dfbf7009fbe78893a9516227629018dea44e85f4
[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/fdtable.h>
28 #include <linux/mm.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.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/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
53 #include <linux/kmod.h>
54
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
57 #include <asm/tlb.h>
58 #include "internal.h"
59
60 #ifdef __alpha__
61 /* for /sbin/loader handling in search_binary_handler() */
62 #include <linux/a.out.h>
63 #endif
64
65 int core_uses_pid;
66 char core_pattern[CORENAME_MAX_SIZE] = "core";
67 int suid_dumpable = 0;
68
69 /* The maximal length of core_pattern is also specified in sysctl.c */
70
71 static LIST_HEAD(formats);
72 static DEFINE_RWLOCK(binfmt_lock);
73
74 int register_binfmt(struct linux_binfmt * fmt)
75 {
76         if (!fmt)
77                 return -EINVAL;
78         write_lock(&binfmt_lock);
79         list_add(&fmt->lh, &formats);
80         write_unlock(&binfmt_lock);
81         return 0;       
82 }
83
84 EXPORT_SYMBOL(register_binfmt);
85
86 void unregister_binfmt(struct linux_binfmt * fmt)
87 {
88         write_lock(&binfmt_lock);
89         list_del(&fmt->lh);
90         write_unlock(&binfmt_lock);
91 }
92
93 EXPORT_SYMBOL(unregister_binfmt);
94
95 static inline void put_binfmt(struct linux_binfmt * fmt)
96 {
97         module_put(fmt->module);
98 }
99
100 /*
101  * Note that a shared library must be both readable and executable due to
102  * security reasons.
103  *
104  * Also note that we take the address to load from from the file itself.
105  */
106 asmlinkage long sys_uselib(const char __user * library)
107 {
108         struct file *file;
109         struct nameidata nd;
110         char *tmp = getname(library);
111         int error = PTR_ERR(tmp);
112
113         if (!IS_ERR(tmp)) {
114                 error = path_lookup_open(AT_FDCWD, tmp,
115                                          LOOKUP_FOLLOW, &nd,
116                                          FMODE_READ|FMODE_EXEC);
117                 putname(tmp);
118         }
119         if (error)
120                 goto out;
121
122         error = -EINVAL;
123         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
124                 goto exit;
125
126         error = -EACCES;
127         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
128                 goto exit;
129
130         error = inode_permission(nd.path.dentry->d_inode,
131                                  MAY_READ | MAY_EXEC | MAY_OPEN);
132         if (error)
133                 goto exit;
134
135         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
136         error = PTR_ERR(file);
137         if (IS_ERR(file))
138                 goto out;
139
140         error = -ENOEXEC;
141         if(file->f_op) {
142                 struct linux_binfmt * fmt;
143
144                 read_lock(&binfmt_lock);
145                 list_for_each_entry(fmt, &formats, lh) {
146                         if (!fmt->load_shlib)
147                                 continue;
148                         if (!try_module_get(fmt->module))
149                                 continue;
150                         read_unlock(&binfmt_lock);
151                         error = fmt->load_shlib(file);
152                         read_lock(&binfmt_lock);
153                         put_binfmt(fmt);
154                         if (error != -ENOEXEC)
155                                 break;
156                 }
157                 read_unlock(&binfmt_lock);
158         }
159         fput(file);
160 out:
161         return error;
162 exit:
163         release_open_intent(&nd);
164         path_put(&nd.path);
165         goto out;
166 }
167
168 #ifdef CONFIG_MMU
169
170 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
171                 int write)
172 {
173         struct page *page;
174         int ret;
175
176 #ifdef CONFIG_STACK_GROWSUP
177         if (write) {
178                 ret = expand_stack_downwards(bprm->vma, pos);
179                 if (ret < 0)
180                         return NULL;
181         }
182 #endif
183         ret = get_user_pages(current, bprm->mm, pos,
184                         1, write, 1, &page, NULL);
185         if (ret <= 0)
186                 return NULL;
187
188         if (write) {
189                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
190                 struct rlimit *rlim;
191
192                 /*
193                  * We've historically supported up to 32 pages (ARG_MAX)
194                  * of argument strings even with small stacks
195                  */
196                 if (size <= ARG_MAX)
197                         return page;
198
199                 /*
200                  * Limit to 1/4-th the stack size for the argv+env strings.
201                  * This ensures that:
202                  *  - the remaining binfmt code will not run out of stack space,
203                  *  - the program will have a reasonable amount of stack left
204                  *    to work from.
205                  */
206                 rlim = current->signal->rlim;
207                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
208                         put_page(page);
209                         return NULL;
210                 }
211         }
212
213         return page;
214 }
215
216 static void put_arg_page(struct page *page)
217 {
218         put_page(page);
219 }
220
221 static void free_arg_page(struct linux_binprm *bprm, int i)
222 {
223 }
224
225 static void free_arg_pages(struct linux_binprm *bprm)
226 {
227 }
228
229 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
230                 struct page *page)
231 {
232         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
233 }
234
235 static int __bprm_mm_init(struct linux_binprm *bprm)
236 {
237         int err = -ENOMEM;
238         struct vm_area_struct *vma = NULL;
239         struct mm_struct *mm = bprm->mm;
240
241         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
242         if (!vma)
243                 goto err;
244
245         down_write(&mm->mmap_sem);
246         vma->vm_mm = mm;
247
248         /*
249          * Place the stack at the largest stack address the architecture
250          * supports. Later, we'll move this to an appropriate place. We don't
251          * use STACK_TOP because that can depend on attributes which aren't
252          * configured yet.
253          */
254         vma->vm_end = STACK_TOP_MAX;
255         vma->vm_start = vma->vm_end - PAGE_SIZE;
256
257         vma->vm_flags = VM_STACK_FLAGS;
258         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
259         err = insert_vm_struct(mm, vma);
260         if (err) {
261                 up_write(&mm->mmap_sem);
262                 goto err;
263         }
264
265         mm->stack_vm = mm->total_vm = 1;
266         up_write(&mm->mmap_sem);
267
268         bprm->p = vma->vm_end - sizeof(void *);
269
270         return 0;
271
272 err:
273         if (vma) {
274                 bprm->vma = NULL;
275                 kmem_cache_free(vm_area_cachep, vma);
276         }
277
278         return err;
279 }
280
281 static bool valid_arg_len(struct linux_binprm *bprm, long len)
282 {
283         return len <= MAX_ARG_STRLEN;
284 }
285
286 #else
287
288 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
289                 int write)
290 {
291         struct page *page;
292
293         page = bprm->page[pos / PAGE_SIZE];
294         if (!page && write) {
295                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
296                 if (!page)
297                         return NULL;
298                 bprm->page[pos / PAGE_SIZE] = page;
299         }
300
301         return page;
302 }
303
304 static void put_arg_page(struct page *page)
305 {
306 }
307
308 static void free_arg_page(struct linux_binprm *bprm, int i)
309 {
310         if (bprm->page[i]) {
311                 __free_page(bprm->page[i]);
312                 bprm->page[i] = NULL;
313         }
314 }
315
316 static void free_arg_pages(struct linux_binprm *bprm)
317 {
318         int i;
319
320         for (i = 0; i < MAX_ARG_PAGES; i++)
321                 free_arg_page(bprm, i);
322 }
323
324 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
325                 struct page *page)
326 {
327 }
328
329 static int __bprm_mm_init(struct linux_binprm *bprm)
330 {
331         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
332         return 0;
333 }
334
335 static bool valid_arg_len(struct linux_binprm *bprm, long len)
336 {
337         return len <= bprm->p;
338 }
339
340 #endif /* CONFIG_MMU */
341
342 /*
343  * Create a new mm_struct and populate it with a temporary stack
344  * vm_area_struct.  We don't have enough context at this point to set the stack
345  * flags, permissions, and offset, so we use temporary values.  We'll update
346  * them later in setup_arg_pages().
347  */
348 int bprm_mm_init(struct linux_binprm *bprm)
349 {
350         int err;
351         struct mm_struct *mm = NULL;
352
353         bprm->mm = mm = mm_alloc();
354         err = -ENOMEM;
355         if (!mm)
356                 goto err;
357
358         err = init_new_context(current, mm);
359         if (err)
360                 goto err;
361
362         err = __bprm_mm_init(bprm);
363         if (err)
364                 goto err;
365
366         return 0;
367
368 err:
369         if (mm) {
370                 bprm->mm = NULL;
371                 mmdrop(mm);
372         }
373
374         return err;
375 }
376
377 /*
378  * count() counts the number of strings in array ARGV.
379  */
380 static int count(char __user * __user * argv, int max)
381 {
382         int i = 0;
383
384         if (argv != NULL) {
385                 for (;;) {
386                         char __user * p;
387
388                         if (get_user(p, argv))
389                                 return -EFAULT;
390                         if (!p)
391                                 break;
392                         argv++;
393                         if (i++ >= max)
394                                 return -E2BIG;
395                         cond_resched();
396                 }
397         }
398         return i;
399 }
400
401 /*
402  * 'copy_strings()' copies argument/environment strings from the old
403  * processes's memory to the new process's stack.  The call to get_user_pages()
404  * ensures the destination page is created and not swapped out.
405  */
406 static int copy_strings(int argc, char __user * __user * argv,
407                         struct linux_binprm *bprm)
408 {
409         struct page *kmapped_page = NULL;
410         char *kaddr = NULL;
411         unsigned long kpos = 0;
412         int ret;
413
414         while (argc-- > 0) {
415                 char __user *str;
416                 int len;
417                 unsigned long pos;
418
419                 if (get_user(str, argv+argc) ||
420                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
421                         ret = -EFAULT;
422                         goto out;
423                 }
424
425                 if (!valid_arg_len(bprm, len)) {
426                         ret = -E2BIG;
427                         goto out;
428                 }
429
430                 /* We're going to work our way backwords. */
431                 pos = bprm->p;
432                 str += len;
433                 bprm->p -= len;
434
435                 while (len > 0) {
436                         int offset, bytes_to_copy;
437
438                         offset = pos % PAGE_SIZE;
439                         if (offset == 0)
440                                 offset = PAGE_SIZE;
441
442                         bytes_to_copy = offset;
443                         if (bytes_to_copy > len)
444                                 bytes_to_copy = len;
445
446                         offset -= bytes_to_copy;
447                         pos -= bytes_to_copy;
448                         str -= bytes_to_copy;
449                         len -= bytes_to_copy;
450
451                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
452                                 struct page *page;
453
454                                 page = get_arg_page(bprm, pos, 1);
455                                 if (!page) {
456                                         ret = -E2BIG;
457                                         goto out;
458                                 }
459
460                                 if (kmapped_page) {
461                                         flush_kernel_dcache_page(kmapped_page);
462                                         kunmap(kmapped_page);
463                                         put_arg_page(kmapped_page);
464                                 }
465                                 kmapped_page = page;
466                                 kaddr = kmap(kmapped_page);
467                                 kpos = pos & PAGE_MASK;
468                                 flush_arg_page(bprm, kpos, kmapped_page);
469                         }
470                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
471                                 ret = -EFAULT;
472                                 goto out;
473                         }
474                 }
475         }
476         ret = 0;
477 out:
478         if (kmapped_page) {
479                 flush_kernel_dcache_page(kmapped_page);
480                 kunmap(kmapped_page);
481                 put_arg_page(kmapped_page);
482         }
483         return ret;
484 }
485
486 /*
487  * Like copy_strings, but get argv and its values from kernel memory.
488  */
489 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
490 {
491         int r;
492         mm_segment_t oldfs = get_fs();
493         set_fs(KERNEL_DS);
494         r = copy_strings(argc, (char __user * __user *)argv, bprm);
495         set_fs(oldfs);
496         return r;
497 }
498 EXPORT_SYMBOL(copy_strings_kernel);
499
500 #ifdef CONFIG_MMU
501
502 /*
503  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
504  * the binfmt code determines where the new stack should reside, we shift it to
505  * its final location.  The process proceeds as follows:
506  *
507  * 1) Use shift to calculate the new vma endpoints.
508  * 2) Extend vma to cover both the old and new ranges.  This ensures the
509  *    arguments passed to subsequent functions are consistent.
510  * 3) Move vma's page tables to the new range.
511  * 4) Free up any cleared pgd range.
512  * 5) Shrink the vma to cover only the new range.
513  */
514 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
515 {
516         struct mm_struct *mm = vma->vm_mm;
517         unsigned long old_start = vma->vm_start;
518         unsigned long old_end = vma->vm_end;
519         unsigned long length = old_end - old_start;
520         unsigned long new_start = old_start - shift;
521         unsigned long new_end = old_end - shift;
522         struct mmu_gather *tlb;
523
524         BUG_ON(new_start > new_end);
525
526         /*
527          * ensure there are no vmas between where we want to go
528          * and where we are
529          */
530         if (vma != find_vma(mm, new_start))
531                 return -EFAULT;
532
533         /*
534          * cover the whole range: [new_start, old_end)
535          */
536         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
537
538         /*
539          * move the page tables downwards, on failure we rely on
540          * process cleanup to remove whatever mess we made.
541          */
542         if (length != move_page_tables(vma, old_start,
543                                        vma, new_start, length))
544                 return -ENOMEM;
545
546         lru_add_drain();
547         tlb = tlb_gather_mmu(mm, 0);
548         if (new_end > old_start) {
549                 /*
550                  * when the old and new regions overlap clear from new_end.
551                  */
552                 free_pgd_range(tlb, new_end, old_end, new_end,
553                         vma->vm_next ? vma->vm_next->vm_start : 0);
554         } else {
555                 /*
556                  * otherwise, clean from old_start; this is done to not touch
557                  * the address space in [new_end, old_start) some architectures
558                  * have constraints on va-space that make this illegal (IA64) -
559                  * for the others its just a little faster.
560                  */
561                 free_pgd_range(tlb, old_start, old_end, new_end,
562                         vma->vm_next ? vma->vm_next->vm_start : 0);
563         }
564         tlb_finish_mmu(tlb, new_end, old_end);
565
566         /*
567          * shrink the vma to just the new range.
568          */
569         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
570
571         return 0;
572 }
573
574 #define EXTRA_STACK_VM_PAGES    20      /* random */
575
576 /*
577  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
578  * the stack is optionally relocated, and some extra space is added.
579  */
580 int setup_arg_pages(struct linux_binprm *bprm,
581                     unsigned long stack_top,
582                     int executable_stack)
583 {
584         unsigned long ret;
585         unsigned long stack_shift;
586         struct mm_struct *mm = current->mm;
587         struct vm_area_struct *vma = bprm->vma;
588         struct vm_area_struct *prev = NULL;
589         unsigned long vm_flags;
590         unsigned long stack_base;
591
592 #ifdef CONFIG_STACK_GROWSUP
593         /* Limit stack size to 1GB */
594         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
595         if (stack_base > (1 << 30))
596                 stack_base = 1 << 30;
597
598         /* Make sure we didn't let the argument array grow too large. */
599         if (vma->vm_end - vma->vm_start > stack_base)
600                 return -ENOMEM;
601
602         stack_base = PAGE_ALIGN(stack_top - stack_base);
603
604         stack_shift = vma->vm_start - stack_base;
605         mm->arg_start = bprm->p - stack_shift;
606         bprm->p = vma->vm_end - stack_shift;
607 #else
608         stack_top = arch_align_stack(stack_top);
609         stack_top = PAGE_ALIGN(stack_top);
610         stack_shift = vma->vm_end - stack_top;
611
612         bprm->p -= stack_shift;
613         mm->arg_start = bprm->p;
614 #endif
615
616         if (bprm->loader)
617                 bprm->loader -= stack_shift;
618         bprm->exec -= stack_shift;
619
620         down_write(&mm->mmap_sem);
621         vm_flags = VM_STACK_FLAGS;
622
623         /*
624          * Adjust stack execute permissions; explicitly enable for
625          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
626          * (arch default) otherwise.
627          */
628         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
629                 vm_flags |= VM_EXEC;
630         else if (executable_stack == EXSTACK_DISABLE_X)
631                 vm_flags &= ~VM_EXEC;
632         vm_flags |= mm->def_flags;
633
634         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
635                         vm_flags);
636         if (ret)
637                 goto out_unlock;
638         BUG_ON(prev != vma);
639
640         /* Move stack pages down in memory. */
641         if (stack_shift) {
642                 ret = shift_arg_pages(vma, stack_shift);
643                 if (ret) {
644                         up_write(&mm->mmap_sem);
645                         return ret;
646                 }
647         }
648
649 #ifdef CONFIG_STACK_GROWSUP
650         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
651 #else
652         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
653 #endif
654         ret = expand_stack(vma, stack_base);
655         if (ret)
656                 ret = -EFAULT;
657
658 out_unlock:
659         up_write(&mm->mmap_sem);
660         return 0;
661 }
662 EXPORT_SYMBOL(setup_arg_pages);
663
664 #endif /* CONFIG_MMU */
665
666 struct file *open_exec(const char *name)
667 {
668         struct nameidata nd;
669         struct file *file;
670         int err;
671
672         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
673                                 FMODE_READ|FMODE_EXEC);
674         if (err)
675                 goto out;
676
677         err = -EACCES;
678         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
679                 goto out_path_put;
680
681         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
682                 goto out_path_put;
683
684         err = inode_permission(nd.path.dentry->d_inode, MAY_EXEC | MAY_OPEN);
685         if (err)
686                 goto out_path_put;
687
688         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
689         if (IS_ERR(file))
690                 return file;
691
692         err = deny_write_access(file);
693         if (err) {
694                 fput(file);
695                 goto out;
696         }
697
698         return file;
699
700  out_path_put:
701         release_open_intent(&nd);
702         path_put(&nd.path);
703  out:
704         return ERR_PTR(err);
705 }
706 EXPORT_SYMBOL(open_exec);
707
708 int kernel_read(struct file *file, unsigned long offset,
709         char *addr, unsigned long count)
710 {
711         mm_segment_t old_fs;
712         loff_t pos = offset;
713         int result;
714
715         old_fs = get_fs();
716         set_fs(get_ds());
717         /* The cast to a user pointer is valid due to the set_fs() */
718         result = vfs_read(file, (void __user *)addr, count, &pos);
719         set_fs(old_fs);
720         return result;
721 }
722
723 EXPORT_SYMBOL(kernel_read);
724
725 static int exec_mmap(struct mm_struct *mm)
726 {
727         struct task_struct *tsk;
728         struct mm_struct * old_mm, *active_mm;
729
730         /* Notify parent that we're no longer interested in the old VM */
731         tsk = current;
732         old_mm = current->mm;
733         mm_release(tsk, old_mm);
734
735         if (old_mm) {
736                 /*
737                  * Make sure that if there is a core dump in progress
738                  * for the old mm, we get out and die instead of going
739                  * through with the exec.  We must hold mmap_sem around
740                  * checking core_state and changing tsk->mm.
741                  */
742                 down_read(&old_mm->mmap_sem);
743                 if (unlikely(old_mm->core_state)) {
744                         up_read(&old_mm->mmap_sem);
745                         return -EINTR;
746                 }
747         }
748         task_lock(tsk);
749         active_mm = tsk->active_mm;
750         tsk->mm = mm;
751         tsk->active_mm = mm;
752         activate_mm(active_mm, mm);
753         task_unlock(tsk);
754         arch_pick_mmap_layout(mm);
755         if (old_mm) {
756                 up_read(&old_mm->mmap_sem);
757                 BUG_ON(active_mm != old_mm);
758                 mm_update_next_owner(old_mm);
759                 mmput(old_mm);
760                 return 0;
761         }
762         mmdrop(active_mm);
763         return 0;
764 }
765
766 /*
767  * This function makes sure the current process has its own signal table,
768  * so that flush_signal_handlers can later reset the handlers without
769  * disturbing other processes.  (Other processes might share the signal
770  * table via the CLONE_SIGHAND option to clone().)
771  */
772 static int de_thread(struct task_struct *tsk)
773 {
774         struct signal_struct *sig = tsk->signal;
775         struct sighand_struct *oldsighand = tsk->sighand;
776         spinlock_t *lock = &oldsighand->siglock;
777         int count;
778
779         if (thread_group_empty(tsk))
780                 goto no_thread_group;
781
782         /*
783          * Kill all other threads in the thread group.
784          */
785         spin_lock_irq(lock);
786         if (signal_group_exit(sig)) {
787                 /*
788                  * Another group action in progress, just
789                  * return so that the signal is processed.
790                  */
791                 spin_unlock_irq(lock);
792                 return -EAGAIN;
793         }
794         sig->group_exit_task = tsk;
795         zap_other_threads(tsk);
796
797         /* Account for the thread group leader hanging around: */
798         count = thread_group_leader(tsk) ? 1 : 2;
799         sig->notify_count = count;
800         while (atomic_read(&sig->count) > count) {
801                 __set_current_state(TASK_UNINTERRUPTIBLE);
802                 spin_unlock_irq(lock);
803                 schedule();
804                 spin_lock_irq(lock);
805         }
806         spin_unlock_irq(lock);
807
808         /*
809          * At this point all other threads have exited, all we have to
810          * do is to wait for the thread group leader to become inactive,
811          * and to assume its PID:
812          */
813         if (!thread_group_leader(tsk)) {
814                 struct task_struct *leader = tsk->group_leader;
815
816                 sig->notify_count = -1; /* for exit_notify() */
817                 for (;;) {
818                         write_lock_irq(&tasklist_lock);
819                         if (likely(leader->exit_state))
820                                 break;
821                         __set_current_state(TASK_UNINTERRUPTIBLE);
822                         write_unlock_irq(&tasklist_lock);
823                         schedule();
824                 }
825
826                 /*
827                  * The only record we have of the real-time age of a
828                  * process, regardless of execs it's done, is start_time.
829                  * All the past CPU time is accumulated in signal_struct
830                  * from sister threads now dead.  But in this non-leader
831                  * exec, nothing survives from the original leader thread,
832                  * whose birth marks the true age of this process now.
833                  * When we take on its identity by switching to its PID, we
834                  * also take its birthdate (always earlier than our own).
835                  */
836                 tsk->start_time = leader->start_time;
837
838                 BUG_ON(!same_thread_group(leader, tsk));
839                 BUG_ON(has_group_leader_pid(tsk));
840                 /*
841                  * An exec() starts a new thread group with the
842                  * TGID of the previous thread group. Rehash the
843                  * two threads with a switched PID, and release
844                  * the former thread group leader:
845                  */
846
847                 /* Become a process group leader with the old leader's pid.
848                  * The old leader becomes a thread of the this thread group.
849                  * Note: The old leader also uses this pid until release_task
850                  *       is called.  Odd but simple and correct.
851                  */
852                 detach_pid(tsk, PIDTYPE_PID);
853                 tsk->pid = leader->pid;
854                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
855                 transfer_pid(leader, tsk, PIDTYPE_PGID);
856                 transfer_pid(leader, tsk, PIDTYPE_SID);
857                 list_replace_rcu(&leader->tasks, &tsk->tasks);
858
859                 tsk->group_leader = tsk;
860                 leader->group_leader = tsk;
861
862                 tsk->exit_signal = SIGCHLD;
863
864                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
865                 leader->exit_state = EXIT_DEAD;
866                 write_unlock_irq(&tasklist_lock);
867
868                 release_task(leader);
869         }
870
871         sig->group_exit_task = NULL;
872         sig->notify_count = 0;
873
874 no_thread_group:
875         exit_itimers(sig);
876         flush_itimer_signals();
877
878         if (atomic_read(&oldsighand->count) != 1) {
879                 struct sighand_struct *newsighand;
880                 /*
881                  * This ->sighand is shared with the CLONE_SIGHAND
882                  * but not CLONE_THREAD task, switch to the new one.
883                  */
884                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
885                 if (!newsighand)
886                         return -ENOMEM;
887
888                 atomic_set(&newsighand->count, 1);
889                 memcpy(newsighand->action, oldsighand->action,
890                        sizeof(newsighand->action));
891
892                 write_lock_irq(&tasklist_lock);
893                 spin_lock(&oldsighand->siglock);
894                 rcu_assign_pointer(tsk->sighand, newsighand);
895                 spin_unlock(&oldsighand->siglock);
896                 write_unlock_irq(&tasklist_lock);
897
898                 __cleanup_sighand(oldsighand);
899         }
900
901         BUG_ON(!thread_group_leader(tsk));
902         return 0;
903 }
904
905 /*
906  * These functions flushes out all traces of the currently running executable
907  * so that a new one can be started
908  */
909 static void flush_old_files(struct files_struct * files)
910 {
911         long j = -1;
912         struct fdtable *fdt;
913
914         spin_lock(&files->file_lock);
915         for (;;) {
916                 unsigned long set, i;
917
918                 j++;
919                 i = j * __NFDBITS;
920                 fdt = files_fdtable(files);
921                 if (i >= fdt->max_fds)
922                         break;
923                 set = fdt->close_on_exec->fds_bits[j];
924                 if (!set)
925                         continue;
926                 fdt->close_on_exec->fds_bits[j] = 0;
927                 spin_unlock(&files->file_lock);
928                 for ( ; set ; i++,set >>= 1) {
929                         if (set & 1) {
930                                 sys_close(i);
931                         }
932                 }
933                 spin_lock(&files->file_lock);
934
935         }
936         spin_unlock(&files->file_lock);
937 }
938
939 char *get_task_comm(char *buf, struct task_struct *tsk)
940 {
941         /* buf must be at least sizeof(tsk->comm) in size */
942         task_lock(tsk);
943         strncpy(buf, tsk->comm, sizeof(tsk->comm));
944         task_unlock(tsk);
945         return buf;
946 }
947
948 void set_task_comm(struct task_struct *tsk, char *buf)
949 {
950         task_lock(tsk);
951         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
952         task_unlock(tsk);
953 }
954
955 int flush_old_exec(struct linux_binprm * bprm)
956 {
957         char * name;
958         int i, ch, retval;
959         char tcomm[sizeof(current->comm)];
960
961         /*
962          * Make sure we have a private signal table and that
963          * we are unassociated from the previous thread group.
964          */
965         retval = de_thread(current);
966         if (retval)
967                 goto out;
968
969         set_mm_exe_file(bprm->mm, bprm->file);
970
971         /*
972          * Release all of the old mmap stuff
973          */
974         retval = exec_mmap(bprm->mm);
975         if (retval)
976                 goto out;
977
978         bprm->mm = NULL;                /* We're using it now */
979
980         /* This is the point of no return */
981         current->sas_ss_sp = current->sas_ss_size = 0;
982
983         if (current_euid() == current_uid() && current_egid() == current_gid())
984                 set_dumpable(current->mm, 1);
985         else
986                 set_dumpable(current->mm, suid_dumpable);
987
988         name = bprm->filename;
989
990         /* Copies the binary name from after last slash */
991         for (i=0; (ch = *(name++)) != '\0';) {
992                 if (ch == '/')
993                         i = 0; /* overwrite what we wrote */
994                 else
995                         if (i < (sizeof(tcomm) - 1))
996                                 tcomm[i++] = ch;
997         }
998         tcomm[i] = '\0';
999         set_task_comm(current, tcomm);
1000
1001         current->flags &= ~PF_RANDOMIZE;
1002         flush_thread();
1003
1004         /* Set the new mm task size. We have to do that late because it may
1005          * depend on TIF_32BIT which is only updated in flush_thread() on
1006          * some architectures like powerpc
1007          */
1008         current->mm->task_size = TASK_SIZE;
1009
1010         /* install the new credentials */
1011         if (bprm->cred->uid != current_euid() ||
1012             bprm->cred->gid != current_egid()) {
1013                 current->pdeath_signal = 0;
1014         } else if (file_permission(bprm->file, MAY_READ) ||
1015                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1016                 set_dumpable(current->mm, suid_dumpable);
1017         }
1018
1019         current->personality &= ~bprm->per_clear;
1020
1021         /* An exec changes our domain. We are no longer part of the thread
1022            group */
1023
1024         current->self_exec_id++;
1025                         
1026         flush_signal_handlers(current, 0);
1027         flush_old_files(current->files);
1028
1029         return 0;
1030
1031 out:
1032         return retval;
1033 }
1034
1035 EXPORT_SYMBOL(flush_old_exec);
1036
1037 /*
1038  * install the new credentials for this executable
1039  */
1040 void install_exec_creds(struct linux_binprm *bprm)
1041 {
1042         security_bprm_committing_creds(bprm);
1043
1044         commit_creds(bprm->cred);
1045         bprm->cred = NULL;
1046
1047         /* cred_exec_mutex must be held at least to this point to prevent
1048          * ptrace_attach() from altering our determination of the task's
1049          * credentials; any time after this it may be unlocked */
1050
1051         security_bprm_committed_creds(bprm);
1052 }
1053 EXPORT_SYMBOL(install_exec_creds);
1054
1055 /*
1056  * determine how safe it is to execute the proposed program
1057  * - the caller must hold current->cred_exec_mutex to protect against
1058  *   PTRACE_ATTACH
1059  */
1060 void check_unsafe_exec(struct linux_binprm *bprm)
1061 {
1062         struct task_struct *p = current;
1063
1064         bprm->unsafe = tracehook_unsafe_exec(p);
1065
1066         if (atomic_read(&p->fs->count) > 1 ||
1067             atomic_read(&p->files->count) > 1 ||
1068             atomic_read(&p->sighand->count) > 1)
1069                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1070 }
1071
1072 /* 
1073  * Fill the binprm structure from the inode. 
1074  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1075  *
1076  * This may be called multiple times for binary chains (scripts for example).
1077  */
1078 int prepare_binprm(struct linux_binprm *bprm)
1079 {
1080         umode_t mode;
1081         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1082         int retval;
1083
1084         mode = inode->i_mode;
1085         if (bprm->file->f_op == NULL)
1086                 return -EACCES;
1087
1088         /* clear any previous set[ug]id data from a previous binary */
1089         bprm->cred->euid = current_euid();
1090         bprm->cred->egid = current_egid();
1091
1092         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1093                 /* Set-uid? */
1094                 if (mode & S_ISUID) {
1095                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1096                         bprm->cred->euid = inode->i_uid;
1097                 }
1098
1099                 /* Set-gid? */
1100                 /*
1101                  * If setgid is set but no group execute bit then this
1102                  * is a candidate for mandatory locking, not a setgid
1103                  * executable.
1104                  */
1105                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1106                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1107                         bprm->cred->egid = inode->i_gid;
1108                 }
1109         }
1110
1111         /* fill in binprm security blob */
1112         retval = security_bprm_set_creds(bprm);
1113         if (retval)
1114                 return retval;
1115         bprm->cred_prepared = 1;
1116
1117         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1118         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1119 }
1120
1121 EXPORT_SYMBOL(prepare_binprm);
1122
1123 /*
1124  * Arguments are '\0' separated strings found at the location bprm->p
1125  * points to; chop off the first by relocating brpm->p to right after
1126  * the first '\0' encountered.
1127  */
1128 int remove_arg_zero(struct linux_binprm *bprm)
1129 {
1130         int ret = 0;
1131         unsigned long offset;
1132         char *kaddr;
1133         struct page *page;
1134
1135         if (!bprm->argc)
1136                 return 0;
1137
1138         do {
1139                 offset = bprm->p & ~PAGE_MASK;
1140                 page = get_arg_page(bprm, bprm->p, 0);
1141                 if (!page) {
1142                         ret = -EFAULT;
1143                         goto out;
1144                 }
1145                 kaddr = kmap_atomic(page, KM_USER0);
1146
1147                 for (; offset < PAGE_SIZE && kaddr[offset];
1148                                 offset++, bprm->p++)
1149                         ;
1150
1151                 kunmap_atomic(kaddr, KM_USER0);
1152                 put_arg_page(page);
1153
1154                 if (offset == PAGE_SIZE)
1155                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1156         } while (offset == PAGE_SIZE);
1157
1158         bprm->p++;
1159         bprm->argc--;
1160         ret = 0;
1161
1162 out:
1163         return ret;
1164 }
1165 EXPORT_SYMBOL(remove_arg_zero);
1166
1167 /*
1168  * cycle the list of binary formats handler, until one recognizes the image
1169  */
1170 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1171 {
1172         unsigned int depth = bprm->recursion_depth;
1173         int try,retval;
1174         struct linux_binfmt *fmt;
1175 #ifdef __alpha__
1176         /* handle /sbin/loader.. */
1177         {
1178             struct exec * eh = (struct exec *) bprm->buf;
1179
1180             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1181                 (eh->fh.f_flags & 0x3000) == 0x3000)
1182             {
1183                 struct file * file;
1184                 unsigned long loader;
1185
1186                 allow_write_access(bprm->file);
1187                 fput(bprm->file);
1188                 bprm->file = NULL;
1189
1190                 loader = bprm->vma->vm_end - sizeof(void *);
1191
1192                 file = open_exec("/sbin/loader");
1193                 retval = PTR_ERR(file);
1194                 if (IS_ERR(file))
1195                         return retval;
1196
1197                 /* Remember if the application is TASO.  */
1198                 bprm->taso = eh->ah.entry < 0x100000000UL;
1199
1200                 bprm->file = file;
1201                 bprm->loader = loader;
1202                 retval = prepare_binprm(bprm);
1203                 if (retval<0)
1204                         return retval;
1205                 /* should call search_binary_handler recursively here,
1206                    but it does not matter */
1207             }
1208         }
1209 #endif
1210         retval = security_bprm_check(bprm);
1211         if (retval)
1212                 return retval;
1213
1214         /* kernel module loader fixup */
1215         /* so we don't try to load run modprobe in kernel space. */
1216         set_fs(USER_DS);
1217
1218         retval = audit_bprm(bprm);
1219         if (retval)
1220                 return retval;
1221
1222         retval = -ENOENT;
1223         for (try=0; try<2; try++) {
1224                 read_lock(&binfmt_lock);
1225                 list_for_each_entry(fmt, &formats, lh) {
1226                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1227                         if (!fn)
1228                                 continue;
1229                         if (!try_module_get(fmt->module))
1230                                 continue;
1231                         read_unlock(&binfmt_lock);
1232                         retval = fn(bprm, regs);
1233                         /*
1234                          * Restore the depth counter to its starting value
1235                          * in this call, so we don't have to rely on every
1236                          * load_binary function to restore it on return.
1237                          */
1238                         bprm->recursion_depth = depth;
1239                         if (retval >= 0) {
1240                                 if (depth == 0)
1241                                         tracehook_report_exec(fmt, bprm, regs);
1242                                 put_binfmt(fmt);
1243                                 allow_write_access(bprm->file);
1244                                 if (bprm->file)
1245                                         fput(bprm->file);
1246                                 bprm->file = NULL;
1247                                 current->did_exec = 1;
1248                                 proc_exec_connector(current);
1249                                 return retval;
1250                         }
1251                         read_lock(&binfmt_lock);
1252                         put_binfmt(fmt);
1253                         if (retval != -ENOEXEC || bprm->mm == NULL)
1254                                 break;
1255                         if (!bprm->file) {
1256                                 read_unlock(&binfmt_lock);
1257                                 return retval;
1258                         }
1259                 }
1260                 read_unlock(&binfmt_lock);
1261                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1262                         break;
1263 #ifdef CONFIG_MODULES
1264                 } else {
1265 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1266                         if (printable(bprm->buf[0]) &&
1267                             printable(bprm->buf[1]) &&
1268                             printable(bprm->buf[2]) &&
1269                             printable(bprm->buf[3]))
1270                                 break; /* -ENOEXEC */
1271                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1272 #endif
1273                 }
1274         }
1275         return retval;
1276 }
1277
1278 EXPORT_SYMBOL(search_binary_handler);
1279
1280 void free_bprm(struct linux_binprm *bprm)
1281 {
1282         free_arg_pages(bprm);
1283         if (bprm->cred)
1284                 abort_creds(bprm->cred);
1285         kfree(bprm);
1286 }
1287
1288 /*
1289  * sys_execve() executes a new program.
1290  */
1291 int do_execve(char * filename,
1292         char __user *__user *argv,
1293         char __user *__user *envp,
1294         struct pt_regs * regs)
1295 {
1296         struct linux_binprm *bprm;
1297         struct file *file;
1298         struct files_struct *displaced;
1299         int retval;
1300
1301         retval = unshare_files(&displaced);
1302         if (retval)
1303                 goto out_ret;
1304
1305         retval = -ENOMEM;
1306         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1307         if (!bprm)
1308                 goto out_files;
1309
1310         retval = mutex_lock_interruptible(&current->cred_exec_mutex);
1311         if (retval < 0)
1312                 goto out_free;
1313
1314         retval = -ENOMEM;
1315         bprm->cred = prepare_exec_creds();
1316         if (!bprm->cred)
1317                 goto out_unlock;
1318         check_unsafe_exec(bprm);
1319
1320         file = open_exec(filename);
1321         retval = PTR_ERR(file);
1322         if (IS_ERR(file))
1323                 goto out_unlock;
1324
1325         sched_exec();
1326
1327         bprm->file = file;
1328         bprm->filename = filename;
1329         bprm->interp = filename;
1330
1331         retval = bprm_mm_init(bprm);
1332         if (retval)
1333                 goto out_file;
1334
1335         bprm->argc = count(argv, MAX_ARG_STRINGS);
1336         if ((retval = bprm->argc) < 0)
1337                 goto out;
1338
1339         bprm->envc = count(envp, MAX_ARG_STRINGS);
1340         if ((retval = bprm->envc) < 0)
1341                 goto out;
1342
1343         retval = prepare_binprm(bprm);
1344         if (retval < 0)
1345                 goto out;
1346
1347         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1348         if (retval < 0)
1349                 goto out;
1350
1351         bprm->exec = bprm->p;
1352         retval = copy_strings(bprm->envc, envp, bprm);
1353         if (retval < 0)
1354                 goto out;
1355
1356         retval = copy_strings(bprm->argc, argv, bprm);
1357         if (retval < 0)
1358                 goto out;
1359
1360         current->flags &= ~PF_KTHREAD;
1361         retval = search_binary_handler(bprm,regs);
1362         if (retval < 0)
1363                 goto out;
1364
1365         /* execve succeeded */
1366         mutex_unlock(&current->cred_exec_mutex);
1367         acct_update_integrals(current);
1368         free_bprm(bprm);
1369         if (displaced)
1370                 put_files_struct(displaced);
1371         return retval;
1372
1373 out:
1374         if (bprm->mm)
1375                 mmput (bprm->mm);
1376
1377 out_file:
1378         if (bprm->file) {
1379                 allow_write_access(bprm->file);
1380                 fput(bprm->file);
1381         }
1382
1383 out_unlock:
1384         mutex_unlock(&current->cred_exec_mutex);
1385
1386 out_free:
1387         free_bprm(bprm);
1388
1389 out_files:
1390         if (displaced)
1391                 reset_files_struct(displaced);
1392 out_ret:
1393         return retval;
1394 }
1395
1396 int set_binfmt(struct linux_binfmt *new)
1397 {
1398         struct linux_binfmt *old = current->binfmt;
1399
1400         if (new) {
1401                 if (!try_module_get(new->module))
1402                         return -1;
1403         }
1404         current->binfmt = new;
1405         if (old)
1406                 module_put(old->module);
1407         return 0;
1408 }
1409
1410 EXPORT_SYMBOL(set_binfmt);
1411
1412 /* format_corename will inspect the pattern parameter, and output a
1413  * name into corename, which must have space for at least
1414  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1415  */
1416 static int format_corename(char *corename, long signr)
1417 {
1418         const struct cred *cred = current_cred();
1419         const char *pat_ptr = core_pattern;
1420         int ispipe = (*pat_ptr == '|');
1421         char *out_ptr = corename;
1422         char *const out_end = corename + CORENAME_MAX_SIZE;
1423         int rc;
1424         int pid_in_pattern = 0;
1425
1426         /* Repeat as long as we have more pattern to process and more output
1427            space */
1428         while (*pat_ptr) {
1429                 if (*pat_ptr != '%') {
1430                         if (out_ptr == out_end)
1431                                 goto out;
1432                         *out_ptr++ = *pat_ptr++;
1433                 } else {
1434                         switch (*++pat_ptr) {
1435                         case 0:
1436                                 goto out;
1437                         /* Double percent, output one percent */
1438                         case '%':
1439                                 if (out_ptr == out_end)
1440                                         goto out;
1441                                 *out_ptr++ = '%';
1442                                 break;
1443                         /* pid */
1444                         case 'p':
1445                                 pid_in_pattern = 1;
1446                                 rc = snprintf(out_ptr, out_end - out_ptr,
1447                                               "%d", task_tgid_vnr(current));
1448                                 if (rc > out_end - out_ptr)
1449                                         goto out;
1450                                 out_ptr += rc;
1451                                 break;
1452                         /* uid */
1453                         case 'u':
1454                                 rc = snprintf(out_ptr, out_end - out_ptr,
1455                                               "%d", cred->uid);
1456                                 if (rc > out_end - out_ptr)
1457                                         goto out;
1458                                 out_ptr += rc;
1459                                 break;
1460                         /* gid */
1461                         case 'g':
1462                                 rc = snprintf(out_ptr, out_end - out_ptr,
1463                                               "%d", cred->gid);
1464                                 if (rc > out_end - out_ptr)
1465                                         goto out;
1466                                 out_ptr += rc;
1467                                 break;
1468                         /* signal that caused the coredump */
1469                         case 's':
1470                                 rc = snprintf(out_ptr, out_end - out_ptr,
1471                                               "%ld", signr);
1472                                 if (rc > out_end - out_ptr)
1473                                         goto out;
1474                                 out_ptr += rc;
1475                                 break;
1476                         /* UNIX time of coredump */
1477                         case 't': {
1478                                 struct timeval tv;
1479                                 do_gettimeofday(&tv);
1480                                 rc = snprintf(out_ptr, out_end - out_ptr,
1481                                               "%lu", tv.tv_sec);
1482                                 if (rc > out_end - out_ptr)
1483                                         goto out;
1484                                 out_ptr += rc;
1485                                 break;
1486                         }
1487                         /* hostname */
1488                         case 'h':
1489                                 down_read(&uts_sem);
1490                                 rc = snprintf(out_ptr, out_end - out_ptr,
1491                                               "%s", utsname()->nodename);
1492                                 up_read(&uts_sem);
1493                                 if (rc > out_end - out_ptr)
1494                                         goto out;
1495                                 out_ptr += rc;
1496                                 break;
1497                         /* executable */
1498                         case 'e':
1499                                 rc = snprintf(out_ptr, out_end - out_ptr,
1500                                               "%s", current->comm);
1501                                 if (rc > out_end - out_ptr)
1502                                         goto out;
1503                                 out_ptr += rc;
1504                                 break;
1505                         /* core limit size */
1506                         case 'c':
1507                                 rc = snprintf(out_ptr, out_end - out_ptr,
1508                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1509                                 if (rc > out_end - out_ptr)
1510                                         goto out;
1511                                 out_ptr += rc;
1512                                 break;
1513                         default:
1514                                 break;
1515                         }
1516                         ++pat_ptr;
1517                 }
1518         }
1519         /* Backward compatibility with core_uses_pid:
1520          *
1521          * If core_pattern does not include a %p (as is the default)
1522          * and core_uses_pid is set, then .%pid will be appended to
1523          * the filename. Do not do this for piped commands. */
1524         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1525                 rc = snprintf(out_ptr, out_end - out_ptr,
1526                               ".%d", task_tgid_vnr(current));
1527                 if (rc > out_end - out_ptr)
1528                         goto out;
1529                 out_ptr += rc;
1530         }
1531 out:
1532         *out_ptr = 0;
1533         return ispipe;
1534 }
1535
1536 static int zap_process(struct task_struct *start)
1537 {
1538         struct task_struct *t;
1539         int nr = 0;
1540
1541         start->signal->flags = SIGNAL_GROUP_EXIT;
1542         start->signal->group_stop_count = 0;
1543
1544         t = start;
1545         do {
1546                 if (t != current && t->mm) {
1547                         sigaddset(&t->pending.signal, SIGKILL);
1548                         signal_wake_up(t, 1);
1549                         nr++;
1550                 }
1551         } while_each_thread(start, t);
1552
1553         return nr;
1554 }
1555
1556 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1557                                 struct core_state *core_state, int exit_code)
1558 {
1559         struct task_struct *g, *p;
1560         unsigned long flags;
1561         int nr = -EAGAIN;
1562
1563         spin_lock_irq(&tsk->sighand->siglock);
1564         if (!signal_group_exit(tsk->signal)) {
1565                 mm->core_state = core_state;
1566                 tsk->signal->group_exit_code = exit_code;
1567                 nr = zap_process(tsk);
1568         }
1569         spin_unlock_irq(&tsk->sighand->siglock);
1570         if (unlikely(nr < 0))
1571                 return nr;
1572
1573         if (atomic_read(&mm->mm_users) == nr + 1)
1574                 goto done;
1575         /*
1576          * We should find and kill all tasks which use this mm, and we should
1577          * count them correctly into ->nr_threads. We don't take tasklist
1578          * lock, but this is safe wrt:
1579          *
1580          * fork:
1581          *      None of sub-threads can fork after zap_process(leader). All
1582          *      processes which were created before this point should be
1583          *      visible to zap_threads() because copy_process() adds the new
1584          *      process to the tail of init_task.tasks list, and lock/unlock
1585          *      of ->siglock provides a memory barrier.
1586          *
1587          * do_exit:
1588          *      The caller holds mm->mmap_sem. This means that the task which
1589          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1590          *      its ->mm.
1591          *
1592          * de_thread:
1593          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1594          *      we must see either old or new leader, this does not matter.
1595          *      However, it can change p->sighand, so lock_task_sighand(p)
1596          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1597          *      it can't fail.
1598          *
1599          *      Note also that "g" can be the old leader with ->mm == NULL
1600          *      and already unhashed and thus removed from ->thread_group.
1601          *      This is OK, __unhash_process()->list_del_rcu() does not
1602          *      clear the ->next pointer, we will find the new leader via
1603          *      next_thread().
1604          */
1605         rcu_read_lock();
1606         for_each_process(g) {
1607                 if (g == tsk->group_leader)
1608                         continue;
1609                 if (g->flags & PF_KTHREAD)
1610                         continue;
1611                 p = g;
1612                 do {
1613                         if (p->mm) {
1614                                 if (unlikely(p->mm == mm)) {
1615                                         lock_task_sighand(p, &flags);
1616                                         nr += zap_process(p);
1617                                         unlock_task_sighand(p, &flags);
1618                                 }
1619                                 break;
1620                         }
1621                 } while_each_thread(g, p);
1622         }
1623         rcu_read_unlock();
1624 done:
1625         atomic_set(&core_state->nr_threads, nr);
1626         return nr;
1627 }
1628
1629 static int coredump_wait(int exit_code, struct core_state *core_state)
1630 {
1631         struct task_struct *tsk = current;
1632         struct mm_struct *mm = tsk->mm;
1633         struct completion *vfork_done;
1634         int core_waiters;
1635
1636         init_completion(&core_state->startup);
1637         core_state->dumper.task = tsk;
1638         core_state->dumper.next = NULL;
1639         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1640         up_write(&mm->mmap_sem);
1641
1642         if (unlikely(core_waiters < 0))
1643                 goto fail;
1644
1645         /*
1646          * Make sure nobody is waiting for us to release the VM,
1647          * otherwise we can deadlock when we wait on each other
1648          */
1649         vfork_done = tsk->vfork_done;
1650         if (vfork_done) {
1651                 tsk->vfork_done = NULL;
1652                 complete(vfork_done);
1653         }
1654
1655         if (core_waiters)
1656                 wait_for_completion(&core_state->startup);
1657 fail:
1658         return core_waiters;
1659 }
1660
1661 static void coredump_finish(struct mm_struct *mm)
1662 {
1663         struct core_thread *curr, *next;
1664         struct task_struct *task;
1665
1666         next = mm->core_state->dumper.next;
1667         while ((curr = next) != NULL) {
1668                 next = curr->next;
1669                 task = curr->task;
1670                 /*
1671                  * see exit_mm(), curr->task must not see
1672                  * ->task == NULL before we read ->next.
1673                  */
1674                 smp_mb();
1675                 curr->task = NULL;
1676                 wake_up_process(task);
1677         }
1678
1679         mm->core_state = NULL;
1680 }
1681
1682 /*
1683  * set_dumpable converts traditional three-value dumpable to two flags and
1684  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1685  * these bits are not changed atomically.  So get_dumpable can observe the
1686  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1687  * return either old dumpable or new one by paying attention to the order of
1688  * modifying the bits.
1689  *
1690  * dumpable |   mm->flags (binary)
1691  * old  new | initial interim  final
1692  * ---------+-----------------------
1693  *  0    1  |   00      01      01
1694  *  0    2  |   00      10(*)   11
1695  *  1    0  |   01      00      00
1696  *  1    2  |   01      11      11
1697  *  2    0  |   11      10(*)   00
1698  *  2    1  |   11      11      01
1699  *
1700  * (*) get_dumpable regards interim value of 10 as 11.
1701  */
1702 void set_dumpable(struct mm_struct *mm, int value)
1703 {
1704         switch (value) {
1705         case 0:
1706                 clear_bit(MMF_DUMPABLE, &mm->flags);
1707                 smp_wmb();
1708                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1709                 break;
1710         case 1:
1711                 set_bit(MMF_DUMPABLE, &mm->flags);
1712                 smp_wmb();
1713                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1714                 break;
1715         case 2:
1716                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1717                 smp_wmb();
1718                 set_bit(MMF_DUMPABLE, &mm->flags);
1719                 break;
1720         }
1721 }
1722
1723 int get_dumpable(struct mm_struct *mm)
1724 {
1725         int ret;
1726
1727         ret = mm->flags & 0x3;
1728         return (ret >= 2) ? 2 : ret;
1729 }
1730
1731 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1732 {
1733         struct core_state core_state;
1734         char corename[CORENAME_MAX_SIZE + 1];
1735         struct mm_struct *mm = current->mm;
1736         struct linux_binfmt * binfmt;
1737         struct inode * inode;
1738         struct file * file;
1739         const struct cred *old_cred;
1740         struct cred *cred;
1741         int retval = 0;
1742         int flag = 0;
1743         int ispipe = 0;
1744         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1745         char **helper_argv = NULL;
1746         int helper_argc = 0;
1747         char *delimit;
1748
1749         audit_core_dumps(signr);
1750
1751         binfmt = current->binfmt;
1752         if (!binfmt || !binfmt->core_dump)
1753                 goto fail;
1754
1755         cred = prepare_creds();
1756         if (!cred) {
1757                 retval = -ENOMEM;
1758                 goto fail;
1759         }
1760
1761         down_write(&mm->mmap_sem);
1762         /*
1763          * If another thread got here first, or we are not dumpable, bail out.
1764          */
1765         if (mm->core_state || !get_dumpable(mm)) {
1766                 up_write(&mm->mmap_sem);
1767                 put_cred(cred);
1768                 goto fail;
1769         }
1770
1771         /*
1772          *      We cannot trust fsuid as being the "true" uid of the
1773          *      process nor do we know its entire history. We only know it
1774          *      was tainted so we dump it as root in mode 2.
1775          */
1776         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1777                 flag = O_EXCL;          /* Stop rewrite attacks */
1778                 cred->fsuid = 0;        /* Dump root private */
1779         }
1780
1781         retval = coredump_wait(exit_code, &core_state);
1782         if (retval < 0) {
1783                 put_cred(cred);
1784                 goto fail;
1785         }
1786
1787         old_cred = override_creds(cred);
1788
1789         /*
1790          * Clear any false indication of pending signals that might
1791          * be seen by the filesystem code called to write the core file.
1792          */
1793         clear_thread_flag(TIF_SIGPENDING);
1794
1795         /*
1796          * lock_kernel() because format_corename() is controlled by sysctl, which
1797          * uses lock_kernel()
1798          */
1799         lock_kernel();
1800         ispipe = format_corename(corename, signr);
1801         unlock_kernel();
1802         /*
1803          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1804          * to a pipe.  Since we're not writing directly to the filesystem
1805          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1806          * created unless the pipe reader choses to write out the core file
1807          * at which point file size limits and permissions will be imposed
1808          * as it does with any other process
1809          */
1810         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1811                 goto fail_unlock;
1812
1813         if (ispipe) {
1814                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1815                 /* Terminate the string before the first option */
1816                 delimit = strchr(corename, ' ');
1817                 if (delimit)
1818                         *delimit = '\0';
1819                 delimit = strrchr(helper_argv[0], '/');
1820                 if (delimit)
1821                         delimit++;
1822                 else
1823                         delimit = helper_argv[0];
1824                 if (!strcmp(delimit, current->comm)) {
1825                         printk(KERN_NOTICE "Recursive core dump detected, "
1826                                         "aborting\n");
1827                         goto fail_unlock;
1828                 }
1829
1830                 core_limit = RLIM_INFINITY;
1831
1832                 /* SIGPIPE can happen, but it's just never processed */
1833                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1834                                 &file)) {
1835                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1836                                corename);
1837                         goto fail_unlock;
1838                 }
1839         } else
1840                 file = filp_open(corename,
1841                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1842                                  0600);
1843         if (IS_ERR(file))
1844                 goto fail_unlock;
1845         inode = file->f_path.dentry->d_inode;
1846         if (inode->i_nlink > 1)
1847                 goto close_fail;        /* multiple links - don't dump */
1848         if (!ispipe && d_unhashed(file->f_path.dentry))
1849                 goto close_fail;
1850
1851         /* AK: actually i see no reason to not allow this for named pipes etc.,
1852            but keep the previous behaviour for now. */
1853         if (!ispipe && !S_ISREG(inode->i_mode))
1854                 goto close_fail;
1855         /*
1856          * Dont allow local users get cute and trick others to coredump
1857          * into their pre-created files:
1858          */
1859         if (inode->i_uid != current_fsuid())
1860                 goto close_fail;
1861         if (!file->f_op)
1862                 goto close_fail;
1863         if (!file->f_op->write)
1864                 goto close_fail;
1865         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1866                 goto close_fail;
1867
1868         retval = binfmt->core_dump(signr, regs, file, core_limit);
1869
1870         if (retval)
1871                 current->signal->group_exit_code |= 0x80;
1872 close_fail:
1873         filp_close(file, NULL);
1874 fail_unlock:
1875         if (helper_argv)
1876                 argv_free(helper_argv);
1877
1878         revert_creds(old_cred);
1879         put_cred(cred);
1880         coredump_finish(mm);
1881 fail:
1882         return retval;
1883 }