Merge commit 'gcl/gcl-next'
[linux-2.6] / fs / exec.c
1 /*
2  *  linux/fs/exec.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  * #!-checking implemented by tytso.
9  */
10 /*
11  * Demand-loading implemented 01.12.91 - no need to read anything but
12  * the header into memory. The inode of the executable is put into
13  * "current->executable", and page faults do the actual loading. Clean.
14  *
15  * Once more I can proudly say that linux stood up to being changed: it
16  * was less than 2 hours work to get demand-loading completely implemented.
17  *
18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
19  * current->executable is only used by the procfs.  This allows a dispatch
20  * table to check for several different types  of binary formats.  We keep
21  * trying until we recognize the file or we run out of supported binary
22  * formats. 
23  */
24
25 #include <linux/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/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/utsname.h>
41 #include <linux/pid_namespace.h>
42 #include <linux/module.h>
43 #include <linux/namei.h>
44 #include <linux/proc_fs.h>
45 #include <linux/mount.h>
46 #include <linux/security.h>
47 #include <linux/syscalls.h>
48 #include <linux/tsacct_kern.h>
49 #include <linux/cn_proc.h>
50 #include <linux/audit.h>
51 #include <linux/tracehook.h>
52
53 #include <asm/uaccess.h>
54 #include <asm/mmu_context.h>
55 #include <asm/tlb.h>
56
57 #ifdef CONFIG_KMOD
58 #include <linux/kmod.h>
59 #endif
60
61 #ifdef __alpha__
62 /* for /sbin/loader handling in search_binary_handler() */
63 #include <linux/a.out.h>
64 #endif
65
66 int core_uses_pid;
67 char core_pattern[CORENAME_MAX_SIZE] = "core";
68 int suid_dumpable = 0;
69
70 /* The maximal length of core_pattern is also specified in sysctl.c */
71
72 static LIST_HEAD(formats);
73 static DEFINE_RWLOCK(binfmt_lock);
74
75 int register_binfmt(struct linux_binfmt * fmt)
76 {
77         if (!fmt)
78                 return -EINVAL;
79         write_lock(&binfmt_lock);
80         list_add(&fmt->lh, &formats);
81         write_unlock(&binfmt_lock);
82         return 0;       
83 }
84
85 EXPORT_SYMBOL(register_binfmt);
86
87 void unregister_binfmt(struct linux_binfmt * fmt)
88 {
89         write_lock(&binfmt_lock);
90         list_del(&fmt->lh);
91         write_unlock(&binfmt_lock);
92 }
93
94 EXPORT_SYMBOL(unregister_binfmt);
95
96 static inline void put_binfmt(struct linux_binfmt * fmt)
97 {
98         module_put(fmt->module);
99 }
100
101 /*
102  * Note that a shared library must be both readable and executable due to
103  * security reasons.
104  *
105  * Also note that we take the address to load from from the file itself.
106  */
107 asmlinkage long sys_uselib(const char __user * library)
108 {
109         struct file *file;
110         struct nameidata nd;
111         char *tmp = getname(library);
112         int error = PTR_ERR(tmp);
113
114         if (!IS_ERR(tmp)) {
115                 error = path_lookup_open(AT_FDCWD, tmp,
116                                          LOOKUP_FOLLOW, &nd,
117                                          FMODE_READ|FMODE_EXEC);
118                 putname(tmp);
119         }
120         if (error)
121                 goto out;
122
123         error = -EINVAL;
124         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
125                 goto exit;
126
127         error = -EACCES;
128         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
129                 goto exit;
130
131         error = vfs_permission(&nd, 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 = vfs_permission(&nd, 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         mm_update_next_owner(old_mm);
755         arch_pick_mmap_layout(mm);
756         if (old_mm) {
757                 up_read(&old_mm->mmap_sem);
758                 BUG_ON(active_mm != 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         struct task_struct *leader = NULL;
778         int count;
779
780         if (thread_group_empty(tsk))
781                 goto no_thread_group;
782
783         /*
784          * Kill all other threads in the thread group.
785          */
786         spin_lock_irq(lock);
787         if (signal_group_exit(sig)) {
788                 /*
789                  * Another group action in progress, just
790                  * return so that the signal is processed.
791                  */
792                 spin_unlock_irq(lock);
793                 return -EAGAIN;
794         }
795         sig->group_exit_task = tsk;
796         zap_other_threads(tsk);
797
798         /* Account for the thread group leader hanging around: */
799         count = thread_group_leader(tsk) ? 1 : 2;
800         sig->notify_count = count;
801         while (atomic_read(&sig->count) > count) {
802                 __set_current_state(TASK_UNINTERRUPTIBLE);
803                 spin_unlock_irq(lock);
804                 schedule();
805                 spin_lock_irq(lock);
806         }
807         spin_unlock_irq(lock);
808
809         /*
810          * At this point all other threads have exited, all we have to
811          * do is to wait for the thread group leader to become inactive,
812          * and to assume its PID:
813          */
814         if (!thread_group_leader(tsk)) {
815                 leader = tsk->group_leader;
816
817                 sig->notify_count = -1; /* for exit_notify() */
818                 for (;;) {
819                         write_lock_irq(&tasklist_lock);
820                         if (likely(leader->exit_state))
821                                 break;
822                         __set_current_state(TASK_UNINTERRUPTIBLE);
823                         write_unlock_irq(&tasklist_lock);
824                         schedule();
825                 }
826
827                 if (unlikely(task_child_reaper(tsk) == leader))
828                         task_active_pid_ns(tsk)->child_reaper = tsk;
829                 /*
830                  * The only record we have of the real-time age of a
831                  * process, regardless of execs it's done, is start_time.
832                  * All the past CPU time is accumulated in signal_struct
833                  * from sister threads now dead.  But in this non-leader
834                  * exec, nothing survives from the original leader thread,
835                  * whose birth marks the true age of this process now.
836                  * When we take on its identity by switching to its PID, we
837                  * also take its birthdate (always earlier than our own).
838                  */
839                 tsk->start_time = leader->start_time;
840
841                 BUG_ON(!same_thread_group(leader, tsk));
842                 BUG_ON(has_group_leader_pid(tsk));
843                 /*
844                  * An exec() starts a new thread group with the
845                  * TGID of the previous thread group. Rehash the
846                  * two threads with a switched PID, and release
847                  * the former thread group leader:
848                  */
849
850                 /* Become a process group leader with the old leader's pid.
851                  * The old leader becomes a thread of the this thread group.
852                  * Note: The old leader also uses this pid until release_task
853                  *       is called.  Odd but simple and correct.
854                  */
855                 detach_pid(tsk, PIDTYPE_PID);
856                 tsk->pid = leader->pid;
857                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
858                 transfer_pid(leader, tsk, PIDTYPE_PGID);
859                 transfer_pid(leader, tsk, PIDTYPE_SID);
860                 list_replace_rcu(&leader->tasks, &tsk->tasks);
861
862                 tsk->group_leader = tsk;
863                 leader->group_leader = tsk;
864
865                 tsk->exit_signal = SIGCHLD;
866
867                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
868                 leader->exit_state = EXIT_DEAD;
869
870                 write_unlock_irq(&tasklist_lock);
871         }
872
873         sig->group_exit_task = NULL;
874         sig->notify_count = 0;
875
876 no_thread_group:
877         exit_itimers(sig);
878         flush_itimer_signals();
879         if (leader)
880                 release_task(leader);
881
882         if (atomic_read(&oldsighand->count) != 1) {
883                 struct sighand_struct *newsighand;
884                 /*
885                  * This ->sighand is shared with the CLONE_SIGHAND
886                  * but not CLONE_THREAD task, switch to the new one.
887                  */
888                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
889                 if (!newsighand)
890                         return -ENOMEM;
891
892                 atomic_set(&newsighand->count, 1);
893                 memcpy(newsighand->action, oldsighand->action,
894                        sizeof(newsighand->action));
895
896                 write_lock_irq(&tasklist_lock);
897                 spin_lock(&oldsighand->siglock);
898                 rcu_assign_pointer(tsk->sighand, newsighand);
899                 spin_unlock(&oldsighand->siglock);
900                 write_unlock_irq(&tasklist_lock);
901
902                 __cleanup_sighand(oldsighand);
903         }
904
905         BUG_ON(!thread_group_leader(tsk));
906         return 0;
907 }
908
909 /*
910  * These functions flushes out all traces of the currently running executable
911  * so that a new one can be started
912  */
913 static void flush_old_files(struct files_struct * files)
914 {
915         long j = -1;
916         struct fdtable *fdt;
917
918         spin_lock(&files->file_lock);
919         for (;;) {
920                 unsigned long set, i;
921
922                 j++;
923                 i = j * __NFDBITS;
924                 fdt = files_fdtable(files);
925                 if (i >= fdt->max_fds)
926                         break;
927                 set = fdt->close_on_exec->fds_bits[j];
928                 if (!set)
929                         continue;
930                 fdt->close_on_exec->fds_bits[j] = 0;
931                 spin_unlock(&files->file_lock);
932                 for ( ; set ; i++,set >>= 1) {
933                         if (set & 1) {
934                                 sys_close(i);
935                         }
936                 }
937                 spin_lock(&files->file_lock);
938
939         }
940         spin_unlock(&files->file_lock);
941 }
942
943 char *get_task_comm(char *buf, struct task_struct *tsk)
944 {
945         /* buf must be at least sizeof(tsk->comm) in size */
946         task_lock(tsk);
947         strncpy(buf, tsk->comm, sizeof(tsk->comm));
948         task_unlock(tsk);
949         return buf;
950 }
951
952 void set_task_comm(struct task_struct *tsk, char *buf)
953 {
954         task_lock(tsk);
955         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
956         task_unlock(tsk);
957 }
958
959 int flush_old_exec(struct linux_binprm * bprm)
960 {
961         char * name;
962         int i, ch, retval;
963         char tcomm[sizeof(current->comm)];
964
965         /*
966          * Make sure we have a private signal table and that
967          * we are unassociated from the previous thread group.
968          */
969         retval = de_thread(current);
970         if (retval)
971                 goto out;
972
973         set_mm_exe_file(bprm->mm, bprm->file);
974
975         /*
976          * Release all of the old mmap stuff
977          */
978         retval = exec_mmap(bprm->mm);
979         if (retval)
980                 goto out;
981
982         bprm->mm = NULL;                /* We're using it now */
983
984         /* This is the point of no return */
985         current->sas_ss_sp = current->sas_ss_size = 0;
986
987         if (current->euid == current->uid && current->egid == current->gid)
988                 set_dumpable(current->mm, 1);
989         else
990                 set_dumpable(current->mm, suid_dumpable);
991
992         name = bprm->filename;
993
994         /* Copies the binary name from after last slash */
995         for (i=0; (ch = *(name++)) != '\0';) {
996                 if (ch == '/')
997                         i = 0; /* overwrite what we wrote */
998                 else
999                         if (i < (sizeof(tcomm) - 1))
1000                                 tcomm[i++] = ch;
1001         }
1002         tcomm[i] = '\0';
1003         set_task_comm(current, tcomm);
1004
1005         current->flags &= ~PF_RANDOMIZE;
1006         flush_thread();
1007
1008         /* Set the new mm task size. We have to do that late because it may
1009          * depend on TIF_32BIT which is only updated in flush_thread() on
1010          * some architectures like powerpc
1011          */
1012         current->mm->task_size = TASK_SIZE;
1013
1014         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1015                 suid_keys(current);
1016                 set_dumpable(current->mm, suid_dumpable);
1017                 current->pdeath_signal = 0;
1018         } else if (file_permission(bprm->file, MAY_READ) ||
1019                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1020                 suid_keys(current);
1021                 set_dumpable(current->mm, suid_dumpable);
1022         }
1023
1024         /* An exec changes our domain. We are no longer part of the thread
1025            group */
1026
1027         current->self_exec_id++;
1028                         
1029         flush_signal_handlers(current, 0);
1030         flush_old_files(current->files);
1031
1032         return 0;
1033
1034 out:
1035         return retval;
1036 }
1037
1038 EXPORT_SYMBOL(flush_old_exec);
1039
1040 /* 
1041  * Fill the binprm structure from the inode. 
1042  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1043  */
1044 int prepare_binprm(struct linux_binprm *bprm)
1045 {
1046         int mode;
1047         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1048         int retval;
1049
1050         mode = inode->i_mode;
1051         if (bprm->file->f_op == NULL)
1052                 return -EACCES;
1053
1054         bprm->e_uid = current->euid;
1055         bprm->e_gid = current->egid;
1056
1057         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1058                 /* Set-uid? */
1059                 if (mode & S_ISUID) {
1060                         current->personality &= ~PER_CLEAR_ON_SETID;
1061                         bprm->e_uid = inode->i_uid;
1062                 }
1063
1064                 /* Set-gid? */
1065                 /*
1066                  * If setgid is set but no group execute bit then this
1067                  * is a candidate for mandatory locking, not a setgid
1068                  * executable.
1069                  */
1070                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1071                         current->personality &= ~PER_CLEAR_ON_SETID;
1072                         bprm->e_gid = inode->i_gid;
1073                 }
1074         }
1075
1076         /* fill in binprm security blob */
1077         retval = security_bprm_set(bprm);
1078         if (retval)
1079                 return retval;
1080
1081         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1082         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1083 }
1084
1085 EXPORT_SYMBOL(prepare_binprm);
1086
1087 static int unsafe_exec(struct task_struct *p)
1088 {
1089         int unsafe = tracehook_unsafe_exec(p);
1090
1091         if (atomic_read(&p->fs->count) > 1 ||
1092             atomic_read(&p->files->count) > 1 ||
1093             atomic_read(&p->sighand->count) > 1)
1094                 unsafe |= LSM_UNSAFE_SHARE;
1095
1096         return unsafe;
1097 }
1098
1099 void compute_creds(struct linux_binprm *bprm)
1100 {
1101         int unsafe;
1102
1103         if (bprm->e_uid != current->uid) {
1104                 suid_keys(current);
1105                 current->pdeath_signal = 0;
1106         }
1107         exec_keys(current);
1108
1109         task_lock(current);
1110         unsafe = unsafe_exec(current);
1111         security_bprm_apply_creds(bprm, unsafe);
1112         task_unlock(current);
1113         security_bprm_post_apply_creds(bprm);
1114 }
1115 EXPORT_SYMBOL(compute_creds);
1116
1117 /*
1118  * Arguments are '\0' separated strings found at the location bprm->p
1119  * points to; chop off the first by relocating brpm->p to right after
1120  * the first '\0' encountered.
1121  */
1122 int remove_arg_zero(struct linux_binprm *bprm)
1123 {
1124         int ret = 0;
1125         unsigned long offset;
1126         char *kaddr;
1127         struct page *page;
1128
1129         if (!bprm->argc)
1130                 return 0;
1131
1132         do {
1133                 offset = bprm->p & ~PAGE_MASK;
1134                 page = get_arg_page(bprm, bprm->p, 0);
1135                 if (!page) {
1136                         ret = -EFAULT;
1137                         goto out;
1138                 }
1139                 kaddr = kmap_atomic(page, KM_USER0);
1140
1141                 for (; offset < PAGE_SIZE && kaddr[offset];
1142                                 offset++, bprm->p++)
1143                         ;
1144
1145                 kunmap_atomic(kaddr, KM_USER0);
1146                 put_arg_page(page);
1147
1148                 if (offset == PAGE_SIZE)
1149                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1150         } while (offset == PAGE_SIZE);
1151
1152         bprm->p++;
1153         bprm->argc--;
1154         ret = 0;
1155
1156 out:
1157         return ret;
1158 }
1159 EXPORT_SYMBOL(remove_arg_zero);
1160
1161 /*
1162  * cycle the list of binary formats handler, until one recognizes the image
1163  */
1164 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1165 {
1166         int try,retval;
1167         struct linux_binfmt *fmt;
1168 #ifdef __alpha__
1169         /* handle /sbin/loader.. */
1170         {
1171             struct exec * eh = (struct exec *) bprm->buf;
1172
1173             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1174                 (eh->fh.f_flags & 0x3000) == 0x3000)
1175             {
1176                 struct file * file;
1177                 unsigned long loader;
1178
1179                 allow_write_access(bprm->file);
1180                 fput(bprm->file);
1181                 bprm->file = NULL;
1182
1183                 loader = bprm->vma->vm_end - sizeof(void *);
1184
1185                 file = open_exec("/sbin/loader");
1186                 retval = PTR_ERR(file);
1187                 if (IS_ERR(file))
1188                         return retval;
1189
1190                 /* Remember if the application is TASO.  */
1191                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1192
1193                 bprm->file = file;
1194                 bprm->loader = loader;
1195                 retval = prepare_binprm(bprm);
1196                 if (retval<0)
1197                         return retval;
1198                 /* should call search_binary_handler recursively here,
1199                    but it does not matter */
1200             }
1201         }
1202 #endif
1203         retval = security_bprm_check(bprm);
1204         if (retval)
1205                 return retval;
1206
1207         /* kernel module loader fixup */
1208         /* so we don't try to load run modprobe in kernel space. */
1209         set_fs(USER_DS);
1210
1211         retval = audit_bprm(bprm);
1212         if (retval)
1213                 return retval;
1214
1215         retval = -ENOENT;
1216         for (try=0; try<2; try++) {
1217                 read_lock(&binfmt_lock);
1218                 list_for_each_entry(fmt, &formats, lh) {
1219                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1220                         if (!fn)
1221                                 continue;
1222                         if (!try_module_get(fmt->module))
1223                                 continue;
1224                         read_unlock(&binfmt_lock);
1225                         retval = fn(bprm, regs);
1226                         if (retval >= 0) {
1227                                 tracehook_report_exec(fmt, bprm, regs);
1228                                 put_binfmt(fmt);
1229                                 allow_write_access(bprm->file);
1230                                 if (bprm->file)
1231                                         fput(bprm->file);
1232                                 bprm->file = NULL;
1233                                 current->did_exec = 1;
1234                                 proc_exec_connector(current);
1235                                 return retval;
1236                         }
1237                         read_lock(&binfmt_lock);
1238                         put_binfmt(fmt);
1239                         if (retval != -ENOEXEC || bprm->mm == NULL)
1240                                 break;
1241                         if (!bprm->file) {
1242                                 read_unlock(&binfmt_lock);
1243                                 return retval;
1244                         }
1245                 }
1246                 read_unlock(&binfmt_lock);
1247                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1248                         break;
1249 #ifdef CONFIG_KMOD
1250                 }else{
1251 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1252                         if (printable(bprm->buf[0]) &&
1253                             printable(bprm->buf[1]) &&
1254                             printable(bprm->buf[2]) &&
1255                             printable(bprm->buf[3]))
1256                                 break; /* -ENOEXEC */
1257                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1258 #endif
1259                 }
1260         }
1261         return retval;
1262 }
1263
1264 EXPORT_SYMBOL(search_binary_handler);
1265
1266 void free_bprm(struct linux_binprm *bprm)
1267 {
1268         free_arg_pages(bprm);
1269         kfree(bprm);
1270 }
1271
1272 /*
1273  * sys_execve() executes a new program.
1274  */
1275 int do_execve(char * filename,
1276         char __user *__user *argv,
1277         char __user *__user *envp,
1278         struct pt_regs * regs)
1279 {
1280         struct linux_binprm *bprm;
1281         struct file *file;
1282         struct files_struct *displaced;
1283         int retval;
1284
1285         retval = unshare_files(&displaced);
1286         if (retval)
1287                 goto out_ret;
1288
1289         retval = -ENOMEM;
1290         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1291         if (!bprm)
1292                 goto out_files;
1293
1294         file = open_exec(filename);
1295         retval = PTR_ERR(file);
1296         if (IS_ERR(file))
1297                 goto out_kfree;
1298
1299         sched_exec();
1300
1301         bprm->file = file;
1302         bprm->filename = filename;
1303         bprm->interp = filename;
1304
1305         retval = bprm_mm_init(bprm);
1306         if (retval)
1307                 goto out_file;
1308
1309         bprm->argc = count(argv, MAX_ARG_STRINGS);
1310         if ((retval = bprm->argc) < 0)
1311                 goto out_mm;
1312
1313         bprm->envc = count(envp, MAX_ARG_STRINGS);
1314         if ((retval = bprm->envc) < 0)
1315                 goto out_mm;
1316
1317         retval = security_bprm_alloc(bprm);
1318         if (retval)
1319                 goto out;
1320
1321         retval = prepare_binprm(bprm);
1322         if (retval < 0)
1323                 goto out;
1324
1325         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1326         if (retval < 0)
1327                 goto out;
1328
1329         bprm->exec = bprm->p;
1330         retval = copy_strings(bprm->envc, envp, bprm);
1331         if (retval < 0)
1332                 goto out;
1333
1334         retval = copy_strings(bprm->argc, argv, bprm);
1335         if (retval < 0)
1336                 goto out;
1337
1338         current->flags &= ~PF_KTHREAD;
1339         retval = search_binary_handler(bprm,regs);
1340         if (retval >= 0) {
1341                 /* execve success */
1342                 security_bprm_free(bprm);
1343                 acct_update_integrals(current);
1344                 free_bprm(bprm);
1345                 if (displaced)
1346                         put_files_struct(displaced);
1347                 return retval;
1348         }
1349
1350 out:
1351         if (bprm->security)
1352                 security_bprm_free(bprm);
1353
1354 out_mm:
1355         if (bprm->mm)
1356                 mmput (bprm->mm);
1357
1358 out_file:
1359         if (bprm->file) {
1360                 allow_write_access(bprm->file);
1361                 fput(bprm->file);
1362         }
1363 out_kfree:
1364         free_bprm(bprm);
1365
1366 out_files:
1367         if (displaced)
1368                 reset_files_struct(displaced);
1369 out_ret:
1370         return retval;
1371 }
1372
1373 int set_binfmt(struct linux_binfmt *new)
1374 {
1375         struct linux_binfmt *old = current->binfmt;
1376
1377         if (new) {
1378                 if (!try_module_get(new->module))
1379                         return -1;
1380         }
1381         current->binfmt = new;
1382         if (old)
1383                 module_put(old->module);
1384         return 0;
1385 }
1386
1387 EXPORT_SYMBOL(set_binfmt);
1388
1389 /* format_corename will inspect the pattern parameter, and output a
1390  * name into corename, which must have space for at least
1391  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1392  */
1393 static int format_corename(char *corename, int nr_threads, long signr)
1394 {
1395         const char *pat_ptr = core_pattern;
1396         int ispipe = (*pat_ptr == '|');
1397         char *out_ptr = corename;
1398         char *const out_end = corename + CORENAME_MAX_SIZE;
1399         int rc;
1400         int pid_in_pattern = 0;
1401
1402         /* Repeat as long as we have more pattern to process and more output
1403            space */
1404         while (*pat_ptr) {
1405                 if (*pat_ptr != '%') {
1406                         if (out_ptr == out_end)
1407                                 goto out;
1408                         *out_ptr++ = *pat_ptr++;
1409                 } else {
1410                         switch (*++pat_ptr) {
1411                         case 0:
1412                                 goto out;
1413                         /* Double percent, output one percent */
1414                         case '%':
1415                                 if (out_ptr == out_end)
1416                                         goto out;
1417                                 *out_ptr++ = '%';
1418                                 break;
1419                         /* pid */
1420                         case 'p':
1421                                 pid_in_pattern = 1;
1422                                 rc = snprintf(out_ptr, out_end - out_ptr,
1423                                               "%d", task_tgid_vnr(current));
1424                                 if (rc > out_end - out_ptr)
1425                                         goto out;
1426                                 out_ptr += rc;
1427                                 break;
1428                         /* uid */
1429                         case 'u':
1430                                 rc = snprintf(out_ptr, out_end - out_ptr,
1431                                               "%d", current->uid);
1432                                 if (rc > out_end - out_ptr)
1433                                         goto out;
1434                                 out_ptr += rc;
1435                                 break;
1436                         /* gid */
1437                         case 'g':
1438                                 rc = snprintf(out_ptr, out_end - out_ptr,
1439                                               "%d", current->gid);
1440                                 if (rc > out_end - out_ptr)
1441                                         goto out;
1442                                 out_ptr += rc;
1443                                 break;
1444                         /* signal that caused the coredump */
1445                         case 's':
1446                                 rc = snprintf(out_ptr, out_end - out_ptr,
1447                                               "%ld", signr);
1448                                 if (rc > out_end - out_ptr)
1449                                         goto out;
1450                                 out_ptr += rc;
1451                                 break;
1452                         /* UNIX time of coredump */
1453                         case 't': {
1454                                 struct timeval tv;
1455                                 do_gettimeofday(&tv);
1456                                 rc = snprintf(out_ptr, out_end - out_ptr,
1457                                               "%lu", tv.tv_sec);
1458                                 if (rc > out_end - out_ptr)
1459                                         goto out;
1460                                 out_ptr += rc;
1461                                 break;
1462                         }
1463                         /* hostname */
1464                         case 'h':
1465                                 down_read(&uts_sem);
1466                                 rc = snprintf(out_ptr, out_end - out_ptr,
1467                                               "%s", utsname()->nodename);
1468                                 up_read(&uts_sem);
1469                                 if (rc > out_end - out_ptr)
1470                                         goto out;
1471                                 out_ptr += rc;
1472                                 break;
1473                         /* executable */
1474                         case 'e':
1475                                 rc = snprintf(out_ptr, out_end - out_ptr,
1476                                               "%s", current->comm);
1477                                 if (rc > out_end - out_ptr)
1478                                         goto out;
1479                                 out_ptr += rc;
1480                                 break;
1481                         /* core limit size */
1482                         case 'c':
1483                                 rc = snprintf(out_ptr, out_end - out_ptr,
1484                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1485                                 if (rc > out_end - out_ptr)
1486                                         goto out;
1487                                 out_ptr += rc;
1488                                 break;
1489                         default:
1490                                 break;
1491                         }
1492                         ++pat_ptr;
1493                 }
1494         }
1495         /* Backward compatibility with core_uses_pid:
1496          *
1497          * If core_pattern does not include a %p (as is the default)
1498          * and core_uses_pid is set, then .%pid will be appended to
1499          * the filename. Do not do this for piped commands. */
1500         if (!ispipe && !pid_in_pattern
1501             && (core_uses_pid || nr_threads)) {
1502                 rc = snprintf(out_ptr, out_end - out_ptr,
1503                               ".%d", task_tgid_vnr(current));
1504                 if (rc > out_end - out_ptr)
1505                         goto out;
1506                 out_ptr += rc;
1507         }
1508 out:
1509         *out_ptr = 0;
1510         return ispipe;
1511 }
1512
1513 static int zap_process(struct task_struct *start)
1514 {
1515         struct task_struct *t;
1516         int nr = 0;
1517
1518         start->signal->flags = SIGNAL_GROUP_EXIT;
1519         start->signal->group_stop_count = 0;
1520
1521         t = start;
1522         do {
1523                 if (t != current && t->mm) {
1524                         sigaddset(&t->pending.signal, SIGKILL);
1525                         signal_wake_up(t, 1);
1526                         nr++;
1527                 }
1528         } while_each_thread(start, t);
1529
1530         return nr;
1531 }
1532
1533 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1534                                 struct core_state *core_state, int exit_code)
1535 {
1536         struct task_struct *g, *p;
1537         unsigned long flags;
1538         int nr = -EAGAIN;
1539
1540         spin_lock_irq(&tsk->sighand->siglock);
1541         if (!signal_group_exit(tsk->signal)) {
1542                 mm->core_state = core_state;
1543                 tsk->signal->group_exit_code = exit_code;
1544                 nr = zap_process(tsk);
1545         }
1546         spin_unlock_irq(&tsk->sighand->siglock);
1547         if (unlikely(nr < 0))
1548                 return nr;
1549
1550         if (atomic_read(&mm->mm_users) == nr + 1)
1551                 goto done;
1552         /*
1553          * We should find and kill all tasks which use this mm, and we should
1554          * count them correctly into ->nr_threads. We don't take tasklist
1555          * lock, but this is safe wrt:
1556          *
1557          * fork:
1558          *      None of sub-threads can fork after zap_process(leader). All
1559          *      processes which were created before this point should be
1560          *      visible to zap_threads() because copy_process() adds the new
1561          *      process to the tail of init_task.tasks list, and lock/unlock
1562          *      of ->siglock provides a memory barrier.
1563          *
1564          * do_exit:
1565          *      The caller holds mm->mmap_sem. This means that the task which
1566          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1567          *      its ->mm.
1568          *
1569          * de_thread:
1570          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1571          *      we must see either old or new leader, this does not matter.
1572          *      However, it can change p->sighand, so lock_task_sighand(p)
1573          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1574          *      it can't fail.
1575          *
1576          *      Note also that "g" can be the old leader with ->mm == NULL
1577          *      and already unhashed and thus removed from ->thread_group.
1578          *      This is OK, __unhash_process()->list_del_rcu() does not
1579          *      clear the ->next pointer, we will find the new leader via
1580          *      next_thread().
1581          */
1582         rcu_read_lock();
1583         for_each_process(g) {
1584                 if (g == tsk->group_leader)
1585                         continue;
1586                 if (g->flags & PF_KTHREAD)
1587                         continue;
1588                 p = g;
1589                 do {
1590                         if (p->mm) {
1591                                 if (unlikely(p->mm == mm)) {
1592                                         lock_task_sighand(p, &flags);
1593                                         nr += zap_process(p);
1594                                         unlock_task_sighand(p, &flags);
1595                                 }
1596                                 break;
1597                         }
1598                 } while_each_thread(g, p);
1599         }
1600         rcu_read_unlock();
1601 done:
1602         atomic_set(&core_state->nr_threads, nr);
1603         return nr;
1604 }
1605
1606 static int coredump_wait(int exit_code, struct core_state *core_state)
1607 {
1608         struct task_struct *tsk = current;
1609         struct mm_struct *mm = tsk->mm;
1610         struct completion *vfork_done;
1611         int core_waiters;
1612
1613         init_completion(&core_state->startup);
1614         core_state->dumper.task = tsk;
1615         core_state->dumper.next = NULL;
1616         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1617         up_write(&mm->mmap_sem);
1618
1619         if (unlikely(core_waiters < 0))
1620                 goto fail;
1621
1622         /*
1623          * Make sure nobody is waiting for us to release the VM,
1624          * otherwise we can deadlock when we wait on each other
1625          */
1626         vfork_done = tsk->vfork_done;
1627         if (vfork_done) {
1628                 tsk->vfork_done = NULL;
1629                 complete(vfork_done);
1630         }
1631
1632         if (core_waiters)
1633                 wait_for_completion(&core_state->startup);
1634 fail:
1635         return core_waiters;
1636 }
1637
1638 static void coredump_finish(struct mm_struct *mm)
1639 {
1640         struct core_thread *curr, *next;
1641         struct task_struct *task;
1642
1643         next = mm->core_state->dumper.next;
1644         while ((curr = next) != NULL) {
1645                 next = curr->next;
1646                 task = curr->task;
1647                 /*
1648                  * see exit_mm(), curr->task must not see
1649                  * ->task == NULL before we read ->next.
1650                  */
1651                 smp_mb();
1652                 curr->task = NULL;
1653                 wake_up_process(task);
1654         }
1655
1656         mm->core_state = NULL;
1657 }
1658
1659 /*
1660  * set_dumpable converts traditional three-value dumpable to two flags and
1661  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1662  * these bits are not changed atomically.  So get_dumpable can observe the
1663  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1664  * return either old dumpable or new one by paying attention to the order of
1665  * modifying the bits.
1666  *
1667  * dumpable |   mm->flags (binary)
1668  * old  new | initial interim  final
1669  * ---------+-----------------------
1670  *  0    1  |   00      01      01
1671  *  0    2  |   00      10(*)   11
1672  *  1    0  |   01      00      00
1673  *  1    2  |   01      11      11
1674  *  2    0  |   11      10(*)   00
1675  *  2    1  |   11      11      01
1676  *
1677  * (*) get_dumpable regards interim value of 10 as 11.
1678  */
1679 void set_dumpable(struct mm_struct *mm, int value)
1680 {
1681         switch (value) {
1682         case 0:
1683                 clear_bit(MMF_DUMPABLE, &mm->flags);
1684                 smp_wmb();
1685                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1686                 break;
1687         case 1:
1688                 set_bit(MMF_DUMPABLE, &mm->flags);
1689                 smp_wmb();
1690                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1691                 break;
1692         case 2:
1693                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1694                 smp_wmb();
1695                 set_bit(MMF_DUMPABLE, &mm->flags);
1696                 break;
1697         }
1698 }
1699
1700 int get_dumpable(struct mm_struct *mm)
1701 {
1702         int ret;
1703
1704         ret = mm->flags & 0x3;
1705         return (ret >= 2) ? 2 : ret;
1706 }
1707
1708 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1709 {
1710         struct core_state core_state;
1711         char corename[CORENAME_MAX_SIZE + 1];
1712         struct mm_struct *mm = current->mm;
1713         struct linux_binfmt * binfmt;
1714         struct inode * inode;
1715         struct file * file;
1716         int retval = 0;
1717         int fsuid = current->fsuid;
1718         int flag = 0;
1719         int ispipe = 0;
1720         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1721         char **helper_argv = NULL;
1722         int helper_argc = 0;
1723         char *delimit;
1724
1725         audit_core_dumps(signr);
1726
1727         binfmt = current->binfmt;
1728         if (!binfmt || !binfmt->core_dump)
1729                 goto fail;
1730         down_write(&mm->mmap_sem);
1731         /*
1732          * If another thread got here first, or we are not dumpable, bail out.
1733          */
1734         if (mm->core_state || !get_dumpable(mm)) {
1735                 up_write(&mm->mmap_sem);
1736                 goto fail;
1737         }
1738
1739         /*
1740          *      We cannot trust fsuid as being the "true" uid of the
1741          *      process nor do we know its entire history. We only know it
1742          *      was tainted so we dump it as root in mode 2.
1743          */
1744         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1745                 flag = O_EXCL;          /* Stop rewrite attacks */
1746                 current->fsuid = 0;     /* Dump root private */
1747         }
1748
1749         retval = coredump_wait(exit_code, &core_state);
1750         if (retval < 0)
1751                 goto fail;
1752
1753         /*
1754          * Clear any false indication of pending signals that might
1755          * be seen by the filesystem code called to write the core file.
1756          */
1757         clear_thread_flag(TIF_SIGPENDING);
1758
1759         /*
1760          * lock_kernel() because format_corename() is controlled by sysctl, which
1761          * uses lock_kernel()
1762          */
1763         lock_kernel();
1764         ispipe = format_corename(corename, retval, signr);
1765         unlock_kernel();
1766         /*
1767          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1768          * to a pipe.  Since we're not writing directly to the filesystem
1769          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1770          * created unless the pipe reader choses to write out the core file
1771          * at which point file size limits and permissions will be imposed
1772          * as it does with any other process
1773          */
1774         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1775                 goto fail_unlock;
1776
1777         if (ispipe) {
1778                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1779                 /* Terminate the string before the first option */
1780                 delimit = strchr(corename, ' ');
1781                 if (delimit)
1782                         *delimit = '\0';
1783                 delimit = strrchr(helper_argv[0], '/');
1784                 if (delimit)
1785                         delimit++;
1786                 else
1787                         delimit = helper_argv[0];
1788                 if (!strcmp(delimit, current->comm)) {
1789                         printk(KERN_NOTICE "Recursive core dump detected, "
1790                                         "aborting\n");
1791                         goto fail_unlock;
1792                 }
1793
1794                 core_limit = RLIM_INFINITY;
1795
1796                 /* SIGPIPE can happen, but it's just never processed */
1797                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1798                                 &file)) {
1799                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1800                                corename);
1801                         goto fail_unlock;
1802                 }
1803         } else
1804                 file = filp_open(corename,
1805                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1806                                  0600);
1807         if (IS_ERR(file))
1808                 goto fail_unlock;
1809         inode = file->f_path.dentry->d_inode;
1810         if (inode->i_nlink > 1)
1811                 goto close_fail;        /* multiple links - don't dump */
1812         if (!ispipe && d_unhashed(file->f_path.dentry))
1813                 goto close_fail;
1814
1815         /* AK: actually i see no reason to not allow this for named pipes etc.,
1816            but keep the previous behaviour for now. */
1817         if (!ispipe && !S_ISREG(inode->i_mode))
1818                 goto close_fail;
1819         /*
1820          * Dont allow local users get cute and trick others to coredump
1821          * into their pre-created files:
1822          */
1823         if (inode->i_uid != current->fsuid)
1824                 goto close_fail;
1825         if (!file->f_op)
1826                 goto close_fail;
1827         if (!file->f_op->write)
1828                 goto close_fail;
1829         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1830                 goto close_fail;
1831
1832         retval = binfmt->core_dump(signr, regs, file, core_limit);
1833
1834         if (retval)
1835                 current->signal->group_exit_code |= 0x80;
1836 close_fail:
1837         filp_close(file, NULL);
1838 fail_unlock:
1839         if (helper_argv)
1840                 argv_free(helper_argv);
1841
1842         current->fsuid = fsuid;
1843         coredump_finish(mm);
1844 fail:
1845         return retval;
1846 }