711bc45d789c54886d1836ab0845b10bd7884714
[linux-2.6.git] / fs / exec.c
1 /*
2  *  linux/fs/exec.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  * #!-checking implemented by tytso.
9  */
10 /*
11  * Demand-loading implemented 01.12.91 - no need to read anything but
12  * the header into memory. The inode of the executable is put into
13  * "current->executable", and page faults do the actual loading. Clean.
14  *
15  * Once more I can proudly say that linux stood up to being changed: it
16  * was less than 2 hours work to get demand-loading completely implemented.
17  *
18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
19  * current->executable is only used by the procfs.  This allows a dispatch
20  * table to check for several different types  of binary formats.  We keep
21  * trying until we recognize the file or we run out of supported binary
22  * formats. 
23  */
24
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/mman.h>
28 #include <linux/a.out.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/highmem.h>
36 #include <linux/spinlock.h>
37 #include <linux/key.h>
38 #include <linux/personality.h>
39 #include <linux/binfmts.h>
40 #include <linux/swap.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/rmap.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
57 #include <asm/tlb.h>
58
59 #ifdef CONFIG_KMOD
60 #include <linux/kmod.h>
61 #endif
62
63 int core_uses_pid;
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 int suid_dumpable = 0;
66
67 /* The maximal length of core_pattern is also specified in sysctl.c */
68
69 static LIST_HEAD(formats);
70 static DEFINE_RWLOCK(binfmt_lock);
71
72 int register_binfmt(struct linux_binfmt * fmt)
73 {
74         if (!fmt)
75                 return -EINVAL;
76         write_lock(&binfmt_lock);
77         list_add(&fmt->lh, &formats);
78         write_unlock(&binfmt_lock);
79         return 0;       
80 }
81
82 EXPORT_SYMBOL(register_binfmt);
83
84 void unregister_binfmt(struct linux_binfmt * fmt)
85 {
86         write_lock(&binfmt_lock);
87         list_del(&fmt->lh);
88         write_unlock(&binfmt_lock);
89 }
90
91 EXPORT_SYMBOL(unregister_binfmt);
92
93 static inline void put_binfmt(struct linux_binfmt * fmt)
94 {
95         module_put(fmt->module);
96 }
97
98 /*
99  * Note that a shared library must be both readable and executable due to
100  * security reasons.
101  *
102  * Also note that we take the address to load from from the file itself.
103  */
104 asmlinkage long sys_uselib(const char __user * library)
105 {
106         struct file * file;
107         struct nameidata nd;
108         int error;
109
110         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
111         if (error)
112                 goto out;
113
114         error = -EINVAL;
115         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
116                 goto exit;
117
118         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
119         if (error)
120                 goto exit;
121
122         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
123         error = PTR_ERR(file);
124         if (IS_ERR(file))
125                 goto out;
126
127         error = -ENOEXEC;
128         if(file->f_op) {
129                 struct linux_binfmt * fmt;
130
131                 read_lock(&binfmt_lock);
132                 list_for_each_entry(fmt, &formats, lh) {
133                         if (!fmt->load_shlib)
134                                 continue;
135                         if (!try_module_get(fmt->module))
136                                 continue;
137                         read_unlock(&binfmt_lock);
138                         error = fmt->load_shlib(file);
139                         read_lock(&binfmt_lock);
140                         put_binfmt(fmt);
141                         if (error != -ENOEXEC)
142                                 break;
143                 }
144                 read_unlock(&binfmt_lock);
145         }
146         fput(file);
147 out:
148         return error;
149 exit:
150         release_open_intent(&nd);
151         path_put(&nd.path);
152         goto out;
153 }
154
155 #ifdef CONFIG_MMU
156
157 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
158                 int write)
159 {
160         struct page *page;
161         int ret;
162
163 #ifdef CONFIG_STACK_GROWSUP
164         if (write) {
165                 ret = expand_stack_downwards(bprm->vma, pos);
166                 if (ret < 0)
167                         return NULL;
168         }
169 #endif
170         ret = get_user_pages(current, bprm->mm, pos,
171                         1, write, 1, &page, NULL);
172         if (ret <= 0)
173                 return NULL;
174
175         if (write) {
176                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
177                 struct rlimit *rlim;
178
179                 /*
180                  * We've historically supported up to 32 pages (ARG_MAX)
181                  * of argument strings even with small stacks
182                  */
183                 if (size <= ARG_MAX)
184                         return page;
185
186                 /*
187                  * Limit to 1/4-th the stack size for the argv+env strings.
188                  * This ensures that:
189                  *  - the remaining binfmt code will not run out of stack space,
190                  *  - the program will have a reasonable amount of stack left
191                  *    to work from.
192                  */
193                 rlim = current->signal->rlim;
194                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
195                         put_page(page);
196                         return NULL;
197                 }
198         }
199
200         return page;
201 }
202
203 static void put_arg_page(struct page *page)
204 {
205         put_page(page);
206 }
207
208 static void free_arg_page(struct linux_binprm *bprm, int i)
209 {
210 }
211
212 static void free_arg_pages(struct linux_binprm *bprm)
213 {
214 }
215
216 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
217                 struct page *page)
218 {
219         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
220 }
221
222 static int __bprm_mm_init(struct linux_binprm *bprm)
223 {
224         int err = -ENOMEM;
225         struct vm_area_struct *vma = NULL;
226         struct mm_struct *mm = bprm->mm;
227
228         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
229         if (!vma)
230                 goto err;
231
232         down_write(&mm->mmap_sem);
233         vma->vm_mm = mm;
234
235         /*
236          * Place the stack at the largest stack address the architecture
237          * supports. Later, we'll move this to an appropriate place. We don't
238          * use STACK_TOP because that can depend on attributes which aren't
239          * configured yet.
240          */
241         vma->vm_end = STACK_TOP_MAX;
242         vma->vm_start = vma->vm_end - PAGE_SIZE;
243
244         vma->vm_flags = VM_STACK_FLAGS;
245         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
246         err = insert_vm_struct(mm, vma);
247         if (err) {
248                 up_write(&mm->mmap_sem);
249                 goto err;
250         }
251
252         mm->stack_vm = mm->total_vm = 1;
253         up_write(&mm->mmap_sem);
254
255         bprm->p = vma->vm_end - sizeof(void *);
256
257         return 0;
258
259 err:
260         if (vma) {
261                 bprm->vma = NULL;
262                 kmem_cache_free(vm_area_cachep, vma);
263         }
264
265         return err;
266 }
267
268 static bool valid_arg_len(struct linux_binprm *bprm, long len)
269 {
270         return len <= MAX_ARG_STRLEN;
271 }
272
273 #else
274
275 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
276                 int write)
277 {
278         struct page *page;
279
280         page = bprm->page[pos / PAGE_SIZE];
281         if (!page && write) {
282                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
283                 if (!page)
284                         return NULL;
285                 bprm->page[pos / PAGE_SIZE] = page;
286         }
287
288         return page;
289 }
290
291 static void put_arg_page(struct page *page)
292 {
293 }
294
295 static void free_arg_page(struct linux_binprm *bprm, int i)
296 {
297         if (bprm->page[i]) {
298                 __free_page(bprm->page[i]);
299                 bprm->page[i] = NULL;
300         }
301 }
302
303 static void free_arg_pages(struct linux_binprm *bprm)
304 {
305         int i;
306
307         for (i = 0; i < MAX_ARG_PAGES; i++)
308                 free_arg_page(bprm, i);
309 }
310
311 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
312                 struct page *page)
313 {
314 }
315
316 static int __bprm_mm_init(struct linux_binprm *bprm)
317 {
318         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
319         return 0;
320 }
321
322 static bool valid_arg_len(struct linux_binprm *bprm, long len)
323 {
324         return len <= bprm->p;
325 }
326
327 #endif /* CONFIG_MMU */
328
329 /*
330  * Create a new mm_struct and populate it with a temporary stack
331  * vm_area_struct.  We don't have enough context at this point to set the stack
332  * flags, permissions, and offset, so we use temporary values.  We'll update
333  * them later in setup_arg_pages().
334  */
335 int bprm_mm_init(struct linux_binprm *bprm)
336 {
337         int err;
338         struct mm_struct *mm = NULL;
339
340         bprm->mm = mm = mm_alloc();
341         err = -ENOMEM;
342         if (!mm)
343                 goto err;
344
345         err = init_new_context(current, mm);
346         if (err)
347                 goto err;
348
349         err = __bprm_mm_init(bprm);
350         if (err)
351                 goto err;
352
353         return 0;
354
355 err:
356         if (mm) {
357                 bprm->mm = NULL;
358                 mmdrop(mm);
359         }
360
361         return err;
362 }
363
364 /*
365  * count() counts the number of strings in array ARGV.
366  */
367 static int count(char __user * __user * argv, int max)
368 {
369         int i = 0;
370
371         if (argv != NULL) {
372                 for (;;) {
373                         char __user * p;
374
375                         if (get_user(p, argv))
376                                 return -EFAULT;
377                         if (!p)
378                                 break;
379                         argv++;
380                         if(++i > max)
381                                 return -E2BIG;
382                         cond_resched();
383                 }
384         }
385         return i;
386 }
387
388 /*
389  * 'copy_strings()' copies argument/environment strings from the old
390  * processes's memory to the new process's stack.  The call to get_user_pages()
391  * ensures the destination page is created and not swapped out.
392  */
393 static int copy_strings(int argc, char __user * __user * argv,
394                         struct linux_binprm *bprm)
395 {
396         struct page *kmapped_page = NULL;
397         char *kaddr = NULL;
398         unsigned long kpos = 0;
399         int ret;
400
401         while (argc-- > 0) {
402                 char __user *str;
403                 int len;
404                 unsigned long pos;
405
406                 if (get_user(str, argv+argc) ||
407                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
408                         ret = -EFAULT;
409                         goto out;
410                 }
411
412                 if (!valid_arg_len(bprm, len)) {
413                         ret = -E2BIG;
414                         goto out;
415                 }
416
417                 /* We're going to work our way backwords. */
418                 pos = bprm->p;
419                 str += len;
420                 bprm->p -= len;
421
422                 while (len > 0) {
423                         int offset, bytes_to_copy;
424
425                         offset = pos % PAGE_SIZE;
426                         if (offset == 0)
427                                 offset = PAGE_SIZE;
428
429                         bytes_to_copy = offset;
430                         if (bytes_to_copy > len)
431                                 bytes_to_copy = len;
432
433                         offset -= bytes_to_copy;
434                         pos -= bytes_to_copy;
435                         str -= bytes_to_copy;
436                         len -= bytes_to_copy;
437
438                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
439                                 struct page *page;
440
441                                 page = get_arg_page(bprm, pos, 1);
442                                 if (!page) {
443                                         ret = -E2BIG;
444                                         goto out;
445                                 }
446
447                                 if (kmapped_page) {
448                                         flush_kernel_dcache_page(kmapped_page);
449                                         kunmap(kmapped_page);
450                                         put_arg_page(kmapped_page);
451                                 }
452                                 kmapped_page = page;
453                                 kaddr = kmap(kmapped_page);
454                                 kpos = pos & PAGE_MASK;
455                                 flush_arg_page(bprm, kpos, kmapped_page);
456                         }
457                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
458                                 ret = -EFAULT;
459                                 goto out;
460                         }
461                 }
462         }
463         ret = 0;
464 out:
465         if (kmapped_page) {
466                 flush_kernel_dcache_page(kmapped_page);
467                 kunmap(kmapped_page);
468                 put_arg_page(kmapped_page);
469         }
470         return ret;
471 }
472
473 /*
474  * Like copy_strings, but get argv and its values from kernel memory.
475  */
476 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
477 {
478         int r;
479         mm_segment_t oldfs = get_fs();
480         set_fs(KERNEL_DS);
481         r = copy_strings(argc, (char __user * __user *)argv, bprm);
482         set_fs(oldfs);
483         return r;
484 }
485 EXPORT_SYMBOL(copy_strings_kernel);
486
487 #ifdef CONFIG_MMU
488
489 /*
490  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
491  * the binfmt code determines where the new stack should reside, we shift it to
492  * its final location.  The process proceeds as follows:
493  *
494  * 1) Use shift to calculate the new vma endpoints.
495  * 2) Extend vma to cover both the old and new ranges.  This ensures the
496  *    arguments passed to subsequent functions are consistent.
497  * 3) Move vma's page tables to the new range.
498  * 4) Free up any cleared pgd range.
499  * 5) Shrink the vma to cover only the new range.
500  */
501 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
502 {
503         struct mm_struct *mm = vma->vm_mm;
504         unsigned long old_start = vma->vm_start;
505         unsigned long old_end = vma->vm_end;
506         unsigned long length = old_end - old_start;
507         unsigned long new_start = old_start - shift;
508         unsigned long new_end = old_end - shift;
509         struct mmu_gather *tlb;
510
511         BUG_ON(new_start > new_end);
512
513         /*
514          * ensure there are no vmas between where we want to go
515          * and where we are
516          */
517         if (vma != find_vma(mm, new_start))
518                 return -EFAULT;
519
520         /*
521          * cover the whole range: [new_start, old_end)
522          */
523         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
524
525         /*
526          * move the page tables downwards, on failure we rely on
527          * process cleanup to remove whatever mess we made.
528          */
529         if (length != move_page_tables(vma, old_start,
530                                        vma, new_start, length))
531                 return -ENOMEM;
532
533         lru_add_drain();
534         tlb = tlb_gather_mmu(mm, 0);
535         if (new_end > old_start) {
536                 /*
537                  * when the old and new regions overlap clear from new_end.
538                  */
539                 free_pgd_range(&tlb, new_end, old_end, new_end,
540                         vma->vm_next ? vma->vm_next->vm_start : 0);
541         } else {
542                 /*
543                  * otherwise, clean from old_start; this is done to not touch
544                  * the address space in [new_end, old_start) some architectures
545                  * have constraints on va-space that make this illegal (IA64) -
546                  * for the others its just a little faster.
547                  */
548                 free_pgd_range(&tlb, old_start, old_end, new_end,
549                         vma->vm_next ? vma->vm_next->vm_start : 0);
550         }
551         tlb_finish_mmu(tlb, new_end, old_end);
552
553         /*
554          * shrink the vma to just the new range.
555          */
556         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
557
558         return 0;
559 }
560
561 #define EXTRA_STACK_VM_PAGES    20      /* random */
562
563 /*
564  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
565  * the stack is optionally relocated, and some extra space is added.
566  */
567 int setup_arg_pages(struct linux_binprm *bprm,
568                     unsigned long stack_top,
569                     int executable_stack)
570 {
571         unsigned long ret;
572         unsigned long stack_shift;
573         struct mm_struct *mm = current->mm;
574         struct vm_area_struct *vma = bprm->vma;
575         struct vm_area_struct *prev = NULL;
576         unsigned long vm_flags;
577         unsigned long stack_base;
578
579 #ifdef CONFIG_STACK_GROWSUP
580         /* Limit stack size to 1GB */
581         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
582         if (stack_base > (1 << 30))
583                 stack_base = 1 << 30;
584
585         /* Make sure we didn't let the argument array grow too large. */
586         if (vma->vm_end - vma->vm_start > stack_base)
587                 return -ENOMEM;
588
589         stack_base = PAGE_ALIGN(stack_top - stack_base);
590
591         stack_shift = vma->vm_start - stack_base;
592         mm->arg_start = bprm->p - stack_shift;
593         bprm->p = vma->vm_end - stack_shift;
594 #else
595         stack_top = arch_align_stack(stack_top);
596         stack_top = PAGE_ALIGN(stack_top);
597         stack_shift = vma->vm_end - stack_top;
598
599         bprm->p -= stack_shift;
600         mm->arg_start = bprm->p;
601 #endif
602
603         if (bprm->loader)
604                 bprm->loader -= stack_shift;
605         bprm->exec -= stack_shift;
606
607         down_write(&mm->mmap_sem);
608         vm_flags = vma->vm_flags;
609
610         /*
611          * Adjust stack execute permissions; explicitly enable for
612          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
613          * (arch default) otherwise.
614          */
615         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
616                 vm_flags |= VM_EXEC;
617         else if (executable_stack == EXSTACK_DISABLE_X)
618                 vm_flags &= ~VM_EXEC;
619         vm_flags |= mm->def_flags;
620
621         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
622                         vm_flags);
623         if (ret)
624                 goto out_unlock;
625         BUG_ON(prev != vma);
626
627         /* Move stack pages down in memory. */
628         if (stack_shift) {
629                 ret = shift_arg_pages(vma, stack_shift);
630                 if (ret) {
631                         up_write(&mm->mmap_sem);
632                         return ret;
633                 }
634         }
635
636 #ifdef CONFIG_STACK_GROWSUP
637         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
638 #else
639         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
640 #endif
641         ret = expand_stack(vma, stack_base);
642         if (ret)
643                 ret = -EFAULT;
644
645 out_unlock:
646         up_write(&mm->mmap_sem);
647         return 0;
648 }
649 EXPORT_SYMBOL(setup_arg_pages);
650
651 #endif /* CONFIG_MMU */
652
653 struct file *open_exec(const char *name)
654 {
655         struct nameidata nd;
656         int err;
657         struct file *file;
658
659         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
660         file = ERR_PTR(err);
661
662         if (!err) {
663                 struct inode *inode = nd.path.dentry->d_inode;
664                 file = ERR_PTR(-EACCES);
665                 if (S_ISREG(inode->i_mode)) {
666                         int err = vfs_permission(&nd, MAY_EXEC);
667                         file = ERR_PTR(err);
668                         if (!err) {
669                                 file = nameidata_to_filp(&nd,
670                                                         O_RDONLY|O_LARGEFILE);
671                                 if (!IS_ERR(file)) {
672                                         err = deny_write_access(file);
673                                         if (err) {
674                                                 fput(file);
675                                                 file = ERR_PTR(err);
676                                         }
677                                 }
678 out:
679                                 return file;
680                         }
681                 }
682                 release_open_intent(&nd);
683                 path_put(&nd.path);
684         }
685         goto out;
686 }
687
688 EXPORT_SYMBOL(open_exec);
689
690 int kernel_read(struct file *file, unsigned long offset,
691         char *addr, unsigned long count)
692 {
693         mm_segment_t old_fs;
694         loff_t pos = offset;
695         int result;
696
697         old_fs = get_fs();
698         set_fs(get_ds());
699         /* The cast to a user pointer is valid due to the set_fs() */
700         result = vfs_read(file, (void __user *)addr, count, &pos);
701         set_fs(old_fs);
702         return result;
703 }
704
705 EXPORT_SYMBOL(kernel_read);
706
707 static int exec_mmap(struct mm_struct *mm)
708 {
709         struct task_struct *tsk;
710         struct mm_struct * old_mm, *active_mm;
711
712         /* Notify parent that we're no longer interested in the old VM */
713         tsk = current;
714         old_mm = current->mm;
715         mm_release(tsk, old_mm);
716
717         if (old_mm) {
718                 /*
719                  * Make sure that if there is a core dump in progress
720                  * for the old mm, we get out and die instead of going
721                  * through with the exec.  We must hold mmap_sem around
722                  * checking core_waiters and changing tsk->mm.  The
723                  * core-inducing thread will increment core_waiters for
724                  * each thread whose ->mm == old_mm.
725                  */
726                 down_read(&old_mm->mmap_sem);
727                 if (unlikely(old_mm->core_waiters)) {
728                         up_read(&old_mm->mmap_sem);
729                         return -EINTR;
730                 }
731         }
732         task_lock(tsk);
733         active_mm = tsk->active_mm;
734         tsk->mm = mm;
735         tsk->active_mm = mm;
736         activate_mm(active_mm, mm);
737         task_unlock(tsk);
738         mm_update_next_owner(mm);
739         arch_pick_mmap_layout(mm);
740         if (old_mm) {
741                 up_read(&old_mm->mmap_sem);
742                 BUG_ON(active_mm != old_mm);
743                 mmput(old_mm);
744                 return 0;
745         }
746         mmdrop(active_mm);
747         return 0;
748 }
749
750 /*
751  * This function makes sure the current process has its own signal table,
752  * so that flush_signal_handlers can later reset the handlers without
753  * disturbing other processes.  (Other processes might share the signal
754  * table via the CLONE_SIGHAND option to clone().)
755  */
756 static int de_thread(struct task_struct *tsk)
757 {
758         struct signal_struct *sig = tsk->signal;
759         struct sighand_struct *oldsighand = tsk->sighand;
760         spinlock_t *lock = &oldsighand->siglock;
761         struct task_struct *leader = NULL;
762         int count;
763
764         if (thread_group_empty(tsk))
765                 goto no_thread_group;
766
767         /*
768          * Kill all other threads in the thread group.
769          * We must hold tasklist_lock to call zap_other_threads.
770          */
771         read_lock(&tasklist_lock);
772         spin_lock_irq(lock);
773         if (signal_group_exit(sig)) {
774                 /*
775                  * Another group action in progress, just
776                  * return so that the signal is processed.
777                  */
778                 spin_unlock_irq(lock);
779                 read_unlock(&tasklist_lock);
780                 return -EAGAIN;
781         }
782
783         /*
784          * child_reaper ignores SIGKILL, change it now.
785          * Reparenting needs write_lock on tasklist_lock,
786          * so it is safe to do it under read_lock.
787          */
788         if (unlikely(tsk->group_leader == task_child_reaper(tsk)))
789                 task_active_pid_ns(tsk)->child_reaper = tsk;
790
791         sig->group_exit_task = tsk;
792         zap_other_threads(tsk);
793         read_unlock(&tasklist_lock);
794
795         /* Account for the thread group leader hanging around: */
796         count = thread_group_leader(tsk) ? 1 : 2;
797         sig->notify_count = count;
798         while (atomic_read(&sig->count) > count) {
799                 __set_current_state(TASK_UNINTERRUPTIBLE);
800                 spin_unlock_irq(lock);
801                 schedule();
802                 spin_lock_irq(lock);
803         }
804         spin_unlock_irq(lock);
805
806         /*
807          * At this point all other threads have exited, all we have to
808          * do is to wait for the thread group leader to become inactive,
809          * and to assume its PID:
810          */
811         if (!thread_group_leader(tsk)) {
812                 leader = tsk->group_leader;
813
814                 sig->notify_count = -1;
815                 for (;;) {
816                         write_lock_irq(&tasklist_lock);
817                         if (likely(leader->exit_state))
818                                 break;
819                         __set_current_state(TASK_UNINTERRUPTIBLE);
820                         write_unlock_irq(&tasklist_lock);
821                         schedule();
822                 }
823
824                 /*
825                  * The only record we have of the real-time age of a
826                  * process, regardless of execs it's done, is start_time.
827                  * All the past CPU time is accumulated in signal_struct
828                  * from sister threads now dead.  But in this non-leader
829                  * exec, nothing survives from the original leader thread,
830                  * whose birth marks the true age of this process now.
831                  * When we take on its identity by switching to its PID, we
832                  * also take its birthdate (always earlier than our own).
833                  */
834                 tsk->start_time = leader->start_time;
835
836                 BUG_ON(!same_thread_group(leader, tsk));
837                 BUG_ON(has_group_leader_pid(tsk));
838                 /*
839                  * An exec() starts a new thread group with the
840                  * TGID of the previous thread group. Rehash the
841                  * two threads with a switched PID, and release
842                  * the former thread group leader:
843                  */
844
845                 /* Become a process group leader with the old leader's pid.
846                  * The old leader becomes a thread of the this thread group.
847                  * Note: The old leader also uses this pid until release_task
848                  *       is called.  Odd but simple and correct.
849                  */
850                 detach_pid(tsk, PIDTYPE_PID);
851                 tsk->pid = leader->pid;
852                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
853                 transfer_pid(leader, tsk, PIDTYPE_PGID);
854                 transfer_pid(leader, tsk, PIDTYPE_SID);
855                 list_replace_rcu(&leader->tasks, &tsk->tasks);
856
857                 tsk->group_leader = tsk;
858                 leader->group_leader = tsk;
859
860                 tsk->exit_signal = SIGCHLD;
861
862                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
863                 leader->exit_state = EXIT_DEAD;
864
865                 write_unlock_irq(&tasklist_lock);
866         }
867
868         sig->group_exit_task = NULL;
869         sig->notify_count = 0;
870
871 no_thread_group:
872         exit_itimers(sig);
873         if (leader)
874                 release_task(leader);
875
876         if (atomic_read(&oldsighand->count) != 1) {
877                 struct sighand_struct *newsighand;
878                 /*
879                  * This ->sighand is shared with the CLONE_SIGHAND
880                  * but not CLONE_THREAD task, switch to the new one.
881                  */
882                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
883                 if (!newsighand)
884                         return -ENOMEM;
885
886                 atomic_set(&newsighand->count, 1);
887                 memcpy(newsighand->action, oldsighand->action,
888                        sizeof(newsighand->action));
889
890                 write_lock_irq(&tasklist_lock);
891                 spin_lock(&oldsighand->siglock);
892                 rcu_assign_pointer(tsk->sighand, newsighand);
893                 spin_unlock(&oldsighand->siglock);
894                 write_unlock_irq(&tasklist_lock);
895
896                 __cleanup_sighand(oldsighand);
897         }
898
899         BUG_ON(!thread_group_leader(tsk));
900         return 0;
901 }
902
903 /*
904  * These functions flushes out all traces of the currently running executable
905  * so that a new one can be started
906  */
907 static void flush_old_files(struct files_struct * files)
908 {
909         long j = -1;
910         struct fdtable *fdt;
911
912         spin_lock(&files->file_lock);
913         for (;;) {
914                 unsigned long set, i;
915
916                 j++;
917                 i = j * __NFDBITS;
918                 fdt = files_fdtable(files);
919                 if (i >= fdt->max_fds)
920                         break;
921                 set = fdt->close_on_exec->fds_bits[j];
922                 if (!set)
923                         continue;
924                 fdt->close_on_exec->fds_bits[j] = 0;
925                 spin_unlock(&files->file_lock);
926                 for ( ; set ; i++,set >>= 1) {
927                         if (set & 1) {
928                                 sys_close(i);
929                         }
930                 }
931                 spin_lock(&files->file_lock);
932
933         }
934         spin_unlock(&files->file_lock);
935 }
936
937 char *get_task_comm(char *buf, struct task_struct *tsk)
938 {
939         /* buf must be at least sizeof(tsk->comm) in size */
940         task_lock(tsk);
941         strncpy(buf, tsk->comm, sizeof(tsk->comm));
942         task_unlock(tsk);
943         return buf;
944 }
945
946 void set_task_comm(struct task_struct *tsk, char *buf)
947 {
948         task_lock(tsk);
949         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
950         task_unlock(tsk);
951 }
952
953 int flush_old_exec(struct linux_binprm * bprm)
954 {
955         char * name;
956         int i, ch, retval;
957         char tcomm[sizeof(current->comm)];
958
959         /*
960          * Make sure we have a private signal table and that
961          * we are unassociated from the previous thread group.
962          */
963         retval = de_thread(current);
964         if (retval)
965                 goto out;
966
967         /*
968          * Release all of the old mmap stuff
969          */
970         retval = exec_mmap(bprm->mm);
971         if (retval)
972                 goto out;
973
974         bprm->mm = NULL;                /* We're using it now */
975
976         /* This is the point of no return */
977         current->sas_ss_sp = current->sas_ss_size = 0;
978
979         if (current->euid == current->uid && current->egid == current->gid)
980                 set_dumpable(current->mm, 1);
981         else
982                 set_dumpable(current->mm, suid_dumpable);
983
984         name = bprm->filename;
985
986         /* Copies the binary name from after last slash */
987         for (i=0; (ch = *(name++)) != '\0';) {
988                 if (ch == '/')
989                         i = 0; /* overwrite what we wrote */
990                 else
991                         if (i < (sizeof(tcomm) - 1))
992                                 tcomm[i++] = ch;
993         }
994         tcomm[i] = '\0';
995         set_task_comm(current, tcomm);
996
997         current->flags &= ~PF_RANDOMIZE;
998         flush_thread();
999
1000         /* Set the new mm task size. We have to do that late because it may
1001          * depend on TIF_32BIT which is only updated in flush_thread() on
1002          * some architectures like powerpc
1003          */
1004         current->mm->task_size = TASK_SIZE;
1005
1006         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1007                 suid_keys(current);
1008                 set_dumpable(current->mm, suid_dumpable);
1009                 current->pdeath_signal = 0;
1010         } else if (file_permission(bprm->file, MAY_READ) ||
1011                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1012                 suid_keys(current);
1013                 set_dumpable(current->mm, suid_dumpable);
1014         }
1015
1016         /* An exec changes our domain. We are no longer part of the thread
1017            group */
1018
1019         current->self_exec_id++;
1020                         
1021         flush_signal_handlers(current, 0);
1022         flush_old_files(current->files);
1023
1024         return 0;
1025
1026 out:
1027         return retval;
1028 }
1029
1030 EXPORT_SYMBOL(flush_old_exec);
1031
1032 /* 
1033  * Fill the binprm structure from the inode. 
1034  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1035  */
1036 int prepare_binprm(struct linux_binprm *bprm)
1037 {
1038         int mode;
1039         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1040         int retval;
1041
1042         mode = inode->i_mode;
1043         if (bprm->file->f_op == NULL)
1044                 return -EACCES;
1045
1046         bprm->e_uid = current->euid;
1047         bprm->e_gid = current->egid;
1048
1049         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1050                 /* Set-uid? */
1051                 if (mode & S_ISUID) {
1052                         current->personality &= ~PER_CLEAR_ON_SETID;
1053                         bprm->e_uid = inode->i_uid;
1054                 }
1055
1056                 /* Set-gid? */
1057                 /*
1058                  * If setgid is set but no group execute bit then this
1059                  * is a candidate for mandatory locking, not a setgid
1060                  * executable.
1061                  */
1062                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1063                         current->personality &= ~PER_CLEAR_ON_SETID;
1064                         bprm->e_gid = inode->i_gid;
1065                 }
1066         }
1067
1068         /* fill in binprm security blob */
1069         retval = security_bprm_set(bprm);
1070         if (retval)
1071                 return retval;
1072
1073         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1074         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1075 }
1076
1077 EXPORT_SYMBOL(prepare_binprm);
1078
1079 static int unsafe_exec(struct task_struct *p)
1080 {
1081         int unsafe = 0;
1082         if (p->ptrace & PT_PTRACED) {
1083                 if (p->ptrace & PT_PTRACE_CAP)
1084                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1085                 else
1086                         unsafe |= LSM_UNSAFE_PTRACE;
1087         }
1088         if (atomic_read(&p->fs->count) > 1 ||
1089             atomic_read(&p->files->count) > 1 ||
1090             atomic_read(&p->sighand->count) > 1)
1091                 unsafe |= LSM_UNSAFE_SHARE;
1092
1093         return unsafe;
1094 }
1095
1096 void compute_creds(struct linux_binprm *bprm)
1097 {
1098         int unsafe;
1099
1100         if (bprm->e_uid != current->uid) {
1101                 suid_keys(current);
1102                 current->pdeath_signal = 0;
1103         }
1104         exec_keys(current);
1105
1106         task_lock(current);
1107         unsafe = unsafe_exec(current);
1108         security_bprm_apply_creds(bprm, unsafe);
1109         task_unlock(current);
1110         security_bprm_post_apply_creds(bprm);
1111 }
1112 EXPORT_SYMBOL(compute_creds);
1113
1114 /*
1115  * Arguments are '\0' separated strings found at the location bprm->p
1116  * points to; chop off the first by relocating brpm->p to right after
1117  * the first '\0' encountered.
1118  */
1119 int remove_arg_zero(struct linux_binprm *bprm)
1120 {
1121         int ret = 0;
1122         unsigned long offset;
1123         char *kaddr;
1124         struct page *page;
1125
1126         if (!bprm->argc)
1127                 return 0;
1128
1129         do {
1130                 offset = bprm->p & ~PAGE_MASK;
1131                 page = get_arg_page(bprm, bprm->p, 0);
1132                 if (!page) {
1133                         ret = -EFAULT;
1134                         goto out;
1135                 }
1136                 kaddr = kmap_atomic(page, KM_USER0);
1137
1138                 for (; offset < PAGE_SIZE && kaddr[offset];
1139                                 offset++, bprm->p++)
1140                         ;
1141
1142                 kunmap_atomic(kaddr, KM_USER0);
1143                 put_arg_page(page);
1144
1145                 if (offset == PAGE_SIZE)
1146                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1147         } while (offset == PAGE_SIZE);
1148
1149         bprm->p++;
1150         bprm->argc--;
1151         ret = 0;
1152
1153 out:
1154         return ret;
1155 }
1156 EXPORT_SYMBOL(remove_arg_zero);
1157
1158 /*
1159  * cycle the list of binary formats handler, until one recognizes the image
1160  */
1161 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1162 {
1163         int try,retval;
1164         struct linux_binfmt *fmt;
1165 #if defined(__alpha__) && defined(CONFIG_ARCH_SUPPORTS_AOUT)
1166         /* handle /sbin/loader.. */
1167         {
1168             struct exec * eh = (struct exec *) bprm->buf;
1169
1170             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1171                 (eh->fh.f_flags & 0x3000) == 0x3000)
1172             {
1173                 struct file * file;
1174                 unsigned long loader;
1175
1176                 allow_write_access(bprm->file);
1177                 fput(bprm->file);
1178                 bprm->file = NULL;
1179
1180                 loader = bprm->vma->vm_end - sizeof(void *);
1181
1182                 file = open_exec("/sbin/loader");
1183                 retval = PTR_ERR(file);
1184                 if (IS_ERR(file))
1185                         return retval;
1186
1187                 /* Remember if the application is TASO.  */
1188                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1189
1190                 bprm->file = file;
1191                 bprm->loader = loader;
1192                 retval = prepare_binprm(bprm);
1193                 if (retval<0)
1194                         return retval;
1195                 /* should call search_binary_handler recursively here,
1196                    but it does not matter */
1197             }
1198         }
1199 #endif
1200         retval = security_bprm_check(bprm);
1201         if (retval)
1202                 return retval;
1203
1204         /* kernel module loader fixup */
1205         /* so we don't try to load run modprobe in kernel space. */
1206         set_fs(USER_DS);
1207
1208         retval = audit_bprm(bprm);
1209         if (retval)
1210                 return retval;
1211
1212         retval = -ENOENT;
1213         for (try=0; try<2; try++) {
1214                 read_lock(&binfmt_lock);
1215                 list_for_each_entry(fmt, &formats, lh) {
1216                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1217                         if (!fn)
1218                                 continue;
1219                         if (!try_module_get(fmt->module))
1220                                 continue;
1221                         read_unlock(&binfmt_lock);
1222                         retval = fn(bprm, regs);
1223                         if (retval >= 0) {
1224                                 put_binfmt(fmt);
1225                                 allow_write_access(bprm->file);
1226                                 if (bprm->file)
1227                                         fput(bprm->file);
1228                                 bprm->file = NULL;
1229                                 current->did_exec = 1;
1230                                 proc_exec_connector(current);
1231                                 return retval;
1232                         }
1233                         read_lock(&binfmt_lock);
1234                         put_binfmt(fmt);
1235                         if (retval != -ENOEXEC || bprm->mm == NULL)
1236                                 break;
1237                         if (!bprm->file) {
1238                                 read_unlock(&binfmt_lock);
1239                                 return retval;
1240                         }
1241                 }
1242                 read_unlock(&binfmt_lock);
1243                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1244                         break;
1245 #ifdef CONFIG_KMOD
1246                 }else{
1247 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1248                         if (printable(bprm->buf[0]) &&
1249                             printable(bprm->buf[1]) &&
1250                             printable(bprm->buf[2]) &&
1251                             printable(bprm->buf[3]))
1252                                 break; /* -ENOEXEC */
1253                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1254 #endif
1255                 }
1256         }
1257         return retval;
1258 }
1259
1260 EXPORT_SYMBOL(search_binary_handler);
1261
1262 /*
1263  * sys_execve() executes a new program.
1264  */
1265 int do_execve(char * filename,
1266         char __user *__user *argv,
1267         char __user *__user *envp,
1268         struct pt_regs * regs)
1269 {
1270         struct linux_binprm *bprm;
1271         struct file *file;
1272         struct files_struct *displaced;
1273         int retval;
1274
1275         retval = unshare_files(&displaced);
1276         if (retval)
1277                 goto out_ret;
1278
1279         retval = -ENOMEM;
1280         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1281         if (!bprm)
1282                 goto out_files;
1283
1284         file = open_exec(filename);
1285         retval = PTR_ERR(file);
1286         if (IS_ERR(file))
1287                 goto out_kfree;
1288
1289         sched_exec();
1290
1291         bprm->file = file;
1292         bprm->filename = filename;
1293         bprm->interp = filename;
1294
1295         retval = bprm_mm_init(bprm);
1296         if (retval)
1297                 goto out_file;
1298
1299         bprm->argc = count(argv, MAX_ARG_STRINGS);
1300         if ((retval = bprm->argc) < 0)
1301                 goto out_mm;
1302
1303         bprm->envc = count(envp, MAX_ARG_STRINGS);
1304         if ((retval = bprm->envc) < 0)
1305                 goto out_mm;
1306
1307         retval = security_bprm_alloc(bprm);
1308         if (retval)
1309                 goto out;
1310
1311         retval = prepare_binprm(bprm);
1312         if (retval < 0)
1313                 goto out;
1314
1315         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1316         if (retval < 0)
1317                 goto out;
1318
1319         bprm->exec = bprm->p;
1320         retval = copy_strings(bprm->envc, envp, bprm);
1321         if (retval < 0)
1322                 goto out;
1323
1324         retval = copy_strings(bprm->argc, argv, bprm);
1325         if (retval < 0)
1326                 goto out;
1327
1328         retval = search_binary_handler(bprm,regs);
1329         if (retval >= 0) {
1330                 /* execve success */
1331                 free_arg_pages(bprm);
1332                 security_bprm_free(bprm);
1333                 acct_update_integrals(current);
1334                 kfree(bprm);
1335                 if (displaced)
1336                         put_files_struct(displaced);
1337                 return retval;
1338         }
1339
1340 out:
1341         free_arg_pages(bprm);
1342         if (bprm->security)
1343                 security_bprm_free(bprm);
1344
1345 out_mm:
1346         if (bprm->mm)
1347                 mmput (bprm->mm);
1348
1349 out_file:
1350         if (bprm->file) {
1351                 allow_write_access(bprm->file);
1352                 fput(bprm->file);
1353         }
1354 out_kfree:
1355         kfree(bprm);
1356
1357 out_files:
1358         if (displaced)
1359                 reset_files_struct(displaced);
1360 out_ret:
1361         return retval;
1362 }
1363
1364 int set_binfmt(struct linux_binfmt *new)
1365 {
1366         struct linux_binfmt *old = current->binfmt;
1367
1368         if (new) {
1369                 if (!try_module_get(new->module))
1370                         return -1;
1371         }
1372         current->binfmt = new;
1373         if (old)
1374                 module_put(old->module);
1375         return 0;
1376 }
1377
1378 EXPORT_SYMBOL(set_binfmt);
1379
1380 /* format_corename will inspect the pattern parameter, and output a
1381  * name into corename, which must have space for at least
1382  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1383  */
1384 static int format_corename(char *corename, const char *pattern, long signr)
1385 {
1386         const char *pat_ptr = pattern;
1387         char *out_ptr = corename;
1388         char *const out_end = corename + CORENAME_MAX_SIZE;
1389         int rc;
1390         int pid_in_pattern = 0;
1391         int ispipe = 0;
1392
1393         if (*pattern == '|')
1394                 ispipe = 1;
1395
1396         /* Repeat as long as we have more pattern to process and more output
1397            space */
1398         while (*pat_ptr) {
1399                 if (*pat_ptr != '%') {
1400                         if (out_ptr == out_end)
1401                                 goto out;
1402                         *out_ptr++ = *pat_ptr++;
1403                 } else {
1404                         switch (*++pat_ptr) {
1405                         case 0:
1406                                 goto out;
1407                         /* Double percent, output one percent */
1408                         case '%':
1409                                 if (out_ptr == out_end)
1410                                         goto out;
1411                                 *out_ptr++ = '%';
1412                                 break;
1413                         /* pid */
1414                         case 'p':
1415                                 pid_in_pattern = 1;
1416                                 rc = snprintf(out_ptr, out_end - out_ptr,
1417                                               "%d", task_tgid_vnr(current));
1418                                 if (rc > out_end - out_ptr)
1419                                         goto out;
1420                                 out_ptr += rc;
1421                                 break;
1422                         /* uid */
1423                         case 'u':
1424                                 rc = snprintf(out_ptr, out_end - out_ptr,
1425                                               "%d", current->uid);
1426                                 if (rc > out_end - out_ptr)
1427                                         goto out;
1428                                 out_ptr += rc;
1429                                 break;
1430                         /* gid */
1431                         case 'g':
1432                                 rc = snprintf(out_ptr, out_end - out_ptr,
1433                                               "%d", current->gid);
1434                                 if (rc > out_end - out_ptr)
1435                                         goto out;
1436                                 out_ptr += rc;
1437                                 break;
1438                         /* signal that caused the coredump */
1439                         case 's':
1440                                 rc = snprintf(out_ptr, out_end - out_ptr,
1441                                               "%ld", signr);
1442                                 if (rc > out_end - out_ptr)
1443                                         goto out;
1444                                 out_ptr += rc;
1445                                 break;
1446                         /* UNIX time of coredump */
1447                         case 't': {
1448                                 struct timeval tv;
1449                                 do_gettimeofday(&tv);
1450                                 rc = snprintf(out_ptr, out_end - out_ptr,
1451                                               "%lu", tv.tv_sec);
1452                                 if (rc > out_end - out_ptr)
1453                                         goto out;
1454                                 out_ptr += rc;
1455                                 break;
1456                         }
1457                         /* hostname */
1458                         case 'h':
1459                                 down_read(&uts_sem);
1460                                 rc = snprintf(out_ptr, out_end - out_ptr,
1461                                               "%s", utsname()->nodename);
1462                                 up_read(&uts_sem);
1463                                 if (rc > out_end - out_ptr)
1464                                         goto out;
1465                                 out_ptr += rc;
1466                                 break;
1467                         /* executable */
1468                         case 'e':
1469                                 rc = snprintf(out_ptr, out_end - out_ptr,
1470                                               "%s", current->comm);
1471                                 if (rc > out_end - out_ptr)
1472                                         goto out;
1473                                 out_ptr += rc;
1474                                 break;
1475                         /* core limit size */
1476                         case 'c':
1477                                 rc = snprintf(out_ptr, out_end - out_ptr,
1478                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1479                                 if (rc > out_end - out_ptr)
1480                                         goto out;
1481                                 out_ptr += rc;
1482                                 break;
1483                         default:
1484                                 break;
1485                         }
1486                         ++pat_ptr;
1487                 }
1488         }
1489         /* Backward compatibility with core_uses_pid:
1490          *
1491          * If core_pattern does not include a %p (as is the default)
1492          * and core_uses_pid is set, then .%pid will be appended to
1493          * the filename. Do not do this for piped commands. */
1494         if (!ispipe && !pid_in_pattern
1495             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1496                 rc = snprintf(out_ptr, out_end - out_ptr,
1497                               ".%d", task_tgid_vnr(current));
1498                 if (rc > out_end - out_ptr)
1499                         goto out;
1500                 out_ptr += rc;
1501         }
1502 out:
1503         *out_ptr = 0;
1504         return ispipe;
1505 }
1506
1507 static void zap_process(struct task_struct *start)
1508 {
1509         struct task_struct *t;
1510
1511         start->signal->flags = SIGNAL_GROUP_EXIT;
1512         start->signal->group_stop_count = 0;
1513
1514         t = start;
1515         do {
1516                 if (t != current && t->mm) {
1517                         t->mm->core_waiters++;
1518                         sigaddset(&t->pending.signal, SIGKILL);
1519                         signal_wake_up(t, 1);
1520                 }
1521         } while ((t = next_thread(t)) != start);
1522 }
1523
1524 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1525                                 int exit_code)
1526 {
1527         struct task_struct *g, *p;
1528         unsigned long flags;
1529         int err = -EAGAIN;
1530
1531         spin_lock_irq(&tsk->sighand->siglock);
1532         if (!signal_group_exit(tsk->signal)) {
1533                 tsk->signal->group_exit_code = exit_code;
1534                 zap_process(tsk);
1535                 err = 0;
1536         }
1537         spin_unlock_irq(&tsk->sighand->siglock);
1538         if (err)
1539                 return err;
1540
1541         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1542                 goto done;
1543
1544         rcu_read_lock();
1545         for_each_process(g) {
1546                 if (g == tsk->group_leader)
1547                         continue;
1548
1549                 p = g;
1550                 do {
1551                         if (p->mm) {
1552                                 if (p->mm == mm) {
1553                                         /*
1554                                          * p->sighand can't disappear, but
1555                                          * may be changed by de_thread()
1556                                          */
1557                                         lock_task_sighand(p, &flags);
1558                                         zap_process(p);
1559                                         unlock_task_sighand(p, &flags);
1560                                 }
1561                                 break;
1562                         }
1563                 } while ((p = next_thread(p)) != g);
1564         }
1565         rcu_read_unlock();
1566 done:
1567         return mm->core_waiters;
1568 }
1569
1570 static int coredump_wait(int exit_code)
1571 {
1572         struct task_struct *tsk = current;
1573         struct mm_struct *mm = tsk->mm;
1574         struct completion startup_done;
1575         struct completion *vfork_done;
1576         int core_waiters;
1577
1578         init_completion(&mm->core_done);
1579         init_completion(&startup_done);
1580         mm->core_startup_done = &startup_done;
1581
1582         core_waiters = zap_threads(tsk, mm, exit_code);
1583         up_write(&mm->mmap_sem);
1584
1585         if (unlikely(core_waiters < 0))
1586                 goto fail;
1587
1588         /*
1589          * Make sure nobody is waiting for us to release the VM,
1590          * otherwise we can deadlock when we wait on each other
1591          */
1592         vfork_done = tsk->vfork_done;
1593         if (vfork_done) {
1594                 tsk->vfork_done = NULL;
1595                 complete(vfork_done);
1596         }
1597
1598         if (core_waiters)
1599                 wait_for_completion(&startup_done);
1600 fail:
1601         BUG_ON(mm->core_waiters);
1602         return core_waiters;
1603 }
1604
1605 /*
1606  * set_dumpable converts traditional three-value dumpable to two flags and
1607  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1608  * these bits are not changed atomically.  So get_dumpable can observe the
1609  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1610  * return either old dumpable or new one by paying attention to the order of
1611  * modifying the bits.
1612  *
1613  * dumpable |   mm->flags (binary)
1614  * old  new | initial interim  final
1615  * ---------+-----------------------
1616  *  0    1  |   00      01      01
1617  *  0    2  |   00      10(*)   11
1618  *  1    0  |   01      00      00
1619  *  1    2  |   01      11      11
1620  *  2    0  |   11      10(*)   00
1621  *  2    1  |   11      11      01
1622  *
1623  * (*) get_dumpable regards interim value of 10 as 11.
1624  */
1625 void set_dumpable(struct mm_struct *mm, int value)
1626 {
1627         switch (value) {
1628         case 0:
1629                 clear_bit(MMF_DUMPABLE, &mm->flags);
1630                 smp_wmb();
1631                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1632                 break;
1633         case 1:
1634                 set_bit(MMF_DUMPABLE, &mm->flags);
1635                 smp_wmb();
1636                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1637                 break;
1638         case 2:
1639                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1640                 smp_wmb();
1641                 set_bit(MMF_DUMPABLE, &mm->flags);
1642                 break;
1643         }
1644 }
1645
1646 int get_dumpable(struct mm_struct *mm)
1647 {
1648         int ret;
1649
1650         ret = mm->flags & 0x3;
1651         return (ret >= 2) ? 2 : ret;
1652 }
1653
1654 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1655 {
1656         char corename[CORENAME_MAX_SIZE + 1];
1657         struct mm_struct *mm = current->mm;
1658         struct linux_binfmt * binfmt;
1659         struct inode * inode;
1660         struct file * file;
1661         int retval = 0;
1662         int fsuid = current->fsuid;
1663         int flag = 0;
1664         int ispipe = 0;
1665         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1666         char **helper_argv = NULL;
1667         int helper_argc = 0;
1668         char *delimit;
1669
1670         audit_core_dumps(signr);
1671
1672         binfmt = current->binfmt;
1673         if (!binfmt || !binfmt->core_dump)
1674                 goto fail;
1675         down_write(&mm->mmap_sem);
1676         /*
1677          * If another thread got here first, or we are not dumpable, bail out.
1678          */
1679         if (mm->core_waiters || !get_dumpable(mm)) {
1680                 up_write(&mm->mmap_sem);
1681                 goto fail;
1682         }
1683
1684         /*
1685          *      We cannot trust fsuid as being the "true" uid of the
1686          *      process nor do we know its entire history. We only know it
1687          *      was tainted so we dump it as root in mode 2.
1688          */
1689         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1690                 flag = O_EXCL;          /* Stop rewrite attacks */
1691                 current->fsuid = 0;     /* Dump root private */
1692         }
1693
1694         retval = coredump_wait(exit_code);
1695         if (retval < 0)
1696                 goto fail;
1697
1698         /*
1699          * Clear any false indication of pending signals that might
1700          * be seen by the filesystem code called to write the core file.
1701          */
1702         clear_thread_flag(TIF_SIGPENDING);
1703
1704         /*
1705          * lock_kernel() because format_corename() is controlled by sysctl, which
1706          * uses lock_kernel()
1707          */
1708         lock_kernel();
1709         ispipe = format_corename(corename, core_pattern, signr);
1710         unlock_kernel();
1711         /*
1712          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1713          * to a pipe.  Since we're not writing directly to the filesystem
1714          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1715          * created unless the pipe reader choses to write out the core file
1716          * at which point file size limits and permissions will be imposed
1717          * as it does with any other process
1718          */
1719         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1720                 goto fail_unlock;
1721
1722         if (ispipe) {
1723                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1724                 /* Terminate the string before the first option */
1725                 delimit = strchr(corename, ' ');
1726                 if (delimit)
1727                         *delimit = '\0';
1728                 delimit = strrchr(helper_argv[0], '/');
1729                 if (delimit)
1730                         delimit++;
1731                 else
1732                         delimit = helper_argv[0];
1733                 if (!strcmp(delimit, current->comm)) {
1734                         printk(KERN_NOTICE "Recursive core dump detected, "
1735                                         "aborting\n");
1736                         goto fail_unlock;
1737                 }
1738
1739                 core_limit = RLIM_INFINITY;
1740
1741                 /* SIGPIPE can happen, but it's just never processed */
1742                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1743                                 &file)) {
1744                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1745                                corename);
1746                         goto fail_unlock;
1747                 }
1748         } else
1749                 file = filp_open(corename,
1750                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1751                                  0600);
1752         if (IS_ERR(file))
1753                 goto fail_unlock;
1754         inode = file->f_path.dentry->d_inode;
1755         if (inode->i_nlink > 1)
1756                 goto close_fail;        /* multiple links - don't dump */
1757         if (!ispipe && d_unhashed(file->f_path.dentry))
1758                 goto close_fail;
1759
1760         /* AK: actually i see no reason to not allow this for named pipes etc.,
1761            but keep the previous behaviour for now. */
1762         if (!ispipe && !S_ISREG(inode->i_mode))
1763                 goto close_fail;
1764         /*
1765          * Dont allow local users get cute and trick others to coredump
1766          * into their pre-created files:
1767          */
1768         if (inode->i_uid != current->fsuid)
1769                 goto close_fail;
1770         if (!file->f_op)
1771                 goto close_fail;
1772         if (!file->f_op->write)
1773                 goto close_fail;
1774         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1775                 goto close_fail;
1776
1777         retval = binfmt->core_dump(signr, regs, file, core_limit);
1778
1779         if (retval)
1780                 current->signal->group_exit_code |= 0x80;
1781 close_fail:
1782         filp_close(file, NULL);
1783 fail_unlock:
1784         if (helper_argv)
1785                 argv_free(helper_argv);
1786
1787         current->fsuid = fsuid;
1788         complete_all(&mm->core_done);
1789 fail:
1790         return retval;
1791 }