thermal: re-document thermal units
[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/mman.h>
28 #include <linux/a.out.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/highmem.h>
36 #include <linux/spinlock.h>
37 #include <linux/key.h>
38 #include <linux/personality.h>
39 #include <linux/binfmts.h>
40 #include <linux/swap.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/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/rmap.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
57 #include <asm/tlb.h>
58
59 #ifdef CONFIG_KMOD
60 #include <linux/kmod.h>
61 #endif
62
63 int core_uses_pid;
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 int suid_dumpable = 0;
66
67 /* The maximal length of core_pattern is also specified in sysctl.c */
68
69 static LIST_HEAD(formats);
70 static DEFINE_RWLOCK(binfmt_lock);
71
72 int register_binfmt(struct linux_binfmt * fmt)
73 {
74         if (!fmt)
75                 return -EINVAL;
76         write_lock(&binfmt_lock);
77         list_add(&fmt->lh, &formats);
78         write_unlock(&binfmt_lock);
79         return 0;       
80 }
81
82 EXPORT_SYMBOL(register_binfmt);
83
84 void unregister_binfmt(struct linux_binfmt * fmt)
85 {
86         write_lock(&binfmt_lock);
87         list_del(&fmt->lh);
88         write_unlock(&binfmt_lock);
89 }
90
91 EXPORT_SYMBOL(unregister_binfmt);
92
93 static inline void put_binfmt(struct linux_binfmt * fmt)
94 {
95         module_put(fmt->module);
96 }
97
98 /*
99  * Note that a shared library must be both readable and executable due to
100  * security reasons.
101  *
102  * Also note that we take the address to load from from the file itself.
103  */
104 asmlinkage long sys_uselib(const char __user * library)
105 {
106         struct file * file;
107         struct nameidata nd;
108         int error;
109
110         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
111         if (error)
112                 goto out;
113
114         error = -EINVAL;
115         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
116                 goto exit;
117
118         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
119         if (error)
120                 goto exit;
121
122         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
123         error = PTR_ERR(file);
124         if (IS_ERR(file))
125                 goto out;
126
127         error = -ENOEXEC;
128         if(file->f_op) {
129                 struct linux_binfmt * fmt;
130
131                 read_lock(&binfmt_lock);
132                 list_for_each_entry(fmt, &formats, lh) {
133                         if (!fmt->load_shlib)
134                                 continue;
135                         if (!try_module_get(fmt->module))
136                                 continue;
137                         read_unlock(&binfmt_lock);
138                         error = fmt->load_shlib(file);
139                         read_lock(&binfmt_lock);
140                         put_binfmt(fmt);
141                         if (error != -ENOEXEC)
142                                 break;
143                 }
144                 read_unlock(&binfmt_lock);
145         }
146         fput(file);
147 out:
148         return error;
149 exit:
150         release_open_intent(&nd);
151         path_put(&nd.path);
152         goto out;
153 }
154
155 #ifdef CONFIG_MMU
156
157 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
158                 int write)
159 {
160         struct page *page;
161         int ret;
162
163 #ifdef CONFIG_STACK_GROWSUP
164         if (write) {
165                 ret = expand_stack_downwards(bprm->vma, pos);
166                 if (ret < 0)
167                         return NULL;
168         }
169 #endif
170         ret = get_user_pages(current, bprm->mm, pos,
171                         1, write, 1, &page, NULL);
172         if (ret <= 0)
173                 return NULL;
174
175         if (write) {
176                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
177                 struct rlimit *rlim;
178
179                 /*
180                  * We've historically supported up to 32 pages (ARG_MAX)
181                  * of argument strings even with small stacks
182                  */
183                 if (size <= ARG_MAX)
184                         return page;
185
186                 /*
187                  * Limit to 1/4-th the stack size for the argv+env strings.
188                  * This ensures that:
189                  *  - the remaining binfmt code will not run out of stack space,
190                  *  - the program will have a reasonable amount of stack left
191                  *    to work from.
192                  */
193                 rlim = current->signal->rlim;
194                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
195                         put_page(page);
196                         return NULL;
197                 }
198         }
199
200         return page;
201 }
202
203 static void put_arg_page(struct page *page)
204 {
205         put_page(page);
206 }
207
208 static void free_arg_page(struct linux_binprm *bprm, int i)
209 {
210 }
211
212 static void free_arg_pages(struct linux_binprm *bprm)
213 {
214 }
215
216 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
217                 struct page *page)
218 {
219         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
220 }
221
222 static int __bprm_mm_init(struct linux_binprm *bprm)
223 {
224         int err = -ENOMEM;
225         struct vm_area_struct *vma = NULL;
226         struct mm_struct *mm = bprm->mm;
227
228         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
229         if (!vma)
230                 goto err;
231
232         down_write(&mm->mmap_sem);
233         vma->vm_mm = mm;
234
235         /*
236          * Place the stack at the largest stack address the architecture
237          * supports. Later, we'll move this to an appropriate place. We don't
238          * use STACK_TOP because that can depend on attributes which aren't
239          * configured yet.
240          */
241         vma->vm_end = STACK_TOP_MAX;
242         vma->vm_start = vma->vm_end - PAGE_SIZE;
243
244         vma->vm_flags = VM_STACK_FLAGS;
245         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
246         err = insert_vm_struct(mm, vma);
247         if (err) {
248                 up_write(&mm->mmap_sem);
249                 goto err;
250         }
251
252         mm->stack_vm = mm->total_vm = 1;
253         up_write(&mm->mmap_sem);
254
255         bprm->p = vma->vm_end - sizeof(void *);
256
257         return 0;
258
259 err:
260         if (vma) {
261                 bprm->vma = NULL;
262                 kmem_cache_free(vm_area_cachep, vma);
263         }
264
265         return err;
266 }
267
268 static bool valid_arg_len(struct linux_binprm *bprm, long len)
269 {
270         return len <= MAX_ARG_STRLEN;
271 }
272
273 #else
274
275 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
276                 int write)
277 {
278         struct page *page;
279
280         page = bprm->page[pos / PAGE_SIZE];
281         if (!page && write) {
282                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
283                 if (!page)
284                         return NULL;
285                 bprm->page[pos / PAGE_SIZE] = page;
286         }
287
288         return page;
289 }
290
291 static void put_arg_page(struct page *page)
292 {
293 }
294
295 static void free_arg_page(struct linux_binprm *bprm, int i)
296 {
297         if (bprm->page[i]) {
298                 __free_page(bprm->page[i]);
299                 bprm->page[i] = NULL;
300         }
301 }
302
303 static void free_arg_pages(struct linux_binprm *bprm)
304 {
305         int i;
306
307         for (i = 0; i < MAX_ARG_PAGES; i++)
308                 free_arg_page(bprm, i);
309 }
310
311 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
312                 struct page *page)
313 {
314 }
315
316 static int __bprm_mm_init(struct linux_binprm *bprm)
317 {
318         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
319         return 0;
320 }
321
322 static bool valid_arg_len(struct linux_binprm *bprm, long len)
323 {
324         return len <= bprm->p;
325 }
326
327 #endif /* CONFIG_MMU */
328
329 /*
330  * Create a new mm_struct and populate it with a temporary stack
331  * vm_area_struct.  We don't have enough context at this point to set the stack
332  * flags, permissions, and offset, so we use temporary values.  We'll update
333  * them later in setup_arg_pages().
334  */
335 int bprm_mm_init(struct linux_binprm *bprm)
336 {
337         int err;
338         struct mm_struct *mm = NULL;
339
340         bprm->mm = mm = mm_alloc();
341         err = -ENOMEM;
342         if (!mm)
343                 goto err;
344
345         err = init_new_context(current, mm);
346         if (err)
347                 goto err;
348
349         err = __bprm_mm_init(bprm);
350         if (err)
351                 goto err;
352
353         return 0;
354
355 err:
356         if (mm) {
357                 bprm->mm = NULL;
358                 mmdrop(mm);
359         }
360
361         return err;
362 }
363
364 /*
365  * count() counts the number of strings in array ARGV.
366  */
367 static int count(char __user * __user * argv, int max)
368 {
369         int i = 0;
370
371         if (argv != NULL) {
372                 for (;;) {
373                         char __user * p;
374
375                         if (get_user(p, argv))
376                                 return -EFAULT;
377                         if (!p)
378                                 break;
379                         argv++;
380                         if(++i > max)
381                                 return -E2BIG;
382                         cond_resched();
383                 }
384         }
385         return i;
386 }
387
388 /*
389  * 'copy_strings()' copies argument/environment strings from the old
390  * processes's memory to the new process's stack.  The call to get_user_pages()
391  * ensures the destination page is created and not swapped out.
392  */
393 static int copy_strings(int argc, char __user * __user * argv,
394                         struct linux_binprm *bprm)
395 {
396         struct page *kmapped_page = NULL;
397         char *kaddr = NULL;
398         unsigned long kpos = 0;
399         int ret;
400
401         while (argc-- > 0) {
402                 char __user *str;
403                 int len;
404                 unsigned long pos;
405
406                 if (get_user(str, argv+argc) ||
407                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
408                         ret = -EFAULT;
409                         goto out;
410                 }
411
412                 if (!valid_arg_len(bprm, len)) {
413                         ret = -E2BIG;
414                         goto out;
415                 }
416
417                 /* We're going to work our way backwords. */
418                 pos = bprm->p;
419                 str += len;
420                 bprm->p -= len;
421
422                 while (len > 0) {
423                         int offset, bytes_to_copy;
424
425                         offset = pos % PAGE_SIZE;
426                         if (offset == 0)
427                                 offset = PAGE_SIZE;
428
429                         bytes_to_copy = offset;
430                         if (bytes_to_copy > len)
431                                 bytes_to_copy = len;
432
433                         offset -= bytes_to_copy;
434                         pos -= bytes_to_copy;
435                         str -= bytes_to_copy;
436                         len -= bytes_to_copy;
437
438                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
439                                 struct page *page;
440
441                                 page = get_arg_page(bprm, pos, 1);
442                                 if (!page) {
443                                         ret = -E2BIG;
444                                         goto out;
445                                 }
446
447                                 if (kmapped_page) {
448                                         flush_kernel_dcache_page(kmapped_page);
449                                         kunmap(kmapped_page);
450                                         put_arg_page(kmapped_page);
451                                 }
452                                 kmapped_page = page;
453                                 kaddr = kmap(kmapped_page);
454                                 kpos = pos & PAGE_MASK;
455                                 flush_arg_page(bprm, kpos, kmapped_page);
456                         }
457                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
458                                 ret = -EFAULT;
459                                 goto out;
460                         }
461                 }
462         }
463         ret = 0;
464 out:
465         if (kmapped_page) {
466                 flush_kernel_dcache_page(kmapped_page);
467                 kunmap(kmapped_page);
468                 put_arg_page(kmapped_page);
469         }
470         return ret;
471 }
472
473 /*
474  * Like copy_strings, but get argv and its values from kernel memory.
475  */
476 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
477 {
478         int r;
479         mm_segment_t oldfs = get_fs();
480         set_fs(KERNEL_DS);
481         r = copy_strings(argc, (char __user * __user *)argv, bprm);
482         set_fs(oldfs);
483         return r;
484 }
485 EXPORT_SYMBOL(copy_strings_kernel);
486
487 #ifdef CONFIG_MMU
488
489 /*
490  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
491  * the binfmt code determines where the new stack should reside, we shift it to
492  * its final location.  The process proceeds as follows:
493  *
494  * 1) Use shift to calculate the new vma endpoints.
495  * 2) Extend vma to cover both the old and new ranges.  This ensures the
496  *    arguments passed to subsequent functions are consistent.
497  * 3) Move vma's page tables to the new range.
498  * 4) Free up any cleared pgd range.
499  * 5) Shrink the vma to cover only the new range.
500  */
501 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
502 {
503         struct mm_struct *mm = vma->vm_mm;
504         unsigned long old_start = vma->vm_start;
505         unsigned long old_end = vma->vm_end;
506         unsigned long length = old_end - old_start;
507         unsigned long new_start = old_start - shift;
508         unsigned long new_end = old_end - shift;
509         struct mmu_gather *tlb;
510
511         BUG_ON(new_start > new_end);
512
513         /*
514          * ensure there are no vmas between where we want to go
515          * and where we are
516          */
517         if (vma != find_vma(mm, new_start))
518                 return -EFAULT;
519
520         /*
521          * cover the whole range: [new_start, old_end)
522          */
523         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
524
525         /*
526          * move the page tables downwards, on failure we rely on
527          * process cleanup to remove whatever mess we made.
528          */
529         if (length != move_page_tables(vma, old_start,
530                                        vma, new_start, length))
531                 return -ENOMEM;
532
533         lru_add_drain();
534         tlb = tlb_gather_mmu(mm, 0);
535         if (new_end > old_start) {
536                 /*
537                  * when the old and new regions overlap clear from new_end.
538                  */
539                 free_pgd_range(&tlb, new_end, old_end, new_end,
540                         vma->vm_next ? vma->vm_next->vm_start : 0);
541         } else {
542                 /*
543                  * otherwise, clean from old_start; this is done to not touch
544                  * the address space in [new_end, old_start) some architectures
545                  * have constraints on va-space that make this illegal (IA64) -
546                  * for the others its just a little faster.
547                  */
548                 free_pgd_range(&tlb, old_start, old_end, new_end,
549                         vma->vm_next ? vma->vm_next->vm_start : 0);
550         }
551         tlb_finish_mmu(tlb, new_end, old_end);
552
553         /*
554          * shrink the vma to just the new range.
555          */
556         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
557
558         return 0;
559 }
560
561 #define EXTRA_STACK_VM_PAGES    20      /* random */
562
563 /*
564  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
565  * the stack is optionally relocated, and some extra space is added.
566  */
567 int setup_arg_pages(struct linux_binprm *bprm,
568                     unsigned long stack_top,
569                     int executable_stack)
570 {
571         unsigned long ret;
572         unsigned long stack_shift;
573         struct mm_struct *mm = current->mm;
574         struct vm_area_struct *vma = bprm->vma;
575         struct vm_area_struct *prev = NULL;
576         unsigned long vm_flags;
577         unsigned long stack_base;
578
579 #ifdef CONFIG_STACK_GROWSUP
580         /* Limit stack size to 1GB */
581         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
582         if (stack_base > (1 << 30))
583                 stack_base = 1 << 30;
584
585         /* Make sure we didn't let the argument array grow too large. */
586         if (vma->vm_end - vma->vm_start > stack_base)
587                 return -ENOMEM;
588
589         stack_base = PAGE_ALIGN(stack_top - stack_base);
590
591         stack_shift = vma->vm_start - stack_base;
592         mm->arg_start = bprm->p - stack_shift;
593         bprm->p = vma->vm_end - stack_shift;
594 #else
595         stack_top = arch_align_stack(stack_top);
596         stack_top = PAGE_ALIGN(stack_top);
597         stack_shift = vma->vm_end - stack_top;
598
599         bprm->p -= stack_shift;
600         mm->arg_start = bprm->p;
601 #endif
602
603         if (bprm->loader)
604                 bprm->loader -= stack_shift;
605         bprm->exec -= stack_shift;
606
607         down_write(&mm->mmap_sem);
608         vm_flags = vma->vm_flags;
609
610         /*
611          * Adjust stack execute permissions; explicitly enable for
612          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
613          * (arch default) otherwise.
614          */
615         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
616                 vm_flags |= VM_EXEC;
617         else if (executable_stack == EXSTACK_DISABLE_X)
618                 vm_flags &= ~VM_EXEC;
619         vm_flags |= mm->def_flags;
620
621         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
622                         vm_flags);
623         if (ret)
624                 goto out_unlock;
625         BUG_ON(prev != vma);
626
627         /* Move stack pages down in memory. */
628         if (stack_shift) {
629                 ret = shift_arg_pages(vma, stack_shift);
630                 if (ret) {
631                         up_write(&mm->mmap_sem);
632                         return ret;
633                 }
634         }
635
636 #ifdef CONFIG_STACK_GROWSUP
637         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
638 #else
639         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
640 #endif
641         ret = expand_stack(vma, stack_base);
642         if (ret)
643                 ret = -EFAULT;
644
645 out_unlock:
646         up_write(&mm->mmap_sem);
647         return 0;
648 }
649 EXPORT_SYMBOL(setup_arg_pages);
650
651 #endif /* CONFIG_MMU */
652
653 struct file *open_exec(const char *name)
654 {
655         struct nameidata nd;
656         int err;
657         struct file *file;
658
659         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
660         file = ERR_PTR(err);
661
662         if (!err) {
663                 struct inode *inode = nd.path.dentry->d_inode;
664                 file = ERR_PTR(-EACCES);
665                 if (S_ISREG(inode->i_mode)) {
666                         int err = vfs_permission(&nd, MAY_EXEC);
667                         file = ERR_PTR(err);
668                         if (!err) {
669                                 file = nameidata_to_filp(&nd,
670                                                         O_RDONLY|O_LARGEFILE);
671                                 if (!IS_ERR(file)) {
672                                         err = deny_write_access(file);
673                                         if (err) {
674                                                 fput(file);
675                                                 file = ERR_PTR(err);
676                                         }
677                                 }
678 out:
679                                 return file;
680                         }
681                 }
682                 release_open_intent(&nd);
683                 path_put(&nd.path);
684         }
685         goto out;
686 }
687
688 EXPORT_SYMBOL(open_exec);
689
690 int kernel_read(struct file *file, unsigned long offset,
691         char *addr, unsigned long count)
692 {
693         mm_segment_t old_fs;
694         loff_t pos = offset;
695         int result;
696
697         old_fs = get_fs();
698         set_fs(get_ds());
699         /* The cast to a user pointer is valid due to the set_fs() */
700         result = vfs_read(file, (void __user *)addr, count, &pos);
701         set_fs(old_fs);
702         return result;
703 }
704
705 EXPORT_SYMBOL(kernel_read);
706
707 static int exec_mmap(struct mm_struct *mm)
708 {
709         struct task_struct *tsk;
710         struct mm_struct * old_mm, *active_mm;
711
712         /* Notify parent that we're no longer interested in the old VM */
713         tsk = current;
714         old_mm = current->mm;
715         mm_release(tsk, old_mm);
716
717         if (old_mm) {
718                 /*
719                  * Make sure that if there is a core dump in progress
720                  * for the old mm, we get out and die instead of going
721                  * through with the exec.  We must hold mmap_sem around
722                  * checking core_waiters and changing tsk->mm.  The
723                  * core-inducing thread will increment core_waiters for
724                  * each thread whose ->mm == old_mm.
725                  */
726                 down_read(&old_mm->mmap_sem);
727                 if (unlikely(old_mm->core_waiters)) {
728                         up_read(&old_mm->mmap_sem);
729                         return -EINTR;
730                 }
731         }
732         task_lock(tsk);
733         active_mm = tsk->active_mm;
734         tsk->mm = mm;
735         tsk->active_mm = mm;
736         activate_mm(active_mm, mm);
737         task_unlock(tsk);
738         arch_pick_mmap_layout(mm);
739         if (old_mm) {
740                 up_read(&old_mm->mmap_sem);
741                 BUG_ON(active_mm != old_mm);
742                 mmput(old_mm);
743                 return 0;
744         }
745         mmdrop(active_mm);
746         return 0;
747 }
748
749 /*
750  * This function makes sure the current process has its own signal table,
751  * so that flush_signal_handlers can later reset the handlers without
752  * disturbing other processes.  (Other processes might share the signal
753  * table via the CLONE_SIGHAND option to clone().)
754  */
755 static int de_thread(struct task_struct *tsk)
756 {
757         struct signal_struct *sig = tsk->signal;
758         struct sighand_struct *oldsighand = tsk->sighand;
759         spinlock_t *lock = &oldsighand->siglock;
760         struct task_struct *leader = NULL;
761         int count;
762
763         if (thread_group_empty(tsk))
764                 goto no_thread_group;
765
766         /*
767          * Kill all other threads in the thread group.
768          * We must hold tasklist_lock to call zap_other_threads.
769          */
770         read_lock(&tasklist_lock);
771         spin_lock_irq(lock);
772         if (signal_group_exit(sig)) {
773                 /*
774                  * Another group action in progress, just
775                  * return so that the signal is processed.
776                  */
777                 spin_unlock_irq(lock);
778                 read_unlock(&tasklist_lock);
779                 return -EAGAIN;
780         }
781
782         /*
783          * child_reaper ignores SIGKILL, change it now.
784          * Reparenting needs write_lock on tasklist_lock,
785          * so it is safe to do it under read_lock.
786          */
787         if (unlikely(tsk->group_leader == task_child_reaper(tsk)))
788                 task_active_pid_ns(tsk)->child_reaper = tsk;
789
790         sig->group_exit_task = tsk;
791         zap_other_threads(tsk);
792         read_unlock(&tasklist_lock);
793
794         /* Account for the thread group leader hanging around: */
795         count = thread_group_leader(tsk) ? 1 : 2;
796         sig->notify_count = count;
797         while (atomic_read(&sig->count) > count) {
798                 __set_current_state(TASK_UNINTERRUPTIBLE);
799                 spin_unlock_irq(lock);
800                 schedule();
801                 spin_lock_irq(lock);
802         }
803         spin_unlock_irq(lock);
804
805         /*
806          * At this point all other threads have exited, all we have to
807          * do is to wait for the thread group leader to become inactive,
808          * and to assume its PID:
809          */
810         if (!thread_group_leader(tsk)) {
811                 leader = tsk->group_leader;
812
813                 sig->notify_count = -1;
814                 for (;;) {
815                         write_lock_irq(&tasklist_lock);
816                         if (likely(leader->exit_state))
817                                 break;
818                         __set_current_state(TASK_UNINTERRUPTIBLE);
819                         write_unlock_irq(&tasklist_lock);
820                         schedule();
821                 }
822
823                 /*
824                  * The only record we have of the real-time age of a
825                  * process, regardless of execs it's done, is start_time.
826                  * All the past CPU time is accumulated in signal_struct
827                  * from sister threads now dead.  But in this non-leader
828                  * exec, nothing survives from the original leader thread,
829                  * whose birth marks the true age of this process now.
830                  * When we take on its identity by switching to its PID, we
831                  * also take its birthdate (always earlier than our own).
832                  */
833                 tsk->start_time = leader->start_time;
834
835                 BUG_ON(!same_thread_group(leader, tsk));
836                 BUG_ON(has_group_leader_pid(tsk));
837                 /*
838                  * An exec() starts a new thread group with the
839                  * TGID of the previous thread group. Rehash the
840                  * two threads with a switched PID, and release
841                  * the former thread group leader:
842                  */
843
844                 /* Become a process group leader with the old leader's pid.
845                  * The old leader becomes a thread of the this thread group.
846                  * Note: The old leader also uses this pid until release_task
847                  *       is called.  Odd but simple and correct.
848                  */
849                 detach_pid(tsk, PIDTYPE_PID);
850                 tsk->pid = leader->pid;
851                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
852                 transfer_pid(leader, tsk, PIDTYPE_PGID);
853                 transfer_pid(leader, tsk, PIDTYPE_SID);
854                 list_replace_rcu(&leader->tasks, &tsk->tasks);
855
856                 tsk->group_leader = tsk;
857                 leader->group_leader = tsk;
858
859                 tsk->exit_signal = SIGCHLD;
860
861                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
862                 leader->exit_state = EXIT_DEAD;
863
864                 write_unlock_irq(&tasklist_lock);
865         }
866
867         sig->group_exit_task = NULL;
868         sig->notify_count = 0;
869
870 no_thread_group:
871         exit_itimers(sig);
872         if (leader)
873                 release_task(leader);
874
875         if (atomic_read(&oldsighand->count) != 1) {
876                 struct sighand_struct *newsighand;
877                 /*
878                  * This ->sighand is shared with the CLONE_SIGHAND
879                  * but not CLONE_THREAD task, switch to the new one.
880                  */
881                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
882                 if (!newsighand)
883                         return -ENOMEM;
884
885                 atomic_set(&newsighand->count, 1);
886                 memcpy(newsighand->action, oldsighand->action,
887                        sizeof(newsighand->action));
888
889                 write_lock_irq(&tasklist_lock);
890                 spin_lock(&oldsighand->siglock);
891                 rcu_assign_pointer(tsk->sighand, newsighand);
892                 spin_unlock(&oldsighand->siglock);
893                 write_unlock_irq(&tasklist_lock);
894
895                 __cleanup_sighand(oldsighand);
896         }
897
898         BUG_ON(!thread_group_leader(tsk));
899         return 0;
900 }
901
902 /*
903  * These functions flushes out all traces of the currently running executable
904  * so that a new one can be started
905  */
906 static void flush_old_files(struct files_struct * files)
907 {
908         long j = -1;
909         struct fdtable *fdt;
910
911         spin_lock(&files->file_lock);
912         for (;;) {
913                 unsigned long set, i;
914
915                 j++;
916                 i = j * __NFDBITS;
917                 fdt = files_fdtable(files);
918                 if (i >= fdt->max_fds)
919                         break;
920                 set = fdt->close_on_exec->fds_bits[j];
921                 if (!set)
922                         continue;
923                 fdt->close_on_exec->fds_bits[j] = 0;
924                 spin_unlock(&files->file_lock);
925                 for ( ; set ; i++,set >>= 1) {
926                         if (set & 1) {
927                                 sys_close(i);
928                         }
929                 }
930                 spin_lock(&files->file_lock);
931
932         }
933         spin_unlock(&files->file_lock);
934 }
935
936 char *get_task_comm(char *buf, struct task_struct *tsk)
937 {
938         /* buf must be at least sizeof(tsk->comm) in size */
939         task_lock(tsk);
940         strncpy(buf, tsk->comm, sizeof(tsk->comm));
941         task_unlock(tsk);
942         return buf;
943 }
944
945 void set_task_comm(struct task_struct *tsk, char *buf)
946 {
947         task_lock(tsk);
948         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
949         task_unlock(tsk);
950 }
951
952 int flush_old_exec(struct linux_binprm * bprm)
953 {
954         char * name;
955         int i, ch, retval;
956         struct files_struct *files;
957         char tcomm[sizeof(current->comm)];
958
959         /*
960          * Make sure we have a private signal table and that
961          * we are unassociated from the previous thread group.
962          */
963         retval = de_thread(current);
964         if (retval)
965                 goto out;
966
967         /*
968          * Make sure we have private file handles. Ask the
969          * fork helper to do the work for us and the exit
970          * helper to do the cleanup of the old one.
971          */
972         files = current->files;         /* refcounted so safe to hold */
973         retval = unshare_files();
974         if (retval)
975                 goto out;
976         /*
977          * Release all of the old mmap stuff
978          */
979         retval = exec_mmap(bprm->mm);
980         if (retval)
981                 goto mmap_failed;
982
983         bprm->mm = NULL;                /* We're using it now */
984
985         /* This is the point of no return */
986         put_files_struct(files);
987
988         current->sas_ss_sp = current->sas_ss_size = 0;
989
990         if (current->euid == current->uid && current->egid == current->gid)
991                 set_dumpable(current->mm, 1);
992         else
993                 set_dumpable(current->mm, suid_dumpable);
994
995         name = bprm->filename;
996
997         /* Copies the binary name from after last slash */
998         for (i=0; (ch = *(name++)) != '\0';) {
999                 if (ch == '/')
1000                         i = 0; /* overwrite what we wrote */
1001                 else
1002                         if (i < (sizeof(tcomm) - 1))
1003                                 tcomm[i++] = ch;
1004         }
1005         tcomm[i] = '\0';
1006         set_task_comm(current, tcomm);
1007
1008         current->flags &= ~PF_RANDOMIZE;
1009         flush_thread();
1010
1011         /* Set the new mm task size. We have to do that late because it may
1012          * depend on TIF_32BIT which is only updated in flush_thread() on
1013          * some architectures like powerpc
1014          */
1015         current->mm->task_size = TASK_SIZE;
1016
1017         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1018                 suid_keys(current);
1019                 set_dumpable(current->mm, suid_dumpable);
1020                 current->pdeath_signal = 0;
1021         } else if (file_permission(bprm->file, MAY_READ) ||
1022                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1023                 suid_keys(current);
1024                 set_dumpable(current->mm, suid_dumpable);
1025         }
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 mmap_failed:
1038         reset_files_struct(current, files);
1039 out:
1040         return retval;
1041 }
1042
1043 EXPORT_SYMBOL(flush_old_exec);
1044
1045 /* 
1046  * Fill the binprm structure from the inode. 
1047  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1048  */
1049 int prepare_binprm(struct linux_binprm *bprm)
1050 {
1051         int mode;
1052         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1053         int retval;
1054
1055         mode = inode->i_mode;
1056         if (bprm->file->f_op == NULL)
1057                 return -EACCES;
1058
1059         bprm->e_uid = current->euid;
1060         bprm->e_gid = current->egid;
1061
1062         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1063                 /* Set-uid? */
1064                 if (mode & S_ISUID) {
1065                         current->personality &= ~PER_CLEAR_ON_SETID;
1066                         bprm->e_uid = inode->i_uid;
1067                 }
1068
1069                 /* Set-gid? */
1070                 /*
1071                  * If setgid is set but no group execute bit then this
1072                  * is a candidate for mandatory locking, not a setgid
1073                  * executable.
1074                  */
1075                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1076                         current->personality &= ~PER_CLEAR_ON_SETID;
1077                         bprm->e_gid = inode->i_gid;
1078                 }
1079         }
1080
1081         /* fill in binprm security blob */
1082         retval = security_bprm_set(bprm);
1083         if (retval)
1084                 return retval;
1085
1086         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1087         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1088 }
1089
1090 EXPORT_SYMBOL(prepare_binprm);
1091
1092 static int unsafe_exec(struct task_struct *p)
1093 {
1094         int unsafe = 0;
1095         if (p->ptrace & PT_PTRACED) {
1096                 if (p->ptrace & PT_PTRACE_CAP)
1097                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1098                 else
1099                         unsafe |= LSM_UNSAFE_PTRACE;
1100         }
1101         if (atomic_read(&p->fs->count) > 1 ||
1102             atomic_read(&p->files->count) > 1 ||
1103             atomic_read(&p->sighand->count) > 1)
1104                 unsafe |= LSM_UNSAFE_SHARE;
1105
1106         return unsafe;
1107 }
1108
1109 void compute_creds(struct linux_binprm *bprm)
1110 {
1111         int unsafe;
1112
1113         if (bprm->e_uid != current->uid) {
1114                 suid_keys(current);
1115                 current->pdeath_signal = 0;
1116         }
1117         exec_keys(current);
1118
1119         task_lock(current);
1120         unsafe = unsafe_exec(current);
1121         security_bprm_apply_creds(bprm, unsafe);
1122         task_unlock(current);
1123         security_bprm_post_apply_creds(bprm);
1124 }
1125 EXPORT_SYMBOL(compute_creds);
1126
1127 /*
1128  * Arguments are '\0' separated strings found at the location bprm->p
1129  * points to; chop off the first by relocating brpm->p to right after
1130  * the first '\0' encountered.
1131  */
1132 int remove_arg_zero(struct linux_binprm *bprm)
1133 {
1134         int ret = 0;
1135         unsigned long offset;
1136         char *kaddr;
1137         struct page *page;
1138
1139         if (!bprm->argc)
1140                 return 0;
1141
1142         do {
1143                 offset = bprm->p & ~PAGE_MASK;
1144                 page = get_arg_page(bprm, bprm->p, 0);
1145                 if (!page) {
1146                         ret = -EFAULT;
1147                         goto out;
1148                 }
1149                 kaddr = kmap_atomic(page, KM_USER0);
1150
1151                 for (; offset < PAGE_SIZE && kaddr[offset];
1152                                 offset++, bprm->p++)
1153                         ;
1154
1155                 kunmap_atomic(kaddr, KM_USER0);
1156                 put_arg_page(page);
1157
1158                 if (offset == PAGE_SIZE)
1159                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1160         } while (offset == PAGE_SIZE);
1161
1162         bprm->p++;
1163         bprm->argc--;
1164         ret = 0;
1165
1166 out:
1167         return ret;
1168 }
1169 EXPORT_SYMBOL(remove_arg_zero);
1170
1171 /*
1172  * cycle the list of binary formats handler, until one recognizes the image
1173  */
1174 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1175 {
1176         int try,retval;
1177         struct linux_binfmt *fmt;
1178 #if defined(__alpha__) && defined(CONFIG_ARCH_SUPPORTS_AOUT)
1179         /* handle /sbin/loader.. */
1180         {
1181             struct exec * eh = (struct exec *) bprm->buf;
1182
1183             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1184                 (eh->fh.f_flags & 0x3000) == 0x3000)
1185             {
1186                 struct file * file;
1187                 unsigned long loader;
1188
1189                 allow_write_access(bprm->file);
1190                 fput(bprm->file);
1191                 bprm->file = NULL;
1192
1193                 loader = bprm->vma->vm_end - sizeof(void *);
1194
1195                 file = open_exec("/sbin/loader");
1196                 retval = PTR_ERR(file);
1197                 if (IS_ERR(file))
1198                         return retval;
1199
1200                 /* Remember if the application is TASO.  */
1201                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1202
1203                 bprm->file = file;
1204                 bprm->loader = loader;
1205                 retval = prepare_binprm(bprm);
1206                 if (retval<0)
1207                         return retval;
1208                 /* should call search_binary_handler recursively here,
1209                    but it does not matter */
1210             }
1211         }
1212 #endif
1213         retval = security_bprm_check(bprm);
1214         if (retval)
1215                 return retval;
1216
1217         /* kernel module loader fixup */
1218         /* so we don't try to load run modprobe in kernel space. */
1219         set_fs(USER_DS);
1220
1221         retval = audit_bprm(bprm);
1222         if (retval)
1223                 return retval;
1224
1225         retval = -ENOENT;
1226         for (try=0; try<2; try++) {
1227                 read_lock(&binfmt_lock);
1228                 list_for_each_entry(fmt, &formats, lh) {
1229                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1230                         if (!fn)
1231                                 continue;
1232                         if (!try_module_get(fmt->module))
1233                                 continue;
1234                         read_unlock(&binfmt_lock);
1235                         retval = fn(bprm, regs);
1236                         if (retval >= 0) {
1237                                 put_binfmt(fmt);
1238                                 allow_write_access(bprm->file);
1239                                 if (bprm->file)
1240                                         fput(bprm->file);
1241                                 bprm->file = NULL;
1242                                 current->did_exec = 1;
1243                                 proc_exec_connector(current);
1244                                 return retval;
1245                         }
1246                         read_lock(&binfmt_lock);
1247                         put_binfmt(fmt);
1248                         if (retval != -ENOEXEC || bprm->mm == NULL)
1249                                 break;
1250                         if (!bprm->file) {
1251                                 read_unlock(&binfmt_lock);
1252                                 return retval;
1253                         }
1254                 }
1255                 read_unlock(&binfmt_lock);
1256                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1257                         break;
1258 #ifdef CONFIG_KMOD
1259                 }else{
1260 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1261                         if (printable(bprm->buf[0]) &&
1262                             printable(bprm->buf[1]) &&
1263                             printable(bprm->buf[2]) &&
1264                             printable(bprm->buf[3]))
1265                                 break; /* -ENOEXEC */
1266                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1267 #endif
1268                 }
1269         }
1270         return retval;
1271 }
1272
1273 EXPORT_SYMBOL(search_binary_handler);
1274
1275 /*
1276  * sys_execve() executes a new program.
1277  */
1278 int do_execve(char * filename,
1279         char __user *__user *argv,
1280         char __user *__user *envp,
1281         struct pt_regs * regs)
1282 {
1283         struct linux_binprm *bprm;
1284         struct file *file;
1285         unsigned long env_p;
1286         int retval;
1287
1288         retval = -ENOMEM;
1289         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1290         if (!bprm)
1291                 goto out_ret;
1292
1293         file = open_exec(filename);
1294         retval = PTR_ERR(file);
1295         if (IS_ERR(file))
1296                 goto out_kfree;
1297
1298         sched_exec();
1299
1300         bprm->file = file;
1301         bprm->filename = filename;
1302         bprm->interp = filename;
1303
1304         retval = bprm_mm_init(bprm);
1305         if (retval)
1306                 goto out_file;
1307
1308         bprm->argc = count(argv, MAX_ARG_STRINGS);
1309         if ((retval = bprm->argc) < 0)
1310                 goto out_mm;
1311
1312         bprm->envc = count(envp, MAX_ARG_STRINGS);
1313         if ((retval = bprm->envc) < 0)
1314                 goto out_mm;
1315
1316         retval = security_bprm_alloc(bprm);
1317         if (retval)
1318                 goto out;
1319
1320         retval = prepare_binprm(bprm);
1321         if (retval < 0)
1322                 goto out;
1323
1324         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1325         if (retval < 0)
1326                 goto out;
1327
1328         bprm->exec = bprm->p;
1329         retval = copy_strings(bprm->envc, envp, bprm);
1330         if (retval < 0)
1331                 goto out;
1332
1333         env_p = bprm->p;
1334         retval = copy_strings(bprm->argc, argv, bprm);
1335         if (retval < 0)
1336                 goto out;
1337         bprm->argv_len = env_p - bprm->p;
1338
1339         retval = search_binary_handler(bprm,regs);
1340         if (retval >= 0) {
1341                 /* execve success */
1342                 free_arg_pages(bprm);
1343                 security_bprm_free(bprm);
1344                 acct_update_integrals(current);
1345                 kfree(bprm);
1346                 return retval;
1347         }
1348
1349 out:
1350         free_arg_pages(bprm);
1351         if (bprm->security)
1352                 security_bprm_free(bprm);
1353
1354 out_mm:
1355         if (bprm->mm)
1356                 mmput (bprm->mm);
1357
1358 out_file:
1359         if (bprm->file) {
1360                 allow_write_access(bprm->file);
1361                 fput(bprm->file);
1362         }
1363 out_kfree:
1364         kfree(bprm);
1365
1366 out_ret:
1367         return retval;
1368 }
1369
1370 int set_binfmt(struct linux_binfmt *new)
1371 {
1372         struct linux_binfmt *old = current->binfmt;
1373
1374         if (new) {
1375                 if (!try_module_get(new->module))
1376                         return -1;
1377         }
1378         current->binfmt = new;
1379         if (old)
1380                 module_put(old->module);
1381         return 0;
1382 }
1383
1384 EXPORT_SYMBOL(set_binfmt);
1385
1386 /* format_corename will inspect the pattern parameter, and output a
1387  * name into corename, which must have space for at least
1388  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1389  */
1390 static int format_corename(char *corename, const char *pattern, long signr)
1391 {
1392         const char *pat_ptr = pattern;
1393         char *out_ptr = corename;
1394         char *const out_end = corename + CORENAME_MAX_SIZE;
1395         int rc;
1396         int pid_in_pattern = 0;
1397         int ispipe = 0;
1398
1399         if (*pattern == '|')
1400                 ispipe = 1;
1401
1402         /* Repeat as long as we have more pattern to process and more output
1403            space */
1404         while (*pat_ptr) {
1405                 if (*pat_ptr != '%') {
1406                         if (out_ptr == out_end)
1407                                 goto out;
1408                         *out_ptr++ = *pat_ptr++;
1409                 } else {
1410                         switch (*++pat_ptr) {
1411                         case 0:
1412                                 goto out;
1413                         /* Double percent, output one percent */
1414                         case '%':
1415                                 if (out_ptr == out_end)
1416                                         goto out;
1417                                 *out_ptr++ = '%';
1418                                 break;
1419                         /* pid */
1420                         case 'p':
1421                                 pid_in_pattern = 1;
1422                                 rc = snprintf(out_ptr, out_end - out_ptr,
1423                                               "%d", task_tgid_vnr(current));
1424                                 if (rc > out_end - out_ptr)
1425                                         goto out;
1426                                 out_ptr += rc;
1427                                 break;
1428                         /* uid */
1429                         case 'u':
1430                                 rc = snprintf(out_ptr, out_end - out_ptr,
1431                                               "%d", current->uid);
1432                                 if (rc > out_end - out_ptr)
1433                                         goto out;
1434                                 out_ptr += rc;
1435                                 break;
1436                         /* gid */
1437                         case 'g':
1438                                 rc = snprintf(out_ptr, out_end - out_ptr,
1439                                               "%d", current->gid);
1440                                 if (rc > out_end - out_ptr)
1441                                         goto out;
1442                                 out_ptr += rc;
1443                                 break;
1444                         /* signal that caused the coredump */
1445                         case 's':
1446                                 rc = snprintf(out_ptr, out_end - out_ptr,
1447                                               "%ld", signr);
1448                                 if (rc > out_end - out_ptr)
1449                                         goto out;
1450                                 out_ptr += rc;
1451                                 break;
1452                         /* UNIX time of coredump */
1453                         case 't': {
1454                                 struct timeval tv;
1455                                 do_gettimeofday(&tv);
1456                                 rc = snprintf(out_ptr, out_end - out_ptr,
1457                                               "%lu", tv.tv_sec);
1458                                 if (rc > out_end - out_ptr)
1459                                         goto out;
1460                                 out_ptr += rc;
1461                                 break;
1462                         }
1463                         /* hostname */
1464                         case 'h':
1465                                 down_read(&uts_sem);
1466                                 rc = snprintf(out_ptr, out_end - out_ptr,
1467                                               "%s", utsname()->nodename);
1468                                 up_read(&uts_sem);
1469                                 if (rc > out_end - out_ptr)
1470                                         goto out;
1471                                 out_ptr += rc;
1472                                 break;
1473                         /* executable */
1474                         case 'e':
1475                                 rc = snprintf(out_ptr, out_end - out_ptr,
1476                                               "%s", current->comm);
1477                                 if (rc > out_end - out_ptr)
1478                                         goto out;
1479                                 out_ptr += rc;
1480                                 break;
1481                         /* core limit size */
1482                         case 'c':
1483                                 rc = snprintf(out_ptr, out_end - out_ptr,
1484                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1485                                 if (rc > out_end - out_ptr)
1486                                         goto out;
1487                                 out_ptr += rc;
1488                                 break;
1489                         default:
1490                                 break;
1491                         }
1492                         ++pat_ptr;
1493                 }
1494         }
1495         /* Backward compatibility with core_uses_pid:
1496          *
1497          * If core_pattern does not include a %p (as is the default)
1498          * and core_uses_pid is set, then .%pid will be appended to
1499          * the filename. Do not do this for piped commands. */
1500         if (!ispipe && !pid_in_pattern
1501             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1502                 rc = snprintf(out_ptr, out_end - out_ptr,
1503                               ".%d", task_tgid_vnr(current));
1504                 if (rc > out_end - out_ptr)
1505                         goto out;
1506                 out_ptr += rc;
1507         }
1508 out:
1509         *out_ptr = 0;
1510         return ispipe;
1511 }
1512
1513 static void zap_process(struct task_struct *start)
1514 {
1515         struct task_struct *t;
1516
1517         start->signal->flags = SIGNAL_GROUP_EXIT;
1518         start->signal->group_stop_count = 0;
1519
1520         t = start;
1521         do {
1522                 if (t != current && t->mm) {
1523                         t->mm->core_waiters++;
1524                         sigaddset(&t->pending.signal, SIGKILL);
1525                         signal_wake_up(t, 1);
1526                 }
1527         } while ((t = next_thread(t)) != start);
1528 }
1529
1530 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1531                                 int exit_code)
1532 {
1533         struct task_struct *g, *p;
1534         unsigned long flags;
1535         int err = -EAGAIN;
1536
1537         spin_lock_irq(&tsk->sighand->siglock);
1538         if (!signal_group_exit(tsk->signal)) {
1539                 tsk->signal->group_exit_code = exit_code;
1540                 zap_process(tsk);
1541                 err = 0;
1542         }
1543         spin_unlock_irq(&tsk->sighand->siglock);
1544         if (err)
1545                 return err;
1546
1547         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1548                 goto done;
1549
1550         rcu_read_lock();
1551         for_each_process(g) {
1552                 if (g == tsk->group_leader)
1553                         continue;
1554
1555                 p = g;
1556                 do {
1557                         if (p->mm) {
1558                                 if (p->mm == mm) {
1559                                         /*
1560                                          * p->sighand can't disappear, but
1561                                          * may be changed by de_thread()
1562                                          */
1563                                         lock_task_sighand(p, &flags);
1564                                         zap_process(p);
1565                                         unlock_task_sighand(p, &flags);
1566                                 }
1567                                 break;
1568                         }
1569                 } while ((p = next_thread(p)) != g);
1570         }
1571         rcu_read_unlock();
1572 done:
1573         return mm->core_waiters;
1574 }
1575
1576 static int coredump_wait(int exit_code)
1577 {
1578         struct task_struct *tsk = current;
1579         struct mm_struct *mm = tsk->mm;
1580         struct completion startup_done;
1581         struct completion *vfork_done;
1582         int core_waiters;
1583
1584         init_completion(&mm->core_done);
1585         init_completion(&startup_done);
1586         mm->core_startup_done = &startup_done;
1587
1588         core_waiters = zap_threads(tsk, mm, exit_code);
1589         up_write(&mm->mmap_sem);
1590
1591         if (unlikely(core_waiters < 0))
1592                 goto fail;
1593
1594         /*
1595          * Make sure nobody is waiting for us to release the VM,
1596          * otherwise we can deadlock when we wait on each other
1597          */
1598         vfork_done = tsk->vfork_done;
1599         if (vfork_done) {
1600                 tsk->vfork_done = NULL;
1601                 complete(vfork_done);
1602         }
1603
1604         if (core_waiters)
1605                 wait_for_completion(&startup_done);
1606 fail:
1607         BUG_ON(mm->core_waiters);
1608         return core_waiters;
1609 }
1610
1611 /*
1612  * set_dumpable converts traditional three-value dumpable to two flags and
1613  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1614  * these bits are not changed atomically.  So get_dumpable can observe the
1615  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1616  * return either old dumpable or new one by paying attention to the order of
1617  * modifying the bits.
1618  *
1619  * dumpable |   mm->flags (binary)
1620  * old  new | initial interim  final
1621  * ---------+-----------------------
1622  *  0    1  |   00      01      01
1623  *  0    2  |   00      10(*)   11
1624  *  1    0  |   01      00      00
1625  *  1    2  |   01      11      11
1626  *  2    0  |   11      10(*)   00
1627  *  2    1  |   11      11      01
1628  *
1629  * (*) get_dumpable regards interim value of 10 as 11.
1630  */
1631 void set_dumpable(struct mm_struct *mm, int value)
1632 {
1633         switch (value) {
1634         case 0:
1635                 clear_bit(MMF_DUMPABLE, &mm->flags);
1636                 smp_wmb();
1637                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1638                 break;
1639         case 1:
1640                 set_bit(MMF_DUMPABLE, &mm->flags);
1641                 smp_wmb();
1642                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1643                 break;
1644         case 2:
1645                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1646                 smp_wmb();
1647                 set_bit(MMF_DUMPABLE, &mm->flags);
1648                 break;
1649         }
1650 }
1651
1652 int get_dumpable(struct mm_struct *mm)
1653 {
1654         int ret;
1655
1656         ret = mm->flags & 0x3;
1657         return (ret >= 2) ? 2 : ret;
1658 }
1659
1660 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1661 {
1662         char corename[CORENAME_MAX_SIZE + 1];
1663         struct mm_struct *mm = current->mm;
1664         struct linux_binfmt * binfmt;
1665         struct inode * inode;
1666         struct file * file;
1667         int retval = 0;
1668         int fsuid = current->fsuid;
1669         int flag = 0;
1670         int ispipe = 0;
1671         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1672         char **helper_argv = NULL;
1673         int helper_argc = 0;
1674         char *delimit;
1675
1676         audit_core_dumps(signr);
1677
1678         binfmt = current->binfmt;
1679         if (!binfmt || !binfmt->core_dump)
1680                 goto fail;
1681         down_write(&mm->mmap_sem);
1682         /*
1683          * If another thread got here first, or we are not dumpable, bail out.
1684          */
1685         if (mm->core_waiters || !get_dumpable(mm)) {
1686                 up_write(&mm->mmap_sem);
1687                 goto fail;
1688         }
1689
1690         /*
1691          *      We cannot trust fsuid as being the "true" uid of the
1692          *      process nor do we know its entire history. We only know it
1693          *      was tainted so we dump it as root in mode 2.
1694          */
1695         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1696                 flag = O_EXCL;          /* Stop rewrite attacks */
1697                 current->fsuid = 0;     /* Dump root private */
1698         }
1699
1700         retval = coredump_wait(exit_code);
1701         if (retval < 0)
1702                 goto fail;
1703
1704         /*
1705          * Clear any false indication of pending signals that might
1706          * be seen by the filesystem code called to write the core file.
1707          */
1708         clear_thread_flag(TIF_SIGPENDING);
1709
1710         /*
1711          * lock_kernel() because format_corename() is controlled by sysctl, which
1712          * uses lock_kernel()
1713          */
1714         lock_kernel();
1715         ispipe = format_corename(corename, core_pattern, signr);
1716         unlock_kernel();
1717         /*
1718          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1719          * to a pipe.  Since we're not writing directly to the filesystem
1720          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1721          * created unless the pipe reader choses to write out the core file
1722          * at which point file size limits and permissions will be imposed
1723          * as it does with any other process
1724          */
1725         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1726                 goto fail_unlock;
1727
1728         if (ispipe) {
1729                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1730                 /* Terminate the string before the first option */
1731                 delimit = strchr(corename, ' ');
1732                 if (delimit)
1733                         *delimit = '\0';
1734                 delimit = strrchr(helper_argv[0], '/');
1735                 if (delimit)
1736                         delimit++;
1737                 else
1738                         delimit = helper_argv[0];
1739                 if (!strcmp(delimit, current->comm)) {
1740                         printk(KERN_NOTICE "Recursive core dump detected, "
1741                                         "aborting\n");
1742                         goto fail_unlock;
1743                 }
1744
1745                 core_limit = RLIM_INFINITY;
1746
1747                 /* SIGPIPE can happen, but it's just never processed */
1748                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1749                                 &file)) {
1750                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1751                                corename);
1752                         goto fail_unlock;
1753                 }
1754         } else
1755                 file = filp_open(corename,
1756                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1757                                  0600);
1758         if (IS_ERR(file))
1759                 goto fail_unlock;
1760         inode = file->f_path.dentry->d_inode;
1761         if (inode->i_nlink > 1)
1762                 goto close_fail;        /* multiple links - don't dump */
1763         if (!ispipe && d_unhashed(file->f_path.dentry))
1764                 goto close_fail;
1765
1766         /* AK: actually i see no reason to not allow this for named pipes etc.,
1767            but keep the previous behaviour for now. */
1768         if (!ispipe && !S_ISREG(inode->i_mode))
1769                 goto close_fail;
1770         /*
1771          * Dont allow local users get cute and trick others to coredump
1772          * into their pre-created files:
1773          */
1774         if (inode->i_uid != current->fsuid)
1775                 goto close_fail;
1776         if (!file->f_op)
1777                 goto close_fail;
1778         if (!file->f_op->write)
1779                 goto close_fail;
1780         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1781                 goto close_fail;
1782
1783         retval = binfmt->core_dump(signr, regs, file, core_limit);
1784
1785         if (retval)
1786                 current->signal->group_exit_code |= 0x80;
1787 close_fail:
1788         filp_close(file, NULL);
1789 fail_unlock:
1790         if (helper_argv)
1791                 argv_free(helper_argv);
1792
1793         current->fsuid = fsuid;
1794         complete_all(&mm->core_done);
1795 fail:
1796         return retval;
1797 }