mm owner: fix race between swapoff and exit
[linux-2.6.git] / fs / exec.c
1 /*
2  *  linux/fs/exec.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  * #!-checking implemented by tytso.
9  */
10 /*
11  * Demand-loading implemented 01.12.91 - no need to read anything but
12  * the header into memory. The inode of the executable is put into
13  * "current->executable", and page faults do the actual loading. Clean.
14  *
15  * Once more I can proudly say that linux stood up to being changed: it
16  * was less than 2 hours work to get demand-loading completely implemented.
17  *
18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
19  * current->executable is only used by the procfs.  This allows a dispatch
20  * table to check for several different types  of binary formats.  We keep
21  * trying until we recognize the file or we run out of supported binary
22  * formats. 
23  */
24
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/mm.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
53
54 #include <asm/uaccess.h>
55 #include <asm/mmu_context.h>
56 #include <asm/tlb.h>
57
58 #ifdef CONFIG_KMOD
59 #include <linux/kmod.h>
60 #endif
61
62 #ifdef __alpha__
63 /* for /sbin/loader handling in search_binary_handler() */
64 #include <linux/a.out.h>
65 #endif
66
67 int core_uses_pid;
68 char core_pattern[CORENAME_MAX_SIZE] = "core";
69 int suid_dumpable = 0;
70
71 /* The maximal length of core_pattern is also specified in sysctl.c */
72
73 static LIST_HEAD(formats);
74 static DEFINE_RWLOCK(binfmt_lock);
75
76 int register_binfmt(struct linux_binfmt * fmt)
77 {
78         if (!fmt)
79                 return -EINVAL;
80         write_lock(&binfmt_lock);
81         list_add(&fmt->lh, &formats);
82         write_unlock(&binfmt_lock);
83         return 0;       
84 }
85
86 EXPORT_SYMBOL(register_binfmt);
87
88 void unregister_binfmt(struct linux_binfmt * fmt)
89 {
90         write_lock(&binfmt_lock);
91         list_del(&fmt->lh);
92         write_unlock(&binfmt_lock);
93 }
94
95 EXPORT_SYMBOL(unregister_binfmt);
96
97 static inline void put_binfmt(struct linux_binfmt * fmt)
98 {
99         module_put(fmt->module);
100 }
101
102 /*
103  * Note that a shared library must be both readable and executable due to
104  * security reasons.
105  *
106  * Also note that we take the address to load from from the file itself.
107  */
108 asmlinkage long sys_uselib(const char __user * library)
109 {
110         struct file *file;
111         struct nameidata nd;
112         char *tmp = getname(library);
113         int error = PTR_ERR(tmp);
114
115         if (!IS_ERR(tmp)) {
116                 error = path_lookup_open(AT_FDCWD, tmp,
117                                          LOOKUP_FOLLOW, &nd,
118                                          FMODE_READ|FMODE_EXEC);
119                 putname(tmp);
120         }
121         if (error)
122                 goto out;
123
124         error = -EINVAL;
125         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
126                 goto exit;
127
128         error = -EACCES;
129         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
130                 goto exit;
131
132         error = vfs_permission(&nd, MAY_READ | MAY_EXEC | MAY_OPEN);
133         if (error)
134                 goto exit;
135
136         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
137         error = PTR_ERR(file);
138         if (IS_ERR(file))
139                 goto out;
140
141         error = -ENOEXEC;
142         if(file->f_op) {
143                 struct linux_binfmt * fmt;
144
145                 read_lock(&binfmt_lock);
146                 list_for_each_entry(fmt, &formats, lh) {
147                         if (!fmt->load_shlib)
148                                 continue;
149                         if (!try_module_get(fmt->module))
150                                 continue;
151                         read_unlock(&binfmt_lock);
152                         error = fmt->load_shlib(file);
153                         read_lock(&binfmt_lock);
154                         put_binfmt(fmt);
155                         if (error != -ENOEXEC)
156                                 break;
157                 }
158                 read_unlock(&binfmt_lock);
159         }
160         fput(file);
161 out:
162         return error;
163 exit:
164         release_open_intent(&nd);
165         path_put(&nd.path);
166         goto out;
167 }
168
169 #ifdef CONFIG_MMU
170
171 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
172                 int write)
173 {
174         struct page *page;
175         int ret;
176
177 #ifdef CONFIG_STACK_GROWSUP
178         if (write) {
179                 ret = expand_stack_downwards(bprm->vma, pos);
180                 if (ret < 0)
181                         return NULL;
182         }
183 #endif
184         ret = get_user_pages(current, bprm->mm, pos,
185                         1, write, 1, &page, NULL);
186         if (ret <= 0)
187                 return NULL;
188
189         if (write) {
190                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
191                 struct rlimit *rlim;
192
193                 /*
194                  * We've historically supported up to 32 pages (ARG_MAX)
195                  * of argument strings even with small stacks
196                  */
197                 if (size <= ARG_MAX)
198                         return page;
199
200                 /*
201                  * Limit to 1/4-th the stack size for the argv+env strings.
202                  * This ensures that:
203                  *  - the remaining binfmt code will not run out of stack space,
204                  *  - the program will have a reasonable amount of stack left
205                  *    to work from.
206                  */
207                 rlim = current->signal->rlim;
208                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
209                         put_page(page);
210                         return NULL;
211                 }
212         }
213
214         return page;
215 }
216
217 static void put_arg_page(struct page *page)
218 {
219         put_page(page);
220 }
221
222 static void free_arg_page(struct linux_binprm *bprm, int i)
223 {
224 }
225
226 static void free_arg_pages(struct linux_binprm *bprm)
227 {
228 }
229
230 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
231                 struct page *page)
232 {
233         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
234 }
235
236 static int __bprm_mm_init(struct linux_binprm *bprm)
237 {
238         int err = -ENOMEM;
239         struct vm_area_struct *vma = NULL;
240         struct mm_struct *mm = bprm->mm;
241
242         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
243         if (!vma)
244                 goto err;
245
246         down_write(&mm->mmap_sem);
247         vma->vm_mm = mm;
248
249         /*
250          * Place the stack at the largest stack address the architecture
251          * supports. Later, we'll move this to an appropriate place. We don't
252          * use STACK_TOP because that can depend on attributes which aren't
253          * configured yet.
254          */
255         vma->vm_end = STACK_TOP_MAX;
256         vma->vm_start = vma->vm_end - PAGE_SIZE;
257
258         vma->vm_flags = VM_STACK_FLAGS;
259         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
260         err = insert_vm_struct(mm, vma);
261         if (err) {
262                 up_write(&mm->mmap_sem);
263                 goto err;
264         }
265
266         mm->stack_vm = mm->total_vm = 1;
267         up_write(&mm->mmap_sem);
268
269         bprm->p = vma->vm_end - sizeof(void *);
270
271         return 0;
272
273 err:
274         if (vma) {
275                 bprm->vma = NULL;
276                 kmem_cache_free(vm_area_cachep, vma);
277         }
278
279         return err;
280 }
281
282 static bool valid_arg_len(struct linux_binprm *bprm, long len)
283 {
284         return len <= MAX_ARG_STRLEN;
285 }
286
287 #else
288
289 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
290                 int write)
291 {
292         struct page *page;
293
294         page = bprm->page[pos / PAGE_SIZE];
295         if (!page && write) {
296                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
297                 if (!page)
298                         return NULL;
299                 bprm->page[pos / PAGE_SIZE] = page;
300         }
301
302         return page;
303 }
304
305 static void put_arg_page(struct page *page)
306 {
307 }
308
309 static void free_arg_page(struct linux_binprm *bprm, int i)
310 {
311         if (bprm->page[i]) {
312                 __free_page(bprm->page[i]);
313                 bprm->page[i] = NULL;
314         }
315 }
316
317 static void free_arg_pages(struct linux_binprm *bprm)
318 {
319         int i;
320
321         for (i = 0; i < MAX_ARG_PAGES; i++)
322                 free_arg_page(bprm, i);
323 }
324
325 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
326                 struct page *page)
327 {
328 }
329
330 static int __bprm_mm_init(struct linux_binprm *bprm)
331 {
332         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
333         return 0;
334 }
335
336 static bool valid_arg_len(struct linux_binprm *bprm, long len)
337 {
338         return len <= bprm->p;
339 }
340
341 #endif /* CONFIG_MMU */
342
343 /*
344  * Create a new mm_struct and populate it with a temporary stack
345  * vm_area_struct.  We don't have enough context at this point to set the stack
346  * flags, permissions, and offset, so we use temporary values.  We'll update
347  * them later in setup_arg_pages().
348  */
349 int bprm_mm_init(struct linux_binprm *bprm)
350 {
351         int err;
352         struct mm_struct *mm = NULL;
353
354         bprm->mm = mm = mm_alloc();
355         err = -ENOMEM;
356         if (!mm)
357                 goto err;
358
359         err = init_new_context(current, mm);
360         if (err)
361                 goto err;
362
363         err = __bprm_mm_init(bprm);
364         if (err)
365                 goto err;
366
367         return 0;
368
369 err:
370         if (mm) {
371                 bprm->mm = NULL;
372                 mmdrop(mm);
373         }
374
375         return err;
376 }
377
378 /*
379  * count() counts the number of strings in array ARGV.
380  */
381 static int count(char __user * __user * argv, int max)
382 {
383         int i = 0;
384
385         if (argv != NULL) {
386                 for (;;) {
387                         char __user * p;
388
389                         if (get_user(p, argv))
390                                 return -EFAULT;
391                         if (!p)
392                                 break;
393                         argv++;
394                         if(++i > max)
395                                 return -E2BIG;
396                         cond_resched();
397                 }
398         }
399         return i;
400 }
401
402 /*
403  * 'copy_strings()' copies argument/environment strings from the old
404  * processes's memory to the new process's stack.  The call to get_user_pages()
405  * ensures the destination page is created and not swapped out.
406  */
407 static int copy_strings(int argc, char __user * __user * argv,
408                         struct linux_binprm *bprm)
409 {
410         struct page *kmapped_page = NULL;
411         char *kaddr = NULL;
412         unsigned long kpos = 0;
413         int ret;
414
415         while (argc-- > 0) {
416                 char __user *str;
417                 int len;
418                 unsigned long pos;
419
420                 if (get_user(str, argv+argc) ||
421                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
422                         ret = -EFAULT;
423                         goto out;
424                 }
425
426                 if (!valid_arg_len(bprm, len)) {
427                         ret = -E2BIG;
428                         goto out;
429                 }
430
431                 /* We're going to work our way backwords. */
432                 pos = bprm->p;
433                 str += len;
434                 bprm->p -= len;
435
436                 while (len > 0) {
437                         int offset, bytes_to_copy;
438
439                         offset = pos % PAGE_SIZE;
440                         if (offset == 0)
441                                 offset = PAGE_SIZE;
442
443                         bytes_to_copy = offset;
444                         if (bytes_to_copy > len)
445                                 bytes_to_copy = len;
446
447                         offset -= bytes_to_copy;
448                         pos -= bytes_to_copy;
449                         str -= bytes_to_copy;
450                         len -= bytes_to_copy;
451
452                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
453                                 struct page *page;
454
455                                 page = get_arg_page(bprm, pos, 1);
456                                 if (!page) {
457                                         ret = -E2BIG;
458                                         goto out;
459                                 }
460
461                                 if (kmapped_page) {
462                                         flush_kernel_dcache_page(kmapped_page);
463                                         kunmap(kmapped_page);
464                                         put_arg_page(kmapped_page);
465                                 }
466                                 kmapped_page = page;
467                                 kaddr = kmap(kmapped_page);
468                                 kpos = pos & PAGE_MASK;
469                                 flush_arg_page(bprm, kpos, kmapped_page);
470                         }
471                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
472                                 ret = -EFAULT;
473                                 goto out;
474                         }
475                 }
476         }
477         ret = 0;
478 out:
479         if (kmapped_page) {
480                 flush_kernel_dcache_page(kmapped_page);
481                 kunmap(kmapped_page);
482                 put_arg_page(kmapped_page);
483         }
484         return ret;
485 }
486
487 /*
488  * Like copy_strings, but get argv and its values from kernel memory.
489  */
490 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
491 {
492         int r;
493         mm_segment_t oldfs = get_fs();
494         set_fs(KERNEL_DS);
495         r = copy_strings(argc, (char __user * __user *)argv, bprm);
496         set_fs(oldfs);
497         return r;
498 }
499 EXPORT_SYMBOL(copy_strings_kernel);
500
501 #ifdef CONFIG_MMU
502
503 /*
504  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
505  * the binfmt code determines where the new stack should reside, we shift it to
506  * its final location.  The process proceeds as follows:
507  *
508  * 1) Use shift to calculate the new vma endpoints.
509  * 2) Extend vma to cover both the old and new ranges.  This ensures the
510  *    arguments passed to subsequent functions are consistent.
511  * 3) Move vma's page tables to the new range.
512  * 4) Free up any cleared pgd range.
513  * 5) Shrink the vma to cover only the new range.
514  */
515 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
516 {
517         struct mm_struct *mm = vma->vm_mm;
518         unsigned long old_start = vma->vm_start;
519         unsigned long old_end = vma->vm_end;
520         unsigned long length = old_end - old_start;
521         unsigned long new_start = old_start - shift;
522         unsigned long new_end = old_end - shift;
523         struct mmu_gather *tlb;
524
525         BUG_ON(new_start > new_end);
526
527         /*
528          * ensure there are no vmas between where we want to go
529          * and where we are
530          */
531         if (vma != find_vma(mm, new_start))
532                 return -EFAULT;
533
534         /*
535          * cover the whole range: [new_start, old_end)
536          */
537         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
538
539         /*
540          * move the page tables downwards, on failure we rely on
541          * process cleanup to remove whatever mess we made.
542          */
543         if (length != move_page_tables(vma, old_start,
544                                        vma, new_start, length))
545                 return -ENOMEM;
546
547         lru_add_drain();
548         tlb = tlb_gather_mmu(mm, 0);
549         if (new_end > old_start) {
550                 /*
551                  * when the old and new regions overlap clear from new_end.
552                  */
553                 free_pgd_range(tlb, new_end, old_end, new_end,
554                         vma->vm_next ? vma->vm_next->vm_start : 0);
555         } else {
556                 /*
557                  * otherwise, clean from old_start; this is done to not touch
558                  * the address space in [new_end, old_start) some architectures
559                  * have constraints on va-space that make this illegal (IA64) -
560                  * for the others its just a little faster.
561                  */
562                 free_pgd_range(tlb, old_start, old_end, new_end,
563                         vma->vm_next ? vma->vm_next->vm_start : 0);
564         }
565         tlb_finish_mmu(tlb, new_end, old_end);
566
567         /*
568          * shrink the vma to just the new range.
569          */
570         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
571
572         return 0;
573 }
574
575 #define EXTRA_STACK_VM_PAGES    20      /* random */
576
577 /*
578  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
579  * the stack is optionally relocated, and some extra space is added.
580  */
581 int setup_arg_pages(struct linux_binprm *bprm,
582                     unsigned long stack_top,
583                     int executable_stack)
584 {
585         unsigned long ret;
586         unsigned long stack_shift;
587         struct mm_struct *mm = current->mm;
588         struct vm_area_struct *vma = bprm->vma;
589         struct vm_area_struct *prev = NULL;
590         unsigned long vm_flags;
591         unsigned long stack_base;
592
593 #ifdef CONFIG_STACK_GROWSUP
594         /* Limit stack size to 1GB */
595         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
596         if (stack_base > (1 << 30))
597                 stack_base = 1 << 30;
598
599         /* Make sure we didn't let the argument array grow too large. */
600         if (vma->vm_end - vma->vm_start > stack_base)
601                 return -ENOMEM;
602
603         stack_base = PAGE_ALIGN(stack_top - stack_base);
604
605         stack_shift = vma->vm_start - stack_base;
606         mm->arg_start = bprm->p - stack_shift;
607         bprm->p = vma->vm_end - stack_shift;
608 #else
609         stack_top = arch_align_stack(stack_top);
610         stack_top = PAGE_ALIGN(stack_top);
611         stack_shift = vma->vm_end - stack_top;
612
613         bprm->p -= stack_shift;
614         mm->arg_start = bprm->p;
615 #endif
616
617         if (bprm->loader)
618                 bprm->loader -= stack_shift;
619         bprm->exec -= stack_shift;
620
621         down_write(&mm->mmap_sem);
622         vm_flags = VM_STACK_FLAGS;
623
624         /*
625          * Adjust stack execute permissions; explicitly enable for
626          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
627          * (arch default) otherwise.
628          */
629         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
630                 vm_flags |= VM_EXEC;
631         else if (executable_stack == EXSTACK_DISABLE_X)
632                 vm_flags &= ~VM_EXEC;
633         vm_flags |= mm->def_flags;
634
635         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
636                         vm_flags);
637         if (ret)
638                 goto out_unlock;
639         BUG_ON(prev != vma);
640
641         /* Move stack pages down in memory. */
642         if (stack_shift) {
643                 ret = shift_arg_pages(vma, stack_shift);
644                 if (ret) {
645                         up_write(&mm->mmap_sem);
646                         return ret;
647                 }
648         }
649
650 #ifdef CONFIG_STACK_GROWSUP
651         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
652 #else
653         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
654 #endif
655         ret = expand_stack(vma, stack_base);
656         if (ret)
657                 ret = -EFAULT;
658
659 out_unlock:
660         up_write(&mm->mmap_sem);
661         return 0;
662 }
663 EXPORT_SYMBOL(setup_arg_pages);
664
665 #endif /* CONFIG_MMU */
666
667 struct file *open_exec(const char *name)
668 {
669         struct nameidata nd;
670         struct file *file;
671         int err;
672
673         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
674                                 FMODE_READ|FMODE_EXEC);
675         if (err)
676                 goto out;
677
678         err = -EACCES;
679         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
680                 goto out_path_put;
681
682         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
683                 goto out_path_put;
684
685         err = vfs_permission(&nd, MAY_EXEC | MAY_OPEN);
686         if (err)
687                 goto out_path_put;
688
689         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
690         if (IS_ERR(file))
691                 return file;
692
693         err = deny_write_access(file);
694         if (err) {
695                 fput(file);
696                 goto out;
697         }
698
699         return file;
700
701  out_path_put:
702         release_open_intent(&nd);
703         path_put(&nd.path);
704  out:
705         return ERR_PTR(err);
706 }
707 EXPORT_SYMBOL(open_exec);
708
709 int kernel_read(struct file *file, unsigned long offset,
710         char *addr, unsigned long count)
711 {
712         mm_segment_t old_fs;
713         loff_t pos = offset;
714         int result;
715
716         old_fs = get_fs();
717         set_fs(get_ds());
718         /* The cast to a user pointer is valid due to the set_fs() */
719         result = vfs_read(file, (void __user *)addr, count, &pos);
720         set_fs(old_fs);
721         return result;
722 }
723
724 EXPORT_SYMBOL(kernel_read);
725
726 static int exec_mmap(struct mm_struct *mm)
727 {
728         struct task_struct *tsk;
729         struct mm_struct * old_mm, *active_mm;
730
731         /* Notify parent that we're no longer interested in the old VM */
732         tsk = current;
733         old_mm = current->mm;
734         mm_release(tsk, old_mm);
735
736         if (old_mm) {
737                 /*
738                  * Make sure that if there is a core dump in progress
739                  * for the old mm, we get out and die instead of going
740                  * through with the exec.  We must hold mmap_sem around
741                  * checking core_state and changing tsk->mm.
742                  */
743                 down_read(&old_mm->mmap_sem);
744                 if (unlikely(old_mm->core_state)) {
745                         up_read(&old_mm->mmap_sem);
746                         return -EINTR;
747                 }
748         }
749         task_lock(tsk);
750         active_mm = tsk->active_mm;
751         tsk->mm = mm;
752         tsk->active_mm = mm;
753         activate_mm(active_mm, mm);
754         task_unlock(tsk);
755         arch_pick_mmap_layout(mm);
756         if (old_mm) {
757                 up_read(&old_mm->mmap_sem);
758                 BUG_ON(active_mm != old_mm);
759                 mm_update_next_owner(old_mm);
760                 mmput(old_mm);
761                 return 0;
762         }
763         mmdrop(active_mm);
764         return 0;
765 }
766
767 /*
768  * This function makes sure the current process has its own signal table,
769  * so that flush_signal_handlers can later reset the handlers without
770  * disturbing other processes.  (Other processes might share the signal
771  * table via the CLONE_SIGHAND option to clone().)
772  */
773 static int de_thread(struct task_struct *tsk)
774 {
775         struct signal_struct *sig = tsk->signal;
776         struct sighand_struct *oldsighand = tsk->sighand;
777         spinlock_t *lock = &oldsighand->siglock;
778         struct task_struct *leader = NULL;
779         int count;
780
781         if (thread_group_empty(tsk))
782                 goto no_thread_group;
783
784         /*
785          * Kill all other threads in the thread group.
786          */
787         spin_lock_irq(lock);
788         if (signal_group_exit(sig)) {
789                 /*
790                  * Another group action in progress, just
791                  * return so that the signal is processed.
792                  */
793                 spin_unlock_irq(lock);
794                 return -EAGAIN;
795         }
796         sig->group_exit_task = tsk;
797         zap_other_threads(tsk);
798
799         /* Account for the thread group leader hanging around: */
800         count = thread_group_leader(tsk) ? 1 : 2;
801         sig->notify_count = count;
802         while (atomic_read(&sig->count) > count) {
803                 __set_current_state(TASK_UNINTERRUPTIBLE);
804                 spin_unlock_irq(lock);
805                 schedule();
806                 spin_lock_irq(lock);
807         }
808         spin_unlock_irq(lock);
809
810         /*
811          * At this point all other threads have exited, all we have to
812          * do is to wait for the thread group leader to become inactive,
813          * and to assume its PID:
814          */
815         if (!thread_group_leader(tsk)) {
816                 leader = tsk->group_leader;
817
818                 sig->notify_count = -1; /* for exit_notify() */
819                 for (;;) {
820                         write_lock_irq(&tasklist_lock);
821                         if (likely(leader->exit_state))
822                                 break;
823                         __set_current_state(TASK_UNINTERRUPTIBLE);
824                         write_unlock_irq(&tasklist_lock);
825                         schedule();
826                 }
827
828                 if (unlikely(task_child_reaper(tsk) == leader))
829                         task_active_pid_ns(tsk)->child_reaper = tsk;
830                 /*
831                  * The only record we have of the real-time age of a
832                  * process, regardless of execs it's done, is start_time.
833                  * All the past CPU time is accumulated in signal_struct
834                  * from sister threads now dead.  But in this non-leader
835                  * exec, nothing survives from the original leader thread,
836                  * whose birth marks the true age of this process now.
837                  * When we take on its identity by switching to its PID, we
838                  * also take its birthdate (always earlier than our own).
839                  */
840                 tsk->start_time = leader->start_time;
841
842                 BUG_ON(!same_thread_group(leader, tsk));
843                 BUG_ON(has_group_leader_pid(tsk));
844                 /*
845                  * An exec() starts a new thread group with the
846                  * TGID of the previous thread group. Rehash the
847                  * two threads with a switched PID, and release
848                  * the former thread group leader:
849                  */
850
851                 /* Become a process group leader with the old leader's pid.
852                  * The old leader becomes a thread of the this thread group.
853                  * Note: The old leader also uses this pid until release_task
854                  *       is called.  Odd but simple and correct.
855                  */
856                 detach_pid(tsk, PIDTYPE_PID);
857                 tsk->pid = leader->pid;
858                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
859                 transfer_pid(leader, tsk, PIDTYPE_PGID);
860                 transfer_pid(leader, tsk, PIDTYPE_SID);
861                 list_replace_rcu(&leader->tasks, &tsk->tasks);
862
863                 tsk->group_leader = tsk;
864                 leader->group_leader = tsk;
865
866                 tsk->exit_signal = SIGCHLD;
867
868                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
869                 leader->exit_state = EXIT_DEAD;
870
871                 write_unlock_irq(&tasklist_lock);
872         }
873
874         sig->group_exit_task = NULL;
875         sig->notify_count = 0;
876
877 no_thread_group:
878         exit_itimers(sig);
879         flush_itimer_signals();
880         if (leader)
881                 release_task(leader);
882
883         if (atomic_read(&oldsighand->count) != 1) {
884                 struct sighand_struct *newsighand;
885                 /*
886                  * This ->sighand is shared with the CLONE_SIGHAND
887                  * but not CLONE_THREAD task, switch to the new one.
888                  */
889                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
890                 if (!newsighand)
891                         return -ENOMEM;
892
893                 atomic_set(&newsighand->count, 1);
894                 memcpy(newsighand->action, oldsighand->action,
895                        sizeof(newsighand->action));
896
897                 write_lock_irq(&tasklist_lock);
898                 spin_lock(&oldsighand->siglock);
899                 rcu_assign_pointer(tsk->sighand, newsighand);
900                 spin_unlock(&oldsighand->siglock);
901                 write_unlock_irq(&tasklist_lock);
902
903                 __cleanup_sighand(oldsighand);
904         }
905
906         BUG_ON(!thread_group_leader(tsk));
907         return 0;
908 }
909
910 /*
911  * These functions flushes out all traces of the currently running executable
912  * so that a new one can be started
913  */
914 static void flush_old_files(struct files_struct * files)
915 {
916         long j = -1;
917         struct fdtable *fdt;
918
919         spin_lock(&files->file_lock);
920         for (;;) {
921                 unsigned long set, i;
922
923                 j++;
924                 i = j * __NFDBITS;
925                 fdt = files_fdtable(files);
926                 if (i >= fdt->max_fds)
927                         break;
928                 set = fdt->close_on_exec->fds_bits[j];
929                 if (!set)
930                         continue;
931                 fdt->close_on_exec->fds_bits[j] = 0;
932                 spin_unlock(&files->file_lock);
933                 for ( ; set ; i++,set >>= 1) {
934                         if (set & 1) {
935                                 sys_close(i);
936                         }
937                 }
938                 spin_lock(&files->file_lock);
939
940         }
941         spin_unlock(&files->file_lock);
942 }
943
944 char *get_task_comm(char *buf, struct task_struct *tsk)
945 {
946         /* buf must be at least sizeof(tsk->comm) in size */
947         task_lock(tsk);
948         strncpy(buf, tsk->comm, sizeof(tsk->comm));
949         task_unlock(tsk);
950         return buf;
951 }
952
953 void set_task_comm(struct task_struct *tsk, char *buf)
954 {
955         task_lock(tsk);
956         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
957         task_unlock(tsk);
958 }
959
960 int flush_old_exec(struct linux_binprm * bprm)
961 {
962         char * name;
963         int i, ch, retval;
964         char tcomm[sizeof(current->comm)];
965
966         /*
967          * Make sure we have a private signal table and that
968          * we are unassociated from the previous thread group.
969          */
970         retval = de_thread(current);
971         if (retval)
972                 goto out;
973
974         set_mm_exe_file(bprm->mm, bprm->file);
975
976         /*
977          * Release all of the old mmap stuff
978          */
979         retval = exec_mmap(bprm->mm);
980         if (retval)
981                 goto out;
982
983         bprm->mm = NULL;                /* We're using it now */
984
985         /* This is the point of no return */
986         current->sas_ss_sp = current->sas_ss_size = 0;
987
988         if (current->euid == current->uid && current->egid == current->gid)
989                 set_dumpable(current->mm, 1);
990         else
991                 set_dumpable(current->mm, suid_dumpable);
992
993         name = bprm->filename;
994
995         /* Copies the binary name from after last slash */
996         for (i=0; (ch = *(name++)) != '\0';) {
997                 if (ch == '/')
998                         i = 0; /* overwrite what we wrote */
999                 else
1000                         if (i < (sizeof(tcomm) - 1))
1001                                 tcomm[i++] = ch;
1002         }
1003         tcomm[i] = '\0';
1004         set_task_comm(current, tcomm);
1005
1006         current->flags &= ~PF_RANDOMIZE;
1007         flush_thread();
1008
1009         /* Set the new mm task size. We have to do that late because it may
1010          * depend on TIF_32BIT which is only updated in flush_thread() on
1011          * some architectures like powerpc
1012          */
1013         current->mm->task_size = TASK_SIZE;
1014
1015         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1016                 suid_keys(current);
1017                 set_dumpable(current->mm, suid_dumpable);
1018                 current->pdeath_signal = 0;
1019         } else if (file_permission(bprm->file, MAY_READ) ||
1020                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1021                 suid_keys(current);
1022                 set_dumpable(current->mm, suid_dumpable);
1023         }
1024
1025         /* An exec changes our domain. We are no longer part of the thread
1026            group */
1027
1028         current->self_exec_id++;
1029                         
1030         flush_signal_handlers(current, 0);
1031         flush_old_files(current->files);
1032
1033         return 0;
1034
1035 out:
1036         return retval;
1037 }
1038
1039 EXPORT_SYMBOL(flush_old_exec);
1040
1041 /* 
1042  * Fill the binprm structure from the inode. 
1043  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1044  */
1045 int prepare_binprm(struct linux_binprm *bprm)
1046 {
1047         int mode;
1048         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1049         int retval;
1050
1051         mode = inode->i_mode;
1052         if (bprm->file->f_op == NULL)
1053                 return -EACCES;
1054
1055         bprm->e_uid = current->euid;
1056         bprm->e_gid = current->egid;
1057
1058         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1059                 /* Set-uid? */
1060                 if (mode & S_ISUID) {
1061                         current->personality &= ~PER_CLEAR_ON_SETID;
1062                         bprm->e_uid = inode->i_uid;
1063                 }
1064
1065                 /* Set-gid? */
1066                 /*
1067                  * If setgid is set but no group execute bit then this
1068                  * is a candidate for mandatory locking, not a setgid
1069                  * executable.
1070                  */
1071                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1072                         current->personality &= ~PER_CLEAR_ON_SETID;
1073                         bprm->e_gid = inode->i_gid;
1074                 }
1075         }
1076
1077         /* fill in binprm security blob */
1078         retval = security_bprm_set(bprm);
1079         if (retval)
1080                 return retval;
1081
1082         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1083         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1084 }
1085
1086 EXPORT_SYMBOL(prepare_binprm);
1087
1088 static int unsafe_exec(struct task_struct *p)
1089 {
1090         int unsafe = tracehook_unsafe_exec(p);
1091
1092         if (atomic_read(&p->fs->count) > 1 ||
1093             atomic_read(&p->files->count) > 1 ||
1094             atomic_read(&p->sighand->count) > 1)
1095                 unsafe |= LSM_UNSAFE_SHARE;
1096
1097         return unsafe;
1098 }
1099
1100 void compute_creds(struct linux_binprm *bprm)
1101 {
1102         int unsafe;
1103
1104         if (bprm->e_uid != current->uid) {
1105                 suid_keys(current);
1106                 current->pdeath_signal = 0;
1107         }
1108         exec_keys(current);
1109
1110         task_lock(current);
1111         unsafe = unsafe_exec(current);
1112         security_bprm_apply_creds(bprm, unsafe);
1113         task_unlock(current);
1114         security_bprm_post_apply_creds(bprm);
1115 }
1116 EXPORT_SYMBOL(compute_creds);
1117
1118 /*
1119  * Arguments are '\0' separated strings found at the location bprm->p
1120  * points to; chop off the first by relocating brpm->p to right after
1121  * the first '\0' encountered.
1122  */
1123 int remove_arg_zero(struct linux_binprm *bprm)
1124 {
1125         int ret = 0;
1126         unsigned long offset;
1127         char *kaddr;
1128         struct page *page;
1129
1130         if (!bprm->argc)
1131                 return 0;
1132
1133         do {
1134                 offset = bprm->p & ~PAGE_MASK;
1135                 page = get_arg_page(bprm, bprm->p, 0);
1136                 if (!page) {
1137                         ret = -EFAULT;
1138                         goto out;
1139                 }
1140                 kaddr = kmap_atomic(page, KM_USER0);
1141
1142                 for (; offset < PAGE_SIZE && kaddr[offset];
1143                                 offset++, bprm->p++)
1144                         ;
1145
1146                 kunmap_atomic(kaddr, KM_USER0);
1147                 put_arg_page(page);
1148
1149                 if (offset == PAGE_SIZE)
1150                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1151         } while (offset == PAGE_SIZE);
1152
1153         bprm->p++;
1154         bprm->argc--;
1155         ret = 0;
1156
1157 out:
1158         return ret;
1159 }
1160 EXPORT_SYMBOL(remove_arg_zero);
1161
1162 /*
1163  * cycle the list of binary formats handler, until one recognizes the image
1164  */
1165 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1166 {
1167         int try,retval;
1168         struct linux_binfmt *fmt;
1169 #ifdef __alpha__
1170         /* handle /sbin/loader.. */
1171         {
1172             struct exec * eh = (struct exec *) bprm->buf;
1173
1174             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1175                 (eh->fh.f_flags & 0x3000) == 0x3000)
1176             {
1177                 struct file * file;
1178                 unsigned long loader;
1179
1180                 allow_write_access(bprm->file);
1181                 fput(bprm->file);
1182                 bprm->file = NULL;
1183
1184                 loader = bprm->vma->vm_end - sizeof(void *);
1185
1186                 file = open_exec("/sbin/loader");
1187                 retval = PTR_ERR(file);
1188                 if (IS_ERR(file))
1189                         return retval;
1190
1191                 /* Remember if the application is TASO.  */
1192                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1193
1194                 bprm->file = file;
1195                 bprm->loader = loader;
1196                 retval = prepare_binprm(bprm);
1197                 if (retval<0)
1198                         return retval;
1199                 /* should call search_binary_handler recursively here,
1200                    but it does not matter */
1201             }
1202         }
1203 #endif
1204         retval = security_bprm_check(bprm);
1205         if (retval)
1206                 return retval;
1207
1208         /* kernel module loader fixup */
1209         /* so we don't try to load run modprobe in kernel space. */
1210         set_fs(USER_DS);
1211
1212         retval = audit_bprm(bprm);
1213         if (retval)
1214                 return retval;
1215
1216         retval = -ENOENT;
1217         for (try=0; try<2; try++) {
1218                 read_lock(&binfmt_lock);
1219                 list_for_each_entry(fmt, &formats, lh) {
1220                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1221                         if (!fn)
1222                                 continue;
1223                         if (!try_module_get(fmt->module))
1224                                 continue;
1225                         read_unlock(&binfmt_lock);
1226                         retval = fn(bprm, regs);
1227                         if (retval >= 0) {
1228                                 tracehook_report_exec(fmt, bprm, regs);
1229                                 put_binfmt(fmt);
1230                                 allow_write_access(bprm->file);
1231                                 if (bprm->file)
1232                                         fput(bprm->file);
1233                                 bprm->file = NULL;
1234                                 current->did_exec = 1;
1235                                 proc_exec_connector(current);
1236                                 return retval;
1237                         }
1238                         read_lock(&binfmt_lock);
1239                         put_binfmt(fmt);
1240                         if (retval != -ENOEXEC || bprm->mm == NULL)
1241                                 break;
1242                         if (!bprm->file) {
1243                                 read_unlock(&binfmt_lock);
1244                                 return retval;
1245                         }
1246                 }
1247                 read_unlock(&binfmt_lock);
1248                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1249                         break;
1250 #ifdef CONFIG_KMOD
1251                 }else{
1252 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1253                         if (printable(bprm->buf[0]) &&
1254                             printable(bprm->buf[1]) &&
1255                             printable(bprm->buf[2]) &&
1256                             printable(bprm->buf[3]))
1257                                 break; /* -ENOEXEC */
1258                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1259 #endif
1260                 }
1261         }
1262         return retval;
1263 }
1264
1265 EXPORT_SYMBOL(search_binary_handler);
1266
1267 void free_bprm(struct linux_binprm *bprm)
1268 {
1269         free_arg_pages(bprm);
1270         kfree(bprm);
1271 }
1272
1273 /*
1274  * sys_execve() executes a new program.
1275  */
1276 int do_execve(char * filename,
1277         char __user *__user *argv,
1278         char __user *__user *envp,
1279         struct pt_regs * regs)
1280 {
1281         struct linux_binprm *bprm;
1282         struct file *file;
1283         struct files_struct *displaced;
1284         int retval;
1285
1286         retval = unshare_files(&displaced);
1287         if (retval)
1288                 goto out_ret;
1289
1290         retval = -ENOMEM;
1291         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1292         if (!bprm)
1293                 goto out_files;
1294
1295         file = open_exec(filename);
1296         retval = PTR_ERR(file);
1297         if (IS_ERR(file))
1298                 goto out_kfree;
1299
1300         sched_exec();
1301
1302         bprm->file = file;
1303         bprm->filename = filename;
1304         bprm->interp = filename;
1305
1306         retval = bprm_mm_init(bprm);
1307         if (retval)
1308                 goto out_file;
1309
1310         bprm->argc = count(argv, MAX_ARG_STRINGS);
1311         if ((retval = bprm->argc) < 0)
1312                 goto out_mm;
1313
1314         bprm->envc = count(envp, MAX_ARG_STRINGS);
1315         if ((retval = bprm->envc) < 0)
1316                 goto out_mm;
1317
1318         retval = security_bprm_alloc(bprm);
1319         if (retval)
1320                 goto out;
1321
1322         retval = prepare_binprm(bprm);
1323         if (retval < 0)
1324                 goto out;
1325
1326         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1327         if (retval < 0)
1328                 goto out;
1329
1330         bprm->exec = bprm->p;
1331         retval = copy_strings(bprm->envc, envp, bprm);
1332         if (retval < 0)
1333                 goto out;
1334
1335         retval = copy_strings(bprm->argc, argv, bprm);
1336         if (retval < 0)
1337                 goto out;
1338
1339         current->flags &= ~PF_KTHREAD;
1340         retval = search_binary_handler(bprm,regs);
1341         if (retval >= 0) {
1342                 /* execve success */
1343                 security_bprm_free(bprm);
1344                 acct_update_integrals(current);
1345                 free_bprm(bprm);
1346                 if (displaced)
1347                         put_files_struct(displaced);
1348                 return retval;
1349         }
1350
1351 out:
1352         if (bprm->security)
1353                 security_bprm_free(bprm);
1354
1355 out_mm:
1356         if (bprm->mm)
1357                 mmput (bprm->mm);
1358
1359 out_file:
1360         if (bprm->file) {
1361                 allow_write_access(bprm->file);
1362                 fput(bprm->file);
1363         }
1364 out_kfree:
1365         free_bprm(bprm);
1366
1367 out_files:
1368         if (displaced)
1369                 reset_files_struct(displaced);
1370 out_ret:
1371         return retval;
1372 }
1373
1374 int set_binfmt(struct linux_binfmt *new)
1375 {
1376         struct linux_binfmt *old = current->binfmt;
1377
1378         if (new) {
1379                 if (!try_module_get(new->module))
1380                         return -1;
1381         }
1382         current->binfmt = new;
1383         if (old)
1384                 module_put(old->module);
1385         return 0;
1386 }
1387
1388 EXPORT_SYMBOL(set_binfmt);
1389
1390 /* format_corename will inspect the pattern parameter, and output a
1391  * name into corename, which must have space for at least
1392  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1393  */
1394 static int format_corename(char *corename, int nr_threads, long signr)
1395 {
1396         const char *pat_ptr = core_pattern;
1397         int ispipe = (*pat_ptr == '|');
1398         char *out_ptr = corename;
1399         char *const out_end = corename + CORENAME_MAX_SIZE;
1400         int rc;
1401         int pid_in_pattern = 0;
1402
1403         /* Repeat as long as we have more pattern to process and more output
1404            space */
1405         while (*pat_ptr) {
1406                 if (*pat_ptr != '%') {
1407                         if (out_ptr == out_end)
1408                                 goto out;
1409                         *out_ptr++ = *pat_ptr++;
1410                 } else {
1411                         switch (*++pat_ptr) {
1412                         case 0:
1413                                 goto out;
1414                         /* Double percent, output one percent */
1415                         case '%':
1416                                 if (out_ptr == out_end)
1417                                         goto out;
1418                                 *out_ptr++ = '%';
1419                                 break;
1420                         /* pid */
1421                         case 'p':
1422                                 pid_in_pattern = 1;
1423                                 rc = snprintf(out_ptr, out_end - out_ptr,
1424                                               "%d", task_tgid_vnr(current));
1425                                 if (rc > out_end - out_ptr)
1426                                         goto out;
1427                                 out_ptr += rc;
1428                                 break;
1429                         /* uid */
1430                         case 'u':
1431                                 rc = snprintf(out_ptr, out_end - out_ptr,
1432                                               "%d", current->uid);
1433                                 if (rc > out_end - out_ptr)
1434                                         goto out;
1435                                 out_ptr += rc;
1436                                 break;
1437                         /* gid */
1438                         case 'g':
1439                                 rc = snprintf(out_ptr, out_end - out_ptr,
1440                                               "%d", current->gid);
1441                                 if (rc > out_end - out_ptr)
1442                                         goto out;
1443                                 out_ptr += rc;
1444                                 break;
1445                         /* signal that caused the coredump */
1446                         case 's':
1447                                 rc = snprintf(out_ptr, out_end - out_ptr,
1448                                               "%ld", signr);
1449                                 if (rc > out_end - out_ptr)
1450                                         goto out;
1451                                 out_ptr += rc;
1452                                 break;
1453                         /* UNIX time of coredump */
1454                         case 't': {
1455                                 struct timeval tv;
1456                                 do_gettimeofday(&tv);
1457                                 rc = snprintf(out_ptr, out_end - out_ptr,
1458                                               "%lu", tv.tv_sec);
1459                                 if (rc > out_end - out_ptr)
1460                                         goto out;
1461                                 out_ptr += rc;
1462                                 break;
1463                         }
1464                         /* hostname */
1465                         case 'h':
1466                                 down_read(&uts_sem);
1467                                 rc = snprintf(out_ptr, out_end - out_ptr,
1468                                               "%s", utsname()->nodename);
1469                                 up_read(&uts_sem);
1470                                 if (rc > out_end - out_ptr)
1471                                         goto out;
1472                                 out_ptr += rc;
1473                                 break;
1474                         /* executable */
1475                         case 'e':
1476                                 rc = snprintf(out_ptr, out_end - out_ptr,
1477                                               "%s", current->comm);
1478                                 if (rc > out_end - out_ptr)
1479                                         goto out;
1480                                 out_ptr += rc;
1481                                 break;
1482                         /* core limit size */
1483                         case 'c':
1484                                 rc = snprintf(out_ptr, out_end - out_ptr,
1485                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1486                                 if (rc > out_end - out_ptr)
1487                                         goto out;
1488                                 out_ptr += rc;
1489                                 break;
1490                         default:
1491                                 break;
1492                         }
1493                         ++pat_ptr;
1494                 }
1495         }
1496         /* Backward compatibility with core_uses_pid:
1497          *
1498          * If core_pattern does not include a %p (as is the default)
1499          * and core_uses_pid is set, then .%pid will be appended to
1500          * the filename. Do not do this for piped commands. */
1501         if (!ispipe && !pid_in_pattern
1502             && (core_uses_pid || nr_threads)) {
1503                 rc = snprintf(out_ptr, out_end - out_ptr,
1504                               ".%d", task_tgid_vnr(current));
1505                 if (rc > out_end - out_ptr)
1506                         goto out;
1507                 out_ptr += rc;
1508         }
1509 out:
1510         *out_ptr = 0;
1511         return ispipe;
1512 }
1513
1514 static int zap_process(struct task_struct *start)
1515 {
1516         struct task_struct *t;
1517         int nr = 0;
1518
1519         start->signal->flags = SIGNAL_GROUP_EXIT;
1520         start->signal->group_stop_count = 0;
1521
1522         t = start;
1523         do {
1524                 if (t != current && t->mm) {
1525                         sigaddset(&t->pending.signal, SIGKILL);
1526                         signal_wake_up(t, 1);
1527                         nr++;
1528                 }
1529         } while_each_thread(start, t);
1530
1531         return nr;
1532 }
1533
1534 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1535                                 struct core_state *core_state, int exit_code)
1536 {
1537         struct task_struct *g, *p;
1538         unsigned long flags;
1539         int nr = -EAGAIN;
1540
1541         spin_lock_irq(&tsk->sighand->siglock);
1542         if (!signal_group_exit(tsk->signal)) {
1543                 mm->core_state = core_state;
1544                 tsk->signal->group_exit_code = exit_code;
1545                 nr = zap_process(tsk);
1546         }
1547         spin_unlock_irq(&tsk->sighand->siglock);
1548         if (unlikely(nr < 0))
1549                 return nr;
1550
1551         if (atomic_read(&mm->mm_users) == nr + 1)
1552                 goto done;
1553         /*
1554          * We should find and kill all tasks which use this mm, and we should
1555          * count them correctly into ->nr_threads. We don't take tasklist
1556          * lock, but this is safe wrt:
1557          *
1558          * fork:
1559          *      None of sub-threads can fork after zap_process(leader). All
1560          *      processes which were created before this point should be
1561          *      visible to zap_threads() because copy_process() adds the new
1562          *      process to the tail of init_task.tasks list, and lock/unlock
1563          *      of ->siglock provides a memory barrier.
1564          *
1565          * do_exit:
1566          *      The caller holds mm->mmap_sem. This means that the task which
1567          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1568          *      its ->mm.
1569          *
1570          * de_thread:
1571          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1572          *      we must see either old or new leader, this does not matter.
1573          *      However, it can change p->sighand, so lock_task_sighand(p)
1574          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1575          *      it can't fail.
1576          *
1577          *      Note also that "g" can be the old leader with ->mm == NULL
1578          *      and already unhashed and thus removed from ->thread_group.
1579          *      This is OK, __unhash_process()->list_del_rcu() does not
1580          *      clear the ->next pointer, we will find the new leader via
1581          *      next_thread().
1582          */
1583         rcu_read_lock();
1584         for_each_process(g) {
1585                 if (g == tsk->group_leader)
1586                         continue;
1587                 if (g->flags & PF_KTHREAD)
1588                         continue;
1589                 p = g;
1590                 do {
1591                         if (p->mm) {
1592                                 if (unlikely(p->mm == mm)) {
1593                                         lock_task_sighand(p, &flags);
1594                                         nr += zap_process(p);
1595                                         unlock_task_sighand(p, &flags);
1596                                 }
1597                                 break;
1598                         }
1599                 } while_each_thread(g, p);
1600         }
1601         rcu_read_unlock();
1602 done:
1603         atomic_set(&core_state->nr_threads, nr);
1604         return nr;
1605 }
1606
1607 static int coredump_wait(int exit_code, struct core_state *core_state)
1608 {
1609         struct task_struct *tsk = current;
1610         struct mm_struct *mm = tsk->mm;
1611         struct completion *vfork_done;
1612         int core_waiters;
1613
1614         init_completion(&core_state->startup);
1615         core_state->dumper.task = tsk;
1616         core_state->dumper.next = NULL;
1617         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1618         up_write(&mm->mmap_sem);
1619
1620         if (unlikely(core_waiters < 0))
1621                 goto fail;
1622
1623         /*
1624          * Make sure nobody is waiting for us to release the VM,
1625          * otherwise we can deadlock when we wait on each other
1626          */
1627         vfork_done = tsk->vfork_done;
1628         if (vfork_done) {
1629                 tsk->vfork_done = NULL;
1630                 complete(vfork_done);
1631         }
1632
1633         if (core_waiters)
1634                 wait_for_completion(&core_state->startup);
1635 fail:
1636         return core_waiters;
1637 }
1638
1639 static void coredump_finish(struct mm_struct *mm)
1640 {
1641         struct core_thread *curr, *next;
1642         struct task_struct *task;
1643
1644         next = mm->core_state->dumper.next;
1645         while ((curr = next) != NULL) {
1646                 next = curr->next;
1647                 task = curr->task;
1648                 /*
1649                  * see exit_mm(), curr->task must not see
1650                  * ->task == NULL before we read ->next.
1651                  */
1652                 smp_mb();
1653                 curr->task = NULL;
1654                 wake_up_process(task);
1655         }
1656
1657         mm->core_state = NULL;
1658 }
1659
1660 /*
1661  * set_dumpable converts traditional three-value dumpable to two flags and
1662  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1663  * these bits are not changed atomically.  So get_dumpable can observe the
1664  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1665  * return either old dumpable or new one by paying attention to the order of
1666  * modifying the bits.
1667  *
1668  * dumpable |   mm->flags (binary)
1669  * old  new | initial interim  final
1670  * ---------+-----------------------
1671  *  0    1  |   00      01      01
1672  *  0    2  |   00      10(*)   11
1673  *  1    0  |   01      00      00
1674  *  1    2  |   01      11      11
1675  *  2    0  |   11      10(*)   00
1676  *  2    1  |   11      11      01
1677  *
1678  * (*) get_dumpable regards interim value of 10 as 11.
1679  */
1680 void set_dumpable(struct mm_struct *mm, int value)
1681 {
1682         switch (value) {
1683         case 0:
1684                 clear_bit(MMF_DUMPABLE, &mm->flags);
1685                 smp_wmb();
1686                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1687                 break;
1688         case 1:
1689                 set_bit(MMF_DUMPABLE, &mm->flags);
1690                 smp_wmb();
1691                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1692                 break;
1693         case 2:
1694                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1695                 smp_wmb();
1696                 set_bit(MMF_DUMPABLE, &mm->flags);
1697                 break;
1698         }
1699 }
1700
1701 int get_dumpable(struct mm_struct *mm)
1702 {
1703         int ret;
1704
1705         ret = mm->flags & 0x3;
1706         return (ret >= 2) ? 2 : ret;
1707 }
1708
1709 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1710 {
1711         struct core_state core_state;
1712         char corename[CORENAME_MAX_SIZE + 1];
1713         struct mm_struct *mm = current->mm;
1714         struct linux_binfmt * binfmt;
1715         struct inode * inode;
1716         struct file * file;
1717         int retval = 0;
1718         int fsuid = current->fsuid;
1719         int flag = 0;
1720         int ispipe = 0;
1721         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1722         char **helper_argv = NULL;
1723         int helper_argc = 0;
1724         char *delimit;
1725
1726         audit_core_dumps(signr);
1727
1728         binfmt = current->binfmt;
1729         if (!binfmt || !binfmt->core_dump)
1730                 goto fail;
1731         down_write(&mm->mmap_sem);
1732         /*
1733          * If another thread got here first, or we are not dumpable, bail out.
1734          */
1735         if (mm->core_state || !get_dumpable(mm)) {
1736                 up_write(&mm->mmap_sem);
1737                 goto fail;
1738         }
1739
1740         /*
1741          *      We cannot trust fsuid as being the "true" uid of the
1742          *      process nor do we know its entire history. We only know it
1743          *      was tainted so we dump it as root in mode 2.
1744          */
1745         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1746                 flag = O_EXCL;          /* Stop rewrite attacks */
1747                 current->fsuid = 0;     /* Dump root private */
1748         }
1749
1750         retval = coredump_wait(exit_code, &core_state);
1751         if (retval < 0)
1752                 goto fail;
1753
1754         /*
1755          * Clear any false indication of pending signals that might
1756          * be seen by the filesystem code called to write the core file.
1757          */
1758         clear_thread_flag(TIF_SIGPENDING);
1759
1760         /*
1761          * lock_kernel() because format_corename() is controlled by sysctl, which
1762          * uses lock_kernel()
1763          */
1764         lock_kernel();
1765         ispipe = format_corename(corename, retval, signr);
1766         unlock_kernel();
1767         /*
1768          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1769          * to a pipe.  Since we're not writing directly to the filesystem
1770          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1771          * created unless the pipe reader choses to write out the core file
1772          * at which point file size limits and permissions will be imposed
1773          * as it does with any other process
1774          */
1775         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1776                 goto fail_unlock;
1777
1778         if (ispipe) {
1779                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1780                 /* Terminate the string before the first option */
1781                 delimit = strchr(corename, ' ');
1782                 if (delimit)
1783                         *delimit = '\0';
1784                 delimit = strrchr(helper_argv[0], '/');
1785                 if (delimit)
1786                         delimit++;
1787                 else
1788                         delimit = helper_argv[0];
1789                 if (!strcmp(delimit, current->comm)) {
1790                         printk(KERN_NOTICE "Recursive core dump detected, "
1791                                         "aborting\n");
1792                         goto fail_unlock;
1793                 }
1794
1795                 core_limit = RLIM_INFINITY;
1796
1797                 /* SIGPIPE can happen, but it's just never processed */
1798                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1799                                 &file)) {
1800                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1801                                corename);
1802                         goto fail_unlock;
1803                 }
1804         } else
1805                 file = filp_open(corename,
1806                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1807                                  0600);
1808         if (IS_ERR(file))
1809                 goto fail_unlock;
1810         inode = file->f_path.dentry->d_inode;
1811         if (inode->i_nlink > 1)
1812                 goto close_fail;        /* multiple links - don't dump */
1813         if (!ispipe && d_unhashed(file->f_path.dentry))
1814                 goto close_fail;
1815
1816         /* AK: actually i see no reason to not allow this for named pipes etc.,
1817            but keep the previous behaviour for now. */
1818         if (!ispipe && !S_ISREG(inode->i_mode))
1819                 goto close_fail;
1820         /*
1821          * Dont allow local users get cute and trick others to coredump
1822          * into their pre-created files:
1823          */
1824         if (inode->i_uid != current->fsuid)
1825                 goto close_fail;
1826         if (!file->f_op)
1827                 goto close_fail;
1828         if (!file->f_op->write)
1829                 goto close_fail;
1830         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1831                 goto close_fail;
1832
1833         retval = binfmt->core_dump(signr, regs, file, core_limit);
1834
1835         if (retval)
1836                 current->signal->group_exit_code |= 0x80;
1837 close_fail:
1838         filp_close(file, NULL);
1839 fail_unlock:
1840         if (helper_argv)
1841                 argv_free(helper_argv);
1842
1843         current->fsuid = fsuid;
1844         coredump_finish(mm);
1845 fail:
1846         return retval;
1847 }