NFS: Separate metadata and page cache revalidation mechanisms
[linux-2.6/mini2440.git] / fs / exec.c
blob3a79d97ac234422c353ad6c37ab973d3363471bd
1 /*
2 * linux/fs/exec.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
7 /*
8 * #!-checking implemented by tytso.
9 */
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
22 * formats.
25 #include <linux/config.h>
26 #include <linux/slab.h>
27 #include <linux/file.h>
28 #include <linux/mman.h>
29 #include <linux/a.out.h>
30 #include <linux/stat.h>
31 #include <linux/fcntl.h>
32 #include <linux/smp_lock.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/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/acct.h>
51 #include <linux/cn_proc.h>
53 #include <asm/uaccess.h>
54 #include <asm/mmu_context.h>
56 #ifdef CONFIG_KMOD
57 #include <linux/kmod.h>
58 #endif
60 int core_uses_pid;
61 char core_pattern[65] = "core";
62 int suid_dumpable = 0;
64 EXPORT_SYMBOL(suid_dumpable);
65 /* The maximal length of core_pattern is also specified in sysctl.c */
67 static struct linux_binfmt *formats;
68 static DEFINE_RWLOCK(binfmt_lock);
70 int register_binfmt(struct linux_binfmt * fmt)
72 struct linux_binfmt ** tmp = &formats;
74 if (!fmt)
75 return -EINVAL;
76 if (fmt->next)
77 return -EBUSY;
78 write_lock(&binfmt_lock);
79 while (*tmp) {
80 if (fmt == *tmp) {
81 write_unlock(&binfmt_lock);
82 return -EBUSY;
84 tmp = &(*tmp)->next;
86 fmt->next = formats;
87 formats = fmt;
88 write_unlock(&binfmt_lock);
89 return 0;
92 EXPORT_SYMBOL(register_binfmt);
94 int unregister_binfmt(struct linux_binfmt * fmt)
96 struct linux_binfmt ** tmp = &formats;
98 write_lock(&binfmt_lock);
99 while (*tmp) {
100 if (fmt == *tmp) {
101 *tmp = fmt->next;
102 write_unlock(&binfmt_lock);
103 return 0;
105 tmp = &(*tmp)->next;
107 write_unlock(&binfmt_lock);
108 return -EINVAL;
111 EXPORT_SYMBOL(unregister_binfmt);
113 static inline void put_binfmt(struct linux_binfmt * fmt)
115 module_put(fmt->module);
119 * Note that a shared library must be both readable and executable due to
120 * security reasons.
122 * Also note that we take the address to load from from the file itself.
124 asmlinkage long sys_uselib(const char __user * library)
126 struct file * file;
127 struct nameidata nd;
128 int error;
130 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
131 if (error)
132 goto out;
134 error = -EINVAL;
135 if (!S_ISREG(nd.dentry->d_inode->i_mode))
136 goto exit;
138 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
139 if (error)
140 goto exit;
142 file = nameidata_to_filp(&nd, O_RDONLY);
143 error = PTR_ERR(file);
144 if (IS_ERR(file))
145 goto out;
147 error = -ENOEXEC;
148 if(file->f_op) {
149 struct linux_binfmt * fmt;
151 read_lock(&binfmt_lock);
152 for (fmt = formats ; fmt ; fmt = fmt->next) {
153 if (!fmt->load_shlib)
154 continue;
155 if (!try_module_get(fmt->module))
156 continue;
157 read_unlock(&binfmt_lock);
158 error = fmt->load_shlib(file);
159 read_lock(&binfmt_lock);
160 put_binfmt(fmt);
161 if (error != -ENOEXEC)
162 break;
164 read_unlock(&binfmt_lock);
166 fput(file);
167 out:
168 return error;
169 exit:
170 release_open_intent(&nd);
171 path_release(&nd);
172 goto out;
176 * count() counts the number of strings in array ARGV.
178 static int count(char __user * __user * argv, int max)
180 int i = 0;
182 if (argv != NULL) {
183 for (;;) {
184 char __user * p;
186 if (get_user(p, argv))
187 return -EFAULT;
188 if (!p)
189 break;
190 argv++;
191 if(++i > max)
192 return -E2BIG;
193 cond_resched();
196 return i;
200 * 'copy_strings()' copies argument/environment strings from user
201 * memory to free pages in kernel mem. These are in a format ready
202 * to be put directly into the top of new user memory.
204 static int copy_strings(int argc, char __user * __user * argv,
205 struct linux_binprm *bprm)
207 struct page *kmapped_page = NULL;
208 char *kaddr = NULL;
209 int ret;
211 while (argc-- > 0) {
212 char __user *str;
213 int len;
214 unsigned long pos;
216 if (get_user(str, argv+argc) ||
217 !(len = strnlen_user(str, bprm->p))) {
218 ret = -EFAULT;
219 goto out;
222 if (bprm->p < len) {
223 ret = -E2BIG;
224 goto out;
227 bprm->p -= len;
228 /* XXX: add architecture specific overflow check here. */
229 pos = bprm->p;
231 while (len > 0) {
232 int i, new, err;
233 int offset, bytes_to_copy;
234 struct page *page;
236 offset = pos % PAGE_SIZE;
237 i = pos/PAGE_SIZE;
238 page = bprm->page[i];
239 new = 0;
240 if (!page) {
241 page = alloc_page(GFP_HIGHUSER);
242 bprm->page[i] = page;
243 if (!page) {
244 ret = -ENOMEM;
245 goto out;
247 new = 1;
250 if (page != kmapped_page) {
251 if (kmapped_page)
252 kunmap(kmapped_page);
253 kmapped_page = page;
254 kaddr = kmap(kmapped_page);
256 if (new && offset)
257 memset(kaddr, 0, offset);
258 bytes_to_copy = PAGE_SIZE - offset;
259 if (bytes_to_copy > len) {
260 bytes_to_copy = len;
261 if (new)
262 memset(kaddr+offset+len, 0,
263 PAGE_SIZE-offset-len);
265 err = copy_from_user(kaddr+offset, str, bytes_to_copy);
266 if (err) {
267 ret = -EFAULT;
268 goto out;
271 pos += bytes_to_copy;
272 str += bytes_to_copy;
273 len -= bytes_to_copy;
276 ret = 0;
277 out:
278 if (kmapped_page)
279 kunmap(kmapped_page);
280 return ret;
284 * Like copy_strings, but get argv and its values from kernel memory.
286 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
288 int r;
289 mm_segment_t oldfs = get_fs();
290 set_fs(KERNEL_DS);
291 r = copy_strings(argc, (char __user * __user *)argv, bprm);
292 set_fs(oldfs);
293 return r;
296 EXPORT_SYMBOL(copy_strings_kernel);
298 #ifdef CONFIG_MMU
300 * This routine is used to map in a page into an address space: needed by
301 * execve() for the initial stack and environment pages.
303 * vma->vm_mm->mmap_sem is held for writing.
305 void install_arg_page(struct vm_area_struct *vma,
306 struct page *page, unsigned long address)
308 struct mm_struct *mm = vma->vm_mm;
309 pte_t * pte;
310 spinlock_t *ptl;
312 if (unlikely(anon_vma_prepare(vma)))
313 goto out;
315 flush_dcache_page(page);
316 pte = get_locked_pte(mm, address, &ptl);
317 if (!pte)
318 goto out;
319 if (!pte_none(*pte)) {
320 pte_unmap_unlock(pte, ptl);
321 goto out;
323 inc_mm_counter(mm, anon_rss);
324 lru_cache_add_active(page);
325 set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
326 page, vma->vm_page_prot))));
327 page_add_new_anon_rmap(page, vma, address);
328 pte_unmap_unlock(pte, ptl);
330 /* no need for flush_tlb */
331 return;
332 out:
333 __free_page(page);
334 force_sig(SIGKILL, current);
337 #define EXTRA_STACK_VM_PAGES 20 /* random */
339 int setup_arg_pages(struct linux_binprm *bprm,
340 unsigned long stack_top,
341 int executable_stack)
343 unsigned long stack_base;
344 struct vm_area_struct *mpnt;
345 struct mm_struct *mm = current->mm;
346 int i, ret;
347 long arg_size;
349 #ifdef CONFIG_STACK_GROWSUP
350 /* Move the argument and environment strings to the bottom of the
351 * stack space.
353 int offset, j;
354 char *to, *from;
356 /* Start by shifting all the pages down */
357 i = 0;
358 for (j = 0; j < MAX_ARG_PAGES; j++) {
359 struct page *page = bprm->page[j];
360 if (!page)
361 continue;
362 bprm->page[i++] = page;
365 /* Now move them within their pages */
366 offset = bprm->p % PAGE_SIZE;
367 to = kmap(bprm->page[0]);
368 for (j = 1; j < i; j++) {
369 memmove(to, to + offset, PAGE_SIZE - offset);
370 from = kmap(bprm->page[j]);
371 memcpy(to + PAGE_SIZE - offset, from, offset);
372 kunmap(bprm->page[j - 1]);
373 to = from;
375 memmove(to, to + offset, PAGE_SIZE - offset);
376 kunmap(bprm->page[j - 1]);
378 /* Limit stack size to 1GB */
379 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
380 if (stack_base > (1 << 30))
381 stack_base = 1 << 30;
382 stack_base = PAGE_ALIGN(stack_top - stack_base);
384 /* Adjust bprm->p to point to the end of the strings. */
385 bprm->p = stack_base + PAGE_SIZE * i - offset;
387 mm->arg_start = stack_base;
388 arg_size = i << PAGE_SHIFT;
390 /* zero pages that were copied above */
391 while (i < MAX_ARG_PAGES)
392 bprm->page[i++] = NULL;
393 #else
394 stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
395 stack_base = PAGE_ALIGN(stack_base);
396 bprm->p += stack_base;
397 mm->arg_start = bprm->p;
398 arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
399 #endif
401 arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
403 if (bprm->loader)
404 bprm->loader += stack_base;
405 bprm->exec += stack_base;
407 mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
408 if (!mpnt)
409 return -ENOMEM;
411 memset(mpnt, 0, sizeof(*mpnt));
413 down_write(&mm->mmap_sem);
415 mpnt->vm_mm = mm;
416 #ifdef CONFIG_STACK_GROWSUP
417 mpnt->vm_start = stack_base;
418 mpnt->vm_end = stack_base + arg_size;
419 #else
420 mpnt->vm_end = stack_top;
421 mpnt->vm_start = mpnt->vm_end - arg_size;
422 #endif
423 /* Adjust stack execute permissions; explicitly enable
424 * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
425 * and leave alone (arch default) otherwise. */
426 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
427 mpnt->vm_flags = VM_STACK_FLAGS | VM_EXEC;
428 else if (executable_stack == EXSTACK_DISABLE_X)
429 mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
430 else
431 mpnt->vm_flags = VM_STACK_FLAGS;
432 mpnt->vm_flags |= mm->def_flags;
433 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
434 if ((ret = insert_vm_struct(mm, mpnt))) {
435 up_write(&mm->mmap_sem);
436 kmem_cache_free(vm_area_cachep, mpnt);
437 return ret;
439 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
442 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
443 struct page *page = bprm->page[i];
444 if (page) {
445 bprm->page[i] = NULL;
446 install_arg_page(mpnt, page, stack_base);
448 stack_base += PAGE_SIZE;
450 up_write(&mm->mmap_sem);
452 return 0;
455 EXPORT_SYMBOL(setup_arg_pages);
457 #define free_arg_pages(bprm) do { } while (0)
459 #else
461 static inline void free_arg_pages(struct linux_binprm *bprm)
463 int i;
465 for (i = 0; i < MAX_ARG_PAGES; i++) {
466 if (bprm->page[i])
467 __free_page(bprm->page[i]);
468 bprm->page[i] = NULL;
472 #endif /* CONFIG_MMU */
474 struct file *open_exec(const char *name)
476 struct nameidata nd;
477 int err;
478 struct file *file;
480 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
481 file = ERR_PTR(err);
483 if (!err) {
484 struct inode *inode = nd.dentry->d_inode;
485 file = ERR_PTR(-EACCES);
486 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
487 S_ISREG(inode->i_mode)) {
488 int err = vfs_permission(&nd, MAY_EXEC);
489 if (!err && !(inode->i_mode & 0111))
490 err = -EACCES;
491 file = ERR_PTR(err);
492 if (!err) {
493 file = nameidata_to_filp(&nd, O_RDONLY);
494 if (!IS_ERR(file)) {
495 err = deny_write_access(file);
496 if (err) {
497 fput(file);
498 file = ERR_PTR(err);
501 out:
502 return file;
505 release_open_intent(&nd);
506 path_release(&nd);
508 goto out;
511 EXPORT_SYMBOL(open_exec);
513 int kernel_read(struct file *file, unsigned long offset,
514 char *addr, unsigned long count)
516 mm_segment_t old_fs;
517 loff_t pos = offset;
518 int result;
520 old_fs = get_fs();
521 set_fs(get_ds());
522 /* The cast to a user pointer is valid due to the set_fs() */
523 result = vfs_read(file, (void __user *)addr, count, &pos);
524 set_fs(old_fs);
525 return result;
528 EXPORT_SYMBOL(kernel_read);
530 static int exec_mmap(struct mm_struct *mm)
532 struct task_struct *tsk;
533 struct mm_struct * old_mm, *active_mm;
535 /* Notify parent that we're no longer interested in the old VM */
536 tsk = current;
537 old_mm = current->mm;
538 mm_release(tsk, old_mm);
540 if (old_mm) {
542 * Make sure that if there is a core dump in progress
543 * for the old mm, we get out and die instead of going
544 * through with the exec. We must hold mmap_sem around
545 * checking core_waiters and changing tsk->mm. The
546 * core-inducing thread will increment core_waiters for
547 * each thread whose ->mm == old_mm.
549 down_read(&old_mm->mmap_sem);
550 if (unlikely(old_mm->core_waiters)) {
551 up_read(&old_mm->mmap_sem);
552 return -EINTR;
555 task_lock(tsk);
556 active_mm = tsk->active_mm;
557 tsk->mm = mm;
558 tsk->active_mm = mm;
559 activate_mm(active_mm, mm);
560 task_unlock(tsk);
561 arch_pick_mmap_layout(mm);
562 if (old_mm) {
563 up_read(&old_mm->mmap_sem);
564 BUG_ON(active_mm != old_mm);
565 mmput(old_mm);
566 return 0;
568 mmdrop(active_mm);
569 return 0;
573 * This function makes sure the current process has its own signal table,
574 * so that flush_signal_handlers can later reset the handlers without
575 * disturbing other processes. (Other processes might share the signal
576 * table via the CLONE_SIGHAND option to clone().)
578 static int de_thread(struct task_struct *tsk)
580 struct signal_struct *sig = tsk->signal;
581 struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
582 spinlock_t *lock = &oldsighand->siglock;
583 struct task_struct *leader = NULL;
584 int count;
587 * If we don't share sighandlers, then we aren't sharing anything
588 * and we can just re-use it all.
590 if (atomic_read(&oldsighand->count) <= 1) {
591 BUG_ON(atomic_read(&sig->count) != 1);
592 exit_itimers(sig);
593 return 0;
596 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
597 if (!newsighand)
598 return -ENOMEM;
600 if (thread_group_empty(current))
601 goto no_thread_group;
604 * Kill all other threads in the thread group.
605 * We must hold tasklist_lock to call zap_other_threads.
607 read_lock(&tasklist_lock);
608 spin_lock_irq(lock);
609 if (sig->flags & SIGNAL_GROUP_EXIT) {
611 * Another group action in progress, just
612 * return so that the signal is processed.
614 spin_unlock_irq(lock);
615 read_unlock(&tasklist_lock);
616 kmem_cache_free(sighand_cachep, newsighand);
617 return -EAGAIN;
621 * child_reaper ignores SIGKILL, change it now.
622 * Reparenting needs write_lock on tasklist_lock,
623 * so it is safe to do it under read_lock.
625 if (unlikely(current->group_leader == child_reaper))
626 child_reaper = current;
628 zap_other_threads(current);
629 read_unlock(&tasklist_lock);
632 * Account for the thread group leader hanging around:
634 count = 1;
635 if (!thread_group_leader(current)) {
636 count = 2;
638 * The SIGALRM timer survives the exec, but needs to point
639 * at us as the new group leader now. We have a race with
640 * a timer firing now getting the old leader, so we need to
641 * synchronize with any firing (by calling del_timer_sync)
642 * before we can safely let the old group leader die.
644 sig->tsk = current;
645 spin_unlock_irq(lock);
646 if (hrtimer_cancel(&sig->real_timer))
647 hrtimer_restart(&sig->real_timer);
648 spin_lock_irq(lock);
650 while (atomic_read(&sig->count) > count) {
651 sig->group_exit_task = current;
652 sig->notify_count = count;
653 __set_current_state(TASK_UNINTERRUPTIBLE);
654 spin_unlock_irq(lock);
655 schedule();
656 spin_lock_irq(lock);
658 sig->group_exit_task = NULL;
659 sig->notify_count = 0;
660 spin_unlock_irq(lock);
663 * At this point all other threads have exited, all we have to
664 * do is to wait for the thread group leader to become inactive,
665 * and to assume its PID:
667 if (!thread_group_leader(current)) {
668 struct dentry *proc_dentry1, *proc_dentry2;
671 * Wait for the thread group leader to be a zombie.
672 * It should already be zombie at this point, most
673 * of the time.
675 leader = current->group_leader;
676 while (leader->exit_state != EXIT_ZOMBIE)
677 yield();
680 * The only record we have of the real-time age of a
681 * process, regardless of execs it's done, is start_time.
682 * All the past CPU time is accumulated in signal_struct
683 * from sister threads now dead. But in this non-leader
684 * exec, nothing survives from the original leader thread,
685 * whose birth marks the true age of this process now.
686 * When we take on its identity by switching to its PID, we
687 * also take its birthdate (always earlier than our own).
689 current->start_time = leader->start_time;
691 spin_lock(&leader->proc_lock);
692 spin_lock(&current->proc_lock);
693 proc_dentry1 = proc_pid_unhash(current);
694 proc_dentry2 = proc_pid_unhash(leader);
695 write_lock_irq(&tasklist_lock);
697 BUG_ON(leader->tgid != current->tgid);
698 BUG_ON(current->pid == current->tgid);
700 * An exec() starts a new thread group with the
701 * TGID of the previous thread group. Rehash the
702 * two threads with a switched PID, and release
703 * the former thread group leader:
706 /* Become a process group leader with the old leader's pid.
707 * Note: The old leader also uses thispid until release_task
708 * is called. Odd but simple and correct.
710 detach_pid(current, PIDTYPE_PID);
711 current->pid = leader->pid;
712 attach_pid(current, PIDTYPE_PID, current->pid);
713 attach_pid(current, PIDTYPE_PGID, current->signal->pgrp);
714 attach_pid(current, PIDTYPE_SID, current->signal->session);
715 list_add_tail_rcu(&current->tasks, &init_task.tasks);
717 current->group_leader = current;
718 leader->group_leader = current;
720 /* Reduce leader to a thread */
721 detach_pid(leader, PIDTYPE_PGID);
722 detach_pid(leader, PIDTYPE_SID);
723 list_del_init(&leader->tasks);
725 current->exit_signal = SIGCHLD;
727 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
728 leader->exit_state = EXIT_DEAD;
730 write_unlock_irq(&tasklist_lock);
731 spin_unlock(&leader->proc_lock);
732 spin_unlock(&current->proc_lock);
733 proc_pid_flush(proc_dentry1);
734 proc_pid_flush(proc_dentry2);
738 * There may be one thread left which is just exiting,
739 * but it's safe to stop telling the group to kill themselves.
741 sig->flags = 0;
743 no_thread_group:
744 exit_itimers(sig);
745 if (leader)
746 release_task(leader);
748 BUG_ON(atomic_read(&sig->count) != 1);
750 if (atomic_read(&oldsighand->count) == 1) {
752 * Now that we nuked the rest of the thread group,
753 * it turns out we are not sharing sighand any more either.
754 * So we can just keep it.
756 kmem_cache_free(sighand_cachep, newsighand);
757 } else {
759 * Move our state over to newsighand and switch it in.
761 atomic_set(&newsighand->count, 1);
762 memcpy(newsighand->action, oldsighand->action,
763 sizeof(newsighand->action));
765 write_lock_irq(&tasklist_lock);
766 spin_lock(&oldsighand->siglock);
767 spin_lock(&newsighand->siglock);
769 rcu_assign_pointer(current->sighand, newsighand);
770 recalc_sigpending();
772 spin_unlock(&newsighand->siglock);
773 spin_unlock(&oldsighand->siglock);
774 write_unlock_irq(&tasklist_lock);
776 if (atomic_dec_and_test(&oldsighand->count))
777 kmem_cache_free(sighand_cachep, oldsighand);
780 BUG_ON(!thread_group_leader(current));
781 return 0;
785 * These functions flushes out all traces of the currently running executable
786 * so that a new one can be started
789 static void flush_old_files(struct files_struct * files)
791 long j = -1;
792 struct fdtable *fdt;
794 spin_lock(&files->file_lock);
795 for (;;) {
796 unsigned long set, i;
798 j++;
799 i = j * __NFDBITS;
800 fdt = files_fdtable(files);
801 if (i >= fdt->max_fds || i >= fdt->max_fdset)
802 break;
803 set = fdt->close_on_exec->fds_bits[j];
804 if (!set)
805 continue;
806 fdt->close_on_exec->fds_bits[j] = 0;
807 spin_unlock(&files->file_lock);
808 for ( ; set ; i++,set >>= 1) {
809 if (set & 1) {
810 sys_close(i);
813 spin_lock(&files->file_lock);
816 spin_unlock(&files->file_lock);
819 void get_task_comm(char *buf, struct task_struct *tsk)
821 /* buf must be at least sizeof(tsk->comm) in size */
822 task_lock(tsk);
823 strncpy(buf, tsk->comm, sizeof(tsk->comm));
824 task_unlock(tsk);
827 void set_task_comm(struct task_struct *tsk, char *buf)
829 task_lock(tsk);
830 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
831 task_unlock(tsk);
834 int flush_old_exec(struct linux_binprm * bprm)
836 char * name;
837 int i, ch, retval;
838 struct files_struct *files;
839 char tcomm[sizeof(current->comm)];
842 * Make sure we have a private signal table and that
843 * we are unassociated from the previous thread group.
845 retval = de_thread(current);
846 if (retval)
847 goto out;
850 * Make sure we have private file handles. Ask the
851 * fork helper to do the work for us and the exit
852 * helper to do the cleanup of the old one.
854 files = current->files; /* refcounted so safe to hold */
855 retval = unshare_files();
856 if (retval)
857 goto out;
859 * Release all of the old mmap stuff
861 retval = exec_mmap(bprm->mm);
862 if (retval)
863 goto mmap_failed;
865 bprm->mm = NULL; /* We're using it now */
867 /* This is the point of no return */
868 steal_locks(files);
869 put_files_struct(files);
871 current->sas_ss_sp = current->sas_ss_size = 0;
873 if (current->euid == current->uid && current->egid == current->gid)
874 current->mm->dumpable = 1;
875 else
876 current->mm->dumpable = suid_dumpable;
878 name = bprm->filename;
880 /* Copies the binary name from after last slash */
881 for (i=0; (ch = *(name++)) != '\0';) {
882 if (ch == '/')
883 i = 0; /* overwrite what we wrote */
884 else
885 if (i < (sizeof(tcomm) - 1))
886 tcomm[i++] = ch;
888 tcomm[i] = '\0';
889 set_task_comm(current, tcomm);
891 current->flags &= ~PF_RANDOMIZE;
892 flush_thread();
894 /* Set the new mm task size. We have to do that late because it may
895 * depend on TIF_32BIT which is only updated in flush_thread() on
896 * some architectures like powerpc
898 current->mm->task_size = TASK_SIZE;
900 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
901 file_permission(bprm->file, MAY_READ) ||
902 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
903 suid_keys(current);
904 current->mm->dumpable = suid_dumpable;
907 /* An exec changes our domain. We are no longer part of the thread
908 group */
910 current->self_exec_id++;
912 flush_signal_handlers(current, 0);
913 flush_old_files(current->files);
915 return 0;
917 mmap_failed:
918 put_files_struct(current->files);
919 current->files = files;
920 out:
921 return retval;
924 EXPORT_SYMBOL(flush_old_exec);
927 * Fill the binprm structure from the inode.
928 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
930 int prepare_binprm(struct linux_binprm *bprm)
932 int mode;
933 struct inode * inode = bprm->file->f_dentry->d_inode;
934 int retval;
936 mode = inode->i_mode;
938 * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
939 * generic_permission lets a non-executable through
941 if (!(mode & 0111)) /* with at least _one_ execute bit set */
942 return -EACCES;
943 if (bprm->file->f_op == NULL)
944 return -EACCES;
946 bprm->e_uid = current->euid;
947 bprm->e_gid = current->egid;
949 if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
950 /* Set-uid? */
951 if (mode & S_ISUID) {
952 current->personality &= ~PER_CLEAR_ON_SETID;
953 bprm->e_uid = inode->i_uid;
956 /* Set-gid? */
958 * If setgid is set but no group execute bit then this
959 * is a candidate for mandatory locking, not a setgid
960 * executable.
962 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
963 current->personality &= ~PER_CLEAR_ON_SETID;
964 bprm->e_gid = inode->i_gid;
968 /* fill in binprm security blob */
969 retval = security_bprm_set(bprm);
970 if (retval)
971 return retval;
973 memset(bprm->buf,0,BINPRM_BUF_SIZE);
974 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
977 EXPORT_SYMBOL(prepare_binprm);
979 static int unsafe_exec(struct task_struct *p)
981 int unsafe = 0;
982 if (p->ptrace & PT_PTRACED) {
983 if (p->ptrace & PT_PTRACE_CAP)
984 unsafe |= LSM_UNSAFE_PTRACE_CAP;
985 else
986 unsafe |= LSM_UNSAFE_PTRACE;
988 if (atomic_read(&p->fs->count) > 1 ||
989 atomic_read(&p->files->count) > 1 ||
990 atomic_read(&p->sighand->count) > 1)
991 unsafe |= LSM_UNSAFE_SHARE;
993 return unsafe;
996 void compute_creds(struct linux_binprm *bprm)
998 int unsafe;
1000 if (bprm->e_uid != current->uid)
1001 suid_keys(current);
1002 exec_keys(current);
1004 task_lock(current);
1005 unsafe = unsafe_exec(current);
1006 security_bprm_apply_creds(bprm, unsafe);
1007 task_unlock(current);
1008 security_bprm_post_apply_creds(bprm);
1011 EXPORT_SYMBOL(compute_creds);
1013 void remove_arg_zero(struct linux_binprm *bprm)
1015 if (bprm->argc) {
1016 unsigned long offset;
1017 char * kaddr;
1018 struct page *page;
1020 offset = bprm->p % PAGE_SIZE;
1021 goto inside;
1023 while (bprm->p++, *(kaddr+offset++)) {
1024 if (offset != PAGE_SIZE)
1025 continue;
1026 offset = 0;
1027 kunmap_atomic(kaddr, KM_USER0);
1028 inside:
1029 page = bprm->page[bprm->p/PAGE_SIZE];
1030 kaddr = kmap_atomic(page, KM_USER0);
1032 kunmap_atomic(kaddr, KM_USER0);
1033 bprm->argc--;
1037 EXPORT_SYMBOL(remove_arg_zero);
1040 * cycle the list of binary formats handler, until one recognizes the image
1042 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1044 int try,retval;
1045 struct linux_binfmt *fmt;
1046 #ifdef __alpha__
1047 /* handle /sbin/loader.. */
1049 struct exec * eh = (struct exec *) bprm->buf;
1051 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1052 (eh->fh.f_flags & 0x3000) == 0x3000)
1054 struct file * file;
1055 unsigned long loader;
1057 allow_write_access(bprm->file);
1058 fput(bprm->file);
1059 bprm->file = NULL;
1061 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1063 file = open_exec("/sbin/loader");
1064 retval = PTR_ERR(file);
1065 if (IS_ERR(file))
1066 return retval;
1068 /* Remember if the application is TASO. */
1069 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1071 bprm->file = file;
1072 bprm->loader = loader;
1073 retval = prepare_binprm(bprm);
1074 if (retval<0)
1075 return retval;
1076 /* should call search_binary_handler recursively here,
1077 but it does not matter */
1080 #endif
1081 retval = security_bprm_check(bprm);
1082 if (retval)
1083 return retval;
1085 /* kernel module loader fixup */
1086 /* so we don't try to load run modprobe in kernel space. */
1087 set_fs(USER_DS);
1088 retval = -ENOENT;
1089 for (try=0; try<2; try++) {
1090 read_lock(&binfmt_lock);
1091 for (fmt = formats ; fmt ; fmt = fmt->next) {
1092 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1093 if (!fn)
1094 continue;
1095 if (!try_module_get(fmt->module))
1096 continue;
1097 read_unlock(&binfmt_lock);
1098 retval = fn(bprm, regs);
1099 if (retval >= 0) {
1100 put_binfmt(fmt);
1101 allow_write_access(bprm->file);
1102 if (bprm->file)
1103 fput(bprm->file);
1104 bprm->file = NULL;
1105 current->did_exec = 1;
1106 proc_exec_connector(current);
1107 return retval;
1109 read_lock(&binfmt_lock);
1110 put_binfmt(fmt);
1111 if (retval != -ENOEXEC || bprm->mm == NULL)
1112 break;
1113 if (!bprm->file) {
1114 read_unlock(&binfmt_lock);
1115 return retval;
1118 read_unlock(&binfmt_lock);
1119 if (retval != -ENOEXEC || bprm->mm == NULL) {
1120 break;
1121 #ifdef CONFIG_KMOD
1122 }else{
1123 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1124 if (printable(bprm->buf[0]) &&
1125 printable(bprm->buf[1]) &&
1126 printable(bprm->buf[2]) &&
1127 printable(bprm->buf[3]))
1128 break; /* -ENOEXEC */
1129 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1130 #endif
1133 return retval;
1136 EXPORT_SYMBOL(search_binary_handler);
1139 * sys_execve() executes a new program.
1141 int do_execve(char * filename,
1142 char __user *__user *argv,
1143 char __user *__user *envp,
1144 struct pt_regs * regs)
1146 struct linux_binprm *bprm;
1147 struct file *file;
1148 int retval;
1149 int i;
1151 retval = -ENOMEM;
1152 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1153 if (!bprm)
1154 goto out_ret;
1156 file = open_exec(filename);
1157 retval = PTR_ERR(file);
1158 if (IS_ERR(file))
1159 goto out_kfree;
1161 sched_exec();
1163 bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1165 bprm->file = file;
1166 bprm->filename = filename;
1167 bprm->interp = filename;
1168 bprm->mm = mm_alloc();
1169 retval = -ENOMEM;
1170 if (!bprm->mm)
1171 goto out_file;
1173 retval = init_new_context(current, bprm->mm);
1174 if (retval < 0)
1175 goto out_mm;
1177 bprm->argc = count(argv, bprm->p / sizeof(void *));
1178 if ((retval = bprm->argc) < 0)
1179 goto out_mm;
1181 bprm->envc = count(envp, bprm->p / sizeof(void *));
1182 if ((retval = bprm->envc) < 0)
1183 goto out_mm;
1185 retval = security_bprm_alloc(bprm);
1186 if (retval)
1187 goto out;
1189 retval = prepare_binprm(bprm);
1190 if (retval < 0)
1191 goto out;
1193 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1194 if (retval < 0)
1195 goto out;
1197 bprm->exec = bprm->p;
1198 retval = copy_strings(bprm->envc, envp, bprm);
1199 if (retval < 0)
1200 goto out;
1202 retval = copy_strings(bprm->argc, argv, bprm);
1203 if (retval < 0)
1204 goto out;
1206 retval = search_binary_handler(bprm,regs);
1207 if (retval >= 0) {
1208 free_arg_pages(bprm);
1210 /* execve success */
1211 security_bprm_free(bprm);
1212 acct_update_integrals(current);
1213 kfree(bprm);
1214 return retval;
1217 out:
1218 /* Something went wrong, return the inode and free the argument pages*/
1219 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1220 struct page * page = bprm->page[i];
1221 if (page)
1222 __free_page(page);
1225 if (bprm->security)
1226 security_bprm_free(bprm);
1228 out_mm:
1229 if (bprm->mm)
1230 mmdrop(bprm->mm);
1232 out_file:
1233 if (bprm->file) {
1234 allow_write_access(bprm->file);
1235 fput(bprm->file);
1238 out_kfree:
1239 kfree(bprm);
1241 out_ret:
1242 return retval;
1245 int set_binfmt(struct linux_binfmt *new)
1247 struct linux_binfmt *old = current->binfmt;
1249 if (new) {
1250 if (!try_module_get(new->module))
1251 return -1;
1253 current->binfmt = new;
1254 if (old)
1255 module_put(old->module);
1256 return 0;
1259 EXPORT_SYMBOL(set_binfmt);
1261 #define CORENAME_MAX_SIZE 64
1263 /* format_corename will inspect the pattern parameter, and output a
1264 * name into corename, which must have space for at least
1265 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1267 static void format_corename(char *corename, const char *pattern, long signr)
1269 const char *pat_ptr = pattern;
1270 char *out_ptr = corename;
1271 char *const out_end = corename + CORENAME_MAX_SIZE;
1272 int rc;
1273 int pid_in_pattern = 0;
1275 /* Repeat as long as we have more pattern to process and more output
1276 space */
1277 while (*pat_ptr) {
1278 if (*pat_ptr != '%') {
1279 if (out_ptr == out_end)
1280 goto out;
1281 *out_ptr++ = *pat_ptr++;
1282 } else {
1283 switch (*++pat_ptr) {
1284 case 0:
1285 goto out;
1286 /* Double percent, output one percent */
1287 case '%':
1288 if (out_ptr == out_end)
1289 goto out;
1290 *out_ptr++ = '%';
1291 break;
1292 /* pid */
1293 case 'p':
1294 pid_in_pattern = 1;
1295 rc = snprintf(out_ptr, out_end - out_ptr,
1296 "%d", current->tgid);
1297 if (rc > out_end - out_ptr)
1298 goto out;
1299 out_ptr += rc;
1300 break;
1301 /* uid */
1302 case 'u':
1303 rc = snprintf(out_ptr, out_end - out_ptr,
1304 "%d", current->uid);
1305 if (rc > out_end - out_ptr)
1306 goto out;
1307 out_ptr += rc;
1308 break;
1309 /* gid */
1310 case 'g':
1311 rc = snprintf(out_ptr, out_end - out_ptr,
1312 "%d", current->gid);
1313 if (rc > out_end - out_ptr)
1314 goto out;
1315 out_ptr += rc;
1316 break;
1317 /* signal that caused the coredump */
1318 case 's':
1319 rc = snprintf(out_ptr, out_end - out_ptr,
1320 "%ld", signr);
1321 if (rc > out_end - out_ptr)
1322 goto out;
1323 out_ptr += rc;
1324 break;
1325 /* UNIX time of coredump */
1326 case 't': {
1327 struct timeval tv;
1328 do_gettimeofday(&tv);
1329 rc = snprintf(out_ptr, out_end - out_ptr,
1330 "%lu", tv.tv_sec);
1331 if (rc > out_end - out_ptr)
1332 goto out;
1333 out_ptr += rc;
1334 break;
1336 /* hostname */
1337 case 'h':
1338 down_read(&uts_sem);
1339 rc = snprintf(out_ptr, out_end - out_ptr,
1340 "%s", system_utsname.nodename);
1341 up_read(&uts_sem);
1342 if (rc > out_end - out_ptr)
1343 goto out;
1344 out_ptr += rc;
1345 break;
1346 /* executable */
1347 case 'e':
1348 rc = snprintf(out_ptr, out_end - out_ptr,
1349 "%s", current->comm);
1350 if (rc > out_end - out_ptr)
1351 goto out;
1352 out_ptr += rc;
1353 break;
1354 default:
1355 break;
1357 ++pat_ptr;
1360 /* Backward compatibility with core_uses_pid:
1362 * If core_pattern does not include a %p (as is the default)
1363 * and core_uses_pid is set, then .%pid will be appended to
1364 * the filename */
1365 if (!pid_in_pattern
1366 && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1367 rc = snprintf(out_ptr, out_end - out_ptr,
1368 ".%d", current->tgid);
1369 if (rc > out_end - out_ptr)
1370 goto out;
1371 out_ptr += rc;
1373 out:
1374 *out_ptr = 0;
1377 static void zap_threads (struct mm_struct *mm)
1379 struct task_struct *g, *p;
1380 struct task_struct *tsk = current;
1381 struct completion *vfork_done = tsk->vfork_done;
1382 int traced = 0;
1385 * Make sure nobody is waiting for us to release the VM,
1386 * otherwise we can deadlock when we wait on each other
1388 if (vfork_done) {
1389 tsk->vfork_done = NULL;
1390 complete(vfork_done);
1393 read_lock(&tasklist_lock);
1394 do_each_thread(g,p)
1395 if (mm == p->mm && p != tsk) {
1396 force_sig_specific(SIGKILL, p);
1397 mm->core_waiters++;
1398 if (unlikely(p->ptrace) &&
1399 unlikely(p->parent->mm == mm))
1400 traced = 1;
1402 while_each_thread(g,p);
1404 read_unlock(&tasklist_lock);
1406 if (unlikely(traced)) {
1408 * We are zapping a thread and the thread it ptraces.
1409 * If the tracee went into a ptrace stop for exit tracing,
1410 * we could deadlock since the tracer is waiting for this
1411 * coredump to finish. Detach them so they can both die.
1413 write_lock_irq(&tasklist_lock);
1414 do_each_thread(g,p) {
1415 if (mm == p->mm && p != tsk &&
1416 p->ptrace && p->parent->mm == mm) {
1417 __ptrace_detach(p, 0);
1419 } while_each_thread(g,p);
1420 write_unlock_irq(&tasklist_lock);
1424 static void coredump_wait(struct mm_struct *mm)
1426 DECLARE_COMPLETION(startup_done);
1427 int core_waiters;
1429 mm->core_startup_done = &startup_done;
1431 zap_threads(mm);
1432 core_waiters = mm->core_waiters;
1433 up_write(&mm->mmap_sem);
1435 if (core_waiters)
1436 wait_for_completion(&startup_done);
1437 BUG_ON(mm->core_waiters);
1440 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1442 char corename[CORENAME_MAX_SIZE + 1];
1443 struct mm_struct *mm = current->mm;
1444 struct linux_binfmt * binfmt;
1445 struct inode * inode;
1446 struct file * file;
1447 int retval = 0;
1448 int fsuid = current->fsuid;
1449 int flag = 0;
1451 binfmt = current->binfmt;
1452 if (!binfmt || !binfmt->core_dump)
1453 goto fail;
1454 down_write(&mm->mmap_sem);
1455 if (!mm->dumpable) {
1456 up_write(&mm->mmap_sem);
1457 goto fail;
1461 * We cannot trust fsuid as being the "true" uid of the
1462 * process nor do we know its entire history. We only know it
1463 * was tainted so we dump it as root in mode 2.
1465 if (mm->dumpable == 2) { /* Setuid core dump mode */
1466 flag = O_EXCL; /* Stop rewrite attacks */
1467 current->fsuid = 0; /* Dump root private */
1469 mm->dumpable = 0;
1471 retval = -EAGAIN;
1472 spin_lock_irq(&current->sighand->siglock);
1473 if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
1474 current->signal->flags = SIGNAL_GROUP_EXIT;
1475 current->signal->group_exit_code = exit_code;
1476 current->signal->group_stop_count = 0;
1477 retval = 0;
1479 spin_unlock_irq(&current->sighand->siglock);
1480 if (retval) {
1481 up_write(&mm->mmap_sem);
1482 goto fail;
1485 init_completion(&mm->core_done);
1486 coredump_wait(mm);
1489 * Clear any false indication of pending signals that might
1490 * be seen by the filesystem code called to write the core file.
1492 clear_thread_flag(TIF_SIGPENDING);
1494 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1495 goto fail_unlock;
1498 * lock_kernel() because format_corename() is controlled by sysctl, which
1499 * uses lock_kernel()
1501 lock_kernel();
1502 format_corename(corename, core_pattern, signr);
1503 unlock_kernel();
1504 file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
1505 if (IS_ERR(file))
1506 goto fail_unlock;
1507 inode = file->f_dentry->d_inode;
1508 if (inode->i_nlink > 1)
1509 goto close_fail; /* multiple links - don't dump */
1510 if (d_unhashed(file->f_dentry))
1511 goto close_fail;
1513 if (!S_ISREG(inode->i_mode))
1514 goto close_fail;
1515 if (!file->f_op)
1516 goto close_fail;
1517 if (!file->f_op->write)
1518 goto close_fail;
1519 if (do_truncate(file->f_dentry, 0, 0, file) != 0)
1520 goto close_fail;
1522 retval = binfmt->core_dump(signr, regs, file);
1524 if (retval)
1525 current->signal->group_exit_code |= 0x80;
1526 close_fail:
1527 filp_close(file, NULL);
1528 fail_unlock:
1529 current->fsuid = fsuid;
1530 complete_all(&mm->core_done);
1531 fail:
1532 return retval;