projects/linux·blob v2.6.21 · refs

1553 lines · 35.1 KiB

1/*
2 * linux/fs/exec.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * #!-checking implemented by tytso.
9 */
10/*
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.
14 *
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.
17 *
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.
23 */
24
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>
53
54#include <asm/uaccess.h>
55#include <asm/mmu_context.h>
56
57#ifdef CONFIG_KMOD
58#include <linux/kmod.h>
59#endif
60
61int core_uses_pid;
62char core_pattern[128] = "core";
63int suid_dumpable = 0;
64
65EXPORT_SYMBOL(suid_dumpable);
66/* The maximal length of core_pattern is also specified in sysctl.c */
67
68static struct linux_binfmt *formats;
69static DEFINE_RWLOCK(binfmt_lock);
70
71int register_binfmt(struct linux_binfmt * fmt)
72{
73 struct linux_binfmt ** tmp = &formats;
74
75 if (!fmt)
76 return -EINVAL;
77 if (fmt->next)
78 return -EBUSY;
79 write_lock(&binfmt_lock);
80 while (*tmp) {
81 if (fmt == *tmp) {
82 write_unlock(&binfmt_lock);
83 return -EBUSY;
84 }
85 tmp = &(*tmp)->next;
86 }
87 fmt->next = formats;
88 formats = fmt;
89 write_unlock(&binfmt_lock);
90 return 0;
91}
92
93EXPORT_SYMBOL(register_binfmt);
94
95int unregister_binfmt(struct linux_binfmt * fmt)
96{
97 struct linux_binfmt ** tmp = &formats;
98
99 write_lock(&binfmt_lock);
100 while (*tmp) {
101 if (fmt == *tmp) {
102 *tmp = fmt->next;
103 write_unlock(&binfmt_lock);
104 return 0;
105 }
106 tmp = &(*tmp)->next;
107 }
108 write_unlock(&binfmt_lock);
109 return -EINVAL;
110}
111
112EXPORT_SYMBOL(unregister_binfmt);
113
114static inline void put_binfmt(struct linux_binfmt * fmt)
115{
116 module_put(fmt->module);
117}
118
119/*
120 * Note that a shared library must be both readable and executable due to
121 * security reasons.
122 *
123 * Also note that we take the address to load from from the file itself.
124 */
125asmlinkage long sys_uselib(const char __user * library)
126{
127 struct file * file;
128 struct nameidata nd;
129 int error;
130
131 error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
132 if (error)
133 goto out;
134
135 error = -EINVAL;
136 if (!S_ISREG(nd.dentry->d_inode->i_mode))
137 goto exit;
138
139 error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
140 if (error)
141 goto exit;
142
143 file = nameidata_to_filp(&nd, O_RDONLY);
144 error = PTR_ERR(file);
145 if (IS_ERR(file))
146 goto out;
147
148 error = -ENOEXEC;
149 if(file->f_op) {
150 struct linux_binfmt * fmt;
151
152 read_lock(&binfmt_lock);
153 for (fmt = formats ; fmt ; fmt = fmt->next) {
154 if (!fmt->load_shlib)
155 continue;
156 if (!try_module_get(fmt->module))
157 continue;
158 read_unlock(&binfmt_lock);
159 error = fmt->load_shlib(file);
160 read_lock(&binfmt_lock);
161 put_binfmt(fmt);
162 if (error != -ENOEXEC)
163 break;
164 }
165 read_unlock(&binfmt_lock);
166 }
167 fput(file);
168out:
169 return error;
170exit:
171 release_open_intent(&nd);
172 path_release(&nd);
173 goto out;
174}
175
176/*
177 * count() counts the number of strings in array ARGV.
178 */
179static int count(char __user * __user * argv, int max)
180{
181 int i = 0;
182
183 if (argv != NULL) {
184 for (;;) {
185 char __user * p;
186
187 if (get_user(p, argv))
188 return -EFAULT;
189 if (!p)
190 break;
191 argv++;
192 if(++i > max)
193 return -E2BIG;
194 cond_resched();
195 }
196 }
197 return i;
198}
199
200/*
201 * 'copy_strings()' copies argument/environment strings from user
202 * memory to free pages in kernel mem. These are in a format ready
203 * to be put directly into the top of new user memory.
204 */
205static int copy_strings(int argc, char __user * __user * argv,
206 struct linux_binprm *bprm)
207{
208 struct page *kmapped_page = NULL;
209 char *kaddr = NULL;
210 int ret;
211
212 while (argc-- > 0) {
213 char __user *str;
214 int len;
215 unsigned long pos;
216
217 if (get_user(str, argv+argc) ||
218 !(len = strnlen_user(str, bprm->p))) {
219 ret = -EFAULT;
220 goto out;
221 }
222
223 if (bprm->p < len) {
224 ret = -E2BIG;
225 goto out;
226 }
227
228 bprm->p -= len;
229 /* XXX: add architecture specific overflow check here. */
230 pos = bprm->p;
231
232 while (len > 0) {
233 int i, new, err;
234 int offset, bytes_to_copy;
235 struct page *page;
236
237 offset = pos % PAGE_SIZE;
238 i = pos/PAGE_SIZE;
239 page = bprm->page[i];
240 new = 0;
241 if (!page) {
242 page = alloc_page(GFP_HIGHUSER);
243 bprm->page[i] = page;
244 if (!page) {
245 ret = -ENOMEM;
246 goto out;
247 }
248 new = 1;
249 }
250
251 if (page != kmapped_page) {
252 if (kmapped_page)
253 kunmap(kmapped_page);
254 kmapped_page = page;
255 kaddr = kmap(kmapped_page);
256 }
257 if (new && offset)
258 memset(kaddr, 0, offset);
259 bytes_to_copy = PAGE_SIZE - offset;
260 if (bytes_to_copy > len) {
261 bytes_to_copy = len;
262 if (new)
263 memset(kaddr+offset+len, 0,
264 PAGE_SIZE-offset-len);
265 }
266 err = copy_from_user(kaddr+offset, str, bytes_to_copy);
267 if (err) {
268 ret = -EFAULT;
269 goto out;
270 }
271
272 pos += bytes_to_copy;
273 str += bytes_to_copy;
274 len -= bytes_to_copy;
275 }
276 }
277 ret = 0;
278out:
279 if (kmapped_page)
280 kunmap(kmapped_page);
281 return ret;
282}
283
284/*
285 * Like copy_strings, but get argv and its values from kernel memory.
286 */
287int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
288{
289 int r;
290 mm_segment_t oldfs = get_fs();
291 set_fs(KERNEL_DS);
292 r = copy_strings(argc, (char __user * __user *)argv, bprm);
293 set_fs(oldfs);
294 return r;
295}
296
297EXPORT_SYMBOL(copy_strings_kernel);
298
299#ifdef CONFIG_MMU
300/*
301 * This routine is used to map in a page into an address space: needed by
302 * execve() for the initial stack and environment pages.
303 *
304 * vma->vm_mm->mmap_sem is held for writing.
305 */
306void install_arg_page(struct vm_area_struct *vma,
307 struct page *page, unsigned long address)
308{
309 struct mm_struct *mm = vma->vm_mm;
310 pte_t * pte;
311 spinlock_t *ptl;
312
313 if (unlikely(anon_vma_prepare(vma)))
314 goto out;
315
316 flush_dcache_page(page);
317 pte = get_locked_pte(mm, address, &ptl);
318 if (!pte)
319 goto out;
320 if (!pte_none(*pte)) {
321 pte_unmap_unlock(pte, ptl);
322 goto out;
323 }
324 inc_mm_counter(mm, anon_rss);
325 lru_cache_add_active(page);
326 set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
327 page, vma->vm_page_prot))));
328 page_add_new_anon_rmap(page, vma, address);
329 pte_unmap_unlock(pte, ptl);
330
331 /* no need for flush_tlb */
332 return;
333out:
334 __free_page(page);
335 force_sig(SIGKILL, current);
336}
337
338#define EXTRA_STACK_VM_PAGES 20 /* random */
339
340int setup_arg_pages(struct linux_binprm *bprm,
341 unsigned long stack_top,
342 int executable_stack)
343{
344 unsigned long stack_base;
345 struct vm_area_struct *mpnt;
346 struct mm_struct *mm = current->mm;
347 int i, ret;
348 long arg_size;
349
350#ifdef CONFIG_STACK_GROWSUP
351 /* Move the argument and environment strings to the bottom of the
352 * stack space.
353 */
354 int offset, j;
355 char *to, *from;
356
357 /* Start by shifting all the pages down */
358 i = 0;
359 for (j = 0; j < MAX_ARG_PAGES; j++) {
360 struct page *page = bprm->page[j];
361 if (!page)
362 continue;
363 bprm->page[i++] = page;
364 }
365
366 /* Now move them within their pages */
367 offset = bprm->p % PAGE_SIZE;
368 to = kmap(bprm->page[0]);
369 for (j = 1; j < i; j++) {
370 memmove(to, to + offset, PAGE_SIZE - offset);
371 from = kmap(bprm->page[j]);
372 memcpy(to + PAGE_SIZE - offset, from, offset);
373 kunmap(bprm->page[j - 1]);
374 to = from;
375 }
376 memmove(to, to + offset, PAGE_SIZE - offset);
377 kunmap(bprm->page[j - 1]);
378
379 /* Limit stack size to 1GB */
380 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
381 if (stack_base > (1 << 30))
382 stack_base = 1 << 30;
383 stack_base = PAGE_ALIGN(stack_top - stack_base);
384
385 /* Adjust bprm->p to point to the end of the strings. */
386 bprm->p = stack_base + PAGE_SIZE * i - offset;
387
388 mm->arg_start = stack_base;
389 arg_size = i << PAGE_SHIFT;
390
391 /* zero pages that were copied above */
392 while (i < MAX_ARG_PAGES)
393 bprm->page[i++] = NULL;
394#else
395 stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
396 stack_base = PAGE_ALIGN(stack_base);
397 bprm->p += stack_base;
398 mm->arg_start = bprm->p;
399 arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
400#endif
401
402 arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
403
404 if (bprm->loader)
405 bprm->loader += stack_base;
406 bprm->exec += stack_base;
407
408 mpnt = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
409 if (!mpnt)
410 return -ENOMEM;
411
412 down_write(&mm->mmap_sem);
413 {
414 mpnt->vm_mm = mm;
415#ifdef CONFIG_STACK_GROWSUP
416 mpnt->vm_start = stack_base;
417 mpnt->vm_end = stack_base + arg_size;
418#else
419 mpnt->vm_end = stack_top;
420 mpnt->vm_start = mpnt->vm_end - arg_size;
421#endif
422 /* Adjust stack execute permissions; explicitly enable
423 * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
424 * and leave alone (arch default) otherwise. */
425 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
426 mpnt->vm_flags = VM_STACK_FLAGS | VM_EXEC;
427 else if (executable_stack == EXSTACK_DISABLE_X)
428 mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
429 else
430 mpnt->vm_flags = VM_STACK_FLAGS;
431 mpnt->vm_flags |= mm->def_flags;
432 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
433 if ((ret = insert_vm_struct(mm, mpnt))) {
434 up_write(&mm->mmap_sem);
435 kmem_cache_free(vm_area_cachep, mpnt);
436 return ret;
437 }
438 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
439 }
440
441 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
442 struct page *page = bprm->page[i];
443 if (page) {
444 bprm->page[i] = NULL;
445 install_arg_page(mpnt, page, stack_base);
446 }
447 stack_base += PAGE_SIZE;
448 }
449 up_write(&mm->mmap_sem);
450
451 return 0;
452}
453
454EXPORT_SYMBOL(setup_arg_pages);
455
456#define free_arg_pages(bprm) do { } while (0)
457
458#else
459
460static inline void free_arg_pages(struct linux_binprm *bprm)
461{
462 int i;
463
464 for (i = 0; i < MAX_ARG_PAGES; i++) {
465 if (bprm->page[i])
466 __free_page(bprm->page[i]);
467 bprm->page[i] = NULL;
468 }
469}
470
471#endif /* CONFIG_MMU */
472
473struct file *open_exec(const char *name)
474{
475 struct nameidata nd;
476 int err;
477 struct file *file;
478
479 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
480 file = ERR_PTR(err);
481
482 if (!err) {
483 struct inode *inode = nd.dentry->d_inode;
484 file = ERR_PTR(-EACCES);
485 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
486 S_ISREG(inode->i_mode)) {
487 int err = vfs_permission(&nd, MAY_EXEC);
488 file = ERR_PTR(err);
489 if (!err) {
490 file = nameidata_to_filp(&nd, O_RDONLY);
491 if (!IS_ERR(file)) {
492 err = deny_write_access(file);
493 if (err) {
494 fput(file);
495 file = ERR_PTR(err);
496 }
497 }
498out:
499 return file;
500 }
501 }
502 release_open_intent(&nd);
503 path_release(&nd);
504 }
505 goto out;
506}
507
508EXPORT_SYMBOL(open_exec);
509
510int kernel_read(struct file *file, unsigned long offset,
511 char *addr, unsigned long count)
512{
513 mm_segment_t old_fs;
514 loff_t pos = offset;
515 int result;
516
517 old_fs = get_fs();
518 set_fs(get_ds());
519 /* The cast to a user pointer is valid due to the set_fs() */
520 result = vfs_read(file, (void __user *)addr, count, &pos);
521 set_fs(old_fs);
522 return result;
523}
524
525EXPORT_SYMBOL(kernel_read);
526
527static int exec_mmap(struct mm_struct *mm)
528{
529 struct task_struct *tsk;
530 struct mm_struct * old_mm, *active_mm;
531
532 /* Notify parent that we're no longer interested in the old VM */
533 tsk = current;
534 old_mm = current->mm;
535 mm_release(tsk, old_mm);
536
537 if (old_mm) {
538 /*
539 * Make sure that if there is a core dump in progress
540 * for the old mm, we get out and die instead of going
541 * through with the exec. We must hold mmap_sem around
542 * checking core_waiters and changing tsk->mm. The
543 * core-inducing thread will increment core_waiters for
544 * each thread whose ->mm == old_mm.
545 */
546 down_read(&old_mm->mmap_sem);
547 if (unlikely(old_mm->core_waiters)) {
548 up_read(&old_mm->mmap_sem);
549 return -EINTR;
550 }
551 }
552 task_lock(tsk);
553 active_mm = tsk->active_mm;
554 tsk->mm = mm;
555 tsk->active_mm = mm;
556 activate_mm(active_mm, mm);
557 task_unlock(tsk);
558 arch_pick_mmap_layout(mm);
559 if (old_mm) {
560 up_read(&old_mm->mmap_sem);
561 BUG_ON(active_mm != old_mm);
562 mmput(old_mm);
563 return 0;
564 }
565 mmdrop(active_mm);
566 return 0;
567}
568
569/*
570 * This function makes sure the current process has its own signal table,
571 * so that flush_signal_handlers can later reset the handlers without
572 * disturbing other processes. (Other processes might share the signal
573 * table via the CLONE_SIGHAND option to clone().)
574 */
575static int de_thread(struct task_struct *tsk)
576{
577 struct signal_struct *sig = tsk->signal;
578 struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
579 spinlock_t *lock = &oldsighand->siglock;
580 struct task_struct *leader = NULL;
581 int count;
582
583 /*
584 * If we don't share sighandlers, then we aren't sharing anything
585 * and we can just re-use it all.
586 */
587 if (atomic_read(&oldsighand->count) <= 1) {
588 BUG_ON(atomic_read(&sig->count) != 1);
589 exit_itimers(sig);
590 return 0;
591 }
592
593 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
594 if (!newsighand)
595 return -ENOMEM;
596
597 if (thread_group_empty(tsk))
598 goto no_thread_group;
599
600 /*
601 * Kill all other threads in the thread group.
602 * We must hold tasklist_lock to call zap_other_threads.
603 */
604 read_lock(&tasklist_lock);
605 spin_lock_irq(lock);
606 if (sig->flags & SIGNAL_GROUP_EXIT) {
607 /*
608 * Another group action in progress, just
609 * return so that the signal is processed.
610 */
611 spin_unlock_irq(lock);
612 read_unlock(&tasklist_lock);
613 kmem_cache_free(sighand_cachep, newsighand);
614 return -EAGAIN;
615 }
616
617 /*
618 * child_reaper ignores SIGKILL, change it now.
619 * Reparenting needs write_lock on tasklist_lock,
620 * so it is safe to do it under read_lock.
621 */
622 if (unlikely(tsk->group_leader == child_reaper(tsk)))
623 tsk->nsproxy->pid_ns->child_reaper = tsk;
624
625 zap_other_threads(tsk);
626 read_unlock(&tasklist_lock);
627
628 /*
629 * Account for the thread group leader hanging around:
630 */
631 count = 1;
632 if (!thread_group_leader(tsk)) {
633 count = 2;
634 /*
635 * The SIGALRM timer survives the exec, but needs to point
636 * at us as the new group leader now. We have a race with
637 * a timer firing now getting the old leader, so we need to
638 * synchronize with any firing (by calling del_timer_sync)
639 * before we can safely let the old group leader die.
640 */
641 sig->tsk = tsk;
642 spin_unlock_irq(lock);
643 if (hrtimer_cancel(&sig->real_timer))
644 hrtimer_restart(&sig->real_timer);
645 spin_lock_irq(lock);
646 }
647 while (atomic_read(&sig->count) > count) {
648 sig->group_exit_task = tsk;
649 sig->notify_count = count;
650 __set_current_state(TASK_UNINTERRUPTIBLE);
651 spin_unlock_irq(lock);
652 schedule();
653 spin_lock_irq(lock);
654 }
655 sig->group_exit_task = NULL;
656 sig->notify_count = 0;
657 spin_unlock_irq(lock);
658
659 /*
660 * At this point all other threads have exited, all we have to
661 * do is to wait for the thread group leader to become inactive,
662 * and to assume its PID:
663 */
664 if (!thread_group_leader(tsk)) {
665 /*
666 * Wait for the thread group leader to be a zombie.
667 * It should already be zombie at this point, most
668 * of the time.
669 */
670 leader = tsk->group_leader;
671 while (leader->exit_state != EXIT_ZOMBIE)
672 yield();
673
674 /*
675 * The only record we have of the real-time age of a
676 * process, regardless of execs it's done, is start_time.
677 * All the past CPU time is accumulated in signal_struct
678 * from sister threads now dead. But in this non-leader
679 * exec, nothing survives from the original leader thread,
680 * whose birth marks the true age of this process now.
681 * When we take on its identity by switching to its PID, we
682 * also take its birthdate (always earlier than our own).
683 */
684 tsk->start_time = leader->start_time;
685
686 write_lock_irq(&tasklist_lock);
687
688 BUG_ON(leader->tgid != tsk->tgid);
689 BUG_ON(tsk->pid == tsk->tgid);
690 /*
691 * An exec() starts a new thread group with the
692 * TGID of the previous thread group. Rehash the
693 * two threads with a switched PID, and release
694 * the former thread group leader:
695 */
696
697 /* Become a process group leader with the old leader's pid.
698 * The old leader becomes a thread of the this thread group.
699 * Note: The old leader also uses this pid until release_task
700 * is called. Odd but simple and correct.
701 */
702 detach_pid(tsk, PIDTYPE_PID);
703 tsk->pid = leader->pid;
704 attach_pid(tsk, PIDTYPE_PID, tsk->pid);
705 transfer_pid(leader, tsk, PIDTYPE_PGID);
706 transfer_pid(leader, tsk, PIDTYPE_SID);
707 list_replace_rcu(&leader->tasks, &tsk->tasks);
708
709 tsk->group_leader = tsk;
710 leader->group_leader = tsk;
711
712 tsk->exit_signal = SIGCHLD;
713
714 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
715 leader->exit_state = EXIT_DEAD;
716
717 write_unlock_irq(&tasklist_lock);
718 }
719
720 /*
721 * There may be one thread left which is just exiting,
722 * but it's safe to stop telling the group to kill themselves.
723 */
724 sig->flags = 0;
725
726no_thread_group:
727 exit_itimers(sig);
728 if (leader)
729 release_task(leader);
730
731 BUG_ON(atomic_read(&sig->count) != 1);
732
733 if (atomic_read(&oldsighand->count) == 1) {
734 /*
735 * Now that we nuked the rest of the thread group,
736 * it turns out we are not sharing sighand any more either.
737 * So we can just keep it.
738 */
739 kmem_cache_free(sighand_cachep, newsighand);
740 } else {
741 /*
742 * Move our state over to newsighand and switch it in.
743 */
744 atomic_set(&newsighand->count, 1);
745 memcpy(newsighand->action, oldsighand->action,
746 sizeof(newsighand->action));
747
748 write_lock_irq(&tasklist_lock);
749 spin_lock(&oldsighand->siglock);
750 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
751
752 rcu_assign_pointer(tsk->sighand, newsighand);
753 recalc_sigpending();
754
755 spin_unlock(&newsighand->siglock);
756 spin_unlock(&oldsighand->siglock);
757 write_unlock_irq(&tasklist_lock);
758
759 if (atomic_dec_and_test(&oldsighand->count))
760 kmem_cache_free(sighand_cachep, oldsighand);
761 }
762
763 BUG_ON(!thread_group_leader(tsk));
764 return 0;
765}
766
767/*
768 * These functions flushes out all traces of the currently running executable
769 * so that a new one can be started
770 */
771
772static void flush_old_files(struct files_struct * files)
773{
774 long j = -1;
775 struct fdtable *fdt;
776
777 spin_lock(&files->file_lock);
778 for (;;) {
779 unsigned long set, i;
780
781 j++;
782 i = j * __NFDBITS;
783 fdt = files_fdtable(files);
784 if (i >= fdt->max_fds)
785 break;
786 set = fdt->close_on_exec->fds_bits[j];
787 if (!set)
788 continue;
789 fdt->close_on_exec->fds_bits[j] = 0;
790 spin_unlock(&files->file_lock);
791 for ( ; set ; i++,set >>= 1) {
792 if (set & 1) {
793 sys_close(i);
794 }
795 }
796 spin_lock(&files->file_lock);
797
798 }
799 spin_unlock(&files->file_lock);
800}
801
802void get_task_comm(char *buf, struct task_struct *tsk)
803{
804 /* buf must be at least sizeof(tsk->comm) in size */
805 task_lock(tsk);
806 strncpy(buf, tsk->comm, sizeof(tsk->comm));
807 task_unlock(tsk);
808}
809
810void set_task_comm(struct task_struct *tsk, char *buf)
811{
812 task_lock(tsk);
813 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
814 task_unlock(tsk);
815}
816
817int flush_old_exec(struct linux_binprm * bprm)
818{
819 char * name;
820 int i, ch, retval;
821 struct files_struct *files;
822 char tcomm[sizeof(current->comm)];
823
824 /*
825 * Make sure we have a private signal table and that
826 * we are unassociated from the previous thread group.
827 */
828 retval = de_thread(current);
829 if (retval)
830 goto out;
831
832 /*
833 * Make sure we have private file handles. Ask the
834 * fork helper to do the work for us and the exit
835 * helper to do the cleanup of the old one.
836 */
837 files = current->files; /* refcounted so safe to hold */
838 retval = unshare_files();
839 if (retval)
840 goto out;
841 /*
842 * Release all of the old mmap stuff
843 */
844 retval = exec_mmap(bprm->mm);
845 if (retval)
846 goto mmap_failed;
847
848 bprm->mm = NULL; /* We're using it now */
849
850 /* This is the point of no return */
851 put_files_struct(files);
852
853 current->sas_ss_sp = current->sas_ss_size = 0;
854
855 if (current->euid == current->uid && current->egid == current->gid)
856 current->mm->dumpable = 1;
857 else
858 current->mm->dumpable = suid_dumpable;
859
860 name = bprm->filename;
861
862 /* Copies the binary name from after last slash */
863 for (i=0; (ch = *(name++)) != '\0';) {
864 if (ch == '/')
865 i = 0; /* overwrite what we wrote */
866 else
867 if (i < (sizeof(tcomm) - 1))
868 tcomm[i++] = ch;
869 }
870 tcomm[i] = '\0';
871 set_task_comm(current, tcomm);
872
873 current->flags &= ~PF_RANDOMIZE;
874 flush_thread();
875
876 /* Set the new mm task size. We have to do that late because it may
877 * depend on TIF_32BIT which is only updated in flush_thread() on
878 * some architectures like powerpc
879 */
880 current->mm->task_size = TASK_SIZE;
881
882 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
883 file_permission(bprm->file, MAY_READ) ||
884 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
885 suid_keys(current);
886 current->mm->dumpable = suid_dumpable;
887 }
888
889 /* An exec changes our domain. We are no longer part of the thread
890 group */
891
892 current->self_exec_id++;
893
894 flush_signal_handlers(current, 0);
895 flush_old_files(current->files);
896
897 return 0;
898
899mmap_failed:
900 reset_files_struct(current, files);
901out:
902 return retval;
903}
904
905EXPORT_SYMBOL(flush_old_exec);
906
907/*
908 * Fill the binprm structure from the inode.
909 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
910 */
911int prepare_binprm(struct linux_binprm *bprm)
912{
913 int mode;
914 struct inode * inode = bprm->file->f_path.dentry->d_inode;
915 int retval;
916
917 mode = inode->i_mode;
918 if (bprm->file->f_op == NULL)
919 return -EACCES;
920
921 bprm->e_uid = current->euid;
922 bprm->e_gid = current->egid;
923
924 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
925 /* Set-uid? */
926 if (mode & S_ISUID) {
927 current->personality &= ~PER_CLEAR_ON_SETID;
928 bprm->e_uid = inode->i_uid;
929 }
930
931 /* Set-gid? */
932 /*
933 * If setgid is set but no group execute bit then this
934 * is a candidate for mandatory locking, not a setgid
935 * executable.
936 */
937 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
938 current->personality &= ~PER_CLEAR_ON_SETID;
939 bprm->e_gid = inode->i_gid;
940 }
941 }
942
943 /* fill in binprm security blob */
944 retval = security_bprm_set(bprm);
945 if (retval)
946 return retval;
947
948 memset(bprm->buf,0,BINPRM_BUF_SIZE);
949 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
950}
951
952EXPORT_SYMBOL(prepare_binprm);
953
954static int unsafe_exec(struct task_struct *p)
955{
956 int unsafe = 0;
957 if (p->ptrace & PT_PTRACED) {
958 if (p->ptrace & PT_PTRACE_CAP)
959 unsafe |= LSM_UNSAFE_PTRACE_CAP;
960 else
961 unsafe |= LSM_UNSAFE_PTRACE;
962 }
963 if (atomic_read(&p->fs->count) > 1 ||
964 atomic_read(&p->files->count) > 1 ||
965 atomic_read(&p->sighand->count) > 1)
966 unsafe |= LSM_UNSAFE_SHARE;
967
968 return unsafe;
969}
970
971void compute_creds(struct linux_binprm *bprm)
972{
973 int unsafe;
974
975 if (bprm->e_uid != current->uid)
976 suid_keys(current);
977 exec_keys(current);
978
979 task_lock(current);
980 unsafe = unsafe_exec(current);
981 security_bprm_apply_creds(bprm, unsafe);
982 task_unlock(current);
983 security_bprm_post_apply_creds(bprm);
984}
985
986EXPORT_SYMBOL(compute_creds);
987
988void remove_arg_zero(struct linux_binprm *bprm)
989{
990 if (bprm->argc) {
991 unsigned long offset;
992 char * kaddr;
993 struct page *page;
994
995 offset = bprm->p % PAGE_SIZE;
996 goto inside;
997
998 while (bprm->p++, *(kaddr+offset++)) {
999 if (offset != PAGE_SIZE)
1000 continue;
1001 offset = 0;
1002 kunmap_atomic(kaddr, KM_USER0);
1003inside:
1004 page = bprm->page[bprm->p/PAGE_SIZE];
1005 kaddr = kmap_atomic(page, KM_USER0);
1006 }
1007 kunmap_atomic(kaddr, KM_USER0);
1008 bprm->argc--;
1009 }
1010}
1011
1012EXPORT_SYMBOL(remove_arg_zero);
1013
1014/*
1015 * cycle the list of binary formats handler, until one recognizes the image
1016 */
1017int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1018{
1019 int try,retval;
1020 struct linux_binfmt *fmt;
1021#ifdef __alpha__
1022 /* handle /sbin/loader.. */
1023 {
1024 struct exec * eh = (struct exec *) bprm->buf;
1025
1026 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1027 (eh->fh.f_flags & 0x3000) == 0x3000)
1028 {
1029 struct file * file;
1030 unsigned long loader;
1031
1032 allow_write_access(bprm->file);
1033 fput(bprm->file);
1034 bprm->file = NULL;
1035
1036 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1037
1038 file = open_exec("/sbin/loader");
1039 retval = PTR_ERR(file);
1040 if (IS_ERR(file))
1041 return retval;
1042
1043 /* Remember if the application is TASO. */
1044 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1045
1046 bprm->file = file;
1047 bprm->loader = loader;
1048 retval = prepare_binprm(bprm);
1049 if (retval<0)
1050 return retval;
1051 /* should call search_binary_handler recursively here,
1052 but it does not matter */
1053 }
1054 }
1055#endif
1056 retval = security_bprm_check(bprm);
1057 if (retval)
1058 return retval;
1059
1060 /* kernel module loader fixup */
1061 /* so we don't try to load run modprobe in kernel space. */
1062 set_fs(USER_DS);
1063
1064 retval = audit_bprm(bprm);
1065 if (retval)
1066 return retval;
1067
1068 retval = -ENOENT;
1069 for (try=0; try<2; try++) {
1070 read_lock(&binfmt_lock);
1071 for (fmt = formats ; fmt ; fmt = fmt->next) {
1072 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1073 if (!fn)
1074 continue;
1075 if (!try_module_get(fmt->module))
1076 continue;
1077 read_unlock(&binfmt_lock);
1078 retval = fn(bprm, regs);
1079 if (retval >= 0) {
1080 put_binfmt(fmt);
1081 allow_write_access(bprm->file);
1082 if (bprm->file)
1083 fput(bprm->file);
1084 bprm->file = NULL;
1085 current->did_exec = 1;
1086 proc_exec_connector(current);
1087 return retval;
1088 }
1089 read_lock(&binfmt_lock);
1090 put_binfmt(fmt);
1091 if (retval != -ENOEXEC || bprm->mm == NULL)
1092 break;
1093 if (!bprm->file) {
1094 read_unlock(&binfmt_lock);
1095 return retval;
1096 }
1097 }
1098 read_unlock(&binfmt_lock);
1099 if (retval != -ENOEXEC || bprm->mm == NULL) {
1100 break;
1101#ifdef CONFIG_KMOD
1102 }else{
1103#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1104 if (printable(bprm->buf[0]) &&
1105 printable(bprm->buf[1]) &&
1106 printable(bprm->buf[2]) &&
1107 printable(bprm->buf[3]))
1108 break; /* -ENOEXEC */
1109 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1110#endif
1111 }
1112 }
1113 return retval;
1114}
1115
1116EXPORT_SYMBOL(search_binary_handler);
1117
1118/*
1119 * sys_execve() executes a new program.
1120 */
1121int do_execve(char * filename,
1122 char __user *__user *argv,
1123 char __user *__user *envp,
1124 struct pt_regs * regs)
1125{
1126 struct linux_binprm *bprm;
1127 struct file *file;
1128 int retval;
1129 int i;
1130
1131 retval = -ENOMEM;
1132 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1133 if (!bprm)
1134 goto out_ret;
1135
1136 file = open_exec(filename);
1137 retval = PTR_ERR(file);
1138 if (IS_ERR(file))
1139 goto out_kfree;
1140
1141 sched_exec();
1142
1143 bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1144
1145 bprm->file = file;
1146 bprm->filename = filename;
1147 bprm->interp = filename;
1148 bprm->mm = mm_alloc();
1149 retval = -ENOMEM;
1150 if (!bprm->mm)
1151 goto out_file;
1152
1153 retval = init_new_context(current, bprm->mm);
1154 if (retval < 0)
1155 goto out_mm;
1156
1157 bprm->argc = count(argv, bprm->p / sizeof(void *));
1158 if ((retval = bprm->argc) < 0)
1159 goto out_mm;
1160
1161 bprm->envc = count(envp, bprm->p / sizeof(void *));
1162 if ((retval = bprm->envc) < 0)
1163 goto out_mm;
1164
1165 retval = security_bprm_alloc(bprm);
1166 if (retval)
1167 goto out;
1168
1169 retval = prepare_binprm(bprm);
1170 if (retval < 0)
1171 goto out;
1172
1173 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1174 if (retval < 0)
1175 goto out;
1176
1177 bprm->exec = bprm->p;
1178 retval = copy_strings(bprm->envc, envp, bprm);
1179 if (retval < 0)
1180 goto out;
1181
1182 retval = copy_strings(bprm->argc, argv, bprm);
1183 if (retval < 0)
1184 goto out;
1185
1186 retval = search_binary_handler(bprm,regs);
1187 if (retval >= 0) {
1188 free_arg_pages(bprm);
1189
1190 /* execve success */
1191 security_bprm_free(bprm);
1192 acct_update_integrals(current);
1193 kfree(bprm);
1194 return retval;
1195 }
1196
1197out:
1198 /* Something went wrong, return the inode and free the argument pages*/
1199 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1200 struct page * page = bprm->page[i];
1201 if (page)
1202 __free_page(page);
1203 }
1204
1205 if (bprm->security)
1206 security_bprm_free(bprm);
1207
1208out_mm:
1209 if (bprm->mm)
1210 mmdrop(bprm->mm);
1211
1212out_file:
1213 if (bprm->file) {
1214 allow_write_access(bprm->file);
1215 fput(bprm->file);
1216 }
1217
1218out_kfree:
1219 kfree(bprm);
1220
1221out_ret:
1222 return retval;
1223}
1224
1225int set_binfmt(struct linux_binfmt *new)
1226{
1227 struct linux_binfmt *old = current->binfmt;
1228
1229 if (new) {
1230 if (!try_module_get(new->module))
1231 return -1;
1232 }
1233 current->binfmt = new;
1234 if (old)
1235 module_put(old->module);
1236 return 0;
1237}
1238
1239EXPORT_SYMBOL(set_binfmt);
1240
1241#define CORENAME_MAX_SIZE 64
1242
1243/* format_corename will inspect the pattern parameter, and output a
1244 * name into corename, which must have space for at least
1245 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1246 */
1247static int format_corename(char *corename, const char *pattern, long signr)
1248{
1249 const char *pat_ptr = pattern;
1250 char *out_ptr = corename;
1251 char *const out_end = corename + CORENAME_MAX_SIZE;
1252 int rc;
1253 int pid_in_pattern = 0;
1254 int ispipe = 0;
1255
1256 if (*pattern == '|')
1257 ispipe = 1;
1258
1259 /* Repeat as long as we have more pattern to process and more output
1260 space */
1261 while (*pat_ptr) {
1262 if (*pat_ptr != '%') {
1263 if (out_ptr == out_end)
1264 goto out;
1265 *out_ptr++ = *pat_ptr++;
1266 } else {
1267 switch (*++pat_ptr) {
1268 case 0:
1269 goto out;
1270 /* Double percent, output one percent */
1271 case '%':
1272 if (out_ptr == out_end)
1273 goto out;
1274 *out_ptr++ = '%';
1275 break;
1276 /* pid */
1277 case 'p':
1278 pid_in_pattern = 1;
1279 rc = snprintf(out_ptr, out_end - out_ptr,
1280 "%d", current->tgid);
1281 if (rc > out_end - out_ptr)
1282 goto out;
1283 out_ptr += rc;
1284 break;
1285 /* uid */
1286 case 'u':
1287 rc = snprintf(out_ptr, out_end - out_ptr,
1288 "%d", current->uid);
1289 if (rc > out_end - out_ptr)
1290 goto out;
1291 out_ptr += rc;
1292 break;
1293 /* gid */
1294 case 'g':
1295 rc = snprintf(out_ptr, out_end - out_ptr,
1296 "%d", current->gid);
1297 if (rc > out_end - out_ptr)
1298 goto out;
1299 out_ptr += rc;
1300 break;
1301 /* signal that caused the coredump */
1302 case 's':
1303 rc = snprintf(out_ptr, out_end - out_ptr,
1304 "%ld", signr);
1305 if (rc > out_end - out_ptr)
1306 goto out;
1307 out_ptr += rc;
1308 break;
1309 /* UNIX time of coredump */
1310 case 't': {
1311 struct timeval tv;
1312 do_gettimeofday(&tv);
1313 rc = snprintf(out_ptr, out_end - out_ptr,
1314 "%lu", tv.tv_sec);
1315 if (rc > out_end - out_ptr)
1316 goto out;
1317 out_ptr += rc;
1318 break;
1319 }
1320 /* hostname */
1321 case 'h':
1322 down_read(&uts_sem);
1323 rc = snprintf(out_ptr, out_end - out_ptr,
1324 "%s", utsname()->nodename);
1325 up_read(&uts_sem);
1326 if (rc > out_end - out_ptr)
1327 goto out;
1328 out_ptr += rc;
1329 break;
1330 /* executable */
1331 case 'e':
1332 rc = snprintf(out_ptr, out_end - out_ptr,
1333 "%s", current->comm);
1334 if (rc > out_end - out_ptr)
1335 goto out;
1336 out_ptr += rc;
1337 break;
1338 default:
1339 break;
1340 }
1341 ++pat_ptr;
1342 }
1343 }
1344 /* Backward compatibility with core_uses_pid:
1345 *
1346 * If core_pattern does not include a %p (as is the default)
1347 * and core_uses_pid is set, then .%pid will be appended to
1348 * the filename. Do not do this for piped commands. */
1349 if (!ispipe && !pid_in_pattern
1350 && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1351 rc = snprintf(out_ptr, out_end - out_ptr,
1352 ".%d", current->tgid);
1353 if (rc > out_end - out_ptr)
1354 goto out;
1355 out_ptr += rc;
1356 }
1357out:
1358 *out_ptr = 0;
1359 return ispipe;
1360}
1361
1362static void zap_process(struct task_struct *start)
1363{
1364 struct task_struct *t;
1365
1366 start->signal->flags = SIGNAL_GROUP_EXIT;
1367 start->signal->group_stop_count = 0;
1368
1369 t = start;
1370 do {
1371 if (t != current && t->mm) {
1372 t->mm->core_waiters++;
1373 sigaddset(&t->pending.signal, SIGKILL);
1374 signal_wake_up(t, 1);
1375 }
1376 } while ((t = next_thread(t)) != start);
1377}
1378
1379static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1380 int exit_code)
1381{
1382 struct task_struct *g, *p;
1383 unsigned long flags;
1384 int err = -EAGAIN;
1385
1386 spin_lock_irq(&tsk->sighand->siglock);
1387 if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1388 tsk->signal->group_exit_code = exit_code;
1389 zap_process(tsk);
1390 err = 0;
1391 }
1392 spin_unlock_irq(&tsk->sighand->siglock);
1393 if (err)
1394 return err;
1395
1396 if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1397 goto done;
1398
1399 rcu_read_lock();
1400 for_each_process(g) {
1401 if (g == tsk->group_leader)
1402 continue;
1403
1404 p = g;
1405 do {
1406 if (p->mm) {
1407 if (p->mm == mm) {
1408 /*
1409 * p->sighand can't disappear, but
1410 * may be changed by de_thread()
1411 */
1412 lock_task_sighand(p, &flags);
1413 zap_process(p);
1414 unlock_task_sighand(p, &flags);
1415 }
1416 break;
1417 }
1418 } while ((p = next_thread(p)) != g);
1419 }
1420 rcu_read_unlock();
1421done:
1422 return mm->core_waiters;
1423}
1424
1425static int coredump_wait(int exit_code)
1426{
1427 struct task_struct *tsk = current;
1428 struct mm_struct *mm = tsk->mm;
1429 struct completion startup_done;
1430 struct completion *vfork_done;
1431 int core_waiters;
1432
1433 init_completion(&mm->core_done);
1434 init_completion(&startup_done);
1435 mm->core_startup_done = &startup_done;
1436
1437 core_waiters = zap_threads(tsk, mm, exit_code);
1438 up_write(&mm->mmap_sem);
1439
1440 if (unlikely(core_waiters < 0))
1441 goto fail;
1442
1443 /*
1444 * Make sure nobody is waiting for us to release the VM,
1445 * otherwise we can deadlock when we wait on each other
1446 */
1447 vfork_done = tsk->vfork_done;
1448 if (vfork_done) {
1449 tsk->vfork_done = NULL;
1450 complete(vfork_done);
1451 }
1452
1453 if (core_waiters)
1454 wait_for_completion(&startup_done);
1455fail:
1456 BUG_ON(mm->core_waiters);
1457 return core_waiters;
1458}
1459
1460int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1461{
1462 char corename[CORENAME_MAX_SIZE + 1];
1463 struct mm_struct *mm = current->mm;
1464 struct linux_binfmt * binfmt;
1465 struct inode * inode;
1466 struct file * file;
1467 int retval = 0;
1468 int fsuid = current->fsuid;
1469 int flag = 0;
1470 int ispipe = 0;
1471
1472 binfmt = current->binfmt;
1473 if (!binfmt || !binfmt->core_dump)
1474 goto fail;
1475 down_write(&mm->mmap_sem);
1476 if (!mm->dumpable) {
1477 up_write(&mm->mmap_sem);
1478 goto fail;
1479 }
1480
1481 /*
1482 * We cannot trust fsuid as being the "true" uid of the
1483 * process nor do we know its entire history. We only know it
1484 * was tainted so we dump it as root in mode 2.
1485 */
1486 if (mm->dumpable == 2) { /* Setuid core dump mode */
1487 flag = O_EXCL; /* Stop rewrite attacks */
1488 current->fsuid = 0; /* Dump root private */
1489 }
1490 mm->dumpable = 0;
1491
1492 retval = coredump_wait(exit_code);
1493 if (retval < 0)
1494 goto fail;
1495
1496 /*
1497 * Clear any false indication of pending signals that might
1498 * be seen by the filesystem code called to write the core file.
1499 */
1500 clear_thread_flag(TIF_SIGPENDING);
1501
1502 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1503 goto fail_unlock;
1504
1505 /*
1506 * lock_kernel() because format_corename() is controlled by sysctl, which
1507 * uses lock_kernel()
1508 */
1509 lock_kernel();
1510 ispipe = format_corename(corename, core_pattern, signr);
1511 unlock_kernel();
1512 if (ispipe) {
1513 /* SIGPIPE can happen, but it's just never processed */
1514 if(call_usermodehelper_pipe(corename+1, NULL, NULL, &file)) {
1515 printk(KERN_INFO "Core dump to %s pipe failed\n",
1516 corename);
1517 goto fail_unlock;
1518 }
1519 } else
1520 file = filp_open(corename,
1521 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1522 0600);
1523 if (IS_ERR(file))
1524 goto fail_unlock;
1525 inode = file->f_path.dentry->d_inode;
1526 if (inode->i_nlink > 1)
1527 goto close_fail; /* multiple links - don't dump */
1528 if (!ispipe && d_unhashed(file->f_path.dentry))
1529 goto close_fail;
1530
1531 /* AK: actually i see no reason to not allow this for named pipes etc.,
1532 but keep the previous behaviour for now. */
1533 if (!ispipe && !S_ISREG(inode->i_mode))
1534 goto close_fail;
1535 if (!file->f_op)
1536 goto close_fail;
1537 if (!file->f_op->write)
1538 goto close_fail;
1539 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1540 goto close_fail;
1541
1542 retval = binfmt->core_dump(signr, regs, file);
1543
1544 if (retval)
1545 current->signal->group_exit_code |= 0x80;
1546close_fail:
1547 filp_close(file, NULL);
1548fail_unlock:
1549 current->fsuid = fsuid;
1550 complete_all(&mm->core_done);
1551fail:
1552 return retval;
1553}