CRED: Make execve() take advantage of copy-on-write credentials
[linux-2.6.git] / fs / exec.c
1 /*
2  *  linux/fs/exec.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  * #!-checking implemented by tytso.
9  */
10 /*
11  * Demand-loading implemented 01.12.91 - no need to read anything but
12  * the header into memory. The inode of the executable is put into
13  * "current->executable", and page faults do the actual loading. Clean.
14  *
15  * Once more I can proudly say that linux stood up to being changed: it
16  * was less than 2 hours work to get demand-loading completely implemented.
17  *
18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
19  * current->executable is only used by the procfs.  This allows a dispatch
20  * table to check for several different types  of binary formats.  We keep
21  * trying until we recognize the file or we run out of supported binary
22  * formats. 
23  */
24
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/mm.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
53 #include <linux/kmod.h>
54
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
57 #include <asm/tlb.h>
58 #include "internal.h"
59
60 #ifdef __alpha__
61 /* for /sbin/loader handling in search_binary_handler() */
62 #include <linux/a.out.h>
63 #endif
64
65 int core_uses_pid;
66 char core_pattern[CORENAME_MAX_SIZE] = "core";
67 int suid_dumpable = 0;
68
69 /* The maximal length of core_pattern is also specified in sysctl.c */
70
71 static LIST_HEAD(formats);
72 static DEFINE_RWLOCK(binfmt_lock);
73
74 int register_binfmt(struct linux_binfmt * fmt)
75 {
76         if (!fmt)
77                 return -EINVAL;
78         write_lock(&binfmt_lock);
79         list_add(&fmt->lh, &formats);
80         write_unlock(&binfmt_lock);
81         return 0;       
82 }
83
84 EXPORT_SYMBOL(register_binfmt);
85
86 void unregister_binfmt(struct linux_binfmt * fmt)
87 {
88         write_lock(&binfmt_lock);
89         list_del(&fmt->lh);
90         write_unlock(&binfmt_lock);
91 }
92
93 EXPORT_SYMBOL(unregister_binfmt);
94
95 static inline void put_binfmt(struct linux_binfmt * fmt)
96 {
97         module_put(fmt->module);
98 }
99
100 /*
101  * Note that a shared library must be both readable and executable due to
102  * security reasons.
103  *
104  * Also note that we take the address to load from from the file itself.
105  */
106 asmlinkage long sys_uselib(const char __user * library)
107 {
108         struct file *file;
109         struct nameidata nd;
110         char *tmp = getname(library);
111         int error = PTR_ERR(tmp);
112
113         if (!IS_ERR(tmp)) {
114                 error = path_lookup_open(AT_FDCWD, tmp,
115                                          LOOKUP_FOLLOW, &nd,
116                                          FMODE_READ|FMODE_EXEC);
117                 putname(tmp);
118         }
119         if (error)
120                 goto out;
121
122         error = -EINVAL;
123         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
124                 goto exit;
125
126         error = -EACCES;
127         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
128                 goto exit;
129
130         error = vfs_permission(&nd, MAY_READ | MAY_EXEC | MAY_OPEN);
131         if (error)
132                 goto exit;
133
134         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
135         error = PTR_ERR(file);
136         if (IS_ERR(file))
137                 goto out;
138
139         error = -ENOEXEC;
140         if(file->f_op) {
141                 struct linux_binfmt * fmt;
142
143                 read_lock(&binfmt_lock);
144                 list_for_each_entry(fmt, &formats, lh) {
145                         if (!fmt->load_shlib)
146                                 continue;
147                         if (!try_module_get(fmt->module))
148                                 continue;
149                         read_unlock(&binfmt_lock);
150                         error = fmt->load_shlib(file);
151                         read_lock(&binfmt_lock);
152                         put_binfmt(fmt);
153                         if (error != -ENOEXEC)
154                                 break;
155                 }
156                 read_unlock(&binfmt_lock);
157         }
158         fput(file);
159 out:
160         return error;
161 exit:
162         release_open_intent(&nd);
163         path_put(&nd.path);
164         goto out;
165 }
166
167 #ifdef CONFIG_MMU
168
169 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
170                 int write)
171 {
172         struct page *page;
173         int ret;
174
175 #ifdef CONFIG_STACK_GROWSUP
176         if (write) {
177                 ret = expand_stack_downwards(bprm->vma, pos);
178                 if (ret < 0)
179                         return NULL;
180         }
181 #endif
182         ret = get_user_pages(current, bprm->mm, pos,
183                         1, write, 1, &page, NULL);
184         if (ret <= 0)
185                 return NULL;
186
187         if (write) {
188                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
189                 struct rlimit *rlim;
190
191                 /*
192                  * We've historically supported up to 32 pages (ARG_MAX)
193                  * of argument strings even with small stacks
194                  */
195                 if (size <= ARG_MAX)
196                         return page;
197
198                 /*
199                  * Limit to 1/4-th the stack size for the argv+env strings.
200                  * This ensures that:
201                  *  - the remaining binfmt code will not run out of stack space,
202                  *  - the program will have a reasonable amount of stack left
203                  *    to work from.
204                  */
205                 rlim = current->signal->rlim;
206                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
207                         put_page(page);
208                         return NULL;
209                 }
210         }
211
212         return page;
213 }
214
215 static void put_arg_page(struct page *page)
216 {
217         put_page(page);
218 }
219
220 static void free_arg_page(struct linux_binprm *bprm, int i)
221 {
222 }
223
224 static void free_arg_pages(struct linux_binprm *bprm)
225 {
226 }
227
228 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
229                 struct page *page)
230 {
231         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
232 }
233
234 static int __bprm_mm_init(struct linux_binprm *bprm)
235 {
236         int err = -ENOMEM;
237         struct vm_area_struct *vma = NULL;
238         struct mm_struct *mm = bprm->mm;
239
240         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
241         if (!vma)
242                 goto err;
243
244         down_write(&mm->mmap_sem);
245         vma->vm_mm = mm;
246
247         /*
248          * Place the stack at the largest stack address the architecture
249          * supports. Later, we'll move this to an appropriate place. We don't
250          * use STACK_TOP because that can depend on attributes which aren't
251          * configured yet.
252          */
253         vma->vm_end = STACK_TOP_MAX;
254         vma->vm_start = vma->vm_end - PAGE_SIZE;
255
256         vma->vm_flags = VM_STACK_FLAGS;
257         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
258         err = insert_vm_struct(mm, vma);
259         if (err) {
260                 up_write(&mm->mmap_sem);
261                 goto err;
262         }
263
264         mm->stack_vm = mm->total_vm = 1;
265         up_write(&mm->mmap_sem);
266
267         bprm->p = vma->vm_end - sizeof(void *);
268
269         return 0;
270
271 err:
272         if (vma) {
273                 bprm->vma = NULL;
274                 kmem_cache_free(vm_area_cachep, vma);
275         }
276
277         return err;
278 }
279
280 static bool valid_arg_len(struct linux_binprm *bprm, long len)
281 {
282         return len <= MAX_ARG_STRLEN;
283 }
284
285 #else
286
287 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
288                 int write)
289 {
290         struct page *page;
291
292         page = bprm->page[pos / PAGE_SIZE];
293         if (!page && write) {
294                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
295                 if (!page)
296                         return NULL;
297                 bprm->page[pos / PAGE_SIZE] = page;
298         }
299
300         return page;
301 }
302
303 static void put_arg_page(struct page *page)
304 {
305 }
306
307 static void free_arg_page(struct linux_binprm *bprm, int i)
308 {
309         if (bprm->page[i]) {
310                 __free_page(bprm->page[i]);
311                 bprm->page[i] = NULL;
312         }
313 }
314
315 static void free_arg_pages(struct linux_binprm *bprm)
316 {
317         int i;
318
319         for (i = 0; i < MAX_ARG_PAGES; i++)
320                 free_arg_page(bprm, i);
321 }
322
323 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
324                 struct page *page)
325 {
326 }
327
328 static int __bprm_mm_init(struct linux_binprm *bprm)
329 {
330         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
331         return 0;
332 }
333
334 static bool valid_arg_len(struct linux_binprm *bprm, long len)
335 {
336         return len <= bprm->p;
337 }
338
339 #endif /* CONFIG_MMU */
340
341 /*
342  * Create a new mm_struct and populate it with a temporary stack
343  * vm_area_struct.  We don't have enough context at this point to set the stack
344  * flags, permissions, and offset, so we use temporary values.  We'll update
345  * them later in setup_arg_pages().
346  */
347 int bprm_mm_init(struct linux_binprm *bprm)
348 {
349         int err;
350         struct mm_struct *mm = NULL;
351
352         bprm->mm = mm = mm_alloc();
353         err = -ENOMEM;
354         if (!mm)
355                 goto err;
356
357         err = init_new_context(current, mm);
358         if (err)
359                 goto err;
360
361         err = __bprm_mm_init(bprm);
362         if (err)
363                 goto err;
364
365         return 0;
366
367 err:
368         if (mm) {
369                 bprm->mm = NULL;
370                 mmdrop(mm);
371         }
372
373         return err;
374 }
375
376 /*
377  * count() counts the number of strings in array ARGV.
378  */
379 static int count(char __user * __user * argv, int max)
380 {
381         int i = 0;
382
383         if (argv != NULL) {
384                 for (;;) {
385                         char __user * p;
386
387                         if (get_user(p, argv))
388                                 return -EFAULT;
389                         if (!p)
390                                 break;
391                         argv++;
392                         if (i++ >= max)
393                                 return -E2BIG;
394                         cond_resched();
395                 }
396         }
397         return i;
398 }
399
400 /*
401  * 'copy_strings()' copies argument/environment strings from the old
402  * processes's memory to the new process's stack.  The call to get_user_pages()
403  * ensures the destination page is created and not swapped out.
404  */
405 static int copy_strings(int argc, char __user * __user * argv,
406                         struct linux_binprm *bprm)
407 {
408         struct page *kmapped_page = NULL;
409         char *kaddr = NULL;
410         unsigned long kpos = 0;
411         int ret;
412
413         while (argc-- > 0) {
414                 char __user *str;
415                 int len;
416                 unsigned long pos;
417
418                 if (get_user(str, argv+argc) ||
419                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
420                         ret = -EFAULT;
421                         goto out;
422                 }
423
424                 if (!valid_arg_len(bprm, len)) {
425                         ret = -E2BIG;
426                         goto out;
427                 }
428
429                 /* We're going to work our way backwords. */
430                 pos = bprm->p;
431                 str += len;
432                 bprm->p -= len;
433
434                 while (len > 0) {
435                         int offset, bytes_to_copy;
436
437                         offset = pos % PAGE_SIZE;
438                         if (offset == 0)
439                                 offset = PAGE_SIZE;
440
441                         bytes_to_copy = offset;
442                         if (bytes_to_copy > len)
443                                 bytes_to_copy = len;
444
445                         offset -= bytes_to_copy;
446                         pos -= bytes_to_copy;
447                         str -= bytes_to_copy;
448                         len -= bytes_to_copy;
449
450                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
451                                 struct page *page;
452
453                                 page = get_arg_page(bprm, pos, 1);
454                                 if (!page) {
455                                         ret = -E2BIG;
456                                         goto out;
457                                 }
458
459                                 if (kmapped_page) {
460                                         flush_kernel_dcache_page(kmapped_page);
461                                         kunmap(kmapped_page);
462                                         put_arg_page(kmapped_page);
463                                 }
464                                 kmapped_page = page;
465                                 kaddr = kmap(kmapped_page);
466                                 kpos = pos & PAGE_MASK;
467                                 flush_arg_page(bprm, kpos, kmapped_page);
468                         }
469                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
470                                 ret = -EFAULT;
471                                 goto out;
472                         }
473                 }
474         }
475         ret = 0;
476 out:
477         if (kmapped_page) {
478                 flush_kernel_dcache_page(kmapped_page);
479                 kunmap(kmapped_page);
480                 put_arg_page(kmapped_page);
481         }
482         return ret;
483 }
484
485 /*
486  * Like copy_strings, but get argv and its values from kernel memory.
487  */
488 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
489 {
490         int r;
491         mm_segment_t oldfs = get_fs();
492         set_fs(KERNEL_DS);
493         r = copy_strings(argc, (char __user * __user *)argv, bprm);
494         set_fs(oldfs);
495         return r;
496 }
497 EXPORT_SYMBOL(copy_strings_kernel);
498
499 #ifdef CONFIG_MMU
500
501 /*
502  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
503  * the binfmt code determines where the new stack should reside, we shift it to
504  * its final location.  The process proceeds as follows:
505  *
506  * 1) Use shift to calculate the new vma endpoints.
507  * 2) Extend vma to cover both the old and new ranges.  This ensures the
508  *    arguments passed to subsequent functions are consistent.
509  * 3) Move vma's page tables to the new range.
510  * 4) Free up any cleared pgd range.
511  * 5) Shrink the vma to cover only the new range.
512  */
513 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
514 {
515         struct mm_struct *mm = vma->vm_mm;
516         unsigned long old_start = vma->vm_start;
517         unsigned long old_end = vma->vm_end;
518         unsigned long length = old_end - old_start;
519         unsigned long new_start = old_start - shift;
520         unsigned long new_end = old_end - shift;
521         struct mmu_gather *tlb;
522
523         BUG_ON(new_start > new_end);
524
525         /*
526          * ensure there are no vmas between where we want to go
527          * and where we are
528          */
529         if (vma != find_vma(mm, new_start))
530                 return -EFAULT;
531
532         /*
533          * cover the whole range: [new_start, old_end)
534          */
535         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
536
537         /*
538          * move the page tables downwards, on failure we rely on
539          * process cleanup to remove whatever mess we made.
540          */
541         if (length != move_page_tables(vma, old_start,
542                                        vma, new_start, length))
543                 return -ENOMEM;
544
545         lru_add_drain();
546         tlb = tlb_gather_mmu(mm, 0);
547         if (new_end > old_start) {
548                 /*
549                  * when the old and new regions overlap clear from new_end.
550                  */
551                 free_pgd_range(tlb, new_end, old_end, new_end,
552                         vma->vm_next ? vma->vm_next->vm_start : 0);
553         } else {
554                 /*
555                  * otherwise, clean from old_start; this is done to not touch
556                  * the address space in [new_end, old_start) some architectures
557                  * have constraints on va-space that make this illegal (IA64) -
558                  * for the others its just a little faster.
559                  */
560                 free_pgd_range(tlb, old_start, old_end, new_end,
561                         vma->vm_next ? vma->vm_next->vm_start : 0);
562         }
563         tlb_finish_mmu(tlb, new_end, old_end);
564
565         /*
566          * shrink the vma to just the new range.
567          */
568         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
569
570         return 0;
571 }
572
573 #define EXTRA_STACK_VM_PAGES    20      /* random */
574
575 /*
576  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
577  * the stack is optionally relocated, and some extra space is added.
578  */
579 int setup_arg_pages(struct linux_binprm *bprm,
580                     unsigned long stack_top,
581                     int executable_stack)
582 {
583         unsigned long ret;
584         unsigned long stack_shift;
585         struct mm_struct *mm = current->mm;
586         struct vm_area_struct *vma = bprm->vma;
587         struct vm_area_struct *prev = NULL;
588         unsigned long vm_flags;
589         unsigned long stack_base;
590
591 #ifdef CONFIG_STACK_GROWSUP
592         /* Limit stack size to 1GB */
593         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
594         if (stack_base > (1 << 30))
595                 stack_base = 1 << 30;
596
597         /* Make sure we didn't let the argument array grow too large. */
598         if (vma->vm_end - vma->vm_start > stack_base)
599                 return -ENOMEM;
600
601         stack_base = PAGE_ALIGN(stack_top - stack_base);
602
603         stack_shift = vma->vm_start - stack_base;
604         mm->arg_start = bprm->p - stack_shift;
605         bprm->p = vma->vm_end - stack_shift;
606 #else
607         stack_top = arch_align_stack(stack_top);
608         stack_top = PAGE_ALIGN(stack_top);
609         stack_shift = vma->vm_end - stack_top;
610
611         bprm->p -= stack_shift;
612         mm->arg_start = bprm->p;
613 #endif
614
615         if (bprm->loader)
616                 bprm->loader -= stack_shift;
617         bprm->exec -= stack_shift;
618
619         down_write(&mm->mmap_sem);
620         vm_flags = VM_STACK_FLAGS;
621
622         /*
623          * Adjust stack execute permissions; explicitly enable for
624          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
625          * (arch default) otherwise.
626          */
627         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
628                 vm_flags |= VM_EXEC;
629         else if (executable_stack == EXSTACK_DISABLE_X)
630                 vm_flags &= ~VM_EXEC;
631         vm_flags |= mm->def_flags;
632
633         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
634                         vm_flags);
635         if (ret)
636                 goto out_unlock;
637         BUG_ON(prev != vma);
638
639         /* Move stack pages down in memory. */
640         if (stack_shift) {
641                 ret = shift_arg_pages(vma, stack_shift);
642                 if (ret) {
643                         up_write(&mm->mmap_sem);
644                         return ret;
645                 }
646         }
647
648 #ifdef CONFIG_STACK_GROWSUP
649         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
650 #else
651         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
652 #endif
653         ret = expand_stack(vma, stack_base);
654         if (ret)
655                 ret = -EFAULT;
656
657 out_unlock:
658         up_write(&mm->mmap_sem);
659         return 0;
660 }
661 EXPORT_SYMBOL(setup_arg_pages);
662
663 #endif /* CONFIG_MMU */
664
665 struct file *open_exec(const char *name)
666 {
667         struct nameidata nd;
668         struct file *file;
669         int err;
670
671         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
672                                 FMODE_READ|FMODE_EXEC);
673         if (err)
674                 goto out;
675
676         err = -EACCES;
677         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
678                 goto out_path_put;
679
680         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
681                 goto out_path_put;
682
683         err = vfs_permission(&nd, MAY_EXEC | MAY_OPEN);
684         if (err)
685                 goto out_path_put;
686
687         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
688         if (IS_ERR(file))
689                 return file;
690
691         err = deny_write_access(file);
692         if (err) {
693                 fput(file);
694                 goto out;
695         }
696
697         return file;
698
699  out_path_put:
700         release_open_intent(&nd);
701         path_put(&nd.path);
702  out:
703         return ERR_PTR(err);
704 }
705 EXPORT_SYMBOL(open_exec);
706
707 int kernel_read(struct file *file, unsigned long offset,
708         char *addr, unsigned long count)
709 {
710         mm_segment_t old_fs;
711         loff_t pos = offset;
712         int result;
713
714         old_fs = get_fs();
715         set_fs(get_ds());
716         /* The cast to a user pointer is valid due to the set_fs() */
717         result = vfs_read(file, (void __user *)addr, count, &pos);
718         set_fs(old_fs);
719         return result;
720 }
721
722 EXPORT_SYMBOL(kernel_read);
723
724 static int exec_mmap(struct mm_struct *mm)
725 {
726         struct task_struct *tsk;
727         struct mm_struct * old_mm, *active_mm;
728
729         /* Notify parent that we're no longer interested in the old VM */
730         tsk = current;
731         old_mm = current->mm;
732         mm_release(tsk, old_mm);
733
734         if (old_mm) {
735                 /*
736                  * Make sure that if there is a core dump in progress
737                  * for the old mm, we get out and die instead of going
738                  * through with the exec.  We must hold mmap_sem around
739                  * checking core_state and changing tsk->mm.
740                  */
741                 down_read(&old_mm->mmap_sem);
742                 if (unlikely(old_mm->core_state)) {
743                         up_read(&old_mm->mmap_sem);
744                         return -EINTR;
745                 }
746         }
747         task_lock(tsk);
748         active_mm = tsk->active_mm;
749         tsk->mm = mm;
750         tsk->active_mm = mm;
751         activate_mm(active_mm, mm);
752         task_unlock(tsk);
753         arch_pick_mmap_layout(mm);
754         if (old_mm) {
755                 up_read(&old_mm->mmap_sem);
756                 BUG_ON(active_mm != old_mm);
757                 mm_update_next_owner(old_mm);
758                 mmput(old_mm);
759                 return 0;
760         }
761         mmdrop(active_mm);
762         return 0;
763 }
764
765 /*
766  * This function makes sure the current process has its own signal table,
767  * so that flush_signal_handlers can later reset the handlers without
768  * disturbing other processes.  (Other processes might share the signal
769  * table via the CLONE_SIGHAND option to clone().)
770  */
771 static int de_thread(struct task_struct *tsk)
772 {
773         struct signal_struct *sig = tsk->signal;
774         struct sighand_struct *oldsighand = tsk->sighand;
775         spinlock_t *lock = &oldsighand->siglock;
776         struct task_struct *leader = NULL;
777         int count;
778
779         if (thread_group_empty(tsk))
780                 goto no_thread_group;
781
782         /*
783          * Kill all other threads in the thread group.
784          */
785         spin_lock_irq(lock);
786         if (signal_group_exit(sig)) {
787                 /*
788                  * Another group action in progress, just
789                  * return so that the signal is processed.
790                  */
791                 spin_unlock_irq(lock);
792                 return -EAGAIN;
793         }
794         sig->group_exit_task = tsk;
795         zap_other_threads(tsk);
796
797         /* Account for the thread group leader hanging around: */
798         count = thread_group_leader(tsk) ? 1 : 2;
799         sig->notify_count = count;
800         while (atomic_read(&sig->count) > count) {
801                 __set_current_state(TASK_UNINTERRUPTIBLE);
802                 spin_unlock_irq(lock);
803                 schedule();
804                 spin_lock_irq(lock);
805         }
806         spin_unlock_irq(lock);
807
808         /*
809          * At this point all other threads have exited, all we have to
810          * do is to wait for the thread group leader to become inactive,
811          * and to assume its PID:
812          */
813         if (!thread_group_leader(tsk)) {
814                 leader = tsk->group_leader;
815
816                 sig->notify_count = -1; /* for exit_notify() */
817                 for (;;) {
818                         write_lock_irq(&tasklist_lock);
819                         if (likely(leader->exit_state))
820                                 break;
821                         __set_current_state(TASK_UNINTERRUPTIBLE);
822                         write_unlock_irq(&tasklist_lock);
823                         schedule();
824                 }
825
826                 /*
827                  * The only record we have of the real-time age of a
828                  * process, regardless of execs it's done, is start_time.
829                  * All the past CPU time is accumulated in signal_struct
830                  * from sister threads now dead.  But in this non-leader
831                  * exec, nothing survives from the original leader thread,
832                  * whose birth marks the true age of this process now.
833                  * When we take on its identity by switching to its PID, we
834                  * also take its birthdate (always earlier than our own).
835                  */
836                 tsk->start_time = leader->start_time;
837
838                 BUG_ON(!same_thread_group(leader, tsk));
839                 BUG_ON(has_group_leader_pid(tsk));
840                 /*
841                  * An exec() starts a new thread group with the
842                  * TGID of the previous thread group. Rehash the
843                  * two threads with a switched PID, and release
844                  * the former thread group leader:
845                  */
846
847                 /* Become a process group leader with the old leader's pid.
848                  * The old leader becomes a thread of the this thread group.
849                  * Note: The old leader also uses this pid until release_task
850                  *       is called.  Odd but simple and correct.
851                  */
852                 detach_pid(tsk, PIDTYPE_PID);
853                 tsk->pid = leader->pid;
854                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
855                 transfer_pid(leader, tsk, PIDTYPE_PGID);
856                 transfer_pid(leader, tsk, PIDTYPE_SID);
857                 list_replace_rcu(&leader->tasks, &tsk->tasks);
858
859                 tsk->group_leader = tsk;
860                 leader->group_leader = tsk;
861
862                 tsk->exit_signal = SIGCHLD;
863
864                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
865                 leader->exit_state = EXIT_DEAD;
866
867                 write_unlock_irq(&tasklist_lock);
868         }
869
870         sig->group_exit_task = NULL;
871         sig->notify_count = 0;
872
873 no_thread_group:
874         exit_itimers(sig);
875         flush_itimer_signals();
876         if (leader)
877                 release_task(leader);
878
879         if (atomic_read(&oldsighand->count) != 1) {
880                 struct sighand_struct *newsighand;
881                 /*
882                  * This ->sighand is shared with the CLONE_SIGHAND
883                  * but not CLONE_THREAD task, switch to the new one.
884                  */
885                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
886                 if (!newsighand)
887                         return -ENOMEM;
888
889                 atomic_set(&newsighand->count, 1);
890                 memcpy(newsighand->action, oldsighand->action,
891                        sizeof(newsighand->action));
892
893                 write_lock_irq(&tasklist_lock);
894                 spin_lock(&oldsighand->siglock);
895                 rcu_assign_pointer(tsk->sighand, newsighand);
896                 spin_unlock(&oldsighand->siglock);
897                 write_unlock_irq(&tasklist_lock);
898
899                 __cleanup_sighand(oldsighand);
900         }
901
902         BUG_ON(!thread_group_leader(tsk));
903         return 0;
904 }
905
906 /*
907  * These functions flushes out all traces of the currently running executable
908  * so that a new one can be started
909  */
910 static void flush_old_files(struct files_struct * files)
911 {
912         long j = -1;
913         struct fdtable *fdt;
914
915         spin_lock(&files->file_lock);
916         for (;;) {
917                 unsigned long set, i;
918
919                 j++;
920                 i = j * __NFDBITS;
921                 fdt = files_fdtable(files);
922                 if (i >= fdt->max_fds)
923                         break;
924                 set = fdt->close_on_exec->fds_bits[j];
925                 if (!set)
926                         continue;
927                 fdt->close_on_exec->fds_bits[j] = 0;
928                 spin_unlock(&files->file_lock);
929                 for ( ; set ; i++,set >>= 1) {
930                         if (set & 1) {
931                                 sys_close(i);
932                         }
933                 }
934                 spin_lock(&files->file_lock);
935
936         }
937         spin_unlock(&files->file_lock);
938 }
939
940 char *get_task_comm(char *buf, struct task_struct *tsk)
941 {
942         /* buf must be at least sizeof(tsk->comm) in size */
943         task_lock(tsk);
944         strncpy(buf, tsk->comm, sizeof(tsk->comm));
945         task_unlock(tsk);
946         return buf;
947 }
948
949 void set_task_comm(struct task_struct *tsk, char *buf)
950 {
951         task_lock(tsk);
952         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
953         task_unlock(tsk);
954 }
955
956 int flush_old_exec(struct linux_binprm * bprm)
957 {
958         char * name;
959         int i, ch, retval;
960         char tcomm[sizeof(current->comm)];
961
962         /*
963          * Make sure we have a private signal table and that
964          * we are unassociated from the previous thread group.
965          */
966         retval = de_thread(current);
967         if (retval)
968                 goto out;
969
970         set_mm_exe_file(bprm->mm, bprm->file);
971
972         /*
973          * Release all of the old mmap stuff
974          */
975         retval = exec_mmap(bprm->mm);
976         if (retval)
977                 goto out;
978
979         bprm->mm = NULL;                /* We're using it now */
980
981         /* This is the point of no return */
982         current->sas_ss_sp = current->sas_ss_size = 0;
983
984         if (current_euid() == current_uid() && current_egid() == current_gid())
985                 set_dumpable(current->mm, 1);
986         else
987                 set_dumpable(current->mm, suid_dumpable);
988
989         name = bprm->filename;
990
991         /* Copies the binary name from after last slash */
992         for (i=0; (ch = *(name++)) != '\0';) {
993                 if (ch == '/')
994                         i = 0; /* overwrite what we wrote */
995                 else
996                         if (i < (sizeof(tcomm) - 1))
997                                 tcomm[i++] = ch;
998         }
999         tcomm[i] = '\0';
1000         set_task_comm(current, tcomm);
1001
1002         current->flags &= ~PF_RANDOMIZE;
1003         flush_thread();
1004
1005         /* Set the new mm task size. We have to do that late because it may
1006          * depend on TIF_32BIT which is only updated in flush_thread() on
1007          * some architectures like powerpc
1008          */
1009         current->mm->task_size = TASK_SIZE;
1010
1011         /* install the new credentials */
1012         if (bprm->cred->uid != current_euid() ||
1013             bprm->cred->gid != current_egid()) {
1014                 current->pdeath_signal = 0;
1015         } else if (file_permission(bprm->file, MAY_READ) ||
1016                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1017                 set_dumpable(current->mm, suid_dumpable);
1018         }
1019
1020         current->personality &= ~bprm->per_clear;
1021
1022         /* An exec changes our domain. We are no longer part of the thread
1023            group */
1024
1025         current->self_exec_id++;
1026                         
1027         flush_signal_handlers(current, 0);
1028         flush_old_files(current->files);
1029
1030         return 0;
1031
1032 out:
1033         return retval;
1034 }
1035
1036 EXPORT_SYMBOL(flush_old_exec);
1037
1038 /*
1039  * install the new credentials for this executable
1040  */
1041 void install_exec_creds(struct linux_binprm *bprm)
1042 {
1043         security_bprm_committing_creds(bprm);
1044
1045         commit_creds(bprm->cred);
1046         bprm->cred = NULL;
1047
1048         /* cred_exec_mutex must be held at least to this point to prevent
1049          * ptrace_attach() from altering our determination of the task's
1050          * credentials; any time after this it may be unlocked */
1051
1052         security_bprm_committed_creds(bprm);
1053 }
1054 EXPORT_SYMBOL(install_exec_creds);
1055
1056 /*
1057  * determine how safe it is to execute the proposed program
1058  * - the caller must hold current->cred_exec_mutex to protect against
1059  *   PTRACE_ATTACH
1060  */
1061 void check_unsafe_exec(struct linux_binprm *bprm)
1062 {
1063         struct task_struct *p = current;
1064
1065         bprm->unsafe = tracehook_unsafe_exec(p);
1066
1067         if (atomic_read(&p->fs->count) > 1 ||
1068             atomic_read(&p->files->count) > 1 ||
1069             atomic_read(&p->sighand->count) > 1)
1070                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1071 }
1072
1073 /* 
1074  * Fill the binprm structure from the inode. 
1075  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1076  *
1077  * This may be called multiple times for binary chains (scripts for example).
1078  */
1079 int prepare_binprm(struct linux_binprm *bprm)
1080 {
1081         umode_t mode;
1082         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1083         int retval;
1084
1085         mode = inode->i_mode;
1086         if (bprm->file->f_op == NULL)
1087                 return -EACCES;
1088
1089         /* clear any previous set[ug]id data from a previous binary */
1090         bprm->cred->euid = current_euid();
1091         bprm->cred->egid = current_egid();
1092
1093         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1094                 /* Set-uid? */
1095                 if (mode & S_ISUID) {
1096                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1097                         bprm->cred->euid = inode->i_uid;
1098                 }
1099
1100                 /* Set-gid? */
1101                 /*
1102                  * If setgid is set but no group execute bit then this
1103                  * is a candidate for mandatory locking, not a setgid
1104                  * executable.
1105                  */
1106                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1107                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1108                         bprm->cred->egid = inode->i_gid;
1109                 }
1110         }
1111
1112         /* fill in binprm security blob */
1113         retval = security_bprm_set_creds(bprm);
1114         if (retval)
1115                 return retval;
1116         bprm->cred_prepared = 1;
1117
1118         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1119         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1120 }
1121
1122 EXPORT_SYMBOL(prepare_binprm);
1123
1124 /*
1125  * Arguments are '\0' separated strings found at the location bprm->p
1126  * points to; chop off the first by relocating brpm->p to right after
1127  * the first '\0' encountered.
1128  */
1129 int remove_arg_zero(struct linux_binprm *bprm)
1130 {
1131         int ret = 0;
1132         unsigned long offset;
1133         char *kaddr;
1134         struct page *page;
1135
1136         if (!bprm->argc)
1137                 return 0;
1138
1139         do {
1140                 offset = bprm->p & ~PAGE_MASK;
1141                 page = get_arg_page(bprm, bprm->p, 0);
1142                 if (!page) {
1143                         ret = -EFAULT;
1144                         goto out;
1145                 }
1146                 kaddr = kmap_atomic(page, KM_USER0);
1147
1148                 for (; offset < PAGE_SIZE && kaddr[offset];
1149                                 offset++, bprm->p++)
1150                         ;
1151
1152                 kunmap_atomic(kaddr, KM_USER0);
1153                 put_arg_page(page);
1154
1155                 if (offset == PAGE_SIZE)
1156                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1157         } while (offset == PAGE_SIZE);
1158
1159         bprm->p++;
1160         bprm->argc--;
1161         ret = 0;
1162
1163 out:
1164         return ret;
1165 }
1166 EXPORT_SYMBOL(remove_arg_zero);
1167
1168 /*
1169  * cycle the list of binary formats handler, until one recognizes the image
1170  */
1171 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1172 {
1173         int try,retval;
1174         struct linux_binfmt *fmt;
1175 #ifdef __alpha__
1176         /* handle /sbin/loader.. */
1177         {
1178             struct exec * eh = (struct exec *) bprm->buf;
1179
1180             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1181                 (eh->fh.f_flags & 0x3000) == 0x3000)
1182             {
1183                 struct file * file;
1184                 unsigned long loader;
1185
1186                 allow_write_access(bprm->file);
1187                 fput(bprm->file);
1188                 bprm->file = NULL;
1189
1190                 loader = bprm->vma->vm_end - sizeof(void *);
1191
1192                 file = open_exec("/sbin/loader");
1193                 retval = PTR_ERR(file);
1194                 if (IS_ERR(file))
1195                         return retval;
1196
1197                 /* Remember if the application is TASO.  */
1198                 bprm->taso = eh->ah.entry < 0x100000000UL;
1199
1200                 bprm->file = file;
1201                 bprm->loader = loader;
1202                 retval = prepare_binprm(bprm);
1203                 if (retval<0)
1204                         return retval;
1205                 /* should call search_binary_handler recursively here,
1206                    but it does not matter */
1207             }
1208         }
1209 #endif
1210         retval = security_bprm_check(bprm);
1211         if (retval)
1212                 return retval;
1213
1214         /* kernel module loader fixup */
1215         /* so we don't try to load run modprobe in kernel space. */
1216         set_fs(USER_DS);
1217
1218         retval = audit_bprm(bprm);
1219         if (retval)
1220                 return retval;
1221
1222         retval = -ENOENT;
1223         for (try=0; try<2; try++) {
1224                 read_lock(&binfmt_lock);
1225                 list_for_each_entry(fmt, &formats, lh) {
1226                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1227                         if (!fn)
1228                                 continue;
1229                         if (!try_module_get(fmt->module))
1230                                 continue;
1231                         read_unlock(&binfmt_lock);
1232                         retval = fn(bprm, regs);
1233                         if (retval >= 0) {
1234                                 tracehook_report_exec(fmt, bprm, regs);
1235                                 put_binfmt(fmt);
1236                                 allow_write_access(bprm->file);
1237                                 if (bprm->file)
1238                                         fput(bprm->file);
1239                                 bprm->file = NULL;
1240                                 current->did_exec = 1;
1241                                 proc_exec_connector(current);
1242                                 return retval;
1243                         }
1244                         read_lock(&binfmt_lock);
1245                         put_binfmt(fmt);
1246                         if (retval != -ENOEXEC || bprm->mm == NULL)
1247                                 break;
1248                         if (!bprm->file) {
1249                                 read_unlock(&binfmt_lock);
1250                                 return retval;
1251                         }
1252                 }
1253                 read_unlock(&binfmt_lock);
1254                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1255                         break;
1256 #ifdef CONFIG_MODULES
1257                 } else {
1258 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1259                         if (printable(bprm->buf[0]) &&
1260                             printable(bprm->buf[1]) &&
1261                             printable(bprm->buf[2]) &&
1262                             printable(bprm->buf[3]))
1263                                 break; /* -ENOEXEC */
1264                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1265 #endif
1266                 }
1267         }
1268         return retval;
1269 }
1270
1271 EXPORT_SYMBOL(search_binary_handler);
1272
1273 void free_bprm(struct linux_binprm *bprm)
1274 {
1275         free_arg_pages(bprm);
1276         if (bprm->cred)
1277                 abort_creds(bprm->cred);
1278         kfree(bprm);
1279 }
1280
1281 /*
1282  * sys_execve() executes a new program.
1283  */
1284 int do_execve(char * filename,
1285         char __user *__user *argv,
1286         char __user *__user *envp,
1287         struct pt_regs * regs)
1288 {
1289         struct linux_binprm *bprm;
1290         struct file *file;
1291         struct files_struct *displaced;
1292         int retval;
1293
1294         retval = unshare_files(&displaced);
1295         if (retval)
1296                 goto out_ret;
1297
1298         retval = -ENOMEM;
1299         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1300         if (!bprm)
1301                 goto out_files;
1302
1303         retval = mutex_lock_interruptible(&current->cred_exec_mutex);
1304         if (retval < 0)
1305                 goto out_free;
1306
1307         retval = -ENOMEM;
1308         bprm->cred = prepare_exec_creds();
1309         if (!bprm->cred)
1310                 goto out_unlock;
1311         check_unsafe_exec(bprm);
1312
1313         file = open_exec(filename);
1314         retval = PTR_ERR(file);
1315         if (IS_ERR(file))
1316                 goto out_unlock;
1317
1318         sched_exec();
1319
1320         bprm->file = file;
1321         bprm->filename = filename;
1322         bprm->interp = filename;
1323
1324         retval = bprm_mm_init(bprm);
1325         if (retval)
1326                 goto out_file;
1327
1328         bprm->argc = count(argv, MAX_ARG_STRINGS);
1329         if ((retval = bprm->argc) < 0)
1330                 goto out;
1331
1332         bprm->envc = count(envp, MAX_ARG_STRINGS);
1333         if ((retval = bprm->envc) < 0)
1334                 goto out;
1335
1336         retval = prepare_binprm(bprm);
1337         if (retval < 0)
1338                 goto out;
1339
1340         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1341         if (retval < 0)
1342                 goto out;
1343
1344         bprm->exec = bprm->p;
1345         retval = copy_strings(bprm->envc, envp, bprm);
1346         if (retval < 0)
1347                 goto out;
1348
1349         retval = copy_strings(bprm->argc, argv, bprm);
1350         if (retval < 0)
1351                 goto out;
1352
1353         current->flags &= ~PF_KTHREAD;
1354         retval = search_binary_handler(bprm,regs);
1355         if (retval < 0)
1356                 goto out;
1357
1358         /* execve succeeded */
1359         mutex_unlock(&current->cred_exec_mutex);
1360         acct_update_integrals(current);
1361         free_bprm(bprm);
1362         if (displaced)
1363                 put_files_struct(displaced);
1364         return retval;
1365
1366 out:
1367         if (bprm->mm)
1368                 mmput (bprm->mm);
1369
1370 out_file:
1371         if (bprm->file) {
1372                 allow_write_access(bprm->file);
1373                 fput(bprm->file);
1374         }
1375
1376 out_unlock:
1377         mutex_unlock(&current->cred_exec_mutex);
1378
1379 out_free:
1380         free_bprm(bprm);
1381
1382 out_files:
1383         if (displaced)
1384                 reset_files_struct(displaced);
1385 out_ret:
1386         return retval;
1387 }
1388
1389 int set_binfmt(struct linux_binfmt *new)
1390 {
1391         struct linux_binfmt *old = current->binfmt;
1392
1393         if (new) {
1394                 if (!try_module_get(new->module))
1395                         return -1;
1396         }
1397         current->binfmt = new;
1398         if (old)
1399                 module_put(old->module);
1400         return 0;
1401 }
1402
1403 EXPORT_SYMBOL(set_binfmt);
1404
1405 /* format_corename will inspect the pattern parameter, and output a
1406  * name into corename, which must have space for at least
1407  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1408  */
1409 static int format_corename(char *corename, long signr)
1410 {
1411         const struct cred *cred = current_cred();
1412         const char *pat_ptr = core_pattern;
1413         int ispipe = (*pat_ptr == '|');
1414         char *out_ptr = corename;
1415         char *const out_end = corename + CORENAME_MAX_SIZE;
1416         int rc;
1417         int pid_in_pattern = 0;
1418
1419         /* Repeat as long as we have more pattern to process and more output
1420            space */
1421         while (*pat_ptr) {
1422                 if (*pat_ptr != '%') {
1423                         if (out_ptr == out_end)
1424                                 goto out;
1425                         *out_ptr++ = *pat_ptr++;
1426                 } else {
1427                         switch (*++pat_ptr) {
1428                         case 0:
1429                                 goto out;
1430                         /* Double percent, output one percent */
1431                         case '%':
1432                                 if (out_ptr == out_end)
1433                                         goto out;
1434                                 *out_ptr++ = '%';
1435                                 break;
1436                         /* pid */
1437                         case 'p':
1438                                 pid_in_pattern = 1;
1439                                 rc = snprintf(out_ptr, out_end - out_ptr,
1440                                               "%d", task_tgid_vnr(current));
1441                                 if (rc > out_end - out_ptr)
1442                                         goto out;
1443                                 out_ptr += rc;
1444                                 break;
1445                         /* uid */
1446                         case 'u':
1447                                 rc = snprintf(out_ptr, out_end - out_ptr,
1448                                               "%d", cred->uid);
1449                                 if (rc > out_end - out_ptr)
1450                                         goto out;
1451                                 out_ptr += rc;
1452                                 break;
1453                         /* gid */
1454                         case 'g':
1455                                 rc = snprintf(out_ptr, out_end - out_ptr,
1456                                               "%d", cred->gid);
1457                                 if (rc > out_end - out_ptr)
1458                                         goto out;
1459                                 out_ptr += rc;
1460                                 break;
1461                         /* signal that caused the coredump */
1462                         case 's':
1463                                 rc = snprintf(out_ptr, out_end - out_ptr,
1464                                               "%ld", signr);
1465                                 if (rc > out_end - out_ptr)
1466                                         goto out;
1467                                 out_ptr += rc;
1468                                 break;
1469                         /* UNIX time of coredump */
1470                         case 't': {
1471                                 struct timeval tv;
1472                                 do_gettimeofday(&tv);
1473                                 rc = snprintf(out_ptr, out_end - out_ptr,
1474                                               "%lu", tv.tv_sec);
1475                                 if (rc > out_end - out_ptr)
1476                                         goto out;
1477                                 out_ptr += rc;
1478                                 break;
1479                         }
1480                         /* hostname */
1481                         case 'h':
1482                                 down_read(&uts_sem);
1483                                 rc = snprintf(out_ptr, out_end - out_ptr,
1484                                               "%s", utsname()->nodename);
1485                                 up_read(&uts_sem);
1486                                 if (rc > out_end - out_ptr)
1487                                         goto out;
1488                                 out_ptr += rc;
1489                                 break;
1490                         /* executable */
1491                         case 'e':
1492                                 rc = snprintf(out_ptr, out_end - out_ptr,
1493                                               "%s", current->comm);
1494                                 if (rc > out_end - out_ptr)
1495                                         goto out;
1496                                 out_ptr += rc;
1497                                 break;
1498                         /* core limit size */
1499                         case 'c':
1500                                 rc = snprintf(out_ptr, out_end - out_ptr,
1501                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1502                                 if (rc > out_end - out_ptr)
1503                                         goto out;
1504                                 out_ptr += rc;
1505                                 break;
1506                         default:
1507                                 break;
1508                         }
1509                         ++pat_ptr;
1510                 }
1511         }
1512         /* Backward compatibility with core_uses_pid:
1513          *
1514          * If core_pattern does not include a %p (as is the default)
1515          * and core_uses_pid is set, then .%pid will be appended to
1516          * the filename. Do not do this for piped commands. */
1517         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1518                 rc = snprintf(out_ptr, out_end - out_ptr,
1519                               ".%d", task_tgid_vnr(current));
1520                 if (rc > out_end - out_ptr)
1521                         goto out;
1522                 out_ptr += rc;
1523         }
1524 out:
1525         *out_ptr = 0;
1526         return ispipe;
1527 }
1528
1529 static int zap_process(struct task_struct *start)
1530 {
1531         struct task_struct *t;
1532         int nr = 0;
1533
1534         start->signal->flags = SIGNAL_GROUP_EXIT;
1535         start->signal->group_stop_count = 0;
1536
1537         t = start;
1538         do {
1539                 if (t != current && t->mm) {
1540                         sigaddset(&t->pending.signal, SIGKILL);
1541                         signal_wake_up(t, 1);
1542                         nr++;
1543                 }
1544         } while_each_thread(start, t);
1545
1546         return nr;
1547 }
1548
1549 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1550                                 struct core_state *core_state, int exit_code)
1551 {
1552         struct task_struct *g, *p;
1553         unsigned long flags;
1554         int nr = -EAGAIN;
1555
1556         spin_lock_irq(&tsk->sighand->siglock);
1557         if (!signal_group_exit(tsk->signal)) {
1558                 mm->core_state = core_state;
1559                 tsk->signal->group_exit_code = exit_code;
1560                 nr = zap_process(tsk);
1561         }
1562         spin_unlock_irq(&tsk->sighand->siglock);
1563         if (unlikely(nr < 0))
1564                 return nr;
1565
1566         if (atomic_read(&mm->mm_users) == nr + 1)
1567                 goto done;
1568         /*
1569          * We should find and kill all tasks which use this mm, and we should
1570          * count them correctly into ->nr_threads. We don't take tasklist
1571          * lock, but this is safe wrt:
1572          *
1573          * fork:
1574          *      None of sub-threads can fork after zap_process(leader). All
1575          *      processes which were created before this point should be
1576          *      visible to zap_threads() because copy_process() adds the new
1577          *      process to the tail of init_task.tasks list, and lock/unlock
1578          *      of ->siglock provides a memory barrier.
1579          *
1580          * do_exit:
1581          *      The caller holds mm->mmap_sem. This means that the task which
1582          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1583          *      its ->mm.
1584          *
1585          * de_thread:
1586          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1587          *      we must see either old or new leader, this does not matter.
1588          *      However, it can change p->sighand, so lock_task_sighand(p)
1589          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1590          *      it can't fail.
1591          *
1592          *      Note also that "g" can be the old leader with ->mm == NULL
1593          *      and already unhashed and thus removed from ->thread_group.
1594          *      This is OK, __unhash_process()->list_del_rcu() does not
1595          *      clear the ->next pointer, we will find the new leader via
1596          *      next_thread().
1597          */
1598         rcu_read_lock();
1599         for_each_process(g) {
1600                 if (g == tsk->group_leader)
1601                         continue;
1602                 if (g->flags & PF_KTHREAD)
1603                         continue;
1604                 p = g;
1605                 do {
1606                         if (p->mm) {
1607                                 if (unlikely(p->mm == mm)) {
1608                                         lock_task_sighand(p, &flags);
1609                                         nr += zap_process(p);
1610                                         unlock_task_sighand(p, &flags);
1611                                 }
1612                                 break;
1613                         }
1614                 } while_each_thread(g, p);
1615         }
1616         rcu_read_unlock();
1617 done:
1618         atomic_set(&core_state->nr_threads, nr);
1619         return nr;
1620 }
1621
1622 static int coredump_wait(int exit_code, struct core_state *core_state)
1623 {
1624         struct task_struct *tsk = current;
1625         struct mm_struct *mm = tsk->mm;
1626         struct completion *vfork_done;
1627         int core_waiters;
1628
1629         init_completion(&core_state->startup);
1630         core_state->dumper.task = tsk;
1631         core_state->dumper.next = NULL;
1632         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1633         up_write(&mm->mmap_sem);
1634
1635         if (unlikely(core_waiters < 0))
1636                 goto fail;
1637
1638         /*
1639          * Make sure nobody is waiting for us to release the VM,
1640          * otherwise we can deadlock when we wait on each other
1641          */
1642         vfork_done = tsk->vfork_done;
1643         if (vfork_done) {
1644                 tsk->vfork_done = NULL;
1645                 complete(vfork_done);
1646         }
1647
1648         if (core_waiters)
1649                 wait_for_completion(&core_state->startup);
1650 fail:
1651         return core_waiters;
1652 }
1653
1654 static void coredump_finish(struct mm_struct *mm)
1655 {
1656         struct core_thread *curr, *next;
1657         struct task_struct *task;
1658
1659         next = mm->core_state->dumper.next;
1660         while ((curr = next) != NULL) {
1661                 next = curr->next;
1662                 task = curr->task;
1663                 /*
1664                  * see exit_mm(), curr->task must not see
1665                  * ->task == NULL before we read ->next.
1666                  */
1667                 smp_mb();
1668                 curr->task = NULL;
1669                 wake_up_process(task);
1670         }
1671
1672         mm->core_state = NULL;
1673 }
1674
1675 /*
1676  * set_dumpable converts traditional three-value dumpable to two flags and
1677  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1678  * these bits are not changed atomically.  So get_dumpable can observe the
1679  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1680  * return either old dumpable or new one by paying attention to the order of
1681  * modifying the bits.
1682  *
1683  * dumpable |   mm->flags (binary)
1684  * old  new | initial interim  final
1685  * ---------+-----------------------
1686  *  0    1  |   00      01      01
1687  *  0    2  |   00      10(*)   11
1688  *  1    0  |   01      00      00
1689  *  1    2  |   01      11      11
1690  *  2    0  |   11      10(*)   00
1691  *  2    1  |   11      11      01
1692  *
1693  * (*) get_dumpable regards interim value of 10 as 11.
1694  */
1695 void set_dumpable(struct mm_struct *mm, int value)
1696 {
1697         switch (value) {
1698         case 0:
1699                 clear_bit(MMF_DUMPABLE, &mm->flags);
1700                 smp_wmb();
1701                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1702                 break;
1703         case 1:
1704                 set_bit(MMF_DUMPABLE, &mm->flags);
1705                 smp_wmb();
1706                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1707                 break;
1708         case 2:
1709                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1710                 smp_wmb();
1711                 set_bit(MMF_DUMPABLE, &mm->flags);
1712                 break;
1713         }
1714 }
1715
1716 int get_dumpable(struct mm_struct *mm)
1717 {
1718         int ret;
1719
1720         ret = mm->flags & 0x3;
1721         return (ret >= 2) ? 2 : ret;
1722 }
1723
1724 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1725 {
1726         struct core_state core_state;
1727         char corename[CORENAME_MAX_SIZE + 1];
1728         struct mm_struct *mm = current->mm;
1729         struct linux_binfmt * binfmt;
1730         struct inode * inode;
1731         struct file * file;
1732         const struct cred *old_cred;
1733         struct cred *cred;
1734         int retval = 0;
1735         int flag = 0;
1736         int ispipe = 0;
1737         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1738         char **helper_argv = NULL;
1739         int helper_argc = 0;
1740         char *delimit;
1741
1742         audit_core_dumps(signr);
1743
1744         binfmt = current->binfmt;
1745         if (!binfmt || !binfmt->core_dump)
1746                 goto fail;
1747
1748         cred = prepare_creds();
1749         if (!cred) {
1750                 retval = -ENOMEM;
1751                 goto fail;
1752         }
1753
1754         down_write(&mm->mmap_sem);
1755         /*
1756          * If another thread got here first, or we are not dumpable, bail out.
1757          */
1758         if (mm->core_state || !get_dumpable(mm)) {
1759                 up_write(&mm->mmap_sem);
1760                 put_cred(cred);
1761                 goto fail;
1762         }
1763
1764         /*
1765          *      We cannot trust fsuid as being the "true" uid of the
1766          *      process nor do we know its entire history. We only know it
1767          *      was tainted so we dump it as root in mode 2.
1768          */
1769         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1770                 flag = O_EXCL;          /* Stop rewrite attacks */
1771                 cred->fsuid = 0;        /* Dump root private */
1772         }
1773
1774         retval = coredump_wait(exit_code, &core_state);
1775         if (retval < 0) {
1776                 put_cred(cred);
1777                 goto fail;
1778         }
1779
1780         old_cred = override_creds(cred);
1781
1782         /*
1783          * Clear any false indication of pending signals that might
1784          * be seen by the filesystem code called to write the core file.
1785          */
1786         clear_thread_flag(TIF_SIGPENDING);
1787
1788         /*
1789          * lock_kernel() because format_corename() is controlled by sysctl, which
1790          * uses lock_kernel()
1791          */
1792         lock_kernel();
1793         ispipe = format_corename(corename, signr);
1794         unlock_kernel();
1795         /*
1796          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1797          * to a pipe.  Since we're not writing directly to the filesystem
1798          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1799          * created unless the pipe reader choses to write out the core file
1800          * at which point file size limits and permissions will be imposed
1801          * as it does with any other process
1802          */
1803         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1804                 goto fail_unlock;
1805
1806         if (ispipe) {
1807                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1808                 /* Terminate the string before the first option */
1809                 delimit = strchr(corename, ' ');
1810                 if (delimit)
1811                         *delimit = '\0';
1812                 delimit = strrchr(helper_argv[0], '/');
1813                 if (delimit)
1814                         delimit++;
1815                 else
1816                         delimit = helper_argv[0];
1817                 if (!strcmp(delimit, current->comm)) {
1818                         printk(KERN_NOTICE "Recursive core dump detected, "
1819                                         "aborting\n");
1820                         goto fail_unlock;
1821                 }
1822
1823                 core_limit = RLIM_INFINITY;
1824
1825                 /* SIGPIPE can happen, but it's just never processed */
1826                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1827                                 &file)) {
1828                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1829                                corename);
1830                         goto fail_unlock;
1831                 }
1832         } else
1833                 file = filp_open(corename,
1834                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1835                                  0600);
1836         if (IS_ERR(file))
1837                 goto fail_unlock;
1838         inode = file->f_path.dentry->d_inode;
1839         if (inode->i_nlink > 1)
1840                 goto close_fail;        /* multiple links - don't dump */
1841         if (!ispipe && d_unhashed(file->f_path.dentry))
1842                 goto close_fail;
1843
1844         /* AK: actually i see no reason to not allow this for named pipes etc.,
1845            but keep the previous behaviour for now. */
1846         if (!ispipe && !S_ISREG(inode->i_mode))
1847                 goto close_fail;
1848         /*
1849          * Dont allow local users get cute and trick others to coredump
1850          * into their pre-created files:
1851          */
1852         if (inode->i_uid != current_fsuid())
1853                 goto close_fail;
1854         if (!file->f_op)
1855                 goto close_fail;
1856         if (!file->f_op->write)
1857                 goto close_fail;
1858         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1859                 goto close_fail;
1860
1861         retval = binfmt->core_dump(signr, regs, file, core_limit);
1862
1863         if (retval)
1864                 current->signal->group_exit_code |= 0x80;
1865 close_fail:
1866         filp_close(file, NULL);
1867 fail_unlock:
1868         if (helper_argv)
1869                 argv_free(helper_argv);
1870
1871         revert_creds(old_cred);
1872         put_cred(cred);
1873         coredump_finish(mm);
1874 fail:
1875         return retval;
1876 }