;
602 down_write(&mm->mmap_sem);
603 vm_flags = vma->vm_flags;
606 * Adjust stack execute permissions; explicitly enable for
607 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
608 * (arch default) otherwise.
610 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
612 else if (executable_stack == EXSTACK_DISABLE_X)
613 vm_flags &= ~VM_EXEC;
614 vm_flags |= mm->def_flags;
616 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
622 /* Move stack pages down in memory. */
624 ret = shift_arg_pages(vma, stack_shift);
626 up_write(&mm->mmap_sem);
631 #ifdef CONFIG_STACK_GROWSUP
632 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
634 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
636 ret = expand_stack(vma, stack_base);
641 up_write(&mm->mmap_sem);
644 EXPORT_SYMBOL(setup_arg_pages);
646 #endif /* CONFIG_MMU */
648 struct file *open_exec(const char *name)
654 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
658 struct inode *inode = nd.dentry->d_inode;
659 file = ERR_PTR(-EACCES);
660 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
661 S_ISREG(inode->i_mode)) {
662 int err = vfs_permission(&nd, MAY_EXEC);
665 file = nameidata_to_filp(&nd, O_RDONLY);
667 err = deny_write_access(file);
677 release_open_intent(&nd);
683 EXPORT_SYMBOL(open_exec);
685 int kernel_read(struct file *file, unsigned long offset,
686 char *addr, unsigned long count)
694 /* The cast to a user pointer is valid due to the set_fs() */
695 result = vfs_read(file, (void __user *)addr, count, &pos);
700 EXPORT_SYMBOL(kernel_read);
702 static int exec_mmap(struct mm_struct *mm)
704 struct task_struct *tsk;
705 struct mm_struct * old_mm, *active_mm;
707 /* Notify parent that we're no longer interested in the old VM */
709 old_mm = current->mm;
710 mm_release(tsk, old_mm);
714 * Make sure that if there is a core dump in progress
715 * for the old mm, we get out and die instead of going
716 * through with the exec. We must hold mmap_sem around
717 * checking core_waiters and changing tsk->mm. The
718 * core-inducing thread will increment core_waiters for
719 * each thread whose ->mm == old_mm.
721 down_read(&old_mm->mmap_sem);
722 if (unlikely(old_mm->core_waiters)) {
723 up_read(&old_mm->mmap_sem);
728 active_mm = tsk->active_mm;
731 activate_mm(active_mm, mm);
733 arch_pick_mmap_layout(mm);
735 up_read(&old_mm->mmap_sem);
736 BUG_ON(active_mm != old_mm);
745 * This function makes sure the current process has its own signal table,
746 * so that flush_signal_handlers can later reset the handlers without
747 * disturbing other processes. (Other processes might share the signal
748 * table via the CLONE_SIGHAND option to clone().)
750 static int de_thread(struct task_struct *tsk)
752 struct signal_struct *sig = tsk->signal;
753 struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
754 spinlock_t *lock = &oldsighand->siglock;
755 struct task_struct *leader = NULL;
759 * If we don't share sighandlers, then we aren't sharing anything
760 * and we can just re-use it all.
762 if (atomic_read(&oldsighand->count) <= 1) {
767 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
771 if (thread_group_empty(tsk))
772 goto no_thread_group;
775 * Kill all other threads in the thread group.
776 * We must hold tasklist_lock to call zap_other_threads.
778 read_lock(&tasklist_lock);
780 if (sig->flags & SIGNAL_GROUP_EXIT) {
782 * Another group action in progress, just
783 * return so that the signal is processed.
785 spin_unlock_irq(lock);
786 read_unlock(&tasklist_lock);
787 kmem_cache_free(sighand_cachep, newsighand);
792 * child_reaper ignores SIGKILL, change it now.
793 * Reparenting needs write_lock on tasklist_lock,
794 * so it is safe to do it under read_lock.
796 if (unlikely(tsk->group_leader == child_reaper(tsk)))
797 tsk->nsproxy->pid_ns->child_reaper = tsk;
799 zap_other_threads(tsk);
800 read_unlock(&tasklist_lock);
803 * Account for the thread group leader hanging around:
806 if (!thread_group_leader(tsk)) {
809 * The SIGALRM timer survives the exec, but needs to point
810 * at us as the new group leader now. We have a race with
811 * a timer firing now getting the old leader, so we need to
812 * synchronize with any firing (by calling del_timer_sync)
813 * before we can safely let the old group leader die.
816 spin_unlock_irq(lock);
817 if (hrtimer_cancel(&sig->real_timer))
818 hrtimer_restart(&sig->real_timer);
821 while (atomic_read(&sig->count) > count) {
822 sig->group_exit_task = tsk;
823 sig->notify_count = count;
824 __set_current_state(TASK_UNINTERRUPTIBLE);
825 spin_unlock_irq(lock);
829 sig->group_exit_task = NULL;
830 sig->notify_count = 0;
831 spin_unlock_irq(lock);
834 * At this point all other threads have exited, all we have to
835 * do is to wait for the thread group leader to become inactive,
836 * and to assume its PID:
838 if (!thread_group_leader(tsk)) {
840 * Wait for the thread group leader to be a zombie.
841 * It should already be zombie at this point, most
844 leader = tsk->group_leader;
845 while (leader->exit_state != EXIT_ZOMBIE)
849 * The only record we have of the real-time age of a
850 * process, regardless of execs it's done, is start_time.
851 * All the past CPU time is accumulated in signal_struct
852 * from sister threads now dead. But in this non-leader
853 * exec, nothing survives from the original leader thread,
854 * whose birth marks the true age of this process now.
855 * When we take on its identity by switching to its PID, we
856 * also take its birthdate (always earlier than our own).
858 tsk->start_time = leader->start_time;
860 write_lock_irq(&tasklist_lock);
862 BUG_ON(leader->tgid != tsk->tgid);
863 BUG_ON(tsk->pid == tsk->tgid);
865 * An exec() starts a new thread group with the
866 * TGID of the previous thread group. Rehash the
867 * two threads with a switched PID, and release
868 * the former thread group leader:
871 /* Become a process group leader with the old leader's pid.
872 * The old leader becomes a thread of the this thread group.
873 * Note: The old leader also uses this pid until release_task
874 * is called. Odd but simple and correct.
876 detach_pid(tsk, PIDTYPE_PID);
877 tsk->pid = leader->pid;
878 attach_pid(tsk, PIDTYPE_PID, find_pid(tsk->pid));
879 transfer_pid(leader, tsk, PIDTYPE_PGID);
880 transfer_pid(leader, tsk, PIDTYPE_SID);
881 list_replace_rcu(&leader->tasks, &tsk->tasks);
883 tsk->group_leader = tsk;
884 leader->group_leader = tsk;
886 tsk->exit_signal = SIGCHLD;
888 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
889 leader->exit_state = EXIT_DEAD;
891 write_unlock_irq(&tasklist_lock);
895 * There may be one thread left which is just exiting,
896 * but it's safe to stop telling the group to kill themselves.
903 release_task(leader);
905 if (atomic_read(&oldsighand->count) == 1) {
907 * Now that we nuked the rest of the thread group,
908 * it turns out we are not sharing sighand any more either.
909 * So we can just keep it.
911 kmem_cache_free(sighand_cachep, newsighand);
914 * Move our state over to newsighand and switch it in.
916 atomic_set(&newsighand->count, 1);
917 memcpy(newsighand->action, oldsighand->action,
918 sizeof(newsighand->action));
920 write_lock_irq(&tasklist_lock);
921 spin_lock(&oldsighand->siglock);
922 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
924 rcu_assign_pointer(tsk->sighand, newsighand);
927 spin_unlock(&newsighand->siglock);
928 spin_unlock(&oldsighand->siglock);
929 write_unlock_irq(&tasklist_lock);
931 __cleanup_sighand(oldsighand);
934 BUG_ON(!thread_group_leader(tsk));
939 * These functions flushes out all traces of the currently running executable
940 * so that a new one can be started
943 static void flush_old_files(struct files_struct * files)
948 spin_lock(&files->file_lock);
950 unsigned long set, i;
954 fdt = files_fdtable(files);
955 if (i >= fdt->max_fds)
957 set = fdt->close_on_exec->fds_bits[j];
960 fdt->close_on_exec->fds_bits[j] = 0;
961 spin_unlock(&files->file_lock);
962 for ( ; set ; i++,set >>= 1) {
967 spin_lock(&files->file_lock);
970 spin_unlock(&files->file_lock);
973 void get_task_comm(char *buf, struct task_struct *tsk)
975 /* buf must be at least sizeof(tsk->comm) in size */
977 strncpy(buf, tsk->comm, sizeof(tsk->comm));
981 void set_task_comm(struct task_struct *tsk, char *buf)
984 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
988 int flush_old_exec(struct linux_binprm * bprm)
992 struct files_struct *files;
993 char tcomm[sizeof(current->comm)];
996 * Make sure we have a private signal table and that
997 * we are unassociated from the previous thread group.
999 retval = de_thread(current);
1004 * Make sure we have private file handles. Ask the
1005 * fork helper to do the work for us and the exit
1006 * helper to do the cleanup of the old one.
1008 files = current->files; /* refcounted so safe to hold */
1009 retval = unshare_files();
1013 * Release all of the old mmap stuff
1015 retval = exec_mmap(bprm->mm);
1019 bprm->mm = NULL; /* We're using it now */
1021 /* This is the point of no return */
1022 put_files_struct(files);
1024 current->sas_ss_sp = current->sas_ss_size = 0;
1026 if (current->euid == current->uid && current->egid == current->gid)
1027 set_dumpable(current->mm, 1);
1029 set_dumpable(current->mm, suid_dumpable);
1031 name = bprm->filename;
1033 /* Copies the binary name from after last slash */
1034 for (i=0; (ch = *(name++)) != '\0';) {
1036 i = 0; /* overwrite what we wrote */
1038 if (i < (sizeof(tcomm) - 1))
1042 set_task_comm(current, tcomm);
1044 current->flags &= ~PF_RANDOMIZE;
1047 /* Set the new mm task size. We have to do that late because it may
1048 * depend on TIF_32BIT which is only updated in flush_thread() on
1049 * some architectures like powerpc
1051 current->mm->task_size = TASK_SIZE;
1053 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1055 set_dumpable(current->mm, suid_dumpable);
1056 current->pdeath_signal = 0;
1057 } else if (file_permission(bprm->file, MAY_READ) ||
1058 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1060 set_dumpable(current->mm, suid_dumpable);
1063 /* An exec changes our domain. We are no longer part of the thread
1066 current->self_exec_id++;
1068 flush_signal_handlers(current, 0);
1069 flush_old_files(current->files);
1074 reset_files_struct(current, files);
1079 EXPORT_SYMBOL(flush_old_exec);
1082 * Fill the binprm structure from the inode.
1083 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1085 int prepare_binprm(struct linux_binprm *bprm)
1088 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1091 mode = inode->i_mode;
1092 if (bprm->file->f_op == NULL)
1095 bprm->e_uid = current->euid;
1096 bprm->e_gid = current->egid;
1098 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1100 if (mode & S_ISUID) {
1101 current->personality &= ~PER_CLEAR_ON_SETID;
1102 bprm->e_uid = inode->i_uid;
1107 * If setgid is set but no group execute bit then this
1108 * is a candidate for mandatory locking, not a setgid
1111 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1112 current->personality &= ~PER_CLEAR_ON_SETID;
1113 bprm->e_gid = inode->i_gid;
1117 /* fill in binprm security blob */
1118 retval = security_bprm_set(bprm);
1122 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1123 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1126 EXPORT_SYMBOL(prepare_binprm);
1128 static int unsafe_exec(struct task_struct *p)
1131 if (p->ptrace & PT_PTRACED) {
1132 if (p->ptrace & PT_PTRACE_CAP)
1133 unsafe |= LSM_UNSAFE_PTRACE_CAP;
1135 unsafe |= LSM_UNSAFE_PTRACE;
1137 if (atomic_read(&p->fs->count) > 1 ||
1138 atomic_read(&p->files->count) > 1 ||
1139 atomic_read(&p->sighand->count) > 1)
1140 unsafe |= LSM_UNSAFE_SHARE;
1145 void compute_creds(struct linux_binprm *bprm)
1149 if (bprm->e_uid != current->uid) {
1151 current->pdeath_signal = 0;
1156 unsafe = unsafe_exec(current);
1157 security_bprm_apply_creds(bprm, unsafe);
1158 task_unlock(current);
1159 security_bprm_post_apply_creds(bprm);
1161 EXPORT_SYMBOL(compute_creds);
1164 * Arguments are '\0' separated strings found at the location bprm->p
1165 * points to; chop off the first by relocating brpm->p to right after
1166 * the first '\0' encountered.
1168 int remove_arg_zero(struct linux_binprm *bprm)
1171 unsigned long offset;
1179 offset = bprm->p & ~PAGE_MASK;
1180 page = get_arg_page(bprm, bprm->p, 0);
1185 kaddr = kmap_atomic(page, KM_USER0);
1187 for (; offset < PAGE_SIZE && kaddr[offset];
1188 offset++, bprm->p++)
1191 kunmap_atomic(kaddr, KM_USER0);
1194 if (offset == PAGE_SIZE)
1195 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1196 } while (offset == PAGE_SIZE);
1205 EXPORT_SYMBOL(remove_arg_zero);
1208 * cycle the list of binary formats handler, until one recognizes the image
1210 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1213 struct linux_binfmt *fmt;
1215 /* handle /sbin/loader.. */
1217 struct exec * eh = (struct exec *) bprm->buf;
1219 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1220 (eh->fh.f_flags & 0x3000) == 0x3000)
1223 unsigned long loader;
1225 allow_write_access(bprm->file);
1229 loader = bprm->vma->vm_end - sizeof(void *);
1231 file = open_exec("/sbin/loader");
1232 retval = PTR_ERR(file);
1236 /* Remember if the application is TASO. */
1237 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1240 bprm->loader = loader;
1241 retval = prepare_binprm(bprm);
1244 /* should call search_binary_handler recursively here,
1245 but it does not matter */
1249 retval = security_bprm_check(bprm);
1253 /* kernel module loader fixup */
1254 /* so we don't try to load run modprobe in kernel space. */
1257 retval = audit_bprm(bprm);
1262 for (try=0; try<2; try++) {
1263 read_lock(&binfmt_lock);
1264 list_for_each_entry(fmt, &formats, lh) {
1265 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1268 if (!try_module_get(fmt->module))
1270 read_unlock(&binfmt_lock);
1271 retval = fn(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) {
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 unsigned long env_p;
1325 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1329 file = open_exec(filename);
1330 retval = PTR_ERR(file);
1337 bprm->filename = filename;
1338 bprm->interp = filename;
1340 retval = bprm_mm_init(bprm);
1344 bprm->argc = count(argv, MAX_ARG_STRINGS);
1345 if ((retval = bprm->argc) < 0)
1348 bprm->envc = count(envp, MAX_ARG_STRINGS);
1349 if ((retval = bprm->envc) < 0)
1352 retval = security_bprm_alloc(bprm);
1356 retval = prepare_binprm(bprm);
1360 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1364 bprm->exec = bprm->p;
1365 retval = copy_strings(bprm->envc, envp, bprm);
1370 retval = copy_strings(bprm->argc, argv, bprm);
1373 bprm->argv_len = env_p - bprm->p;
1375 retval = search_binary_handler(bprm,regs);
1377 /* execve success */
1378 free_arg_pages(bprm);
1379 security_bprm_free(bprm);
1380 acct_update_integrals(current);
1386 free_arg_pages(bprm);
1388 security_bprm_free(bprm);
1396 allow_write_access(bprm->file);
1406 int set_binfmt(struct linux_binfmt *new)
1408 struct linux_binfmt *old = current->binfmt;
1411 if (!try_module_get(new->module))
1414 current->binfmt = new;
1416 module_put(old->module);
1420 EXPORT_SYMBOL(set_binfmt);
1422 /* format_corename will inspect the pattern parameter, and output a
1423 * name into corename, which must have space for at least
1424 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1426 static int format_corename(char *corename, const char *pattern, long signr)
1428 const char *pat_ptr = pattern;
1429 char *out_ptr = corename;
1430 char *const out_end = corename + CORENAME_MAX_SIZE;
1432 int pid_in_pattern = 0;
1435 if (*pattern == '|')
1438 /* Repeat as long as we have more pattern to process and more output
1441 if (*pat_ptr != '%') {
1442 if (out_ptr == out_end)
1444 *out_ptr++ = *pat_ptr++;
1446 switch (*++pat_ptr) {
1449 /* Double percent, output one percent */
1451 if (out_ptr == out_end)
1458 rc = snprintf(out_ptr, out_end - out_ptr,
1459 "%d", current->tgid);
1460 if (rc > out_end - out_ptr)
1466 rc = snprintf(out_ptr, out_end - out_ptr,
1467 "%d", current->uid);
1468 if (rc > out_end - out_ptr)
1474 rc = snprintf(out_ptr, out_end - out_ptr,
1475 "%d", current->gid);
1476 if (rc > out_end - out_ptr)
1480 /* signal that caused the coredump */
1482 rc = snprintf(out_ptr, out_end - out_ptr,
1484 if (rc > out_end - out_ptr)
1488 /* UNIX time of coredump */
1491 do_gettimeofday(&tv);
1492 rc = snprintf(out_ptr, out_end - out_ptr,
1494 if (rc > out_end - out_ptr)
1501 down_read(&uts_sem);
1502 rc = snprintf(out_ptr, out_end - out_ptr,
1503 "%s", utsname()->nodename);
1505 if (rc > out_end - out_ptr)
1511 rc = snprintf(out_ptr, out_end - out_ptr,
1512 "%s", current->comm);
1513 if (rc > out_end - out_ptr)
1523 /* Backward compatibility with core_uses_pid:
1525 * If core_pattern does not include a %p (as is the default)
1526 * and core_uses_pid is set, then .%pid will be appended to
1527 * the filename. Do not do this for piped commands. */
1528 if (!ispipe && !pid_in_pattern
1529 && (core_uses_pid || atomic_read(¤t->mm->mm_users) != 1)) {
1530 rc = snprintf(out_ptr, out_end - out_ptr,
1531 ".%d", current->tgid);
1532 if (rc > out_end - out_ptr)
1541 static void zap_process(struct task_struct *start)
1543 struct task_struct *t;
1545 start->signal->flags = SIGNAL_GROUP_EXIT;
1546 start->signal->group_stop_count = 0;
1550 if (t != current && t->mm) {
1551 t->mm->core_waiters++;
1552 sigaddset(&t->pending.signal, SIGKILL);
1553 signal_wake_up(t, 1);
1555 } while ((t = next_thread(t)) != start);
1558 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1561 struct task_struct *g, *p;
1562 unsigned long flags;
1565 spin_lock_irq(&tsk->sighand->siglock);
1566 if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1567 tsk->signal->group_exit_code = exit_code;
1571 spin_unlock_irq(&tsk->sighand->siglock);
1575 if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1579 for_each_process(g) {
1580 if (g == tsk->group_leader)
1588 * p->sighand can't disappear, but
1589 * may be changed by de_thread()
1591 lock_task_sighand(p, &flags);
1593 unlock_task_sighand(p, &flags);
1597 } while ((p = next_thread(p)) != g);
1601 return mm->core_waiters;
1604 static int coredump_wait(int exit_code)
1606 struct task_struct *tsk = current;
1607 struct mm_struct *mm = tsk->mm;
1608 struct completion startup_done;
1609 struct completion *vfork_done;
1612 init_completion(&mm->core_done);
1613 init_completion(&startup_done);
1614 mm->core_startup_done = &startup_done;
1616 core_waiters = zap_threads(tsk, mm, exit_code);
1617 up_write(&mm->mmap_sem);
1619 if (unlikely(core_waiters < 0))
1623 * Make sure nobody is waiting for us to release the VM,
1624 * otherwise we can deadlock when we wait on each other
1626 vfork_done = tsk->vfork_done;
1628 tsk->vfork_done = NULL;
1629 complete(vfork_done);
1633 wait_for_completion(&startup_done);
1635 BUG_ON(mm->core_waiters);
1636 return core_waiters;
1640 * set_dumpable converts traditional three-value dumpable to two flags and
1641 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1642 * these bits are not changed atomically. So get_dumpable can observe the
1643 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1644 * return either old dumpable or new one by paying attention to the order of
1645 * modifying the bits.
1647 * dumpable | mm->flags (binary)
1648 * old new | initial interim final
1649 * ---------+-----------------------
1657 * (*) get_dumpable regards interim value of 10 as 11.
1659 void set_dumpable(struct mm_struct *mm, int value)
1663 clear_bit(MMF_DUMPABLE, &mm->flags);
1665 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1668 set_bit(MMF_DUMPABLE, &mm->flags);
1670 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1673 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1675 set_bit(MMF_DUMPABLE, &mm->flags);
1679 EXPORT_SYMBOL_GPL(set_dumpable);
1681 int get_dumpable(struct mm_struct *mm)
1685 ret = mm->flags & 0x3;
1686 return (ret >= 2) ? 2 : ret;
1689 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1691 char corename[CORENAME_MAX_SIZE + 1];
1692 struct mm_struct *mm = current->mm;
1693 struct linux_binfmt * binfmt;
1694 struct inode * inode;
1697 int fsuid = current->fsuid;
1701 audit_core_dumps(signr);
1703 binfmt = current->binfmt;
1704 if (!binfmt || !binfmt->core_dump)
1706 down_write(&mm->mmap_sem);
1707 if (!get_dumpable(mm)) {
1708 up_write(&mm->mmap_sem);
1713 * We cannot trust fsuid as being the "true" uid of the
1714 * process nor do we know its entire history. We only know it
1715 * was tainted so we dump it as root in mode 2.
1717 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1718 flag = O_EXCL; /* Stop rewrite attacks */
1719 current->fsuid = 0; /* Dump root private */
1721 set_dumpable(mm, 0);
1723 retval = coredump_wait(exit_code);
1728 * Clear any false indication of pending signals that might
1729 * be seen by the filesystem code called to write the core file.
1731 clear_thread_flag(TIF_SIGPENDING);
1733 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1737 * lock_kernel() because format_corename() is controlled by sysctl, which
1738 * uses lock_kernel()
1741 ispipe = format_corename(corename, core_pattern, signr);
1744 /* SIGPIPE can happen, but it's just never processed */
1745 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1746 printk(KERN_INFO "Core dump to %s pipe failed\n",
1751 file = filp_open(corename,
1752 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1756 inode = file->f_path.dentry->d_inode;
1757 if (inode->i_nlink > 1)
1758 goto close_fail; /* multiple links - don't dump */
1759 if (!ispipe && d_unhashed(file->f_path.dentry))
1762 /* AK: actually i see no reason to not allow this for named pipes etc.,
1763 but keep the previous behaviour for now. */
1764 if (!ispipe && !S_ISREG(inode->i_mode))
1768 if (!file->f_op->write)
1770 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1773 retval = binfmt->core_dump(signr, regs, file);
1776 current->signal->group_exit_code |= 0x80;
1778 filp_close(file, NULL);
1780 current->fsuid = fsuid;
1781 complete_all(&mm->core_done);