blob: 3a4b35a14c0d4504b0b3b8598c9ffa0d50ad6b40 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
52#include <asm/uaccess.h>
53#include <asm/mmu_context.h>
54
55#ifdef CONFIG_KMOD
56#include <linux/kmod.h>
57#endif
58
59int core_uses_pid;
60char core_pattern[65] = "core";
61/* The maximal length of core_pattern is also specified in sysctl.c */
62
63static struct linux_binfmt *formats;
64static DEFINE_RWLOCK(binfmt_lock);
65
66int register_binfmt(struct linux_binfmt * fmt)
67{
68 struct linux_binfmt ** tmp = &formats;
69
70 if (!fmt)
71 return -EINVAL;
72 if (fmt->next)
73 return -EBUSY;
74 write_lock(&binfmt_lock);
75 while (*tmp) {
76 if (fmt == *tmp) {
77 write_unlock(&binfmt_lock);
78 return -EBUSY;
79 }
80 tmp = &(*tmp)->next;
81 }
82 fmt->next = formats;
83 formats = fmt;
84 write_unlock(&binfmt_lock);
85 return 0;
86}
87
88EXPORT_SYMBOL(register_binfmt);
89
90int unregister_binfmt(struct linux_binfmt * fmt)
91{
92 struct linux_binfmt ** tmp = &formats;
93
94 write_lock(&binfmt_lock);
95 while (*tmp) {
96 if (fmt == *tmp) {
97 *tmp = fmt->next;
98 write_unlock(&binfmt_lock);
99 return 0;
100 }
101 tmp = &(*tmp)->next;
102 }
103 write_unlock(&binfmt_lock);
104 return -EINVAL;
105}
106
107EXPORT_SYMBOL(unregister_binfmt);
108
109static inline void put_binfmt(struct linux_binfmt * fmt)
110{
111 module_put(fmt->module);
112}
113
114/*
115 * Note that a shared library must be both readable and executable due to
116 * security reasons.
117 *
118 * Also note that we take the address to load from from the file itself.
119 */
120asmlinkage long sys_uselib(const char __user * library)
121{
122 struct file * file;
123 struct nameidata nd;
124 int error;
125
126 nd.intent.open.flags = FMODE_READ;
127 error = __user_walk(library, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
128 if (error)
129 goto out;
130
131 error = -EINVAL;
132 if (!S_ISREG(nd.dentry->d_inode->i_mode))
133 goto exit;
134
135 error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
136 if (error)
137 goto exit;
138
139 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
140 error = PTR_ERR(file);
141 if (IS_ERR(file))
142 goto out;
143
144 error = -ENOEXEC;
145 if(file->f_op) {
146 struct linux_binfmt * fmt;
147
148 read_lock(&binfmt_lock);
149 for (fmt = formats ; fmt ; fmt = fmt->next) {
150 if (!fmt->load_shlib)
151 continue;
152 if (!try_module_get(fmt->module))
153 continue;
154 read_unlock(&binfmt_lock);
155 error = fmt->load_shlib(file);
156 read_lock(&binfmt_lock);
157 put_binfmt(fmt);
158 if (error != -ENOEXEC)
159 break;
160 }
161 read_unlock(&binfmt_lock);
162 }
163 fput(file);
164out:
165 return error;
166exit:
167 path_release(&nd);
168 goto out;
169}
170
171/*
172 * count() counts the number of strings in array ARGV.
173 */
174static int count(char __user * __user * argv, int max)
175{
176 int i = 0;
177
178 if (argv != NULL) {
179 for (;;) {
180 char __user * p;
181
182 if (get_user(p, argv))
183 return -EFAULT;
184 if (!p)
185 break;
186 argv++;
187 if(++i > max)
188 return -E2BIG;
189 cond_resched();
190 }
191 }
192 return i;
193}
194
195/*
196 * 'copy_strings()' copies argument/environment strings from user
197 * memory to free pages in kernel mem. These are in a format ready
198 * to be put directly into the top of new user memory.
199 */
Adrian Bunk75c96f82005-05-05 16:16:09 -0700200static int copy_strings(int argc, char __user * __user * argv,
201 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
203 struct page *kmapped_page = NULL;
204 char *kaddr = NULL;
205 int ret;
206
207 while (argc-- > 0) {
208 char __user *str;
209 int len;
210 unsigned long pos;
211
212 if (get_user(str, argv+argc) ||
213 !(len = strnlen_user(str, bprm->p))) {
214 ret = -EFAULT;
215 goto out;
216 }
217
218 if (bprm->p < len) {
219 ret = -E2BIG;
220 goto out;
221 }
222
223 bprm->p -= len;
224 /* XXX: add architecture specific overflow check here. */
225 pos = bprm->p;
226
227 while (len > 0) {
228 int i, new, err;
229 int offset, bytes_to_copy;
230 struct page *page;
231
232 offset = pos % PAGE_SIZE;
233 i = pos/PAGE_SIZE;
234 page = bprm->page[i];
235 new = 0;
236 if (!page) {
237 page = alloc_page(GFP_HIGHUSER);
238 bprm->page[i] = page;
239 if (!page) {
240 ret = -ENOMEM;
241 goto out;
242 }
243 new = 1;
244 }
245
246 if (page != kmapped_page) {
247 if (kmapped_page)
248 kunmap(kmapped_page);
249 kmapped_page = page;
250 kaddr = kmap(kmapped_page);
251 }
252 if (new && offset)
253 memset(kaddr, 0, offset);
254 bytes_to_copy = PAGE_SIZE - offset;
255 if (bytes_to_copy > len) {
256 bytes_to_copy = len;
257 if (new)
258 memset(kaddr+offset+len, 0,
259 PAGE_SIZE-offset-len);
260 }
261 err = copy_from_user(kaddr+offset, str, bytes_to_copy);
262 if (err) {
263 ret = -EFAULT;
264 goto out;
265 }
266
267 pos += bytes_to_copy;
268 str += bytes_to_copy;
269 len -= bytes_to_copy;
270 }
271 }
272 ret = 0;
273out:
274 if (kmapped_page)
275 kunmap(kmapped_page);
276 return ret;
277}
278
279/*
280 * Like copy_strings, but get argv and its values from kernel memory.
281 */
282int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
283{
284 int r;
285 mm_segment_t oldfs = get_fs();
286 set_fs(KERNEL_DS);
287 r = copy_strings(argc, (char __user * __user *)argv, bprm);
288 set_fs(oldfs);
289 return r;
290}
291
292EXPORT_SYMBOL(copy_strings_kernel);
293
294#ifdef CONFIG_MMU
295/*
296 * This routine is used to map in a page into an address space: needed by
297 * execve() for the initial stack and environment pages.
298 *
299 * vma->vm_mm->mmap_sem is held for writing.
300 */
301void install_arg_page(struct vm_area_struct *vma,
302 struct page *page, unsigned long address)
303{
304 struct mm_struct *mm = vma->vm_mm;
305 pgd_t * pgd;
306 pud_t * pud;
307 pmd_t * pmd;
308 pte_t * pte;
309
310 if (unlikely(anon_vma_prepare(vma)))
311 goto out_sig;
312
313 flush_dcache_page(page);
314 pgd = pgd_offset(mm, address);
315
316 spin_lock(&mm->page_table_lock);
317 pud = pud_alloc(mm, pgd, address);
318 if (!pud)
319 goto out;
320 pmd = pmd_alloc(mm, pud, address);
321 if (!pmd)
322 goto out;
323 pte = pte_alloc_map(mm, pmd, address);
324 if (!pte)
325 goto out;
326 if (!pte_none(*pte)) {
327 pte_unmap(pte);
328 goto out;
329 }
330 inc_mm_counter(mm, rss);
331 lru_cache_add_active(page);
332 set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
333 page, vma->vm_page_prot))));
334 page_add_anon_rmap(page, vma, address);
335 pte_unmap(pte);
336 spin_unlock(&mm->page_table_lock);
337
338 /* no need for flush_tlb */
339 return;
340out:
341 spin_unlock(&mm->page_table_lock);
342out_sig:
343 __free_page(page);
344 force_sig(SIGKILL, current);
345}
346
347#define EXTRA_STACK_VM_PAGES 20 /* random */
348
349int setup_arg_pages(struct linux_binprm *bprm,
350 unsigned long stack_top,
351 int executable_stack)
352{
353 unsigned long stack_base;
354 struct vm_area_struct *mpnt;
355 struct mm_struct *mm = current->mm;
356 int i, ret;
357 long arg_size;
358
359#ifdef CONFIG_STACK_GROWSUP
360 /* Move the argument and environment strings to the bottom of the
361 * stack space.
362 */
363 int offset, j;
364 char *to, *from;
365
366 /* Start by shifting all the pages down */
367 i = 0;
368 for (j = 0; j < MAX_ARG_PAGES; j++) {
369 struct page *page = bprm->page[j];
370 if (!page)
371 continue;
372 bprm->page[i++] = page;
373 }
374
375 /* Now move them within their pages */
376 offset = bprm->p % PAGE_SIZE;
377 to = kmap(bprm->page[0]);
378 for (j = 1; j < i; j++) {
379 memmove(to, to + offset, PAGE_SIZE - offset);
380 from = kmap(bprm->page[j]);
381 memcpy(to + PAGE_SIZE - offset, from, offset);
382 kunmap(bprm->page[j - 1]);
383 to = from;
384 }
385 memmove(to, to + offset, PAGE_SIZE - offset);
386 kunmap(bprm->page[j - 1]);
387
388 /* Limit stack size to 1GB */
389 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
390 if (stack_base > (1 << 30))
391 stack_base = 1 << 30;
392 stack_base = PAGE_ALIGN(stack_top - stack_base);
393
394 /* Adjust bprm->p to point to the end of the strings. */
395 bprm->p = stack_base + PAGE_SIZE * i - offset;
396
397 mm->arg_start = stack_base;
398 arg_size = i << PAGE_SHIFT;
399
400 /* zero pages that were copied above */
401 while (i < MAX_ARG_PAGES)
402 bprm->page[i++] = NULL;
403#else
404 stack_base = arch_align_stack(stack_top - MAX_ARG_PAGES*PAGE_SIZE);
405 stack_base = PAGE_ALIGN(stack_base);
406 bprm->p += stack_base;
407 mm->arg_start = bprm->p;
408 arg_size = stack_top - (PAGE_MASK & (unsigned long) mm->arg_start);
409#endif
410
411 arg_size += EXTRA_STACK_VM_PAGES * PAGE_SIZE;
412
413 if (bprm->loader)
414 bprm->loader += stack_base;
415 bprm->exec += stack_base;
416
417 mpnt = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
418 if (!mpnt)
419 return -ENOMEM;
420
421 if (security_vm_enough_memory(arg_size >> PAGE_SHIFT)) {
422 kmem_cache_free(vm_area_cachep, mpnt);
423 return -ENOMEM;
424 }
425
426 memset(mpnt, 0, sizeof(*mpnt));
427
428 down_write(&mm->mmap_sem);
429 {
430 mpnt->vm_mm = mm;
431#ifdef CONFIG_STACK_GROWSUP
432 mpnt->vm_start = stack_base;
433 mpnt->vm_end = stack_base + arg_size;
434#else
435 mpnt->vm_end = stack_top;
436 mpnt->vm_start = mpnt->vm_end - arg_size;
437#endif
438 /* Adjust stack execute permissions; explicitly enable
439 * for EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X
440 * and leave alone (arch default) otherwise. */
441 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
442 mpnt->vm_flags = VM_STACK_FLAGS | VM_EXEC;
443 else if (executable_stack == EXSTACK_DISABLE_X)
444 mpnt->vm_flags = VM_STACK_FLAGS & ~VM_EXEC;
445 else
446 mpnt->vm_flags = VM_STACK_FLAGS;
447 mpnt->vm_flags |= mm->def_flags;
448 mpnt->vm_page_prot = protection_map[mpnt->vm_flags & 0x7];
449 if ((ret = insert_vm_struct(mm, mpnt))) {
450 up_write(&mm->mmap_sem);
451 kmem_cache_free(vm_area_cachep, mpnt);
452 return ret;
453 }
454 mm->stack_vm = mm->total_vm = vma_pages(mpnt);
455 }
456
457 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
458 struct page *page = bprm->page[i];
459 if (page) {
460 bprm->page[i] = NULL;
461 install_arg_page(mpnt, page, stack_base);
462 }
463 stack_base += PAGE_SIZE;
464 }
465 up_write(&mm->mmap_sem);
466
467 return 0;
468}
469
470EXPORT_SYMBOL(setup_arg_pages);
471
472#define free_arg_pages(bprm) do { } while (0)
473
474#else
475
476static inline void free_arg_pages(struct linux_binprm *bprm)
477{
478 int i;
479
480 for (i = 0; i < MAX_ARG_PAGES; i++) {
481 if (bprm->page[i])
482 __free_page(bprm->page[i]);
483 bprm->page[i] = NULL;
484 }
485}
486
487#endif /* CONFIG_MMU */
488
489struct file *open_exec(const char *name)
490{
491 struct nameidata nd;
492 int err;
493 struct file *file;
494
495 nd.intent.open.flags = FMODE_READ;
496 err = path_lookup(name, LOOKUP_FOLLOW|LOOKUP_OPEN, &nd);
497 file = ERR_PTR(err);
498
499 if (!err) {
500 struct inode *inode = nd.dentry->d_inode;
501 file = ERR_PTR(-EACCES);
502 if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
503 S_ISREG(inode->i_mode)) {
504 int err = permission(inode, MAY_EXEC, &nd);
505 if (!err && !(inode->i_mode & 0111))
506 err = -EACCES;
507 file = ERR_PTR(err);
508 if (!err) {
509 file = dentry_open(nd.dentry, nd.mnt, O_RDONLY);
510 if (!IS_ERR(file)) {
511 err = deny_write_access(file);
512 if (err) {
513 fput(file);
514 file = ERR_PTR(err);
515 }
516 }
517out:
518 return file;
519 }
520 }
521 path_release(&nd);
522 }
523 goto out;
524}
525
526EXPORT_SYMBOL(open_exec);
527
528int kernel_read(struct file *file, unsigned long offset,
529 char *addr, unsigned long count)
530{
531 mm_segment_t old_fs;
532 loff_t pos = offset;
533 int result;
534
535 old_fs = get_fs();
536 set_fs(get_ds());
537 /* The cast to a user pointer is valid due to the set_fs() */
538 result = vfs_read(file, (void __user *)addr, count, &pos);
539 set_fs(old_fs);
540 return result;
541}
542
543EXPORT_SYMBOL(kernel_read);
544
545static int exec_mmap(struct mm_struct *mm)
546{
547 struct task_struct *tsk;
548 struct mm_struct * old_mm, *active_mm;
549
550 /* Notify parent that we're no longer interested in the old VM */
551 tsk = current;
552 old_mm = current->mm;
553 mm_release(tsk, old_mm);
554
555 if (old_mm) {
556 /*
557 * Make sure that if there is a core dump in progress
558 * for the old mm, we get out and die instead of going
559 * through with the exec. We must hold mmap_sem around
560 * checking core_waiters and changing tsk->mm. The
561 * core-inducing thread will increment core_waiters for
562 * each thread whose ->mm == old_mm.
563 */
564 down_read(&old_mm->mmap_sem);
565 if (unlikely(old_mm->core_waiters)) {
566 up_read(&old_mm->mmap_sem);
567 return -EINTR;
568 }
569 }
570 task_lock(tsk);
571 active_mm = tsk->active_mm;
572 tsk->mm = mm;
573 tsk->active_mm = mm;
574 activate_mm(active_mm, mm);
575 task_unlock(tsk);
576 arch_pick_mmap_layout(mm);
577 if (old_mm) {
578 up_read(&old_mm->mmap_sem);
579 if (active_mm != old_mm) BUG();
580 mmput(old_mm);
581 return 0;
582 }
583 mmdrop(active_mm);
584 return 0;
585}
586
587/*
588 * This function makes sure the current process has its own signal table,
589 * so that flush_signal_handlers can later reset the handlers without
590 * disturbing other processes. (Other processes might share the signal
591 * table via the CLONE_SIGHAND option to clone().)
592 */
593static inline int de_thread(struct task_struct *tsk)
594{
595 struct signal_struct *sig = tsk->signal;
596 struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
597 spinlock_t *lock = &oldsighand->siglock;
598 int count;
599
600 /*
601 * If we don't share sighandlers, then we aren't sharing anything
602 * and we can just re-use it all.
603 */
604 if (atomic_read(&oldsighand->count) <= 1) {
605 BUG_ON(atomic_read(&sig->count) != 1);
606 exit_itimers(sig);
607 return 0;
608 }
609
610 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
611 if (!newsighand)
612 return -ENOMEM;
613
614 if (thread_group_empty(current))
615 goto no_thread_group;
616
617 /*
618 * Kill all other threads in the thread group.
619 * We must hold tasklist_lock to call zap_other_threads.
620 */
621 read_lock(&tasklist_lock);
622 spin_lock_irq(lock);
623 if (sig->flags & SIGNAL_GROUP_EXIT) {
624 /*
625 * Another group action in progress, just
626 * return so that the signal is processed.
627 */
628 spin_unlock_irq(lock);
629 read_unlock(&tasklist_lock);
630 kmem_cache_free(sighand_cachep, newsighand);
631 return -EAGAIN;
632 }
633 zap_other_threads(current);
634 read_unlock(&tasklist_lock);
635
636 /*
637 * Account for the thread group leader hanging around:
638 */
639 count = 2;
640 if (thread_group_leader(current))
641 count = 1;
642 while (atomic_read(&sig->count) > count) {
643 sig->group_exit_task = current;
644 sig->notify_count = count;
645 __set_current_state(TASK_UNINTERRUPTIBLE);
646 spin_unlock_irq(lock);
647 schedule();
648 spin_lock_irq(lock);
649 }
650 sig->group_exit_task = NULL;
651 sig->notify_count = 0;
Linus Torvaldsc2a0f592005-06-18 13:06:22 -0700652 sig->real_timer.data = (unsigned long)current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 spin_unlock_irq(lock);
654
655 /*
656 * At this point all other threads have exited, all we have to
657 * do is to wait for the thread group leader to become inactive,
658 * and to assume its PID:
659 */
660 if (!thread_group_leader(current)) {
661 struct task_struct *leader = current->group_leader, *parent;
662 struct dentry *proc_dentry1, *proc_dentry2;
663 unsigned long exit_state, ptrace;
664
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 while (leader->exit_state != EXIT_ZOMBIE)
671 yield();
672
673 spin_lock(&leader->proc_lock);
674 spin_lock(&current->proc_lock);
675 proc_dentry1 = proc_pid_unhash(current);
676 proc_dentry2 = proc_pid_unhash(leader);
677 write_lock_irq(&tasklist_lock);
678
Linus Torvaldsc2a0f592005-06-18 13:06:22 -0700679 BUG_ON(leader->tgid != current->tgid);
680 BUG_ON(current->pid == current->tgid);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 /*
682 * An exec() starts a new thread group with the
683 * TGID of the previous thread group. Rehash the
684 * two threads with a switched PID, and release
685 * the former thread group leader:
686 */
687 ptrace = leader->ptrace;
688 parent = leader->parent;
689 if (unlikely(ptrace) && unlikely(parent == current)) {
690 /*
691 * Joker was ptracing his own group leader,
692 * and now he wants to be his own parent!
693 * We can't have that.
694 */
695 ptrace = 0;
696 }
697
698 ptrace_unlink(current);
699 ptrace_unlink(leader);
700 remove_parent(current);
701 remove_parent(leader);
702
703 switch_exec_pids(leader, current);
704
705 current->parent = current->real_parent = leader->real_parent;
706 leader->parent = leader->real_parent = child_reaper;
707 current->group_leader = current;
708 leader->group_leader = leader;
709
710 add_parent(current, current->parent);
711 add_parent(leader, leader->parent);
712 if (ptrace) {
713 current->ptrace = ptrace;
714 __ptrace_link(current, parent);
715 }
716
717 list_del(&current->tasks);
718 list_add_tail(&current->tasks, &init_task.tasks);
719 current->exit_signal = SIGCHLD;
720 exit_state = leader->exit_state;
721
722 write_unlock_irq(&tasklist_lock);
723 spin_unlock(&leader->proc_lock);
724 spin_unlock(&current->proc_lock);
725 proc_pid_flush(proc_dentry1);
726 proc_pid_flush(proc_dentry2);
727
Linus Torvaldsc2a0f592005-06-18 13:06:22 -0700728 BUG_ON(exit_state != EXIT_ZOMBIE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 release_task(leader);
730 }
731
732 /*
733 * Now there are really no other threads at all,
734 * so it's safe to stop telling them to kill themselves.
735 */
736 sig->flags = 0;
737
738no_thread_group:
739 BUG_ON(atomic_read(&sig->count) != 1);
740 exit_itimers(sig);
741
742 if (atomic_read(&oldsighand->count) == 1) {
743 /*
744 * Now that we nuked the rest of the thread group,
745 * it turns out we are not sharing sighand any more either.
746 * So we can just keep it.
747 */
748 kmem_cache_free(sighand_cachep, newsighand);
749 } else {
750 /*
751 * Move our state over to newsighand and switch it in.
752 */
753 spin_lock_init(&newsighand->siglock);
754 atomic_set(&newsighand->count, 1);
755 memcpy(newsighand->action, oldsighand->action,
756 sizeof(newsighand->action));
757
758 write_lock_irq(&tasklist_lock);
759 spin_lock(&oldsighand->siglock);
760 spin_lock(&newsighand->siglock);
761
762 current->sighand = newsighand;
763 recalc_sigpending();
764
765 spin_unlock(&newsighand->siglock);
766 spin_unlock(&oldsighand->siglock);
767 write_unlock_irq(&tasklist_lock);
768
769 if (atomic_dec_and_test(&oldsighand->count))
770 kmem_cache_free(sighand_cachep, oldsighand);
771 }
772
Linus Torvaldsc2a0f592005-06-18 13:06:22 -0700773 BUG_ON(!thread_group_empty(current));
774 BUG_ON(!thread_group_leader(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 return 0;
776}
777
778/*
779 * These functions flushes out all traces of the currently running executable
780 * so that a new one can be started
781 */
782
783static inline void flush_old_files(struct files_struct * files)
784{
785 long j = -1;
786
787 spin_lock(&files->file_lock);
788 for (;;) {
789 unsigned long set, i;
790
791 j++;
792 i = j * __NFDBITS;
793 if (i >= files->max_fds || i >= files->max_fdset)
794 break;
795 set = files->close_on_exec->fds_bits[j];
796 if (!set)
797 continue;
798 files->close_on_exec->fds_bits[j] = 0;
799 spin_unlock(&files->file_lock);
800 for ( ; set ; i++,set >>= 1) {
801 if (set & 1) {
802 sys_close(i);
803 }
804 }
805 spin_lock(&files->file_lock);
806
807 }
808 spin_unlock(&files->file_lock);
809}
810
811void get_task_comm(char *buf, struct task_struct *tsk)
812{
813 /* buf must be at least sizeof(tsk->comm) in size */
814 task_lock(tsk);
815 strncpy(buf, tsk->comm, sizeof(tsk->comm));
816 task_unlock(tsk);
817}
818
819void set_task_comm(struct task_struct *tsk, char *buf)
820{
821 task_lock(tsk);
822 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
823 task_unlock(tsk);
824}
825
826int flush_old_exec(struct linux_binprm * bprm)
827{
828 char * name;
829 int i, ch, retval;
830 struct files_struct *files;
831 char tcomm[sizeof(current->comm)];
832
833 /*
834 * Make sure we have a private signal table and that
835 * we are unassociated from the previous thread group.
836 */
837 retval = de_thread(current);
838 if (retval)
839 goto out;
840
841 /*
842 * Make sure we have private file handles. Ask the
843 * fork helper to do the work for us and the exit
844 * helper to do the cleanup of the old one.
845 */
846 files = current->files; /* refcounted so safe to hold */
847 retval = unshare_files();
848 if (retval)
849 goto out;
850 /*
851 * Release all of the old mmap stuff
852 */
853 retval = exec_mmap(bprm->mm);
854 if (retval)
855 goto mmap_failed;
856
857 bprm->mm = NULL; /* We're using it now */
858
859 /* This is the point of no return */
860 steal_locks(files);
861 put_files_struct(files);
862
863 current->sas_ss_sp = current->sas_ss_size = 0;
864
865 if (current->euid == current->uid && current->egid == current->gid)
866 current->mm->dumpable = 1;
867 name = bprm->filename;
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -0700868
869 /* Copies the binary name from after last slash */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 for (i=0; (ch = *(name++)) != '\0';) {
871 if (ch == '/')
Paolo 'Blaisorblade' Giarrusso36772092005-05-05 16:16:12 -0700872 i = 0; /* overwrite what we wrote */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 else
874 if (i < (sizeof(tcomm) - 1))
875 tcomm[i++] = ch;
876 }
877 tcomm[i] = '\0';
878 set_task_comm(current, tcomm);
879
880 current->flags &= ~PF_RANDOMIZE;
881 flush_thread();
882
883 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid ||
884 permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
885 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
886 suid_keys(current);
887 current->mm->dumpable = 0;
888 }
889
890 /* An exec changes our domain. We are no longer part of the thread
891 group */
892
893 current->self_exec_id++;
894
895 flush_signal_handlers(current, 0);
896 flush_old_files(current->files);
897
898 return 0;
899
900mmap_failed:
901 put_files_struct(current->files);
902 current->files = files;
903out:
904 return retval;
905}
906
907EXPORT_SYMBOL(flush_old_exec);
908
909/*
910 * Fill the binprm structure from the inode.
911 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
912 */
913int prepare_binprm(struct linux_binprm *bprm)
914{
915 int mode;
916 struct inode * inode = bprm->file->f_dentry->d_inode;
917 int retval;
918
919 mode = inode->i_mode;
920 /*
921 * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
922 * generic_permission lets a non-executable through
923 */
924 if (!(mode & 0111)) /* with at least _one_ execute bit set */
925 return -EACCES;
926 if (bprm->file->f_op == NULL)
927 return -EACCES;
928
929 bprm->e_uid = current->euid;
930 bprm->e_gid = current->egid;
931
932 if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
933 /* Set-uid? */
934 if (mode & S_ISUID) {
935 current->personality &= ~PER_CLEAR_ON_SETID;
936 bprm->e_uid = inode->i_uid;
937 }
938
939 /* Set-gid? */
940 /*
941 * If setgid is set but no group execute bit then this
942 * is a candidate for mandatory locking, not a setgid
943 * executable.
944 */
945 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
946 current->personality &= ~PER_CLEAR_ON_SETID;
947 bprm->e_gid = inode->i_gid;
948 }
949 }
950
951 /* fill in binprm security blob */
952 retval = security_bprm_set(bprm);
953 if (retval)
954 return retval;
955
956 memset(bprm->buf,0,BINPRM_BUF_SIZE);
957 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
958}
959
960EXPORT_SYMBOL(prepare_binprm);
961
962static inline int unsafe_exec(struct task_struct *p)
963{
964 int unsafe = 0;
965 if (p->ptrace & PT_PTRACED) {
966 if (p->ptrace & PT_PTRACE_CAP)
967 unsafe |= LSM_UNSAFE_PTRACE_CAP;
968 else
969 unsafe |= LSM_UNSAFE_PTRACE;
970 }
971 if (atomic_read(&p->fs->count) > 1 ||
972 atomic_read(&p->files->count) > 1 ||
973 atomic_read(&p->sighand->count) > 1)
974 unsafe |= LSM_UNSAFE_SHARE;
975
976 return unsafe;
977}
978
979void compute_creds(struct linux_binprm *bprm)
980{
981 int unsafe;
982
983 if (bprm->e_uid != current->uid)
984 suid_keys(current);
985 exec_keys(current);
986
987 task_lock(current);
988 unsafe = unsafe_exec(current);
989 security_bprm_apply_creds(bprm, unsafe);
990 task_unlock(current);
991 security_bprm_post_apply_creds(bprm);
992}
993
994EXPORT_SYMBOL(compute_creds);
995
996void remove_arg_zero(struct linux_binprm *bprm)
997{
998 if (bprm->argc) {
999 unsigned long offset;
1000 char * kaddr;
1001 struct page *page;
1002
1003 offset = bprm->p % PAGE_SIZE;
1004 goto inside;
1005
1006 while (bprm->p++, *(kaddr+offset++)) {
1007 if (offset != PAGE_SIZE)
1008 continue;
1009 offset = 0;
1010 kunmap_atomic(kaddr, KM_USER0);
1011inside:
1012 page = bprm->page[bprm->p/PAGE_SIZE];
1013 kaddr = kmap_atomic(page, KM_USER0);
1014 }
1015 kunmap_atomic(kaddr, KM_USER0);
1016 bprm->argc--;
1017 }
1018}
1019
1020EXPORT_SYMBOL(remove_arg_zero);
1021
1022/*
1023 * cycle the list of binary formats handler, until one recognizes the image
1024 */
1025int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1026{
1027 int try,retval;
1028 struct linux_binfmt *fmt;
1029#ifdef __alpha__
1030 /* handle /sbin/loader.. */
1031 {
1032 struct exec * eh = (struct exec *) bprm->buf;
1033
1034 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1035 (eh->fh.f_flags & 0x3000) == 0x3000)
1036 {
1037 struct file * file;
1038 unsigned long loader;
1039
1040 allow_write_access(bprm->file);
1041 fput(bprm->file);
1042 bprm->file = NULL;
1043
1044 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1045
1046 file = open_exec("/sbin/loader");
1047 retval = PTR_ERR(file);
1048 if (IS_ERR(file))
1049 return retval;
1050
1051 /* Remember if the application is TASO. */
1052 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1053
1054 bprm->file = file;
1055 bprm->loader = loader;
1056 retval = prepare_binprm(bprm);
1057 if (retval<0)
1058 return retval;
1059 /* should call search_binary_handler recursively here,
1060 but it does not matter */
1061 }
1062 }
1063#endif
1064 retval = security_bprm_check(bprm);
1065 if (retval)
1066 return retval;
1067
1068 /* kernel module loader fixup */
1069 /* so we don't try to load run modprobe in kernel space. */
1070 set_fs(USER_DS);
1071 retval = -ENOENT;
1072 for (try=0; try<2; try++) {
1073 read_lock(&binfmt_lock);
1074 for (fmt = formats ; fmt ; fmt = fmt->next) {
1075 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1076 if (!fn)
1077 continue;
1078 if (!try_module_get(fmt->module))
1079 continue;
1080 read_unlock(&binfmt_lock);
1081 retval = fn(bprm, regs);
1082 if (retval >= 0) {
1083 put_binfmt(fmt);
1084 allow_write_access(bprm->file);
1085 if (bprm->file)
1086 fput(bprm->file);
1087 bprm->file = NULL;
1088 current->did_exec = 1;
1089 return retval;
1090 }
1091 read_lock(&binfmt_lock);
1092 put_binfmt(fmt);
1093 if (retval != -ENOEXEC || bprm->mm == NULL)
1094 break;
1095 if (!bprm->file) {
1096 read_unlock(&binfmt_lock);
1097 return retval;
1098 }
1099 }
1100 read_unlock(&binfmt_lock);
1101 if (retval != -ENOEXEC || bprm->mm == NULL) {
1102 break;
1103#ifdef CONFIG_KMOD
1104 }else{
1105#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1106 if (printable(bprm->buf[0]) &&
1107 printable(bprm->buf[1]) &&
1108 printable(bprm->buf[2]) &&
1109 printable(bprm->buf[3]))
1110 break; /* -ENOEXEC */
1111 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1112#endif
1113 }
1114 }
1115 return retval;
1116}
1117
1118EXPORT_SYMBOL(search_binary_handler);
1119
1120/*
1121 * sys_execve() executes a new program.
1122 */
1123int do_execve(char * filename,
1124 char __user *__user *argv,
1125 char __user *__user *envp,
1126 struct pt_regs * regs)
1127{
1128 struct linux_binprm *bprm;
1129 struct file *file;
1130 int retval;
1131 int i;
1132
1133 retval = -ENOMEM;
1134 bprm = kmalloc(sizeof(*bprm), GFP_KERNEL);
1135 if (!bprm)
1136 goto out_ret;
1137 memset(bprm, 0, sizeof(*bprm));
1138
1139 file = open_exec(filename);
1140 retval = PTR_ERR(file);
1141 if (IS_ERR(file))
1142 goto out_kfree;
1143
1144 sched_exec();
1145
1146 bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1147
1148 bprm->file = file;
1149 bprm->filename = filename;
1150 bprm->interp = filename;
1151 bprm->mm = mm_alloc();
1152 retval = -ENOMEM;
1153 if (!bprm->mm)
1154 goto out_file;
1155
1156 retval = init_new_context(current, bprm->mm);
1157 if (retval < 0)
1158 goto out_mm;
1159
1160 bprm->argc = count(argv, bprm->p / sizeof(void *));
1161 if ((retval = bprm->argc) < 0)
1162 goto out_mm;
1163
1164 bprm->envc = count(envp, bprm->p / sizeof(void *));
1165 if ((retval = bprm->envc) < 0)
1166 goto out_mm;
1167
1168 retval = security_bprm_alloc(bprm);
1169 if (retval)
1170 goto out;
1171
1172 retval = prepare_binprm(bprm);
1173 if (retval < 0)
1174 goto out;
1175
1176 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1177 if (retval < 0)
1178 goto out;
1179
1180 bprm->exec = bprm->p;
1181 retval = copy_strings(bprm->envc, envp, bprm);
1182 if (retval < 0)
1183 goto out;
1184
1185 retval = copy_strings(bprm->argc, argv, bprm);
1186 if (retval < 0)
1187 goto out;
1188
1189 retval = search_binary_handler(bprm,regs);
1190 if (retval >= 0) {
1191 free_arg_pages(bprm);
1192
1193 /* execve success */
1194 security_bprm_free(bprm);
1195 acct_update_integrals(current);
1196 update_mem_hiwater(current);
1197 kfree(bprm);
1198 return retval;
1199 }
1200
1201out:
1202 /* Something went wrong, return the inode and free the argument pages*/
1203 for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1204 struct page * page = bprm->page[i];
1205 if (page)
1206 __free_page(page);
1207 }
1208
1209 if (bprm->security)
1210 security_bprm_free(bprm);
1211
1212out_mm:
1213 if (bprm->mm)
1214 mmdrop(bprm->mm);
1215
1216out_file:
1217 if (bprm->file) {
1218 allow_write_access(bprm->file);
1219 fput(bprm->file);
1220 }
1221
1222out_kfree:
1223 kfree(bprm);
1224
1225out_ret:
1226 return retval;
1227}
1228
1229int set_binfmt(struct linux_binfmt *new)
1230{
1231 struct linux_binfmt *old = current->binfmt;
1232
1233 if (new) {
1234 if (!try_module_get(new->module))
1235 return -1;
1236 }
1237 current->binfmt = new;
1238 if (old)
1239 module_put(old->module);
1240 return 0;
1241}
1242
1243EXPORT_SYMBOL(set_binfmt);
1244
1245#define CORENAME_MAX_SIZE 64
1246
1247/* format_corename will inspect the pattern parameter, and output a
1248 * name into corename, which must have space for at least
1249 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1250 */
1251static void format_corename(char *corename, const char *pattern, long signr)
1252{
1253 const char *pat_ptr = pattern;
1254 char *out_ptr = corename;
1255 char *const out_end = corename + CORENAME_MAX_SIZE;
1256 int rc;
1257 int pid_in_pattern = 0;
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", system_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 */
1349 if (!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 }
1357 out:
1358 *out_ptr = 0;
1359}
1360
1361static void zap_threads (struct mm_struct *mm)
1362{
1363 struct task_struct *g, *p;
1364 struct task_struct *tsk = current;
1365 struct completion *vfork_done = tsk->vfork_done;
1366 int traced = 0;
1367
1368 /*
1369 * Make sure nobody is waiting for us to release the VM,
1370 * otherwise we can deadlock when we wait on each other
1371 */
1372 if (vfork_done) {
1373 tsk->vfork_done = NULL;
1374 complete(vfork_done);
1375 }
1376
1377 read_lock(&tasklist_lock);
1378 do_each_thread(g,p)
1379 if (mm == p->mm && p != tsk) {
1380 force_sig_specific(SIGKILL, p);
1381 mm->core_waiters++;
1382 if (unlikely(p->ptrace) &&
1383 unlikely(p->parent->mm == mm))
1384 traced = 1;
1385 }
1386 while_each_thread(g,p);
1387
1388 read_unlock(&tasklist_lock);
1389
1390 if (unlikely(traced)) {
1391 /*
1392 * We are zapping a thread and the thread it ptraces.
1393 * If the tracee went into a ptrace stop for exit tracing,
1394 * we could deadlock since the tracer is waiting for this
1395 * coredump to finish. Detach them so they can both die.
1396 */
1397 write_lock_irq(&tasklist_lock);
1398 do_each_thread(g,p) {
1399 if (mm == p->mm && p != tsk &&
1400 p->ptrace && p->parent->mm == mm) {
1401 __ptrace_unlink(p);
1402 }
1403 } while_each_thread(g,p);
1404 write_unlock_irq(&tasklist_lock);
1405 }
1406}
1407
1408static void coredump_wait(struct mm_struct *mm)
1409{
1410 DECLARE_COMPLETION(startup_done);
1411
1412 mm->core_waiters++; /* let other threads block */
1413 mm->core_startup_done = &startup_done;
1414
1415 /* give other threads a chance to run: */
1416 yield();
1417
1418 zap_threads(mm);
1419 if (--mm->core_waiters) {
1420 up_write(&mm->mmap_sem);
1421 wait_for_completion(&startup_done);
1422 } else
1423 up_write(&mm->mmap_sem);
1424 BUG_ON(mm->core_waiters);
1425}
1426
1427int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1428{
1429 char corename[CORENAME_MAX_SIZE + 1];
1430 struct mm_struct *mm = current->mm;
1431 struct linux_binfmt * binfmt;
1432 struct inode * inode;
1433 struct file * file;
1434 int retval = 0;
1435
1436 binfmt = current->binfmt;
1437 if (!binfmt || !binfmt->core_dump)
1438 goto fail;
1439 down_write(&mm->mmap_sem);
1440 if (!mm->dumpable) {
1441 up_write(&mm->mmap_sem);
1442 goto fail;
1443 }
1444 mm->dumpable = 0;
1445 init_completion(&mm->core_done);
1446 spin_lock_irq(&current->sighand->siglock);
1447 current->signal->flags = SIGNAL_GROUP_EXIT;
1448 current->signal->group_exit_code = exit_code;
1449 spin_unlock_irq(&current->sighand->siglock);
1450 coredump_wait(mm);
1451
1452 /*
1453 * Clear any false indication of pending signals that might
1454 * be seen by the filesystem code called to write the core file.
1455 */
1456 current->signal->group_stop_count = 0;
1457 clear_thread_flag(TIF_SIGPENDING);
1458
1459 if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1460 goto fail_unlock;
1461
1462 /*
1463 * lock_kernel() because format_corename() is controlled by sysctl, which
1464 * uses lock_kernel()
1465 */
1466 lock_kernel();
1467 format_corename(corename, core_pattern, signr);
1468 unlock_kernel();
1469 file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE, 0600);
1470 if (IS_ERR(file))
1471 goto fail_unlock;
1472 inode = file->f_dentry->d_inode;
1473 if (inode->i_nlink > 1)
1474 goto close_fail; /* multiple links - don't dump */
1475 if (d_unhashed(file->f_dentry))
1476 goto close_fail;
1477
1478 if (!S_ISREG(inode->i_mode))
1479 goto close_fail;
1480 if (!file->f_op)
1481 goto close_fail;
1482 if (!file->f_op->write)
1483 goto close_fail;
1484 if (do_truncate(file->f_dentry, 0) != 0)
1485 goto close_fail;
1486
1487 retval = binfmt->core_dump(signr, regs, file);
1488
1489 if (retval)
1490 current->signal->group_exit_code |= 0x80;
1491close_fail:
1492 filp_close(file, NULL);
1493fail_unlock:
1494 complete_all(&mm->core_done);
1495fail:
1496 return retval;
1497}