Get rid of indirect include of fs_struct.h
[linux-2.6.git] / fs / exec.c
1 /*
2  *  linux/fs/exec.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  * #!-checking implemented by tytso.
9  */
10 /*
11  * Demand-loading implemented 01.12.91 - no need to read anything but
12  * the header into memory. The inode of the executable is put into
13  * "current->executable", and page faults do the actual loading. Clean.
14  *
15  * Once more I can proudly say that linux stood up to being changed: it
16  * was less than 2 hours work to get demand-loading completely implemented.
17  *
18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
19  * current->executable is only used by the procfs.  This allows a dispatch
20  * table to check for several different types  of binary formats.  We keep
21  * trying until we recognize the file or we run out of supported binary
22  * formats. 
23  */
24
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/mm.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/ima.h>
49 #include <linux/syscalls.h>
50 #include <linux/tsacct_kern.h>
51 #include <linux/cn_proc.h>
52 #include <linux/audit.h>
53 #include <linux/tracehook.h>
54 #include <linux/kmod.h>
55 #include <linux/fsnotify.h>
56 #include <linux/fs_struct.h>
57
58 #include <asm/uaccess.h>
59 #include <asm/mmu_context.h>
60 #include <asm/tlb.h>
61 #include "internal.h"
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 SYSCALL_DEFINE1(uselib, const char __user *, library)
105 {
106         struct file *file;
107         struct nameidata nd;
108         char *tmp = getname(library);
109         int error = PTR_ERR(tmp);
110
111         if (!IS_ERR(tmp)) {
112                 error = path_lookup_open(AT_FDCWD, tmp,
113                                          LOOKUP_FOLLOW, &nd,
114                                          FMODE_READ|FMODE_EXEC);
115                 putname(tmp);
116         }
117         if (error)
118                 goto out;
119
120         error = -EINVAL;
121         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
122                 goto exit;
123
124         error = -EACCES;
125         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
126                 goto exit;
127
128         error = inode_permission(nd.path.dentry->d_inode,
129                                  MAY_READ | MAY_EXEC | MAY_OPEN);
130         if (error)
131                 goto exit;
132         error = ima_path_check(&nd.path, MAY_READ | MAY_EXEC | MAY_OPEN);
133         if (error)
134                 goto exit;
135
136         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
137         error = PTR_ERR(file);
138         if (IS_ERR(file))
139                 goto out;
140
141         fsnotify_open(file->f_path.dentry);
142
143         error = -ENOEXEC;
144         if(file->f_op) {
145                 struct linux_binfmt * fmt;
146
147                 read_lock(&binfmt_lock);
148                 list_for_each_entry(fmt, &formats, lh) {
149                         if (!fmt->load_shlib)
150                                 continue;
151                         if (!try_module_get(fmt->module))
152                                 continue;
153                         read_unlock(&binfmt_lock);
154                         error = fmt->load_shlib(file);
155                         read_lock(&binfmt_lock);
156                         put_binfmt(fmt);
157                         if (error != -ENOEXEC)
158                                 break;
159                 }
160                 read_unlock(&binfmt_lock);
161         }
162         fput(file);
163 out:
164         return error;
165 exit:
166         release_open_intent(&nd);
167         path_put(&nd.path);
168         goto out;
169 }
170
171 #ifdef CONFIG_MMU
172
173 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
174                 int write)
175 {
176         struct page *page;
177         int ret;
178
179 #ifdef CONFIG_STACK_GROWSUP
180         if (write) {
181                 ret = expand_stack_downwards(bprm->vma, pos);
182                 if (ret < 0)
183                         return NULL;
184         }
185 #endif
186         ret = get_user_pages(current, bprm->mm, pos,
187                         1, write, 1, &page, NULL);
188         if (ret <= 0)
189                 return NULL;
190
191         if (write) {
192                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
193                 struct rlimit *rlim;
194
195                 /*
196                  * We've historically supported up to 32 pages (ARG_MAX)
197                  * of argument strings even with small stacks
198                  */
199                 if (size <= ARG_MAX)
200                         return page;
201
202                 /*
203                  * Limit to 1/4-th the stack size for the argv+env strings.
204                  * This ensures that:
205                  *  - the remaining binfmt code will not run out of stack space,
206                  *  - the program will have a reasonable amount of stack left
207                  *    to work from.
208                  */
209                 rlim = current->signal->rlim;
210                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
211                         put_page(page);
212                         return NULL;
213                 }
214         }
215
216         return page;
217 }
218
219 static void put_arg_page(struct page *page)
220 {
221         put_page(page);
222 }
223
224 static void free_arg_page(struct linux_binprm *bprm, int i)
225 {
226 }
227
228 static void free_arg_pages(struct linux_binprm *bprm)
229 {
230 }
231
232 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
233                 struct page *page)
234 {
235         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
236 }
237
238 static int __bprm_mm_init(struct linux_binprm *bprm)
239 {
240         int err;
241         struct vm_area_struct *vma = NULL;
242         struct mm_struct *mm = bprm->mm;
243
244         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
245         if (!vma)
246                 return -ENOMEM;
247
248         down_write(&mm->mmap_sem);
249         vma->vm_mm = mm;
250
251         /*
252          * Place the stack at the largest stack address the architecture
253          * supports. Later, we'll move this to an appropriate place. We don't
254          * use STACK_TOP because that can depend on attributes which aren't
255          * configured yet.
256          */
257         vma->vm_end = STACK_TOP_MAX;
258         vma->vm_start = vma->vm_end - PAGE_SIZE;
259         vma->vm_flags = VM_STACK_FLAGS;
260         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
261         err = insert_vm_struct(mm, vma);
262         if (err)
263                 goto err;
264
265         mm->stack_vm = mm->total_vm = 1;
266         up_write(&mm->mmap_sem);
267         bprm->p = vma->vm_end - sizeof(void *);
268         return 0;
269 err:
270         up_write(&mm->mmap_sem);
271         bprm->vma = NULL;
272         kmem_cache_free(vm_area_cachep, vma);
273         return err;
274 }
275
276 static bool valid_arg_len(struct linux_binprm *bprm, long len)
277 {
278         return len <= MAX_ARG_STRLEN;
279 }
280
281 #else
282
283 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
284                 int write)
285 {
286         struct page *page;
287
288         page = bprm->page[pos / PAGE_SIZE];
289         if (!page && write) {
290                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
291                 if (!page)
292                         return NULL;
293                 bprm->page[pos / PAGE_SIZE] = page;
294         }
295
296         return page;
297 }
298
299 static void put_arg_page(struct page *page)
300 {
301 }
302
303 static void free_arg_page(struct linux_binprm *bprm, int i)
304 {
305         if (bprm->page[i]) {
306                 __free_page(bprm->page[i]);
307                 bprm->page[i] = NULL;
308         }
309 }
310
311 static void free_arg_pages(struct linux_binprm *bprm)
312 {
313         int i;
314
315         for (i = 0; i < MAX_ARG_PAGES; i++)
316                 free_arg_page(bprm, i);
317 }
318
319 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
320                 struct page *page)
321 {
322 }
323
324 static int __bprm_mm_init(struct linux_binprm *bprm)
325 {
326         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
327         return 0;
328 }
329
330 static bool valid_arg_len(struct linux_binprm *bprm, long len)
331 {
332         return len <= bprm->p;
333 }
334
335 #endif /* CONFIG_MMU */
336
337 /*
338  * Create a new mm_struct and populate it with a temporary stack
339  * vm_area_struct.  We don't have enough context at this point to set the stack
340  * flags, permissions, and offset, so we use temporary values.  We'll update
341  * them later in setup_arg_pages().
342  */
343 int bprm_mm_init(struct linux_binprm *bprm)
344 {
345         int err;
346         struct mm_struct *mm = NULL;
347
348         bprm->mm = mm = mm_alloc();
349         err = -ENOMEM;
350         if (!mm)
351                 goto err;
352
353         err = init_new_context(current, mm);
354         if (err)
355                 goto err;
356
357         err = __bprm_mm_init(bprm);
358         if (err)
359                 goto err;
360
361         return 0;
362
363 err:
364         if (mm) {
365                 bprm->mm = NULL;
366                 mmdrop(mm);
367         }
368
369         return err;
370 }
371
372 /*
373  * count() counts the number of strings in array ARGV.
374  */
375 static int count(char __user * __user * argv, int max)
376 {
377         int i = 0;
378
379         if (argv != NULL) {
380                 for (;;) {
381                         char __user * p;
382
383                         if (get_user(p, argv))
384                                 return -EFAULT;
385                         if (!p)
386                                 break;
387                         argv++;
388                         if (i++ >= max)
389                                 return -E2BIG;
390                         cond_resched();
391                 }
392         }
393         return i;
394 }
395
396 /*
397  * 'copy_strings()' copies argument/environment strings from the old
398  * processes's memory to the new process's stack.  The call to get_user_pages()
399  * ensures the destination page is created and not swapped out.
400  */
401 static int copy_strings(int argc, char __user * __user * argv,
402                         struct linux_binprm *bprm)
403 {
404         struct page *kmapped_page = NULL;
405         char *kaddr = NULL;
406         unsigned long kpos = 0;
407         int ret;
408
409         while (argc-- > 0) {
410                 char __user *str;
411                 int len;
412                 unsigned long pos;
413
414                 if (get_user(str, argv+argc) ||
415                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
416                         ret = -EFAULT;
417                         goto out;
418                 }
419
420                 if (!valid_arg_len(bprm, len)) {
421                         ret = -E2BIG;
422                         goto out;
423                 }
424
425                 /* We're going to work our way backwords. */
426                 pos = bprm->p;
427                 str += len;
428                 bprm->p -= len;
429
430                 while (len > 0) {
431                         int offset, bytes_to_copy;
432
433                         offset = pos % PAGE_SIZE;
434                         if (offset == 0)
435                                 offset = PAGE_SIZE;
436
437                         bytes_to_copy = offset;
438                         if (bytes_to_copy > len)
439                                 bytes_to_copy = len;
440
441                         offset -= bytes_to_copy;
442                         pos -= bytes_to_copy;
443                         str -= bytes_to_copy;
444                         len -= bytes_to_copy;
445
446                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
447                                 struct page *page;
448
449                                 page = get_arg_page(bprm, pos, 1);
450                                 if (!page) {
451                                         ret = -E2BIG;
452                                         goto out;
453                                 }
454
455                                 if (kmapped_page) {
456                                         flush_kernel_dcache_page(kmapped_page);
457                                         kunmap(kmapped_page);
458                                         put_arg_page(kmapped_page);
459                                 }
460                                 kmapped_page = page;
461                                 kaddr = kmap(kmapped_page);
462                                 kpos = pos & PAGE_MASK;
463                                 flush_arg_page(bprm, kpos, kmapped_page);
464                         }
465                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
466                                 ret = -EFAULT;
467                                 goto out;
468                         }
469                 }
470         }
471         ret = 0;
472 out:
473         if (kmapped_page) {
474                 flush_kernel_dcache_page(kmapped_page);
475                 kunmap(kmapped_page);
476                 put_arg_page(kmapped_page);
477         }
478         return ret;
479 }
480
481 /*
482  * Like copy_strings, but get argv and its values from kernel memory.
483  */
484 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
485 {
486         int r;
487         mm_segment_t oldfs = get_fs();
488         set_fs(KERNEL_DS);
489         r = copy_strings(argc, (char __user * __user *)argv, bprm);
490         set_fs(oldfs);
491         return r;
492 }
493 EXPORT_SYMBOL(copy_strings_kernel);
494
495 #ifdef CONFIG_MMU
496
497 /*
498  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
499  * the binfmt code determines where the new stack should reside, we shift it to
500  * its final location.  The process proceeds as follows:
501  *
502  * 1) Use shift to calculate the new vma endpoints.
503  * 2) Extend vma to cover both the old and new ranges.  This ensures the
504  *    arguments passed to subsequent functions are consistent.
505  * 3) Move vma's page tables to the new range.
506  * 4) Free up any cleared pgd range.
507  * 5) Shrink the vma to cover only the new range.
508  */
509 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
510 {
511         struct mm_struct *mm = vma->vm_mm;
512         unsigned long old_start = vma->vm_start;
513         unsigned long old_end = vma->vm_end;
514         unsigned long length = old_end - old_start;
515         unsigned long new_start = old_start - shift;
516         unsigned long new_end = old_end - shift;
517         struct mmu_gather *tlb;
518
519         BUG_ON(new_start > new_end);
520
521         /*
522          * ensure there are no vmas between where we want to go
523          * and where we are
524          */
525         if (vma != find_vma(mm, new_start))
526                 return -EFAULT;
527
528         /*
529          * cover the whole range: [new_start, old_end)
530          */
531         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
532
533         /*
534          * move the page tables downwards, on failure we rely on
535          * process cleanup to remove whatever mess we made.
536          */
537         if (length != move_page_tables(vma, old_start,
538                                        vma, new_start, length))
539                 return -ENOMEM;
540
541         lru_add_drain();
542         tlb = tlb_gather_mmu(mm, 0);
543         if (new_end > old_start) {
544                 /*
545                  * when the old and new regions overlap clear from new_end.
546                  */
547                 free_pgd_range(tlb, new_end, old_end, new_end,
548                         vma->vm_next ? vma->vm_next->vm_start : 0);
549         } else {
550                 /*
551                  * otherwise, clean from old_start; this is done to not touch
552                  * the address space in [new_end, old_start) some architectures
553                  * have constraints on va-space that make this illegal (IA64) -
554                  * for the others its just a little faster.
555                  */
556                 free_pgd_range(tlb, old_start, old_end, new_end,
557                         vma->vm_next ? vma->vm_next->vm_start : 0);
558         }
559         tlb_finish_mmu(tlb, new_end, old_end);
560
561         /*
562          * shrink the vma to just the new range.
563          */
564         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
565
566         return 0;
567 }
568
569 #define EXTRA_STACK_VM_PAGES    20      /* random */
570
571 /*
572  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
573  * the stack is optionally relocated, and some extra space is added.
574  */
575 int setup_arg_pages(struct linux_binprm *bprm,
576                     unsigned long stack_top,
577                     int executable_stack)
578 {
579         unsigned long ret;
580         unsigned long stack_shift;
581         struct mm_struct *mm = current->mm;
582         struct vm_area_struct *vma = bprm->vma;
583         struct vm_area_struct *prev = NULL;
584         unsigned long vm_flags;
585         unsigned long stack_base;
586
587 #ifdef CONFIG_STACK_GROWSUP
588         /* Limit stack size to 1GB */
589         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
590         if (stack_base > (1 << 30))
591                 stack_base = 1 << 30;
592
593         /* Make sure we didn't let the argument array grow too large. */
594         if (vma->vm_end - vma->vm_start > stack_base)
595                 return -ENOMEM;
596
597         stack_base = PAGE_ALIGN(stack_top - stack_base);
598
599         stack_shift = vma->vm_start - stack_base;
600         mm->arg_start = bprm->p - stack_shift;
601         bprm->p = vma->vm_end - stack_shift;
602 #else
603         stack_top = arch_align_stack(stack_top);
604         stack_top = PAGE_ALIGN(stack_top);
605         stack_shift = vma->vm_end - stack_top;
606
607         bprm->p -= stack_shift;
608         mm->arg_start = bprm->p;
609 #endif
610
611         if (bprm->loader)
612                 bprm->loader -= stack_shift;
613         bprm->exec -= stack_shift;
614
615         down_write(&mm->mmap_sem);
616         vm_flags = VM_STACK_FLAGS;
617
618         /*
619          * Adjust stack execute permissions; explicitly enable for
620          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
621          * (arch default) otherwise.
622          */
623         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
624                 vm_flags |= VM_EXEC;
625         else if (executable_stack == EXSTACK_DISABLE_X)
626                 vm_flags &= ~VM_EXEC;
627         vm_flags |= mm->def_flags;
628
629         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
630                         vm_flags);
631         if (ret)
632                 goto out_unlock;
633         BUG_ON(prev != vma);
634
635         /* Move stack pages down in memory. */
636         if (stack_shift) {
637                 ret = shift_arg_pages(vma, stack_shift);
638                 if (ret) {
639                         up_write(&mm->mmap_sem);
640                         return ret;
641                 }
642         }
643
644 #ifdef CONFIG_STACK_GROWSUP
645         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
646 #else
647         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
648 #endif
649         ret = expand_stack(vma, stack_base);
650         if (ret)
651                 ret = -EFAULT;
652
653 out_unlock:
654         up_write(&mm->mmap_sem);
655         return 0;
656 }
657 EXPORT_SYMBOL(setup_arg_pages);
658
659 #endif /* CONFIG_MMU */
660
661 struct file *open_exec(const char *name)
662 {
663         struct nameidata nd;
664         struct file *file;
665         int err;
666
667         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
668                                 FMODE_READ|FMODE_EXEC);
669         if (err)
670                 goto out;
671
672         err = -EACCES;
673         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
674                 goto out_path_put;
675
676         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
677                 goto out_path_put;
678
679         err = inode_permission(nd.path.dentry->d_inode, MAY_EXEC | MAY_OPEN);
680         if (err)
681                 goto out_path_put;
682         err = ima_path_check(&nd.path, MAY_EXEC | MAY_OPEN);
683         if (err)
684                 goto out_path_put;
685
686         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
687         if (IS_ERR(file))
688                 return file;
689
690         fsnotify_open(file->f_path.dentry);
691
692         err = deny_write_access(file);
693         if (err) {
694                 fput(file);
695                 goto out;
696         }
697
698         return file;
699
700  out_path_put:
701         release_open_intent(&nd);
702         path_put(&nd.path);
703  out:
704         return ERR_PTR(err);
705 }
706 EXPORT_SYMBOL(open_exec);
707
708 int kernel_read(struct file *file, unsigned long offset,
709         char *addr, unsigned long count)
710 {
711         mm_segment_t old_fs;
712         loff_t pos = offset;
713         int result;
714
715         old_fs = get_fs();
716         set_fs(get_ds());
717         /* The cast to a user pointer is valid due to the set_fs() */
718         result = vfs_read(file, (void __user *)addr, count, &pos);
719         set_fs(old_fs);
720         return result;
721 }
722
723 EXPORT_SYMBOL(kernel_read);
724
725 static int exec_mmap(struct mm_struct *mm)
726 {
727         struct task_struct *tsk;
728         struct mm_struct * old_mm, *active_mm;
729
730         /* Notify parent that we're no longer interested in the old VM */
731         tsk = current;
732         old_mm = current->mm;
733         mm_release(tsk, old_mm);
734
735         if (old_mm) {
736                 /*
737                  * Make sure that if there is a core dump in progress
738                  * for the old mm, we get out and die instead of going
739                  * through with the exec.  We must hold mmap_sem around
740                  * checking core_state and changing tsk->mm.
741                  */
742                 down_read(&old_mm->mmap_sem);
743                 if (unlikely(old_mm->core_state)) {
744                         up_read(&old_mm->mmap_sem);
745                         return -EINTR;
746                 }
747         }
748         task_lock(tsk);
749         active_mm = tsk->active_mm;
750         tsk->mm = mm;
751         tsk->active_mm = mm;
752         activate_mm(active_mm, mm);
753         task_unlock(tsk);
754         arch_pick_mmap_layout(mm);
755         if (old_mm) {
756                 up_read(&old_mm->mmap_sem);
757                 BUG_ON(active_mm != old_mm);
758                 mm_update_next_owner(old_mm);
759                 mmput(old_mm);
760                 return 0;
761         }
762         mmdrop(active_mm);
763         return 0;
764 }
765
766 /*
767  * This function makes sure the current process has its own signal table,
768  * so that flush_signal_handlers can later reset the handlers without
769  * disturbing other processes.  (Other processes might share the signal
770  * table via the CLONE_SIGHAND option to clone().)
771  */
772 static int de_thread(struct task_struct *tsk)
773 {
774         struct signal_struct *sig = tsk->signal;
775         struct sighand_struct *oldsighand = tsk->sighand;
776         spinlock_t *lock = &oldsighand->siglock;
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                 struct task_struct *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                 write_unlock_irq(&tasklist_lock);
867
868                 release_task(leader);
869         }
870
871         sig->group_exit_task = NULL;
872         sig->notify_count = 0;
873
874 no_thread_group:
875         exit_itimers(sig);
876         flush_itimer_signals();
877
878         if (atomic_read(&oldsighand->count) != 1) {
879                 struct sighand_struct *newsighand;
880                 /*
881                  * This ->sighand is shared with the CLONE_SIGHAND
882                  * but not CLONE_THREAD task, switch to the new one.
883                  */
884                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
885                 if (!newsighand)
886                         return -ENOMEM;
887
888                 atomic_set(&newsighand->count, 1);
889                 memcpy(newsighand->action, oldsighand->action,
890                        sizeof(newsighand->action));
891
892                 write_lock_irq(&tasklist_lock);
893                 spin_lock(&oldsighand->siglock);
894                 rcu_assign_pointer(tsk->sighand, newsighand);
895                 spin_unlock(&oldsighand->siglock);
896                 write_unlock_irq(&tasklist_lock);
897
898                 __cleanup_sighand(oldsighand);
899         }
900
901         BUG_ON(!thread_group_leader(tsk));
902         return 0;
903 }
904
905 /*
906  * These functions flushes out all traces of the currently running executable
907  * so that a new one can be started
908  */
909 static void flush_old_files(struct files_struct * files)
910 {
911         long j = -1;
912         struct fdtable *fdt;
913
914         spin_lock(&files->file_lock);
915         for (;;) {
916                 unsigned long set, i;
917
918                 j++;
919                 i = j * __NFDBITS;
920                 fdt = files_fdtable(files);
921                 if (i >= fdt->max_fds)
922                         break;
923                 set = fdt->close_on_exec->fds_bits[j];
924                 if (!set)
925                         continue;
926                 fdt->close_on_exec->fds_bits[j] = 0;
927                 spin_unlock(&files->file_lock);
928                 for ( ; set ; i++,set >>= 1) {
929                         if (set & 1) {
930                                 sys_close(i);
931                         }
932                 }
933                 spin_lock(&files->file_lock);
934
935         }
936         spin_unlock(&files->file_lock);
937 }
938
939 char *get_task_comm(char *buf, struct task_struct *tsk)
940 {
941         /* buf must be at least sizeof(tsk->comm) in size */
942         task_lock(tsk);
943         strncpy(buf, tsk->comm, sizeof(tsk->comm));
944         task_unlock(tsk);
945         return buf;
946 }
947
948 void set_task_comm(struct task_struct *tsk, char *buf)
949 {
950         task_lock(tsk);
951         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
952         task_unlock(tsk);
953 }
954
955 int flush_old_exec(struct linux_binprm * bprm)
956 {
957         char * name;
958         int i, ch, retval;
959         char tcomm[sizeof(current->comm)];
960
961         /*
962          * Make sure we have a private signal table and that
963          * we are unassociated from the previous thread group.
964          */
965         retval = de_thread(current);
966         if (retval)
967                 goto out;
968
969         set_mm_exe_file(bprm->mm, bprm->file);
970
971         /*
972          * Release all of the old mmap stuff
973          */
974         retval = exec_mmap(bprm->mm);
975         if (retval)
976                 goto out;
977
978         bprm->mm = NULL;                /* We're using it now */
979
980         /* This is the point of no return */
981         current->sas_ss_sp = current->sas_ss_size = 0;
982
983         if (current_euid() == current_uid() && current_egid() == current_gid())
984                 set_dumpable(current->mm, 1);
985         else
986                 set_dumpable(current->mm, suid_dumpable);
987
988         name = bprm->filename;
989
990         /* Copies the binary name from after last slash */
991         for (i=0; (ch = *(name++)) != '\0';) {
992                 if (ch == '/')
993                         i = 0; /* overwrite what we wrote */
994                 else
995                         if (i < (sizeof(tcomm) - 1))
996                                 tcomm[i++] = ch;
997         }
998         tcomm[i] = '\0';
999         set_task_comm(current, tcomm);
1000
1001         current->flags &= ~PF_RANDOMIZE;
1002         flush_thread();
1003
1004         /* Set the new mm task size. We have to do that late because it may
1005          * depend on TIF_32BIT which is only updated in flush_thread() on
1006          * some architectures like powerpc
1007          */
1008         current->mm->task_size = TASK_SIZE;
1009
1010         /* install the new credentials */
1011         if (bprm->cred->uid != current_euid() ||
1012             bprm->cred->gid != current_egid()) {
1013                 current->pdeath_signal = 0;
1014         } else if (file_permission(bprm->file, MAY_READ) ||
1015                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1016                 set_dumpable(current->mm, suid_dumpable);
1017         }
1018
1019         current->personality &= ~bprm->per_clear;
1020
1021         /* An exec changes our domain. We are no longer part of the thread
1022            group */
1023
1024         current->self_exec_id++;
1025                         
1026         flush_signal_handlers(current, 0);
1027         flush_old_files(current->files);
1028
1029         return 0;
1030
1031 out:
1032         return retval;
1033 }
1034
1035 EXPORT_SYMBOL(flush_old_exec);
1036
1037 /*
1038  * install the new credentials for this executable
1039  */
1040 void install_exec_creds(struct linux_binprm *bprm)
1041 {
1042         security_bprm_committing_creds(bprm);
1043
1044         commit_creds(bprm->cred);
1045         bprm->cred = NULL;
1046
1047         /* cred_exec_mutex must be held at least to this point to prevent
1048          * ptrace_attach() from altering our determination of the task's
1049          * credentials; any time after this it may be unlocked */
1050
1051         security_bprm_committed_creds(bprm);
1052 }
1053 EXPORT_SYMBOL(install_exec_creds);
1054
1055 /*
1056  * determine how safe it is to execute the proposed program
1057  * - the caller must hold current->cred_exec_mutex to protect against
1058  *   PTRACE_ATTACH
1059  */
1060 int check_unsafe_exec(struct linux_binprm *bprm)
1061 {
1062         struct task_struct *p = current, *t;
1063         unsigned long flags;
1064         unsigned n_fs;
1065         int res = 0;
1066
1067         bprm->unsafe = tracehook_unsafe_exec(p);
1068
1069         n_fs = 1;
1070         write_lock(&p->fs->lock);
1071         lock_task_sighand(p, &flags);
1072         for (t = next_thread(p); t != p; t = next_thread(t)) {
1073                 if (t->fs == p->fs)
1074                         n_fs++;
1075         }
1076
1077         if (p->fs->users > n_fs) {
1078                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1079         } else {
1080                 if (p->fs->in_exec)
1081                         res = -EAGAIN;
1082                 p->fs->in_exec = 1;
1083         }
1084
1085         unlock_task_sighand(p, &flags);
1086         write_unlock(&p->fs->lock);
1087
1088         return res;
1089 }
1090
1091 /* 
1092  * Fill the binprm structure from the inode. 
1093  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1094  *
1095  * This may be called multiple times for binary chains (scripts for example).
1096  */
1097 int prepare_binprm(struct linux_binprm *bprm)
1098 {
1099         umode_t mode;
1100         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1101         int retval;
1102
1103         mode = inode->i_mode;
1104         if (bprm->file->f_op == NULL)
1105                 return -EACCES;
1106
1107         /* clear any previous set[ug]id data from a previous binary */
1108         bprm->cred->euid = current_euid();
1109         bprm->cred->egid = current_egid();
1110
1111         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1112                 /* Set-uid? */
1113                 if (mode & S_ISUID) {
1114                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1115                         bprm->cred->euid = inode->i_uid;
1116                 }
1117
1118                 /* Set-gid? */
1119                 /*
1120                  * If setgid is set but no group execute bit then this
1121                  * is a candidate for mandatory locking, not a setgid
1122                  * executable.
1123                  */
1124                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1125                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1126                         bprm->cred->egid = inode->i_gid;
1127                 }
1128         }
1129
1130         /* fill in binprm security blob */
1131         retval = security_bprm_set_creds(bprm);
1132         if (retval)
1133                 return retval;
1134         bprm->cred_prepared = 1;
1135
1136         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1137         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1138 }
1139
1140 EXPORT_SYMBOL(prepare_binprm);
1141
1142 /*
1143  * Arguments are '\0' separated strings found at the location bprm->p
1144  * points to; chop off the first by relocating brpm->p to right after
1145  * the first '\0' encountered.
1146  */
1147 int remove_arg_zero(struct linux_binprm *bprm)
1148 {
1149         int ret = 0;
1150         unsigned long offset;
1151         char *kaddr;
1152         struct page *page;
1153
1154         if (!bprm->argc)
1155                 return 0;
1156
1157         do {
1158                 offset = bprm->p & ~PAGE_MASK;
1159                 page = get_arg_page(bprm, bprm->p, 0);
1160                 if (!page) {
1161                         ret = -EFAULT;
1162                         goto out;
1163                 }
1164                 kaddr = kmap_atomic(page, KM_USER0);
1165
1166                 for (; offset < PAGE_SIZE && kaddr[offset];
1167                                 offset++, bprm->p++)
1168                         ;
1169
1170                 kunmap_atomic(kaddr, KM_USER0);
1171                 put_arg_page(page);
1172
1173                 if (offset == PAGE_SIZE)
1174                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1175         } while (offset == PAGE_SIZE);
1176
1177         bprm->p++;
1178         bprm->argc--;
1179         ret = 0;
1180
1181 out:
1182         return ret;
1183 }
1184 EXPORT_SYMBOL(remove_arg_zero);
1185
1186 /*
1187  * cycle the list of binary formats handler, until one recognizes the image
1188  */
1189 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1190 {
1191         unsigned int depth = bprm->recursion_depth;
1192         int try,retval;
1193         struct linux_binfmt *fmt;
1194
1195         retval = security_bprm_check(bprm);
1196         if (retval)
1197                 return retval;
1198         retval = ima_bprm_check(bprm);
1199         if (retval)
1200                 return retval;
1201
1202         /* kernel module loader fixup */
1203         /* so we don't try to load run modprobe in kernel space. */
1204         set_fs(USER_DS);
1205
1206         retval = audit_bprm(bprm);
1207         if (retval)
1208                 return retval;
1209
1210         retval = -ENOENT;
1211         for (try=0; try<2; try++) {
1212                 read_lock(&binfmt_lock);
1213                 list_for_each_entry(fmt, &formats, lh) {
1214                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1215                         if (!fn)
1216                                 continue;
1217                         if (!try_module_get(fmt->module))
1218                                 continue;
1219                         read_unlock(&binfmt_lock);
1220                         retval = fn(bprm, regs);
1221                         /*
1222                          * Restore the depth counter to its starting value
1223                          * in this call, so we don't have to rely on every
1224                          * load_binary function to restore it on return.
1225                          */
1226                         bprm->recursion_depth = depth;
1227                         if (retval >= 0) {
1228                                 if (depth == 0)
1229                                         tracehook_report_exec(fmt, bprm, regs);
1230                                 put_binfmt(fmt);
1231                                 allow_write_access(bprm->file);
1232                                 if (bprm->file)
1233                                         fput(bprm->file);
1234                                 bprm->file = NULL;
1235                                 current->did_exec = 1;
1236                                 proc_exec_connector(current);
1237                                 return retval;
1238                         }
1239                         read_lock(&binfmt_lock);
1240                         put_binfmt(fmt);
1241                         if (retval != -ENOEXEC || bprm->mm == NULL)
1242                                 break;
1243                         if (!bprm->file) {
1244                                 read_unlock(&binfmt_lock);
1245                                 return retval;
1246                         }
1247                 }
1248                 read_unlock(&binfmt_lock);
1249                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1250                         break;
1251 #ifdef CONFIG_MODULES
1252                 } else {
1253 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1254                         if (printable(bprm->buf[0]) &&
1255                             printable(bprm->buf[1]) &&
1256                             printable(bprm->buf[2]) &&
1257                             printable(bprm->buf[3]))
1258                                 break; /* -ENOEXEC */
1259                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1260 #endif
1261                 }
1262         }
1263         return retval;
1264 }
1265
1266 EXPORT_SYMBOL(search_binary_handler);
1267
1268 void free_bprm(struct linux_binprm *bprm)
1269 {
1270         free_arg_pages(bprm);
1271         if (bprm->cred)
1272                 abort_creds(bprm->cred);
1273         kfree(bprm);
1274 }
1275
1276 /*
1277  * sys_execve() executes a new program.
1278  */
1279 int do_execve(char * filename,
1280         char __user *__user *argv,
1281         char __user *__user *envp,
1282         struct pt_regs * regs)
1283 {
1284         struct linux_binprm *bprm;
1285         struct file *file;
1286         struct files_struct *displaced;
1287         int retval;
1288
1289         retval = unshare_files(&displaced);
1290         if (retval)
1291                 goto out_ret;
1292
1293         retval = -ENOMEM;
1294         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1295         if (!bprm)
1296                 goto out_files;
1297
1298         retval = mutex_lock_interruptible(&current->cred_exec_mutex);
1299         if (retval < 0)
1300                 goto out_free;
1301         current->in_execve = 1;
1302
1303         retval = -ENOMEM;
1304         bprm->cred = prepare_exec_creds();
1305         if (!bprm->cred)
1306                 goto out_unlock;
1307
1308         retval = check_unsafe_exec(bprm);
1309         if (retval)
1310                 goto out_unlock;
1311
1312         file = open_exec(filename);
1313         retval = PTR_ERR(file);
1314         if (IS_ERR(file))
1315                 goto out_unmark;
1316
1317         sched_exec();
1318
1319         bprm->file = file;
1320         bprm->filename = filename;
1321         bprm->interp = filename;
1322
1323         retval = bprm_mm_init(bprm);
1324         if (retval)
1325                 goto out_file;
1326
1327         bprm->argc = count(argv, MAX_ARG_STRINGS);
1328         if ((retval = bprm->argc) < 0)
1329                 goto out;
1330
1331         bprm->envc = count(envp, MAX_ARG_STRINGS);
1332         if ((retval = bprm->envc) < 0)
1333                 goto out;
1334
1335         retval = prepare_binprm(bprm);
1336         if (retval < 0)
1337                 goto out;
1338
1339         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1340         if (retval < 0)
1341                 goto out;
1342
1343         bprm->exec = bprm->p;
1344         retval = copy_strings(bprm->envc, envp, bprm);
1345         if (retval < 0)
1346                 goto out;
1347
1348         retval = copy_strings(bprm->argc, argv, bprm);
1349         if (retval < 0)
1350                 goto out;
1351
1352         current->flags &= ~PF_KTHREAD;
1353         retval = search_binary_handler(bprm,regs);
1354         if (retval < 0)
1355                 goto out;
1356
1357         /* execve succeeded */
1358         write_lock(&current->fs->lock);
1359         current->fs->in_exec = 0;
1360         write_unlock(&current->fs->lock);
1361         current->in_execve = 0;
1362         mutex_unlock(&current->cred_exec_mutex);
1363         acct_update_integrals(current);
1364         free_bprm(bprm);
1365         if (displaced)
1366                 put_files_struct(displaced);
1367         return retval;
1368
1369 out:
1370         if (bprm->mm)
1371                 mmput (bprm->mm);
1372
1373 out_file:
1374         if (bprm->file) {
1375                 allow_write_access(bprm->file);
1376                 fput(bprm->file);
1377         }
1378
1379 out_unmark:
1380         write_lock(&current->fs->lock);
1381         current->fs->in_exec = 0;
1382         write_unlock(&current->fs->lock);
1383
1384 out_unlock:
1385         current->in_execve = 0;
1386         mutex_unlock(&current->cred_exec_mutex);
1387
1388 out_free:
1389         free_bprm(bprm);
1390
1391 out_files:
1392         if (displaced)
1393                 reset_files_struct(displaced);
1394 out_ret:
1395         return retval;
1396 }
1397
1398 int set_binfmt(struct linux_binfmt *new)
1399 {
1400         struct linux_binfmt *old = current->binfmt;
1401
1402         if (new) {
1403                 if (!try_module_get(new->module))
1404                         return -1;
1405         }
1406         current->binfmt = new;
1407         if (old)
1408                 module_put(old->module);
1409         return 0;
1410 }
1411
1412 EXPORT_SYMBOL(set_binfmt);
1413
1414 /* format_corename will inspect the pattern parameter, and output a
1415  * name into corename, which must have space for at least
1416  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1417  */
1418 static int format_corename(char *corename, long signr)
1419 {
1420         const struct cred *cred = current_cred();
1421         const char *pat_ptr = core_pattern;
1422         int ispipe = (*pat_ptr == '|');
1423         char *out_ptr = corename;
1424         char *const out_end = corename + CORENAME_MAX_SIZE;
1425         int rc;
1426         int pid_in_pattern = 0;
1427
1428         /* Repeat as long as we have more pattern to process and more output
1429            space */
1430         while (*pat_ptr) {
1431                 if (*pat_ptr != '%') {
1432                         if (out_ptr == out_end)
1433                                 goto out;
1434                         *out_ptr++ = *pat_ptr++;
1435                 } else {
1436                         switch (*++pat_ptr) {
1437                         case 0:
1438                                 goto out;
1439                         /* Double percent, output one percent */
1440                         case '%':
1441                                 if (out_ptr == out_end)
1442                                         goto out;
1443                                 *out_ptr++ = '%';
1444                                 break;
1445                         /* pid */
1446                         case 'p':
1447                                 pid_in_pattern = 1;
1448                                 rc = snprintf(out_ptr, out_end - out_ptr,
1449                                               "%d", task_tgid_vnr(current));
1450                                 if (rc > out_end - out_ptr)
1451                                         goto out;
1452                                 out_ptr += rc;
1453                                 break;
1454                         /* uid */
1455                         case 'u':
1456                                 rc = snprintf(out_ptr, out_end - out_ptr,
1457                                               "%d", cred->uid);
1458                                 if (rc > out_end - out_ptr)
1459                                         goto out;
1460                                 out_ptr += rc;
1461                                 break;
1462                         /* gid */
1463                         case 'g':
1464                                 rc = snprintf(out_ptr, out_end - out_ptr,
1465                                               "%d", cred->gid);
1466                                 if (rc > out_end - out_ptr)
1467                                         goto out;
1468                                 out_ptr += rc;
1469                                 break;
1470                         /* signal that caused the coredump */
1471                         case 's':
1472                                 rc = snprintf(out_ptr, out_end - out_ptr,
1473                                               "%ld", signr);
1474                                 if (rc > out_end - out_ptr)
1475                                         goto out;
1476                                 out_ptr += rc;
1477                                 break;
1478                         /* UNIX time of coredump */
1479                         case 't': {
1480                                 struct timeval tv;
1481                                 do_gettimeofday(&tv);
1482                                 rc = snprintf(out_ptr, out_end - out_ptr,
1483                                               "%lu", tv.tv_sec);
1484                                 if (rc > out_end - out_ptr)
1485                                         goto out;
1486                                 out_ptr += rc;
1487                                 break;
1488                         }
1489                         /* hostname */
1490                         case 'h':
1491                                 down_read(&uts_sem);
1492                                 rc = snprintf(out_ptr, out_end - out_ptr,
1493                                               "%s", utsname()->nodename);
1494                                 up_read(&uts_sem);
1495                                 if (rc > out_end - out_ptr)
1496                                         goto out;
1497                                 out_ptr += rc;
1498                                 break;
1499                         /* executable */
1500                         case 'e':
1501                                 rc = snprintf(out_ptr, out_end - out_ptr,
1502                                               "%s", current->comm);
1503                                 if (rc > out_end - out_ptr)
1504                                         goto out;
1505                                 out_ptr += rc;
1506                                 break;
1507                         /* core limit size */
1508                         case 'c':
1509                                 rc = snprintf(out_ptr, out_end - out_ptr,
1510                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1511                                 if (rc > out_end - out_ptr)
1512                                         goto out;
1513                                 out_ptr += rc;
1514                                 break;
1515                         default:
1516                                 break;
1517                         }
1518                         ++pat_ptr;
1519                 }
1520         }
1521         /* Backward compatibility with core_uses_pid:
1522          *
1523          * If core_pattern does not include a %p (as is the default)
1524          * and core_uses_pid is set, then .%pid will be appended to
1525          * the filename. Do not do this for piped commands. */
1526         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1527                 rc = snprintf(out_ptr, out_end - out_ptr,
1528                               ".%d", task_tgid_vnr(current));
1529                 if (rc > out_end - out_ptr)
1530                         goto out;
1531                 out_ptr += rc;
1532         }
1533 out:
1534         *out_ptr = 0;
1535         return ispipe;
1536 }
1537
1538 static int zap_process(struct task_struct *start)
1539 {
1540         struct task_struct *t;
1541         int nr = 0;
1542
1543         start->signal->flags = SIGNAL_GROUP_EXIT;
1544         start->signal->group_stop_count = 0;
1545
1546         t = start;
1547         do {
1548                 if (t != current && t->mm) {
1549                         sigaddset(&t->pending.signal, SIGKILL);
1550                         signal_wake_up(t, 1);
1551                         nr++;
1552                 }
1553         } while_each_thread(start, t);
1554
1555         return nr;
1556 }
1557
1558 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1559                                 struct core_state *core_state, int exit_code)
1560 {
1561         struct task_struct *g, *p;
1562         unsigned long flags;
1563         int nr = -EAGAIN;
1564
1565         spin_lock_irq(&tsk->sighand->siglock);
1566         if (!signal_group_exit(tsk->signal)) {
1567                 mm->core_state = core_state;
1568                 tsk->signal->group_exit_code = exit_code;
1569                 nr = zap_process(tsk);
1570         }
1571         spin_unlock_irq(&tsk->sighand->siglock);
1572         if (unlikely(nr < 0))
1573                 return nr;
1574
1575         if (atomic_read(&mm->mm_users) == nr + 1)
1576                 goto done;
1577         /*
1578          * We should find and kill all tasks which use this mm, and we should
1579          * count them correctly into ->nr_threads. We don't take tasklist
1580          * lock, but this is safe wrt:
1581          *
1582          * fork:
1583          *      None of sub-threads can fork after zap_process(leader). All
1584          *      processes which were created before this point should be
1585          *      visible to zap_threads() because copy_process() adds the new
1586          *      process to the tail of init_task.tasks list, and lock/unlock
1587          *      of ->siglock provides a memory barrier.
1588          *
1589          * do_exit:
1590          *      The caller holds mm->mmap_sem. This means that the task which
1591          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1592          *      its ->mm.
1593          *
1594          * de_thread:
1595          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1596          *      we must see either old or new leader, this does not matter.
1597          *      However, it can change p->sighand, so lock_task_sighand(p)
1598          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1599          *      it can't fail.
1600          *
1601          *      Note also that "g" can be the old leader with ->mm == NULL
1602          *      and already unhashed and thus removed from ->thread_group.
1603          *      This is OK, __unhash_process()->list_del_rcu() does not
1604          *      clear the ->next pointer, we will find the new leader via
1605          *      next_thread().
1606          */
1607         rcu_read_lock();
1608         for_each_process(g) {
1609                 if (g == tsk->group_leader)
1610                         continue;
1611                 if (g->flags & PF_KTHREAD)
1612                         continue;
1613                 p = g;
1614                 do {
1615                         if (p->mm) {
1616                                 if (unlikely(p->mm == mm)) {
1617                                         lock_task_sighand(p, &flags);
1618                                         nr += zap_process(p);
1619                                         unlock_task_sighand(p, &flags);
1620                                 }
1621                                 break;
1622                         }
1623                 } while_each_thread(g, p);
1624         }
1625         rcu_read_unlock();
1626 done:
1627         atomic_set(&core_state->nr_threads, nr);
1628         return nr;
1629 }
1630
1631 static int coredump_wait(int exit_code, struct core_state *core_state)
1632 {
1633         struct task_struct *tsk = current;
1634         struct mm_struct *mm = tsk->mm;
1635         struct completion *vfork_done;
1636         int core_waiters;
1637
1638         init_completion(&core_state->startup);
1639         core_state->dumper.task = tsk;
1640         core_state->dumper.next = NULL;
1641         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1642         up_write(&mm->mmap_sem);
1643
1644         if (unlikely(core_waiters < 0))
1645                 goto fail;
1646
1647         /*
1648          * Make sure nobody is waiting for us to release the VM,
1649          * otherwise we can deadlock when we wait on each other
1650          */
1651         vfork_done = tsk->vfork_done;
1652         if (vfork_done) {
1653                 tsk->vfork_done = NULL;
1654                 complete(vfork_done);
1655         }
1656
1657         if (core_waiters)
1658                 wait_for_completion(&core_state->startup);
1659 fail:
1660         return core_waiters;
1661 }
1662
1663 static void coredump_finish(struct mm_struct *mm)
1664 {
1665         struct core_thread *curr, *next;
1666         struct task_struct *task;
1667
1668         next = mm->core_state->dumper.next;
1669         while ((curr = next) != NULL) {
1670                 next = curr->next;
1671                 task = curr->task;
1672                 /*
1673                  * see exit_mm(), curr->task must not see
1674                  * ->task == NULL before we read ->next.
1675                  */
1676                 smp_mb();
1677                 curr->task = NULL;
1678                 wake_up_process(task);
1679         }
1680
1681         mm->core_state = NULL;
1682 }
1683
1684 /*
1685  * set_dumpable converts traditional three-value dumpable to two flags and
1686  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1687  * these bits are not changed atomically.  So get_dumpable can observe the
1688  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1689  * return either old dumpable or new one by paying attention to the order of
1690  * modifying the bits.
1691  *
1692  * dumpable |   mm->flags (binary)
1693  * old  new | initial interim  final
1694  * ---------+-----------------------
1695  *  0    1  |   00      01      01
1696  *  0    2  |   00      10(*)   11
1697  *  1    0  |   01      00      00
1698  *  1    2  |   01      11      11
1699  *  2    0  |   11      10(*)   00
1700  *  2    1  |   11      11      01
1701  *
1702  * (*) get_dumpable regards interim value of 10 as 11.
1703  */
1704 void set_dumpable(struct mm_struct *mm, int value)
1705 {
1706         switch (value) {
1707         case 0:
1708                 clear_bit(MMF_DUMPABLE, &mm->flags);
1709                 smp_wmb();
1710                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1711                 break;
1712         case 1:
1713                 set_bit(MMF_DUMPABLE, &mm->flags);
1714                 smp_wmb();
1715                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1716                 break;
1717         case 2:
1718                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1719                 smp_wmb();
1720                 set_bit(MMF_DUMPABLE, &mm->flags);
1721                 break;
1722         }
1723 }
1724
1725 int get_dumpable(struct mm_struct *mm)
1726 {
1727         int ret;
1728
1729         ret = mm->flags & 0x3;
1730         return (ret >= 2) ? 2 : ret;
1731 }
1732
1733 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1734 {
1735         struct core_state core_state;
1736         char corename[CORENAME_MAX_SIZE + 1];
1737         struct mm_struct *mm = current->mm;
1738         struct linux_binfmt * binfmt;
1739         struct inode * inode;
1740         struct file * file;
1741         const struct cred *old_cred;
1742         struct cred *cred;
1743         int retval = 0;
1744         int flag = 0;
1745         int ispipe = 0;
1746         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1747         char **helper_argv = NULL;
1748         int helper_argc = 0;
1749         char *delimit;
1750
1751         audit_core_dumps(signr);
1752
1753         binfmt = current->binfmt;
1754         if (!binfmt || !binfmt->core_dump)
1755                 goto fail;
1756
1757         cred = prepare_creds();
1758         if (!cred) {
1759                 retval = -ENOMEM;
1760                 goto fail;
1761         }
1762
1763         down_write(&mm->mmap_sem);
1764         /*
1765          * If another thread got here first, or we are not dumpable, bail out.
1766          */
1767         if (mm->core_state || !get_dumpable(mm)) {
1768                 up_write(&mm->mmap_sem);
1769                 put_cred(cred);
1770                 goto fail;
1771         }
1772
1773         /*
1774          *      We cannot trust fsuid as being the "true" uid of the
1775          *      process nor do we know its entire history. We only know it
1776          *      was tainted so we dump it as root in mode 2.
1777          */
1778         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1779                 flag = O_EXCL;          /* Stop rewrite attacks */
1780                 cred->fsuid = 0;        /* Dump root private */
1781         }
1782
1783         retval = coredump_wait(exit_code, &core_state);
1784         if (retval < 0) {
1785                 put_cred(cred);
1786                 goto fail;
1787         }
1788
1789         old_cred = override_creds(cred);
1790
1791         /*
1792          * Clear any false indication of pending signals that might
1793          * be seen by the filesystem code called to write the core file.
1794          */
1795         clear_thread_flag(TIF_SIGPENDING);
1796
1797         /*
1798          * lock_kernel() because format_corename() is controlled by sysctl, which
1799          * uses lock_kernel()
1800          */
1801         lock_kernel();
1802         ispipe = format_corename(corename, signr);
1803         unlock_kernel();
1804         /*
1805          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1806          * to a pipe.  Since we're not writing directly to the filesystem
1807          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1808          * created unless the pipe reader choses to write out the core file
1809          * at which point file size limits and permissions will be imposed
1810          * as it does with any other process
1811          */
1812         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1813                 goto fail_unlock;
1814
1815         if (ispipe) {
1816                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1817                 if (!helper_argv) {
1818                         printk(KERN_WARNING "%s failed to allocate memory\n",
1819                                __func__);
1820                         goto fail_unlock;
1821                 }
1822                 /* Terminate the string before the first option */
1823                 delimit = strchr(corename, ' ');
1824                 if (delimit)
1825                         *delimit = '\0';
1826                 delimit = strrchr(helper_argv[0], '/');
1827                 if (delimit)
1828                         delimit++;
1829                 else
1830                         delimit = helper_argv[0];
1831                 if (!strcmp(delimit, current->comm)) {
1832                         printk(KERN_NOTICE "Recursive core dump detected, "
1833                                         "aborting\n");
1834                         goto fail_unlock;
1835                 }
1836
1837                 core_limit = RLIM_INFINITY;
1838
1839                 /* SIGPIPE can happen, but it's just never processed */
1840                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1841                                 &file)) {
1842                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1843                                corename);
1844                         goto fail_unlock;
1845                 }
1846         } else
1847                 file = filp_open(corename,
1848                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1849                                  0600);
1850         if (IS_ERR(file))
1851                 goto fail_unlock;
1852         inode = file->f_path.dentry->d_inode;
1853         if (inode->i_nlink > 1)
1854                 goto close_fail;        /* multiple links - don't dump */
1855         if (!ispipe && d_unhashed(file->f_path.dentry))
1856                 goto close_fail;
1857
1858         /* AK: actually i see no reason to not allow this for named pipes etc.,
1859            but keep the previous behaviour for now. */
1860         if (!ispipe && !S_ISREG(inode->i_mode))
1861                 goto close_fail;
1862         /*
1863          * Dont allow local users get cute and trick others to coredump
1864          * into their pre-created files:
1865          */
1866         if (inode->i_uid != current_fsuid())
1867                 goto close_fail;
1868         if (!file->f_op)
1869                 goto close_fail;
1870         if (!file->f_op->write)
1871                 goto close_fail;
1872         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1873                 goto close_fail;
1874
1875         retval = binfmt->core_dump(signr, regs, file, core_limit);
1876
1877         if (retval)
1878                 current->signal->group_exit_code |= 0x80;
1879 close_fail:
1880         filp_close(file, NULL);
1881 fail_unlock:
1882         if (helper_argv)
1883                 argv_free(helper_argv);
1884
1885         revert_creds(old_cred);
1886         put_cred(cred);
1887         coredump_finish(mm);
1888 fail:
1889         return;
1890 }