ia64/linux-2.6.18-xen.hg

view fs/exec.c @ 452:c7ed6fe5dca0

kexec: dont initialise regions in reserve_memory()

There is no need to initialise efi_memmap_res and boot_param_res in
reserve_memory() for the initial xen domain as it is done in
machine_kexec_setup_resources() using values from the kexec hypercall.

Signed-off-by: Simon Horman <horms@verge.net.au>
author Keir Fraser <keir.fraser@citrix.com>
date Thu Feb 28 10:55:18 2008 +0000 (2008-02-28)
parents 831230e53067
children
line source
1 /*
2 * linux/fs/exec.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
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 */
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/module.h>
42 #include <linux/namei.h>
43 #include <linux/proc_fs.h>
44 #include <linux/ptrace.h>
45 #include <linux/mount.h>
46 #include <linux/security.h>
47 #include <linux/syscalls.h>
48 #include <linux/rmap.h>
49 #include <linux/acct.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.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)
71 {
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;
83 }
84 tmp = &(*tmp)->next;
85 }
86 fmt->next = formats;
87 formats = fmt;
88 write_unlock(&binfmt_lock);
89 return 0;
90 }
92 EXPORT_SYMBOL(register_binfmt);
94 int unregister_binfmt(struct linux_binfmt * fmt)
95 {
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;
104 }
105 tmp = &(*tmp)->next;
106 }
107 write_unlock(&binfmt_lock);
108 return -EINVAL;
109 }
111 EXPORT_SYMBOL(unregister_binfmt);
113 static inline void put_binfmt(struct linux_binfmt * fmt)
114 {
115 module_put(fmt->module);
116 }
118 /*
119 * Note that a shared library must be both readable and executable due to
120 * security reasons.
121 *
122 * Also note that we take the address to load from from the file itself.
123 */
124 asmlinkage long sys_uselib(const char __user * library)
125 {
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;
163 }
164 read_unlock(&binfmt_lock);
165 }
166 fput(file);
167 out:
168 return error;
169 exit:
170 release_open_intent(&nd);
171 path_release(&nd);
172 goto out;
173 }
175 /*
176 * count() counts the number of strings in array ARGV.
177 */
178 static int count(char __user * __user * argv, int max)
179 {
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();
194 }
195 }
196 return i;
197 }
199 /*
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.
203 */
204 static int copy_strings(int argc, char __user * __user * argv,
205 struct linux_binprm *bprm)
206 {
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;
220 }
222 if (bprm->p < len) {
223 ret = -E2BIG;
224 goto out;
225 }
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;
246 }
247 new = 1;
248 }
250 if (page != kmapped_page) {
251 if (kmapped_page)
252 kunmap(kmapped_page);
253 kmapped_page = page;
254 kaddr = kmap(kmapped_page);
255 }
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);
264 }
265 err = copy_from_user(kaddr+offset, str, bytes_to_copy);
266 if (err) {
267 ret = -EFAULT;
268 goto out;
269 }
271 pos += bytes_to_copy;
272 str += bytes_to_copy;
273 len -= bytes_to_copy;
274 }
275 }
276 ret = 0;
277 out:
278 if (kmapped_page)
279 kunmap(kmapped_page);
280 return ret;
281 }
283 /*
284 * Like copy_strings, but get argv and its values from kernel memory.
285 */
286 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
287 {
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;
294 }
296 EXPORT_SYMBOL(copy_strings_kernel);
298 #ifdef CONFIG_MMU
299 /*
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.
302 *
303 * vma->vm_mm->mmap_sem is held for writing.
304 */
305 void install_arg_page(struct vm_area_struct *vma,
306 struct page *page, unsigned long address)
307 {
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;
322 }
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);
335 }
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)
342 {
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.
352 */
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;
363 }
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;
374 }
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);
414 {
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;
438 }
439 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
440 }
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);
447 }
448 stack_base += PAGE_SIZE;
449 }
450 up_write(&mm->mmap_sem);
452 return 0;
453 }
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)
462 {
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;
469 }
470 }
472 #endif /* CONFIG_MMU */
474 struct file *open_exec(const char *name)
475 {
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 file = ERR_PTR(err);
490 if (!err) {
491 file = nameidata_to_filp(&nd, O_RDONLY);
492 if (!IS_ERR(file)) {
493 err = deny_write_access(file);
494 if (err) {
495 fput(file);
496 file = ERR_PTR(err);
497 }
498 }
499 out:
500 return file;
501 }
502 }
503 release_open_intent(&nd);
504 path_release(&nd);
505 }
506 goto out;
507 }
509 EXPORT_SYMBOL(open_exec);
511 int kernel_read(struct file *file, unsigned long offset,
512 char *addr, unsigned long count)
513 {
514 mm_segment_t old_fs;
515 loff_t pos = offset;
516 int result;
518 old_fs = get_fs();
519 set_fs(get_ds());
520 /* The cast to a user pointer is valid due to the set_fs() */
521 result = vfs_read(file, (void __user *)addr, count, &pos);
522 set_fs(old_fs);
523 return result;
524 }
526 EXPORT_SYMBOL(kernel_read);
528 static int exec_mmap(struct mm_struct *mm)
529 {
530 struct task_struct *tsk;
531 struct mm_struct * old_mm, *active_mm;
533 /* Notify parent that we're no longer interested in the old VM */
534 tsk = current;
535 old_mm = current->mm;
536 mm_release(tsk, old_mm);
538 if (old_mm) {
539 /*
540 * Make sure that if there is a core dump in progress
541 * for the old mm, we get out and die instead of going
542 * through with the exec. We must hold mmap_sem around
543 * checking core_waiters and changing tsk->mm. The
544 * core-inducing thread will increment core_waiters for
545 * each thread whose ->mm == old_mm.
546 */
547 down_read(&old_mm->mmap_sem);
548 if (unlikely(old_mm->core_waiters)) {
549 up_read(&old_mm->mmap_sem);
550 return -EINTR;
551 }
552 }
553 task_lock(tsk);
554 active_mm = tsk->active_mm;
555 tsk->mm = mm;
556 tsk->active_mm = mm;
557 activate_mm(active_mm, mm);
558 task_unlock(tsk);
559 arch_pick_mmap_layout(mm);
560 if (old_mm) {
561 up_read(&old_mm->mmap_sem);
562 BUG_ON(active_mm != old_mm);
563 mmput(old_mm);
564 return 0;
565 }
566 mmdrop(active_mm);
567 return 0;
568 }
570 /*
571 * This function makes sure the current process has its own signal table,
572 * so that flush_signal_handlers can later reset the handlers without
573 * disturbing other processes. (Other processes might share the signal
574 * table via the CLONE_SIGHAND option to clone().)
575 */
576 static int de_thread(struct task_struct *tsk)
577 {
578 struct signal_struct *sig = tsk->signal;
579 struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
580 spinlock_t *lock = &oldsighand->siglock;
581 struct task_struct *leader = NULL;
582 int count;
584 /*
585 * If we don't share sighandlers, then we aren't sharing anything
586 * and we can just re-use it all.
587 */
588 if (atomic_read(&oldsighand->count) <= 1) {
589 BUG_ON(atomic_read(&sig->count) != 1);
590 exit_itimers(sig);
591 return 0;
592 }
594 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
595 if (!newsighand)
596 return -ENOMEM;
598 if (thread_group_empty(current))
599 goto no_thread_group;
601 /*
602 * Kill all other threads in the thread group.
603 * We must hold tasklist_lock to call zap_other_threads.
604 */
605 read_lock(&tasklist_lock);
606 spin_lock_irq(lock);
607 if (sig->flags & SIGNAL_GROUP_EXIT) {
608 /*
609 * Another group action in progress, just
610 * return so that the signal is processed.
611 */
612 spin_unlock_irq(lock);
613 read_unlock(&tasklist_lock);
614 kmem_cache_free(sighand_cachep, newsighand);
615 return -EAGAIN;
616 }
618 /*
619 * child_reaper ignores SIGKILL, change it now.
620 * Reparenting needs write_lock on tasklist_lock,
621 * so it is safe to do it under read_lock.
622 */
623 if (unlikely(current->group_leader == child_reaper))
624 child_reaper = current;
626 zap_other_threads(current);
627 read_unlock(&tasklist_lock);
629 /*
630 * Account for the thread group leader hanging around:
631 */
632 count = 1;
633 if (!thread_group_leader(current)) {
634 count = 2;
635 /*
636 * The SIGALRM timer survives the exec, but needs to point
637 * at us as the new group leader now. We have a race with
638 * a timer firing now getting the old leader, so we need to
639 * synchronize with any firing (by calling del_timer_sync)
640 * before we can safely let the old group leader die.
641 */
642 sig->tsk = current;
643 spin_unlock_irq(lock);
644 if (hrtimer_cancel(&sig->real_timer))
645 hrtimer_restart(&sig->real_timer);
646 spin_lock_irq(lock);
647 }
648 while (atomic_read(&sig->count) > count) {
649 sig->group_exit_task = current;
650 sig->notify_count = count;
651 __set_current_state(TASK_UNINTERRUPTIBLE);
652 spin_unlock_irq(lock);
653 schedule();
654 spin_lock_irq(lock);
655 }
656 sig->group_exit_task = NULL;
657 sig->notify_count = 0;
658 spin_unlock_irq(lock);
660 /*
661 * At this point all other threads have exited, all we have to
662 * do is to wait for the thread group leader to become inactive,
663 * and to assume its PID:
664 */
665 if (!thread_group_leader(current)) {
666 /*
667 * Wait for the thread group leader to be a zombie.
668 * It should already be zombie at this point, most
669 * of the time.
670 */
671 leader = current->group_leader;
672 while (leader->exit_state != EXIT_ZOMBIE)
673 yield();
675 /*
676 * The only record we have of the real-time age of a
677 * process, regardless of execs it's done, is start_time.
678 * All the past CPU time is accumulated in signal_struct
679 * from sister threads now dead. But in this non-leader
680 * exec, nothing survives from the original leader thread,
681 * whose birth marks the true age of this process now.
682 * When we take on its identity by switching to its PID, we
683 * also take its birthdate (always earlier than our own).
684 */
685 current->start_time = leader->start_time;
687 write_lock_irq(&tasklist_lock);
689 BUG_ON(leader->tgid != current->tgid);
690 BUG_ON(current->pid == current->tgid);
691 /*
692 * An exec() starts a new thread group with the
693 * TGID of the previous thread group. Rehash the
694 * two threads with a switched PID, and release
695 * the former thread group leader:
696 */
698 /* Become a process group leader with the old leader's pid.
699 * Note: The old leader also uses thispid until release_task
700 * is called. Odd but simple and correct.
701 */
702 detach_pid(current, PIDTYPE_PID);
703 current->pid = leader->pid;
704 attach_pid(current, PIDTYPE_PID, current->pid);
705 attach_pid(current, PIDTYPE_PGID, current->signal->pgrp);
706 attach_pid(current, PIDTYPE_SID, current->signal->session);
707 list_replace_rcu(&leader->tasks, &current->tasks);
709 current->group_leader = current;
710 leader->group_leader = current;
712 /* Reduce leader to a thread */
713 detach_pid(leader, PIDTYPE_PGID);
714 detach_pid(leader, PIDTYPE_SID);
716 current->exit_signal = SIGCHLD;
718 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
719 leader->exit_state = EXIT_DEAD;
721 write_unlock_irq(&tasklist_lock);
722 }
724 /*
725 * There may be one thread left which is just exiting,
726 * but it's safe to stop telling the group to kill themselves.
727 */
728 sig->flags = 0;
730 no_thread_group:
731 exit_itimers(sig);
732 if (leader)
733 release_task(leader);
735 BUG_ON(atomic_read(&sig->count) != 1);
737 if (atomic_read(&oldsighand->count) == 1) {
738 /*
739 * Now that we nuked the rest of the thread group,
740 * it turns out we are not sharing sighand any more either.
741 * So we can just keep it.
742 */
743 kmem_cache_free(sighand_cachep, newsighand);
744 } else {
745 /*
746 * Move our state over to newsighand and switch it in.
747 */
748 atomic_set(&newsighand->count, 1);
749 memcpy(newsighand->action, oldsighand->action,
750 sizeof(newsighand->action));
752 write_lock_irq(&tasklist_lock);
753 spin_lock(&oldsighand->siglock);
754 spin_lock_nested(&newsighand->siglock, SINGLE_DEPTH_NESTING);
756 rcu_assign_pointer(current->sighand, newsighand);
757 recalc_sigpending();
759 spin_unlock(&newsighand->siglock);
760 spin_unlock(&oldsighand->siglock);
761 write_unlock_irq(&tasklist_lock);
763 if (atomic_dec_and_test(&oldsighand->count))
764 kmem_cache_free(sighand_cachep, oldsighand);
765 }
767 BUG_ON(!thread_group_leader(current));
768 return 0;
769 }
771 /*
772 * These functions flushes out all traces of the currently running executable
773 * so that a new one can be started
774 */
776 static void flush_old_files(struct files_struct * files)
777 {
778 long j = -1;
779 struct fdtable *fdt;
781 spin_lock(&files->file_lock);
782 for (;;) {
783 unsigned long set, i;
785 j++;
786 i = j * __NFDBITS;
787 fdt = files_fdtable(files);
788 if (i >= fdt->max_fds || i >= fdt->max_fdset)
789 break;
790 set = fdt->close_on_exec->fds_bits[j];
791 if (!set)
792 continue;
793 fdt->close_on_exec->fds_bits[j] = 0;
794 spin_unlock(&files->file_lock);
795 for ( ; set ; i++,set >>= 1) {
796 if (set & 1) {
797 sys_close(i);
798 }
799 }
800 spin_lock(&files->file_lock);
802 }
803 spin_unlock(&files->file_lock);
804 }
806 void get_task_comm(char *buf, struct task_struct *tsk)
807 {
808 /* buf must be at least sizeof(tsk->comm) in size */
809 task_lock(tsk);
810 strncpy(buf, tsk->comm, sizeof(tsk->comm));
811 task_unlock(tsk);
812 }
814 void set_task_comm(struct task_struct *tsk, char *buf)
815 {
816 task_lock(tsk);
817 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
818 task_unlock(tsk);
819 }
821 int flush_old_exec(struct linux_binprm * bprm)
822 {
823 char * name;
824 int i, ch, retval;
825 struct files_struct *files;
826 char tcomm[sizeof(current->comm)];
828 /*
829 * Make sure we have a private signal table and that
830 * we are unassociated from the previous thread group.
831 */
832 retval = de_thread(current);
833 if (retval)
834 goto out;
836 /*
837 * Make sure we have private file handles. Ask the
838 * fork helper to do the work for us and the exit
839 * helper to do the cleanup of the old one.
840 */
841 files = current->files; /* refcounted so safe to hold */
842 retval = unshare_files();
843 if (retval)
844 goto out;
845 /*
846 * Release all of the old mmap stuff
847 */
848 retval = exec_mmap(bprm->mm);
849 if (retval)
850 goto mmap_failed;
852 bprm->mm = NULL; /* We're using it now */
854 /* This is the point of no return */
855 put_files_struct(files);
857 current->sas_ss_sp = current->sas_ss_size = 0;
859 if (current->euid == current->uid && current->egid == current->gid)
860 current->mm->dumpable = 1;
861 else
862 current->mm->dumpable = suid_dumpable;
864 name = bprm->filename;
866 /* Copies the binary name from after last slash */
867 for (i=0; (ch = *(name++)) != '\0';) {
868 if (ch == '/')
869 i = 0; /* overwrite what we wrote */
870 else
871 if (i < (sizeof(tcomm) - 1))
872 tcomm[i++] = ch;
873 }
874 tcomm[i] = '\0';
875 set_task_comm(current, tcomm);
877 current->flags &= ~PF_RANDOMIZE;
878 flush_thread();
880 /* Set the new mm task size. We have to do that late because it may
881 * depend on TIF_32BIT which is only updated in flush_thread() on
882 * some architectures like powerpc
883 */
884 current->mm->task_size = TASK_SIZE;
886 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
887 file_permission(bprm->file, MAY_READ) ||
888 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
889 suid_keys(current);
890 current->mm->dumpable = suid_dumpable;
891 }
893 /* An exec changes our domain. We are no longer part of the thread
894 group */
896 current->self_exec_id++;
898 flush_signal_handlers(current, 0);
899 flush_old_files(current->files);
901 return 0;
903 mmap_failed:
904 put_files_struct(current->files);
905 current->files = files;
906 out:
907 return retval;
908 }
910 EXPORT_SYMBOL(flush_old_exec);
912 /*
913 * Fill the binprm structure from the inode.
914 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
915 */
916 int prepare_binprm(struct linux_binprm *bprm)
917 {
918 int mode;
919 struct inode * inode = bprm->file->f_dentry->d_inode;
920 int retval;
922 mode = inode->i_mode;
923 if (bprm->file->f_op == NULL)
924 return -EACCES;
926 bprm->e_uid = current->euid;
927 bprm->e_gid = current->egid;
929 if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
930 /* Set-uid? */
931 if (mode & S_ISUID) {
932 current->personality &= ~PER_CLEAR_ON_SETID;
933 bprm->e_uid = inode->i_uid;
934 }
936 /* Set-gid? */
937 /*
938 * If setgid is set but no group execute bit then this
939 * is a candidate for mandatory locking, not a setgid
940 * executable.
941 */
942 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
943 current->personality &= ~PER_CLEAR_ON_SETID;
944 bprm->e_gid = inode->i_gid;
945 }
946 }
948 /* fill in binprm security blob */
949 retval = security_bprm_set(bprm);
950 if (retval)
951 return retval;
953 memset(bprm->buf,0,BINPRM_BUF_SIZE);
954 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
955 }
957 EXPORT_SYMBOL(prepare_binprm);
959 static int unsafe_exec(struct task_struct *p)
960 {
961 int unsafe = 0;
962 if (p->ptrace & PT_PTRACED) {
963 if (p->ptrace & PT_PTRACE_CAP)
964 unsafe |= LSM_UNSAFE_PTRACE_CAP;
965 else
966 unsafe |= LSM_UNSAFE_PTRACE;
967 }
968 if (atomic_read(&p->fs->count) > 1 ||
969 atomic_read(&p->files->count) > 1 ||
970 atomic_read(&p->sighand->count) > 1)
971 unsafe |= LSM_UNSAFE_SHARE;
973 return unsafe;
974 }
976 void compute_creds(struct linux_binprm *bprm)
977 {
978 int unsafe;
980 if (bprm->e_uid != current->uid)
981 suid_keys(current);
982 exec_keys(current);
984 task_lock(current);
985 unsafe = unsafe_exec(current);
986 security_bprm_apply_creds(bprm, unsafe);
987 task_unlock(current);
988 security_bprm_post_apply_creds(bprm);
989 }
991 EXPORT_SYMBOL(compute_creds);
993 void remove_arg_zero(struct linux_binprm *bprm)
994 {
995 if (bprm->argc) {
996 unsigned long offset;
997 char * kaddr;
998 struct page *page;
1000 offset = bprm->p % PAGE_SIZE;
1001 goto inside;
1003 while (bprm->p++, *(kaddr+offset++)) {
1004 if (offset != PAGE_SIZE)
1005 continue;
1006 offset = 0;
1007 kunmap_atomic(kaddr, KM_USER0);
1008 inside:
1009 page = bprm->page[bprm->p/PAGE_SIZE];
1010 kaddr = kmap_atomic(page, KM_USER0);
1012 kunmap_atomic(kaddr, KM_USER0);
1013 bprm->argc--;
1017 EXPORT_SYMBOL(remove_arg_zero);
1019 /*
1020 * cycle the list of binary formats handler, until one recognizes the image
1021 */
1022 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1024 int try,retval;
1025 struct linux_binfmt *fmt;
1026 #ifdef __alpha__
1027 /* handle /sbin/loader.. */
1029 struct exec * eh = (struct exec *) bprm->buf;
1031 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1032 (eh->fh.f_flags & 0x3000) == 0x3000)
1034 struct file * file;
1035 unsigned long loader;
1037 allow_write_access(bprm->file);
1038 fput(bprm->file);
1039 bprm->file = NULL;
1041 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1043 file = open_exec("/sbin/loader");
1044 retval = PTR_ERR(file);
1045 if (IS_ERR(file))
1046 return retval;
1048 /* Remember if the application is TASO. */
1049 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1051 bprm->file = file;
1052 bprm->loader = loader;
1053 retval = prepare_binprm(bprm);
1054 if (retval<0)
1055 return retval;
1056 /* should call search_binary_handler recursively here,
1057 but it does not matter */
1060 #endif
1061 retval = security_bprm_check(bprm);
1062 if (retval)
1063 return retval;
1065 /* kernel module loader fixup */
1066 /* so we don't try to load run modprobe in kernel space. */
1067 set_fs(USER_DS);
1069 retval = audit_bprm(bprm);
1070 if (retval)
1071 return retval;
1073 retval = -ENOENT;
1074 for (try=0; try<2; try++) {
1075 read_lock(&binfmt_lock);
1076 for (fmt = formats ; fmt ; fmt = fmt->next) {
1077 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1078 if (!fn)
1079 continue;
1080 if (!try_module_get(fmt->module))
1081 continue;
1082 read_unlock(&binfmt_lock);
1083 retval = fn(bprm, regs);
1084 if (retval >= 0) {
1085 put_binfmt(fmt);
1086 allow_write_access(bprm->file);
1087 if (bprm->file)
1088 fput(bprm->file);
1089 bprm->file = NULL;
1090 current->did_exec = 1;
1091 proc_exec_connector(current);
1092 return retval;
1094 read_lock(&binfmt_lock);
1095 put_binfmt(fmt);
1096 if (retval != -ENOEXEC || bprm->mm == NULL)
1097 break;
1098 if (!bprm->file) {
1099 read_unlock(&binfmt_lock);
1100 return retval;
1103 read_unlock(&binfmt_lock);
1104 if (retval != -ENOEXEC || bprm->mm == NULL) {
1105 break;
1106 #ifdef CONFIG_KMOD
1107 }else{
1108 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1109 if (printable(bprm->buf[0]) &&
1110 printable(bprm->buf[1]) &&
1111 printable(bprm->buf[2]) &&
1112 printable(bprm->buf[3]))
1113 break; /* -ENOEXEC */
1114 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1115 #endif
1118 return retval;
1121 EXPORT_SYMBOL(search_binary_handler);
1123 /*
1124 * sys_execve() executes a new program.
1125 */
1126 int do_execve(char * filename,
1127 char __user *__user *argv,
1128 char __user *__user *envp,
1129 struct pt_regs * regs)
1131 struct linux_binprm *bprm;
1132 struct file *file;
1133 int retval;
1134 int i;
1136 retval = -ENOMEM;
1137 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1138 if (!bprm)
1139 goto out_ret;
1141 file = open_exec(filename);
1142 retval = PTR_ERR(file);
1143 if (IS_ERR(file))
1144 goto out_kfree;
1146 sched_exec();
1148 bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1150 bprm->file = file;
1151 bprm->filename = filename;
1152 bprm->interp = filename;
1153 bprm->mm = mm_alloc();
1154 retval = -ENOMEM;
1155 if (!bprm->mm)
1156 goto out_file;
1158 retval = init_new_context(current, bprm->mm);
1159 if (retval < 0)
1160 goto out_mm;
1162 bprm->argc = count(argv, bprm->p / sizeof(void *));
1163 if ((retval = bprm->argc) < 0)
1164 goto out_mm;
1166 bprm->envc = count(envp, bprm->p / sizeof(void *));
1167 if ((retval = bprm->envc) < 0)
1168 goto out_mm;
1170 retval = security_bprm_alloc(bprm);
1171 if (retval)
1172 goto out;
1174 retval = prepare_binprm(bprm);
1175 if (retval < 0)
1176 goto out;
1178 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1179 if (retval < 0)
1180 goto out;
1182 bprm->exec = bprm->p;
1183 retval = copy_strings(bprm->envc, envp, bprm);
1184 if (retval < 0)
1185 goto out;
1187 retval = copy_strings(bprm->argc, argv, bprm);
1188 if (retval < 0)
1189 goto out;
1191 retval = search_binary_handler(bprm,regs);
1192 if (retval >= 0) {
1193 free_arg_pages(bprm);
1195 /* execve success */
1196 security_bprm_free(bprm);
1197 acct_update_integrals(current);
1198 kfree(bprm);
1199 return retval;
1202 out:
1203 /* Something went wrong, return the inode and free the argument pages*/
1204 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1205 struct page * page = bprm->page[i];
1206 if (page)
1207 __free_page(page);
1210 if (bprm->security)
1211 security_bprm_free(bprm);
1213 out_mm:
1214 if (bprm->mm)
1215 mmdrop(bprm->mm);
1217 out_file:
1218 if (bprm->file) {
1219 allow_write_access(bprm->file);
1220 fput(bprm->file);
1223 out_kfree:
1224 kfree(bprm);
1226 out_ret:
1227 return retval;
1230 int set_binfmt(struct linux_binfmt *new)
1232 struct linux_binfmt *old = current->binfmt;
1234 if (new) {
1235 if (!try_module_get(new->module))
1236 return -1;
1238 current->binfmt = new;
1239 if (old)
1240 module_put(old->module);
1241 return 0;
1244 EXPORT_SYMBOL(set_binfmt);
1246 #define CORENAME_MAX_SIZE 64
1248 /* format_corename will inspect the pattern parameter, and output a
1249 * name into corename, which must have space for at least
1250 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1251 */
1252 static void format_corename(char *corename, const char *pattern, long signr)
1254 const char *pat_ptr = pattern;
1255 char *out_ptr = corename;
1256 char *const out_end = corename + CORENAME_MAX_SIZE;
1257 int rc;
1258 int pid_in_pattern = 0;
1260 /* Repeat as long as we have more pattern to process and more output
1261 space */
1262 while (*pat_ptr) {
1263 if (*pat_ptr != '%') {
1264 if (out_ptr == out_end)
1265 goto out;
1266 *out_ptr++ = *pat_ptr++;
1267 } else {
1268 switch (*++pat_ptr) {
1269 case 0:
1270 goto out;
1271 /* Double percent, output one percent */
1272 case '%':
1273 if (out_ptr == out_end)
1274 goto out;
1275 *out_ptr++ = '%';
1276 break;
1277 /* pid */
1278 case 'p':
1279 pid_in_pattern = 1;
1280 rc = snprintf(out_ptr, out_end - out_ptr,
1281 "%d", current->tgid);
1282 if (rc > out_end - out_ptr)
1283 goto out;
1284 out_ptr += rc;
1285 break;
1286 /* uid */
1287 case 'u':
1288 rc = snprintf(out_ptr, out_end - out_ptr,
1289 "%d", current->uid);
1290 if (rc > out_end - out_ptr)
1291 goto out;
1292 out_ptr += rc;
1293 break;
1294 /* gid */
1295 case 'g':
1296 rc = snprintf(out_ptr, out_end - out_ptr,
1297 "%d", current->gid);
1298 if (rc > out_end - out_ptr)
1299 goto out;
1300 out_ptr += rc;
1301 break;
1302 /* signal that caused the coredump */
1303 case 's':
1304 rc = snprintf(out_ptr, out_end - out_ptr,
1305 "%ld", signr);
1306 if (rc > out_end - out_ptr)
1307 goto out;
1308 out_ptr += rc;
1309 break;
1310 /* UNIX time of coredump */
1311 case 't': {
1312 struct timeval tv;
1313 do_gettimeofday(&tv);
1314 rc = snprintf(out_ptr, out_end - out_ptr,
1315 "%lu", tv.tv_sec);
1316 if (rc > out_end - out_ptr)
1317 goto out;
1318 out_ptr += rc;
1319 break;
1321 /* hostname */
1322 case 'h':
1323 down_read(&uts_sem);
1324 rc = snprintf(out_ptr, out_end - out_ptr,
1325 "%s", system_utsname.nodename);
1326 up_read(&uts_sem);
1327 if (rc > out_end - out_ptr)
1328 goto out;
1329 out_ptr += rc;
1330 break;
1331 /* executable */
1332 case 'e':
1333 rc = snprintf(out_ptr, out_end - out_ptr,
1334 "%s", current->comm);
1335 if (rc > out_end - out_ptr)
1336 goto out;
1337 out_ptr += rc;
1338 break;
1339 default:
1340 break;
1342 ++pat_ptr;
1345 /* Backward compatibility with core_uses_pid:
1347 * If core_pattern does not include a %p (as is the default)
1348 * and core_uses_pid is set, then .%pid will be appended to
1349 * the filename */
1350 if (!pid_in_pattern
1351 && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1352 rc = snprintf(out_ptr, out_end - out_ptr,
1353 ".%d", current->tgid);
1354 if (rc > out_end - out_ptr)
1355 goto out;
1356 out_ptr += rc;
1358 out:
1359 *out_ptr = 0;
1362 static void zap_process(struct task_struct *start)
1364 struct task_struct *t;
1366 start->signal->flags = SIGNAL_GROUP_EXIT;
1367 start->signal->group_stop_count = 0;
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);
1376 } while ((t = next_thread(t)) != start);
1379 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1380 int exit_code)
1382 struct task_struct *g, *p;
1383 unsigned long flags;
1384 int err = -EAGAIN;
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;
1392 spin_unlock_irq(&tsk->sighand->siglock);
1393 if (err)
1394 return err;
1396 if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1397 goto done;
1399 rcu_read_lock();
1400 for_each_process(g) {
1401 if (g == tsk->group_leader)
1402 continue;
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);
1416 break;
1418 } while ((p = next_thread(p)) != g);
1420 rcu_read_unlock();
1421 done:
1422 return mm->core_waiters;
1425 static int coredump_wait(int exit_code)
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;
1433 init_completion(&mm->core_done);
1434 init_completion(&startup_done);
1435 mm->core_startup_done = &startup_done;
1437 core_waiters = zap_threads(tsk, mm, exit_code);
1438 up_write(&mm->mmap_sem);
1440 if (unlikely(core_waiters < 0))
1441 goto fail;
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);
1453 if (core_waiters)
1454 wait_for_completion(&startup_done);
1455 fail:
1456 BUG_ON(mm->core_waiters);
1457 return core_waiters;
1460 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
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;
1471 binfmt = current->binfmt;
1472 if (!binfmt || !binfmt->core_dump)
1473 goto fail;
1474 down_write(&mm->mmap_sem);
1475 if (!mm->dumpable) {
1476 up_write(&mm->mmap_sem);
1477 goto fail;
1480 /*
1481 * We cannot trust fsuid as being the "true" uid of the
1482 * process nor do we know its entire history. We only know it
1483 * was tainted so we dump it as root in mode 2.
1484 */
1485 if (mm->dumpable == 2) { /* Setuid core dump mode */
1486 flag = O_EXCL; /* Stop rewrite attacks */
1487 current->fsuid = 0; /* Dump root private */
1489 mm->dumpable = 0;
1491 retval = coredump_wait(exit_code);
1492 if (retval < 0)
1493 goto fail;
1495 /*
1496 * Clear any false indication of pending signals that might
1497 * be seen by the filesystem code called to write the core file.
1498 */
1499 clear_thread_flag(TIF_SIGPENDING);
1501 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1502 goto fail_unlock;
1504 /*
1505 * lock_kernel() because format_corename() is controlled by sysctl, which
1506 * uses lock_kernel()
1507 */
1508 lock_kernel();
1509 format_corename(corename, core_pattern, signr);
1510 unlock_kernel();
1511 file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
1512 if (IS_ERR(file))
1513 goto fail_unlock;
1514 inode = file->f_dentry->d_inode;
1515 if (inode->i_nlink > 1)
1516 goto close_fail; /* multiple links - don't dump */
1517 if (d_unhashed(file->f_dentry))
1518 goto close_fail;
1520 if (!S_ISREG(inode->i_mode))
1521 goto close_fail;
1522 if (!file->f_op)
1523 goto close_fail;
1524 if (!file->f_op->write)
1525 goto close_fail;
1526 if (do_truncate(file->f_dentry, 0, 0, file) != 0)
1527 goto close_fail;
1529 retval = binfmt->core_dump(signr, regs, file);
1531 if (retval)
1532 current->signal->group_exit_code |= 0x80;
1533 close_fail:
1534 filp_close(file, NULL);
1535 fail_unlock:
1536 current->fsuid = fsuid;
1537 complete_all(&mm->core_done);
1538 fail:
1539 return retval;