signals: de_thread: simplify the ->child_reaper switching
[linux-2.6.git] / fs / exec.c
1 /*
2  *  linux/fs/exec.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  * #!-checking implemented by tytso.
9  */
10 /*
11  * Demand-loading implemented 01.12.91 - no need to read anything but
12  * the header into memory. The inode of the executable is put into
13  * "current->executable", and page faults do the actual loading. Clean.
14  *
15  * Once more I can proudly say that linux stood up to being changed: it
16  * was less than 2 hours work to get demand-loading completely implemented.
17  *
18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
19  * current->executable is only used by the procfs.  This allows a dispatch
20  * table to check for several different types  of binary formats.  We keep
21  * trying until we recognize the file or we run out of supported binary
22  * formats. 
23  */
24
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/mman.h>
28 #include <linux/a.out.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/string.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/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/rmap.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
57 #include <asm/tlb.h>
58
59 #ifdef CONFIG_KMOD
60 #include <linux/kmod.h>
61 #endif
62
63 int core_uses_pid;
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 int suid_dumpable = 0;
66
67 /* The maximal length of core_pattern is also specified in sysctl.c */
68
69 static LIST_HEAD(formats);
70 static DEFINE_RWLOCK(binfmt_lock);
71
72 int register_binfmt(struct linux_binfmt * fmt)
73 {
74         if (!fmt)
75                 return -EINVAL;
76         write_lock(&binfmt_lock);
77         list_add(&fmt->lh, &formats);
78         write_unlock(&binfmt_lock);
79         return 0;       
80 }
81
82 EXPORT_SYMBOL(register_binfmt);
83
84 void unregister_binfmt(struct linux_binfmt * fmt)
85 {
86         write_lock(&binfmt_lock);
87         list_del(&fmt->lh);
88         write_unlock(&binfmt_lock);
89 }
90
91 EXPORT_SYMBOL(unregister_binfmt);
92
93 static inline void put_binfmt(struct linux_binfmt * fmt)
94 {
95         module_put(fmt->module);
96 }
97
98 /*
99  * Note that a shared library must be both readable and executable due to
100  * security reasons.
101  *
102  * Also note that we take the address to load from from the file itself.
103  */
104 asmlinkage long sys_uselib(const char __user * library)
105 {
106         struct file * file;
107         struct nameidata nd;
108         int error;
109
110         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
111         if (error)
112                 goto out;
113
114         error = -EINVAL;
115         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
116                 goto exit;
117
118         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
119         if (error)
120                 goto exit;
121
122         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
123         error = PTR_ERR(file);
124         if (IS_ERR(file))
125                 goto out;
126
127         error = -ENOEXEC;
128         if(file->f_op) {
129                 struct linux_binfmt * fmt;
130
131                 read_lock(&binfmt_lock);
132                 list_for_each_entry(fmt, &formats, lh) {
133                         if (!fmt->load_shlib)
134                                 continue;
135                         if (!try_module_get(fmt->module))
136                                 continue;
137                         read_unlock(&binfmt_lock);
138                         error = fmt->load_shlib(file);
139                         read_lock(&binfmt_lock);
140                         put_binfmt(fmt);
141                         if (error != -ENOEXEC)
142                                 break;
143                 }
144                 read_unlock(&binfmt_lock);
145         }
146         fput(file);
147 out:
148         return error;
149 exit:
150         release_open_intent(&nd);
151         path_put(&nd.path);
152         goto out;
153 }
154
155 #ifdef CONFIG_MMU
156
157 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
158                 int write)
159 {
160         struct page *page;
161         int ret;
162
163 #ifdef CONFIG_STACK_GROWSUP
164         if (write) {
165                 ret = expand_stack_downwards(bprm->vma, pos);
166                 if (ret < 0)
167                         return NULL;
168         }
169 #endif
170         ret = get_user_pages(current, bprm->mm, pos,
171                         1, write, 1, &page, NULL);
172         if (ret <= 0)
173                 return NULL;
174
175         if (write) {
176                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
177                 struct rlimit *rlim;
178
179                 /*
180                  * We've historically supported up to 32 pages (ARG_MAX)
181                  * of argument strings even with small stacks
182                  */
183                 if (size <= ARG_MAX)
184                         return page;
185
186                 /*
187                  * Limit to 1/4-th the stack size for the argv+env strings.
188                  * This ensures that:
189                  *  - the remaining binfmt code will not run out of stack space,
190                  *  - the program will have a reasonable amount of stack left
191                  *    to work from.
192                  */
193                 rlim = current->signal->rlim;
194                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
195                         put_page(page);
196                         return NULL;
197                 }
198         }
199
200         return page;
201 }
202
203 static void put_arg_page(struct page *page)
204 {
205         put_page(page);
206 }
207
208 static void free_arg_page(struct linux_binprm *bprm, int i)
209 {
210 }
211
212 static void free_arg_pages(struct linux_binprm *bprm)
213 {
214 }
215
216 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
217                 struct page *page)
218 {
219         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
220 }
221
222 static int __bprm_mm_init(struct linux_binprm *bprm)
223 {
224         int err = -ENOMEM;
225         struct vm_area_struct *vma = NULL;
226         struct mm_struct *mm = bprm->mm;
227
228         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
229         if (!vma)
230                 goto err;
231
232         down_write(&mm->mmap_sem);
233         vma->vm_mm = mm;
234
235         /*
236          * Place the stack at the largest stack address the architecture
237          * supports. Later, we'll move this to an appropriate place. We don't
238          * use STACK_TOP because that can depend on attributes which aren't
239          * configured yet.
240          */
241         vma->vm_end = STACK_TOP_MAX;
242         vma->vm_start = vma->vm_end - PAGE_SIZE;
243
244         vma->vm_flags = VM_STACK_FLAGS;
245         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
246         err = insert_vm_struct(mm, vma);
247         if (err) {
248                 up_write(&mm->mmap_sem);
249                 goto err;
250         }
251
252         mm->stack_vm = mm->total_vm = 1;
253         up_write(&mm->mmap_sem);
254
255         bprm->p = vma->vm_end - sizeof(void *);
256
257         return 0;
258
259 err:
260         if (vma) {
261                 bprm->vma = NULL;
262                 kmem_cache_free(vm_area_cachep, vma);
263         }
264
265         return err;
266 }
267
268 static bool valid_arg_len(struct linux_binprm *bprm, long len)
269 {
270         return len <= MAX_ARG_STRLEN;
271 }
272
273 #else
274
275 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
276                 int write)
277 {
278         struct page *page;
279
280         page = bprm->page[pos / PAGE_SIZE];
281         if (!page && write) {
282                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
283                 if (!page)
284                         return NULL;
285                 bprm->page[pos / PAGE_SIZE] = page;
286         }
287
288         return page;
289 }
290
291 static void put_arg_page(struct page *page)
292 {
293 }
294
295 static void free_arg_page(struct linux_binprm *bprm, int i)
296 {
297         if (bprm->page[i]) {
298                 __free_page(bprm->page[i]);
299                 bprm->page[i] = NULL;
300         }
301 }
302
303 static void free_arg_pages(struct linux_binprm *bprm)
304 {
305         int i;
306
307         for (i = 0; i < MAX_ARG_PAGES; i++)
308                 free_arg_page(bprm, i);
309 }
310
311 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
312                 struct page *page)
313 {
314 }
315
316 static int __bprm_mm_init(struct linux_binprm *bprm)
317 {
318         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
319         return 0;
320 }
321
322 static bool valid_arg_len(struct linux_binprm *bprm, long len)
323 {
324         return len <= bprm->p;
325 }
326
327 #endif /* CONFIG_MMU */
328
329 /*
330  * Create a new mm_struct and populate it with a temporary stack
331  * vm_area_struct.  We don't have enough context at this point to set the stack
332  * flags, permissions, and offset, so we use temporary values.  We'll update
333  * them later in setup_arg_pages().
334  */
335 int bprm_mm_init(struct linux_binprm *bprm)
336 {
337         int err;
338         struct mm_struct *mm = NULL;
339
340         bprm->mm = mm = mm_alloc();
341         err = -ENOMEM;
342         if (!mm)
343                 goto err;
344
345         err = init_new_context(current, mm);
346         if (err)
347                 goto err;
348
349         err = __bprm_mm_init(bprm);
350         if (err)
351                 goto err;
352
353         return 0;
354
355 err:
356         if (mm) {
357                 bprm->mm = NULL;
358                 mmdrop(mm);
359         }
360
361         return err;
362 }
363
364 /*
365  * count() counts the number of strings in array ARGV.
366  */
367 static int count(char __user * __user * argv, int max)
368 {
369         int i = 0;
370
371         if (argv != NULL) {
372                 for (;;) {
373                         char __user * p;
374
375                         if (get_user(p, argv))
376                                 return -EFAULT;
377                         if (!p)
378                                 break;
379                         argv++;
380                         if(++i > max)
381                                 return -E2BIG;
382                         cond_resched();
383                 }
384         }
385         return i;
386 }
387
388 /*
389  * 'copy_strings()' copies argument/environment strings from the old
390  * processes's memory to the new process's stack.  The call to get_user_pages()
391  * ensures the destination page is created and not swapped out.
392  */
393 static int copy_strings(int argc, char __user * __user * argv,
394                         struct linux_binprm *bprm)
395 {
396         struct page *kmapped_page = NULL;
397         char *kaddr = NULL;
398         unsigned long kpos = 0;
399         int ret;
400
401         while (argc-- > 0) {
402                 char __user *str;
403                 int len;
404                 unsigned long pos;
405
406                 if (get_user(str, argv+argc) ||
407                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
408                         ret = -EFAULT;
409                         goto out;
410                 }
411
412                 if (!valid_arg_len(bprm, len)) {
413                         ret = -E2BIG;
414                         goto out;
415                 }
416
417                 /* We're going to work our way backwords. */
418                 pos = bprm->p;
419                 str += len;
420                 bprm->p -= len;
421
422                 while (len > 0) {
423                         int offset, bytes_to_copy;
424
425                         offset = pos % PAGE_SIZE;
426                         if (offset == 0)
427                                 offset = PAGE_SIZE;
428
429                         bytes_to_copy = offset;
430                         if (bytes_to_copy > len)
431                                 bytes_to_copy = len;
432
433                         offset -= bytes_to_copy;
434                         pos -= bytes_to_copy;
435                         str -= bytes_to_copy;
436                         len -= bytes_to_copy;
437
438                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
439                                 struct page *page;
440
441                                 page = get_arg_page(bprm, pos, 1);
442                                 if (!page) {
443                                         ret = -E2BIG;
444                                         goto out;
445                                 }
446
447                                 if (kmapped_page) {
448                                         flush_kernel_dcache_page(kmapped_page);
449                                         kunmap(kmapped_page);
450                                         put_arg_page(kmapped_page);
451                                 }
452                                 kmapped_page = page;
453                                 kaddr = kmap(kmapped_page);
454                                 kpos = pos & PAGE_MASK;
455                                 flush_arg_page(bprm, kpos, kmapped_page);
456                         }
457                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
458                                 ret = -EFAULT;
459                                 goto out;
460                         }
461                 }
462         }
463         ret = 0;
464 out:
465         if (kmapped_page) {
466                 flush_kernel_dcache_page(kmapped_page);
467                 kunmap(kmapped_page);
468                 put_arg_page(kmapped_page);
469         }
470         return ret;
471 }
472
473 /*
474  * Like copy_strings, but get argv and its values from kernel memory.
475  */
476 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
477 {
478         int r;
479         mm_segment_t oldfs = get_fs();
480         set_fs(KERNEL_DS);
481         r = copy_strings(argc, (char __user * __user *)argv, bprm);
482         set_fs(oldfs);
483         return r;
484 }
485 EXPORT_SYMBOL(copy_strings_kernel);
486
487 #ifdef CONFIG_MMU
488
489 /*
490  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
491  * the binfmt code determines where the new stack should reside, we shift it to
492  * its final location.  The process proceeds as follows:
493  *
494  * 1) Use shift to calculate the new vma endpoints.
495  * 2) Extend vma to cover both the old and new ranges.  This ensures the
496  *    arguments passed to subsequent functions are consistent.
497  * 3) Move vma's page tables to the new range.
498  * 4) Free up any cleared pgd range.
499  * 5) Shrink the vma to cover only the new range.
500  */
501 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
502 {
503         struct mm_struct *mm = vma->vm_mm;
504         unsigned long old_start = vma->vm_start;
505         unsigned long old_end = vma->vm_end;
506         unsigned long length = old_end - old_start;
507         unsigned long new_start = old_start - shift;
508         unsigned long new_end = old_end - shift;
509         struct mmu_gather *tlb;
510
511         BUG_ON(new_start > new_end);
512
513         /*
514          * ensure there are no vmas between where we want to go
515          * and where we are
516          */
517         if (vma != find_vma(mm, new_start))
518                 return -EFAULT;
519
520         /*
521          * cover the whole range: [new_start, old_end)
522          */
523         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
524
525         /*
526          * move the page tables downwards, on failure we rely on
527          * process cleanup to remove whatever mess we made.
528          */
529         if (length != move_page_tables(vma, old_start,
530                                        vma, new_start, length))
531                 return -ENOMEM;
532
533         lru_add_drain();
534         tlb = tlb_gather_mmu(mm, 0);
535         if (new_end > old_start) {
536                 /*
537                  * when the old and new regions overlap clear from new_end.
538                  */
539                 free_pgd_range(&tlb, new_end, old_end, new_end,
540                         vma->vm_next ? vma->vm_next->vm_start : 0);
541         } else {
542                 /*
543                  * otherwise, clean from old_start; this is done to not touch
544                  * the address space in [new_end, old_start) some architectures
545                  * have constraints on va-space that make this illegal (IA64) -
546                  * for the others its just a little faster.
547                  */
548                 free_pgd_range(&tlb, old_start, old_end, new_end,
549                         vma->vm_next ? vma->vm_next->vm_start : 0);
550         }
551         tlb_finish_mmu(tlb, new_end, old_end);
552
553         /*
554          * shrink the vma to just the new range.
555          */
556         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
557
558         return 0;
559 }
560
561 #define EXTRA_STACK_VM_PAGES    20      /* random */
562
563 /*
564  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
565  * the stack is optionally relocated, and some extra space is added.
566  */
567 int setup_arg_pages(struct linux_binprm *bprm,
568                     unsigned long stack_top,
569                     int executable_stack)
570 {
571         unsigned long ret;
572         unsigned long stack_shift;
573         struct mm_struct *mm = current->mm;
574         struct vm_area_struct *vma = bprm->vma;
575         struct vm_area_struct *prev = NULL;
576         unsigned long vm_flags;
577         unsigned long stack_base;
578
579 #ifdef CONFIG_STACK_GROWSUP
580         /* Limit stack size to 1GB */
581         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
582         if (stack_base > (1 << 30))
583                 stack_base = 1 << 30;
584
585         /* Make sure we didn't let the argument array grow too large. */
586         if (vma->vm_end - vma->vm_start > stack_base)
587                 return -ENOMEM;
588
589         stack_base = PAGE_ALIGN(stack_top - stack_base);
590
591         stack_shift = vma->vm_start - stack_base;
592         mm->arg_start = bprm->p - stack_shift;
593         bprm->p = vma->vm_end - stack_shift;
594 #else
595         stack_top = arch_align_stack(stack_top);
596         stack_top = PAGE_ALIGN(stack_top);
597         stack_shift = vma->vm_end - stack_top;
598
599         bprm->p -= stack_shift;
600         mm->arg_start = bprm->p;
601 #endif
602
603         if (bprm->loader)
604                 bprm->loader -= stack_shift;
605         bprm->exec -= stack_shift;
606
607         down_write(&mm->mmap_sem);
608         vm_flags = vma->vm_flags;
609
610         /*
611          * Adjust stack execute permissions; explicitly enable for
612          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
613          * (arch default) otherwise.
614          */
615         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
616                 vm_flags |= VM_EXEC;
617         else if (executable_stack == EXSTACK_DISABLE_X)
618                 vm_flags &= ~VM_EXEC;
619         vm_flags |= mm->def_flags;
620
621         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
622                         vm_flags);
623         if (ret)
624                 goto out_unlock;
625         BUG_ON(prev != vma);
626
627         /* Move stack pages down in memory. */
628         if (stack_shift) {
629                 ret = shift_arg_pages(vma, stack_shift);
630                 if (ret) {
631                         up_write(&mm->mmap_sem);
632                         return ret;
633                 }
634         }
635
636 #ifdef CONFIG_STACK_GROWSUP
637         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
638 #else
639         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
640 #endif
641         ret = expand_stack(vma, stack_base);
642         if (ret)
643                 ret = -EFAULT;
644
645 out_unlock:
646         up_write(&mm->mmap_sem);
647         return 0;
648 }
649 EXPORT_SYMBOL(setup_arg_pages);
650
651 #endif /* CONFIG_MMU */
652
653 struct file *open_exec(const char *name)
654 {
655         struct nameidata nd;
656         int err;
657         struct file *file;
658
659         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
660         file = ERR_PTR(err);
661
662         if (!err) {
663                 struct inode *inode = nd.path.dentry->d_inode;
664                 file = ERR_PTR(-EACCES);
665                 if (S_ISREG(inode->i_mode)) {
666                         int err = vfs_permission(&nd, MAY_EXEC);
667                         file = ERR_PTR(err);
668                         if (!err) {
669                                 file = nameidata_to_filp(&nd,
670                                                         O_RDONLY|O_LARGEFILE);
671                                 if (!IS_ERR(file)) {
672                                         err = deny_write_access(file);
673                                         if (err) {
674                                                 fput(file);
675                                                 file = ERR_PTR(err);
676                                         }
677                                 }
678 out:
679                                 return file;
680                         }
681                 }
682                 release_open_intent(&nd);
683                 path_put(&nd.path);
684         }
685         goto out;
686 }
687
688 EXPORT_SYMBOL(open_exec);
689
690 int kernel_read(struct file *file, unsigned long offset,
691         char *addr, unsigned long count)
692 {
693         mm_segment_t old_fs;
694         loff_t pos = offset;
695         int result;
696
697         old_fs = get_fs();
698         set_fs(get_ds());
699         /* The cast to a user pointer is valid due to the set_fs() */
700         result = vfs_read(file, (void __user *)addr, count, &pos);
701         set_fs(old_fs);
702         return result;
703 }
704
705 EXPORT_SYMBOL(kernel_read);
706
707 static int exec_mmap(struct mm_struct *mm)
708 {
709         struct task_struct *tsk;
710         struct mm_struct * old_mm, *active_mm;
711
712         /* Notify parent that we're no longer interested in the old VM */
713         tsk = current;
714         old_mm = current->mm;
715         mm_release(tsk, old_mm);
716
717         if (old_mm) {
718                 /*
719                  * Make sure that if there is a core dump in progress
720                  * for the old mm, we get out and die instead of going
721                  * through with the exec.  We must hold mmap_sem around
722                  * checking core_waiters and changing tsk->mm.  The
723                  * core-inducing thread will increment core_waiters for
724                  * each thread whose ->mm == old_mm.
725                  */
726                 down_read(&old_mm->mmap_sem);
727                 if (unlikely(old_mm->core_waiters)) {
728                         up_read(&old_mm->mmap_sem);
729                         return -EINTR;
730                 }
731         }
732         task_lock(tsk);
733         active_mm = tsk->active_mm;
734         tsk->mm = mm;
735         tsk->active_mm = mm;
736         activate_mm(active_mm, mm);
737         task_unlock(tsk);
738         mm_update_next_owner(mm);
739         arch_pick_mmap_layout(mm);
740         if (old_mm) {
741                 up_read(&old_mm->mmap_sem);
742                 BUG_ON(active_mm != old_mm);
743                 mmput(old_mm);
744                 return 0;
745         }
746         mmdrop(active_mm);
747         return 0;
748 }
749
750 /*
751  * This function makes sure the current process has its own signal table,
752  * so that flush_signal_handlers can later reset the handlers without
753  * disturbing other processes.  (Other processes might share the signal
754  * table via the CLONE_SIGHAND option to clone().)
755  */
756 static int de_thread(struct task_struct *tsk)
757 {
758         struct signal_struct *sig = tsk->signal;
759         struct sighand_struct *oldsighand = tsk->sighand;
760         spinlock_t *lock = &oldsighand->siglock;
761         struct task_struct *leader = NULL;
762         int count;
763
764         if (thread_group_empty(tsk))
765                 goto no_thread_group;
766
767         /*
768          * Kill all other threads in the thread group.
769          */
770         spin_lock_irq(lock);
771         if (signal_group_exit(sig)) {
772                 /*
773                  * Another group action in progress, just
774                  * return so that the signal is processed.
775                  */
776                 spin_unlock_irq(lock);
777                 return -EAGAIN;
778         }
779         sig->group_exit_task = tsk;
780         zap_other_threads(tsk);
781
782         /* Account for the thread group leader hanging around: */
783         count = thread_group_leader(tsk) ? 1 : 2;
784         sig->notify_count = count;
785         while (atomic_read(&sig->count) > count) {
786                 __set_current_state(TASK_UNINTERRUPTIBLE);
787                 spin_unlock_irq(lock);
788                 schedule();
789                 spin_lock_irq(lock);
790         }
791         spin_unlock_irq(lock);
792
793         /*
794          * At this point all other threads have exited, all we have to
795          * do is to wait for the thread group leader to become inactive,
796          * and to assume its PID:
797          */
798         if (!thread_group_leader(tsk)) {
799                 leader = tsk->group_leader;
800
801                 sig->notify_count = -1;
802                 for (;;) {
803                         write_lock_irq(&tasklist_lock);
804                         if (likely(leader->exit_state))
805                                 break;
806                         __set_current_state(TASK_UNINTERRUPTIBLE);
807                         write_unlock_irq(&tasklist_lock);
808                         schedule();
809                 }
810
811                 if (unlikely(task_child_reaper(tsk) == leader))
812                         task_active_pid_ns(tsk)->child_reaper = tsk;
813                 /*
814                  * The only record we have of the real-time age of a
815                  * process, regardless of execs it's done, is start_time.
816                  * All the past CPU time is accumulated in signal_struct
817                  * from sister threads now dead.  But in this non-leader
818                  * exec, nothing survives from the original leader thread,
819                  * whose birth marks the true age of this process now.
820                  * When we take on its identity by switching to its PID, we
821                  * also take its birthdate (always earlier than our own).
822                  */
823                 tsk->start_time = leader->start_time;
824
825                 BUG_ON(!same_thread_group(leader, tsk));
826                 BUG_ON(has_group_leader_pid(tsk));
827                 /*
828                  * An exec() starts a new thread group with the
829                  * TGID of the previous thread group. Rehash the
830                  * two threads with a switched PID, and release
831                  * the former thread group leader:
832                  */
833
834                 /* Become a process group leader with the old leader's pid.
835                  * The old leader becomes a thread of the this thread group.
836                  * Note: The old leader also uses this pid until release_task
837                  *       is called.  Odd but simple and correct.
838                  */
839                 detach_pid(tsk, PIDTYPE_PID);
840                 tsk->pid = leader->pid;
841                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
842                 transfer_pid(leader, tsk, PIDTYPE_PGID);
843                 transfer_pid(leader, tsk, PIDTYPE_SID);
844                 list_replace_rcu(&leader->tasks, &tsk->tasks);
845
846                 tsk->group_leader = tsk;
847                 leader->group_leader = tsk;
848
849                 tsk->exit_signal = SIGCHLD;
850
851                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
852                 leader->exit_state = EXIT_DEAD;
853
854                 write_unlock_irq(&tasklist_lock);
855         }
856
857         sig->group_exit_task = NULL;
858         sig->notify_count = 0;
859
860 no_thread_group:
861         exit_itimers(sig);
862         if (leader)
863                 release_task(leader);
864
865         if (atomic_read(&oldsighand->count) != 1) {
866                 struct sighand_struct *newsighand;
867                 /*
868                  * This ->sighand is shared with the CLONE_SIGHAND
869                  * but not CLONE_THREAD task, switch to the new one.
870                  */
871                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
872                 if (!newsighand)
873                         return -ENOMEM;
874
875                 atomic_set(&newsighand->count, 1);
876                 memcpy(newsighand->action, oldsighand->action,
877                        sizeof(newsighand->action));
878
879                 write_lock_irq(&tasklist_lock);
880                 spin_lock(&oldsighand->siglock);
881                 rcu_assign_pointer(tsk->sighand, newsighand);
882                 spin_unlock(&oldsighand->siglock);
883                 write_unlock_irq(&tasklist_lock);
884
885                 __cleanup_sighand(oldsighand);
886         }
887
888         BUG_ON(!thread_group_leader(tsk));
889         return 0;
890 }
891
892 /*
893  * These functions flushes out all traces of the currently running executable
894  * so that a new one can be started
895  */
896 static void flush_old_files(struct files_struct * files)
897 {
898         long j = -1;
899         struct fdtable *fdt;
900
901         spin_lock(&files->file_lock);
902         for (;;) {
903                 unsigned long set, i;
904
905                 j++;
906                 i = j * __NFDBITS;
907                 fdt = files_fdtable(files);
908                 if (i >= fdt->max_fds)
909                         break;
910                 set = fdt->close_on_exec->fds_bits[j];
911                 if (!set)
912                         continue;
913                 fdt->close_on_exec->fds_bits[j] = 0;
914                 spin_unlock(&files->file_lock);
915                 for ( ; set ; i++,set >>= 1) {
916                         if (set & 1) {
917                                 sys_close(i);
918                         }
919                 }
920                 spin_lock(&files->file_lock);
921
922         }
923         spin_unlock(&files->file_lock);
924 }
925
926 char *get_task_comm(char *buf, struct task_struct *tsk)
927 {
928         /* buf must be at least sizeof(tsk->comm) in size */
929         task_lock(tsk);
930         strncpy(buf, tsk->comm, sizeof(tsk->comm));
931         task_unlock(tsk);
932         return buf;
933 }
934
935 void set_task_comm(struct task_struct *tsk, char *buf)
936 {
937         task_lock(tsk);
938         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
939         task_unlock(tsk);
940 }
941
942 int flush_old_exec(struct linux_binprm * bprm)
943 {
944         char * name;
945         int i, ch, retval;
946         char tcomm[sizeof(current->comm)];
947
948         /*
949          * Make sure we have a private signal table and that
950          * we are unassociated from the previous thread group.
951          */
952         retval = de_thread(current);
953         if (retval)
954                 goto out;
955
956         set_mm_exe_file(bprm->mm, bprm->file);
957
958         /*
959          * Release all of the old mmap stuff
960          */
961         retval = exec_mmap(bprm->mm);
962         if (retval)
963                 goto out;
964
965         bprm->mm = NULL;                /* We're using it now */
966
967         /* This is the point of no return */
968         current->sas_ss_sp = current->sas_ss_size = 0;
969
970         if (current->euid == current->uid && current->egid == current->gid)
971                 set_dumpable(current->mm, 1);
972         else
973                 set_dumpable(current->mm, suid_dumpable);
974
975         name = bprm->filename;
976
977         /* Copies the binary name from after last slash */
978         for (i=0; (ch = *(name++)) != '\0';) {
979                 if (ch == '/')
980                         i = 0; /* overwrite what we wrote */
981                 else
982                         if (i < (sizeof(tcomm) - 1))
983                                 tcomm[i++] = ch;
984         }
985         tcomm[i] = '\0';
986         set_task_comm(current, tcomm);
987
988         current->flags &= ~PF_RANDOMIZE;
989         flush_thread();
990
991         /* Set the new mm task size. We have to do that late because it may
992          * depend on TIF_32BIT which is only updated in flush_thread() on
993          * some architectures like powerpc
994          */
995         current->mm->task_size = TASK_SIZE;
996
997         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
998                 suid_keys(current);
999                 set_dumpable(current->mm, suid_dumpable);
1000                 current->pdeath_signal = 0;
1001         } else if (file_permission(bprm->file, MAY_READ) ||
1002                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1003                 suid_keys(current);
1004                 set_dumpable(current->mm, suid_dumpable);
1005         }
1006
1007         /* An exec changes our domain. We are no longer part of the thread
1008            group */
1009
1010         current->self_exec_id++;
1011                         
1012         flush_signal_handlers(current, 0);
1013         flush_old_files(current->files);
1014
1015         return 0;
1016
1017 out:
1018         return retval;
1019 }
1020
1021 EXPORT_SYMBOL(flush_old_exec);
1022
1023 /* 
1024  * Fill the binprm structure from the inode. 
1025  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1026  */
1027 int prepare_binprm(struct linux_binprm *bprm)
1028 {
1029         int mode;
1030         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1031         int retval;
1032
1033         mode = inode->i_mode;
1034         if (bprm->file->f_op == NULL)
1035                 return -EACCES;
1036
1037         bprm->e_uid = current->euid;
1038         bprm->e_gid = current->egid;
1039
1040         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1041                 /* Set-uid? */
1042                 if (mode & S_ISUID) {
1043                         current->personality &= ~PER_CLEAR_ON_SETID;
1044                         bprm->e_uid = inode->i_uid;
1045                 }
1046
1047                 /* Set-gid? */
1048                 /*
1049                  * If setgid is set but no group execute bit then this
1050                  * is a candidate for mandatory locking, not a setgid
1051                  * executable.
1052                  */
1053                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1054                         current->personality &= ~PER_CLEAR_ON_SETID;
1055                         bprm->e_gid = inode->i_gid;
1056                 }
1057         }
1058
1059         /* fill in binprm security blob */
1060         retval = security_bprm_set(bprm);
1061         if (retval)
1062                 return retval;
1063
1064         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1065         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1066 }
1067
1068 EXPORT_SYMBOL(prepare_binprm);
1069
1070 static int unsafe_exec(struct task_struct *p)
1071 {
1072         int unsafe = 0;
1073         if (p->ptrace & PT_PTRACED) {
1074                 if (p->ptrace & PT_PTRACE_CAP)
1075                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1076                 else
1077                         unsafe |= LSM_UNSAFE_PTRACE;
1078         }
1079         if (atomic_read(&p->fs->count) > 1 ||
1080             atomic_read(&p->files->count) > 1 ||
1081             atomic_read(&p->sighand->count) > 1)
1082                 unsafe |= LSM_UNSAFE_SHARE;
1083
1084         return unsafe;
1085 }
1086
1087 void compute_creds(struct linux_binprm *bprm)
1088 {
1089         int unsafe;
1090
1091         if (bprm->e_uid != current->uid) {
1092                 suid_keys(current);
1093                 current->pdeath_signal = 0;
1094         }
1095         exec_keys(current);
1096
1097         task_lock(current);
1098         unsafe = unsafe_exec(current);
1099         security_bprm_apply_creds(bprm, unsafe);
1100         task_unlock(current);
1101         security_bprm_post_apply_creds(bprm);
1102 }
1103 EXPORT_SYMBOL(compute_creds);
1104
1105 /*
1106  * Arguments are '\0' separated strings found at the location bprm->p
1107  * points to; chop off the first by relocating brpm->p to right after
1108  * the first '\0' encountered.
1109  */
1110 int remove_arg_zero(struct linux_binprm *bprm)
1111 {
1112         int ret = 0;
1113         unsigned long offset;
1114         char *kaddr;
1115         struct page *page;
1116
1117         if (!bprm->argc)
1118                 return 0;
1119
1120         do {
1121                 offset = bprm->p & ~PAGE_MASK;
1122                 page = get_arg_page(bprm, bprm->p, 0);
1123                 if (!page) {
1124                         ret = -EFAULT;
1125                         goto out;
1126                 }
1127                 kaddr = kmap_atomic(page, KM_USER0);
1128
1129                 for (; offset < PAGE_SIZE && kaddr[offset];
1130                                 offset++, bprm->p++)
1131                         ;
1132
1133                 kunmap_atomic(kaddr, KM_USER0);
1134                 put_arg_page(page);
1135
1136                 if (offset == PAGE_SIZE)
1137                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1138         } while (offset == PAGE_SIZE);
1139
1140         bprm->p++;
1141         bprm->argc--;
1142         ret = 0;
1143
1144 out:
1145         return ret;
1146 }
1147 EXPORT_SYMBOL(remove_arg_zero);
1148
1149 /*
1150  * cycle the list of binary formats handler, until one recognizes the image
1151  */
1152 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1153 {
1154         int try,retval;
1155         struct linux_binfmt *fmt;
1156 #if defined(__alpha__) && defined(CONFIG_ARCH_SUPPORTS_AOUT)
1157         /* handle /sbin/loader.. */
1158         {
1159             struct exec * eh = (struct exec *) bprm->buf;
1160
1161             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1162                 (eh->fh.f_flags & 0x3000) == 0x3000)
1163             {
1164                 struct file * file;
1165                 unsigned long loader;
1166
1167                 allow_write_access(bprm->file);
1168                 fput(bprm->file);
1169                 bprm->file = NULL;
1170
1171                 loader = bprm->vma->vm_end - sizeof(void *);
1172
1173                 file = open_exec("/sbin/loader");
1174                 retval = PTR_ERR(file);
1175                 if (IS_ERR(file))
1176                         return retval;
1177
1178                 /* Remember if the application is TASO.  */
1179                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1180
1181                 bprm->file = file;
1182                 bprm->loader = loader;
1183                 retval = prepare_binprm(bprm);
1184                 if (retval<0)
1185                         return retval;
1186                 /* should call search_binary_handler recursively here,
1187                    but it does not matter */
1188             }
1189         }
1190 #endif
1191         retval = security_bprm_check(bprm);
1192         if (retval)
1193                 return retval;
1194
1195         /* kernel module loader fixup */
1196         /* so we don't try to load run modprobe in kernel space. */
1197         set_fs(USER_DS);
1198
1199         retval = audit_bprm(bprm);
1200         if (retval)
1201                 return retval;
1202
1203         retval = -ENOENT;
1204         for (try=0; try<2; try++) {
1205                 read_lock(&binfmt_lock);
1206                 list_for_each_entry(fmt, &formats, lh) {
1207                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1208                         if (!fn)
1209                                 continue;
1210                         if (!try_module_get(fmt->module))
1211                                 continue;
1212                         read_unlock(&binfmt_lock);
1213                         retval = fn(bprm, regs);
1214                         if (retval >= 0) {
1215                                 put_binfmt(fmt);
1216                                 allow_write_access(bprm->file);
1217                                 if (bprm->file)
1218                                         fput(bprm->file);
1219                                 bprm->file = NULL;
1220                                 current->did_exec = 1;
1221                                 proc_exec_connector(current);
1222                                 return retval;
1223                         }
1224                         read_lock(&binfmt_lock);
1225                         put_binfmt(fmt);
1226                         if (retval != -ENOEXEC || bprm->mm == NULL)
1227                                 break;
1228                         if (!bprm->file) {
1229                                 read_unlock(&binfmt_lock);
1230                                 return retval;
1231                         }
1232                 }
1233                 read_unlock(&binfmt_lock);
1234                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1235                         break;
1236 #ifdef CONFIG_KMOD
1237                 }else{
1238 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1239                         if (printable(bprm->buf[0]) &&
1240                             printable(bprm->buf[1]) &&
1241                             printable(bprm->buf[2]) &&
1242                             printable(bprm->buf[3]))
1243                                 break; /* -ENOEXEC */
1244                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1245 #endif
1246                 }
1247         }
1248         return retval;
1249 }
1250
1251 EXPORT_SYMBOL(search_binary_handler);
1252
1253 /*
1254  * sys_execve() executes a new program.
1255  */
1256 int do_execve(char * filename,
1257         char __user *__user *argv,
1258         char __user *__user *envp,
1259         struct pt_regs * regs)
1260 {
1261         struct linux_binprm *bprm;
1262         struct file *file;
1263         struct files_struct *displaced;
1264         int retval;
1265
1266         retval = unshare_files(&displaced);
1267         if (retval)
1268                 goto out_ret;
1269
1270         retval = -ENOMEM;
1271         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1272         if (!bprm)
1273                 goto out_files;
1274
1275         file = open_exec(filename);
1276         retval = PTR_ERR(file);
1277         if (IS_ERR(file))
1278                 goto out_kfree;
1279
1280         sched_exec();
1281
1282         bprm->file = file;
1283         bprm->filename = filename;
1284         bprm->interp = filename;
1285
1286         retval = bprm_mm_init(bprm);
1287         if (retval)
1288                 goto out_file;
1289
1290         bprm->argc = count(argv, MAX_ARG_STRINGS);
1291         if ((retval = bprm->argc) < 0)
1292                 goto out_mm;
1293
1294         bprm->envc = count(envp, MAX_ARG_STRINGS);
1295         if ((retval = bprm->envc) < 0)
1296                 goto out_mm;
1297
1298         retval = security_bprm_alloc(bprm);
1299         if (retval)
1300                 goto out;
1301
1302         retval = prepare_binprm(bprm);
1303         if (retval < 0)
1304                 goto out;
1305
1306         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1307         if (retval < 0)
1308                 goto out;
1309
1310         bprm->exec = bprm->p;
1311         retval = copy_strings(bprm->envc, envp, bprm);
1312         if (retval < 0)
1313                 goto out;
1314
1315         retval = copy_strings(bprm->argc, argv, bprm);
1316         if (retval < 0)
1317                 goto out;
1318
1319         retval = search_binary_handler(bprm,regs);
1320         if (retval >= 0) {
1321                 /* execve success */
1322                 free_arg_pages(bprm);
1323                 security_bprm_free(bprm);
1324                 acct_update_integrals(current);
1325                 kfree(bprm);
1326                 if (displaced)
1327                         put_files_struct(displaced);
1328                 return retval;
1329         }
1330
1331 out:
1332         free_arg_pages(bprm);
1333         if (bprm->security)
1334                 security_bprm_free(bprm);
1335
1336 out_mm:
1337         if (bprm->mm)
1338                 mmput (bprm->mm);
1339
1340 out_file:
1341         if (bprm->file) {
1342                 allow_write_access(bprm->file);
1343                 fput(bprm->file);
1344         }
1345 out_kfree:
1346         kfree(bprm);
1347
1348 out_files:
1349         if (displaced)
1350                 reset_files_struct(displaced);
1351 out_ret:
1352         return retval;
1353 }
1354
1355 int set_binfmt(struct linux_binfmt *new)
1356 {
1357         struct linux_binfmt *old = current->binfmt;
1358
1359         if (new) {
1360                 if (!try_module_get(new->module))
1361                         return -1;
1362         }
1363         current->binfmt = new;
1364         if (old)
1365                 module_put(old->module);
1366         return 0;
1367 }
1368
1369 EXPORT_SYMBOL(set_binfmt);
1370
1371 /* format_corename will inspect the pattern parameter, and output a
1372  * name into corename, which must have space for at least
1373  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1374  */
1375 static int format_corename(char *corename, const char *pattern, long signr)
1376 {
1377         const char *pat_ptr = pattern;
1378         char *out_ptr = corename;
1379         char *const out_end = corename + CORENAME_MAX_SIZE;
1380         int rc;
1381         int pid_in_pattern = 0;
1382         int ispipe = 0;
1383
1384         if (*pattern == '|')
1385                 ispipe = 1;
1386
1387         /* Repeat as long as we have more pattern to process and more output
1388            space */
1389         while (*pat_ptr) {
1390                 if (*pat_ptr != '%') {
1391                         if (out_ptr == out_end)
1392                                 goto out;
1393                         *out_ptr++ = *pat_ptr++;
1394                 } else {
1395                         switch (*++pat_ptr) {
1396                         case 0:
1397                                 goto out;
1398                         /* Double percent, output one percent */
1399                         case '%':
1400                                 if (out_ptr == out_end)
1401                                         goto out;
1402                                 *out_ptr++ = '%';
1403                                 break;
1404                         /* pid */
1405                         case 'p':
1406                                 pid_in_pattern = 1;
1407                                 rc = snprintf(out_ptr, out_end - out_ptr,
1408                                               "%d", task_tgid_vnr(current));
1409                                 if (rc > out_end - out_ptr)
1410                                         goto out;
1411                                 out_ptr += rc;
1412                                 break;
1413                         /* uid */
1414                         case 'u':
1415                                 rc = snprintf(out_ptr, out_end - out_ptr,
1416                                               "%d", current->uid);
1417                                 if (rc > out_end - out_ptr)
1418                                         goto out;
1419                                 out_ptr += rc;
1420                                 break;
1421                         /* gid */
1422                         case 'g':
1423                                 rc = snprintf(out_ptr, out_end - out_ptr,
1424                                               "%d", current->gid);
1425                                 if (rc > out_end - out_ptr)
1426                                         goto out;
1427                                 out_ptr += rc;
1428                                 break;
1429                         /* signal that caused the coredump */
1430                         case 's':
1431                                 rc = snprintf(out_ptr, out_end - out_ptr,
1432                                               "%ld", signr);
1433                                 if (rc > out_end - out_ptr)
1434                                         goto out;
1435                                 out_ptr += rc;
1436                                 break;
1437                         /* UNIX time of coredump */
1438                         case 't': {
1439                                 struct timeval tv;
1440                                 do_gettimeofday(&tv);
1441                                 rc = snprintf(out_ptr, out_end - out_ptr,
1442                                               "%lu", tv.tv_sec);
1443                                 if (rc > out_end - out_ptr)
1444                                         goto out;
1445                                 out_ptr += rc;
1446                                 break;
1447                         }
1448                         /* hostname */
1449                         case 'h':
1450                                 down_read(&uts_sem);
1451                                 rc = snprintf(out_ptr, out_end - out_ptr,
1452                                               "%s", utsname()->nodename);
1453                                 up_read(&uts_sem);
1454                                 if (rc > out_end - out_ptr)
1455                                         goto out;
1456                                 out_ptr += rc;
1457                                 break;
1458                         /* executable */
1459                         case 'e':
1460                                 rc = snprintf(out_ptr, out_end - out_ptr,
1461                                               "%s", current->comm);
1462                                 if (rc > out_end - out_ptr)
1463                                         goto out;
1464                                 out_ptr += rc;
1465                                 break;
1466                         /* core limit size */
1467                         case 'c':
1468                                 rc = snprintf(out_ptr, out_end - out_ptr,
1469                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1470                                 if (rc > out_end - out_ptr)
1471                                         goto out;
1472                                 out_ptr += rc;
1473                                 break;
1474                         default:
1475                                 break;
1476                         }
1477                         ++pat_ptr;
1478                 }
1479         }
1480         /* Backward compatibility with core_uses_pid:
1481          *
1482          * If core_pattern does not include a %p (as is the default)
1483          * and core_uses_pid is set, then .%pid will be appended to
1484          * the filename. Do not do this for piped commands. */
1485         if (!ispipe && !pid_in_pattern
1486             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1487                 rc = snprintf(out_ptr, out_end - out_ptr,
1488                               ".%d", task_tgid_vnr(current));
1489                 if (rc > out_end - out_ptr)
1490                         goto out;
1491                 out_ptr += rc;
1492         }
1493 out:
1494         *out_ptr = 0;
1495         return ispipe;
1496 }
1497
1498 static void zap_process(struct task_struct *start)
1499 {
1500         struct task_struct *t;
1501
1502         start->signal->flags = SIGNAL_GROUP_EXIT;
1503         start->signal->group_stop_count = 0;
1504
1505         t = start;
1506         do {
1507                 if (t != current && t->mm) {
1508                         t->mm->core_waiters++;
1509                         sigaddset(&t->pending.signal, SIGKILL);
1510                         signal_wake_up(t, 1);
1511                 }
1512         } while ((t = next_thread(t)) != start);
1513 }
1514
1515 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1516                                 int exit_code)
1517 {
1518         struct task_struct *g, *p;
1519         unsigned long flags;
1520         int err = -EAGAIN;
1521
1522         spin_lock_irq(&tsk->sighand->siglock);
1523         if (!signal_group_exit(tsk->signal)) {
1524                 tsk->signal->group_exit_code = exit_code;
1525                 zap_process(tsk);
1526                 err = 0;
1527         }
1528         spin_unlock_irq(&tsk->sighand->siglock);
1529         if (err)
1530                 return err;
1531
1532         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1533                 goto done;
1534
1535         rcu_read_lock();
1536         for_each_process(g) {
1537                 if (g == tsk->group_leader)
1538                         continue;
1539
1540                 p = g;
1541                 do {
1542                         if (p->mm) {
1543                                 if (p->mm == mm) {
1544                                         /*
1545                                          * p->sighand can't disappear, but
1546                                          * may be changed by de_thread()
1547                                          */
1548                                         lock_task_sighand(p, &flags);
1549                                         zap_process(p);
1550                                         unlock_task_sighand(p, &flags);
1551                                 }
1552                                 break;
1553                         }
1554                 } while ((p = next_thread(p)) != g);
1555         }
1556         rcu_read_unlock();
1557 done:
1558         return mm->core_waiters;
1559 }
1560
1561 static int coredump_wait(int exit_code)
1562 {
1563         struct task_struct *tsk = current;
1564         struct mm_struct *mm = tsk->mm;
1565         struct completion startup_done;
1566         struct completion *vfork_done;
1567         int core_waiters;
1568
1569         init_completion(&mm->core_done);
1570         init_completion(&startup_done);
1571         mm->core_startup_done = &startup_done;
1572
1573         core_waiters = zap_threads(tsk, mm, exit_code);
1574         up_write(&mm->mmap_sem);
1575
1576         if (unlikely(core_waiters < 0))
1577                 goto fail;
1578
1579         /*
1580          * Make sure nobody is waiting for us to release the VM,
1581          * otherwise we can deadlock when we wait on each other
1582          */
1583         vfork_done = tsk->vfork_done;
1584         if (vfork_done) {
1585                 tsk->vfork_done = NULL;
1586                 complete(vfork_done);
1587         }
1588
1589         if (core_waiters)
1590                 wait_for_completion(&startup_done);
1591 fail:
1592         BUG_ON(mm->core_waiters);
1593         return core_waiters;
1594 }
1595
1596 /*
1597  * set_dumpable converts traditional three-value dumpable to two flags and
1598  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1599  * these bits are not changed atomically.  So get_dumpable can observe the
1600  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1601  * return either old dumpable or new one by paying attention to the order of
1602  * modifying the bits.
1603  *
1604  * dumpable |   mm->flags (binary)
1605  * old  new | initial interim  final
1606  * ---------+-----------------------
1607  *  0    1  |   00      01      01
1608  *  0    2  |   00      10(*)   11
1609  *  1    0  |   01      00      00
1610  *  1    2  |   01      11      11
1611  *  2    0  |   11      10(*)   00
1612  *  2    1  |   11      11      01
1613  *
1614  * (*) get_dumpable regards interim value of 10 as 11.
1615  */
1616 void set_dumpable(struct mm_struct *mm, int value)
1617 {
1618         switch (value) {
1619         case 0:
1620                 clear_bit(MMF_DUMPABLE, &mm->flags);
1621                 smp_wmb();
1622                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1623                 break;
1624         case 1:
1625                 set_bit(MMF_DUMPABLE, &mm->flags);
1626                 smp_wmb();
1627                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1628                 break;
1629         case 2:
1630                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1631                 smp_wmb();
1632                 set_bit(MMF_DUMPABLE, &mm->flags);
1633                 break;
1634         }
1635 }
1636
1637 int get_dumpable(struct mm_struct *mm)
1638 {
1639         int ret;
1640
1641         ret = mm->flags & 0x3;
1642         return (ret >= 2) ? 2 : ret;
1643 }
1644
1645 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1646 {
1647         char corename[CORENAME_MAX_SIZE + 1];
1648         struct mm_struct *mm = current->mm;
1649         struct linux_binfmt * binfmt;
1650         struct inode * inode;
1651         struct file * file;
1652         int retval = 0;
1653         int fsuid = current->fsuid;
1654         int flag = 0;
1655         int ispipe = 0;
1656         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1657         char **helper_argv = NULL;
1658         int helper_argc = 0;
1659         char *delimit;
1660
1661         audit_core_dumps(signr);
1662
1663         binfmt = current->binfmt;
1664         if (!binfmt || !binfmt->core_dump)
1665                 goto fail;
1666         down_write(&mm->mmap_sem);
1667         /*
1668          * If another thread got here first, or we are not dumpable, bail out.
1669          */
1670         if (mm->core_waiters || !get_dumpable(mm)) {
1671                 up_write(&mm->mmap_sem);
1672                 goto fail;
1673         }
1674
1675         /*
1676          *      We cannot trust fsuid as being the "true" uid of the
1677          *      process nor do we know its entire history. We only know it
1678          *      was tainted so we dump it as root in mode 2.
1679          */
1680         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1681                 flag = O_EXCL;          /* Stop rewrite attacks */
1682                 current->fsuid = 0;     /* Dump root private */
1683         }
1684
1685         retval = coredump_wait(exit_code);
1686         if (retval < 0)
1687                 goto fail;
1688
1689         /*
1690          * Clear any false indication of pending signals that might
1691          * be seen by the filesystem code called to write the core file.
1692          */
1693         clear_thread_flag(TIF_SIGPENDING);
1694
1695         /*
1696          * lock_kernel() because format_corename() is controlled by sysctl, which
1697          * uses lock_kernel()
1698          */
1699         lock_kernel();
1700         ispipe = format_corename(corename, core_pattern, signr);
1701         unlock_kernel();
1702         /*
1703          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1704          * to a pipe.  Since we're not writing directly to the filesystem
1705          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1706          * created unless the pipe reader choses to write out the core file
1707          * at which point file size limits and permissions will be imposed
1708          * as it does with any other process
1709          */
1710         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1711                 goto fail_unlock;
1712
1713         if (ispipe) {
1714                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1715                 /* Terminate the string before the first option */
1716                 delimit = strchr(corename, ' ');
1717                 if (delimit)
1718                         *delimit = '\0';
1719                 delimit = strrchr(helper_argv[0], '/');
1720                 if (delimit)
1721                         delimit++;
1722                 else
1723                         delimit = helper_argv[0];
1724                 if (!strcmp(delimit, current->comm)) {
1725                         printk(KERN_NOTICE "Recursive core dump detected, "
1726                                         "aborting\n");
1727                         goto fail_unlock;
1728                 }
1729
1730                 core_limit = RLIM_INFINITY;
1731
1732                 /* SIGPIPE can happen, but it's just never processed */
1733                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1734                                 &file)) {
1735                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1736                                corename);
1737                         goto fail_unlock;
1738                 }
1739         } else
1740                 file = filp_open(corename,
1741                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1742                                  0600);
1743         if (IS_ERR(file))
1744                 goto fail_unlock;
1745         inode = file->f_path.dentry->d_inode;
1746         if (inode->i_nlink > 1)
1747                 goto close_fail;        /* multiple links - don't dump */
1748         if (!ispipe && d_unhashed(file->f_path.dentry))
1749                 goto close_fail;
1750
1751         /* AK: actually i see no reason to not allow this for named pipes etc.,
1752            but keep the previous behaviour for now. */
1753         if (!ispipe && !S_ISREG(inode->i_mode))
1754                 goto close_fail;
1755         /*
1756          * Dont allow local users get cute and trick others to coredump
1757          * into their pre-created files:
1758          */
1759         if (inode->i_uid != current->fsuid)
1760                 goto close_fail;
1761         if (!file->f_op)
1762                 goto close_fail;
1763         if (!file->f_op->write)
1764                 goto close_fail;
1765         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1766                 goto close_fail;
1767
1768         retval = binfmt->core_dump(signr, regs, file, core_limit);
1769
1770         if (retval)
1771                 current->signal->group_exit_code |= 0x80;
1772 close_fail:
1773         filp_close(file, NULL);
1774 fail_unlock:
1775         if (helper_argv)
1776                 argv_free(helper_argv);
1777
1778         current->fsuid = fsuid;
1779         complete_all(&mm->core_done);
1780 fail:
1781         return retval;
1782 }