TPM: integrity fix
[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/ima.h>
49 #include <linux/syscalls.h>
50 #include <linux/tsacct_kern.h>
51 #include <linux/cn_proc.h>
52 #include <linux/audit.h>
53 #include <linux/tracehook.h>
54 #include <linux/kmod.h>
55
56 #include <asm/uaccess.h>
57 #include <asm/mmu_context.h>
58 #include <asm/tlb.h>
59 #include "internal.h"
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         char *tmp = getname(library);
112         int error = PTR_ERR(tmp);
113
114         if (!IS_ERR(tmp)) {
115                 error = path_lookup_open(AT_FDCWD, tmp,
116                                          LOOKUP_FOLLOW, &nd,
117                                          FMODE_READ|FMODE_EXEC);
118                 putname(tmp);
119         }
120         if (error)
121                 goto out;
122
123         error = -EINVAL;
124         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
125                 goto exit;
126
127         error = -EACCES;
128         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
129                 goto exit;
130
131         error = vfs_permission(&nd, MAY_READ | MAY_EXEC | MAY_OPEN);
132         if (error)
133                 goto exit;
134         error = ima_path_check(&nd.path, MAY_READ | MAY_EXEC | MAY_OPEN);
135         if (error)
136                 goto exit;
137
138         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
139         error = PTR_ERR(file);
140         if (IS_ERR(file))
141                 goto out;
142
143         error = -ENOEXEC;
144         if(file->f_op) {
145                 struct linux_binfmt * fmt;
146
147                 read_lock(&binfmt_lock);
148                 list_for_each_entry(fmt, &formats, lh) {
149                         if (!fmt->load_shlib)
150                                 continue;
151                         if (!try_module_get(fmt->module))
152                                 continue;
153                         read_unlock(&binfmt_lock);
154                         error = fmt->load_shlib(file);
155                         read_lock(&binfmt_lock);
156                         put_binfmt(fmt);
157                         if (error != -ENOEXEC)
158                                 break;
159                 }
160                 read_unlock(&binfmt_lock);
161         }
162         fput(file);
163 out:
164         return error;
165 exit:
166         release_open_intent(&nd);
167         path_put(&nd.path);
168         goto out;
169 }
170
171 #ifdef CONFIG_MMU
172
173 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
174                 int write)
175 {
176         struct page *page;
177         int ret;
178
179 #ifdef CONFIG_STACK_GROWSUP
180         if (write) {
181                 ret = expand_stack_downwards(bprm->vma, pos);
182                 if (ret < 0)
183                         return NULL;
184         }
185 #endif
186         ret = get_user_pages(current, bprm->mm, pos,
187                         1, write, 1, &page, NULL);
188         if (ret <= 0)
189                 return NULL;
190
191         if (write) {
192                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
193                 struct rlimit *rlim;
194
195                 /*
196                  * We've historically supported up to 32 pages (ARG_MAX)
197                  * of argument strings even with small stacks
198                  */
199                 if (size <= ARG_MAX)
200                         return page;
201
202                 /*
203                  * Limit to 1/4-th the stack size for the argv+env strings.
204                  * This ensures that:
205                  *  - the remaining binfmt code will not run out of stack space,
206                  *  - the program will have a reasonable amount of stack left
207                  *    to work from.
208                  */
209                 rlim = current->signal->rlim;
210                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
211                         put_page(page);
212                         return NULL;
213                 }
214         }
215
216         return page;
217 }
218
219 static void put_arg_page(struct page *page)
220 {
221         put_page(page);
222 }
223
224 static void free_arg_page(struct linux_binprm *bprm, int i)
225 {
226 }
227
228 static void free_arg_pages(struct linux_binprm *bprm)
229 {
230 }
231
232 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
233                 struct page *page)
234 {
235         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
236 }
237
238 static int __bprm_mm_init(struct linux_binprm *bprm)
239 {
240         int err = -ENOMEM;
241         struct vm_area_struct *vma = NULL;
242         struct mm_struct *mm = bprm->mm;
243
244         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
245         if (!vma)
246                 goto err;
247
248         down_write(&mm->mmap_sem);
249         vma->vm_mm = mm;
250
251         /*
252          * Place the stack at the largest stack address the architecture
253          * supports. Later, we'll move this to an appropriate place. We don't
254          * use STACK_TOP because that can depend on attributes which aren't
255          * configured yet.
256          */
257         vma->vm_end = STACK_TOP_MAX;
258         vma->vm_start = vma->vm_end - PAGE_SIZE;
259
260         vma->vm_flags = VM_STACK_FLAGS;
261         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
262         err = insert_vm_struct(mm, vma);
263         if (err) {
264                 up_write(&mm->mmap_sem);
265                 goto err;
266         }
267
268         mm->stack_vm = mm->total_vm = 1;
269         up_write(&mm->mmap_sem);
270
271         bprm->p = vma->vm_end - sizeof(void *);
272
273         return 0;
274
275 err:
276         if (vma) {
277                 bprm->vma = NULL;
278                 kmem_cache_free(vm_area_cachep, vma);
279         }
280
281         return err;
282 }
283
284 static bool valid_arg_len(struct linux_binprm *bprm, long len)
285 {
286         return len <= MAX_ARG_STRLEN;
287 }
288
289 #else
290
291 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
292                 int write)
293 {
294         struct page *page;
295
296         page = bprm->page[pos / PAGE_SIZE];
297         if (!page && write) {
298                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
299                 if (!page)
300                         return NULL;
301                 bprm->page[pos / PAGE_SIZE] = page;
302         }
303
304         return page;
305 }
306
307 static void put_arg_page(struct page *page)
308 {
309 }
310
311 static void free_arg_page(struct linux_binprm *bprm, int i)
312 {
313         if (bprm->page[i]) {
314                 __free_page(bprm->page[i]);
315                 bprm->page[i] = NULL;
316         }
317 }
318
319 static void free_arg_pages(struct linux_binprm *bprm)
320 {
321         int i;
322
323         for (i = 0; i < MAX_ARG_PAGES; i++)
324                 free_arg_page(bprm, i);
325 }
326
327 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
328                 struct page *page)
329 {
330 }
331
332 static int __bprm_mm_init(struct linux_binprm *bprm)
333 {
334         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
335         return 0;
336 }
337
338 static bool valid_arg_len(struct linux_binprm *bprm, long len)
339 {
340         return len <= bprm->p;
341 }
342
343 #endif /* CONFIG_MMU */
344
345 /*
346  * Create a new mm_struct and populate it with a temporary stack
347  * vm_area_struct.  We don't have enough context at this point to set the stack
348  * flags, permissions, and offset, so we use temporary values.  We'll update
349  * them later in setup_arg_pages().
350  */
351 int bprm_mm_init(struct linux_binprm *bprm)
352 {
353         int err;
354         struct mm_struct *mm = NULL;
355
356         bprm->mm = mm = mm_alloc();
357         err = -ENOMEM;
358         if (!mm)
359                 goto err;
360
361         err = init_new_context(current, mm);
362         if (err)
363                 goto err;
364
365         err = __bprm_mm_init(bprm);
366         if (err)
367                 goto err;
368
369         return 0;
370
371 err:
372         if (mm) {
373                 bprm->mm = NULL;
374                 mmdrop(mm);
375         }
376
377         return err;
378 }
379
380 /*
381  * count() counts the number of strings in array ARGV.
382  */
383 static int count(char __user * __user * argv, int max)
384 {
385         int i = 0;
386
387         if (argv != NULL) {
388                 for (;;) {
389                         char __user * p;
390
391                         if (get_user(p, argv))
392                                 return -EFAULT;
393                         if (!p)
394                                 break;
395                         argv++;
396                         if (i++ >= max)
397                                 return -E2BIG;
398                         cond_resched();
399                 }
400         }
401         return i;
402 }
403
404 /*
405  * 'copy_strings()' copies argument/environment strings from the old
406  * processes's memory to the new process's stack.  The call to get_user_pages()
407  * ensures the destination page is created and not swapped out.
408  */
409 static int copy_strings(int argc, char __user * __user * argv,
410                         struct linux_binprm *bprm)
411 {
412         struct page *kmapped_page = NULL;
413         char *kaddr = NULL;
414         unsigned long kpos = 0;
415         int ret;
416
417         while (argc-- > 0) {
418                 char __user *str;
419                 int len;
420                 unsigned long pos;
421
422                 if (get_user(str, argv+argc) ||
423                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
424                         ret = -EFAULT;
425                         goto out;
426                 }
427
428                 if (!valid_arg_len(bprm, len)) {
429                         ret = -E2BIG;
430                         goto out;
431                 }
432
433                 /* We're going to work our way backwords. */
434                 pos = bprm->p;
435                 str += len;
436                 bprm->p -= len;
437
438                 while (len > 0) {
439                         int offset, bytes_to_copy;
440
441                         offset = pos % PAGE_SIZE;
442                         if (offset == 0)
443                                 offset = PAGE_SIZE;
444
445                         bytes_to_copy = offset;
446                         if (bytes_to_copy > len)
447                                 bytes_to_copy = len;
448
449                         offset -= bytes_to_copy;
450                         pos -= bytes_to_copy;
451                         str -= bytes_to_copy;
452                         len -= bytes_to_copy;
453
454                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
455                                 struct page *page;
456
457                                 page = get_arg_page(bprm, pos, 1);
458                                 if (!page) {
459                                         ret = -E2BIG;
460                                         goto out;
461                                 }
462
463                                 if (kmapped_page) {
464                                         flush_kernel_dcache_page(kmapped_page);
465                                         kunmap(kmapped_page);
466                                         put_arg_page(kmapped_page);
467                                 }
468                                 kmapped_page = page;
469                                 kaddr = kmap(kmapped_page);
470                                 kpos = pos & PAGE_MASK;
471                                 flush_arg_page(bprm, kpos, kmapped_page);
472                         }
473                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
474                                 ret = -EFAULT;
475                                 goto out;
476                         }
477                 }
478         }
479         ret = 0;
480 out:
481         if (kmapped_page) {
482                 flush_kernel_dcache_page(kmapped_page);
483                 kunmap(kmapped_page);
484                 put_arg_page(kmapped_page);
485         }
486         return ret;
487 }
488
489 /*
490  * Like copy_strings, but get argv and its values from kernel memory.
491  */
492 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
493 {
494         int r;
495         mm_segment_t oldfs = get_fs();
496         set_fs(KERNEL_DS);
497         r = copy_strings(argc, (char __user * __user *)argv, bprm);
498         set_fs(oldfs);
499         return r;
500 }
501 EXPORT_SYMBOL(copy_strings_kernel);
502
503 #ifdef CONFIG_MMU
504
505 /*
506  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
507  * the binfmt code determines where the new stack should reside, we shift it to
508  * its final location.  The process proceeds as follows:
509  *
510  * 1) Use shift to calculate the new vma endpoints.
511  * 2) Extend vma to cover both the old and new ranges.  This ensures the
512  *    arguments passed to subsequent functions are consistent.
513  * 3) Move vma's page tables to the new range.
514  * 4) Free up any cleared pgd range.
515  * 5) Shrink the vma to cover only the new range.
516  */
517 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
518 {
519         struct mm_struct *mm = vma->vm_mm;
520         unsigned long old_start = vma->vm_start;
521         unsigned long old_end = vma->vm_end;
522         unsigned long length = old_end - old_start;
523         unsigned long new_start = old_start - shift;
524         unsigned long new_end = old_end - shift;
525         struct mmu_gather *tlb;
526
527         BUG_ON(new_start > new_end);
528
529         /*
530          * ensure there are no vmas between where we want to go
531          * and where we are
532          */
533         if (vma != find_vma(mm, new_start))
534                 return -EFAULT;
535
536         /*
537          * cover the whole range: [new_start, old_end)
538          */
539         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
540
541         /*
542          * move the page tables downwards, on failure we rely on
543          * process cleanup to remove whatever mess we made.
544          */
545         if (length != move_page_tables(vma, old_start,
546                                        vma, new_start, length))
547                 return -ENOMEM;
548
549         lru_add_drain();
550         tlb = tlb_gather_mmu(mm, 0);
551         if (new_end > old_start) {
552                 /*
553                  * when the old and new regions overlap clear from new_end.
554                  */
555                 free_pgd_range(tlb, new_end, old_end, new_end,
556                         vma->vm_next ? vma->vm_next->vm_start : 0);
557         } else {
558                 /*
559                  * otherwise, clean from old_start; this is done to not touch
560                  * the address space in [new_end, old_start) some architectures
561                  * have constraints on va-space that make this illegal (IA64) -
562                  * for the others its just a little faster.
563                  */
564                 free_pgd_range(tlb, old_start, old_end, new_end,
565                         vma->vm_next ? vma->vm_next->vm_start : 0);
566         }
567         tlb_finish_mmu(tlb, new_end, old_end);
568
569         /*
570          * shrink the vma to just the new range.
571          */
572         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
573
574         return 0;
575 }
576
577 #define EXTRA_STACK_VM_PAGES    20      /* random */
578
579 /*
580  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
581  * the stack is optionally relocated, and some extra space is added.
582  */
583 int setup_arg_pages(struct linux_binprm *bprm,
584                     unsigned long stack_top,
585                     int executable_stack)
586 {
587         unsigned long ret;
588         unsigned long stack_shift;
589         struct mm_struct *mm = current->mm;
590         struct vm_area_struct *vma = bprm->vma;
591         struct vm_area_struct *prev = NULL;
592         unsigned long vm_flags;
593         unsigned long stack_base;
594
595 #ifdef CONFIG_STACK_GROWSUP
596         /* Limit stack size to 1GB */
597         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
598         if (stack_base > (1 << 30))
599                 stack_base = 1 << 30;
600
601         /* Make sure we didn't let the argument array grow too large. */
602         if (vma->vm_end - vma->vm_start > stack_base)
603                 return -ENOMEM;
604
605         stack_base = PAGE_ALIGN(stack_top - stack_base);
606
607         stack_shift = vma->vm_start - stack_base;
608         mm->arg_start = bprm->p - stack_shift;
609         bprm->p = vma->vm_end - stack_shift;
610 #else
611         stack_top = arch_align_stack(stack_top);
612         stack_top = PAGE_ALIGN(stack_top);
613         stack_shift = vma->vm_end - stack_top;
614
615         bprm->p -= stack_shift;
616         mm->arg_start = bprm->p;
617 #endif
618
619         if (bprm->loader)
620                 bprm->loader -= stack_shift;
621         bprm->exec -= stack_shift;
622
623         down_write(&mm->mmap_sem);
624         vm_flags = VM_STACK_FLAGS;
625
626         /*
627          * Adjust stack execute permissions; explicitly enable for
628          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
629          * (arch default) otherwise.
630          */
631         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
632                 vm_flags |= VM_EXEC;
633         else if (executable_stack == EXSTACK_DISABLE_X)
634                 vm_flags &= ~VM_EXEC;
635         vm_flags |= mm->def_flags;
636
637         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
638                         vm_flags);
639         if (ret)
640                 goto out_unlock;
641         BUG_ON(prev != vma);
642
643         /* Move stack pages down in memory. */
644         if (stack_shift) {
645                 ret = shift_arg_pages(vma, stack_shift);
646                 if (ret) {
647                         up_write(&mm->mmap_sem);
648                         return ret;
649                 }
650         }
651
652 #ifdef CONFIG_STACK_GROWSUP
653         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
654 #else
655         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
656 #endif
657         ret = expand_stack(vma, stack_base);
658         if (ret)
659                 ret = -EFAULT;
660
661 out_unlock:
662         up_write(&mm->mmap_sem);
663         return 0;
664 }
665 EXPORT_SYMBOL(setup_arg_pages);
666
667 #endif /* CONFIG_MMU */
668
669 struct file *open_exec(const char *name)
670 {
671         struct nameidata nd;
672         struct file *file;
673         int err;
674
675         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
676                                 FMODE_READ|FMODE_EXEC);
677         if (err)
678                 goto out;
679
680         err = -EACCES;
681         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
682                 goto out_path_put;
683
684         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
685                 goto out_path_put;
686
687         err = vfs_permission(&nd, MAY_EXEC | MAY_OPEN);
688         if (err)
689                 goto out_path_put;
690         err = ima_path_check(&nd.path, MAY_EXEC | MAY_OPEN);
691         if (err)
692                 goto out_path_put;
693
694         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
695         if (IS_ERR(file))
696                 return file;
697
698         err = deny_write_access(file);
699         if (err) {
700                 fput(file);
701                 goto out;
702         }
703
704         return file;
705
706  out_path_put:
707         release_open_intent(&nd);
708         path_put(&nd.path);
709  out:
710         return ERR_PTR(err);
711 }
712 EXPORT_SYMBOL(open_exec);
713
714 int kernel_read(struct file *file, unsigned long offset,
715         char *addr, unsigned long count)
716 {
717         mm_segment_t old_fs;
718         loff_t pos = offset;
719         int result;
720
721         old_fs = get_fs();
722         set_fs(get_ds());
723         /* The cast to a user pointer is valid due to the set_fs() */
724         result = vfs_read(file, (void __user *)addr, count, &pos);
725         set_fs(old_fs);
726         return result;
727 }
728
729 EXPORT_SYMBOL(kernel_read);
730
731 static int exec_mmap(struct mm_struct *mm)
732 {
733         struct task_struct *tsk;
734         struct mm_struct * old_mm, *active_mm;
735
736         /* Notify parent that we're no longer interested in the old VM */
737         tsk = current;
738         old_mm = current->mm;
739         mm_release(tsk, old_mm);
740
741         if (old_mm) {
742                 /*
743                  * Make sure that if there is a core dump in progress
744                  * for the old mm, we get out and die instead of going
745                  * through with the exec.  We must hold mmap_sem around
746                  * checking core_state and changing tsk->mm.
747                  */
748                 down_read(&old_mm->mmap_sem);
749                 if (unlikely(old_mm->core_state)) {
750                         up_read(&old_mm->mmap_sem);
751                         return -EINTR;
752                 }
753         }
754         task_lock(tsk);
755         active_mm = tsk->active_mm;
756         tsk->mm = mm;
757         tsk->active_mm = mm;
758         activate_mm(active_mm, mm);
759         task_unlock(tsk);
760         arch_pick_mmap_layout(mm);
761         if (old_mm) {
762                 up_read(&old_mm->mmap_sem);
763                 BUG_ON(active_mm != old_mm);
764                 mm_update_next_owner(old_mm);
765                 mmput(old_mm);
766                 return 0;
767         }
768         mmdrop(active_mm);
769         return 0;
770 }
771
772 /*
773  * This function makes sure the current process has its own signal table,
774  * so that flush_signal_handlers can later reset the handlers without
775  * disturbing other processes.  (Other processes might share the signal
776  * table via the CLONE_SIGHAND option to clone().)
777  */
778 static int de_thread(struct task_struct *tsk)
779 {
780         struct signal_struct *sig = tsk->signal;
781         struct sighand_struct *oldsighand = tsk->sighand;
782         spinlock_t *lock = &oldsighand->siglock;
783         int count;
784
785         if (thread_group_empty(tsk))
786                 goto no_thread_group;
787
788         /*
789          * Kill all other threads in the thread group.
790          */
791         spin_lock_irq(lock);
792         if (signal_group_exit(sig)) {
793                 /*
794                  * Another group action in progress, just
795                  * return so that the signal is processed.
796                  */
797                 spin_unlock_irq(lock);
798                 return -EAGAIN;
799         }
800         sig->group_exit_task = tsk;
801         zap_other_threads(tsk);
802
803         /* Account for the thread group leader hanging around: */
804         count = thread_group_leader(tsk) ? 1 : 2;
805         sig->notify_count = count;
806         while (atomic_read(&sig->count) > count) {
807                 __set_current_state(TASK_UNINTERRUPTIBLE);
808                 spin_unlock_irq(lock);
809                 schedule();
810                 spin_lock_irq(lock);
811         }
812         spin_unlock_irq(lock);
813
814         /*
815          * At this point all other threads have exited, all we have to
816          * do is to wait for the thread group leader to become inactive,
817          * and to assume its PID:
818          */
819         if (!thread_group_leader(tsk)) {
820                 struct task_struct *leader = tsk->group_leader;
821
822                 sig->notify_count = -1; /* for exit_notify() */
823                 for (;;) {
824                         write_lock_irq(&tasklist_lock);
825                         if (likely(leader->exit_state))
826                                 break;
827                         __set_current_state(TASK_UNINTERRUPTIBLE);
828                         write_unlock_irq(&tasklist_lock);
829                         schedule();
830                 }
831
832                 /*
833                  * The only record we have of the real-time age of a
834                  * process, regardless of execs it's done, is start_time.
835                  * All the past CPU time is accumulated in signal_struct
836                  * from sister threads now dead.  But in this non-leader
837                  * exec, nothing survives from the original leader thread,
838                  * whose birth marks the true age of this process now.
839                  * When we take on its identity by switching to its PID, we
840                  * also take its birthdate (always earlier than our own).
841                  */
842                 tsk->start_time = leader->start_time;
843
844                 BUG_ON(!same_thread_group(leader, tsk));
845                 BUG_ON(has_group_leader_pid(tsk));
846                 /*
847                  * An exec() starts a new thread group with the
848                  * TGID of the previous thread group. Rehash the
849                  * two threads with a switched PID, and release
850                  * the former thread group leader:
851                  */
852
853                 /* Become a process group leader with the old leader's pid.
854                  * The old leader becomes a thread of the this thread group.
855                  * Note: The old leader also uses this pid until release_task
856                  *       is called.  Odd but simple and correct.
857                  */
858                 detach_pid(tsk, PIDTYPE_PID);
859                 tsk->pid = leader->pid;
860                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
861                 transfer_pid(leader, tsk, PIDTYPE_PGID);
862                 transfer_pid(leader, tsk, PIDTYPE_SID);
863                 list_replace_rcu(&leader->tasks, &tsk->tasks);
864
865                 tsk->group_leader = tsk;
866                 leader->group_leader = tsk;
867
868                 tsk->exit_signal = SIGCHLD;
869
870                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
871                 leader->exit_state = EXIT_DEAD;
872                 write_unlock_irq(&tasklist_lock);
873
874                 release_task(leader);
875         }
876
877         sig->group_exit_task = NULL;
878         sig->notify_count = 0;
879
880 no_thread_group:
881         exit_itimers(sig);
882         flush_itimer_signals();
883
884         if (atomic_read(&oldsighand->count) != 1) {
885                 struct sighand_struct *newsighand;
886                 /*
887                  * This ->sighand is shared with the CLONE_SIGHAND
888                  * but not CLONE_THREAD task, switch to the new one.
889                  */
890                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
891                 if (!newsighand)
892                         return -ENOMEM;
893
894                 atomic_set(&newsighand->count, 1);
895                 memcpy(newsighand->action, oldsighand->action,
896                        sizeof(newsighand->action));
897
898                 write_lock_irq(&tasklist_lock);
899                 spin_lock(&oldsighand->siglock);
900                 rcu_assign_pointer(tsk->sighand, newsighand);
901                 spin_unlock(&oldsighand->siglock);
902                 write_unlock_irq(&tasklist_lock);
903
904                 __cleanup_sighand(oldsighand);
905         }
906
907         BUG_ON(!thread_group_leader(tsk));
908         return 0;
909 }
910
911 /*
912  * These functions flushes out all traces of the currently running executable
913  * so that a new one can be started
914  */
915 static void flush_old_files(struct files_struct * files)
916 {
917         long j = -1;
918         struct fdtable *fdt;
919
920         spin_lock(&files->file_lock);
921         for (;;) {
922                 unsigned long set, i;
923
924                 j++;
925                 i = j * __NFDBITS;
926                 fdt = files_fdtable(files);
927                 if (i >= fdt->max_fds)
928                         break;
929                 set = fdt->close_on_exec->fds_bits[j];
930                 if (!set)
931                         continue;
932                 fdt->close_on_exec->fds_bits[j] = 0;
933                 spin_unlock(&files->file_lock);
934                 for ( ; set ; i++,set >>= 1) {
935                         if (set & 1) {
936                                 sys_close(i);
937                         }
938                 }
939                 spin_lock(&files->file_lock);
940
941         }
942         spin_unlock(&files->file_lock);
943 }
944
945 char *get_task_comm(char *buf, struct task_struct *tsk)
946 {
947         /* buf must be at least sizeof(tsk->comm) in size */
948         task_lock(tsk);
949         strncpy(buf, tsk->comm, sizeof(tsk->comm));
950         task_unlock(tsk);
951         return buf;
952 }
953
954 void set_task_comm(struct task_struct *tsk, char *buf)
955 {
956         task_lock(tsk);
957         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
958         task_unlock(tsk);
959 }
960
961 int flush_old_exec(struct linux_binprm * bprm)
962 {
963         char * name;
964         int i, ch, retval;
965         char tcomm[sizeof(current->comm)];
966
967         /*
968          * Make sure we have a private signal table and that
969          * we are unassociated from the previous thread group.
970          */
971         retval = de_thread(current);
972         if (retval)
973                 goto out;
974
975         set_mm_exe_file(bprm->mm, bprm->file);
976
977         /*
978          * Release all of the old mmap stuff
979          */
980         retval = exec_mmap(bprm->mm);
981         if (retval)
982                 goto out;
983
984         bprm->mm = NULL;                /* We're using it now */
985
986         /* This is the point of no return */
987         current->sas_ss_sp = current->sas_ss_size = 0;
988
989         if (current_euid() == current_uid() && current_egid() == current_gid())
990                 set_dumpable(current->mm, 1);
991         else
992                 set_dumpable(current->mm, suid_dumpable);
993
994         name = bprm->filename;
995
996         /* Copies the binary name from after last slash */
997         for (i=0; (ch = *(name++)) != '\0';) {
998                 if (ch == '/')
999                         i = 0; /* overwrite what we wrote */
1000                 else
1001                         if (i < (sizeof(tcomm) - 1))
1002                                 tcomm[i++] = ch;
1003         }
1004         tcomm[i] = '\0';
1005         set_task_comm(current, tcomm);
1006
1007         current->flags &= ~PF_RANDOMIZE;
1008         flush_thread();
1009
1010         /* Set the new mm task size. We have to do that late because it may
1011          * depend on TIF_32BIT which is only updated in flush_thread() on
1012          * some architectures like powerpc
1013          */
1014         current->mm->task_size = TASK_SIZE;
1015
1016         /* install the new credentials */
1017         if (bprm->cred->uid != current_euid() ||
1018             bprm->cred->gid != current_egid()) {
1019                 current->pdeath_signal = 0;
1020         } else if (file_permission(bprm->file, MAY_READ) ||
1021                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1022                 set_dumpable(current->mm, suid_dumpable);
1023         }
1024
1025         current->personality &= ~bprm->per_clear;
1026
1027         /* An exec changes our domain. We are no longer part of the thread
1028            group */
1029
1030         current->self_exec_id++;
1031                         
1032         flush_signal_handlers(current, 0);
1033         flush_old_files(current->files);
1034
1035         return 0;
1036
1037 out:
1038         return retval;
1039 }
1040
1041 EXPORT_SYMBOL(flush_old_exec);
1042
1043 /*
1044  * install the new credentials for this executable
1045  */
1046 void install_exec_creds(struct linux_binprm *bprm)
1047 {
1048         security_bprm_committing_creds(bprm);
1049
1050         commit_creds(bprm->cred);
1051         bprm->cred = NULL;
1052
1053         /* cred_exec_mutex must be held at least to this point to prevent
1054          * ptrace_attach() from altering our determination of the task's
1055          * credentials; any time after this it may be unlocked */
1056
1057         security_bprm_committed_creds(bprm);
1058 }
1059 EXPORT_SYMBOL(install_exec_creds);
1060
1061 /*
1062  * determine how safe it is to execute the proposed program
1063  * - the caller must hold current->cred_exec_mutex to protect against
1064  *   PTRACE_ATTACH
1065  */
1066 void check_unsafe_exec(struct linux_binprm *bprm)
1067 {
1068         struct task_struct *p = current;
1069
1070         bprm->unsafe = tracehook_unsafe_exec(p);
1071
1072         if (atomic_read(&p->fs->count) > 1 ||
1073             atomic_read(&p->files->count) > 1 ||
1074             atomic_read(&p->sighand->count) > 1)
1075                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1076 }
1077
1078 /* 
1079  * Fill the binprm structure from the inode. 
1080  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1081  *
1082  * This may be called multiple times for binary chains (scripts for example).
1083  */
1084 int prepare_binprm(struct linux_binprm *bprm)
1085 {
1086         umode_t mode;
1087         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1088         int retval;
1089
1090         mode = inode->i_mode;
1091         if (bprm->file->f_op == NULL)
1092                 return -EACCES;
1093
1094         /* clear any previous set[ug]id data from a previous binary */
1095         bprm->cred->euid = current_euid();
1096         bprm->cred->egid = current_egid();
1097
1098         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1099                 /* Set-uid? */
1100                 if (mode & S_ISUID) {
1101                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1102                         bprm->cred->euid = inode->i_uid;
1103                 }
1104
1105                 /* Set-gid? */
1106                 /*
1107                  * If setgid is set but no group execute bit then this
1108                  * is a candidate for mandatory locking, not a setgid
1109                  * executable.
1110                  */
1111                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1112                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1113                         bprm->cred->egid = inode->i_gid;
1114                 }
1115         }
1116
1117         /* fill in binprm security blob */
1118         retval = security_bprm_set_creds(bprm);
1119         if (retval)
1120                 return retval;
1121         bprm->cred_prepared = 1;
1122
1123         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1124         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1125 }
1126
1127 EXPORT_SYMBOL(prepare_binprm);
1128
1129 /*
1130  * Arguments are '\0' separated strings found at the location bprm->p
1131  * points to; chop off the first by relocating brpm->p to right after
1132  * the first '\0' encountered.
1133  */
1134 int remove_arg_zero(struct linux_binprm *bprm)
1135 {
1136         int ret = 0;
1137         unsigned long offset;
1138         char *kaddr;
1139         struct page *page;
1140
1141         if (!bprm->argc)
1142                 return 0;
1143
1144         do {
1145                 offset = bprm->p & ~PAGE_MASK;
1146                 page = get_arg_page(bprm, bprm->p, 0);
1147                 if (!page) {
1148                         ret = -EFAULT;
1149                         goto out;
1150                 }
1151                 kaddr = kmap_atomic(page, KM_USER0);
1152
1153                 for (; offset < PAGE_SIZE && kaddr[offset];
1154                                 offset++, bprm->p++)
1155                         ;
1156
1157                 kunmap_atomic(kaddr, KM_USER0);
1158                 put_arg_page(page);
1159
1160                 if (offset == PAGE_SIZE)
1161                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1162         } while (offset == PAGE_SIZE);
1163
1164         bprm->p++;
1165         bprm->argc--;
1166         ret = 0;
1167
1168 out:
1169         return ret;
1170 }
1171 EXPORT_SYMBOL(remove_arg_zero);
1172
1173 /*
1174  * cycle the list of binary formats handler, until one recognizes the image
1175  */
1176 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1177 {
1178         unsigned int depth = bprm->recursion_depth;
1179         int try,retval;
1180         struct linux_binfmt *fmt;
1181 #ifdef __alpha__
1182         /* handle /sbin/loader.. */
1183         {
1184             struct exec * eh = (struct exec *) bprm->buf;
1185
1186             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1187                 (eh->fh.f_flags & 0x3000) == 0x3000)
1188             {
1189                 struct file * file;
1190                 unsigned long loader;
1191
1192                 allow_write_access(bprm->file);
1193                 fput(bprm->file);
1194                 bprm->file = NULL;
1195
1196                 loader = bprm->vma->vm_end - sizeof(void *);
1197
1198                 file = open_exec("/sbin/loader");
1199                 retval = PTR_ERR(file);
1200                 if (IS_ERR(file))
1201                         return retval;
1202
1203                 /* Remember if the application is TASO.  */
1204                 bprm->taso = eh->ah.entry < 0x100000000UL;
1205
1206                 bprm->file = file;
1207                 bprm->loader = loader;
1208                 retval = prepare_binprm(bprm);
1209                 if (retval<0)
1210                         return retval;
1211                 /* should call search_binary_handler recursively here,
1212                    but it does not matter */
1213             }
1214         }
1215 #endif
1216         retval = security_bprm_check(bprm);
1217         if (retval)
1218                 return retval;
1219         retval = ima_bprm_check(bprm);
1220         if (retval)
1221                 return retval;
1222
1223         /* kernel module loader fixup */
1224         /* so we don't try to load run modprobe in kernel space. */
1225         set_fs(USER_DS);
1226
1227         retval = audit_bprm(bprm);
1228         if (retval)
1229                 return retval;
1230
1231         retval = -ENOENT;
1232         for (try=0; try<2; try++) {
1233                 read_lock(&binfmt_lock);
1234                 list_for_each_entry(fmt, &formats, lh) {
1235                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1236                         if (!fn)
1237                                 continue;
1238                         if (!try_module_get(fmt->module))
1239                                 continue;
1240                         read_unlock(&binfmt_lock);
1241                         retval = fn(bprm, regs);
1242                         /*
1243                          * Restore the depth counter to its starting value
1244                          * in this call, so we don't have to rely on every
1245                          * load_binary function to restore it on return.
1246                          */
1247                         bprm->recursion_depth = depth;
1248                         if (retval >= 0) {
1249                                 if (depth == 0)
1250                                         tracehook_report_exec(fmt, bprm, regs);
1251                                 put_binfmt(fmt);
1252                                 allow_write_access(bprm->file);
1253                                 if (bprm->file)
1254                                         fput(bprm->file);
1255                                 bprm->file = NULL;
1256                                 current->did_exec = 1;
1257                                 proc_exec_connector(current);
1258                                 return retval;
1259                         }
1260                         read_lock(&binfmt_lock);
1261                         put_binfmt(fmt);
1262                         if (retval != -ENOEXEC || bprm->mm == NULL)
1263                                 break;
1264                         if (!bprm->file) {
1265                                 read_unlock(&binfmt_lock);
1266                                 return retval;
1267                         }
1268                 }
1269                 read_unlock(&binfmt_lock);
1270                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1271                         break;
1272 #ifdef CONFIG_MODULES
1273                 } else {
1274 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1275                         if (printable(bprm->buf[0]) &&
1276                             printable(bprm->buf[1]) &&
1277                             printable(bprm->buf[2]) &&
1278                             printable(bprm->buf[3]))
1279                                 break; /* -ENOEXEC */
1280                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1281 #endif
1282                 }
1283         }
1284         return retval;
1285 }
1286
1287 EXPORT_SYMBOL(search_binary_handler);
1288
1289 void free_bprm(struct linux_binprm *bprm)
1290 {
1291         free_arg_pages(bprm);
1292         if (bprm->cred)
1293                 abort_creds(bprm->cred);
1294         kfree(bprm);
1295 }
1296
1297 /*
1298  * sys_execve() executes a new program.
1299  */
1300 int do_execve(char * filename,
1301         char __user *__user *argv,
1302         char __user *__user *envp,
1303         struct pt_regs * regs)
1304 {
1305         struct linux_binprm *bprm;
1306         struct file *file;
1307         struct files_struct *displaced;
1308         int retval;
1309
1310         retval = unshare_files(&displaced);
1311         if (retval)
1312                 goto out_ret;
1313
1314         retval = -ENOMEM;
1315         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1316         if (!bprm)
1317                 goto out_files;
1318
1319         retval = mutex_lock_interruptible(&current->cred_exec_mutex);
1320         if (retval < 0)
1321                 goto out_free;
1322
1323         retval = -ENOMEM;
1324         bprm->cred = prepare_exec_creds();
1325         if (!bprm->cred)
1326                 goto out_unlock;
1327         check_unsafe_exec(bprm);
1328
1329         file = open_exec(filename);
1330         retval = PTR_ERR(file);
1331         if (IS_ERR(file))
1332                 goto out_unlock;
1333
1334         sched_exec();
1335
1336         bprm->file = file;
1337         bprm->filename = filename;
1338         bprm->interp = filename;
1339
1340         retval = bprm_mm_init(bprm);
1341         if (retval)
1342                 goto out_file;
1343
1344         bprm->argc = count(argv, MAX_ARG_STRINGS);
1345         if ((retval = bprm->argc) < 0)
1346                 goto out;
1347
1348         bprm->envc = count(envp, MAX_ARG_STRINGS);
1349         if ((retval = bprm->envc) < 0)
1350                 goto out;
1351
1352         retval = prepare_binprm(bprm);
1353         if (retval < 0)
1354                 goto out;
1355
1356         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1357         if (retval < 0)
1358                 goto out;
1359
1360         bprm->exec = bprm->p;
1361         retval = copy_strings(bprm->envc, envp, bprm);
1362         if (retval < 0)
1363                 goto out;
1364
1365         retval = copy_strings(bprm->argc, argv, bprm);
1366         if (retval < 0)
1367                 goto out;
1368
1369         current->flags &= ~PF_KTHREAD;
1370         retval = search_binary_handler(bprm,regs);
1371         if (retval < 0)
1372                 goto out;
1373
1374         /* execve succeeded */
1375         mutex_unlock(&current->cred_exec_mutex);
1376         acct_update_integrals(current);
1377         free_bprm(bprm);
1378         if (displaced)
1379                 put_files_struct(displaced);
1380         return retval;
1381
1382 out:
1383         if (bprm->mm)
1384                 mmput (bprm->mm);
1385
1386 out_file:
1387         if (bprm->file) {
1388                 allow_write_access(bprm->file);
1389                 fput(bprm->file);
1390         }
1391
1392 out_unlock:
1393         mutex_unlock(&current->cred_exec_mutex);
1394
1395 out_free:
1396         free_bprm(bprm);
1397
1398 out_files:
1399         if (displaced)
1400                 reset_files_struct(displaced);
1401 out_ret:
1402         return retval;
1403 }
1404
1405 int set_binfmt(struct linux_binfmt *new)
1406 {
1407         struct linux_binfmt *old = current->binfmt;
1408
1409         if (new) {
1410                 if (!try_module_get(new->module))
1411                         return -1;
1412         }
1413         current->binfmt = new;
1414         if (old)
1415                 module_put(old->module);
1416         return 0;
1417 }
1418
1419 EXPORT_SYMBOL(set_binfmt);
1420
1421 /* format_corename will inspect the pattern parameter, and output a
1422  * name into corename, which must have space for at least
1423  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1424  */
1425 static int format_corename(char *corename, long signr)
1426 {
1427         const struct cred *cred = current_cred();
1428         const char *pat_ptr = core_pattern;
1429         int ispipe = (*pat_ptr == '|');
1430         char *out_ptr = corename;
1431         char *const out_end = corename + CORENAME_MAX_SIZE;
1432         int rc;
1433         int pid_in_pattern = 0;
1434
1435         /* Repeat as long as we have more pattern to process and more output
1436            space */
1437         while (*pat_ptr) {
1438                 if (*pat_ptr != '%') {
1439                         if (out_ptr == out_end)
1440                                 goto out;
1441                         *out_ptr++ = *pat_ptr++;
1442                 } else {
1443                         switch (*++pat_ptr) {
1444                         case 0:
1445                                 goto out;
1446                         /* Double percent, output one percent */
1447                         case '%':
1448                                 if (out_ptr == out_end)
1449                                         goto out;
1450                                 *out_ptr++ = '%';
1451                                 break;
1452                         /* pid */
1453                         case 'p':
1454                                 pid_in_pattern = 1;
1455                                 rc = snprintf(out_ptr, out_end - out_ptr,
1456                                               "%d", task_tgid_vnr(current));
1457                                 if (rc > out_end - out_ptr)
1458                                         goto out;
1459                                 out_ptr += rc;
1460                                 break;
1461                         /* uid */
1462                         case 'u':
1463                                 rc = snprintf(out_ptr, out_end - out_ptr,
1464                                               "%d", cred->uid);
1465                                 if (rc > out_end - out_ptr)
1466                                         goto out;
1467                                 out_ptr += rc;
1468                                 break;
1469                         /* gid */
1470                         case 'g':
1471                                 rc = snprintf(out_ptr, out_end - out_ptr,
1472                                               "%d", cred->gid);
1473                                 if (rc > out_end - out_ptr)
1474                                         goto out;
1475                                 out_ptr += rc;
1476                                 break;
1477                         /* signal that caused the coredump */
1478                         case 's':
1479                                 rc = snprintf(out_ptr, out_end - out_ptr,
1480                                               "%ld", signr);
1481                                 if (rc > out_end - out_ptr)
1482                                         goto out;
1483                                 out_ptr += rc;
1484                                 break;
1485                         /* UNIX time of coredump */
1486                         case 't': {
1487                                 struct timeval tv;
1488                                 do_gettimeofday(&tv);
1489                                 rc = snprintf(out_ptr, out_end - out_ptr,
1490                                               "%lu", tv.tv_sec);
1491                                 if (rc > out_end - out_ptr)
1492                                         goto out;
1493                                 out_ptr += rc;
1494                                 break;
1495                         }
1496                         /* hostname */
1497                         case 'h':
1498                                 down_read(&uts_sem);
1499                                 rc = snprintf(out_ptr, out_end - out_ptr,
1500                                               "%s", utsname()->nodename);
1501                                 up_read(&uts_sem);
1502                                 if (rc > out_end - out_ptr)
1503                                         goto out;
1504                                 out_ptr += rc;
1505                                 break;
1506                         /* executable */
1507                         case 'e':
1508                                 rc = snprintf(out_ptr, out_end - out_ptr,
1509                                               "%s", current->comm);
1510                                 if (rc > out_end - out_ptr)
1511                                         goto out;
1512                                 out_ptr += rc;
1513                                 break;
1514                         /* core limit size */
1515                         case 'c':
1516                                 rc = snprintf(out_ptr, out_end - out_ptr,
1517                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1518                                 if (rc > out_end - out_ptr)
1519                                         goto out;
1520                                 out_ptr += rc;
1521                                 break;
1522                         default:
1523                                 break;
1524                         }
1525                         ++pat_ptr;
1526                 }
1527         }
1528         /* Backward compatibility with core_uses_pid:
1529          *
1530          * If core_pattern does not include a %p (as is the default)
1531          * and core_uses_pid is set, then .%pid will be appended to
1532          * the filename. Do not do this for piped commands. */
1533         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1534                 rc = snprintf(out_ptr, out_end - out_ptr,
1535                               ".%d", task_tgid_vnr(current));
1536                 if (rc > out_end - out_ptr)
1537                         goto out;
1538                 out_ptr += rc;
1539         }
1540 out:
1541         *out_ptr = 0;
1542         return ispipe;
1543 }
1544
1545 static int zap_process(struct task_struct *start)
1546 {
1547         struct task_struct *t;
1548         int nr = 0;
1549
1550         start->signal->flags = SIGNAL_GROUP_EXIT;
1551         start->signal->group_stop_count = 0;
1552
1553         t = start;
1554         do {
1555                 if (t != current && t->mm) {
1556                         sigaddset(&t->pending.signal, SIGKILL);
1557                         signal_wake_up(t, 1);
1558                         nr++;
1559                 }
1560         } while_each_thread(start, t);
1561
1562         return nr;
1563 }
1564
1565 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1566                                 struct core_state *core_state, int exit_code)
1567 {
1568         struct task_struct *g, *p;
1569         unsigned long flags;
1570         int nr = -EAGAIN;
1571
1572         spin_lock_irq(&tsk->sighand->siglock);
1573         if (!signal_group_exit(tsk->signal)) {
1574                 mm->core_state = core_state;
1575                 tsk->signal->group_exit_code = exit_code;
1576                 nr = zap_process(tsk);
1577         }
1578         spin_unlock_irq(&tsk->sighand->siglock);
1579         if (unlikely(nr < 0))
1580                 return nr;
1581
1582         if (atomic_read(&mm->mm_users) == nr + 1)
1583                 goto done;
1584         /*
1585          * We should find and kill all tasks which use this mm, and we should
1586          * count them correctly into ->nr_threads. We don't take tasklist
1587          * lock, but this is safe wrt:
1588          *
1589          * fork:
1590          *      None of sub-threads can fork after zap_process(leader). All
1591          *      processes which were created before this point should be
1592          *      visible to zap_threads() because copy_process() adds the new
1593          *      process to the tail of init_task.tasks list, and lock/unlock
1594          *      of ->siglock provides a memory barrier.
1595          *
1596          * do_exit:
1597          *      The caller holds mm->mmap_sem. This means that the task which
1598          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1599          *      its ->mm.
1600          *
1601          * de_thread:
1602          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1603          *      we must see either old or new leader, this does not matter.
1604          *      However, it can change p->sighand, so lock_task_sighand(p)
1605          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1606          *      it can't fail.
1607          *
1608          *      Note also that "g" can be the old leader with ->mm == NULL
1609          *      and already unhashed and thus removed from ->thread_group.
1610          *      This is OK, __unhash_process()->list_del_rcu() does not
1611          *      clear the ->next pointer, we will find the new leader via
1612          *      next_thread().
1613          */
1614         rcu_read_lock();
1615         for_each_process(g) {
1616                 if (g == tsk->group_leader)
1617                         continue;
1618                 if (g->flags & PF_KTHREAD)
1619                         continue;
1620                 p = g;
1621                 do {
1622                         if (p->mm) {
1623                                 if (unlikely(p->mm == mm)) {
1624                                         lock_task_sighand(p, &flags);
1625                                         nr += zap_process(p);
1626                                         unlock_task_sighand(p, &flags);
1627                                 }
1628                                 break;
1629                         }
1630                 } while_each_thread(g, p);
1631         }
1632         rcu_read_unlock();
1633 done:
1634         atomic_set(&core_state->nr_threads, nr);
1635         return nr;
1636 }
1637
1638 static int coredump_wait(int exit_code, struct core_state *core_state)
1639 {
1640         struct task_struct *tsk = current;
1641         struct mm_struct *mm = tsk->mm;
1642         struct completion *vfork_done;
1643         int core_waiters;
1644
1645         init_completion(&core_state->startup);
1646         core_state->dumper.task = tsk;
1647         core_state->dumper.next = NULL;
1648         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1649         up_write(&mm->mmap_sem);
1650
1651         if (unlikely(core_waiters < 0))
1652                 goto fail;
1653
1654         /*
1655          * Make sure nobody is waiting for us to release the VM,
1656          * otherwise we can deadlock when we wait on each other
1657          */
1658         vfork_done = tsk->vfork_done;
1659         if (vfork_done) {
1660                 tsk->vfork_done = NULL;
1661                 complete(vfork_done);
1662         }
1663
1664         if (core_waiters)
1665                 wait_for_completion(&core_state->startup);
1666 fail:
1667         return core_waiters;
1668 }
1669
1670 static void coredump_finish(struct mm_struct *mm)
1671 {
1672         struct core_thread *curr, *next;
1673         struct task_struct *task;
1674
1675         next = mm->core_state->dumper.next;
1676         while ((curr = next) != NULL) {
1677                 next = curr->next;
1678                 task = curr->task;
1679                 /*
1680                  * see exit_mm(), curr->task must not see
1681                  * ->task == NULL before we read ->next.
1682                  */
1683                 smp_mb();
1684                 curr->task = NULL;
1685                 wake_up_process(task);
1686         }
1687
1688         mm->core_state = NULL;
1689 }
1690
1691 /*
1692  * set_dumpable converts traditional three-value dumpable to two flags and
1693  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1694  * these bits are not changed atomically.  So get_dumpable can observe the
1695  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1696  * return either old dumpable or new one by paying attention to the order of
1697  * modifying the bits.
1698  *
1699  * dumpable |   mm->flags (binary)
1700  * old  new | initial interim  final
1701  * ---------+-----------------------
1702  *  0    1  |   00      01      01
1703  *  0    2  |   00      10(*)   11
1704  *  1    0  |   01      00      00
1705  *  1    2  |   01      11      11
1706  *  2    0  |   11      10(*)   00
1707  *  2    1  |   11      11      01
1708  *
1709  * (*) get_dumpable regards interim value of 10 as 11.
1710  */
1711 void set_dumpable(struct mm_struct *mm, int value)
1712 {
1713         switch (value) {
1714         case 0:
1715                 clear_bit(MMF_DUMPABLE, &mm->flags);
1716                 smp_wmb();
1717                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1718                 break;
1719         case 1:
1720                 set_bit(MMF_DUMPABLE, &mm->flags);
1721                 smp_wmb();
1722                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1723                 break;
1724         case 2:
1725                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1726                 smp_wmb();
1727                 set_bit(MMF_DUMPABLE, &mm->flags);
1728                 break;
1729         }
1730 }
1731
1732 int get_dumpable(struct mm_struct *mm)
1733 {
1734         int ret;
1735
1736         ret = mm->flags & 0x3;
1737         return (ret >= 2) ? 2 : ret;
1738 }
1739
1740 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1741 {
1742         struct core_state core_state;
1743         char corename[CORENAME_MAX_SIZE + 1];
1744         struct mm_struct *mm = current->mm;
1745         struct linux_binfmt * binfmt;
1746         struct inode * inode;
1747         struct file * file;
1748         const struct cred *old_cred;
1749         struct cred *cred;
1750         int retval = 0;
1751         int flag = 0;
1752         int ispipe = 0;
1753         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1754         char **helper_argv = NULL;
1755         int helper_argc = 0;
1756         char *delimit;
1757
1758         audit_core_dumps(signr);
1759
1760         binfmt = current->binfmt;
1761         if (!binfmt || !binfmt->core_dump)
1762                 goto fail;
1763
1764         cred = prepare_creds();
1765         if (!cred) {
1766                 retval = -ENOMEM;
1767                 goto fail;
1768         }
1769
1770         down_write(&mm->mmap_sem);
1771         /*
1772          * If another thread got here first, or we are not dumpable, bail out.
1773          */
1774         if (mm->core_state || !get_dumpable(mm)) {
1775                 up_write(&mm->mmap_sem);
1776                 put_cred(cred);
1777                 goto fail;
1778         }
1779
1780         /*
1781          *      We cannot trust fsuid as being the "true" uid of the
1782          *      process nor do we know its entire history. We only know it
1783          *      was tainted so we dump it as root in mode 2.
1784          */
1785         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1786                 flag = O_EXCL;          /* Stop rewrite attacks */
1787                 cred->fsuid = 0;        /* Dump root private */
1788         }
1789
1790         retval = coredump_wait(exit_code, &core_state);
1791         if (retval < 0) {
1792                 put_cred(cred);
1793                 goto fail;
1794         }
1795
1796         old_cred = override_creds(cred);
1797
1798         /*
1799          * Clear any false indication of pending signals that might
1800          * be seen by the filesystem code called to write the core file.
1801          */
1802         clear_thread_flag(TIF_SIGPENDING);
1803
1804         /*
1805          * lock_kernel() because format_corename() is controlled by sysctl, which
1806          * uses lock_kernel()
1807          */
1808         lock_kernel();
1809         ispipe = format_corename(corename, signr);
1810         unlock_kernel();
1811         /*
1812          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1813          * to a pipe.  Since we're not writing directly to the filesystem
1814          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1815          * created unless the pipe reader choses to write out the core file
1816          * at which point file size limits and permissions will be imposed
1817          * as it does with any other process
1818          */
1819         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1820                 goto fail_unlock;
1821
1822         if (ispipe) {
1823                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1824                 /* Terminate the string before the first option */
1825                 delimit = strchr(corename, ' ');
1826                 if (delimit)
1827                         *delimit = '\0';
1828                 delimit = strrchr(helper_argv[0], '/');
1829                 if (delimit)
1830                         delimit++;
1831                 else
1832                         delimit = helper_argv[0];
1833                 if (!strcmp(delimit, current->comm)) {
1834                         printk(KERN_NOTICE "Recursive core dump detected, "
1835                                         "aborting\n");
1836                         goto fail_unlock;
1837                 }
1838
1839                 core_limit = RLIM_INFINITY;
1840
1841                 /* SIGPIPE can happen, but it's just never processed */
1842                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1843                                 &file)) {
1844                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1845                                corename);
1846                         goto fail_unlock;
1847                 }
1848         } else
1849                 file = filp_open(corename,
1850                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1851                                  0600);
1852         if (IS_ERR(file))
1853                 goto fail_unlock;
1854         inode = file->f_path.dentry->d_inode;
1855         if (inode->i_nlink > 1)
1856                 goto close_fail;        /* multiple links - don't dump */
1857         if (!ispipe && d_unhashed(file->f_path.dentry))
1858                 goto close_fail;
1859
1860         /* AK: actually i see no reason to not allow this for named pipes etc.,
1861            but keep the previous behaviour for now. */
1862         if (!ispipe && !S_ISREG(inode->i_mode))
1863                 goto close_fail;
1864         /*
1865          * Dont allow local users get cute and trick others to coredump
1866          * into their pre-created files:
1867          */
1868         if (inode->i_uid != current_fsuid())
1869                 goto close_fail;
1870         if (!file->f_op)
1871                 goto close_fail;
1872         if (!file->f_op->write)
1873                 goto close_fail;
1874         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1875                 goto close_fail;
1876
1877         retval = binfmt->core_dump(signr, regs, file, core_limit);
1878
1879         if (retval)
1880                 current->signal->group_exit_code |= 0x80;
1881 close_fail:
1882         filp_close(file, NULL);
1883 fail_unlock:
1884         if (helper_argv)
1885                 argv_free(helper_argv);
1886
1887         revert_creds(old_cred);
1888         put_cred(cred);
1889         coredump_finish(mm);
1890 fail:
1891         return retval;
1892 }