CRED: Fix SUID exec regression
[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 #include <linux/kmod.h>
54 #include <linux/fsnotify.h>
55
56 #include <asm/uaccess.h>
57 #include <asm/mmu_context.h>
58 #include <asm/tlb.h>
59 #include "internal.h"
60
61 int core_uses_pid;
62 char core_pattern[CORENAME_MAX_SIZE] = "core";
63 int suid_dumpable = 0;
64
65 /* The maximal length of core_pattern is also specified in sysctl.c */
66
67 static LIST_HEAD(formats);
68 static DEFINE_RWLOCK(binfmt_lock);
69
70 int register_binfmt(struct linux_binfmt * fmt)
71 {
72         if (!fmt)
73                 return -EINVAL;
74         write_lock(&binfmt_lock);
75         list_add(&fmt->lh, &formats);
76         write_unlock(&binfmt_lock);
77         return 0;       
78 }
79
80 EXPORT_SYMBOL(register_binfmt);
81
82 void unregister_binfmt(struct linux_binfmt * fmt)
83 {
84         write_lock(&binfmt_lock);
85         list_del(&fmt->lh);
86         write_unlock(&binfmt_lock);
87 }
88
89 EXPORT_SYMBOL(unregister_binfmt);
90
91 static inline void put_binfmt(struct linux_binfmt * fmt)
92 {
93         module_put(fmt->module);
94 }
95
96 /*
97  * Note that a shared library must be both readable and executable due to
98  * security reasons.
99  *
100  * Also note that we take the address to load from from the file itself.
101  */
102 SYSCALL_DEFINE1(uselib, const char __user *, library)
103 {
104         struct file *file;
105         struct nameidata nd;
106         char *tmp = getname(library);
107         int error = PTR_ERR(tmp);
108
109         if (!IS_ERR(tmp)) {
110                 error = path_lookup_open(AT_FDCWD, tmp,
111                                          LOOKUP_FOLLOW, &nd,
112                                          FMODE_READ|FMODE_EXEC);
113                 putname(tmp);
114         }
115         if (error)
116                 goto out;
117
118         error = -EINVAL;
119         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
120                 goto exit;
121
122         error = -EACCES;
123         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
124                 goto exit;
125
126         error = inode_permission(nd.path.dentry->d_inode,
127                                  MAY_READ | MAY_EXEC | MAY_OPEN);
128         if (error)
129                 goto exit;
130
131         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
132         error = PTR_ERR(file);
133         if (IS_ERR(file))
134                 goto out;
135
136         fsnotify_open(file->f_path.dentry);
137
138         error = -ENOEXEC;
139         if(file->f_op) {
140                 struct linux_binfmt * fmt;
141
142                 read_lock(&binfmt_lock);
143                 list_for_each_entry(fmt, &formats, lh) {
144                         if (!fmt->load_shlib)
145                                 continue;
146                         if (!try_module_get(fmt->module))
147                                 continue;
148                         read_unlock(&binfmt_lock);
149                         error = fmt->load_shlib(file);
150                         read_lock(&binfmt_lock);
151                         put_binfmt(fmt);
152                         if (error != -ENOEXEC)
153                                 break;
154                 }
155                 read_unlock(&binfmt_lock);
156         }
157         fput(file);
158 out:
159         return error;
160 exit:
161         release_open_intent(&nd);
162         path_put(&nd.path);
163         goto out;
164 }
165
166 #ifdef CONFIG_MMU
167
168 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
169                 int write)
170 {
171         struct page *page;
172         int ret;
173
174 #ifdef CONFIG_STACK_GROWSUP
175         if (write) {
176                 ret = expand_stack_downwards(bprm->vma, pos);
177                 if (ret < 0)
178                         return NULL;
179         }
180 #endif
181         ret = get_user_pages(current, bprm->mm, pos,
182                         1, write, 1, &page, NULL);
183         if (ret <= 0)
184                 return NULL;
185
186         if (write) {
187                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
188                 struct rlimit *rlim;
189
190                 /*
191                  * We've historically supported up to 32 pages (ARG_MAX)
192                  * of argument strings even with small stacks
193                  */
194                 if (size <= ARG_MAX)
195                         return page;
196
197                 /*
198                  * Limit to 1/4-th the stack size for the argv+env strings.
199                  * This ensures that:
200                  *  - the remaining binfmt code will not run out of stack space,
201                  *  - the program will have a reasonable amount of stack left
202                  *    to work from.
203                  */
204                 rlim = current->signal->rlim;
205                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
206                         put_page(page);
207                         return NULL;
208                 }
209         }
210
211         return page;
212 }
213
214 static void put_arg_page(struct page *page)
215 {
216         put_page(page);
217 }
218
219 static void free_arg_page(struct linux_binprm *bprm, int i)
220 {
221 }
222
223 static void free_arg_pages(struct linux_binprm *bprm)
224 {
225 }
226
227 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
228                 struct page *page)
229 {
230         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
231 }
232
233 static int __bprm_mm_init(struct linux_binprm *bprm)
234 {
235         int err;
236         struct vm_area_struct *vma = NULL;
237         struct mm_struct *mm = bprm->mm;
238
239         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
240         if (!vma)
241                 return -ENOMEM;
242
243         down_write(&mm->mmap_sem);
244         vma->vm_mm = mm;
245
246         /*
247          * Place the stack at the largest stack address the architecture
248          * supports. Later, we'll move this to an appropriate place. We don't
249          * use STACK_TOP because that can depend on attributes which aren't
250          * configured yet.
251          */
252         vma->vm_end = STACK_TOP_MAX;
253         vma->vm_start = vma->vm_end - PAGE_SIZE;
254         vma->vm_flags = VM_STACK_FLAGS;
255         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
256         err = insert_vm_struct(mm, vma);
257         if (err)
258                 goto err;
259
260         mm->stack_vm = mm->total_vm = 1;
261         up_write(&mm->mmap_sem);
262         bprm->p = vma->vm_end - sizeof(void *);
263         return 0;
264 err:
265         up_write(&mm->mmap_sem);
266         bprm->vma = NULL;
267         kmem_cache_free(vm_area_cachep, vma);
268         return err;
269 }
270
271 static bool valid_arg_len(struct linux_binprm *bprm, long len)
272 {
273         return len <= MAX_ARG_STRLEN;
274 }
275
276 #else
277
278 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
279                 int write)
280 {
281         struct page *page;
282
283         page = bprm->page[pos / PAGE_SIZE];
284         if (!page && write) {
285                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
286                 if (!page)
287                         return NULL;
288                 bprm->page[pos / PAGE_SIZE] = page;
289         }
290
291         return page;
292 }
293
294 static void put_arg_page(struct page *page)
295 {
296 }
297
298 static void free_arg_page(struct linux_binprm *bprm, int i)
299 {
300         if (bprm->page[i]) {
301                 __free_page(bprm->page[i]);
302                 bprm->page[i] = NULL;
303         }
304 }
305
306 static void free_arg_pages(struct linux_binprm *bprm)
307 {
308         int i;
309
310         for (i = 0; i < MAX_ARG_PAGES; i++)
311                 free_arg_page(bprm, i);
312 }
313
314 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
315                 struct page *page)
316 {
317 }
318
319 static int __bprm_mm_init(struct linux_binprm *bprm)
320 {
321         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
322         return 0;
323 }
324
325 static bool valid_arg_len(struct linux_binprm *bprm, long len)
326 {
327         return len <= bprm->p;
328 }
329
330 #endif /* CONFIG_MMU */
331
332 /*
333  * Create a new mm_struct and populate it with a temporary stack
334  * vm_area_struct.  We don't have enough context at this point to set the stack
335  * flags, permissions, and offset, so we use temporary values.  We'll update
336  * them later in setup_arg_pages().
337  */
338 int bprm_mm_init(struct linux_binprm *bprm)
339 {
340         int err;
341         struct mm_struct *mm = NULL;
342
343         bprm->mm = mm = mm_alloc();
344         err = -ENOMEM;
345         if (!mm)
346                 goto err;
347
348         err = init_new_context(current, mm);
349         if (err)
350                 goto err;
351
352         err = __bprm_mm_init(bprm);
353         if (err)
354                 goto err;
355
356         return 0;
357
358 err:
359         if (mm) {
360                 bprm->mm = NULL;
361                 mmdrop(mm);
362         }
363
364         return err;
365 }
366
367 /*
368  * count() counts the number of strings in array ARGV.
369  */
370 static int count(char __user * __user * argv, int max)
371 {
372         int i = 0;
373
374         if (argv != NULL) {
375                 for (;;) {
376                         char __user * p;
377
378                         if (get_user(p, argv))
379                                 return -EFAULT;
380                         if (!p)
381                                 break;
382                         argv++;
383                         if (i++ >= max)
384                                 return -E2BIG;
385                         cond_resched();
386                 }
387         }
388         return i;
389 }
390
391 /*
392  * 'copy_strings()' copies argument/environment strings from the old
393  * processes's memory to the new process's stack.  The call to get_user_pages()
394  * ensures the destination page is created and not swapped out.
395  */
396 static int copy_strings(int argc, char __user * __user * argv,
397                         struct linux_binprm *bprm)
398 {
399         struct page *kmapped_page = NULL;
400         char *kaddr = NULL;
401         unsigned long kpos = 0;
402         int ret;
403
404         while (argc-- > 0) {
405                 char __user *str;
406                 int len;
407                 unsigned long pos;
408
409                 if (get_user(str, argv+argc) ||
410                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
411                         ret = -EFAULT;
412                         goto out;
413                 }
414
415                 if (!valid_arg_len(bprm, len)) {
416                         ret = -E2BIG;
417                         goto out;
418                 }
419
420                 /* We're going to work our way backwords. */
421                 pos = bprm->p;
422                 str += len;
423                 bprm->p -= len;
424
425                 while (len > 0) {
426                         int offset, bytes_to_copy;
427
428                         offset = pos % PAGE_SIZE;
429                         if (offset == 0)
430                                 offset = PAGE_SIZE;
431
432                         bytes_to_copy = offset;
433                         if (bytes_to_copy > len)
434                                 bytes_to_copy = len;
435
436                         offset -= bytes_to_copy;
437                         pos -= bytes_to_copy;
438                         str -= bytes_to_copy;
439                         len -= bytes_to_copy;
440
441                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
442                                 struct page *page;
443
444                                 page = get_arg_page(bprm, pos, 1);
445                                 if (!page) {
446                                         ret = -E2BIG;
447                                         goto out;
448                                 }
449
450                                 if (kmapped_page) {
451                                         flush_kernel_dcache_page(kmapped_page);
452                                         kunmap(kmapped_page);
453                                         put_arg_page(kmapped_page);
454                                 }
455                                 kmapped_page = page;
456                                 kaddr = kmap(kmapped_page);
457                                 kpos = pos & PAGE_MASK;
458                                 flush_arg_page(bprm, kpos, kmapped_page);
459                         }
460                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
461                                 ret = -EFAULT;
462                                 goto out;
463                         }
464                 }
465         }
466         ret = 0;
467 out:
468         if (kmapped_page) {
469                 flush_kernel_dcache_page(kmapped_page);
470                 kunmap(kmapped_page);
471                 put_arg_page(kmapped_page);
472         }
473         return ret;
474 }
475
476 /*
477  * Like copy_strings, but get argv and its values from kernel memory.
478  */
479 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
480 {
481         int r;
482         mm_segment_t oldfs = get_fs();
483         set_fs(KERNEL_DS);
484         r = copy_strings(argc, (char __user * __user *)argv, bprm);
485         set_fs(oldfs);
486         return r;
487 }
488 EXPORT_SYMBOL(copy_strings_kernel);
489
490 #ifdef CONFIG_MMU
491
492 /*
493  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
494  * the binfmt code determines where the new stack should reside, we shift it to
495  * its final location.  The process proceeds as follows:
496  *
497  * 1) Use shift to calculate the new vma endpoints.
498  * 2) Extend vma to cover both the old and new ranges.  This ensures the
499  *    arguments passed to subsequent functions are consistent.
500  * 3) Move vma's page tables to the new range.
501  * 4) Free up any cleared pgd range.
502  * 5) Shrink the vma to cover only the new range.
503  */
504 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
505 {
506         struct mm_struct *mm = vma->vm_mm;
507         unsigned long old_start = vma->vm_start;
508         unsigned long old_end = vma->vm_end;
509         unsigned long length = old_end - old_start;
510         unsigned long new_start = old_start - shift;
511         unsigned long new_end = old_end - shift;
512         struct mmu_gather *tlb;
513
514         BUG_ON(new_start > new_end);
515
516         /*
517          * ensure there are no vmas between where we want to go
518          * and where we are
519          */
520         if (vma != find_vma(mm, new_start))
521                 return -EFAULT;
522
523         /*
524          * cover the whole range: [new_start, old_end)
525          */
526         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
527
528         /*
529          * move the page tables downwards, on failure we rely on
530          * process cleanup to remove whatever mess we made.
531          */
532         if (length != move_page_tables(vma, old_start,
533                                        vma, new_start, length))
534                 return -ENOMEM;
535
536         lru_add_drain();
537         tlb = tlb_gather_mmu(mm, 0);
538         if (new_end > old_start) {
539                 /*
540                  * when the old and new regions overlap clear from new_end.
541                  */
542                 free_pgd_range(tlb, new_end, old_end, new_end,
543                         vma->vm_next ? vma->vm_next->vm_start : 0);
544         } else {
545                 /*
546                  * otherwise, clean from old_start; this is done to not touch
547                  * the address space in [new_end, old_start) some architectures
548                  * have constraints on va-space that make this illegal (IA64) -
549                  * for the others its just a little faster.
550                  */
551                 free_pgd_range(tlb, old_start, old_end, new_end,
552                         vma->vm_next ? vma->vm_next->vm_start : 0);
553         }
554         tlb_finish_mmu(tlb, new_end, old_end);
555
556         /*
557          * shrink the vma to just the new range.
558          */
559         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
560
561         return 0;
562 }
563
564 #define EXTRA_STACK_VM_PAGES    20      /* random */
565
566 /*
567  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
568  * the stack is optionally relocated, and some extra space is added.
569  */
570 int setup_arg_pages(struct linux_binprm *bprm,
571                     unsigned long stack_top,
572                     int executable_stack)
573 {
574         unsigned long ret;
575         unsigned long stack_shift;
576         struct mm_struct *mm = current->mm;
577         struct vm_area_struct *vma = bprm->vma;
578         struct vm_area_struct *prev = NULL;
579         unsigned long vm_flags;
580         unsigned long stack_base;
581
582 #ifdef CONFIG_STACK_GROWSUP
583         /* Limit stack size to 1GB */
584         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
585         if (stack_base > (1 << 30))
586                 stack_base = 1 << 30;
587
588         /* Make sure we didn't let the argument array grow too large. */
589         if (vma->vm_end - vma->vm_start > stack_base)
590                 return -ENOMEM;
591
592         stack_base = PAGE_ALIGN(stack_top - stack_base);
593
594         stack_shift = vma->vm_start - stack_base;
595         mm->arg_start = bprm->p - stack_shift;
596         bprm->p = vma->vm_end - stack_shift;
597 #else
598         stack_top = arch_align_stack(stack_top);
599         stack_top = PAGE_ALIGN(stack_top);
600         stack_shift = vma->vm_end - stack_top;
601
602         bprm->p -= stack_shift;
603         mm->arg_start = bprm->p;
604 #endif
605
606         if (bprm->loader)
607                 bprm->loader -= stack_shift;
608         bprm->exec -= stack_shift;
609
610         down_write(&mm->mmap_sem);
611         vm_flags = VM_STACK_FLAGS;
612
613         /*
614          * Adjust stack execute permissions; explicitly enable for
615          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
616          * (arch default) otherwise.
617          */
618         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
619                 vm_flags |= VM_EXEC;
620         else if (executable_stack == EXSTACK_DISABLE_X)
621                 vm_flags &= ~VM_EXEC;
622         vm_flags |= mm->def_flags;
623
624         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
625                         vm_flags);
626         if (ret)
627                 goto out_unlock;
628         BUG_ON(prev != vma);
629
630         /* Move stack pages down in memory. */
631         if (stack_shift) {
632                 ret = shift_arg_pages(vma, stack_shift);
633                 if (ret) {
634                         up_write(&mm->mmap_sem);
635                         return ret;
636                 }
637         }
638
639 #ifdef CONFIG_STACK_GROWSUP
640         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
641 #else
642         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
643 #endif
644         ret = expand_stack(vma, stack_base);
645         if (ret)
646                 ret = -EFAULT;
647
648 out_unlock:
649         up_write(&mm->mmap_sem);
650         return 0;
651 }
652 EXPORT_SYMBOL(setup_arg_pages);
653
654 #endif /* CONFIG_MMU */
655
656 struct file *open_exec(const char *name)
657 {
658         struct nameidata nd;
659         struct file *file;
660         int err;
661
662         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
663                                 FMODE_READ|FMODE_EXEC);
664         if (err)
665                 goto out;
666
667         err = -EACCES;
668         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
669                 goto out_path_put;
670
671         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
672                 goto out_path_put;
673
674         err = inode_permission(nd.path.dentry->d_inode, MAY_EXEC | MAY_OPEN);
675         if (err)
676                 goto out_path_put;
677
678         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
679         if (IS_ERR(file))
680                 return file;
681
682         fsnotify_open(file->f_path.dentry);
683
684         err = deny_write_access(file);
685         if (err) {
686                 fput(file);
687                 goto out;
688         }
689
690         return file;
691
692  out_path_put:
693         release_open_intent(&nd);
694         path_put(&nd.path);
695  out:
696         return ERR_PTR(err);
697 }
698 EXPORT_SYMBOL(open_exec);
699
700 int kernel_read(struct file *file, unsigned long offset,
701         char *addr, unsigned long count)
702 {
703         mm_segment_t old_fs;
704         loff_t pos = offset;
705         int result;
706
707         old_fs = get_fs();
708         set_fs(get_ds());
709         /* The cast to a user pointer is valid due to the set_fs() */
710         result = vfs_read(file, (void __user *)addr, count, &pos);
711         set_fs(old_fs);
712         return result;
713 }
714
715 EXPORT_SYMBOL(kernel_read);
716
717 static int exec_mmap(struct mm_struct *mm)
718 {
719         struct task_struct *tsk;
720         struct mm_struct * old_mm, *active_mm;
721
722         /* Notify parent that we're no longer interested in the old VM */
723         tsk = current;
724         old_mm = current->mm;
725         mm_release(tsk, old_mm);
726
727         if (old_mm) {
728                 /*
729                  * Make sure that if there is a core dump in progress
730                  * for the old mm, we get out and die instead of going
731                  * through with the exec.  We must hold mmap_sem around
732                  * checking core_state and changing tsk->mm.
733                  */
734                 down_read(&old_mm->mmap_sem);
735                 if (unlikely(old_mm->core_state)) {
736                         up_read(&old_mm->mmap_sem);
737                         return -EINTR;
738                 }
739         }
740         task_lock(tsk);
741         active_mm = tsk->active_mm;
742         tsk->mm = mm;
743         tsk->active_mm = mm;
744         activate_mm(active_mm, mm);
745         task_unlock(tsk);
746         arch_pick_mmap_layout(mm);
747         if (old_mm) {
748                 up_read(&old_mm->mmap_sem);
749                 BUG_ON(active_mm != old_mm);
750                 mm_update_next_owner(old_mm);
751                 mmput(old_mm);
752                 return 0;
753         }
754         mmdrop(active_mm);
755         return 0;
756 }
757
758 /*
759  * This function makes sure the current process has its own signal table,
760  * so that flush_signal_handlers can later reset the handlers without
761  * disturbing other processes.  (Other processes might share the signal
762  * table via the CLONE_SIGHAND option to clone().)
763  */
764 static int de_thread(struct task_struct *tsk)
765 {
766         struct signal_struct *sig = tsk->signal;
767         struct sighand_struct *oldsighand = tsk->sighand;
768         spinlock_t *lock = &oldsighand->siglock;
769         int count;
770
771         if (thread_group_empty(tsk))
772                 goto no_thread_group;
773
774         /*
775          * Kill all other threads in the thread group.
776          */
777         spin_lock_irq(lock);
778         if (signal_group_exit(sig)) {
779                 /*
780                  * Another group action in progress, just
781                  * return so that the signal is processed.
782                  */
783                 spin_unlock_irq(lock);
784                 return -EAGAIN;
785         }
786         sig->group_exit_task = tsk;
787         zap_other_threads(tsk);
788
789         /* Account for the thread group leader hanging around: */
790         count = thread_group_leader(tsk) ? 1 : 2;
791         sig->notify_count = count;
792         while (atomic_read(&sig->count) > count) {
793                 __set_current_state(TASK_UNINTERRUPTIBLE);
794                 spin_unlock_irq(lock);
795                 schedule();
796                 spin_lock_irq(lock);
797         }
798         spin_unlock_irq(lock);
799
800         /*
801          * At this point all other threads have exited, all we have to
802          * do is to wait for the thread group leader to become inactive,
803          * and to assume its PID:
804          */
805         if (!thread_group_leader(tsk)) {
806                 struct task_struct *leader = tsk->group_leader;
807
808                 sig->notify_count = -1; /* for exit_notify() */
809                 for (;;) {
810                         write_lock_irq(&tasklist_lock);
811                         if (likely(leader->exit_state))
812                                 break;
813                         __set_current_state(TASK_UNINTERRUPTIBLE);
814                         write_unlock_irq(&tasklist_lock);
815                         schedule();
816                 }
817
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                 write_unlock_irq(&tasklist_lock);
859
860                 release_task(leader);
861         }
862
863         sig->group_exit_task = NULL;
864         sig->notify_count = 0;
865
866 no_thread_group:
867         exit_itimers(sig);
868         flush_itimer_signals();
869
870         if (atomic_read(&oldsighand->count) != 1) {
871                 struct sighand_struct *newsighand;
872                 /*
873                  * This ->sighand is shared with the CLONE_SIGHAND
874                  * but not CLONE_THREAD task, switch to the new one.
875                  */
876                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
877                 if (!newsighand)
878                         return -ENOMEM;
879
880                 atomic_set(&newsighand->count, 1);
881                 memcpy(newsighand->action, oldsighand->action,
882                        sizeof(newsighand->action));
883
884                 write_lock_irq(&tasklist_lock);
885                 spin_lock(&oldsighand->siglock);
886                 rcu_assign_pointer(tsk->sighand, newsighand);
887                 spin_unlock(&oldsighand->siglock);
888                 write_unlock_irq(&tasklist_lock);
889
890                 __cleanup_sighand(oldsighand);
891         }
892
893         BUG_ON(!thread_group_leader(tsk));
894         return 0;
895 }
896
897 /*
898  * These functions flushes out all traces of the currently running executable
899  * so that a new one can be started
900  */
901 static void flush_old_files(struct files_struct * files)
902 {
903         long j = -1;
904         struct fdtable *fdt;
905
906         spin_lock(&files->file_lock);
907         for (;;) {
908                 unsigned long set, i;
909
910                 j++;
911                 i = j * __NFDBITS;
912                 fdt = files_fdtable(files);
913                 if (i >= fdt->max_fds)
914                         break;
915                 set = fdt->close_on_exec->fds_bits[j];
916                 if (!set)
917                         continue;
918                 fdt->close_on_exec->fds_bits[j] = 0;
919                 spin_unlock(&files->file_lock);
920                 for ( ; set ; i++,set >>= 1) {
921                         if (set & 1) {
922                                 sys_close(i);
923                         }
924                 }
925                 spin_lock(&files->file_lock);
926
927         }
928         spin_unlock(&files->file_lock);
929 }
930
931 char *get_task_comm(char *buf, struct task_struct *tsk)
932 {
933         /* buf must be at least sizeof(tsk->comm) in size */
934         task_lock(tsk);
935         strncpy(buf, tsk->comm, sizeof(tsk->comm));
936         task_unlock(tsk);
937         return buf;
938 }
939
940 void set_task_comm(struct task_struct *tsk, char *buf)
941 {
942         task_lock(tsk);
943         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
944         task_unlock(tsk);
945 }
946
947 int flush_old_exec(struct linux_binprm * bprm)
948 {
949         char * name;
950         int i, ch, retval;
951         char tcomm[sizeof(current->comm)];
952
953         /*
954          * Make sure we have a private signal table and that
955          * we are unassociated from the previous thread group.
956          */
957         retval = de_thread(current);
958         if (retval)
959                 goto out;
960
961         set_mm_exe_file(bprm->mm, bprm->file);
962
963         /*
964          * Release all of the old mmap stuff
965          */
966         retval = exec_mmap(bprm->mm);
967         if (retval)
968                 goto out;
969
970         bprm->mm = NULL;                /* We're using it now */
971
972         /* This is the point of no return */
973         current->sas_ss_sp = current->sas_ss_size = 0;
974
975         if (current_euid() == current_uid() && current_egid() == current_gid())
976                 set_dumpable(current->mm, 1);
977         else
978                 set_dumpable(current->mm, suid_dumpable);
979
980         name = bprm->filename;
981
982         /* Copies the binary name from after last slash */
983         for (i=0; (ch = *(name++)) != '\0';) {
984                 if (ch == '/')
985                         i = 0; /* overwrite what we wrote */
986                 else
987                         if (i < (sizeof(tcomm) - 1))
988                                 tcomm[i++] = ch;
989         }
990         tcomm[i] = '\0';
991         set_task_comm(current, tcomm);
992
993         current->flags &= ~PF_RANDOMIZE;
994         flush_thread();
995
996         /* Set the new mm task size. We have to do that late because it may
997          * depend on TIF_32BIT which is only updated in flush_thread() on
998          * some architectures like powerpc
999          */
1000         current->mm->task_size = TASK_SIZE;
1001
1002         /* install the new credentials */
1003         if (bprm->cred->uid != current_euid() ||
1004             bprm->cred->gid != current_egid()) {
1005                 current->pdeath_signal = 0;
1006         } else if (file_permission(bprm->file, MAY_READ) ||
1007                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1008                 set_dumpable(current->mm, suid_dumpable);
1009         }
1010
1011         current->personality &= ~bprm->per_clear;
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  * install the new credentials for this executable
1031  */
1032 void install_exec_creds(struct linux_binprm *bprm)
1033 {
1034         security_bprm_committing_creds(bprm);
1035
1036         commit_creds(bprm->cred);
1037         bprm->cred = NULL;
1038
1039         /* cred_exec_mutex must be held at least to this point to prevent
1040          * ptrace_attach() from altering our determination of the task's
1041          * credentials; any time after this it may be unlocked */
1042
1043         security_bprm_committed_creds(bprm);
1044 }
1045 EXPORT_SYMBOL(install_exec_creds);
1046
1047 /*
1048  * determine how safe it is to execute the proposed program
1049  * - the caller must hold current->cred_exec_mutex to protect against
1050  *   PTRACE_ATTACH
1051  */
1052 void check_unsafe_exec(struct linux_binprm *bprm, struct files_struct *files)
1053 {
1054         struct task_struct *p = current, *t;
1055         unsigned long flags;
1056         unsigned n_fs, n_files, n_sighand;
1057
1058         bprm->unsafe = tracehook_unsafe_exec(p);
1059
1060         n_fs = 1;
1061         n_files = 1;
1062         n_sighand = 1;
1063         lock_task_sighand(p, &flags);
1064         for (t = next_thread(p); t != p; t = next_thread(t)) {
1065                 if (t->fs == p->fs)
1066                         n_fs++;
1067                 if (t->files == files)
1068                         n_files++;
1069                 n_sighand++;
1070         }
1071
1072         if (atomic_read(&p->fs->count) > n_fs ||
1073             atomic_read(&p->files->count) > n_files ||
1074             atomic_read(&p->sighand->count) > n_sighand)
1075                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1076
1077         unlock_task_sighand(p, &flags);
1078 }
1079
1080 /* 
1081  * Fill the binprm structure from the inode. 
1082  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1083  *
1084  * This may be called multiple times for binary chains (scripts for example).
1085  */
1086 int prepare_binprm(struct linux_binprm *bprm)
1087 {
1088         umode_t mode;
1089         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1090         int retval;
1091
1092         mode = inode->i_mode;
1093         if (bprm->file->f_op == NULL)
1094                 return -EACCES;
1095
1096         /* clear any previous set[ug]id data from a previous binary */
1097         bprm->cred->euid = current_euid();
1098         bprm->cred->egid = current_egid();
1099
1100         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1101                 /* Set-uid? */
1102                 if (mode & S_ISUID) {
1103                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1104                         bprm->cred->euid = inode->i_uid;
1105                 }
1106
1107                 /* Set-gid? */
1108                 /*
1109                  * If setgid is set but no group execute bit then this
1110                  * is a candidate for mandatory locking, not a setgid
1111                  * executable.
1112                  */
1113                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1114                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1115                         bprm->cred->egid = inode->i_gid;
1116                 }
1117         }
1118
1119         /* fill in binprm security blob */
1120         retval = security_bprm_set_creds(bprm);
1121         if (retval)
1122                 return retval;
1123         bprm->cred_prepared = 1;
1124
1125         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1126         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1127 }
1128
1129 EXPORT_SYMBOL(prepare_binprm);
1130
1131 /*
1132  * Arguments are '\0' separated strings found at the location bprm->p
1133  * points to; chop off the first by relocating brpm->p to right after
1134  * the first '\0' encountered.
1135  */
1136 int remove_arg_zero(struct linux_binprm *bprm)
1137 {
1138         int ret = 0;
1139         unsigned long offset;
1140         char *kaddr;
1141         struct page *page;
1142
1143         if (!bprm->argc)
1144                 return 0;
1145
1146         do {
1147                 offset = bprm->p & ~PAGE_MASK;
1148                 page = get_arg_page(bprm, bprm->p, 0);
1149                 if (!page) {
1150                         ret = -EFAULT;
1151                         goto out;
1152                 }
1153                 kaddr = kmap_atomic(page, KM_USER0);
1154
1155                 for (; offset < PAGE_SIZE && kaddr[offset];
1156                                 offset++, bprm->p++)
1157                         ;
1158
1159                 kunmap_atomic(kaddr, KM_USER0);
1160                 put_arg_page(page);
1161
1162                 if (offset == PAGE_SIZE)
1163                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1164         } while (offset == PAGE_SIZE);
1165
1166         bprm->p++;
1167         bprm->argc--;
1168         ret = 0;
1169
1170 out:
1171         return ret;
1172 }
1173 EXPORT_SYMBOL(remove_arg_zero);
1174
1175 /*
1176  * cycle the list of binary formats handler, until one recognizes the image
1177  */
1178 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1179 {
1180         unsigned int depth = bprm->recursion_depth;
1181         int try,retval;
1182         struct linux_binfmt *fmt;
1183
1184         retval = security_bprm_check(bprm);
1185         if (retval)
1186                 return retval;
1187
1188         /* kernel module loader fixup */
1189         /* so we don't try to load run modprobe in kernel space. */
1190         set_fs(USER_DS);
1191
1192         retval = audit_bprm(bprm);
1193         if (retval)
1194                 return retval;
1195
1196         retval = -ENOENT;
1197         for (try=0; try<2; try++) {
1198                 read_lock(&binfmt_lock);
1199                 list_for_each_entry(fmt, &formats, lh) {
1200                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1201                         if (!fn)
1202                                 continue;
1203                         if (!try_module_get(fmt->module))
1204                                 continue;
1205                         read_unlock(&binfmt_lock);
1206                         retval = fn(bprm, regs);
1207                         /*
1208                          * Restore the depth counter to its starting value
1209                          * in this call, so we don't have to rely on every
1210                          * load_binary function to restore it on return.
1211                          */
1212                         bprm->recursion_depth = depth;
1213                         if (retval >= 0) {
1214                                 if (depth == 0)
1215                                         tracehook_report_exec(fmt, bprm, regs);
1216                                 put_binfmt(fmt);
1217                                 allow_write_access(bprm->file);
1218                                 if (bprm->file)
1219                                         fput(bprm->file);
1220                                 bprm->file = NULL;
1221                                 current->did_exec = 1;
1222                                 proc_exec_connector(current);
1223                                 return retval;
1224                         }
1225                         read_lock(&binfmt_lock);
1226                         put_binfmt(fmt);
1227                         if (retval != -ENOEXEC || bprm->mm == NULL)
1228                                 break;
1229                         if (!bprm->file) {
1230                                 read_unlock(&binfmt_lock);
1231                                 return retval;
1232                         }
1233                 }
1234                 read_unlock(&binfmt_lock);
1235                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1236                         break;
1237 #ifdef CONFIG_MODULES
1238                 } else {
1239 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1240                         if (printable(bprm->buf[0]) &&
1241                             printable(bprm->buf[1]) &&
1242                             printable(bprm->buf[2]) &&
1243                             printable(bprm->buf[3]))
1244                                 break; /* -ENOEXEC */
1245                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1246 #endif
1247                 }
1248         }
1249         return retval;
1250 }
1251
1252 EXPORT_SYMBOL(search_binary_handler);
1253
1254 void free_bprm(struct linux_binprm *bprm)
1255 {
1256         free_arg_pages(bprm);
1257         if (bprm->cred)
1258                 abort_creds(bprm->cred);
1259         kfree(bprm);
1260 }
1261
1262 /*
1263  * sys_execve() executes a new program.
1264  */
1265 int do_execve(char * filename,
1266         char __user *__user *argv,
1267         char __user *__user *envp,
1268         struct pt_regs * regs)
1269 {
1270         struct linux_binprm *bprm;
1271         struct file *file;
1272         struct files_struct *displaced;
1273         int retval;
1274
1275         retval = unshare_files(&displaced);
1276         if (retval)
1277                 goto out_ret;
1278
1279         retval = -ENOMEM;
1280         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1281         if (!bprm)
1282                 goto out_files;
1283
1284         retval = mutex_lock_interruptible(&current->cred_exec_mutex);
1285         if (retval < 0)
1286                 goto out_free;
1287
1288         retval = -ENOMEM;
1289         bprm->cred = prepare_exec_creds();
1290         if (!bprm->cred)
1291                 goto out_unlock;
1292         check_unsafe_exec(bprm, displaced);
1293
1294         file = open_exec(filename);
1295         retval = PTR_ERR(file);
1296         if (IS_ERR(file))
1297                 goto out_unlock;
1298
1299         sched_exec();
1300
1301         bprm->file = file;
1302         bprm->filename = filename;
1303         bprm->interp = filename;
1304
1305         retval = bprm_mm_init(bprm);
1306         if (retval)
1307                 goto out_file;
1308
1309         bprm->argc = count(argv, MAX_ARG_STRINGS);
1310         if ((retval = bprm->argc) < 0)
1311                 goto out;
1312
1313         bprm->envc = count(envp, MAX_ARG_STRINGS);
1314         if ((retval = bprm->envc) < 0)
1315                 goto out;
1316
1317         retval = prepare_binprm(bprm);
1318         if (retval < 0)
1319                 goto out;
1320
1321         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1322         if (retval < 0)
1323                 goto out;
1324
1325         bprm->exec = bprm->p;
1326         retval = copy_strings(bprm->envc, envp, bprm);
1327         if (retval < 0)
1328                 goto out;
1329
1330         retval = copy_strings(bprm->argc, argv, bprm);
1331         if (retval < 0)
1332                 goto out;
1333
1334         current->flags &= ~PF_KTHREAD;
1335         retval = search_binary_handler(bprm,regs);
1336         if (retval < 0)
1337                 goto out;
1338
1339         /* execve succeeded */
1340         mutex_unlock(&current->cred_exec_mutex);
1341         acct_update_integrals(current);
1342         free_bprm(bprm);
1343         if (displaced)
1344                 put_files_struct(displaced);
1345         return retval;
1346
1347 out:
1348         if (bprm->mm)
1349                 mmput (bprm->mm);
1350
1351 out_file:
1352         if (bprm->file) {
1353                 allow_write_access(bprm->file);
1354                 fput(bprm->file);
1355         }
1356
1357 out_unlock:
1358         mutex_unlock(&current->cred_exec_mutex);
1359
1360 out_free:
1361         free_bprm(bprm);
1362
1363 out_files:
1364         if (displaced)
1365                 reset_files_struct(displaced);
1366 out_ret:
1367         return retval;
1368 }
1369
1370 int set_binfmt(struct linux_binfmt *new)
1371 {
1372         struct linux_binfmt *old = current->binfmt;
1373
1374         if (new) {
1375                 if (!try_module_get(new->module))
1376                         return -1;
1377         }
1378         current->binfmt = new;
1379         if (old)
1380                 module_put(old->module);
1381         return 0;
1382 }
1383
1384 EXPORT_SYMBOL(set_binfmt);
1385
1386 /* format_corename will inspect the pattern parameter, and output a
1387  * name into corename, which must have space for at least
1388  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1389  */
1390 static int format_corename(char *corename, long signr)
1391 {
1392         const struct cred *cred = current_cred();
1393         const char *pat_ptr = core_pattern;
1394         int ispipe = (*pat_ptr == '|');
1395         char *out_ptr = corename;
1396         char *const out_end = corename + CORENAME_MAX_SIZE;
1397         int rc;
1398         int pid_in_pattern = 0;
1399
1400         /* Repeat as long as we have more pattern to process and more output
1401            space */
1402         while (*pat_ptr) {
1403                 if (*pat_ptr != '%') {
1404                         if (out_ptr == out_end)
1405                                 goto out;
1406                         *out_ptr++ = *pat_ptr++;
1407                 } else {
1408                         switch (*++pat_ptr) {
1409                         case 0:
1410                                 goto out;
1411                         /* Double percent, output one percent */
1412                         case '%':
1413                                 if (out_ptr == out_end)
1414                                         goto out;
1415                                 *out_ptr++ = '%';
1416                                 break;
1417                         /* pid */
1418                         case 'p':
1419                                 pid_in_pattern = 1;
1420                                 rc = snprintf(out_ptr, out_end - out_ptr,
1421                                               "%d", task_tgid_vnr(current));
1422                                 if (rc > out_end - out_ptr)
1423                                         goto out;
1424                                 out_ptr += rc;
1425                                 break;
1426                         /* uid */
1427                         case 'u':
1428                                 rc = snprintf(out_ptr, out_end - out_ptr,
1429                                               "%d", cred->uid);
1430                                 if (rc > out_end - out_ptr)
1431                                         goto out;
1432                                 out_ptr += rc;
1433                                 break;
1434                         /* gid */
1435                         case 'g':
1436                                 rc = snprintf(out_ptr, out_end - out_ptr,
1437                                               "%d", cred->gid);
1438                                 if (rc > out_end - out_ptr)
1439                                         goto out;
1440                                 out_ptr += rc;
1441                                 break;
1442                         /* signal that caused the coredump */
1443                         case 's':
1444                                 rc = snprintf(out_ptr, out_end - out_ptr,
1445                                               "%ld", signr);
1446                                 if (rc > out_end - out_ptr)
1447                                         goto out;
1448                                 out_ptr += rc;
1449                                 break;
1450                         /* UNIX time of coredump */
1451                         case 't': {
1452                                 struct timeval tv;
1453                                 do_gettimeofday(&tv);
1454                                 rc = snprintf(out_ptr, out_end - out_ptr,
1455                                               "%lu", tv.tv_sec);
1456                                 if (rc > out_end - out_ptr)
1457                                         goto out;
1458                                 out_ptr += rc;
1459                                 break;
1460                         }
1461                         /* hostname */
1462                         case 'h':
1463                                 down_read(&uts_sem);
1464                                 rc = snprintf(out_ptr, out_end - out_ptr,
1465                                               "%s", utsname()->nodename);
1466                                 up_read(&uts_sem);
1467                                 if (rc > out_end - out_ptr)
1468                                         goto out;
1469                                 out_ptr += rc;
1470                                 break;
1471                         /* executable */
1472                         case 'e':
1473                                 rc = snprintf(out_ptr, out_end - out_ptr,
1474                                               "%s", current->comm);
1475                                 if (rc > out_end - out_ptr)
1476                                         goto out;
1477                                 out_ptr += rc;
1478                                 break;
1479                         /* core limit size */
1480                         case 'c':
1481                                 rc = snprintf(out_ptr, out_end - out_ptr,
1482                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1483                                 if (rc > out_end - out_ptr)
1484                                         goto out;
1485                                 out_ptr += rc;
1486                                 break;
1487                         default:
1488                                 break;
1489                         }
1490                         ++pat_ptr;
1491                 }
1492         }
1493         /* Backward compatibility with core_uses_pid:
1494          *
1495          * If core_pattern does not include a %p (as is the default)
1496          * and core_uses_pid is set, then .%pid will be appended to
1497          * the filename. Do not do this for piped commands. */
1498         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1499                 rc = snprintf(out_ptr, out_end - out_ptr,
1500                               ".%d", task_tgid_vnr(current));
1501                 if (rc > out_end - out_ptr)
1502                         goto out;
1503                 out_ptr += rc;
1504         }
1505 out:
1506         *out_ptr = 0;
1507         return ispipe;
1508 }
1509
1510 static int zap_process(struct task_struct *start)
1511 {
1512         struct task_struct *t;
1513         int nr = 0;
1514
1515         start->signal->flags = SIGNAL_GROUP_EXIT;
1516         start->signal->group_stop_count = 0;
1517
1518         t = start;
1519         do {
1520                 if (t != current && t->mm) {
1521                         sigaddset(&t->pending.signal, SIGKILL);
1522                         signal_wake_up(t, 1);
1523                         nr++;
1524                 }
1525         } while_each_thread(start, t);
1526
1527         return nr;
1528 }
1529
1530 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1531                                 struct core_state *core_state, int exit_code)
1532 {
1533         struct task_struct *g, *p;
1534         unsigned long flags;
1535         int nr = -EAGAIN;
1536
1537         spin_lock_irq(&tsk->sighand->siglock);
1538         if (!signal_group_exit(tsk->signal)) {
1539                 mm->core_state = core_state;
1540                 tsk->signal->group_exit_code = exit_code;
1541                 nr = zap_process(tsk);
1542         }
1543         spin_unlock_irq(&tsk->sighand->siglock);
1544         if (unlikely(nr < 0))
1545                 return nr;
1546
1547         if (atomic_read(&mm->mm_users) == nr + 1)
1548                 goto done;
1549         /*
1550          * We should find and kill all tasks which use this mm, and we should
1551          * count them correctly into ->nr_threads. We don't take tasklist
1552          * lock, but this is safe wrt:
1553          *
1554          * fork:
1555          *      None of sub-threads can fork after zap_process(leader). All
1556          *      processes which were created before this point should be
1557          *      visible to zap_threads() because copy_process() adds the new
1558          *      process to the tail of init_task.tasks list, and lock/unlock
1559          *      of ->siglock provides a memory barrier.
1560          *
1561          * do_exit:
1562          *      The caller holds mm->mmap_sem. This means that the task which
1563          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1564          *      its ->mm.
1565          *
1566          * de_thread:
1567          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1568          *      we must see either old or new leader, this does not matter.
1569          *      However, it can change p->sighand, so lock_task_sighand(p)
1570          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1571          *      it can't fail.
1572          *
1573          *      Note also that "g" can be the old leader with ->mm == NULL
1574          *      and already unhashed and thus removed from ->thread_group.
1575          *      This is OK, __unhash_process()->list_del_rcu() does not
1576          *      clear the ->next pointer, we will find the new leader via
1577          *      next_thread().
1578          */
1579         rcu_read_lock();
1580         for_each_process(g) {
1581                 if (g == tsk->group_leader)
1582                         continue;
1583                 if (g->flags & PF_KTHREAD)
1584                         continue;
1585                 p = g;
1586                 do {
1587                         if (p->mm) {
1588                                 if (unlikely(p->mm == mm)) {
1589                                         lock_task_sighand(p, &flags);
1590                                         nr += zap_process(p);
1591                                         unlock_task_sighand(p, &flags);
1592                                 }
1593                                 break;
1594                         }
1595                 } while_each_thread(g, p);
1596         }
1597         rcu_read_unlock();
1598 done:
1599         atomic_set(&core_state->nr_threads, nr);
1600         return nr;
1601 }
1602
1603 static int coredump_wait(int exit_code, struct core_state *core_state)
1604 {
1605         struct task_struct *tsk = current;
1606         struct mm_struct *mm = tsk->mm;
1607         struct completion *vfork_done;
1608         int core_waiters;
1609
1610         init_completion(&core_state->startup);
1611         core_state->dumper.task = tsk;
1612         core_state->dumper.next = NULL;
1613         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1614         up_write(&mm->mmap_sem);
1615
1616         if (unlikely(core_waiters < 0))
1617                 goto fail;
1618
1619         /*
1620          * Make sure nobody is waiting for us to release the VM,
1621          * otherwise we can deadlock when we wait on each other
1622          */
1623         vfork_done = tsk->vfork_done;
1624         if (vfork_done) {
1625                 tsk->vfork_done = NULL;
1626                 complete(vfork_done);
1627         }
1628
1629         if (core_waiters)
1630                 wait_for_completion(&core_state->startup);
1631 fail:
1632         return core_waiters;
1633 }
1634
1635 static void coredump_finish(struct mm_struct *mm)
1636 {
1637         struct core_thread *curr, *next;
1638         struct task_struct *task;
1639
1640         next = mm->core_state->dumper.next;
1641         while ((curr = next) != NULL) {
1642                 next = curr->next;
1643                 task = curr->task;
1644                 /*
1645                  * see exit_mm(), curr->task must not see
1646                  * ->task == NULL before we read ->next.
1647                  */
1648                 smp_mb();
1649                 curr->task = NULL;
1650                 wake_up_process(task);
1651         }
1652
1653         mm->core_state = NULL;
1654 }
1655
1656 /*
1657  * set_dumpable converts traditional three-value dumpable to two flags and
1658  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1659  * these bits are not changed atomically.  So get_dumpable can observe the
1660  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1661  * return either old dumpable or new one by paying attention to the order of
1662  * modifying the bits.
1663  *
1664  * dumpable |   mm->flags (binary)
1665  * old  new | initial interim  final
1666  * ---------+-----------------------
1667  *  0    1  |   00      01      01
1668  *  0    2  |   00      10(*)   11
1669  *  1    0  |   01      00      00
1670  *  1    2  |   01      11      11
1671  *  2    0  |   11      10(*)   00
1672  *  2    1  |   11      11      01
1673  *
1674  * (*) get_dumpable regards interim value of 10 as 11.
1675  */
1676 void set_dumpable(struct mm_struct *mm, int value)
1677 {
1678         switch (value) {
1679         case 0:
1680                 clear_bit(MMF_DUMPABLE, &mm->flags);
1681                 smp_wmb();
1682                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1683                 break;
1684         case 1:
1685                 set_bit(MMF_DUMPABLE, &mm->flags);
1686                 smp_wmb();
1687                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1688                 break;
1689         case 2:
1690                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1691                 smp_wmb();
1692                 set_bit(MMF_DUMPABLE, &mm->flags);
1693                 break;
1694         }
1695 }
1696
1697 int get_dumpable(struct mm_struct *mm)
1698 {
1699         int ret;
1700
1701         ret = mm->flags & 0x3;
1702         return (ret >= 2) ? 2 : ret;
1703 }
1704
1705 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1706 {
1707         struct core_state core_state;
1708         char corename[CORENAME_MAX_SIZE + 1];
1709         struct mm_struct *mm = current->mm;
1710         struct linux_binfmt * binfmt;
1711         struct inode * inode;
1712         struct file * file;
1713         const struct cred *old_cred;
1714         struct cred *cred;
1715         int retval = 0;
1716         int flag = 0;
1717         int ispipe = 0;
1718         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1719         char **helper_argv = NULL;
1720         int helper_argc = 0;
1721         char *delimit;
1722
1723         audit_core_dumps(signr);
1724
1725         binfmt = current->binfmt;
1726         if (!binfmt || !binfmt->core_dump)
1727                 goto fail;
1728
1729         cred = prepare_creds();
1730         if (!cred) {
1731                 retval = -ENOMEM;
1732                 goto fail;
1733         }
1734
1735         down_write(&mm->mmap_sem);
1736         /*
1737          * If another thread got here first, or we are not dumpable, bail out.
1738          */
1739         if (mm->core_state || !get_dumpable(mm)) {
1740                 up_write(&mm->mmap_sem);
1741                 put_cred(cred);
1742                 goto fail;
1743         }
1744
1745         /*
1746          *      We cannot trust fsuid as being the "true" uid of the
1747          *      process nor do we know its entire history. We only know it
1748          *      was tainted so we dump it as root in mode 2.
1749          */
1750         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1751                 flag = O_EXCL;          /* Stop rewrite attacks */
1752                 cred->fsuid = 0;        /* Dump root private */
1753         }
1754
1755         retval = coredump_wait(exit_code, &core_state);
1756         if (retval < 0) {
1757                 put_cred(cred);
1758                 goto fail;
1759         }
1760
1761         old_cred = override_creds(cred);
1762
1763         /*
1764          * Clear any false indication of pending signals that might
1765          * be seen by the filesystem code called to write the core file.
1766          */
1767         clear_thread_flag(TIF_SIGPENDING);
1768
1769         /*
1770          * lock_kernel() because format_corename() is controlled by sysctl, which
1771          * uses lock_kernel()
1772          */
1773         lock_kernel();
1774         ispipe = format_corename(corename, signr);
1775         unlock_kernel();
1776         /*
1777          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1778          * to a pipe.  Since we're not writing directly to the filesystem
1779          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1780          * created unless the pipe reader choses to write out the core file
1781          * at which point file size limits and permissions will be imposed
1782          * as it does with any other process
1783          */
1784         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1785                 goto fail_unlock;
1786
1787         if (ispipe) {
1788                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1789                 if (!helper_argv) {
1790                         printk(KERN_WARNING "%s failed to allocate memory\n",
1791                                __func__);
1792                         goto fail_unlock;
1793                 }
1794                 /* Terminate the string before the first option */
1795                 delimit = strchr(corename, ' ');
1796                 if (delimit)
1797                         *delimit = '\0';
1798                 delimit = strrchr(helper_argv[0], '/');
1799                 if (delimit)
1800                         delimit++;
1801                 else
1802                         delimit = helper_argv[0];
1803                 if (!strcmp(delimit, current->comm)) {
1804                         printk(KERN_NOTICE "Recursive core dump detected, "
1805                                         "aborting\n");
1806                         goto fail_unlock;
1807                 }
1808
1809                 core_limit = RLIM_INFINITY;
1810
1811                 /* SIGPIPE can happen, but it's just never processed */
1812                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1813                                 &file)) {
1814                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1815                                corename);
1816                         goto fail_unlock;
1817                 }
1818         } else
1819                 file = filp_open(corename,
1820                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1821                                  0600);
1822         if (IS_ERR(file))
1823                 goto fail_unlock;
1824         inode = file->f_path.dentry->d_inode;
1825         if (inode->i_nlink > 1)
1826                 goto close_fail;        /* multiple links - don't dump */
1827         if (!ispipe && d_unhashed(file->f_path.dentry))
1828                 goto close_fail;
1829
1830         /* AK: actually i see no reason to not allow this for named pipes etc.,
1831            but keep the previous behaviour for now. */
1832         if (!ispipe && !S_ISREG(inode->i_mode))
1833                 goto close_fail;
1834         /*
1835          * Dont allow local users get cute and trick others to coredump
1836          * into their pre-created files:
1837          */
1838         if (inode->i_uid != current_fsuid())
1839                 goto close_fail;
1840         if (!file->f_op)
1841                 goto close_fail;
1842         if (!file->f_op->write)
1843                 goto close_fail;
1844         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1845                 goto close_fail;
1846
1847         retval = binfmt->core_dump(signr, regs, file, core_limit);
1848
1849         if (retval)
1850                 current->signal->group_exit_code |= 0x80;
1851 close_fail:
1852         filp_close(file, NULL);
1853 fail_unlock:
1854         if (helper_argv)
1855                 argv_free(helper_argv);
1856
1857         revert_creds(old_cred);
1858         put_cred(cred);
1859         coredump_finish(mm);
1860 fail:
1861         return;
1862 }