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