projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
[PATCH] proc: Rewrite the proc dentry flush on exit optimization
[deliverable/linux.git] / fs / exec.c
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/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>
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
61 int core_uses_pid;
62 char core_pattern[65] = "core";
63 int suid_dumpable = 0;
64
65 EXPORT_SYMBOL(suid_dumpable);
66 /* The maximal length of core_pattern is also specified in sysctl.c */
67
68 static struct linux_binfmt *formats;
69 static DEFINE_RWLOCK(binfmt_lock);
70
71 int 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
93 EXPORT_SYMBOL(register_binfmt);
94
95 int 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
112 EXPORT_SYMBOL(unregister_binfmt);
113
114 static 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 */
125 asmlinkage 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);
168 out:
169 return error;
170 exit:
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 */
179 static 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 */
205 static 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;
278 out:
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 */
287 int 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
297 EXPORT_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 */
306 void 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;
333 out:
334 __free_page(page);
335 force_sig(SIGKILL, current);
336 }
337
338 #define EXTRA_STACK_VM_PAGES 20 /* random */
339
340 int 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_alloc(vm_area_cachep, SLAB_KERNEL);
409 if (!mpnt)
410 return -ENOMEM;
411
412 memset(mpnt, 0, sizeof(*mpnt));
413
414 down_write(&mm->mmap_sem);
415 {
416 mpnt->vm_mm = mm;
417 #ifdef CONFIG_STACK_GROWSUP
418 mpnt->vm_start = stack_base;
419 mpnt->vm_end = stack_base + arg_size;
420 #else
421 mpnt->vm_end = stack_top;
422 mpnt->vm_start = mpnt->vm_end - arg_size;
423 #endif
424 /* Adjust stack execute permissions; explicitly enable
425 * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
426 * and leave alone (arch default) otherwise. */
427 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
428 mpnt->vm_flags = VM_STACK_FLAGS | VM_EXEC;
429 else if (executable_stack == EXSTACK_DISABLE_X)
430 mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
431 else
432 mpnt->vm_flags = VM_STACK_FLAGS;
433 mpnt->vm_flags |= mm->def_flags;
434 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
435 if ((ret = insert_vm_struct(mm, mpnt))) {
436 up_write(&mm->mmap_sem);
437 kmem_cache_free(vm_area_cachep, mpnt);
438 return ret;
439 }
440 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
441 }
442
443 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
444 struct page *page = bprm->page[i];
445 if (page) {
446 bprm->page[i] = NULL;
447 install_arg_page(mpnt, page, stack_base);
448 }
449 stack_base += PAGE_SIZE;
450 }
451 up_write(&mm->mmap_sem);
452
453 return 0;
454 }
455
456 EXPORT_SYMBOL(setup_arg_pages);
457
458 #define free_arg_pages(bprm) do { } while (0)
459
460 #else
461
462 static inline void free_arg_pages(struct linux_binprm *bprm)
463 {
464 int i;
465
466 for (i = 0; i < MAX_ARG_PAGES; i++) {
467 if (bprm->page[i])
468 __free_page(bprm->page[i]);
469 bprm->page[i] = NULL;
470 }
471 }
472
473 #endif /* CONFIG_MMU */
474
475 struct file *open_exec(const char *name)
476 {
477 struct nameidata nd;
478 int err;
479 struct file *file;
480
481 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
482 file = ERR_PTR(err);
483
484 if (!err) {
485 struct inode *inode = nd.dentry->d_inode;
486 file = ERR_PTR(-EACCES);
487 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
488 S_ISREG(inode->i_mode)) {
489 int err = vfs_permission(&nd, MAY_EXEC);
490 if (!err && !(inode->i_mode & 0111))
491 err = -EACCES;
492 file = ERR_PTR(err);
493 if (!err) {
494 file = nameidata_to_filp(&nd, O_RDONLY);
495 if (!IS_ERR(file)) {
496 err = deny_write_access(file);
497 if (err) {
498 fput(file);
499 file = ERR_PTR(err);
500 }
501 }
502 out:
503 return file;
504 }
505 }
506 release_open_intent(&nd);
507 path_release(&nd);
508 }
509 goto out;
510 }
511
512 EXPORT_SYMBOL(open_exec);
513
514 int kernel_read(struct file *file, unsigned long offset,
515 char *addr, unsigned long count)
516 {
517 mm_segment_t old_fs;
518 loff_t pos = offset;
519 int result;
520
521 old_fs = get_fs();
522 set_fs(get_ds());
523 /* The cast to a user pointer is valid due to the set_fs() */
524 result = vfs_read(file, (void __user *)addr, count, &pos);
525 set_fs(old_fs);
526 return result;
527 }
528
529 EXPORT_SYMBOL(kernel_read);
530
531 static int exec_mmap(struct mm_struct *mm)
532 {
533 struct task_struct *tsk;
534 struct mm_struct * old_mm, *active_mm;
535
536 /* Notify parent that we're no longer interested in the old VM */
537 tsk = current;
538 old_mm = current->mm;
539 mm_release(tsk, old_mm);
540
541 if (old_mm) {
542 /*
543 * Make sure that if there is a core dump in progress
544 * for the old mm, we get out and die instead of going
545 * through with the exec. We must hold mmap_sem around
546 * checking core_waiters and changing tsk->mm. The
547 * core-inducing thread will increment core_waiters for
548 * each thread whose ->mm == old_mm.
549 */
550 down_read(&old_mm->mmap_sem);
551 if (unlikely(old_mm->core_waiters)) {
552 up_read(&old_mm->mmap_sem);
553 return -EINTR;
554 }
555 }
556 task_lock(tsk);
557 active_mm = tsk->active_mm;
558 tsk->mm = mm;
559 tsk->active_mm = mm;
560 activate_mm(active_mm, mm);
561 task_unlock(tsk);
562 arch_pick_mmap_layout(mm);
563 if (old_mm) {
564 up_read(&old_mm->mmap_sem);
565 BUG_ON(active_mm != old_mm);
566 mmput(old_mm);
567 return 0;
568 }
569 mmdrop(active_mm);
570 return 0;
571 }
572
573 /*
574 * This function makes sure the current process has its own signal table,
575 * so that flush_signal_handlers can later reset the handlers without
576 * disturbing other processes. (Other processes might share the signal
577 * table via the CLONE_SIGHAND option to clone().)
578 */
579 static int de_thread(struct task_struct *tsk)
580 {
581 struct signal_struct *sig = tsk->signal;
582 struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
583 spinlock_t *lock = &oldsighand->siglock;
584 struct task_struct *leader = NULL;
585 int count;
586
587 /*
588 * If we don't share sighandlers, then we aren't sharing anything
589 * and we can just re-use it all.
590 */
591 if (atomic_read(&oldsighand->count) <= 1) {
592 BUG_ON(atomic_read(&sig->count) != 1);
593 exit_itimers(sig);
594 return 0;
595 }
596
597 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
598 if (!newsighand)
599 return -ENOMEM;
600
601 if (thread_group_empty(current))
602 goto no_thread_group;
603
604 /*
605 * Kill all other threads in the thread group.
606 * We must hold tasklist_lock to call zap_other_threads.
607 */
608 read_lock(&tasklist_lock);
609 spin_lock_irq(lock);
610 if (sig->flags & SIGNAL_GROUP_EXIT) {
611 /*
612 * Another group action in progress, just
613 * return so that the signal is processed.
614 */
615 spin_unlock_irq(lock);
616 read_unlock(&tasklist_lock);
617 kmem_cache_free(sighand_cachep, newsighand);
618 return -EAGAIN;
619 }
620
621 /*
622 * child_reaper ignores SIGKILL, change it now.
623 * Reparenting needs write_lock on tasklist_lock,
624 * so it is safe to do it under read_lock.
625 */
626 if (unlikely(current->group_leader == child_reaper))
627 child_reaper = current;
628
629 zap_other_threads(current);
630 read_unlock(&tasklist_lock);
631
632 /*
633 * Account for the thread group leader hanging around:
634 */
635 count = 1;
636 if (!thread_group_leader(current)) {
637 count = 2;
638 /*
639 * The SIGALRM timer survives the exec, but needs to point
640 * at us as the new group leader now. We have a race with
641 * a timer firing now getting the old leader, so we need to
642 * synchronize with any firing (by calling del_timer_sync)
643 * before we can safely let the old group leader die.
644 */
645 sig->tsk = current;
646 spin_unlock_irq(lock);
647 if (hrtimer_cancel(&sig->real_timer))
648 hrtimer_restart(&sig->real_timer);
649 spin_lock_irq(lock);
650 }
651 while (atomic_read(&sig->count) > count) {
652 sig->group_exit_task = current;
653 sig->notify_count = count;
654 __set_current_state(TASK_UNINTERRUPTIBLE);
655 spin_unlock_irq(lock);
656 schedule();
657 spin_lock_irq(lock);
658 }
659 sig->group_exit_task = NULL;
660 sig->notify_count = 0;
661 spin_unlock_irq(lock);
662
663 /*
664 * At this point all other threads have exited, all we have to
665 * do is to wait for the thread group leader to become inactive,
666 * and to assume its PID:
667 */
668 if (!thread_group_leader(current)) {
669 /*
670 * Wait for the thread group leader to be a zombie.
671 * It should already be zombie at this point, most
672 * of the time.
673 */
674 leader = current->group_leader;
675 while (leader->exit_state != EXIT_ZOMBIE)
676 yield();
677
678 /*
679 * The only record we have of the real-time age of a
680 * process, regardless of execs it's done, is start_time.
681 * All the past CPU time is accumulated in signal_struct
682 * from sister threads now dead. But in this non-leader
683 * exec, nothing survives from the original leader thread,
684 * whose birth marks the true age of this process now.
685 * When we take on its identity by switching to its PID, we
686 * also take its birthdate (always earlier than our own).
687 */
688 current->start_time = leader->start_time;
689
690 write_lock_irq(&tasklist_lock);
691
692 BUG_ON(leader->tgid != current->tgid);
693 BUG_ON(current->pid == current->tgid);
694 /*
695 * An exec() starts a new thread group with the
696 * TGID of the previous thread group. Rehash the
697 * two threads with a switched PID, and release
698 * the former thread group leader:
699 */
700
701 /* Become a process group leader with the old leader's pid.
702 * Note: The old leader also uses thispid until release_task
703 * is called. Odd but simple and correct.
704 */
705 detach_pid(current, PIDTYPE_PID);
706 current->pid = leader->pid;
707 attach_pid(current, PIDTYPE_PID, current->pid);
708 attach_pid(current, PIDTYPE_PGID, current->signal->pgrp);
709 attach_pid(current, PIDTYPE_SID, current->signal->session);
710 list_add_tail_rcu(&current->tasks, &init_task.tasks);
711
712 current->group_leader = current;
713 leader->group_leader = current;
714
715 /* Reduce leader to a thread */
716 detach_pid(leader, PIDTYPE_PGID);
717 detach_pid(leader, PIDTYPE_SID);
718 list_del_init(&leader->tasks);
719
720 current->exit_signal = SIGCHLD;
721
722 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
723 leader->exit_state = EXIT_DEAD;
724
725 write_unlock_irq(&tasklist_lock);
726 }
727
728 /*
729 * There may be one thread left which is just exiting,
730 * but it's safe to stop telling the group to kill themselves.
731 */
732 sig->flags = 0;
733
734 no_thread_group:
735 exit_itimers(sig);
736 if (leader)
737 release_task(leader);
738
739 BUG_ON(atomic_read(&sig->count) != 1);
740
741 if (atomic_read(&oldsighand->count) == 1) {
742 /*
743 * Now that we nuked the rest of the thread group,
744 * it turns out we are not sharing sighand any more either.
745 * So we can just keep it.
746 */
747 kmem_cache_free(sighand_cachep, newsighand);
748 } else {
749 /*
750 * Move our state over to newsighand and switch it in.
751 */
752 atomic_set(&newsighand->count, 1);
753 memcpy(newsighand->action, oldsighand->action,
754 sizeof(newsighand->action));
755
756 write_lock_irq(&tasklist_lock);
757 spin_lock(&oldsighand->siglock);
758 spin_lock(&newsighand->siglock);
759
760 rcu_assign_pointer(current->sighand, newsighand);
761 recalc_sigpending();
762
763 spin_unlock(&newsighand->siglock);
764 spin_unlock(&oldsighand->siglock);
765 write_unlock_irq(&tasklist_lock);
766
767 if (atomic_dec_and_test(&oldsighand->count))
768 kmem_cache_free(sighand_cachep, oldsighand);
769 }
770
771 BUG_ON(!thread_group_leader(current));
772 return 0;
773 }
774
775 /*
776 * These functions flushes out all traces of the currently running executable
777 * so that a new one can be started
778 */
779
780 static void flush_old_files(struct files_struct * files)
781 {
782 long j = -1;
783 struct fdtable *fdt;
784
785 spin_lock(&files->file_lock);
786 for (;;) {
787 unsigned long set, i;
788
789 j++;
790 i = j * __NFDBITS;
791 fdt = files_fdtable(files);
792 if (i >= fdt->max_fds || i >= fdt->max_fdset)
793 break;
794 set = fdt->close_on_exec->fds_bits[j];
795 if (!set)
796 continue;
797 fdt->close_on_exec->fds_bits[j] = 0;
798 spin_unlock(&files->file_lock);
799 for ( ; set ; i++,set >>= 1) {
800 if (set & 1) {
801 sys_close(i);
802 }
803 }
804 spin_lock(&files->file_lock);
805
806 }
807 spin_unlock(&files->file_lock);
808 }
809
810 void get_task_comm(char *buf, struct task_struct *tsk)
811 {
812 /* buf must be at least sizeof(tsk->comm) in size */
813 task_lock(tsk);
814 strncpy(buf, tsk->comm, sizeof(tsk->comm));
815 task_unlock(tsk);
816 }
817
818 void set_task_comm(struct task_struct *tsk, char *buf)
819 {
820 task_lock(tsk);
821 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
822 task_unlock(tsk);
823 }
824
825 int flush_old_exec(struct linux_binprm * bprm)
826 {
827 char * name;
828 int i, ch, retval;
829 struct files_struct *files;
830 char tcomm[sizeof(current->comm)];
831
832 /*
833 * Make sure we have a private signal table and that
834 * we are unassociated from the previous thread group.
835 */
836 retval = de_thread(current);
837 if (retval)
838 goto out;
839
840 /*
841 * Make sure we have private file handles. Ask the
842 * fork helper to do the work for us and the exit
843 * helper to do the cleanup of the old one.
844 */
845 files = current->files; /* refcounted so safe to hold */
846 retval = unshare_files();
847 if (retval)
848 goto out;
849 /*
850 * Release all of the old mmap stuff
851 */
852 retval = exec_mmap(bprm->mm);
853 if (retval)
854 goto mmap_failed;
855
856 bprm->mm = NULL; /* We're using it now */
857
858 /* This is the point of no return */
859 put_files_struct(files);
860
861 current->sas_ss_sp = current->sas_ss_size = 0;
862
863 if (current->euid == current->uid && current->egid == current->gid)
864 current->mm->dumpable = 1;
865 else
866 current->mm->dumpable = suid_dumpable;
867
868 name = bprm->filename;
869
870 /* Copies the binary name from after last slash */
871 for (i=0; (ch = *(name++)) != '\0';) {
872 if (ch == '/')
873 i = 0; /* overwrite what we wrote */
874 else
875 if (i < (sizeof(tcomm) - 1))
876 tcomm[i++] = ch;
877 }
878 tcomm[i] = '\0';
879 set_task_comm(current, tcomm);
880
881 current->flags &= ~PF_RANDOMIZE;
882 flush_thread();
883
884 /* Set the new mm task size. We have to do that late because it may
885 * depend on TIF_32BIT which is only updated in flush_thread() on
886 * some architectures like powerpc
887 */
888 current->mm->task_size = TASK_SIZE;
889
890 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
891 file_permission(bprm->file, MAY_READ) ||
892 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
893 suid_keys(current);
894 current->mm->dumpable = suid_dumpable;
895 }
896
897 /* An exec changes our domain. We are no longer part of the thread
898 group */
899
900 current->self_exec_id++;
901
902 flush_signal_handlers(current, 0);
903 flush_old_files(current->files);
904
905 return 0;
906
907 mmap_failed:
908 put_files_struct(current->files);
909 current->files = files;
910 out:
911 return retval;
912 }
913
914 EXPORT_SYMBOL(flush_old_exec);
915
916 /*
917 * Fill the binprm structure from the inode.
918 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
919 */
920 int prepare_binprm(struct linux_binprm *bprm)
921 {
922 int mode;
923 struct inode * inode = bprm->file->f_dentry->d_inode;
924 int retval;
925
926 mode = inode->i_mode;
927 /*
928 * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
929 * generic_permission lets a non-executable through
930 */
931 if (!(mode & 0111)) /* with at least _one_ execute bit set */
932 return -EACCES;
933 if (bprm->file->f_op == NULL)
934 return -EACCES;
935
936 bprm->e_uid = current->euid;
937 bprm->e_gid = current->egid;
938
939 if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
940 /* Set-uid? */
941 if (mode & S_ISUID) {
942 current->personality &= ~PER_CLEAR_ON_SETID;
943 bprm->e_uid = inode->i_uid;
944 }
945
946 /* Set-gid? */
947 /*
948 * If setgid is set but no group execute bit then this
949 * is a candidate for mandatory locking, not a setgid
950 * executable.
951 */
952 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
953 current->personality &= ~PER_CLEAR_ON_SETID;
954 bprm->e_gid = inode->i_gid;
955 }
956 }
957
958 /* fill in binprm security blob */
959 retval = security_bprm_set(bprm);
960 if (retval)
961 return retval;
962
963 memset(bprm->buf,0,BINPRM_BUF_SIZE);
964 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
965 }
966
967 EXPORT_SYMBOL(prepare_binprm);
968
969 static int unsafe_exec(struct task_struct *p)
970 {
971 int unsafe = 0;
972 if (p->ptrace & PT_PTRACED) {
973 if (p->ptrace & PT_PTRACE_CAP)
974 unsafe |= LSM_UNSAFE_PTRACE_CAP;
975 else
976 unsafe |= LSM_UNSAFE_PTRACE;
977 }
978 if (atomic_read(&p->fs->count) > 1 ||
979 atomic_read(&p->files->count) > 1 ||
980 atomic_read(&p->sighand->count) > 1)
981 unsafe |= LSM_UNSAFE_SHARE;
982
983 return unsafe;
984 }
985
986 void compute_creds(struct linux_binprm *bprm)
987 {
988 int unsafe;
989
990 if (bprm->e_uid != current->uid)
991 suid_keys(current);
992 exec_keys(current);
993
994 task_lock(current);
995 unsafe = unsafe_exec(current);
996 security_bprm_apply_creds(bprm, unsafe);
997 task_unlock(current);
998 security_bprm_post_apply_creds(bprm);
999 }
1000
1001 EXPORT_SYMBOL(compute_creds);
1002
1003 void remove_arg_zero(struct linux_binprm *bprm)
1004 {
1005 if (bprm->argc) {
1006 unsigned long offset;
1007 char * kaddr;
1008 struct page *page;
1009
1010 offset = bprm->p % PAGE_SIZE;
1011 goto inside;
1012
1013 while (bprm->p++, *(kaddr+offset++)) {
1014 if (offset != PAGE_SIZE)
1015 continue;
1016 offset = 0;
1017 kunmap_atomic(kaddr, KM_USER0);
1018 inside:
1019 page = bprm->page[bprm->p/PAGE_SIZE];
1020 kaddr = kmap_atomic(page, KM_USER0);
1021 }
1022 kunmap_atomic(kaddr, KM_USER0);
1023 bprm->argc--;
1024 }
1025 }
1026
1027 EXPORT_SYMBOL(remove_arg_zero);
1028
1029 /*
1030 * cycle the list of binary formats handler, until one recognizes the image
1031 */
1032 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1033 {
1034 int try,retval;
1035 struct linux_binfmt *fmt;
1036 #ifdef __alpha__
1037 /* handle /sbin/loader.. */
1038 {
1039 struct exec * eh = (struct exec *) bprm->buf;
1040
1041 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1042 (eh->fh.f_flags & 0x3000) == 0x3000)
1043 {
1044 struct file * file;
1045 unsigned long loader;
1046
1047 allow_write_access(bprm->file);
1048 fput(bprm->file);
1049 bprm->file = NULL;
1050
1051 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1052
1053 file = open_exec("/sbin/loader");
1054 retval = PTR_ERR(file);
1055 if (IS_ERR(file))
1056 return retval;
1057
1058 /* Remember if the application is TASO. */
1059 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1060
1061 bprm->file = file;
1062 bprm->loader = loader;
1063 retval = prepare_binprm(bprm);
1064 if (retval<0)
1065 return retval;
1066 /* should call search_binary_handler recursively here,
1067 but it does not matter */
1068 }
1069 }
1070 #endif
1071 retval = security_bprm_check(bprm);
1072 if (retval)
1073 return retval;
1074
1075 /* kernel module loader fixup */
1076 /* so we don't try to load run modprobe in kernel space. */
1077 set_fs(USER_DS);
1078
1079 retval = audit_bprm(bprm);
1080 if (retval)
1081 return retval;
1082
1083 retval = -ENOENT;
1084 for (try=0; try<2; try++) {
1085 read_lock(&binfmt_lock);
1086 for (fmt = formats ; fmt ; fmt = fmt->next) {
1087 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1088 if (!fn)
1089 continue;
1090 if (!try_module_get(fmt->module))
1091 continue;
1092 read_unlock(&binfmt_lock);
1093 retval = fn(bprm, regs);
1094 if (retval >= 0) {
1095 put_binfmt(fmt);
1096 allow_write_access(bprm->file);
1097 if (bprm->file)
1098 fput(bprm->file);
1099 bprm->file = NULL;
1100 current->did_exec = 1;
1101 proc_exec_connector(current);
1102 return retval;
1103 }
1104 read_lock(&binfmt_lock);
1105 put_binfmt(fmt);
1106 if (retval != -ENOEXEC || bprm->mm == NULL)
1107 break;
1108 if (!bprm->file) {
1109 read_unlock(&binfmt_lock);
1110 return retval;
1111 }
1112 }
1113 read_unlock(&binfmt_lock);
1114 if (retval != -ENOEXEC || bprm->mm == NULL) {
1115 break;
1116 #ifdef CONFIG_KMOD
1117 }else{
1118 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1119 if (printable(bprm->buf[0]) &&
1120 printable(bprm->buf[1]) &&
1121 printable(bprm->buf[2]) &&
1122 printable(bprm->buf[3]))
1123 break; /* -ENOEXEC */
1124 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1125 #endif
1126 }
1127 }
1128 return retval;
1129 }
1130
1131 EXPORT_SYMBOL(search_binary_handler);
1132
1133 /*
1134 * sys_execve() executes a new program.
1135 */
1136 int do_execve(char * filename,
1137 char __user *__user *argv,
1138 char __user *__user *envp,
1139 struct pt_regs * regs)
1140 {
1141 struct linux_binprm *bprm;
1142 struct file *file;
1143 int retval;
1144 int i;
1145
1146 retval = -ENOMEM;
1147 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1148 if (!bprm)
1149 goto out_ret;
1150
1151 file = open_exec(filename);
1152 retval = PTR_ERR(file);
1153 if (IS_ERR(file))
1154 goto out_kfree;
1155
1156 sched_exec();
1157
1158 bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1159
1160 bprm->file = file;
1161 bprm->filename = filename;
1162 bprm->interp = filename;
1163 bprm->mm = mm_alloc();
1164 retval = -ENOMEM;
1165 if (!bprm->mm)
1166 goto out_file;
1167
1168 retval = init_new_context(current, bprm->mm);
1169 if (retval < 0)
1170 goto out_mm;
1171
1172 bprm->argc = count(argv, bprm->p / sizeof(void *));
1173 if ((retval = bprm->argc) < 0)
1174 goto out_mm;
1175
1176 bprm->envc = count(envp, bprm->p / sizeof(void *));
1177 if ((retval = bprm->envc) < 0)
1178 goto out_mm;
1179
1180 retval = security_bprm_alloc(bprm);
1181 if (retval)
1182 goto out;
1183
1184 retval = prepare_binprm(bprm);
1185 if (retval < 0)
1186 goto out;
1187
1188 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1189 if (retval < 0)
1190 goto out;
1191
1192 bprm->exec = bprm->p;
1193 retval = copy_strings(bprm->envc, envp, bprm);
1194 if (retval < 0)
1195 goto out;
1196
1197 retval = copy_strings(bprm->argc, argv, bprm);
1198 if (retval < 0)
1199 goto out;
1200
1201 retval = search_binary_handler(bprm,regs);
1202 if (retval >= 0) {
1203 free_arg_pages(bprm);
1204
1205 /* execve success */
1206 security_bprm_free(bprm);
1207 acct_update_integrals(current);
1208 kfree(bprm);
1209 return retval;
1210 }
1211
1212 out:
1213 /* Something went wrong, return the inode and free the argument pages*/
1214 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1215 struct page * page = bprm->page[i];
1216 if (page)
1217 __free_page(page);
1218 }
1219
1220 if (bprm->security)
1221 security_bprm_free(bprm);
1222
1223 out_mm:
1224 if (bprm->mm)
1225 mmdrop(bprm->mm);
1226
1227 out_file:
1228 if (bprm->file) {
1229 allow_write_access(bprm->file);
1230 fput(bprm->file);
1231 }
1232
1233 out_kfree:
1234 kfree(bprm);
1235
1236 out_ret:
1237 return retval;
1238 }
1239
1240 int set_binfmt(struct linux_binfmt *new)
1241 {
1242 struct linux_binfmt *old = current->binfmt;
1243
1244 if (new) {
1245 if (!try_module_get(new->module))
1246 return -1;
1247 }
1248 current->binfmt = new;
1249 if (old)
1250 module_put(old->module);
1251 return 0;
1252 }
1253
1254 EXPORT_SYMBOL(set_binfmt);
1255
1256 #define CORENAME_MAX_SIZE 64
1257
1258 /* format_corename will inspect the pattern parameter, and output a
1259 * name into corename, which must have space for at least
1260 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1261 */
1262 static void format_corename(char *corename, const char *pattern, long signr)
1263 {
1264 const char *pat_ptr = pattern;
1265 char *out_ptr = corename;
1266 char *const out_end = corename + CORENAME_MAX_SIZE;
1267 int rc;
1268 int pid_in_pattern = 0;
1269
1270 /* Repeat as long as we have more pattern to process and more output
1271 space */
1272 while (*pat_ptr) {
1273 if (*pat_ptr != '%') {
1274 if (out_ptr == out_end)
1275 goto out;
1276 *out_ptr++ = *pat_ptr++;
1277 } else {
1278 switch (*++pat_ptr) {
1279 case 0:
1280 goto out;
1281 /* Double percent, output one percent */
1282 case '%':
1283 if (out_ptr == out_end)
1284 goto out;
1285 *out_ptr++ = '%';
1286 break;
1287 /* pid */
1288 case 'p':
1289 pid_in_pattern = 1;
1290 rc = snprintf(out_ptr, out_end - out_ptr,
1291 "%d", current->tgid);
1292 if (rc > out_end - out_ptr)
1293 goto out;
1294 out_ptr += rc;
1295 break;
1296 /* uid */
1297 case 'u':
1298 rc = snprintf(out_ptr, out_end - out_ptr,
1299 "%d", current->uid);
1300 if (rc > out_end - out_ptr)
1301 goto out;
1302 out_ptr += rc;
1303 break;
1304 /* gid */
1305 case 'g':
1306 rc = snprintf(out_ptr, out_end - out_ptr,
1307 "%d", current->gid);
1308 if (rc > out_end - out_ptr)
1309 goto out;
1310 out_ptr += rc;
1311 break;
1312 /* signal that caused the coredump */
1313 case 's':
1314 rc = snprintf(out_ptr, out_end - out_ptr,
1315 "%ld", signr);
1316 if (rc > out_end - out_ptr)
1317 goto out;
1318 out_ptr += rc;
1319 break;
1320 /* UNIX time of coredump */
1321 case 't': {
1322 struct timeval tv;
1323 do_gettimeofday(&tv);
1324 rc = snprintf(out_ptr, out_end - out_ptr,
1325 "%lu", tv.tv_sec);
1326 if (rc > out_end - out_ptr)
1327 goto out;
1328 out_ptr += rc;
1329 break;
1330 }
1331 /* hostname */
1332 case 'h':
1333 down_read(&uts_sem);
1334 rc = snprintf(out_ptr, out_end - out_ptr,
1335 "%s", system_utsname.nodename);
1336 up_read(&uts_sem);
1337 if (rc > out_end - out_ptr)
1338 goto out;
1339 out_ptr += rc;
1340 break;
1341 /* executable */
1342 case 'e':
1343 rc = snprintf(out_ptr, out_end - out_ptr,
1344 "%s", current->comm);
1345 if (rc > out_end - out_ptr)
1346 goto out;
1347 out_ptr += rc;
1348 break;
1349 default:
1350 break;
1351 }
1352 ++pat_ptr;
1353 }
1354 }
1355 /* Backward compatibility with core_uses_pid:
1356 *
1357 * If core_pattern does not include a %p (as is the default)
1358 * and core_uses_pid is set, then .%pid will be appended to
1359 * the filename */
1360 if (!pid_in_pattern
1361 && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1362 rc = snprintf(out_ptr, out_end - out_ptr,
1363 ".%d", current->tgid);
1364 if (rc > out_end - out_ptr)
1365 goto out;
1366 out_ptr += rc;
1367 }
1368 out:
1369 *out_ptr = 0;
1370 }
1371
1372 static void zap_threads (struct mm_struct *mm)
1373 {
1374 struct task_struct *g, *p;
1375 struct task_struct *tsk = current;
1376 struct completion *vfork_done = tsk->vfork_done;
1377 int traced = 0;
1378
1379 /*
1380 * Make sure nobody is waiting for us to release the VM,
1381 * otherwise we can deadlock when we wait on each other
1382 */
1383 if (vfork_done) {
1384 tsk->vfork_done = NULL;
1385 complete(vfork_done);
1386 }
1387
1388 read_lock(&tasklist_lock);
1389 do_each_thread(g,p)
1390 if (mm == p->mm && p != tsk) {
1391 force_sig_specific(SIGKILL, p);
1392 mm->core_waiters++;
1393 if (unlikely(p->ptrace) &&
1394 unlikely(p->parent->mm == mm))
1395 traced = 1;
1396 }
1397 while_each_thread(g,p);
1398
1399 read_unlock(&tasklist_lock);
1400
1401 if (unlikely(traced)) {
1402 /*
1403 * We are zapping a thread and the thread it ptraces.
1404 * If the tracee went into a ptrace stop for exit tracing,
1405 * we could deadlock since the tracer is waiting for this
1406 * coredump to finish. Detach them so they can both die.
1407 */
1408 write_lock_irq(&tasklist_lock);
1409 do_each_thread(g,p) {
1410 if (mm == p->mm && p != tsk &&
1411 p->ptrace && p->parent->mm == mm) {
1412 __ptrace_detach(p, 0);
1413 }
1414 } while_each_thread(g,p);
1415 write_unlock_irq(&tasklist_lock);
1416 }
1417 }
1418
1419 static void coredump_wait(struct mm_struct *mm)
1420 {
1421 DECLARE_COMPLETION(startup_done);
1422 int core_waiters;
1423
1424 mm->core_startup_done = &startup_done;
1425
1426 zap_threads(mm);
1427 core_waiters = mm->core_waiters;
1428 up_write(&mm->mmap_sem);
1429
1430 if (core_waiters)
1431 wait_for_completion(&startup_done);
1432 BUG_ON(mm->core_waiters);
1433 }
1434
1435 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1436 {
1437 char corename[CORENAME_MAX_SIZE + 1];
1438 struct mm_struct *mm = current->mm;
1439 struct linux_binfmt * binfmt;
1440 struct inode * inode;
1441 struct file * file;
1442 int retval = 0;
1443 int fsuid = current->fsuid;
1444 int flag = 0;
1445
1446 binfmt = current->binfmt;
1447 if (!binfmt || !binfmt->core_dump)
1448 goto fail;
1449 down_write(&mm->mmap_sem);
1450 if (!mm->dumpable) {
1451 up_write(&mm->mmap_sem);
1452 goto fail;
1453 }
1454
1455 /*
1456 * We cannot trust fsuid as being the "true" uid of the
1457 * process nor do we know its entire history. We only know it
1458 * was tainted so we dump it as root in mode 2.
1459 */
1460 if (mm->dumpable == 2) { /* Setuid core dump mode */
1461 flag = O_EXCL; /* Stop rewrite attacks */
1462 current->fsuid = 0; /* Dump root private */
1463 }
1464 mm->dumpable = 0;
1465
1466 retval = -EAGAIN;
1467 spin_lock_irq(&current->sighand->siglock);
1468 if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
1469 current->signal->flags = SIGNAL_GROUP_EXIT;
1470 current->signal->group_exit_code = exit_code;
1471 current->signal->group_stop_count = 0;
1472 retval = 0;
1473 }
1474 spin_unlock_irq(&current->sighand->siglock);
1475 if (retval) {
1476 up_write(&mm->mmap_sem);
1477 goto fail;
1478 }
1479
1480 init_completion(&mm->core_done);
1481 coredump_wait(mm);
1482
1483 /*
1484 * Clear any false indication of pending signals that might
1485 * be seen by the filesystem code called to write the core file.
1486 */
1487 clear_thread_flag(TIF_SIGPENDING);
1488
1489 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1490 goto fail_unlock;
1491
1492 /*
1493 * lock_kernel() because format_corename() is controlled by sysctl, which
1494 * uses lock_kernel()
1495 */
1496 lock_kernel();
1497 format_corename(corename, core_pattern, signr);
1498 unlock_kernel();
1499 file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
1500 if (IS_ERR(file))
1501 goto fail_unlock;
1502 inode = file->f_dentry->d_inode;
1503 if (inode->i_nlink > 1)
1504 goto close_fail; /* multiple links - don't dump */
1505 if (d_unhashed(file->f_dentry))
1506 goto close_fail;
1507
1508 if (!S_ISREG(inode->i_mode))
1509 goto close_fail;
1510 if (!file->f_op)
1511 goto close_fail;
1512 if (!file->f_op->write)
1513 goto close_fail;
1514 if (do_truncate(file->f_dentry, 0, 0, file) != 0)
1515 goto close_fail;
1516
1517 retval = binfmt->core_dump(signr, regs, file);
1518
1519 if (retval)
1520 current->signal->group_exit_code |= 0x80;
1521 close_fail:
1522 filp_close(file, NULL);
1523 fail_unlock:
1524 current->fsuid = fsuid;
1525 complete_all(&mm->core_done);
1526 fail:
1527 return retval;
1528 }
Linux kernel - Deliverables
This page took 0.073418 seconds and 6 git commands to generate.