pid namespaces: define and use task_active_pid_ns() wrapper
[linux-2.6.git] / fs / exec.c
1 /*
2  *  linux/fs/exec.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  * #!-checking implemented by tytso.
9  */
10 /*
11  * Demand-loading implemented 01.12.91 - no need to read anything but
12  * the header into memory. The inode of the executable is put into
13  * "current->executable", and page faults do the actual loading. Clean.
14  *
15  * Once more I can proudly say that linux stood up to being changed: it
16  * was less than 2 hours work to get demand-loading completely implemented.
17  *
18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
19  * current->executable is only used by the procfs.  This allows a dispatch
20  * table to check for several different types  of binary formats.  We keep
21  * trying until we recognize the file or we run out of supported binary
22  * formats. 
23  */
24
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/mman.h>
28 #include <linux/a.out.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/highmem.h>
36 #include <linux/spinlock.h>
37 #include <linux/key.h>
38 #include <linux/personality.h>
39 #include <linux/binfmts.h>
40 #include <linux/swap.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/rmap.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
57 #include <asm/tlb.h>
58
59 #ifdef CONFIG_KMOD
60 #include <linux/kmod.h>
61 #endif
62
63 int core_uses_pid;
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 int suid_dumpable = 0;
66
67 /* The maximal length of core_pattern is also specified in sysctl.c */
68
69 static LIST_HEAD(formats);
70 static DEFINE_RWLOCK(binfmt_lock);
71
72 int register_binfmt(struct linux_binfmt * fmt)
73 {
74         if (!fmt)
75                 return -EINVAL;
76         write_lock(&binfmt_lock);
77         list_add(&fmt->lh, &formats);
78         write_unlock(&binfmt_lock);
79         return 0;       
80 }
81
82 EXPORT_SYMBOL(register_binfmt);
83
84 void unregister_binfmt(struct linux_binfmt * fmt)
85 {
86         write_lock(&binfmt_lock);
87         list_del(&fmt->lh);
88         write_unlock(&binfmt_lock);
89 }
90
91 EXPORT_SYMBOL(unregister_binfmt);
92
93 static inline void put_binfmt(struct linux_binfmt * fmt)
94 {
95         module_put(fmt->module);
96 }
97
98 /*
99  * Note that a shared library must be both readable and executable due to
100  * security reasons.
101  *
102  * Also note that we take the address to load from from the file itself.
103  */
104 asmlinkage long sys_uselib(const char __user * library)
105 {
106         struct file * file;
107         struct nameidata nd;
108         int error;
109
110         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
111         if (error)
112                 goto out;
113
114         error = -EINVAL;
115         if (!S_ISREG(nd.dentry->d_inode->i_mode))
116                 goto exit;
117
118         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
119         if (error)
120                 goto exit;
121
122         file = nameidata_to_filp(&nd, O_RDONLY);
123         error = PTR_ERR(file);
124         if (IS_ERR(file))
125                 goto out;
126
127         error = -ENOEXEC;
128         if(file->f_op) {
129                 struct linux_binfmt * fmt;
130
131                 read_lock(&binfmt_lock);
132                 list_for_each_entry(fmt, &formats, lh) {
133                         if (!fmt->load_shlib)
134                                 continue;
135                         if (!try_module_get(fmt->module))
136                                 continue;
137                         read_unlock(&binfmt_lock);
138                         error = fmt->load_shlib(file);
139                         read_lock(&binfmt_lock);
140                         put_binfmt(fmt);
141                         if (error != -ENOEXEC)
142                                 break;
143                 }
144                 read_unlock(&binfmt_lock);
145         }
146         fput(file);
147 out:
148         return error;
149 exit:
150         release_open_intent(&nd);
151         path_release(&nd);
152         goto out;
153 }
154
155 #ifdef CONFIG_MMU
156
157 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
158                 int write)
159 {
160         struct page *page;
161         int ret;
162
163 #ifdef CONFIG_STACK_GROWSUP
164         if (write) {
165                 ret = expand_stack_downwards(bprm->vma, pos);
166                 if (ret < 0)
167                         return NULL;
168         }
169 #endif
170         ret = get_user_pages(current, bprm->mm, pos,
171                         1, write, 1, &page, NULL);
172         if (ret <= 0)
173                 return NULL;
174
175         if (write) {
176                 struct rlimit *rlim = current->signal->rlim;
177                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
178
179                 /*
180                  * Limit to 1/4-th the stack size for the argv+env strings.
181                  * This ensures that:
182                  *  - the remaining binfmt code will not run out of stack space,
183                  *  - the program will have a reasonable amount of stack left
184                  *    to work from.
185                  */
186                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
187                         put_page(page);
188                         return NULL;
189                 }
190         }
191
192         return page;
193 }
194
195 static void put_arg_page(struct page *page)
196 {
197         put_page(page);
198 }
199
200 static void free_arg_page(struct linux_binprm *bprm, int i)
201 {
202 }
203
204 static void free_arg_pages(struct linux_binprm *bprm)
205 {
206 }
207
208 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
209                 struct page *page)
210 {
211         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
212 }
213
214 static int __bprm_mm_init(struct linux_binprm *bprm)
215 {
216         int err = -ENOMEM;
217         struct vm_area_struct *vma = NULL;
218         struct mm_struct *mm = bprm->mm;
219
220         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
221         if (!vma)
222                 goto err;
223
224         down_write(&mm->mmap_sem);
225         vma->vm_mm = mm;
226
227         /*
228          * Place the stack at the largest stack address the architecture
229          * supports. Later, we'll move this to an appropriate place. We don't
230          * use STACK_TOP because that can depend on attributes which aren't
231          * configured yet.
232          */
233         vma->vm_end = STACK_TOP_MAX;
234         vma->vm_start = vma->vm_end - PAGE_SIZE;
235
236         vma->vm_flags = VM_STACK_FLAGS;
237         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
238         err = insert_vm_struct(mm, vma);
239         if (err) {
240                 up_write(&mm->mmap_sem);
241                 goto err;
242         }
243
244         mm->stack_vm = mm->total_vm = 1;
245         up_write(&mm->mmap_sem);
246
247         bprm->p = vma->vm_end - sizeof(void *);
248
249         return 0;
250
251 err:
252         if (vma) {
253                 bprm->vma = NULL;
254                 kmem_cache_free(vm_area_cachep, vma);
255         }
256
257         return err;
258 }
259
260 static bool valid_arg_len(struct linux_binprm *bprm, long len)
261 {
262         return len <= MAX_ARG_STRLEN;
263 }
264
265 #else
266
267 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
268                 int write)
269 {
270         struct page *page;
271
272         page = bprm->page[pos / PAGE_SIZE];
273         if (!page && write) {
274                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
275                 if (!page)
276                         return NULL;
277                 bprm->page[pos / PAGE_SIZE] = page;
278         }
279
280         return page;
281 }
282
283 static void put_arg_page(struct page *page)
284 {
285 }
286
287 static void free_arg_page(struct linux_binprm *bprm, int i)
288 {
289         if (bprm->page[i]) {
290                 __free_page(bprm->page[i]);
291                 bprm->page[i] = NULL;
292         }
293 }
294
295 static void free_arg_pages(struct linux_binprm *bprm)
296 {
297         int i;
298
299         for (i = 0; i < MAX_ARG_PAGES; i++)
300                 free_arg_page(bprm, i);
301 }
302
303 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
304                 struct page *page)
305 {
306 }
307
308 static int __bprm_mm_init(struct linux_binprm *bprm)
309 {
310         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
311         return 0;
312 }
313
314 static bool valid_arg_len(struct linux_binprm *bprm, long len)
315 {
316         return len <= bprm->p;
317 }
318
319 #endif /* CONFIG_MMU */
320
321 /*
322  * Create a new mm_struct and populate it with a temporary stack
323  * vm_area_struct.  We don't have enough context at this point to set the stack
324  * flags, permissions, and offset, so we use temporary values.  We'll update
325  * them later in setup_arg_pages().
326  */
327 int bprm_mm_init(struct linux_binprm *bprm)
328 {
329         int err;
330         struct mm_struct *mm = NULL;
331
332         bprm->mm = mm = mm_alloc();
333         err = -ENOMEM;
334         if (!mm)
335                 goto err;
336
337         err = init_new_context(current, mm);
338         if (err)
339                 goto err;
340
341         err = __bprm_mm_init(bprm);
342         if (err)
343                 goto err;
344
345         return 0;
346
347 err:
348         if (mm) {
349                 bprm->mm = NULL;
350                 mmdrop(mm);
351         }
352
353         return err;
354 }
355
356 /*
357  * count() counts the number of strings in array ARGV.
358  */
359 static int count(char __user * __user * argv, int max)
360 {
361         int i = 0;
362
363         if (argv != NULL) {
364                 for (;;) {
365                         char __user * p;
366
367                         if (get_user(p, argv))
368                                 return -EFAULT;
369                         if (!p)
370                                 break;
371                         argv++;
372                         if(++i > max)
373                                 return -E2BIG;
374                         cond_resched();
375                 }
376         }
377         return i;
378 }
379
380 /*
381  * 'copy_strings()' copies argument/environment strings from the old
382  * processes's memory to the new process's stack.  The call to get_user_pages()
383  * ensures the destination page is created and not swapped out.
384  */
385 static int copy_strings(int argc, char __user * __user * argv,
386                         struct linux_binprm *bprm)
387 {
388         struct page *kmapped_page = NULL;
389         char *kaddr = NULL;
390         unsigned long kpos = 0;
391         int ret;
392
393         while (argc-- > 0) {
394                 char __user *str;
395                 int len;
396                 unsigned long pos;
397
398                 if (get_user(str, argv+argc) ||
399                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
400                         ret = -EFAULT;
401                         goto out;
402                 }
403
404                 if (!valid_arg_len(bprm, len)) {
405                         ret = -E2BIG;
406                         goto out;
407                 }
408
409                 /* We're going to work our way backwords. */
410                 pos = bprm->p;
411                 str += len;
412                 bprm->p -= len;
413
414                 while (len > 0) {
415                         int offset, bytes_to_copy;
416
417                         offset = pos % PAGE_SIZE;
418                         if (offset == 0)
419                                 offset = PAGE_SIZE;
420
421                         bytes_to_copy = offset;
422                         if (bytes_to_copy > len)
423                                 bytes_to_copy = len;
424
425                         offset -= bytes_to_copy;
426                         pos -= bytes_to_copy;
427                         str -= bytes_to_copy;
428                         len -= bytes_to_copy;
429
430                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
431                                 struct page *page;
432
433                                 page = get_arg_page(bprm, pos, 1);
434                                 if (!page) {
435                                         ret = -E2BIG;
436                                         goto out;
437                                 }
438
439                                 if (kmapped_page) {
440                                         flush_kernel_dcache_page(kmapped_page);
441                                         kunmap(kmapped_page);
442                                         put_arg_page(kmapped_page);
443                                 }
444                                 kmapped_page = page;
445                                 kaddr = kmap(kmapped_page);
446                                 kpos = pos & PAGE_MASK;
447                                 flush_arg_page(bprm, kpos, kmapped_page);
448                         }
449                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
450                                 ret = -EFAULT;
451                                 goto out;
452                         }
453                 }
454         }
455         ret = 0;
456 out:
457         if (kmapped_page) {
458                 flush_kernel_dcache_page(kmapped_page);
459                 kunmap(kmapped_page);
460                 put_arg_page(kmapped_page);
461         }
462         return ret;
463 }
464
465 /*
466  * Like copy_strings, but get argv and its values from kernel memory.
467  */
468 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
469 {
470         int r;
471         mm_segment_t oldfs = get_fs();
472         set_fs(KERNEL_DS);
473         r = copy_strings(argc, (char __user * __user *)argv, bprm);
474         set_fs(oldfs);
475         return r;
476 }
477 EXPORT_SYMBOL(copy_strings_kernel);
478
479 #ifdef CONFIG_MMU
480
481 /*
482  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
483  * the binfmt code determines where the new stack should reside, we shift it to
484  * its final location.  The process proceeds as follows:
485  *
486  * 1) Use shift to calculate the new vma endpoints.
487  * 2) Extend vma to cover both the old and new ranges.  This ensures the
488  *    arguments passed to subsequent functions are consistent.
489  * 3) Move vma's page tables to the new range.
490  * 4) Free up any cleared pgd range.
491  * 5) Shrink the vma to cover only the new range.
492  */
493 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
494 {
495         struct mm_struct *mm = vma->vm_mm;
496         unsigned long old_start = vma->vm_start;
497         unsigned long old_end = vma->vm_end;
498         unsigned long length = old_end - old_start;
499         unsigned long new_start = old_start - shift;
500         unsigned long new_end = old_end - shift;
501         struct mmu_gather *tlb;
502
503         BUG_ON(new_start > new_end);
504
505         /*
506          * ensure there are no vmas between where we want to go
507          * and where we are
508          */
509         if (vma != find_vma(mm, new_start))
510                 return -EFAULT;
511
512         /*
513          * cover the whole range: [new_start, old_end)
514          */
515         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
516
517         /*
518          * move the page tables downwards, on failure we rely on
519          * process cleanup to remove whatever mess we made.
520          */
521         if (length != move_page_tables(vma, old_start,
522                                        vma, new_start, length))
523                 return -ENOMEM;
524
525         lru_add_drain();
526         tlb = tlb_gather_mmu(mm, 0);
527         if (new_end > old_start) {
528                 /*
529                  * when the old and new regions overlap clear from new_end.
530                  */
531                 free_pgd_range(&tlb, new_end, old_end, new_end,
532                         vma->vm_next ? vma->vm_next->vm_start : 0);
533         } else {
534                 /*
535                  * otherwise, clean from old_start; this is done to not touch
536                  * the address space in [new_end, old_start) some architectures
537                  * have constraints on va-space that make this illegal (IA64) -
538                  * for the others its just a little faster.
539                  */
540                 free_pgd_range(&tlb, old_start, old_end, new_end,
541                         vma->vm_next ? vma->vm_next->vm_start : 0);
542         }
543         tlb_finish_mmu(tlb, new_end, old_end);
544
545         /*
546          * shrink the vma to just the new range.
547          */
548         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
549
550         return 0;
551 }
552
553 #define EXTRA_STACK_VM_PAGES    20      /* random */
554
555 /*
556  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
557  * the stack is optionally relocated, and some extra space is added.
558  */
559 int setup_arg_pages(struct linux_binprm *bprm,
560                     unsigned long stack_top,
561                     int executable_stack)
562 {
563         unsigned long ret;
564         unsigned long stack_shift;
565         struct mm_struct *mm = current->mm;
566         struct vm_area_struct *vma = bprm->vma;
567         struct vm_area_struct *prev = NULL;
568         unsigned long vm_flags;
569         unsigned long stack_base;
570
571 #ifdef CONFIG_STACK_GROWSUP
572         /* Limit stack size to 1GB */
573         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
574         if (stack_base > (1 << 30))
575                 stack_base = 1 << 30;
576
577         /* Make sure we didn't let the argument array grow too large. */
578         if (vma->vm_end - vma->vm_start > stack_base)
579                 return -ENOMEM;
580
581         stack_base = PAGE_ALIGN(stack_top - stack_base);
582
583         stack_shift = vma->vm_start - stack_base;
584         mm->arg_start = bprm->p - stack_shift;
585         bprm->p = vma->vm_end - stack_shift;
586 #else
587         stack_top = arch_align_stack(stack_top);
588         stack_top = PAGE_ALIGN(stack_top);
589         stack_shift = vma->vm_end - stack_top;
590
591         bprm->p -= stack_shift;
592         mm->arg_start = bprm->p;
593 #endif
594
595         if (bprm->loader)
596                 bprm->loader -= stack_shift;
597         bprm->exec -= stack_shift;
598
599         down_write(&mm->mmap_sem);
600         vm_flags = vma->vm_flags;
601
602         /*
603          * Adjust stack execute permissions; explicitly enable for
604          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
605          * (arch default) otherwise.
606          */
607         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
608                 vm_flags |= VM_EXEC;
609         else if (executable_stack == EXSTACK_DISABLE_X)
610                 vm_flags &= ~VM_EXEC;
611         vm_flags |= mm->def_flags;
612
613         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
614                         vm_flags);
615         if (ret)
616                 goto out_unlock;
617         BUG_ON(prev != vma);
618
619         /* Move stack pages down in memory. */
620         if (stack_shift) {
621                 ret = shift_arg_pages(vma, stack_shift);
622                 if (ret) {
623                         up_write(&mm->mmap_sem);
624                         return ret;
625                 }
626         }
627
628 #ifdef CONFIG_STACK_GROWSUP
629         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
630 #else
631         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
632 #endif
633         ret = expand_stack(vma, stack_base);
634         if (ret)
635                 ret = -EFAULT;
636
637 out_unlock:
638         up_write(&mm->mmap_sem);
639         return 0;
640 }
641 EXPORT_SYMBOL(setup_arg_pages);
642
643 #endif /* CONFIG_MMU */
644
645 struct file *open_exec(const char *name)
646 {
647         struct nameidata nd;
648         int err;
649         struct file *file;
650
651         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
652         file = ERR_PTR(err);
653
654         if (!err) {
655                 struct inode *inode = nd.dentry->d_inode;
656                 file = ERR_PTR(-EACCES);
657                 if (S_ISREG(inode->i_mode)) {
658                         int err = vfs_permission(&nd, MAY_EXEC);
659                         file = ERR_PTR(err);
660                         if (!err) {
661                                 file = nameidata_to_filp(&nd, O_RDONLY);
662                                 if (!IS_ERR(file)) {
663                                         err = deny_write_access(file);
664                                         if (err) {
665                                                 fput(file);
666                                                 file = ERR_PTR(err);
667                                         }
668                                 }
669 out:
670                                 return file;
671                         }
672                 }
673                 release_open_intent(&nd);
674                 path_release(&nd);
675         }
676         goto out;
677 }
678
679 EXPORT_SYMBOL(open_exec);
680
681 int kernel_read(struct file *file, unsigned long offset,
682         char *addr, unsigned long count)
683 {
684         mm_segment_t old_fs;
685         loff_t pos = offset;
686         int result;
687
688         old_fs = get_fs();
689         set_fs(get_ds());
690         /* The cast to a user pointer is valid due to the set_fs() */
691         result = vfs_read(file, (void __user *)addr, count, &pos);
692         set_fs(old_fs);
693         return result;
694 }
695
696 EXPORT_SYMBOL(kernel_read);
697
698 static int exec_mmap(struct mm_struct *mm)
699 {
700         struct task_struct *tsk;
701         struct mm_struct * old_mm, *active_mm;
702
703         /* Notify parent that we're no longer interested in the old VM */
704         tsk = current;
705         old_mm = current->mm;
706         mm_release(tsk, old_mm);
707
708         if (old_mm) {
709                 /*
710                  * Make sure that if there is a core dump in progress
711                  * for the old mm, we get out and die instead of going
712                  * through with the exec.  We must hold mmap_sem around
713                  * checking core_waiters and changing tsk->mm.  The
714                  * core-inducing thread will increment core_waiters for
715                  * each thread whose ->mm == old_mm.
716                  */
717                 down_read(&old_mm->mmap_sem);
718                 if (unlikely(old_mm->core_waiters)) {
719                         up_read(&old_mm->mmap_sem);
720                         return -EINTR;
721                 }
722         }
723         task_lock(tsk);
724         active_mm = tsk->active_mm;
725         tsk->mm = mm;
726         tsk->active_mm = mm;
727         activate_mm(active_mm, mm);
728         task_unlock(tsk);
729         arch_pick_mmap_layout(mm);
730         if (old_mm) {
731                 up_read(&old_mm->mmap_sem);
732                 BUG_ON(active_mm != old_mm);
733                 mmput(old_mm);
734                 return 0;
735         }
736         mmdrop(active_mm);
737         return 0;
738 }
739
740 /*
741  * This function makes sure the current process has its own signal table,
742  * so that flush_signal_handlers can later reset the handlers without
743  * disturbing other processes.  (Other processes might share the signal
744  * table via the CLONE_SIGHAND option to clone().)
745  */
746 static int de_thread(struct task_struct *tsk)
747 {
748         struct signal_struct *sig = tsk->signal;
749         struct sighand_struct *oldsighand = tsk->sighand;
750         spinlock_t *lock = &oldsighand->siglock;
751         struct task_struct *leader = NULL;
752         int count;
753
754         if (thread_group_empty(tsk))
755                 goto no_thread_group;
756
757         /*
758          * Kill all other threads in the thread group.
759          * We must hold tasklist_lock to call zap_other_threads.
760          */
761         read_lock(&tasklist_lock);
762         spin_lock_irq(lock);
763         if (sig->flags & SIGNAL_GROUP_EXIT) {
764                 /*
765                  * Another group action in progress, just
766                  * return so that the signal is processed.
767                  */
768                 spin_unlock_irq(lock);
769                 read_unlock(&tasklist_lock);
770                 return -EAGAIN;
771         }
772
773         /*
774          * child_reaper ignores SIGKILL, change it now.
775          * Reparenting needs write_lock on tasklist_lock,
776          * so it is safe to do it under read_lock.
777          */
778         if (unlikely(tsk->group_leader == child_reaper(tsk)))
779                 task_active_pid_ns(tsk)->child_reaper = tsk;
780
781         zap_other_threads(tsk);
782         read_unlock(&tasklist_lock);
783
784         /*
785          * Account for the thread group leader hanging around:
786          */
787         count = 1;
788         if (!thread_group_leader(tsk)) {
789                 count = 2;
790                 /*
791                  * The SIGALRM timer survives the exec, but needs to point
792                  * at us as the new group leader now.  We have a race with
793                  * a timer firing now getting the old leader, so we need to
794                  * synchronize with any firing (by calling del_timer_sync)
795                  * before we can safely let the old group leader die.
796                  */
797                 sig->tsk = tsk;
798                 spin_unlock_irq(lock);
799                 if (hrtimer_cancel(&sig->real_timer))
800                         hrtimer_restart(&sig->real_timer);
801                 spin_lock_irq(lock);
802         }
803
804         sig->notify_count = count;
805         sig->group_exit_task = tsk;
806         while (atomic_read(&sig->count) > count) {
807                 __set_current_state(TASK_UNINTERRUPTIBLE);
808                 spin_unlock_irq(lock);
809                 schedule();
810                 spin_lock_irq(lock);
811         }
812         spin_unlock_irq(lock);
813
814         /*
815          * At this point all other threads have exited, all we have to
816          * do is to wait for the thread group leader to become inactive,
817          * and to assume its PID:
818          */
819         if (!thread_group_leader(tsk)) {
820                 leader = tsk->group_leader;
821
822                 sig->notify_count = -1;
823                 for (;;) {
824                         write_lock_irq(&tasklist_lock);
825                         if (likely(leader->exit_state))
826                                 break;
827                         __set_current_state(TASK_UNINTERRUPTIBLE);
828                         write_unlock_irq(&tasklist_lock);
829                         schedule();
830                 }
831
832                 /*
833                  * The only record we have of the real-time age of a
834                  * process, regardless of execs it's done, is start_time.
835                  * All the past CPU time is accumulated in signal_struct
836                  * from sister threads now dead.  But in this non-leader
837                  * exec, nothing survives from the original leader thread,
838                  * whose birth marks the true age of this process now.
839                  * When we take on its identity by switching to its PID, we
840                  * also take its birthdate (always earlier than our own).
841                  */
842                 tsk->start_time = leader->start_time;
843
844                 BUG_ON(leader->tgid != tsk->tgid);
845                 BUG_ON(tsk->pid == tsk->tgid);
846                 /*
847                  * An exec() starts a new thread group with the
848                  * TGID of the previous thread group. Rehash the
849                  * two threads with a switched PID, and release
850                  * the former thread group leader:
851                  */
852
853                 /* Become a process group leader with the old leader's pid.
854                  * The old leader becomes a thread of the this thread group.
855                  * Note: The old leader also uses this pid until release_task
856                  *       is called.  Odd but simple and correct.
857                  */
858                 detach_pid(tsk, PIDTYPE_PID);
859                 tsk->pid = leader->pid;
860                 attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
861                 transfer_pid(leader, tsk, PIDTYPE_PGID);
862                 transfer_pid(leader, tsk, PIDTYPE_SID);
863                 list_replace_rcu(&leader->tasks, &tsk->tasks);
864
865                 tsk->group_leader = tsk;
866                 leader->group_leader = tsk;
867
868                 tsk->exit_signal = SIGCHLD;
869
870                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
871                 leader->exit_state = EXIT_DEAD;
872
873                 write_unlock_irq(&tasklist_lock);
874         }
875
876         sig->group_exit_task = NULL;
877         sig->notify_count = 0;
878         /*
879          * There may be one thread left which is just exiting,
880          * but it's safe to stop telling the group to kill themselves.
881          */
882         sig->flags = 0;
883
884 no_thread_group:
885         exit_itimers(sig);
886         if (leader)
887                 release_task(leader);
888
889         if (atomic_read(&oldsighand->count) != 1) {
890                 struct sighand_struct *newsighand;
891                 /*
892                  * This ->sighand is shared with the CLONE_SIGHAND
893                  * but not CLONE_THREAD task, switch to the new one.
894                  */
895                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
896                 if (!newsighand)
897                         return -ENOMEM;
898
899                 atomic_set(&newsighand->count, 1);
900                 memcpy(newsighand->action, oldsighand->action,
901                        sizeof(newsighand->action));
902
903                 write_lock_irq(&tasklist_lock);
904                 spin_lock(&oldsighand->siglock);
905                 rcu_assign_pointer(tsk->sighand, newsighand);
906                 spin_unlock(&oldsighand->siglock);
907                 write_unlock_irq(&tasklist_lock);
908
909                 __cleanup_sighand(oldsighand);
910         }
911
912         BUG_ON(!thread_group_leader(tsk));
913         return 0;
914 }
915
916 /*
917  * These functions flushes out all traces of the currently running executable
918  * so that a new one can be started
919  */
920 static void flush_old_files(struct files_struct * files)
921 {
922         long j = -1;
923         struct fdtable *fdt;
924
925         spin_lock(&files->file_lock);
926         for (;;) {
927                 unsigned long set, i;
928
929                 j++;
930                 i = j * __NFDBITS;
931                 fdt = files_fdtable(files);
932                 if (i >= fdt->max_fds)
933                         break;
934                 set = fdt->close_on_exec->fds_bits[j];
935                 if (!set)
936                         continue;
937                 fdt->close_on_exec->fds_bits[j] = 0;
938                 spin_unlock(&files->file_lock);
939                 for ( ; set ; i++,set >>= 1) {
940                         if (set & 1) {
941                                 sys_close(i);
942                         }
943                 }
944                 spin_lock(&files->file_lock);
945
946         }
947         spin_unlock(&files->file_lock);
948 }
949
950 void get_task_comm(char *buf, struct task_struct *tsk)
951 {
952         /* buf must be at least sizeof(tsk->comm) in size */
953         task_lock(tsk);
954         strncpy(buf, tsk->comm, sizeof(tsk->comm));
955         task_unlock(tsk);
956 }
957
958 void set_task_comm(struct task_struct *tsk, char *buf)
959 {
960         task_lock(tsk);
961         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
962         task_unlock(tsk);
963 }
964
965 int flush_old_exec(struct linux_binprm * bprm)
966 {
967         char * name;
968         int i, ch, retval;
969         struct files_struct *files;
970         char tcomm[sizeof(current->comm)];
971
972         /*
973          * Make sure we have a private signal table and that
974          * we are unassociated from the previous thread group.
975          */
976         retval = de_thread(current);
977         if (retval)
978                 goto out;
979
980         /*
981          * Make sure we have private file handles. Ask the
982          * fork helper to do the work for us and the exit
983          * helper to do the cleanup of the old one.
984          */
985         files = current->files;         /* refcounted so safe to hold */
986         retval = unshare_files();
987         if (retval)
988                 goto out;
989         /*
990          * Release all of the old mmap stuff
991          */
992         retval = exec_mmap(bprm->mm);
993         if (retval)
994                 goto mmap_failed;
995
996         bprm->mm = NULL;                /* We're using it now */
997
998         /* This is the point of no return */
999         put_files_struct(files);
1000
1001         current->sas_ss_sp = current->sas_ss_size = 0;
1002
1003         if (current->euid == current->uid && current->egid == current->gid)
1004                 set_dumpable(current->mm, 1);
1005         else
1006                 set_dumpable(current->mm, suid_dumpable);
1007
1008         name = bprm->filename;
1009
1010         /* Copies the binary name from after last slash */
1011         for (i=0; (ch = *(name++)) != '\0';) {
1012                 if (ch == '/')
1013                         i = 0; /* overwrite what we wrote */
1014                 else
1015                         if (i < (sizeof(tcomm) - 1))
1016                                 tcomm[i++] = ch;
1017         }
1018         tcomm[i] = '\0';
1019         set_task_comm(current, tcomm);
1020
1021         current->flags &= ~PF_RANDOMIZE;
1022         flush_thread();
1023
1024         /* Set the new mm task size. We have to do that late because it may
1025          * depend on TIF_32BIT which is only updated in flush_thread() on
1026          * some architectures like powerpc
1027          */
1028         current->mm->task_size = TASK_SIZE;
1029
1030         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1031                 suid_keys(current);
1032                 set_dumpable(current->mm, suid_dumpable);
1033                 current->pdeath_signal = 0;
1034         } else if (file_permission(bprm->file, MAY_READ) ||
1035                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1036                 suid_keys(current);
1037                 set_dumpable(current->mm, suid_dumpable);
1038         }
1039
1040         /* An exec changes our domain. We are no longer part of the thread
1041            group */
1042
1043         current->self_exec_id++;
1044                         
1045         flush_signal_handlers(current, 0);
1046         flush_old_files(current->files);
1047
1048         return 0;
1049
1050 mmap_failed:
1051         reset_files_struct(current, files);
1052 out:
1053         return retval;
1054 }
1055
1056 EXPORT_SYMBOL(flush_old_exec);
1057
1058 /* 
1059  * Fill the binprm structure from the inode. 
1060  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1061  */
1062 int prepare_binprm(struct linux_binprm *bprm)
1063 {
1064         int mode;
1065         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1066         int retval;
1067
1068         mode = inode->i_mode;
1069         if (bprm->file->f_op == NULL)
1070                 return -EACCES;
1071
1072         bprm->e_uid = current->euid;
1073         bprm->e_gid = current->egid;
1074
1075         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1076                 /* Set-uid? */
1077                 if (mode & S_ISUID) {
1078                         current->personality &= ~PER_CLEAR_ON_SETID;
1079                         bprm->e_uid = inode->i_uid;
1080                 }
1081
1082                 /* Set-gid? */
1083                 /*
1084                  * If setgid is set but no group execute bit then this
1085                  * is a candidate for mandatory locking, not a setgid
1086                  * executable.
1087                  */
1088                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1089                         current->personality &= ~PER_CLEAR_ON_SETID;
1090                         bprm->e_gid = inode->i_gid;
1091                 }
1092         }
1093
1094         /* fill in binprm security blob */
1095         retval = security_bprm_set(bprm);
1096         if (retval)
1097                 return retval;
1098
1099         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1100         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1101 }
1102
1103 EXPORT_SYMBOL(prepare_binprm);
1104
1105 static int unsafe_exec(struct task_struct *p)
1106 {
1107         int unsafe = 0;
1108         if (p->ptrace & PT_PTRACED) {
1109                 if (p->ptrace & PT_PTRACE_CAP)
1110                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1111                 else
1112                         unsafe |= LSM_UNSAFE_PTRACE;
1113         }
1114         if (atomic_read(&p->fs->count) > 1 ||
1115             atomic_read(&p->files->count) > 1 ||
1116             atomic_read(&p->sighand->count) > 1)
1117                 unsafe |= LSM_UNSAFE_SHARE;
1118
1119         return unsafe;
1120 }
1121
1122 void compute_creds(struct linux_binprm *bprm)
1123 {
1124         int unsafe;
1125
1126         if (bprm->e_uid != current->uid) {
1127                 suid_keys(current);
1128                 current->pdeath_signal = 0;
1129         }
1130         exec_keys(current);
1131
1132         task_lock(current);
1133         unsafe = unsafe_exec(current);
1134         security_bprm_apply_creds(bprm, unsafe);
1135         task_unlock(current);
1136         security_bprm_post_apply_creds(bprm);
1137 }
1138 EXPORT_SYMBOL(compute_creds);
1139
1140 /*
1141  * Arguments are '\0' separated strings found at the location bprm->p
1142  * points to; chop off the first by relocating brpm->p to right after
1143  * the first '\0' encountered.
1144  */
1145 int remove_arg_zero(struct linux_binprm *bprm)
1146 {
1147         int ret = 0;
1148         unsigned long offset;
1149         char *kaddr;
1150         struct page *page;
1151
1152         if (!bprm->argc)
1153                 return 0;
1154
1155         do {
1156                 offset = bprm->p & ~PAGE_MASK;
1157                 page = get_arg_page(bprm, bprm->p, 0);
1158                 if (!page) {
1159                         ret = -EFAULT;
1160                         goto out;
1161                 }
1162                 kaddr = kmap_atomic(page, KM_USER0);
1163
1164                 for (; offset < PAGE_SIZE && kaddr[offset];
1165                                 offset++, bprm->p++)
1166                         ;
1167
1168                 kunmap_atomic(kaddr, KM_USER0);
1169                 put_arg_page(page);
1170
1171                 if (offset == PAGE_SIZE)
1172                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1173         } while (offset == PAGE_SIZE);
1174
1175         bprm->p++;
1176         bprm->argc--;
1177         ret = 0;
1178
1179 out:
1180         return ret;
1181 }
1182 EXPORT_SYMBOL(remove_arg_zero);
1183
1184 /*
1185  * cycle the list of binary formats handler, until one recognizes the image
1186  */
1187 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1188 {
1189         int try,retval;
1190         struct linux_binfmt *fmt;
1191 #ifdef __alpha__
1192         /* handle /sbin/loader.. */
1193         {
1194             struct exec * eh = (struct exec *) bprm->buf;
1195
1196             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1197                 (eh->fh.f_flags & 0x3000) == 0x3000)
1198             {
1199                 struct file * file;
1200                 unsigned long loader;
1201
1202                 allow_write_access(bprm->file);
1203                 fput(bprm->file);
1204                 bprm->file = NULL;
1205
1206                 loader = bprm->vma->vm_end - sizeof(void *);
1207
1208                 file = open_exec("/sbin/loader");
1209                 retval = PTR_ERR(file);
1210                 if (IS_ERR(file))
1211                         return retval;
1212
1213                 /* Remember if the application is TASO.  */
1214                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1215
1216                 bprm->file = file;
1217                 bprm->loader = loader;
1218                 retval = prepare_binprm(bprm);
1219                 if (retval<0)
1220                         return retval;
1221                 /* should call search_binary_handler recursively here,
1222                    but it does not matter */
1223             }
1224         }
1225 #endif
1226         retval = security_bprm_check(bprm);
1227         if (retval)
1228                 return retval;
1229
1230         /* kernel module loader fixup */
1231         /* so we don't try to load run modprobe in kernel space. */
1232         set_fs(USER_DS);
1233
1234         retval = audit_bprm(bprm);
1235         if (retval)
1236                 return retval;
1237
1238         retval = -ENOENT;
1239         for (try=0; try<2; try++) {
1240                 read_lock(&binfmt_lock);
1241                 list_for_each_entry(fmt, &formats, lh) {
1242                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1243                         if (!fn)
1244                                 continue;
1245                         if (!try_module_get(fmt->module))
1246                                 continue;
1247                         read_unlock(&binfmt_lock);
1248                         retval = fn(bprm, regs);
1249                         if (retval >= 0) {
1250                                 put_binfmt(fmt);
1251                                 allow_write_access(bprm->file);
1252                                 if (bprm->file)
1253                                         fput(bprm->file);
1254                                 bprm->file = NULL;
1255                                 current->did_exec = 1;
1256                                 proc_exec_connector(current);
1257                                 return retval;
1258                         }
1259                         read_lock(&binfmt_lock);
1260                         put_binfmt(fmt);
1261                         if (retval != -ENOEXEC || bprm->mm == NULL)
1262                                 break;
1263                         if (!bprm->file) {
1264                                 read_unlock(&binfmt_lock);
1265                                 return retval;
1266                         }
1267                 }
1268                 read_unlock(&binfmt_lock);
1269                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1270                         break;
1271 #ifdef CONFIG_KMOD
1272                 }else{
1273 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1274                         if (printable(bprm->buf[0]) &&
1275                             printable(bprm->buf[1]) &&
1276                             printable(bprm->buf[2]) &&
1277                             printable(bprm->buf[3]))
1278                                 break; /* -ENOEXEC */
1279                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1280 #endif
1281                 }
1282         }
1283         return retval;
1284 }
1285
1286 EXPORT_SYMBOL(search_binary_handler);
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         unsigned long env_p;
1299         int retval;
1300
1301         retval = -ENOMEM;
1302         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1303         if (!bprm)
1304                 goto out_ret;
1305
1306         file = open_exec(filename);
1307         retval = PTR_ERR(file);
1308         if (IS_ERR(file))
1309                 goto out_kfree;
1310
1311         sched_exec();
1312
1313         bprm->file = file;
1314         bprm->filename = filename;
1315         bprm->interp = filename;
1316
1317         retval = bprm_mm_init(bprm);
1318         if (retval)
1319                 goto out_file;
1320
1321         bprm->argc = count(argv, MAX_ARG_STRINGS);
1322         if ((retval = bprm->argc) < 0)
1323                 goto out_mm;
1324
1325         bprm->envc = count(envp, MAX_ARG_STRINGS);
1326         if ((retval = bprm->envc) < 0)
1327                 goto out_mm;
1328
1329         retval = security_bprm_alloc(bprm);
1330         if (retval)
1331                 goto out;
1332
1333         retval = prepare_binprm(bprm);
1334         if (retval < 0)
1335                 goto out;
1336
1337         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1338         if (retval < 0)
1339                 goto out;
1340
1341         bprm->exec = bprm->p;
1342         retval = copy_strings(bprm->envc, envp, bprm);
1343         if (retval < 0)
1344                 goto out;
1345
1346         env_p = bprm->p;
1347         retval = copy_strings(bprm->argc, argv, bprm);
1348         if (retval < 0)
1349                 goto out;
1350         bprm->argv_len = env_p - bprm->p;
1351
1352         retval = search_binary_handler(bprm,regs);
1353         if (retval >= 0) {
1354                 /* execve success */
1355                 free_arg_pages(bprm);
1356                 security_bprm_free(bprm);
1357                 acct_update_integrals(current);
1358                 kfree(bprm);
1359                 return retval;
1360         }
1361
1362 out:
1363         free_arg_pages(bprm);
1364         if (bprm->security)
1365                 security_bprm_free(bprm);
1366
1367 out_mm:
1368         if (bprm->mm)
1369                 mmput (bprm->mm);
1370
1371 out_file:
1372         if (bprm->file) {
1373                 allow_write_access(bprm->file);
1374                 fput(bprm->file);
1375         }
1376 out_kfree:
1377         kfree(bprm);
1378
1379 out_ret:
1380         return retval;
1381 }
1382
1383 int set_binfmt(struct linux_binfmt *new)
1384 {
1385         struct linux_binfmt *old = current->binfmt;
1386
1387         if (new) {
1388                 if (!try_module_get(new->module))
1389                         return -1;
1390         }
1391         current->binfmt = new;
1392         if (old)
1393                 module_put(old->module);
1394         return 0;
1395 }
1396
1397 EXPORT_SYMBOL(set_binfmt);
1398
1399 /* format_corename will inspect the pattern parameter, and output a
1400  * name into corename, which must have space for at least
1401  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1402  */
1403 static int format_corename(char *corename, const char *pattern, long signr)
1404 {
1405         const char *pat_ptr = pattern;
1406         char *out_ptr = corename;
1407         char *const out_end = corename + CORENAME_MAX_SIZE;
1408         int rc;
1409         int pid_in_pattern = 0;
1410         int ispipe = 0;
1411
1412         if (*pattern == '|')
1413                 ispipe = 1;
1414
1415         /* Repeat as long as we have more pattern to process and more output
1416            space */
1417         while (*pat_ptr) {
1418                 if (*pat_ptr != '%') {
1419                         if (out_ptr == out_end)
1420                                 goto out;
1421                         *out_ptr++ = *pat_ptr++;
1422                 } else {
1423                         switch (*++pat_ptr) {
1424                         case 0:
1425                                 goto out;
1426                         /* Double percent, output one percent */
1427                         case '%':
1428                                 if (out_ptr == out_end)
1429                                         goto out;
1430                                 *out_ptr++ = '%';
1431                                 break;
1432                         /* pid */
1433                         case 'p':
1434                                 pid_in_pattern = 1;
1435                                 rc = snprintf(out_ptr, out_end - out_ptr,
1436                                               "%d", current->tgid);
1437                                 if (rc > out_end - out_ptr)
1438                                         goto out;
1439                                 out_ptr += rc;
1440                                 break;
1441                         /* uid */
1442                         case 'u':
1443                                 rc = snprintf(out_ptr, out_end - out_ptr,
1444                                               "%d", current->uid);
1445                                 if (rc > out_end - out_ptr)
1446                                         goto out;
1447                                 out_ptr += rc;
1448                                 break;
1449                         /* gid */
1450                         case 'g':
1451                                 rc = snprintf(out_ptr, out_end - out_ptr,
1452                                               "%d", current->gid);
1453                                 if (rc > out_end - out_ptr)
1454                                         goto out;
1455                                 out_ptr += rc;
1456                                 break;
1457                         /* signal that caused the coredump */
1458                         case 's':
1459                                 rc = snprintf(out_ptr, out_end - out_ptr,
1460                                               "%ld", signr);
1461                                 if (rc > out_end - out_ptr)
1462                                         goto out;
1463                                 out_ptr += rc;
1464                                 break;
1465                         /* UNIX time of coredump */
1466                         case 't': {
1467                                 struct timeval tv;
1468                                 do_gettimeofday(&tv);
1469                                 rc = snprintf(out_ptr, out_end - out_ptr,
1470                                               "%lu", tv.tv_sec);
1471                                 if (rc > out_end - out_ptr)
1472                                         goto out;
1473                                 out_ptr += rc;
1474                                 break;
1475                         }
1476                         /* hostname */
1477                         case 'h':
1478                                 down_read(&uts_sem);
1479                                 rc = snprintf(out_ptr, out_end - out_ptr,
1480                                               "%s", utsname()->nodename);
1481                                 up_read(&uts_sem);
1482                                 if (rc > out_end - out_ptr)
1483                                         goto out;
1484                                 out_ptr += rc;
1485                                 break;
1486                         /* executable */
1487                         case 'e':
1488                                 rc = snprintf(out_ptr, out_end - out_ptr,
1489                                               "%s", current->comm);
1490                                 if (rc > out_end - out_ptr)
1491                                         goto out;
1492                                 out_ptr += rc;
1493                                 break;
1494                         /* core limit size */
1495                         case 'c':
1496                                 rc = snprintf(out_ptr, out_end - out_ptr,
1497                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1498                                 if (rc > out_end - out_ptr)
1499                                         goto out;
1500                                 out_ptr += rc;
1501                                 break;
1502                         default:
1503                                 break;
1504                         }
1505                         ++pat_ptr;
1506                 }
1507         }
1508         /* Backward compatibility with core_uses_pid:
1509          *
1510          * If core_pattern does not include a %p (as is the default)
1511          * and core_uses_pid is set, then .%pid will be appended to
1512          * the filename. Do not do this for piped commands. */
1513         if (!ispipe && !pid_in_pattern
1514             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1515                 rc = snprintf(out_ptr, out_end - out_ptr,
1516                               ".%d", current->tgid);
1517                 if (rc > out_end - out_ptr)
1518                         goto out;
1519                 out_ptr += rc;
1520         }
1521 out:
1522         *out_ptr = 0;
1523         return ispipe;
1524 }
1525
1526 static void zap_process(struct task_struct *start)
1527 {
1528         struct task_struct *t;
1529
1530         start->signal->flags = SIGNAL_GROUP_EXIT;
1531         start->signal->group_stop_count = 0;
1532
1533         t = start;
1534         do {
1535                 if (t != current && t->mm) {
1536                         t->mm->core_waiters++;
1537                         sigaddset(&t->pending.signal, SIGKILL);
1538                         signal_wake_up(t, 1);
1539                 }
1540         } while ((t = next_thread(t)) != start);
1541 }
1542
1543 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1544                                 int exit_code)
1545 {
1546         struct task_struct *g, *p;
1547         unsigned long flags;
1548         int err = -EAGAIN;
1549
1550         spin_lock_irq(&tsk->sighand->siglock);
1551         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1552                 tsk->signal->group_exit_code = exit_code;
1553                 zap_process(tsk);
1554                 err = 0;
1555         }
1556         spin_unlock_irq(&tsk->sighand->siglock);
1557         if (err)
1558                 return err;
1559
1560         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1561                 goto done;
1562
1563         rcu_read_lock();
1564         for_each_process(g) {
1565                 if (g == tsk->group_leader)
1566                         continue;
1567
1568                 p = g;
1569                 do {
1570                         if (p->mm) {
1571                                 if (p->mm == mm) {
1572                                         /*
1573                                          * p->sighand can't disappear, but
1574                                          * may be changed by de_thread()
1575                                          */
1576                                         lock_task_sighand(p, &flags);
1577                                         zap_process(p);
1578                                         unlock_task_sighand(p, &flags);
1579                                 }
1580                                 break;
1581                         }
1582                 } while ((p = next_thread(p)) != g);
1583         }
1584         rcu_read_unlock();
1585 done:
1586         return mm->core_waiters;
1587 }
1588
1589 static int coredump_wait(int exit_code)
1590 {
1591         struct task_struct *tsk = current;
1592         struct mm_struct *mm = tsk->mm;
1593         struct completion startup_done;
1594         struct completion *vfork_done;
1595         int core_waiters;
1596
1597         init_completion(&mm->core_done);
1598         init_completion(&startup_done);
1599         mm->core_startup_done = &startup_done;
1600
1601         core_waiters = zap_threads(tsk, mm, exit_code);
1602         up_write(&mm->mmap_sem);
1603
1604         if (unlikely(core_waiters < 0))
1605                 goto fail;
1606
1607         /*
1608          * Make sure nobody is waiting for us to release the VM,
1609          * otherwise we can deadlock when we wait on each other
1610          */
1611         vfork_done = tsk->vfork_done;
1612         if (vfork_done) {
1613                 tsk->vfork_done = NULL;
1614                 complete(vfork_done);
1615         }
1616
1617         if (core_waiters)
1618                 wait_for_completion(&startup_done);
1619 fail:
1620         BUG_ON(mm->core_waiters);
1621         return core_waiters;
1622 }
1623
1624 /*
1625  * set_dumpable converts traditional three-value dumpable to two flags and
1626  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1627  * these bits are not changed atomically.  So get_dumpable can observe the
1628  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1629  * return either old dumpable or new one by paying attention to the order of
1630  * modifying the bits.
1631  *
1632  * dumpable |   mm->flags (binary)
1633  * old  new | initial interim  final
1634  * ---------+-----------------------
1635  *  0    1  |   00      01      01
1636  *  0    2  |   00      10(*)   11
1637  *  1    0  |   01      00      00
1638  *  1    2  |   01      11      11
1639  *  2    0  |   11      10(*)   00
1640  *  2    1  |   11      11      01
1641  *
1642  * (*) get_dumpable regards interim value of 10 as 11.
1643  */
1644 void set_dumpable(struct mm_struct *mm, int value)
1645 {
1646         switch (value) {
1647         case 0:
1648                 clear_bit(MMF_DUMPABLE, &mm->flags);
1649                 smp_wmb();
1650                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1651                 break;
1652         case 1:
1653                 set_bit(MMF_DUMPABLE, &mm->flags);
1654                 smp_wmb();
1655                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1656                 break;
1657         case 2:
1658                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1659                 smp_wmb();
1660                 set_bit(MMF_DUMPABLE, &mm->flags);
1661                 break;
1662         }
1663 }
1664
1665 int get_dumpable(struct mm_struct *mm)
1666 {
1667         int ret;
1668
1669         ret = mm->flags & 0x3;
1670         return (ret >= 2) ? 2 : ret;
1671 }
1672
1673 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1674 {
1675         char corename[CORENAME_MAX_SIZE + 1];
1676         struct mm_struct *mm = current->mm;
1677         struct linux_binfmt * binfmt;
1678         struct inode * inode;
1679         struct file * file;
1680         int retval = 0;
1681         int fsuid = current->fsuid;
1682         int flag = 0;
1683         int ispipe = 0;
1684         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1685         char **helper_argv = NULL;
1686         int helper_argc = 0;
1687         char *delimit;
1688
1689         audit_core_dumps(signr);
1690
1691         binfmt = current->binfmt;
1692         if (!binfmt || !binfmt->core_dump)
1693                 goto fail;
1694         down_write(&mm->mmap_sem);
1695         if (!get_dumpable(mm)) {
1696                 up_write(&mm->mmap_sem);
1697                 goto fail;
1698         }
1699
1700         /*
1701          *      We cannot trust fsuid as being the "true" uid of the
1702          *      process nor do we know its entire history. We only know it
1703          *      was tainted so we dump it as root in mode 2.
1704          */
1705         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1706                 flag = O_EXCL;          /* Stop rewrite attacks */
1707                 current->fsuid = 0;     /* Dump root private */
1708         }
1709         set_dumpable(mm, 0);
1710
1711         retval = coredump_wait(exit_code);
1712         if (retval < 0)
1713                 goto fail;
1714
1715         /*
1716          * Clear any false indication of pending signals that might
1717          * be seen by the filesystem code called to write the core file.
1718          */
1719         clear_thread_flag(TIF_SIGPENDING);
1720
1721         /*
1722          * lock_kernel() because format_corename() is controlled by sysctl, which
1723          * uses lock_kernel()
1724          */
1725         lock_kernel();
1726         ispipe = format_corename(corename, core_pattern, signr);
1727         unlock_kernel();
1728         /*
1729          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1730          * to a pipe.  Since we're not writing directly to the filesystem
1731          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1732          * created unless the pipe reader choses to write out the core file
1733          * at which point file size limits and permissions will be imposed
1734          * as it does with any other process
1735          */
1736         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1737                 goto fail_unlock;
1738
1739         if (ispipe) {
1740                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1741                 /* Terminate the string before the first option */
1742                 delimit = strchr(corename, ' ');
1743                 if (delimit)
1744                         *delimit = '\0';
1745                 delimit = strrchr(helper_argv[0], '/');
1746                 if (delimit)
1747                         delimit++;
1748                 else
1749                         delimit = helper_argv[0];
1750                 if (!strcmp(delimit, current->comm)) {
1751                         printk(KERN_NOTICE "Recursive core dump detected, "
1752                                         "aborting\n");
1753                         goto fail_unlock;
1754                 }
1755
1756                 core_limit = RLIM_INFINITY;
1757
1758                 /* SIGPIPE can happen, but it's just never processed */
1759                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1760                                 &file)) {
1761                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1762                                corename);
1763                         goto fail_unlock;
1764                 }
1765         } else
1766                 file = filp_open(corename,
1767                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1768                                  0600);
1769         if (IS_ERR(file))
1770                 goto fail_unlock;
1771         inode = file->f_path.dentry->d_inode;
1772         if (inode->i_nlink > 1)
1773                 goto close_fail;        /* multiple links - don't dump */
1774         if (!ispipe && d_unhashed(file->f_path.dentry))
1775                 goto close_fail;
1776
1777         /* AK: actually i see no reason to not allow this for named pipes etc.,
1778            but keep the previous behaviour for now. */
1779         if (!ispipe && !S_ISREG(inode->i_mode))
1780                 goto close_fail;
1781         if (!file->f_op)
1782                 goto close_fail;
1783         if (!file->f_op->write)
1784                 goto close_fail;
1785         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1786                 goto close_fail;
1787
1788         retval = binfmt->core_dump(signr, regs, file, core_limit);
1789
1790         if (retval)
1791                 current->signal->group_exit_code |= 0x80;
1792 close_fail:
1793         filp_close(file, NULL);
1794 fail_unlock:
1795         if (helper_argv)
1796                 argv_free(helper_argv);
1797
1798         current->fsuid = fsuid;
1799         complete_all(&mm->core_done);
1800 fail:
1801         return retval;
1802 }