exec: fix stack excutability without PT_GNU_STACK
[linux-2.6.git] / fs / exec.c
1 /*
2  *  linux/fs/exec.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  * #!-checking implemented by tytso.
9  */
10 /*
11  * Demand-loading implemented 01.12.91 - no need to read anything but
12  * the header into memory. The inode of the executable is put into
13  * "current->executable", and page faults do the actual loading. Clean.
14  *
15  * Once more I can proudly say that linux stood up to being changed: it
16  * was less than 2 hours work to get demand-loading completely implemented.
17  *
18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
19  * current->executable is only used by the procfs.  This allows a dispatch
20  * table to check for several different types  of binary formats.  We keep
21  * trying until we recognize the file or we run out of supported binary
22  * formats. 
23  */
24
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/mman.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 #ifdef __alpha__
64 /* for /sbin/loader handling in search_binary_handler() */
65 #include <linux/a.out.h>
66 #endif
67
68 int core_uses_pid;
69 char core_pattern[CORENAME_MAX_SIZE] = "core";
70 int suid_dumpable = 0;
71
72 /* The maximal length of core_pattern is also specified in sysctl.c */
73
74 static LIST_HEAD(formats);
75 static DEFINE_RWLOCK(binfmt_lock);
76
77 int register_binfmt(struct linux_binfmt * fmt)
78 {
79         if (!fmt)
80                 return -EINVAL;
81         write_lock(&binfmt_lock);
82         list_add(&fmt->lh, &formats);
83         write_unlock(&binfmt_lock);
84         return 0;       
85 }
86
87 EXPORT_SYMBOL(register_binfmt);
88
89 void unregister_binfmt(struct linux_binfmt * fmt)
90 {
91         write_lock(&binfmt_lock);
92         list_del(&fmt->lh);
93         write_unlock(&binfmt_lock);
94 }
95
96 EXPORT_SYMBOL(unregister_binfmt);
97
98 static inline void put_binfmt(struct linux_binfmt * fmt)
99 {
100         module_put(fmt->module);
101 }
102
103 /*
104  * Note that a shared library must be both readable and executable due to
105  * security reasons.
106  *
107  * Also note that we take the address to load from from the file itself.
108  */
109 asmlinkage long sys_uselib(const char __user * library)
110 {
111         struct file * file;
112         struct nameidata nd;
113         int error;
114
115         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
116         if (error)
117                 goto out;
118
119         error = -EINVAL;
120         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
121                 goto exit;
122
123         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
124         if (error)
125                 goto exit;
126
127         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
128         error = PTR_ERR(file);
129         if (IS_ERR(file))
130                 goto out;
131
132         error = -ENOEXEC;
133         if(file->f_op) {
134                 struct linux_binfmt * fmt;
135
136                 read_lock(&binfmt_lock);
137                 list_for_each_entry(fmt, &formats, lh) {
138                         if (!fmt->load_shlib)
139                                 continue;
140                         if (!try_module_get(fmt->module))
141                                 continue;
142                         read_unlock(&binfmt_lock);
143                         error = fmt->load_shlib(file);
144                         read_lock(&binfmt_lock);
145                         put_binfmt(fmt);
146                         if (error != -ENOEXEC)
147                                 break;
148                 }
149                 read_unlock(&binfmt_lock);
150         }
151         fput(file);
152 out:
153         return error;
154 exit:
155         release_open_intent(&nd);
156         path_put(&nd.path);
157         goto out;
158 }
159
160 #ifdef CONFIG_MMU
161
162 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
163                 int write)
164 {
165         struct page *page;
166         int ret;
167
168 #ifdef CONFIG_STACK_GROWSUP
169         if (write) {
170                 ret = expand_stack_downwards(bprm->vma, pos);
171                 if (ret < 0)
172                         return NULL;
173         }
174 #endif
175         ret = get_user_pages(current, bprm->mm, pos,
176                         1, write, 1, &page, NULL);
177         if (ret <= 0)
178                 return NULL;
179
180         if (write) {
181                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
182                 struct rlimit *rlim;
183
184                 /*
185                  * We've historically supported up to 32 pages (ARG_MAX)
186                  * of argument strings even with small stacks
187                  */
188                 if (size <= ARG_MAX)
189                         return page;
190
191                 /*
192                  * Limit to 1/4-th the stack size for the argv+env strings.
193                  * This ensures that:
194                  *  - the remaining binfmt code will not run out of stack space,
195                  *  - the program will have a reasonable amount of stack left
196                  *    to work from.
197                  */
198                 rlim = current->signal->rlim;
199                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
200                         put_page(page);
201                         return NULL;
202                 }
203         }
204
205         return page;
206 }
207
208 static void put_arg_page(struct page *page)
209 {
210         put_page(page);
211 }
212
213 static void free_arg_page(struct linux_binprm *bprm, int i)
214 {
215 }
216
217 static void free_arg_pages(struct linux_binprm *bprm)
218 {
219 }
220
221 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
222                 struct page *page)
223 {
224         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
225 }
226
227 static int __bprm_mm_init(struct linux_binprm *bprm)
228 {
229         int err = -ENOMEM;
230         struct vm_area_struct *vma = NULL;
231         struct mm_struct *mm = bprm->mm;
232
233         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
234         if (!vma)
235                 goto err;
236
237         down_write(&mm->mmap_sem);
238         vma->vm_mm = mm;
239
240         /*
241          * Place the stack at the largest stack address the architecture
242          * supports. Later, we'll move this to an appropriate place. We don't
243          * use STACK_TOP because that can depend on attributes which aren't
244          * configured yet.
245          */
246         vma->vm_end = STACK_TOP_MAX;
247         vma->vm_start = vma->vm_end - PAGE_SIZE;
248
249         vma->vm_flags = VM_STACK_FLAGS;
250         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
251         err = insert_vm_struct(mm, vma);
252         if (err) {
253                 up_write(&mm->mmap_sem);
254                 goto err;
255         }
256
257         mm->stack_vm = mm->total_vm = 1;
258         up_write(&mm->mmap_sem);
259
260         bprm->p = vma->vm_end - sizeof(void *);
261
262         return 0;
263
264 err:
265         if (vma) {
266                 bprm->vma = NULL;
267                 kmem_cache_free(vm_area_cachep, vma);
268         }
269
270         return err;
271 }
272
273 static bool valid_arg_len(struct linux_binprm *bprm, long len)
274 {
275         return len <= MAX_ARG_STRLEN;
276 }
277
278 #else
279
280 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
281                 int write)
282 {
283         struct page *page;
284
285         page = bprm->page[pos / PAGE_SIZE];
286         if (!page && write) {
287                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
288                 if (!page)
289                         return NULL;
290                 bprm->page[pos / PAGE_SIZE] = page;
291         }
292
293         return page;
294 }
295
296 static void put_arg_page(struct page *page)
297 {
298 }
299
300 static void free_arg_page(struct linux_binprm *bprm, int i)
301 {
302         if (bprm->page[i]) {
303                 __free_page(bprm->page[i]);
304                 bprm->page[i] = NULL;
305         }
306 }
307
308 static void free_arg_pages(struct linux_binprm *bprm)
309 {
310         int i;
311
312         for (i = 0; i < MAX_ARG_PAGES; i++)
313                 free_arg_page(bprm, i);
314 }
315
316 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
317                 struct page *page)
318 {
319 }
320
321 static int __bprm_mm_init(struct linux_binprm *bprm)
322 {
323         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
324         return 0;
325 }
326
327 static bool valid_arg_len(struct linux_binprm *bprm, long len)
328 {
329         return len <= bprm->p;
330 }
331
332 #endif /* CONFIG_MMU */
333
334 /*
335  * Create a new mm_struct and populate it with a temporary stack
336  * vm_area_struct.  We don't have enough context at this point to set the stack
337  * flags, permissions, and offset, so we use temporary values.  We'll update
338  * them later in setup_arg_pages().
339  */
340 int bprm_mm_init(struct linux_binprm *bprm)
341 {
342         int err;
343         struct mm_struct *mm = NULL;
344
345         bprm->mm = mm = mm_alloc();
346         err = -ENOMEM;
347         if (!mm)
348                 goto err;
349
350         err = init_new_context(current, mm);
351         if (err)
352                 goto err;
353
354         err = __bprm_mm_init(bprm);
355         if (err)
356                 goto err;
357
358         return 0;
359
360 err:
361         if (mm) {
362                 bprm->mm = NULL;
363                 mmdrop(mm);
364         }
365
366         return err;
367 }
368
369 /*
370  * count() counts the number of strings in array ARGV.
371  */
372 static int count(char __user * __user * argv, int max)
373 {
374         int i = 0;
375
376         if (argv != NULL) {
377                 for (;;) {
378                         char __user * p;
379
380                         if (get_user(p, argv))
381                                 return -EFAULT;
382                         if (!p)
383                                 break;
384                         argv++;
385                         if(++i > max)
386                                 return -E2BIG;
387                         cond_resched();
388                 }
389         }
390         return i;
391 }
392
393 /*
394  * 'copy_strings()' copies argument/environment strings from the old
395  * processes's memory to the new process's stack.  The call to get_user_pages()
396  * ensures the destination page is created and not swapped out.
397  */
398 static int copy_strings(int argc, char __user * __user * argv,
399                         struct linux_binprm *bprm)
400 {
401         struct page *kmapped_page = NULL;
402         char *kaddr = NULL;
403         unsigned long kpos = 0;
404         int ret;
405
406         while (argc-- > 0) {
407                 char __user *str;
408                 int len;
409                 unsigned long pos;
410
411                 if (get_user(str, argv+argc) ||
412                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
413                         ret = -EFAULT;
414                         goto out;
415                 }
416
417                 if (!valid_arg_len(bprm, len)) {
418                         ret = -E2BIG;
419                         goto out;
420                 }
421
422                 /* We're going to work our way backwords. */
423                 pos = bprm->p;
424                 str += len;
425                 bprm->p -= len;
426
427                 while (len > 0) {
428                         int offset, bytes_to_copy;
429
430                         offset = pos % PAGE_SIZE;
431                         if (offset == 0)
432                                 offset = PAGE_SIZE;
433
434                         bytes_to_copy = offset;
435                         if (bytes_to_copy > len)
436                                 bytes_to_copy = len;
437
438                         offset -= bytes_to_copy;
439                         pos -= bytes_to_copy;
440                         str -= bytes_to_copy;
441                         len -= bytes_to_copy;
442
443                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
444                                 struct page *page;
445
446                                 page = get_arg_page(bprm, pos, 1);
447                                 if (!page) {
448                                         ret = -E2BIG;
449                                         goto out;
450                                 }
451
452                                 if (kmapped_page) {
453                                         flush_kernel_dcache_page(kmapped_page);
454                                         kunmap(kmapped_page);
455                                         put_arg_page(kmapped_page);
456                                 }
457                                 kmapped_page = page;
458                                 kaddr = kmap(kmapped_page);
459                                 kpos = pos & PAGE_MASK;
460                                 flush_arg_page(bprm, kpos, kmapped_page);
461                         }
462                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
463                                 ret = -EFAULT;
464                                 goto out;
465                         }
466                 }
467         }
468         ret = 0;
469 out:
470         if (kmapped_page) {
471                 flush_kernel_dcache_page(kmapped_page);
472                 kunmap(kmapped_page);
473                 put_arg_page(kmapped_page);
474         }
475         return ret;
476 }
477
478 /*
479  * Like copy_strings, but get argv and its values from kernel memory.
480  */
481 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
482 {
483         int r;
484         mm_segment_t oldfs = get_fs();
485         set_fs(KERNEL_DS);
486         r = copy_strings(argc, (char __user * __user *)argv, bprm);
487         set_fs(oldfs);
488         return r;
489 }
490 EXPORT_SYMBOL(copy_strings_kernel);
491
492 #ifdef CONFIG_MMU
493
494 /*
495  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
496  * the binfmt code determines where the new stack should reside, we shift it to
497  * its final location.  The process proceeds as follows:
498  *
499  * 1) Use shift to calculate the new vma endpoints.
500  * 2) Extend vma to cover both the old and new ranges.  This ensures the
501  *    arguments passed to subsequent functions are consistent.
502  * 3) Move vma's page tables to the new range.
503  * 4) Free up any cleared pgd range.
504  * 5) Shrink the vma to cover only the new range.
505  */
506 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
507 {
508         struct mm_struct *mm = vma->vm_mm;
509         unsigned long old_start = vma->vm_start;
510         unsigned long old_end = vma->vm_end;
511         unsigned long length = old_end - old_start;
512         unsigned long new_start = old_start - shift;
513         unsigned long new_end = old_end - shift;
514         struct mmu_gather *tlb;
515
516         BUG_ON(new_start > new_end);
517
518         /*
519          * ensure there are no vmas between where we want to go
520          * and where we are
521          */
522         if (vma != find_vma(mm, new_start))
523                 return -EFAULT;
524
525         /*
526          * cover the whole range: [new_start, old_end)
527          */
528         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
529
530         /*
531          * move the page tables downwards, on failure we rely on
532          * process cleanup to remove whatever mess we made.
533          */
534         if (length != move_page_tables(vma, old_start,
535                                        vma, new_start, length))
536                 return -ENOMEM;
537
538         lru_add_drain();
539         tlb = tlb_gather_mmu(mm, 0);
540         if (new_end > old_start) {
541                 /*
542                  * when the old and new regions overlap clear from new_end.
543                  */
544                 free_pgd_range(&tlb, new_end, old_end, new_end,
545                         vma->vm_next ? vma->vm_next->vm_start : 0);
546         } else {
547                 /*
548                  * otherwise, clean from old_start; this is done to not touch
549                  * the address space in [new_end, old_start) some architectures
550                  * have constraints on va-space that make this illegal (IA64) -
551                  * for the others its just a little faster.
552                  */
553                 free_pgd_range(&tlb, old_start, old_end, new_end,
554                         vma->vm_next ? vma->vm_next->vm_start : 0);
555         }
556         tlb_finish_mmu(tlb, new_end, old_end);
557
558         /*
559          * shrink the vma to just the new range.
560          */
561         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
562
563         return 0;
564 }
565
566 #define EXTRA_STACK_VM_PAGES    20      /* random */
567
568 /*
569  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
570  * the stack is optionally relocated, and some extra space is added.
571  */
572 int setup_arg_pages(struct linux_binprm *bprm,
573                     unsigned long stack_top,
574                     int executable_stack)
575 {
576         unsigned long ret;
577         unsigned long stack_shift;
578         struct mm_struct *mm = current->mm;
579         struct vm_area_struct *vma = bprm->vma;
580         struct vm_area_struct *prev = NULL;
581         unsigned long vm_flags;
582         unsigned long stack_base;
583
584 #ifdef CONFIG_STACK_GROWSUP
585         /* Limit stack size to 1GB */
586         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
587         if (stack_base > (1 << 30))
588                 stack_base = 1 << 30;
589
590         /* Make sure we didn't let the argument array grow too large. */
591         if (vma->vm_end - vma->vm_start > stack_base)
592                 return -ENOMEM;
593
594         stack_base = PAGE_ALIGN(stack_top - stack_base);
595
596         stack_shift = vma->vm_start - stack_base;
597         mm->arg_start = bprm->p - stack_shift;
598         bprm->p = vma->vm_end - stack_shift;
599 #else
600         stack_top = arch_align_stack(stack_top);
601         stack_top = PAGE_ALIGN(stack_top);
602         stack_shift = vma->vm_end - stack_top;
603
604         bprm->p -= stack_shift;
605         mm->arg_start = bprm->p;
606 #endif
607
608         if (bprm->loader)
609                 bprm->loader -= stack_shift;
610         bprm->exec -= stack_shift;
611
612         down_write(&mm->mmap_sem);
613         vm_flags = VM_STACK_FLAGS;
614
615         /*
616          * Adjust stack execute permissions; explicitly enable for
617          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
618          * (arch default) otherwise.
619          */
620         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
621                 vm_flags |= VM_EXEC;
622         else if (executable_stack == EXSTACK_DISABLE_X)
623                 vm_flags &= ~VM_EXEC;
624         vm_flags |= mm->def_flags;
625
626         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
627                         vm_flags);
628         if (ret)
629                 goto out_unlock;
630         BUG_ON(prev != vma);
631
632         /* Move stack pages down in memory. */
633         if (stack_shift) {
634                 ret = shift_arg_pages(vma, stack_shift);
635                 if (ret) {
636                         up_write(&mm->mmap_sem);
637                         return ret;
638                 }
639         }
640
641 #ifdef CONFIG_STACK_GROWSUP
642         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
643 #else
644         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
645 #endif
646         ret = expand_stack(vma, stack_base);
647         if (ret)
648                 ret = -EFAULT;
649
650 out_unlock:
651         up_write(&mm->mmap_sem);
652         return 0;
653 }
654 EXPORT_SYMBOL(setup_arg_pages);
655
656 #endif /* CONFIG_MMU */
657
658 struct file *open_exec(const char *name)
659 {
660         struct nameidata nd;
661         int err;
662         struct file *file;
663
664         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
665         file = ERR_PTR(err);
666
667         if (!err) {
668                 struct inode *inode = nd.path.dentry->d_inode;
669                 file = ERR_PTR(-EACCES);
670                 if (S_ISREG(inode->i_mode)) {
671                         int err = vfs_permission(&nd, MAY_EXEC);
672                         file = ERR_PTR(err);
673                         if (!err) {
674                                 file = nameidata_to_filp(&nd,
675                                                         O_RDONLY|O_LARGEFILE);
676                                 if (!IS_ERR(file)) {
677                                         err = deny_write_access(file);
678                                         if (err) {
679                                                 fput(file);
680                                                 file = ERR_PTR(err);
681                                         }
682                                 }
683 out:
684                                 return file;
685                         }
686                 }
687                 release_open_intent(&nd);
688                 path_put(&nd.path);
689         }
690         goto out;
691 }
692
693 EXPORT_SYMBOL(open_exec);
694
695 int kernel_read(struct file *file, unsigned long offset,
696         char *addr, unsigned long count)
697 {
698         mm_segment_t old_fs;
699         loff_t pos = offset;
700         int result;
701
702         old_fs = get_fs();
703         set_fs(get_ds());
704         /* The cast to a user pointer is valid due to the set_fs() */
705         result = vfs_read(file, (void __user *)addr, count, &pos);
706         set_fs(old_fs);
707         return result;
708 }
709
710 EXPORT_SYMBOL(kernel_read);
711
712 static int exec_mmap(struct mm_struct *mm)
713 {
714         struct task_struct *tsk;
715         struct mm_struct * old_mm, *active_mm;
716
717         /* Notify parent that we're no longer interested in the old VM */
718         tsk = current;
719         old_mm = current->mm;
720         mm_release(tsk, old_mm);
721
722         if (old_mm) {
723                 /*
724                  * Make sure that if there is a core dump in progress
725                  * for the old mm, we get out and die instead of going
726                  * through with the exec.  We must hold mmap_sem around
727                  * checking core_waiters and changing tsk->mm.  The
728                  * core-inducing thread will increment core_waiters for
729                  * each thread whose ->mm == old_mm.
730                  */
731                 down_read(&old_mm->mmap_sem);
732                 if (unlikely(old_mm->core_waiters)) {
733                         up_read(&old_mm->mmap_sem);
734                         return -EINTR;
735                 }
736         }
737         task_lock(tsk);
738         active_mm = tsk->active_mm;
739         tsk->mm = mm;
740         tsk->active_mm = mm;
741         activate_mm(active_mm, mm);
742         task_unlock(tsk);
743         mm_update_next_owner(old_mm);
744         arch_pick_mmap_layout(mm);
745         if (old_mm) {
746                 up_read(&old_mm->mmap_sem);
747                 BUG_ON(active_mm != old_mm);
748                 mmput(old_mm);
749                 return 0;
750         }
751         mmdrop(active_mm);
752         return 0;
753 }
754
755 /*
756  * This function makes sure the current process has its own signal table,
757  * so that flush_signal_handlers can later reset the handlers without
758  * disturbing other processes.  (Other processes might share the signal
759  * table via the CLONE_SIGHAND option to clone().)
760  */
761 static int de_thread(struct task_struct *tsk)
762 {
763         struct signal_struct *sig = tsk->signal;
764         struct sighand_struct *oldsighand = tsk->sighand;
765         spinlock_t *lock = &oldsighand->siglock;
766         struct task_struct *leader = NULL;
767         int count;
768
769         if (thread_group_empty(tsk))
770                 goto no_thread_group;
771
772         /*
773          * Kill all other threads in the thread group.
774          */
775         spin_lock_irq(lock);
776         if (signal_group_exit(sig)) {
777                 /*
778                  * Another group action in progress, just
779                  * return so that the signal is processed.
780                  */
781                 spin_unlock_irq(lock);
782                 return -EAGAIN;
783         }
784         sig->group_exit_task = tsk;
785         zap_other_threads(tsk);
786
787         /* Account for the thread group leader hanging around: */
788         count = thread_group_leader(tsk) ? 1 : 2;
789         sig->notify_count = count;
790         while (atomic_read(&sig->count) > count) {
791                 __set_current_state(TASK_UNINTERRUPTIBLE);
792                 spin_unlock_irq(lock);
793                 schedule();
794                 spin_lock_irq(lock);
795         }
796         spin_unlock_irq(lock);
797
798         /*
799          * At this point all other threads have exited, all we have to
800          * do is to wait for the thread group leader to become inactive,
801          * and to assume its PID:
802          */
803         if (!thread_group_leader(tsk)) {
804                 leader = tsk->group_leader;
805
806                 sig->notify_count = -1; /* for exit_notify() */
807                 for (;;) {
808                         write_lock_irq(&tasklist_lock);
809                         if (likely(leader->exit_state))
810                                 break;
811                         __set_current_state(TASK_UNINTERRUPTIBLE);
812                         write_unlock_irq(&tasklist_lock);
813                         schedule();
814                 }
815
816                 if (unlikely(task_child_reaper(tsk) == leader))
817                         task_active_pid_ns(tsk)->child_reaper = tsk;
818                 /*
819                  * The only record we have of the real-time age of a
820                  * process, regardless of execs it's done, is start_time.
821                  * All the past CPU time is accumulated in signal_struct
822                  * from sister threads now dead.  But in this non-leader
823                  * exec, nothing survives from the original leader thread,
824                  * whose birth marks the true age of this process now.
825                  * When we take on its identity by switching to its PID, we
826                  * also take its birthdate (always earlier than our own).
827                  */
828                 tsk->start_time = leader->start_time;
829
830                 BUG_ON(!same_thread_group(leader, tsk));
831                 BUG_ON(has_group_leader_pid(tsk));
832                 /*
833                  * An exec() starts a new thread group with the
834                  * TGID of the previous thread group. Rehash the
835                  * two threads with a switched PID, and release
836                  * the former thread group leader:
837                  */
838
839                 /* Become a process group leader with the old leader's pid.
840                  * The old leader becomes a thread of the this thread group.
841                  * Note: The old leader also uses this pid until release_task
842                  *       is called.  Odd but simple and correct.
843                  */
844                 detach_pid(tsk, PIDTYPE_PID);
845                 tsk->pid = leader->pid;
846                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
847                 transfer_pid(leader, tsk, PIDTYPE_PGID);
848                 transfer_pid(leader, tsk, PIDTYPE_SID);
849                 list_replace_rcu(&leader->tasks, &tsk->tasks);
850
851                 tsk->group_leader = tsk;
852                 leader->group_leader = tsk;
853
854                 tsk->exit_signal = SIGCHLD;
855
856                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
857                 leader->exit_state = EXIT_DEAD;
858
859                 write_unlock_irq(&tasklist_lock);
860         }
861
862         sig->group_exit_task = NULL;
863         sig->notify_count = 0;
864
865 no_thread_group:
866         exit_itimers(sig);
867         flush_itimer_signals();
868         if (leader)
869                 release_task(leader);
870
871         if (atomic_read(&oldsighand->count) != 1) {
872                 struct sighand_struct *newsighand;
873                 /*
874                  * This ->sighand is shared with the CLONE_SIGHAND
875                  * but not CLONE_THREAD task, switch to the new one.
876                  */
877                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
878                 if (!newsighand)
879                         return -ENOMEM;
880
881                 atomic_set(&newsighand->count, 1);
882                 memcpy(newsighand->action, oldsighand->action,
883                        sizeof(newsighand->action));
884
885                 write_lock_irq(&tasklist_lock);
886                 spin_lock(&oldsighand->siglock);
887                 rcu_assign_pointer(tsk->sighand, newsighand);
888                 spin_unlock(&oldsighand->siglock);
889                 write_unlock_irq(&tasklist_lock);
890
891                 __cleanup_sighand(oldsighand);
892         }
893
894         BUG_ON(!thread_group_leader(tsk));
895         return 0;
896 }
897
898 /*
899  * These functions flushes out all traces of the currently running executable
900  * so that a new one can be started
901  */
902 static void flush_old_files(struct files_struct * files)
903 {
904         long j = -1;
905         struct fdtable *fdt;
906
907         spin_lock(&files->file_lock);
908         for (;;) {
909                 unsigned long set, i;
910
911                 j++;
912                 i = j * __NFDBITS;
913                 fdt = files_fdtable(files);
914                 if (i >= fdt->max_fds)
915                         break;
916                 set = fdt->close_on_exec->fds_bits[j];
917                 if (!set)
918                         continue;
919                 fdt->close_on_exec->fds_bits[j] = 0;
920                 spin_unlock(&files->file_lock);
921                 for ( ; set ; i++,set >>= 1) {
922                         if (set & 1) {
923                                 sys_close(i);
924                         }
925                 }
926                 spin_lock(&files->file_lock);
927
928         }
929         spin_unlock(&files->file_lock);
930 }
931
932 char *get_task_comm(char *buf, struct task_struct *tsk)
933 {
934         /* buf must be at least sizeof(tsk->comm) in size */
935         task_lock(tsk);
936         strncpy(buf, tsk->comm, sizeof(tsk->comm));
937         task_unlock(tsk);
938         return buf;
939 }
940
941 void set_task_comm(struct task_struct *tsk, char *buf)
942 {
943         task_lock(tsk);
944         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
945         task_unlock(tsk);
946 }
947
948 int flush_old_exec(struct linux_binprm * bprm)
949 {
950         char * name;
951         int i, ch, retval;
952         char tcomm[sizeof(current->comm)];
953
954         /*
955          * Make sure we have a private signal table and that
956          * we are unassociated from the previous thread group.
957          */
958         retval = de_thread(current);
959         if (retval)
960                 goto out;
961
962         set_mm_exe_file(bprm->mm, bprm->file);
963
964         /*
965          * Release all of the old mmap stuff
966          */
967         retval = exec_mmap(bprm->mm);
968         if (retval)
969                 goto out;
970
971         bprm->mm = NULL;                /* We're using it now */
972
973         /* This is the point of no return */
974         current->sas_ss_sp = current->sas_ss_size = 0;
975
976         if (current->euid == current->uid && current->egid == current->gid)
977                 set_dumpable(current->mm, 1);
978         else
979                 set_dumpable(current->mm, suid_dumpable);
980
981         name = bprm->filename;
982
983         /* Copies the binary name from after last slash */
984         for (i=0; (ch = *(name++)) != '\0';) {
985                 if (ch == '/')
986                         i = 0; /* overwrite what we wrote */
987                 else
988                         if (i < (sizeof(tcomm) - 1))
989                                 tcomm[i++] = ch;
990         }
991         tcomm[i] = '\0';
992         set_task_comm(current, tcomm);
993
994         current->flags &= ~PF_RANDOMIZE;
995         flush_thread();
996
997         /* Set the new mm task size. We have to do that late because it may
998          * depend on TIF_32BIT which is only updated in flush_thread() on
999          * some architectures like powerpc
1000          */
1001         current->mm->task_size = TASK_SIZE;
1002
1003         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1004                 suid_keys(current);
1005                 set_dumpable(current->mm, suid_dumpable);
1006                 current->pdeath_signal = 0;
1007         } else if (file_permission(bprm->file, MAY_READ) ||
1008                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1009                 suid_keys(current);
1010                 set_dumpable(current->mm, suid_dumpable);
1011         }
1012
1013         /* An exec changes our domain. We are no longer part of the thread
1014            group */
1015
1016         current->self_exec_id++;
1017                         
1018         flush_signal_handlers(current, 0);
1019         flush_old_files(current->files);
1020
1021         return 0;
1022
1023 out:
1024         return retval;
1025 }
1026
1027 EXPORT_SYMBOL(flush_old_exec);
1028
1029 /* 
1030  * Fill the binprm structure from the inode. 
1031  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1032  */
1033 int prepare_binprm(struct linux_binprm *bprm)
1034 {
1035         int mode;
1036         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1037         int retval;
1038
1039         mode = inode->i_mode;
1040         if (bprm->file->f_op == NULL)
1041                 return -EACCES;
1042
1043         bprm->e_uid = current->euid;
1044         bprm->e_gid = current->egid;
1045
1046         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1047                 /* Set-uid? */
1048                 if (mode & S_ISUID) {
1049                         current->personality &= ~PER_CLEAR_ON_SETID;
1050                         bprm->e_uid = inode->i_uid;
1051                 }
1052
1053                 /* Set-gid? */
1054                 /*
1055                  * If setgid is set but no group execute bit then this
1056                  * is a candidate for mandatory locking, not a setgid
1057                  * executable.
1058                  */
1059                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1060                         current->personality &= ~PER_CLEAR_ON_SETID;
1061                         bprm->e_gid = inode->i_gid;
1062                 }
1063         }
1064
1065         /* fill in binprm security blob */
1066         retval = security_bprm_set(bprm);
1067         if (retval)
1068                 return retval;
1069
1070         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1071         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1072 }
1073
1074 EXPORT_SYMBOL(prepare_binprm);
1075
1076 static int unsafe_exec(struct task_struct *p)
1077 {
1078         int unsafe = 0;
1079         if (p->ptrace & PT_PTRACED) {
1080                 if (p->ptrace & PT_PTRACE_CAP)
1081                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1082                 else
1083                         unsafe |= LSM_UNSAFE_PTRACE;
1084         }
1085         if (atomic_read(&p->fs->count) > 1 ||
1086             atomic_read(&p->files->count) > 1 ||
1087             atomic_read(&p->sighand->count) > 1)
1088                 unsafe |= LSM_UNSAFE_SHARE;
1089
1090         return unsafe;
1091 }
1092
1093 void compute_creds(struct linux_binprm *bprm)
1094 {
1095         int unsafe;
1096
1097         if (bprm->e_uid != current->uid) {
1098                 suid_keys(current);
1099                 current->pdeath_signal = 0;
1100         }
1101         exec_keys(current);
1102
1103         task_lock(current);
1104         unsafe = unsafe_exec(current);
1105         security_bprm_apply_creds(bprm, unsafe);
1106         task_unlock(current);
1107         security_bprm_post_apply_creds(bprm);
1108 }
1109 EXPORT_SYMBOL(compute_creds);
1110
1111 /*
1112  * Arguments are '\0' separated strings found at the location bprm->p
1113  * points to; chop off the first by relocating brpm->p to right after
1114  * the first '\0' encountered.
1115  */
1116 int remove_arg_zero(struct linux_binprm *bprm)
1117 {
1118         int ret = 0;
1119         unsigned long offset;
1120         char *kaddr;
1121         struct page *page;
1122
1123         if (!bprm->argc)
1124                 return 0;
1125
1126         do {
1127                 offset = bprm->p & ~PAGE_MASK;
1128                 page = get_arg_page(bprm, bprm->p, 0);
1129                 if (!page) {
1130                         ret = -EFAULT;
1131                         goto out;
1132                 }
1133                 kaddr = kmap_atomic(page, KM_USER0);
1134
1135                 for (; offset < PAGE_SIZE && kaddr[offset];
1136                                 offset++, bprm->p++)
1137                         ;
1138
1139                 kunmap_atomic(kaddr, KM_USER0);
1140                 put_arg_page(page);
1141
1142                 if (offset == PAGE_SIZE)
1143                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1144         } while (offset == PAGE_SIZE);
1145
1146         bprm->p++;
1147         bprm->argc--;
1148         ret = 0;
1149
1150 out:
1151         return ret;
1152 }
1153 EXPORT_SYMBOL(remove_arg_zero);
1154
1155 /*
1156  * cycle the list of binary formats handler, until one recognizes the image
1157  */
1158 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1159 {
1160         int try,retval;
1161         struct linux_binfmt *fmt;
1162 #ifdef __alpha__
1163         /* handle /sbin/loader.. */
1164         {
1165             struct exec * eh = (struct exec *) bprm->buf;
1166
1167             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1168                 (eh->fh.f_flags & 0x3000) == 0x3000)
1169             {
1170                 struct file * file;
1171                 unsigned long loader;
1172
1173                 allow_write_access(bprm->file);
1174                 fput(bprm->file);
1175                 bprm->file = NULL;
1176
1177                 loader = bprm->vma->vm_end - sizeof(void *);
1178
1179                 file = open_exec("/sbin/loader");
1180                 retval = PTR_ERR(file);
1181                 if (IS_ERR(file))
1182                         return retval;
1183
1184                 /* Remember if the application is TASO.  */
1185                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1186
1187                 bprm->file = file;
1188                 bprm->loader = loader;
1189                 retval = prepare_binprm(bprm);
1190                 if (retval<0)
1191                         return retval;
1192                 /* should call search_binary_handler recursively here,
1193                    but it does not matter */
1194             }
1195         }
1196 #endif
1197         retval = security_bprm_check(bprm);
1198         if (retval)
1199                 return retval;
1200
1201         /* kernel module loader fixup */
1202         /* so we don't try to load run modprobe in kernel space. */
1203         set_fs(USER_DS);
1204
1205         retval = audit_bprm(bprm);
1206         if (retval)
1207                 return retval;
1208
1209         retval = -ENOENT;
1210         for (try=0; try<2; try++) {
1211                 read_lock(&binfmt_lock);
1212                 list_for_each_entry(fmt, &formats, lh) {
1213                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1214                         if (!fn)
1215                                 continue;
1216                         if (!try_module_get(fmt->module))
1217                                 continue;
1218                         read_unlock(&binfmt_lock);
1219                         retval = fn(bprm, regs);
1220                         if (retval >= 0) {
1221                                 put_binfmt(fmt);
1222                                 allow_write_access(bprm->file);
1223                                 if (bprm->file)
1224                                         fput(bprm->file);
1225                                 bprm->file = NULL;
1226                                 current->did_exec = 1;
1227                                 proc_exec_connector(current);
1228                                 return retval;
1229                         }
1230                         read_lock(&binfmt_lock);
1231                         put_binfmt(fmt);
1232                         if (retval != -ENOEXEC || bprm->mm == NULL)
1233                                 break;
1234                         if (!bprm->file) {
1235                                 read_unlock(&binfmt_lock);
1236                                 return retval;
1237                         }
1238                 }
1239                 read_unlock(&binfmt_lock);
1240                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1241                         break;
1242 #ifdef CONFIG_KMOD
1243                 }else{
1244 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1245                         if (printable(bprm->buf[0]) &&
1246                             printable(bprm->buf[1]) &&
1247                             printable(bprm->buf[2]) &&
1248                             printable(bprm->buf[3]))
1249                                 break; /* -ENOEXEC */
1250                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1251 #endif
1252                 }
1253         }
1254         return retval;
1255 }
1256
1257 EXPORT_SYMBOL(search_binary_handler);
1258
1259 void free_bprm(struct linux_binprm *bprm)
1260 {
1261         free_arg_pages(bprm);
1262         kfree(bprm);
1263 }
1264
1265 /*
1266  * sys_execve() executes a new program.
1267  */
1268 int do_execve(char * filename,
1269         char __user *__user *argv,
1270         char __user *__user *envp,
1271         struct pt_regs * regs)
1272 {
1273         struct linux_binprm *bprm;
1274         struct file *file;
1275         struct files_struct *displaced;
1276         int retval;
1277
1278         retval = unshare_files(&displaced);
1279         if (retval)
1280                 goto out_ret;
1281
1282         retval = -ENOMEM;
1283         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1284         if (!bprm)
1285                 goto out_files;
1286
1287         file = open_exec(filename);
1288         retval = PTR_ERR(file);
1289         if (IS_ERR(file))
1290                 goto out_kfree;
1291
1292         sched_exec();
1293
1294         bprm->file = file;
1295         bprm->filename = filename;
1296         bprm->interp = filename;
1297
1298         retval = bprm_mm_init(bprm);
1299         if (retval)
1300                 goto out_file;
1301
1302         bprm->argc = count(argv, MAX_ARG_STRINGS);
1303         if ((retval = bprm->argc) < 0)
1304                 goto out_mm;
1305
1306         bprm->envc = count(envp, MAX_ARG_STRINGS);
1307         if ((retval = bprm->envc) < 0)
1308                 goto out_mm;
1309
1310         retval = security_bprm_alloc(bprm);
1311         if (retval)
1312                 goto out;
1313
1314         retval = prepare_binprm(bprm);
1315         if (retval < 0)
1316                 goto out;
1317
1318         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1319         if (retval < 0)
1320                 goto out;
1321
1322         bprm->exec = bprm->p;
1323         retval = copy_strings(bprm->envc, envp, bprm);
1324         if (retval < 0)
1325                 goto out;
1326
1327         retval = copy_strings(bprm->argc, argv, bprm);
1328         if (retval < 0)
1329                 goto out;
1330
1331         retval = search_binary_handler(bprm,regs);
1332         if (retval >= 0) {
1333                 /* execve success */
1334                 security_bprm_free(bprm);
1335                 acct_update_integrals(current);
1336                 free_bprm(bprm);
1337                 if (displaced)
1338                         put_files_struct(displaced);
1339                 return retval;
1340         }
1341
1342 out:
1343         if (bprm->security)
1344                 security_bprm_free(bprm);
1345
1346 out_mm:
1347         if (bprm->mm)
1348                 mmput (bprm->mm);
1349
1350 out_file:
1351         if (bprm->file) {
1352                 allow_write_access(bprm->file);
1353                 fput(bprm->file);
1354         }
1355 out_kfree:
1356         free_bprm(bprm);
1357
1358 out_files:
1359         if (displaced)
1360                 reset_files_struct(displaced);
1361 out_ret:
1362         return retval;
1363 }
1364
1365 int set_binfmt(struct linux_binfmt *new)
1366 {
1367         struct linux_binfmt *old = current->binfmt;
1368
1369         if (new) {
1370                 if (!try_module_get(new->module))
1371                         return -1;
1372         }
1373         current->binfmt = new;
1374         if (old)
1375                 module_put(old->module);
1376         return 0;
1377 }
1378
1379 EXPORT_SYMBOL(set_binfmt);
1380
1381 /* format_corename will inspect the pattern parameter, and output a
1382  * name into corename, which must have space for at least
1383  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1384  */
1385 static int format_corename(char *corename, const char *pattern, long signr)
1386 {
1387         const char *pat_ptr = pattern;
1388         char *out_ptr = corename;
1389         char *const out_end = corename + CORENAME_MAX_SIZE;
1390         int rc;
1391         int pid_in_pattern = 0;
1392         int ispipe = 0;
1393
1394         if (*pattern == '|')
1395                 ispipe = 1;
1396
1397         /* Repeat as long as we have more pattern to process and more output
1398            space */
1399         while (*pat_ptr) {
1400                 if (*pat_ptr != '%') {
1401                         if (out_ptr == out_end)
1402                                 goto out;
1403                         *out_ptr++ = *pat_ptr++;
1404                 } else {
1405                         switch (*++pat_ptr) {
1406                         case 0:
1407                                 goto out;
1408                         /* Double percent, output one percent */
1409                         case '%':
1410                                 if (out_ptr == out_end)
1411                                         goto out;
1412                                 *out_ptr++ = '%';
1413                                 break;
1414                         /* pid */
1415                         case 'p':
1416                                 pid_in_pattern = 1;
1417                                 rc = snprintf(out_ptr, out_end - out_ptr,
1418                                               "%d", task_tgid_vnr(current));
1419                                 if (rc > out_end - out_ptr)
1420                                         goto out;
1421                                 out_ptr += rc;
1422                                 break;
1423                         /* uid */
1424                         case 'u':
1425                                 rc = snprintf(out_ptr, out_end - out_ptr,
1426                                               "%d", current->uid);
1427                                 if (rc > out_end - out_ptr)
1428                                         goto out;
1429                                 out_ptr += rc;
1430                                 break;
1431                         /* gid */
1432                         case 'g':
1433                                 rc = snprintf(out_ptr, out_end - out_ptr,
1434                                               "%d", current->gid);
1435                                 if (rc > out_end - out_ptr)
1436                                         goto out;
1437                                 out_ptr += rc;
1438                                 break;
1439                         /* signal that caused the coredump */
1440                         case 's':
1441                                 rc = snprintf(out_ptr, out_end - out_ptr,
1442                                               "%ld", signr);
1443                                 if (rc > out_end - out_ptr)
1444                                         goto out;
1445                                 out_ptr += rc;
1446                                 break;
1447                         /* UNIX time of coredump */
1448                         case 't': {
1449                                 struct timeval tv;
1450                                 do_gettimeofday(&tv);
1451                                 rc = snprintf(out_ptr, out_end - out_ptr,
1452                                               "%lu", tv.tv_sec);
1453                                 if (rc > out_end - out_ptr)
1454                                         goto out;
1455                                 out_ptr += rc;
1456                                 break;
1457                         }
1458                         /* hostname */
1459                         case 'h':
1460                                 down_read(&uts_sem);
1461                                 rc = snprintf(out_ptr, out_end - out_ptr,
1462                                               "%s", utsname()->nodename);
1463                                 up_read(&uts_sem);
1464                                 if (rc > out_end - out_ptr)
1465                                         goto out;
1466                                 out_ptr += rc;
1467                                 break;
1468                         /* executable */
1469                         case 'e':
1470                                 rc = snprintf(out_ptr, out_end - out_ptr,
1471                                               "%s", current->comm);
1472                                 if (rc > out_end - out_ptr)
1473                                         goto out;
1474                                 out_ptr += rc;
1475                                 break;
1476                         /* core limit size */
1477                         case 'c':
1478                                 rc = snprintf(out_ptr, out_end - out_ptr,
1479                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1480                                 if (rc > out_end - out_ptr)
1481                                         goto out;
1482                                 out_ptr += rc;
1483                                 break;
1484                         default:
1485                                 break;
1486                         }
1487                         ++pat_ptr;
1488                 }
1489         }
1490         /* Backward compatibility with core_uses_pid:
1491          *
1492          * If core_pattern does not include a %p (as is the default)
1493          * and core_uses_pid is set, then .%pid will be appended to
1494          * the filename. Do not do this for piped commands. */
1495         if (!ispipe && !pid_in_pattern
1496             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1497                 rc = snprintf(out_ptr, out_end - out_ptr,
1498                               ".%d", task_tgid_vnr(current));
1499                 if (rc > out_end - out_ptr)
1500                         goto out;
1501                 out_ptr += rc;
1502         }
1503 out:
1504         *out_ptr = 0;
1505         return ispipe;
1506 }
1507
1508 static void zap_process(struct task_struct *start)
1509 {
1510         struct task_struct *t;
1511
1512         start->signal->flags = SIGNAL_GROUP_EXIT;
1513         start->signal->group_stop_count = 0;
1514
1515         t = start;
1516         do {
1517                 if (t != current && t->mm) {
1518                         t->mm->core_waiters++;
1519                         sigaddset(&t->pending.signal, SIGKILL);
1520                         signal_wake_up(t, 1);
1521                 }
1522         } while ((t = next_thread(t)) != start);
1523 }
1524
1525 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1526                                 int exit_code)
1527 {
1528         struct task_struct *g, *p;
1529         unsigned long flags;
1530         int err = -EAGAIN;
1531
1532         spin_lock_irq(&tsk->sighand->siglock);
1533         if (!signal_group_exit(tsk->signal)) {
1534                 tsk->signal->group_exit_code = exit_code;
1535                 zap_process(tsk);
1536                 err = 0;
1537         }
1538         spin_unlock_irq(&tsk->sighand->siglock);
1539         if (err)
1540                 return err;
1541
1542         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1543                 goto done;
1544
1545         rcu_read_lock();
1546         for_each_process(g) {
1547                 if (g == tsk->group_leader)
1548                         continue;
1549
1550                 p = g;
1551                 do {
1552                         if (p->mm) {
1553                                 if (p->mm == mm) {
1554                                         /*
1555                                          * p->sighand can't disappear, but
1556                                          * may be changed by de_thread()
1557                                          */
1558                                         lock_task_sighand(p, &flags);
1559                                         zap_process(p);
1560                                         unlock_task_sighand(p, &flags);
1561                                 }
1562                                 break;
1563                         }
1564                 } while ((p = next_thread(p)) != g);
1565         }
1566         rcu_read_unlock();
1567 done:
1568         return mm->core_waiters;
1569 }
1570
1571 static int coredump_wait(int exit_code)
1572 {
1573         struct task_struct *tsk = current;
1574         struct mm_struct *mm = tsk->mm;
1575         struct completion startup_done;
1576         struct completion *vfork_done;
1577         int core_waiters;
1578
1579         init_completion(&mm->core_done);
1580         init_completion(&startup_done);
1581         mm->core_startup_done = &startup_done;
1582
1583         core_waiters = zap_threads(tsk, mm, exit_code);
1584         up_write(&mm->mmap_sem);
1585
1586         if (unlikely(core_waiters < 0))
1587                 goto fail;
1588
1589         /*
1590          * Make sure nobody is waiting for us to release the VM,
1591          * otherwise we can deadlock when we wait on each other
1592          */
1593         vfork_done = tsk->vfork_done;
1594         if (vfork_done) {
1595                 tsk->vfork_done = NULL;
1596                 complete(vfork_done);
1597         }
1598
1599         if (core_waiters)
1600                 wait_for_completion(&startup_done);
1601 fail:
1602         BUG_ON(mm->core_waiters);
1603         return core_waiters;
1604 }
1605
1606 /*
1607  * set_dumpable converts traditional three-value dumpable to two flags and
1608  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1609  * these bits are not changed atomically.  So get_dumpable can observe the
1610  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1611  * return either old dumpable or new one by paying attention to the order of
1612  * modifying the bits.
1613  *
1614  * dumpable |   mm->flags (binary)
1615  * old  new | initial interim  final
1616  * ---------+-----------------------
1617  *  0    1  |   00      01      01
1618  *  0    2  |   00      10(*)   11
1619  *  1    0  |   01      00      00
1620  *  1    2  |   01      11      11
1621  *  2    0  |   11      10(*)   00
1622  *  2    1  |   11      11      01
1623  *
1624  * (*) get_dumpable regards interim value of 10 as 11.
1625  */
1626 void set_dumpable(struct mm_struct *mm, int value)
1627 {
1628         switch (value) {
1629         case 0:
1630                 clear_bit(MMF_DUMPABLE, &mm->flags);
1631                 smp_wmb();
1632                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1633                 break;
1634         case 1:
1635                 set_bit(MMF_DUMPABLE, &mm->flags);
1636                 smp_wmb();
1637                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1638                 break;
1639         case 2:
1640                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1641                 smp_wmb();
1642                 set_bit(MMF_DUMPABLE, &mm->flags);
1643                 break;
1644         }
1645 }
1646
1647 int get_dumpable(struct mm_struct *mm)
1648 {
1649         int ret;
1650
1651         ret = mm->flags & 0x3;
1652         return (ret >= 2) ? 2 : ret;
1653 }
1654
1655 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1656 {
1657         char corename[CORENAME_MAX_SIZE + 1];
1658         struct mm_struct *mm = current->mm;
1659         struct linux_binfmt * binfmt;
1660         struct inode * inode;
1661         struct file * file;
1662         int retval = 0;
1663         int fsuid = current->fsuid;
1664         int flag = 0;
1665         int ispipe = 0;
1666         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1667         char **helper_argv = NULL;
1668         int helper_argc = 0;
1669         char *delimit;
1670
1671         audit_core_dumps(signr);
1672
1673         binfmt = current->binfmt;
1674         if (!binfmt || !binfmt->core_dump)
1675                 goto fail;
1676         down_write(&mm->mmap_sem);
1677         /*
1678          * If another thread got here first, or we are not dumpable, bail out.
1679          */
1680         if (mm->core_waiters || !get_dumpable(mm)) {
1681                 up_write(&mm->mmap_sem);
1682                 goto fail;
1683         }
1684
1685         /*
1686          *      We cannot trust fsuid as being the "true" uid of the
1687          *      process nor do we know its entire history. We only know it
1688          *      was tainted so we dump it as root in mode 2.
1689          */
1690         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1691                 flag = O_EXCL;          /* Stop rewrite attacks */
1692                 current->fsuid = 0;     /* Dump root private */
1693         }
1694
1695         retval = coredump_wait(exit_code);
1696         if (retval < 0)
1697                 goto fail;
1698
1699         /*
1700          * Clear any false indication of pending signals that might
1701          * be seen by the filesystem code called to write the core file.
1702          */
1703         clear_thread_flag(TIF_SIGPENDING);
1704
1705         /*
1706          * lock_kernel() because format_corename() is controlled by sysctl, which
1707          * uses lock_kernel()
1708          */
1709         lock_kernel();
1710         ispipe = format_corename(corename, core_pattern, signr);
1711         unlock_kernel();
1712         /*
1713          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1714          * to a pipe.  Since we're not writing directly to the filesystem
1715          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1716          * created unless the pipe reader choses to write out the core file
1717          * at which point file size limits and permissions will be imposed
1718          * as it does with any other process
1719          */
1720         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1721                 goto fail_unlock;
1722
1723         if (ispipe) {
1724                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1725                 /* Terminate the string before the first option */
1726                 delimit = strchr(corename, ' ');
1727                 if (delimit)
1728                         *delimit = '\0';
1729                 delimit = strrchr(helper_argv[0], '/');
1730                 if (delimit)
1731                         delimit++;
1732                 else
1733                         delimit = helper_argv[0];
1734                 if (!strcmp(delimit, current->comm)) {
1735                         printk(KERN_NOTICE "Recursive core dump detected, "
1736                                         "aborting\n");
1737                         goto fail_unlock;
1738                 }
1739
1740                 core_limit = RLIM_INFINITY;
1741
1742                 /* SIGPIPE can happen, but it's just never processed */
1743                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1744                                 &file)) {
1745                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1746                                corename);
1747                         goto fail_unlock;
1748                 }
1749         } else
1750                 file = filp_open(corename,
1751                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1752                                  0600);
1753         if (IS_ERR(file))
1754                 goto fail_unlock;
1755         inode = file->f_path.dentry->d_inode;
1756         if (inode->i_nlink > 1)
1757                 goto close_fail;        /* multiple links - don't dump */
1758         if (!ispipe && d_unhashed(file->f_path.dentry))
1759                 goto close_fail;
1760
1761         /* AK: actually i see no reason to not allow this for named pipes etc.,
1762            but keep the previous behaviour for now. */
1763         if (!ispipe && !S_ISREG(inode->i_mode))
1764                 goto close_fail;
1765         /*
1766          * Dont allow local users get cute and trick others to coredump
1767          * into their pre-created files:
1768          */
1769         if (inode->i_uid != current->fsuid)
1770                 goto close_fail;
1771         if (!file->f_op)
1772                 goto close_fail;
1773         if (!file->f_op->write)
1774                 goto close_fail;
1775         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1776                 goto close_fail;
1777
1778         retval = binfmt->core_dump(signr, regs, file, core_limit);
1779
1780         if (retval)
1781                 current->signal->group_exit_code |= 0x80;
1782 close_fail:
1783         filp_close(file, NULL);
1784 fail_unlock:
1785         if (helper_argv)
1786                 argv_free(helper_argv);
1787
1788         current->fsuid = fsuid;
1789         complete_all(&mm->core_done);
1790 fail:
1791         return retval;
1792 }