[PATCH] de_thread: fix lockless do_each_thread
[linux-2.6] / 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_replace_rcu(&leader->tasks, &current->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
719                 current->exit_signal = SIGCHLD;
720
721                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
722                 leader->exit_state = EXIT_DEAD;
723
724                 write_unlock_irq(&tasklist_lock);
725         }
726
727         /*
728          * There may be one thread left which is just exiting,
729          * but it's safe to stop telling the group to kill themselves.
730          */
731         sig->flags = 0;
732
733 no_thread_group:
734         exit_itimers(sig);
735         if (leader)
736                 release_task(leader);
737
738         BUG_ON(atomic_read(&sig->count) != 1);
739
740         if (atomic_read(&oldsighand->count) == 1) {
741                 /*
742                  * Now that we nuked the rest of the thread group,
743                  * it turns out we are not sharing sighand any more either.
744                  * So we can just keep it.
745                  */
746                 kmem_cache_free(sighand_cachep, newsighand);
747         } else {
748                 /*
749                  * Move our state over to newsighand and switch it in.
750                  */
751                 atomic_set(&newsighand->count, 1);
752                 memcpy(newsighand->action, oldsighand->action,
753                        sizeof(newsighand->action));
754
755                 write_lock_irq(&tasklist_lock);
756                 spin_lock(&oldsighand->siglock);
757                 spin_lock(&newsighand->siglock);
758
759                 rcu_assign_pointer(current->sighand, newsighand);
760                 recalc_sigpending();
761
762                 spin_unlock(&newsighand->siglock);
763                 spin_unlock(&oldsighand->siglock);
764                 write_unlock_irq(&tasklist_lock);
765
766                 if (atomic_dec_and_test(&oldsighand->count))
767                         kmem_cache_free(sighand_cachep, oldsighand);
768         }
769
770         BUG_ON(!thread_group_leader(current));
771         return 0;
772 }
773         
774 /*
775  * These functions flushes out all traces of the currently running executable
776  * so that a new one can be started
777  */
778
779 static void flush_old_files(struct files_struct * files)
780 {
781         long j = -1;
782         struct fdtable *fdt;
783
784         spin_lock(&files->file_lock);
785         for (;;) {
786                 unsigned long set, i;
787
788                 j++;
789                 i = j * __NFDBITS;
790                 fdt = files_fdtable(files);
791                 if (i >= fdt->max_fds || i >= fdt->max_fdset)
792                         break;
793                 set = fdt->close_on_exec->fds_bits[j];
794                 if (!set)
795                         continue;
796                 fdt->close_on_exec->fds_bits[j] = 0;
797                 spin_unlock(&files->file_lock);
798                 for ( ; set ; i++,set >>= 1) {
799                         if (set & 1) {
800                                 sys_close(i);
801                         }
802                 }
803                 spin_lock(&files->file_lock);
804
805         }
806         spin_unlock(&files->file_lock);
807 }
808
809 void get_task_comm(char *buf, struct task_struct *tsk)
810 {
811         /* buf must be at least sizeof(tsk->comm) in size */
812         task_lock(tsk);
813         strncpy(buf, tsk->comm, sizeof(tsk->comm));
814         task_unlock(tsk);
815 }
816
817 void set_task_comm(struct task_struct *tsk, char *buf)
818 {
819         task_lock(tsk);
820         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
821         task_unlock(tsk);
822 }
823
824 int flush_old_exec(struct linux_binprm * bprm)
825 {
826         char * name;
827         int i, ch, retval;
828         struct files_struct *files;
829         char tcomm[sizeof(current->comm)];
830
831         /*
832          * Make sure we have a private signal table and that
833          * we are unassociated from the previous thread group.
834          */
835         retval = de_thread(current);
836         if (retval)
837                 goto out;
838
839         /*
840          * Make sure we have private file handles. Ask the
841          * fork helper to do the work for us and the exit
842          * helper to do the cleanup of the old one.
843          */
844         files = current->files;         /* refcounted so safe to hold */
845         retval = unshare_files();
846         if (retval)
847                 goto out;
848         /*
849          * Release all of the old mmap stuff
850          */
851         retval = exec_mmap(bprm->mm);
852         if (retval)
853                 goto mmap_failed;
854
855         bprm->mm = NULL;                /* We're using it now */
856
857         /* This is the point of no return */
858         put_files_struct(files);
859
860         current->sas_ss_sp = current->sas_ss_size = 0;
861
862         if (current->euid == current->uid && current->egid == current->gid)
863                 current->mm->dumpable = 1;
864         else
865                 current->mm->dumpable = suid_dumpable;
866
867         name = bprm->filename;
868
869         /* Copies the binary name from after last slash */
870         for (i=0; (ch = *(name++)) != '\0';) {
871                 if (ch == '/')
872                         i = 0; /* overwrite what we wrote */
873                 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         /* Set the new mm task size. We have to do that late because it may
884          * depend on TIF_32BIT which is only updated in flush_thread() on
885          * some architectures like powerpc
886          */
887         current->mm->task_size = TASK_SIZE;
888
889         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid || 
890             file_permission(bprm->file, MAY_READ) ||
891             (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
892                 suid_keys(current);
893                 current->mm->dumpable = suid_dumpable;
894         }
895
896         /* An exec changes our domain. We are no longer part of the thread
897            group */
898
899         current->self_exec_id++;
900                         
901         flush_signal_handlers(current, 0);
902         flush_old_files(current->files);
903
904         return 0;
905
906 mmap_failed:
907         put_files_struct(current->files);
908         current->files = files;
909 out:
910         return retval;
911 }
912
913 EXPORT_SYMBOL(flush_old_exec);
914
915 /* 
916  * Fill the binprm structure from the inode. 
917  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
918  */
919 int prepare_binprm(struct linux_binprm *bprm)
920 {
921         int mode;
922         struct inode * inode = bprm->file->f_dentry->d_inode;
923         int retval;
924
925         mode = inode->i_mode;
926         /*
927          * Check execute perms again - if the caller has CAP_DAC_OVERRIDE,
928          * generic_permission lets a non-executable through
929          */
930         if (!(mode & 0111))     /* with at least _one_ execute bit set */
931                 return -EACCES;
932         if (bprm->file->f_op == NULL)
933                 return -EACCES;
934
935         bprm->e_uid = current->euid;
936         bprm->e_gid = current->egid;
937
938         if(!(bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)) {
939                 /* Set-uid? */
940                 if (mode & S_ISUID) {
941                         current->personality &= ~PER_CLEAR_ON_SETID;
942                         bprm->e_uid = inode->i_uid;
943                 }
944
945                 /* Set-gid? */
946                 /*
947                  * If setgid is set but no group execute bit then this
948                  * is a candidate for mandatory locking, not a setgid
949                  * executable.
950                  */
951                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
952                         current->personality &= ~PER_CLEAR_ON_SETID;
953                         bprm->e_gid = inode->i_gid;
954                 }
955         }
956
957         /* fill in binprm security blob */
958         retval = security_bprm_set(bprm);
959         if (retval)
960                 return retval;
961
962         memset(bprm->buf,0,BINPRM_BUF_SIZE);
963         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
964 }
965
966 EXPORT_SYMBOL(prepare_binprm);
967
968 static int unsafe_exec(struct task_struct *p)
969 {
970         int unsafe = 0;
971         if (p->ptrace & PT_PTRACED) {
972                 if (p->ptrace & PT_PTRACE_CAP)
973                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
974                 else
975                         unsafe |= LSM_UNSAFE_PTRACE;
976         }
977         if (atomic_read(&p->fs->count) > 1 ||
978             atomic_read(&p->files->count) > 1 ||
979             atomic_read(&p->sighand->count) > 1)
980                 unsafe |= LSM_UNSAFE_SHARE;
981
982         return unsafe;
983 }
984
985 void compute_creds(struct linux_binprm *bprm)
986 {
987         int unsafe;
988
989         if (bprm->e_uid != current->uid)
990                 suid_keys(current);
991         exec_keys(current);
992
993         task_lock(current);
994         unsafe = unsafe_exec(current);
995         security_bprm_apply_creds(bprm, unsafe);
996         task_unlock(current);
997         security_bprm_post_apply_creds(bprm);
998 }
999
1000 EXPORT_SYMBOL(compute_creds);
1001
1002 void remove_arg_zero(struct linux_binprm *bprm)
1003 {
1004         if (bprm->argc) {
1005                 unsigned long offset;
1006                 char * kaddr;
1007                 struct page *page;
1008
1009                 offset = bprm->p % PAGE_SIZE;
1010                 goto inside;
1011
1012                 while (bprm->p++, *(kaddr+offset++)) {
1013                         if (offset != PAGE_SIZE)
1014                                 continue;
1015                         offset = 0;
1016                         kunmap_atomic(kaddr, KM_USER0);
1017 inside:
1018                         page = bprm->page[bprm->p/PAGE_SIZE];
1019                         kaddr = kmap_atomic(page, KM_USER0);
1020                 }
1021                 kunmap_atomic(kaddr, KM_USER0);
1022                 bprm->argc--;
1023         }
1024 }
1025
1026 EXPORT_SYMBOL(remove_arg_zero);
1027
1028 /*
1029  * cycle the list of binary formats handler, until one recognizes the image
1030  */
1031 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1032 {
1033         int try,retval;
1034         struct linux_binfmt *fmt;
1035 #ifdef __alpha__
1036         /* handle /sbin/loader.. */
1037         {
1038             struct exec * eh = (struct exec *) bprm->buf;
1039
1040             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1041                 (eh->fh.f_flags & 0x3000) == 0x3000)
1042             {
1043                 struct file * file;
1044                 unsigned long loader;
1045
1046                 allow_write_access(bprm->file);
1047                 fput(bprm->file);
1048                 bprm->file = NULL;
1049
1050                 loader = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1051
1052                 file = open_exec("/sbin/loader");
1053                 retval = PTR_ERR(file);
1054                 if (IS_ERR(file))
1055                         return retval;
1056
1057                 /* Remember if the application is TASO.  */
1058                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1059
1060                 bprm->file = file;
1061                 bprm->loader = loader;
1062                 retval = prepare_binprm(bprm);
1063                 if (retval<0)
1064                         return retval;
1065                 /* should call search_binary_handler recursively here,
1066                    but it does not matter */
1067             }
1068         }
1069 #endif
1070         retval = security_bprm_check(bprm);
1071         if (retval)
1072                 return retval;
1073
1074         /* kernel module loader fixup */
1075         /* so we don't try to load run modprobe in kernel space. */
1076         set_fs(USER_DS);
1077
1078         retval = audit_bprm(bprm);
1079         if (retval)
1080                 return retval;
1081
1082         retval = -ENOENT;
1083         for (try=0; try<2; try++) {
1084                 read_lock(&binfmt_lock);
1085                 for (fmt = formats ; fmt ; fmt = fmt->next) {
1086                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1087                         if (!fn)
1088                                 continue;
1089                         if (!try_module_get(fmt->module))
1090                                 continue;
1091                         read_unlock(&binfmt_lock);
1092                         retval = fn(bprm, regs);
1093                         if (retval >= 0) {
1094                                 put_binfmt(fmt);
1095                                 allow_write_access(bprm->file);
1096                                 if (bprm->file)
1097                                         fput(bprm->file);
1098                                 bprm->file = NULL;
1099                                 current->did_exec = 1;
1100                                 proc_exec_connector(current);
1101                                 return retval;
1102                         }
1103                         read_lock(&binfmt_lock);
1104                         put_binfmt(fmt);
1105                         if (retval != -ENOEXEC || bprm->mm == NULL)
1106                                 break;
1107                         if (!bprm->file) {
1108                                 read_unlock(&binfmt_lock);
1109                                 return retval;
1110                         }
1111                 }
1112                 read_unlock(&binfmt_lock);
1113                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1114                         break;
1115 #ifdef CONFIG_KMOD
1116                 }else{
1117 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1118                         if (printable(bprm->buf[0]) &&
1119                             printable(bprm->buf[1]) &&
1120                             printable(bprm->buf[2]) &&
1121                             printable(bprm->buf[3]))
1122                                 break; /* -ENOEXEC */
1123                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1124 #endif
1125                 }
1126         }
1127         return retval;
1128 }
1129
1130 EXPORT_SYMBOL(search_binary_handler);
1131
1132 /*
1133  * sys_execve() executes a new program.
1134  */
1135 int do_execve(char * filename,
1136         char __user *__user *argv,
1137         char __user *__user *envp,
1138         struct pt_regs * regs)
1139 {
1140         struct linux_binprm *bprm;
1141         struct file *file;
1142         int retval;
1143         int i;
1144
1145         retval = -ENOMEM;
1146         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1147         if (!bprm)
1148                 goto out_ret;
1149
1150         file = open_exec(filename);
1151         retval = PTR_ERR(file);
1152         if (IS_ERR(file))
1153                 goto out_kfree;
1154
1155         sched_exec();
1156
1157         bprm->p = PAGE_SIZE*MAX_ARG_PAGES-sizeof(void *);
1158
1159         bprm->file = file;
1160         bprm->filename = filename;
1161         bprm->interp = filename;
1162         bprm->mm = mm_alloc();
1163         retval = -ENOMEM;
1164         if (!bprm->mm)
1165                 goto out_file;
1166
1167         retval = init_new_context(current, bprm->mm);
1168         if (retval < 0)
1169                 goto out_mm;
1170
1171         bprm->argc = count(argv, bprm->p / sizeof(void *));
1172         if ((retval = bprm->argc) < 0)
1173                 goto out_mm;
1174
1175         bprm->envc = count(envp, bprm->p / sizeof(void *));
1176         if ((retval = bprm->envc) < 0)
1177                 goto out_mm;
1178
1179         retval = security_bprm_alloc(bprm);
1180         if (retval)
1181                 goto out;
1182
1183         retval = prepare_binprm(bprm);
1184         if (retval < 0)
1185                 goto out;
1186
1187         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1188         if (retval < 0)
1189                 goto out;
1190
1191         bprm->exec = bprm->p;
1192         retval = copy_strings(bprm->envc, envp, bprm);
1193         if (retval < 0)
1194                 goto out;
1195
1196         retval = copy_strings(bprm->argc, argv, bprm);
1197         if (retval < 0)
1198                 goto out;
1199
1200         retval = search_binary_handler(bprm,regs);
1201         if (retval >= 0) {
1202                 free_arg_pages(bprm);
1203
1204                 /* execve success */
1205                 security_bprm_free(bprm);
1206                 acct_update_integrals(current);
1207                 kfree(bprm);
1208                 return retval;
1209         }
1210
1211 out:
1212         /* Something went wrong, return the inode and free the argument pages*/
1213         for (i = 0 ; i < MAX_ARG_PAGES ; i++) {
1214                 struct page * page = bprm->page[i];
1215                 if (page)
1216                         __free_page(page);
1217         }
1218
1219         if (bprm->security)
1220                 security_bprm_free(bprm);
1221
1222 out_mm:
1223         if (bprm->mm)
1224                 mmdrop(bprm->mm);
1225
1226 out_file:
1227         if (bprm->file) {
1228                 allow_write_access(bprm->file);
1229                 fput(bprm->file);
1230         }
1231
1232 out_kfree:
1233         kfree(bprm);
1234
1235 out_ret:
1236         return retval;
1237 }
1238
1239 int set_binfmt(struct linux_binfmt *new)
1240 {
1241         struct linux_binfmt *old = current->binfmt;
1242
1243         if (new) {
1244                 if (!try_module_get(new->module))
1245                         return -1;
1246         }
1247         current->binfmt = new;
1248         if (old)
1249                 module_put(old->module);
1250         return 0;
1251 }
1252
1253 EXPORT_SYMBOL(set_binfmt);
1254
1255 #define CORENAME_MAX_SIZE 64
1256
1257 /* format_corename will inspect the pattern parameter, and output a
1258  * name into corename, which must have space for at least
1259  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1260  */
1261 static void format_corename(char *corename, const char *pattern, long signr)
1262 {
1263         const char *pat_ptr = pattern;
1264         char *out_ptr = corename;
1265         char *const out_end = corename + CORENAME_MAX_SIZE;
1266         int rc;
1267         int pid_in_pattern = 0;
1268
1269         /* Repeat as long as we have more pattern to process and more output
1270            space */
1271         while (*pat_ptr) {
1272                 if (*pat_ptr != '%') {
1273                         if (out_ptr == out_end)
1274                                 goto out;
1275                         *out_ptr++ = *pat_ptr++;
1276                 } else {
1277                         switch (*++pat_ptr) {
1278                         case 0:
1279                                 goto out;
1280                         /* Double percent, output one percent */
1281                         case '%':
1282                                 if (out_ptr == out_end)
1283                                         goto out;
1284                                 *out_ptr++ = '%';
1285                                 break;
1286                         /* pid */
1287                         case 'p':
1288                                 pid_in_pattern = 1;
1289                                 rc = snprintf(out_ptr, out_end - out_ptr,
1290                                               "%d", current->tgid);
1291                                 if (rc > out_end - out_ptr)
1292                                         goto out;
1293                                 out_ptr += rc;
1294                                 break;
1295                         /* uid */
1296                         case 'u':
1297                                 rc = snprintf(out_ptr, out_end - out_ptr,
1298                                               "%d", current->uid);
1299                                 if (rc > out_end - out_ptr)
1300                                         goto out;
1301                                 out_ptr += rc;
1302                                 break;
1303                         /* gid */
1304                         case 'g':
1305                                 rc = snprintf(out_ptr, out_end - out_ptr,
1306                                               "%d", current->gid);
1307                                 if (rc > out_end - out_ptr)
1308                                         goto out;
1309                                 out_ptr += rc;
1310                                 break;
1311                         /* signal that caused the coredump */
1312                         case 's':
1313                                 rc = snprintf(out_ptr, out_end - out_ptr,
1314                                               "%ld", signr);
1315                                 if (rc > out_end - out_ptr)
1316                                         goto out;
1317                                 out_ptr += rc;
1318                                 break;
1319                         /* UNIX time of coredump */
1320                         case 't': {
1321                                 struct timeval tv;
1322                                 do_gettimeofday(&tv);
1323                                 rc = snprintf(out_ptr, out_end - out_ptr,
1324                                               "%lu", tv.tv_sec);
1325                                 if (rc > out_end - out_ptr)
1326                                         goto out;
1327                                 out_ptr += rc;
1328                                 break;
1329                         }
1330                         /* hostname */
1331                         case 'h':
1332                                 down_read(&uts_sem);
1333                                 rc = snprintf(out_ptr, out_end - out_ptr,
1334                                               "%s", system_utsname.nodename);
1335                                 up_read(&uts_sem);
1336                                 if (rc > out_end - out_ptr)
1337                                         goto out;
1338                                 out_ptr += rc;
1339                                 break;
1340                         /* executable */
1341                         case 'e':
1342                                 rc = snprintf(out_ptr, out_end - out_ptr,
1343                                               "%s", current->comm);
1344                                 if (rc > out_end - out_ptr)
1345                                         goto out;
1346                                 out_ptr += rc;
1347                                 break;
1348                         default:
1349                                 break;
1350                         }
1351                         ++pat_ptr;
1352                 }
1353         }
1354         /* Backward compatibility with core_uses_pid:
1355          *
1356          * If core_pattern does not include a %p (as is the default)
1357          * and core_uses_pid is set, then .%pid will be appended to
1358          * the filename */
1359         if (!pid_in_pattern
1360             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1361                 rc = snprintf(out_ptr, out_end - out_ptr,
1362                               ".%d", current->tgid);
1363                 if (rc > out_end - out_ptr)
1364                         goto out;
1365                 out_ptr += rc;
1366         }
1367       out:
1368         *out_ptr = 0;
1369 }
1370
1371 static void zap_threads (struct mm_struct *mm)
1372 {
1373         struct task_struct *g, *p;
1374         struct task_struct *tsk = current;
1375         struct completion *vfork_done = tsk->vfork_done;
1376         int traced = 0;
1377
1378         /*
1379          * Make sure nobody is waiting for us to release the VM,
1380          * otherwise we can deadlock when we wait on each other
1381          */
1382         if (vfork_done) {
1383                 tsk->vfork_done = NULL;
1384                 complete(vfork_done);
1385         }
1386
1387         read_lock(&tasklist_lock);
1388         do_each_thread(g,p)
1389                 if (mm == p->mm && p != tsk) {
1390                         force_sig_specific(SIGKILL, p);
1391                         mm->core_waiters++;
1392                         if (unlikely(p->ptrace) &&
1393                             unlikely(p->parent->mm == mm))
1394                                 traced = 1;
1395                 }
1396         while_each_thread(g,p);
1397
1398         read_unlock(&tasklist_lock);
1399
1400         if (unlikely(traced)) {
1401                 /*
1402                  * We are zapping a thread and the thread it ptraces.
1403                  * If the tracee went into a ptrace stop for exit tracing,
1404                  * we could deadlock since the tracer is waiting for this
1405                  * coredump to finish.  Detach them so they can both die.
1406                  */
1407                 write_lock_irq(&tasklist_lock);
1408                 do_each_thread(g,p) {
1409                         if (mm == p->mm && p != tsk &&
1410                             p->ptrace && p->parent->mm == mm) {
1411                                 __ptrace_detach(p, 0);
1412                         }
1413                 } while_each_thread(g,p);
1414                 write_unlock_irq(&tasklist_lock);
1415         }
1416 }
1417
1418 static void coredump_wait(struct mm_struct *mm)
1419 {
1420         DECLARE_COMPLETION(startup_done);
1421         int core_waiters;
1422
1423         mm->core_startup_done = &startup_done;
1424
1425         zap_threads(mm);
1426         core_waiters = mm->core_waiters;
1427         up_write(&mm->mmap_sem);
1428
1429         if (core_waiters)
1430                 wait_for_completion(&startup_done);
1431         BUG_ON(mm->core_waiters);
1432 }
1433
1434 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1435 {
1436         char corename[CORENAME_MAX_SIZE + 1];
1437         struct mm_struct *mm = current->mm;
1438         struct linux_binfmt * binfmt;
1439         struct inode * inode;
1440         struct file * file;
1441         int retval = 0;
1442         int fsuid = current->fsuid;
1443         int flag = 0;
1444
1445         binfmt = current->binfmt;
1446         if (!binfmt || !binfmt->core_dump)
1447                 goto fail;
1448         down_write(&mm->mmap_sem);
1449         if (!mm->dumpable) {
1450                 up_write(&mm->mmap_sem);
1451                 goto fail;
1452         }
1453
1454         /*
1455          *      We cannot trust fsuid as being the "true" uid of the
1456          *      process nor do we know its entire history. We only know it
1457          *      was tainted so we dump it as root in mode 2.
1458          */
1459         if (mm->dumpable == 2) {        /* Setuid core dump mode */
1460                 flag = O_EXCL;          /* Stop rewrite attacks */
1461                 current->fsuid = 0;     /* Dump root private */
1462         }
1463         mm->dumpable = 0;
1464
1465         retval = -EAGAIN;
1466         spin_lock_irq(&current->sighand->siglock);
1467         if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
1468                 current->signal->flags = SIGNAL_GROUP_EXIT;
1469                 current->signal->group_exit_code = exit_code;
1470                 current->signal->group_stop_count = 0;
1471                 retval = 0;
1472         }
1473         spin_unlock_irq(&current->sighand->siglock);
1474         if (retval) {
1475                 up_write(&mm->mmap_sem);
1476                 goto fail;
1477         }
1478
1479         init_completion(&mm->core_done);
1480         coredump_wait(mm);
1481
1482         /*
1483          * Clear any false indication of pending signals that might
1484          * be seen by the filesystem code called to write the core file.
1485          */
1486         clear_thread_flag(TIF_SIGPENDING);
1487
1488         if (current->signal->rlim[RLIMIT_CORE].rlim_cur < binfmt->min_coredump)
1489                 goto fail_unlock;
1490
1491         /*
1492          * lock_kernel() because format_corename() is controlled by sysctl, which
1493          * uses lock_kernel()
1494          */
1495         lock_kernel();
1496         format_corename(corename, core_pattern, signr);
1497         unlock_kernel();
1498         file = filp_open(corename, O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag, 0600);
1499         if (IS_ERR(file))
1500                 goto fail_unlock;
1501         inode = file->f_dentry->d_inode;
1502         if (inode->i_nlink > 1)
1503                 goto close_fail;        /* multiple links - don't dump */
1504         if (d_unhashed(file->f_dentry))
1505                 goto close_fail;
1506
1507         if (!S_ISREG(inode->i_mode))
1508                 goto close_fail;
1509         if (!file->f_op)
1510                 goto close_fail;
1511         if (!file->f_op->write)
1512                 goto close_fail;
1513         if (do_truncate(file->f_dentry, 0, 0, file) != 0)
1514                 goto close_fail;
1515
1516         retval = binfmt->core_dump(signr, regs, file);
1517
1518         if (retval)
1519                 current->signal->group_exit_code |= 0x80;
1520 close_fail:
1521         filp_close(file, NULL);
1522 fail_unlock:
1523         current->fsuid = fsuid;
1524         complete_all(&mm->core_done);
1525 fail:
1526         return retval;
1527 }