[PATCH] pass MAY_OPEN to vfs_permission() explicitly
[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/highmem.h>
36 #include <linux/spinlock.h>
37 #include <linux/key.h>
38 #include <linux/personality.h>
39 #include <linux/binfmts.h>
40 #include <linux/utsname.h>
41 #include <linux/pid_namespace.h>
42 #include <linux/module.h>
43 #include <linux/namei.h>
44 #include <linux/proc_fs.h>
45 #include <linux/mount.h>
46 #include <linux/security.h>
47 #include <linux/syscalls.h>
48 #include <linux/tsacct_kern.h>
49 #include <linux/cn_proc.h>
50 #include <linux/audit.h>
51 #include <linux/tracehook.h>
52
53 #include <asm/uaccess.h>
54 #include <asm/mmu_context.h>
55 #include <asm/tlb.h>
56
57 #ifdef CONFIG_KMOD
58 #include <linux/kmod.h>
59 #endif
60
61 #ifdef __alpha__
62 /* for /sbin/loader handling in search_binary_handler() */
63 #include <linux/a.out.h>
64 #endif
65
66 int core_uses_pid;
67 char core_pattern[CORENAME_MAX_SIZE] = "core";
68 int suid_dumpable = 0;
69
70 /* The maximal length of core_pattern is also specified in sysctl.c */
71
72 static LIST_HEAD(formats);
73 static DEFINE_RWLOCK(binfmt_lock);
74
75 int register_binfmt(struct linux_binfmt * fmt)
76 {
77         if (!fmt)
78                 return -EINVAL;
79         write_lock(&binfmt_lock);
80         list_add(&fmt->lh, &formats);
81         write_unlock(&binfmt_lock);
82         return 0;       
83 }
84
85 EXPORT_SYMBOL(register_binfmt);
86
87 void unregister_binfmt(struct linux_binfmt * fmt)
88 {
89         write_lock(&binfmt_lock);
90         list_del(&fmt->lh);
91         write_unlock(&binfmt_lock);
92 }
93
94 EXPORT_SYMBOL(unregister_binfmt);
95
96 static inline void put_binfmt(struct linux_binfmt * fmt)
97 {
98         module_put(fmt->module);
99 }
100
101 /*
102  * Note that a shared library must be both readable and executable due to
103  * security reasons.
104  *
105  * Also note that we take the address to load from from the file itself.
106  */
107 asmlinkage long sys_uselib(const char __user * library)
108 {
109         struct file * file;
110         struct nameidata nd;
111         int error;
112
113         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
114         if (error)
115                 goto out;
116
117         error = -EINVAL;
118         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
119                 goto exit;
120
121         error = vfs_permission(&nd, MAY_READ | MAY_EXEC | MAY_OPEN);
122         if (error)
123                 goto exit;
124
125         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
126         error = PTR_ERR(file);
127         if (IS_ERR(file))
128                 goto out;
129
130         error = -ENOEXEC;
131         if(file->f_op) {
132                 struct linux_binfmt * fmt;
133
134                 read_lock(&binfmt_lock);
135                 list_for_each_entry(fmt, &formats, lh) {
136                         if (!fmt->load_shlib)
137                                 continue;
138                         if (!try_module_get(fmt->module))
139                                 continue;
140                         read_unlock(&binfmt_lock);
141                         error = fmt->load_shlib(file);
142                         read_lock(&binfmt_lock);
143                         put_binfmt(fmt);
144                         if (error != -ENOEXEC)
145                                 break;
146                 }
147                 read_unlock(&binfmt_lock);
148         }
149         fput(file);
150 out:
151         return error;
152 exit:
153         release_open_intent(&nd);
154         path_put(&nd.path);
155         goto out;
156 }
157
158 #ifdef CONFIG_MMU
159
160 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
161                 int write)
162 {
163         struct page *page;
164         int ret;
165
166 #ifdef CONFIG_STACK_GROWSUP
167         if (write) {
168                 ret = expand_stack_downwards(bprm->vma, pos);
169                 if (ret < 0)
170                         return NULL;
171         }
172 #endif
173         ret = get_user_pages(current, bprm->mm, pos,
174                         1, write, 1, &page, NULL);
175         if (ret <= 0)
176                 return NULL;
177
178         if (write) {
179                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
180                 struct rlimit *rlim;
181
182                 /*
183                  * We've historically supported up to 32 pages (ARG_MAX)
184                  * of argument strings even with small stacks
185                  */
186                 if (size <= ARG_MAX)
187                         return page;
188
189                 /*
190                  * Limit to 1/4-th the stack size for the argv+env strings.
191                  * This ensures that:
192                  *  - the remaining binfmt code will not run out of stack space,
193                  *  - the program will have a reasonable amount of stack left
194                  *    to work from.
195                  */
196                 rlim = current->signal->rlim;
197                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
198                         put_page(page);
199                         return NULL;
200                 }
201         }
202
203         return page;
204 }
205
206 static void put_arg_page(struct page *page)
207 {
208         put_page(page);
209 }
210
211 static void free_arg_page(struct linux_binprm *bprm, int i)
212 {
213 }
214
215 static void free_arg_pages(struct linux_binprm *bprm)
216 {
217 }
218
219 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
220                 struct page *page)
221 {
222         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
223 }
224
225 static int __bprm_mm_init(struct linux_binprm *bprm)
226 {
227         int err = -ENOMEM;
228         struct vm_area_struct *vma = NULL;
229         struct mm_struct *mm = bprm->mm;
230
231         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
232         if (!vma)
233                 goto err;
234
235         down_write(&mm->mmap_sem);
236         vma->vm_mm = mm;
237
238         /*
239          * Place the stack at the largest stack address the architecture
240          * supports. Later, we'll move this to an appropriate place. We don't
241          * use STACK_TOP because that can depend on attributes which aren't
242          * configured yet.
243          */
244         vma->vm_end = STACK_TOP_MAX;
245         vma->vm_start = vma->vm_end - PAGE_SIZE;
246
247         vma->vm_flags = VM_STACK_FLAGS;
248         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
249         err = insert_vm_struct(mm, vma);
250         if (err) {
251                 up_write(&mm->mmap_sem);
252                 goto err;
253         }
254
255         mm->stack_vm = mm->total_vm = 1;
256         up_write(&mm->mmap_sem);
257
258         bprm->p = vma->vm_end - sizeof(void *);
259
260         return 0;
261
262 err:
263         if (vma) {
264                 bprm->vma = NULL;
265                 kmem_cache_free(vm_area_cachep, vma);
266         }
267
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         int err;
660         struct file *file;
661
662         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
663         file = ERR_PTR(err);
664
665         if (!err) {
666                 struct inode *inode = nd.path.dentry->d_inode;
667                 file = ERR_PTR(-EACCES);
668                 if (S_ISREG(inode->i_mode)) {
669                         int err = vfs_permission(&nd, MAY_EXEC | MAY_OPEN);
670                         file = ERR_PTR(err);
671                         if (!err) {
672                                 file = nameidata_to_filp(&nd,
673                                                         O_RDONLY|O_LARGEFILE);
674                                 if (!IS_ERR(file)) {
675                                         err = deny_write_access(file);
676                                         if (err) {
677                                                 fput(file);
678                                                 file = ERR_PTR(err);
679                                         }
680                                 }
681 out:
682                                 return file;
683                         }
684                 }
685                 release_open_intent(&nd);
686                 path_put(&nd.path);
687         }
688         goto out;
689 }
690
691 EXPORT_SYMBOL(open_exec);
692
693 int kernel_read(struct file *file, unsigned long offset,
694         char *addr, unsigned long count)
695 {
696         mm_segment_t old_fs;
697         loff_t pos = offset;
698         int result;
699
700         old_fs = get_fs();
701         set_fs(get_ds());
702         /* The cast to a user pointer is valid due to the set_fs() */
703         result = vfs_read(file, (void __user *)addr, count, &pos);
704         set_fs(old_fs);
705         return result;
706 }
707
708 EXPORT_SYMBOL(kernel_read);
709
710 static int exec_mmap(struct mm_struct *mm)
711 {
712         struct task_struct *tsk;
713         struct mm_struct * old_mm, *active_mm;
714
715         /* Notify parent that we're no longer interested in the old VM */
716         tsk = current;
717         old_mm = current->mm;
718         mm_release(tsk, old_mm);
719
720         if (old_mm) {
721                 /*
722                  * Make sure that if there is a core dump in progress
723                  * for the old mm, we get out and die instead of going
724                  * through with the exec.  We must hold mmap_sem around
725                  * checking core_state and changing tsk->mm.
726                  */
727                 down_read(&old_mm->mmap_sem);
728                 if (unlikely(old_mm->core_state)) {
729                         up_read(&old_mm->mmap_sem);
730                         return -EINTR;
731                 }
732         }
733         task_lock(tsk);
734         active_mm = tsk->active_mm;
735         tsk->mm = mm;
736         tsk->active_mm = mm;
737         activate_mm(active_mm, mm);
738         task_unlock(tsk);
739         mm_update_next_owner(old_mm);
740         arch_pick_mmap_layout(mm);
741         if (old_mm) {
742                 up_read(&old_mm->mmap_sem);
743                 BUG_ON(active_mm != old_mm);
744                 mmput(old_mm);
745                 return 0;
746         }
747         mmdrop(active_mm);
748         return 0;
749 }
750
751 /*
752  * This function makes sure the current process has its own signal table,
753  * so that flush_signal_handlers can later reset the handlers without
754  * disturbing other processes.  (Other processes might share the signal
755  * table via the CLONE_SIGHAND option to clone().)
756  */
757 static int de_thread(struct task_struct *tsk)
758 {
759         struct signal_struct *sig = tsk->signal;
760         struct sighand_struct *oldsighand = tsk->sighand;
761         spinlock_t *lock = &oldsighand->siglock;
762         struct task_struct *leader = NULL;
763         int count;
764
765         if (thread_group_empty(tsk))
766                 goto no_thread_group;
767
768         /*
769          * Kill all other threads in the thread group.
770          */
771         spin_lock_irq(lock);
772         if (signal_group_exit(sig)) {
773                 /*
774                  * Another group action in progress, just
775                  * return so that the signal is processed.
776                  */
777                 spin_unlock_irq(lock);
778                 return -EAGAIN;
779         }
780         sig->group_exit_task = tsk;
781         zap_other_threads(tsk);
782
783         /* Account for the thread group leader hanging around: */
784         count = thread_group_leader(tsk) ? 1 : 2;
785         sig->notify_count = count;
786         while (atomic_read(&sig->count) > count) {
787                 __set_current_state(TASK_UNINTERRUPTIBLE);
788                 spin_unlock_irq(lock);
789                 schedule();
790                 spin_lock_irq(lock);
791         }
792         spin_unlock_irq(lock);
793
794         /*
795          * At this point all other threads have exited, all we have to
796          * do is to wait for the thread group leader to become inactive,
797          * and to assume its PID:
798          */
799         if (!thread_group_leader(tsk)) {
800                 leader = tsk->group_leader;
801
802                 sig->notify_count = -1; /* for exit_notify() */
803                 for (;;) {
804                         write_lock_irq(&tasklist_lock);
805                         if (likely(leader->exit_state))
806                                 break;
807                         __set_current_state(TASK_UNINTERRUPTIBLE);
808                         write_unlock_irq(&tasklist_lock);
809                         schedule();
810                 }
811
812                 if (unlikely(task_child_reaper(tsk) == leader))
813                         task_active_pid_ns(tsk)->child_reaper = tsk;
814                 /*
815                  * The only record we have of the real-time age of a
816                  * process, regardless of execs it's done, is start_time.
817                  * All the past CPU time is accumulated in signal_struct
818                  * from sister threads now dead.  But in this non-leader
819                  * exec, nothing survives from the original leader thread,
820                  * whose birth marks the true age of this process now.
821                  * When we take on its identity by switching to its PID, we
822                  * also take its birthdate (always earlier than our own).
823                  */
824                 tsk->start_time = leader->start_time;
825
826                 BUG_ON(!same_thread_group(leader, tsk));
827                 BUG_ON(has_group_leader_pid(tsk));
828                 /*
829                  * An exec() starts a new thread group with the
830                  * TGID of the previous thread group. Rehash the
831                  * two threads with a switched PID, and release
832                  * the former thread group leader:
833                  */
834
835                 /* Become a process group leader with the old leader's pid.
836                  * The old leader becomes a thread of the this thread group.
837                  * Note: The old leader also uses this pid until release_task
838                  *       is called.  Odd but simple and correct.
839                  */
840                 detach_pid(tsk, PIDTYPE_PID);
841                 tsk->pid = leader->pid;
842                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
843                 transfer_pid(leader, tsk, PIDTYPE_PGID);
844                 transfer_pid(leader, tsk, PIDTYPE_SID);
845                 list_replace_rcu(&leader->tasks, &tsk->tasks);
846
847                 tsk->group_leader = tsk;
848                 leader->group_leader = tsk;
849
850                 tsk->exit_signal = SIGCHLD;
851
852                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
853                 leader->exit_state = EXIT_DEAD;
854
855                 write_unlock_irq(&tasklist_lock);
856         }
857
858         sig->group_exit_task = NULL;
859         sig->notify_count = 0;
860
861 no_thread_group:
862         exit_itimers(sig);
863         flush_itimer_signals();
864         if (leader)
865                 release_task(leader);
866
867         if (atomic_read(&oldsighand->count) != 1) {
868                 struct sighand_struct *newsighand;
869                 /*
870                  * This ->sighand is shared with the CLONE_SIGHAND
871                  * but not CLONE_THREAD task, switch to the new one.
872                  */
873                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
874                 if (!newsighand)
875                         return -ENOMEM;
876
877                 atomic_set(&newsighand->count, 1);
878                 memcpy(newsighand->action, oldsighand->action,
879                        sizeof(newsighand->action));
880
881                 write_lock_irq(&tasklist_lock);
882                 spin_lock(&oldsighand->siglock);
883                 rcu_assign_pointer(tsk->sighand, newsighand);
884                 spin_unlock(&oldsighand->siglock);
885                 write_unlock_irq(&tasklist_lock);
886
887                 __cleanup_sighand(oldsighand);
888         }
889
890         BUG_ON(!thread_group_leader(tsk));
891         return 0;
892 }
893
894 /*
895  * These functions flushes out all traces of the currently running executable
896  * so that a new one can be started
897  */
898 static void flush_old_files(struct files_struct * files)
899 {
900         long j = -1;
901         struct fdtable *fdt;
902
903         spin_lock(&files->file_lock);
904         for (;;) {
905                 unsigned long set, i;
906
907                 j++;
908                 i = j * __NFDBITS;
909                 fdt = files_fdtable(files);
910                 if (i >= fdt->max_fds)
911                         break;
912                 set = fdt->close_on_exec->fds_bits[j];
913                 if (!set)
914                         continue;
915                 fdt->close_on_exec->fds_bits[j] = 0;
916                 spin_unlock(&files->file_lock);
917                 for ( ; set ; i++,set >>= 1) {
918                         if (set & 1) {
919                                 sys_close(i);
920                         }
921                 }
922                 spin_lock(&files->file_lock);
923
924         }
925         spin_unlock(&files->file_lock);
926 }
927
928 char *get_task_comm(char *buf, struct task_struct *tsk)
929 {
930         /* buf must be at least sizeof(tsk->comm) in size */
931         task_lock(tsk);
932         strncpy(buf, tsk->comm, sizeof(tsk->comm));
933         task_unlock(tsk);
934         return buf;
935 }
936
937 void set_task_comm(struct task_struct *tsk, char *buf)
938 {
939         task_lock(tsk);
940         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
941         task_unlock(tsk);
942 }
943
944 int flush_old_exec(struct linux_binprm * bprm)
945 {
946         char * name;
947         int i, ch, retval;
948         char tcomm[sizeof(current->comm)];
949
950         /*
951          * Make sure we have a private signal table and that
952          * we are unassociated from the previous thread group.
953          */
954         retval = de_thread(current);
955         if (retval)
956                 goto out;
957
958         set_mm_exe_file(bprm->mm, bprm->file);
959
960         /*
961          * Release all of the old mmap stuff
962          */
963         retval = exec_mmap(bprm->mm);
964         if (retval)
965                 goto out;
966
967         bprm->mm = NULL;                /* We're using it now */
968
969         /* This is the point of no return */
970         current->sas_ss_sp = current->sas_ss_size = 0;
971
972         if (current->euid == current->uid && current->egid == current->gid)
973                 set_dumpable(current->mm, 1);
974         else
975                 set_dumpable(current->mm, suid_dumpable);
976
977         name = bprm->filename;
978
979         /* Copies the binary name from after last slash */
980         for (i=0; (ch = *(name++)) != '\0';) {
981                 if (ch == '/')
982                         i = 0; /* overwrite what we wrote */
983                 else
984                         if (i < (sizeof(tcomm) - 1))
985                                 tcomm[i++] = ch;
986         }
987         tcomm[i] = '\0';
988         set_task_comm(current, tcomm);
989
990         current->flags &= ~PF_RANDOMIZE;
991         flush_thread();
992
993         /* Set the new mm task size. We have to do that late because it may
994          * depend on TIF_32BIT which is only updated in flush_thread() on
995          * some architectures like powerpc
996          */
997         current->mm->task_size = TASK_SIZE;
998
999         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1000                 suid_keys(current);
1001                 set_dumpable(current->mm, suid_dumpable);
1002                 current->pdeath_signal = 0;
1003         } else if (file_permission(bprm->file, MAY_READ) ||
1004                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1005                 suid_keys(current);
1006                 set_dumpable(current->mm, suid_dumpable);
1007         }
1008
1009         /* An exec changes our domain. We are no longer part of the thread
1010            group */
1011
1012         current->self_exec_id++;
1013                         
1014         flush_signal_handlers(current, 0);
1015         flush_old_files(current->files);
1016
1017         return 0;
1018
1019 out:
1020         return retval;
1021 }
1022
1023 EXPORT_SYMBOL(flush_old_exec);
1024
1025 /* 
1026  * Fill the binprm structure from the inode. 
1027  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1028  */
1029 int prepare_binprm(struct linux_binprm *bprm)
1030 {
1031         int mode;
1032         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1033         int retval;
1034
1035         mode = inode->i_mode;
1036         if (bprm->file->f_op == NULL)
1037                 return -EACCES;
1038
1039         bprm->e_uid = current->euid;
1040         bprm->e_gid = current->egid;
1041
1042         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1043                 /* Set-uid? */
1044                 if (mode & S_ISUID) {
1045                         current->personality &= ~PER_CLEAR_ON_SETID;
1046                         bprm->e_uid = inode->i_uid;
1047                 }
1048
1049                 /* Set-gid? */
1050                 /*
1051                  * If setgid is set but no group execute bit then this
1052                  * is a candidate for mandatory locking, not a setgid
1053                  * executable.
1054                  */
1055                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1056                         current->personality &= ~PER_CLEAR_ON_SETID;
1057                         bprm->e_gid = inode->i_gid;
1058                 }
1059         }
1060
1061         /* fill in binprm security blob */
1062         retval = security_bprm_set(bprm);
1063         if (retval)
1064                 return retval;
1065
1066         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1067         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1068 }
1069
1070 EXPORT_SYMBOL(prepare_binprm);
1071
1072 static int unsafe_exec(struct task_struct *p)
1073 {
1074         int unsafe = tracehook_unsafe_exec(p);
1075
1076         if (atomic_read(&p->fs->count) > 1 ||
1077             atomic_read(&p->files->count) > 1 ||
1078             atomic_read(&p->sighand->count) > 1)
1079                 unsafe |= LSM_UNSAFE_SHARE;
1080
1081         return unsafe;
1082 }
1083
1084 void compute_creds(struct linux_binprm *bprm)
1085 {
1086         int unsafe;
1087
1088         if (bprm->e_uid != current->uid) {
1089                 suid_keys(current);
1090                 current->pdeath_signal = 0;
1091         }
1092         exec_keys(current);
1093
1094         task_lock(current);
1095         unsafe = unsafe_exec(current);
1096         security_bprm_apply_creds(bprm, unsafe);
1097         task_unlock(current);
1098         security_bprm_post_apply_creds(bprm);
1099 }
1100 EXPORT_SYMBOL(compute_creds);
1101
1102 /*
1103  * Arguments are '\0' separated strings found at the location bprm->p
1104  * points to; chop off the first by relocating brpm->p to right after
1105  * the first '\0' encountered.
1106  */
1107 int remove_arg_zero(struct linux_binprm *bprm)
1108 {
1109         int ret = 0;
1110         unsigned long offset;
1111         char *kaddr;
1112         struct page *page;
1113
1114         if (!bprm->argc)
1115                 return 0;
1116
1117         do {
1118                 offset = bprm->p & ~PAGE_MASK;
1119                 page = get_arg_page(bprm, bprm->p, 0);
1120                 if (!page) {
1121                         ret = -EFAULT;
1122                         goto out;
1123                 }
1124                 kaddr = kmap_atomic(page, KM_USER0);
1125
1126                 for (; offset < PAGE_SIZE && kaddr[offset];
1127                                 offset++, bprm->p++)
1128                         ;
1129
1130                 kunmap_atomic(kaddr, KM_USER0);
1131                 put_arg_page(page);
1132
1133                 if (offset == PAGE_SIZE)
1134                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1135         } while (offset == PAGE_SIZE);
1136
1137         bprm->p++;
1138         bprm->argc--;
1139         ret = 0;
1140
1141 out:
1142         return ret;
1143 }
1144 EXPORT_SYMBOL(remove_arg_zero);
1145
1146 /*
1147  * cycle the list of binary formats handler, until one recognizes the image
1148  */
1149 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1150 {
1151         int try,retval;
1152         struct linux_binfmt *fmt;
1153 #ifdef __alpha__
1154         /* handle /sbin/loader.. */
1155         {
1156             struct exec * eh = (struct exec *) bprm->buf;
1157
1158             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1159                 (eh->fh.f_flags & 0x3000) == 0x3000)
1160             {
1161                 struct file * file;
1162                 unsigned long loader;
1163
1164                 allow_write_access(bprm->file);
1165                 fput(bprm->file);
1166                 bprm->file = NULL;
1167
1168                 loader = bprm->vma->vm_end - sizeof(void *);
1169
1170                 file = open_exec("/sbin/loader");
1171                 retval = PTR_ERR(file);
1172                 if (IS_ERR(file))
1173                         return retval;
1174
1175                 /* Remember if the application is TASO.  */
1176                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1177
1178                 bprm->file = file;
1179                 bprm->loader = loader;
1180                 retval = prepare_binprm(bprm);
1181                 if (retval<0)
1182                         return retval;
1183                 /* should call search_binary_handler recursively here,
1184                    but it does not matter */
1185             }
1186         }
1187 #endif
1188         retval = security_bprm_check(bprm);
1189         if (retval)
1190                 return retval;
1191
1192         /* kernel module loader fixup */
1193         /* so we don't try to load run modprobe in kernel space. */
1194         set_fs(USER_DS);
1195
1196         retval = audit_bprm(bprm);
1197         if (retval)
1198                 return retval;
1199
1200         retval = -ENOENT;
1201         for (try=0; try<2; try++) {
1202                 read_lock(&binfmt_lock);
1203                 list_for_each_entry(fmt, &formats, lh) {
1204                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1205                         if (!fn)
1206                                 continue;
1207                         if (!try_module_get(fmt->module))
1208                                 continue;
1209                         read_unlock(&binfmt_lock);
1210                         retval = fn(bprm, regs);
1211                         if (retval >= 0) {
1212                                 tracehook_report_exec(fmt, bprm, regs);
1213                                 put_binfmt(fmt);
1214                                 allow_write_access(bprm->file);
1215                                 if (bprm->file)
1216                                         fput(bprm->file);
1217                                 bprm->file = NULL;
1218                                 current->did_exec = 1;
1219                                 proc_exec_connector(current);
1220                                 return retval;
1221                         }
1222                         read_lock(&binfmt_lock);
1223                         put_binfmt(fmt);
1224                         if (retval != -ENOEXEC || bprm->mm == NULL)
1225                                 break;
1226                         if (!bprm->file) {
1227                                 read_unlock(&binfmt_lock);
1228                                 return retval;
1229                         }
1230                 }
1231                 read_unlock(&binfmt_lock);
1232                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1233                         break;
1234 #ifdef CONFIG_KMOD
1235                 }else{
1236 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1237                         if (printable(bprm->buf[0]) &&
1238                             printable(bprm->buf[1]) &&
1239                             printable(bprm->buf[2]) &&
1240                             printable(bprm->buf[3]))
1241                                 break; /* -ENOEXEC */
1242                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1243 #endif
1244                 }
1245         }
1246         return retval;
1247 }
1248
1249 EXPORT_SYMBOL(search_binary_handler);
1250
1251 void free_bprm(struct linux_binprm *bprm)
1252 {
1253         free_arg_pages(bprm);
1254         kfree(bprm);
1255 }
1256
1257 /*
1258  * sys_execve() executes a new program.
1259  */
1260 int do_execve(char * filename,
1261         char __user *__user *argv,
1262         char __user *__user *envp,
1263         struct pt_regs * regs)
1264 {
1265         struct linux_binprm *bprm;
1266         struct file *file;
1267         struct files_struct *displaced;
1268         int retval;
1269
1270         retval = unshare_files(&displaced);
1271         if (retval)
1272                 goto out_ret;
1273
1274         retval = -ENOMEM;
1275         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1276         if (!bprm)
1277                 goto out_files;
1278
1279         file = open_exec(filename);
1280         retval = PTR_ERR(file);
1281         if (IS_ERR(file))
1282                 goto out_kfree;
1283
1284         sched_exec();
1285
1286         bprm->file = file;
1287         bprm->filename = filename;
1288         bprm->interp = filename;
1289
1290         retval = bprm_mm_init(bprm);
1291         if (retval)
1292                 goto out_file;
1293
1294         bprm->argc = count(argv, MAX_ARG_STRINGS);
1295         if ((retval = bprm->argc) < 0)
1296                 goto out_mm;
1297
1298         bprm->envc = count(envp, MAX_ARG_STRINGS);
1299         if ((retval = bprm->envc) < 0)
1300                 goto out_mm;
1301
1302         retval = security_bprm_alloc(bprm);
1303         if (retval)
1304                 goto out;
1305
1306         retval = prepare_binprm(bprm);
1307         if (retval < 0)
1308                 goto out;
1309
1310         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1311         if (retval < 0)
1312                 goto out;
1313
1314         bprm->exec = bprm->p;
1315         retval = copy_strings(bprm->envc, envp, bprm);
1316         if (retval < 0)
1317                 goto out;
1318
1319         retval = copy_strings(bprm->argc, argv, bprm);
1320         if (retval < 0)
1321                 goto out;
1322
1323         current->flags &= ~PF_KTHREAD;
1324         retval = search_binary_handler(bprm,regs);
1325         if (retval >= 0) {
1326                 /* execve success */
1327                 security_bprm_free(bprm);
1328                 acct_update_integrals(current);
1329                 free_bprm(bprm);
1330                 if (displaced)
1331                         put_files_struct(displaced);
1332                 return retval;
1333         }
1334
1335 out:
1336         if (bprm->security)
1337                 security_bprm_free(bprm);
1338
1339 out_mm:
1340         if (bprm->mm)
1341                 mmput (bprm->mm);
1342
1343 out_file:
1344         if (bprm->file) {
1345                 allow_write_access(bprm->file);
1346                 fput(bprm->file);
1347         }
1348 out_kfree:
1349         free_bprm(bprm);
1350
1351 out_files:
1352         if (displaced)
1353                 reset_files_struct(displaced);
1354 out_ret:
1355         return retval;
1356 }
1357
1358 int set_binfmt(struct linux_binfmt *new)
1359 {
1360         struct linux_binfmt *old = current->binfmt;
1361
1362         if (new) {
1363                 if (!try_module_get(new->module))
1364                         return -1;
1365         }
1366         current->binfmt = new;
1367         if (old)
1368                 module_put(old->module);
1369         return 0;
1370 }
1371
1372 EXPORT_SYMBOL(set_binfmt);
1373
1374 /* format_corename will inspect the pattern parameter, and output a
1375  * name into corename, which must have space for at least
1376  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1377  */
1378 static int format_corename(char *corename, int nr_threads, long signr)
1379 {
1380         const char *pat_ptr = core_pattern;
1381         int ispipe = (*pat_ptr == '|');
1382         char *out_ptr = corename;
1383         char *const out_end = corename + CORENAME_MAX_SIZE;
1384         int rc;
1385         int pid_in_pattern = 0;
1386
1387         /* Repeat as long as we have more pattern to process and more output
1388            space */
1389         while (*pat_ptr) {
1390                 if (*pat_ptr != '%') {
1391                         if (out_ptr == out_end)
1392                                 goto out;
1393                         *out_ptr++ = *pat_ptr++;
1394                 } else {
1395                         switch (*++pat_ptr) {
1396                         case 0:
1397                                 goto out;
1398                         /* Double percent, output one percent */
1399                         case '%':
1400                                 if (out_ptr == out_end)
1401                                         goto out;
1402                                 *out_ptr++ = '%';
1403                                 break;
1404                         /* pid */
1405                         case 'p':
1406                                 pid_in_pattern = 1;
1407                                 rc = snprintf(out_ptr, out_end - out_ptr,
1408                                               "%d", task_tgid_vnr(current));
1409                                 if (rc > out_end - out_ptr)
1410                                         goto out;
1411                                 out_ptr += rc;
1412                                 break;
1413                         /* uid */
1414                         case 'u':
1415                                 rc = snprintf(out_ptr, out_end - out_ptr,
1416                                               "%d", current->uid);
1417                                 if (rc > out_end - out_ptr)
1418                                         goto out;
1419                                 out_ptr += rc;
1420                                 break;
1421                         /* gid */
1422                         case 'g':
1423                                 rc = snprintf(out_ptr, out_end - out_ptr,
1424                                               "%d", current->gid);
1425                                 if (rc > out_end - out_ptr)
1426                                         goto out;
1427                                 out_ptr += rc;
1428                                 break;
1429                         /* signal that caused the coredump */
1430                         case 's':
1431                                 rc = snprintf(out_ptr, out_end - out_ptr,
1432                                               "%ld", signr);
1433                                 if (rc > out_end - out_ptr)
1434                                         goto out;
1435                                 out_ptr += rc;
1436                                 break;
1437                         /* UNIX time of coredump */
1438                         case 't': {
1439                                 struct timeval tv;
1440                                 do_gettimeofday(&tv);
1441                                 rc = snprintf(out_ptr, out_end - out_ptr,
1442                                               "%lu", tv.tv_sec);
1443                                 if (rc > out_end - out_ptr)
1444                                         goto out;
1445                                 out_ptr += rc;
1446                                 break;
1447                         }
1448                         /* hostname */
1449                         case 'h':
1450                                 down_read(&uts_sem);
1451                                 rc = snprintf(out_ptr, out_end - out_ptr,
1452                                               "%s", utsname()->nodename);
1453                                 up_read(&uts_sem);
1454                                 if (rc > out_end - out_ptr)
1455                                         goto out;
1456                                 out_ptr += rc;
1457                                 break;
1458                         /* executable */
1459                         case 'e':
1460                                 rc = snprintf(out_ptr, out_end - out_ptr,
1461                                               "%s", current->comm);
1462                                 if (rc > out_end - out_ptr)
1463                                         goto out;
1464                                 out_ptr += rc;
1465                                 break;
1466                         /* core limit size */
1467                         case 'c':
1468                                 rc = snprintf(out_ptr, out_end - out_ptr,
1469                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1470                                 if (rc > out_end - out_ptr)
1471                                         goto out;
1472                                 out_ptr += rc;
1473                                 break;
1474                         default:
1475                                 break;
1476                         }
1477                         ++pat_ptr;
1478                 }
1479         }
1480         /* Backward compatibility with core_uses_pid:
1481          *
1482          * If core_pattern does not include a %p (as is the default)
1483          * and core_uses_pid is set, then .%pid will be appended to
1484          * the filename. Do not do this for piped commands. */
1485         if (!ispipe && !pid_in_pattern
1486             && (core_uses_pid || nr_threads)) {
1487                 rc = snprintf(out_ptr, out_end - out_ptr,
1488                               ".%d", task_tgid_vnr(current));
1489                 if (rc > out_end - out_ptr)
1490                         goto out;
1491                 out_ptr += rc;
1492         }
1493 out:
1494         *out_ptr = 0;
1495         return ispipe;
1496 }
1497
1498 static int zap_process(struct task_struct *start)
1499 {
1500         struct task_struct *t;
1501         int nr = 0;
1502
1503         start->signal->flags = SIGNAL_GROUP_EXIT;
1504         start->signal->group_stop_count = 0;
1505
1506         t = start;
1507         do {
1508                 if (t != current && t->mm) {
1509                         sigaddset(&t->pending.signal, SIGKILL);
1510                         signal_wake_up(t, 1);
1511                         nr++;
1512                 }
1513         } while_each_thread(start, t);
1514
1515         return nr;
1516 }
1517
1518 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1519                                 struct core_state *core_state, int exit_code)
1520 {
1521         struct task_struct *g, *p;
1522         unsigned long flags;
1523         int nr = -EAGAIN;
1524
1525         spin_lock_irq(&tsk->sighand->siglock);
1526         if (!signal_group_exit(tsk->signal)) {
1527                 mm->core_state = core_state;
1528                 tsk->signal->group_exit_code = exit_code;
1529                 nr = zap_process(tsk);
1530         }
1531         spin_unlock_irq(&tsk->sighand->siglock);
1532         if (unlikely(nr < 0))
1533                 return nr;
1534
1535         if (atomic_read(&mm->mm_users) == nr + 1)
1536                 goto done;
1537         /*
1538          * We should find and kill all tasks which use this mm, and we should
1539          * count them correctly into ->nr_threads. We don't take tasklist
1540          * lock, but this is safe wrt:
1541          *
1542          * fork:
1543          *      None of sub-threads can fork after zap_process(leader). All
1544          *      processes which were created before this point should be
1545          *      visible to zap_threads() because copy_process() adds the new
1546          *      process to the tail of init_task.tasks list, and lock/unlock
1547          *      of ->siglock provides a memory barrier.
1548          *
1549          * do_exit:
1550          *      The caller holds mm->mmap_sem. This means that the task which
1551          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1552          *      its ->mm.
1553          *
1554          * de_thread:
1555          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1556          *      we must see either old or new leader, this does not matter.
1557          *      However, it can change p->sighand, so lock_task_sighand(p)
1558          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1559          *      it can't fail.
1560          *
1561          *      Note also that "g" can be the old leader with ->mm == NULL
1562          *      and already unhashed and thus removed from ->thread_group.
1563          *      This is OK, __unhash_process()->list_del_rcu() does not
1564          *      clear the ->next pointer, we will find the new leader via
1565          *      next_thread().
1566          */
1567         rcu_read_lock();
1568         for_each_process(g) {
1569                 if (g == tsk->group_leader)
1570                         continue;
1571                 if (g->flags & PF_KTHREAD)
1572                         continue;
1573                 p = g;
1574                 do {
1575                         if (p->mm) {
1576                                 if (unlikely(p->mm == mm)) {
1577                                         lock_task_sighand(p, &flags);
1578                                         nr += zap_process(p);
1579                                         unlock_task_sighand(p, &flags);
1580                                 }
1581                                 break;
1582                         }
1583                 } while_each_thread(g, p);
1584         }
1585         rcu_read_unlock();
1586 done:
1587         atomic_set(&core_state->nr_threads, nr);
1588         return nr;
1589 }
1590
1591 static int coredump_wait(int exit_code, struct core_state *core_state)
1592 {
1593         struct task_struct *tsk = current;
1594         struct mm_struct *mm = tsk->mm;
1595         struct completion *vfork_done;
1596         int core_waiters;
1597
1598         init_completion(&core_state->startup);
1599         core_state->dumper.task = tsk;
1600         core_state->dumper.next = NULL;
1601         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1602         up_write(&mm->mmap_sem);
1603
1604         if (unlikely(core_waiters < 0))
1605                 goto fail;
1606
1607         /*
1608          * Make sure nobody is waiting for us to release the VM,
1609          * otherwise we can deadlock when we wait on each other
1610          */
1611         vfork_done = tsk->vfork_done;
1612         if (vfork_done) {
1613                 tsk->vfork_done = NULL;
1614                 complete(vfork_done);
1615         }
1616
1617         if (core_waiters)
1618                 wait_for_completion(&core_state->startup);
1619 fail:
1620         return core_waiters;
1621 }
1622
1623 static void coredump_finish(struct mm_struct *mm)
1624 {
1625         struct core_thread *curr, *next;
1626         struct task_struct *task;
1627
1628         next = mm->core_state->dumper.next;
1629         while ((curr = next) != NULL) {
1630                 next = curr->next;
1631                 task = curr->task;
1632                 /*
1633                  * see exit_mm(), curr->task must not see
1634                  * ->task == NULL before we read ->next.
1635                  */
1636                 smp_mb();
1637                 curr->task = NULL;
1638                 wake_up_process(task);
1639         }
1640
1641         mm->core_state = NULL;
1642 }
1643
1644 /*
1645  * set_dumpable converts traditional three-value dumpable to two flags and
1646  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1647  * these bits are not changed atomically.  So get_dumpable can observe the
1648  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1649  * return either old dumpable or new one by paying attention to the order of
1650  * modifying the bits.
1651  *
1652  * dumpable |   mm->flags (binary)
1653  * old  new | initial interim  final
1654  * ---------+-----------------------
1655  *  0    1  |   00      01      01
1656  *  0    2  |   00      10(*)   11
1657  *  1    0  |   01      00      00
1658  *  1    2  |   01      11      11
1659  *  2    0  |   11      10(*)   00
1660  *  2    1  |   11      11      01
1661  *
1662  * (*) get_dumpable regards interim value of 10 as 11.
1663  */
1664 void set_dumpable(struct mm_struct *mm, int value)
1665 {
1666         switch (value) {
1667         case 0:
1668                 clear_bit(MMF_DUMPABLE, &mm->flags);
1669                 smp_wmb();
1670                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1671                 break;
1672         case 1:
1673                 set_bit(MMF_DUMPABLE, &mm->flags);
1674                 smp_wmb();
1675                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1676                 break;
1677         case 2:
1678                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1679                 smp_wmb();
1680                 set_bit(MMF_DUMPABLE, &mm->flags);
1681                 break;
1682         }
1683 }
1684
1685 int get_dumpable(struct mm_struct *mm)
1686 {
1687         int ret;
1688
1689         ret = mm->flags & 0x3;
1690         return (ret >= 2) ? 2 : ret;
1691 }
1692
1693 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1694 {
1695         struct core_state core_state;
1696         char corename[CORENAME_MAX_SIZE + 1];
1697         struct mm_struct *mm = current->mm;
1698         struct linux_binfmt * binfmt;
1699         struct inode * inode;
1700         struct file * file;
1701         int retval = 0;
1702         int fsuid = current->fsuid;
1703         int flag = 0;
1704         int ispipe = 0;
1705         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1706         char **helper_argv = NULL;
1707         int helper_argc = 0;
1708         char *delimit;
1709
1710         audit_core_dumps(signr);
1711
1712         binfmt = current->binfmt;
1713         if (!binfmt || !binfmt->core_dump)
1714                 goto fail;
1715         down_write(&mm->mmap_sem);
1716         /*
1717          * If another thread got here first, or we are not dumpable, bail out.
1718          */
1719         if (mm->core_state || !get_dumpable(mm)) {
1720                 up_write(&mm->mmap_sem);
1721                 goto fail;
1722         }
1723
1724         /*
1725          *      We cannot trust fsuid as being the "true" uid of the
1726          *      process nor do we know its entire history. We only know it
1727          *      was tainted so we dump it as root in mode 2.
1728          */
1729         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1730                 flag = O_EXCL;          /* Stop rewrite attacks */
1731                 current->fsuid = 0;     /* Dump root private */
1732         }
1733
1734         retval = coredump_wait(exit_code, &core_state);
1735         if (retval < 0)
1736                 goto fail;
1737
1738         /*
1739          * Clear any false indication of pending signals that might
1740          * be seen by the filesystem code called to write the core file.
1741          */
1742         clear_thread_flag(TIF_SIGPENDING);
1743
1744         /*
1745          * lock_kernel() because format_corename() is controlled by sysctl, which
1746          * uses lock_kernel()
1747          */
1748         lock_kernel();
1749         ispipe = format_corename(corename, retval, signr);
1750         unlock_kernel();
1751         /*
1752          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1753          * to a pipe.  Since we're not writing directly to the filesystem
1754          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1755          * created unless the pipe reader choses to write out the core file
1756          * at which point file size limits and permissions will be imposed
1757          * as it does with any other process
1758          */
1759         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1760                 goto fail_unlock;
1761
1762         if (ispipe) {
1763                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1764                 /* Terminate the string before the first option */
1765                 delimit = strchr(corename, ' ');
1766                 if (delimit)
1767                         *delimit = '\0';
1768                 delimit = strrchr(helper_argv[0], '/');
1769                 if (delimit)
1770                         delimit++;
1771                 else
1772                         delimit = helper_argv[0];
1773                 if (!strcmp(delimit, current->comm)) {
1774                         printk(KERN_NOTICE "Recursive core dump detected, "
1775                                         "aborting\n");
1776                         goto fail_unlock;
1777                 }
1778
1779                 core_limit = RLIM_INFINITY;
1780
1781                 /* SIGPIPE can happen, but it's just never processed */
1782                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1783                                 &file)) {
1784                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1785                                corename);
1786                         goto fail_unlock;
1787                 }
1788         } else
1789                 file = filp_open(corename,
1790                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1791                                  0600);
1792         if (IS_ERR(file))
1793                 goto fail_unlock;
1794         inode = file->f_path.dentry->d_inode;
1795         if (inode->i_nlink > 1)
1796                 goto close_fail;        /* multiple links - don't dump */
1797         if (!ispipe && d_unhashed(file->f_path.dentry))
1798                 goto close_fail;
1799
1800         /* AK: actually i see no reason to not allow this for named pipes etc.,
1801            but keep the previous behaviour for now. */
1802         if (!ispipe && !S_ISREG(inode->i_mode))
1803                 goto close_fail;
1804         /*
1805          * Dont allow local users get cute and trick others to coredump
1806          * into their pre-created files:
1807          */
1808         if (inode->i_uid != current->fsuid)
1809                 goto close_fail;
1810         if (!file->f_op)
1811                 goto close_fail;
1812         if (!file->f_op->write)
1813                 goto close_fail;
1814         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1815                 goto close_fail;
1816
1817         retval = binfmt->core_dump(signr, regs, file, core_limit);
1818
1819         if (retval)
1820                 current->signal->group_exit_code |= 0x80;
1821 close_fail:
1822         filp_close(file, NULL);
1823 fail_unlock:
1824         if (helper_argv)
1825                 argv_free(helper_argv);
1826
1827         current->fsuid = fsuid;
1828         coredump_finish(mm);
1829 fail:
1830         return retval;
1831 }