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