New locking/refcounting for fs_struct
[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
57 #include <asm/uaccess.h>
58 #include <asm/mmu_context.h>
59 #include <asm/tlb.h>
60 #include "internal.h"
61
62 int core_uses_pid;
63 char core_pattern[CORENAME_MAX_SIZE] = "core";
64 int suid_dumpable = 0;
65
66 /* The maximal length of core_pattern is also specified in sysctl.c */
67
68 static LIST_HEAD(formats);
69 static DEFINE_RWLOCK(binfmt_lock);
70
71 int register_binfmt(struct linux_binfmt * fmt)
72 {
73         if (!fmt)
74                 return -EINVAL;
75         write_lock(&binfmt_lock);
76         list_add(&fmt->lh, &formats);
77         write_unlock(&binfmt_lock);
78         return 0;       
79 }
80
81 EXPORT_SYMBOL(register_binfmt);
82
83 void unregister_binfmt(struct linux_binfmt * fmt)
84 {
85         write_lock(&binfmt_lock);
86         list_del(&fmt->lh);
87         write_unlock(&binfmt_lock);
88 }
89
90 EXPORT_SYMBOL(unregister_binfmt);
91
92 static inline void put_binfmt(struct linux_binfmt * fmt)
93 {
94         module_put(fmt->module);
95 }
96
97 /*
98  * Note that a shared library must be both readable and executable due to
99  * security reasons.
100  *
101  * Also note that we take the address to load from from the file itself.
102  */
103 SYSCALL_DEFINE1(uselib, const char __user *, library)
104 {
105         struct file *file;
106         struct nameidata nd;
107         char *tmp = getname(library);
108         int error = PTR_ERR(tmp);
109
110         if (!IS_ERR(tmp)) {
111                 error = path_lookup_open(AT_FDCWD, tmp,
112                                          LOOKUP_FOLLOW, &nd,
113                                          FMODE_READ|FMODE_EXEC);
114                 putname(tmp);
115         }
116         if (error)
117                 goto out;
118
119         error = -EINVAL;
120         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
121                 goto exit;
122
123         error = -EACCES;
124         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
125                 goto exit;
126
127         error = inode_permission(nd.path.dentry->d_inode,
128                                  MAY_READ | MAY_EXEC | MAY_OPEN);
129         if (error)
130                 goto exit;
131         error = ima_path_check(&nd.path, MAY_READ | MAY_EXEC | MAY_OPEN);
132         if (error)
133                 goto exit;
134
135         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
136         error = PTR_ERR(file);
137         if (IS_ERR(file))
138                 goto out;
139
140         fsnotify_open(file->f_path.dentry);
141
142         error = -ENOEXEC;
143         if(file->f_op) {
144                 struct linux_binfmt * fmt;
145
146                 read_lock(&binfmt_lock);
147                 list_for_each_entry(fmt, &formats, lh) {
148                         if (!fmt->load_shlib)
149                                 continue;
150                         if (!try_module_get(fmt->module))
151                                 continue;
152                         read_unlock(&binfmt_lock);
153                         error = fmt->load_shlib(file);
154                         read_lock(&binfmt_lock);
155                         put_binfmt(fmt);
156                         if (error != -ENOEXEC)
157                                 break;
158                 }
159                 read_unlock(&binfmt_lock);
160         }
161         fput(file);
162 out:
163         return error;
164 exit:
165         release_open_intent(&nd);
166         path_put(&nd.path);
167         goto out;
168 }
169
170 #ifdef CONFIG_MMU
171
172 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
173                 int write)
174 {
175         struct page *page;
176         int ret;
177
178 #ifdef CONFIG_STACK_GROWSUP
179         if (write) {
180                 ret = expand_stack_downwards(bprm->vma, pos);
181                 if (ret < 0)
182                         return NULL;
183         }
184 #endif
185         ret = get_user_pages(current, bprm->mm, pos,
186                         1, write, 1, &page, NULL);
187         if (ret <= 0)
188                 return NULL;
189
190         if (write) {
191                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
192                 struct rlimit *rlim;
193
194                 /*
195                  * We've historically supported up to 32 pages (ARG_MAX)
196                  * of argument strings even with small stacks
197                  */
198                 if (size <= ARG_MAX)
199                         return page;
200
201                 /*
202                  * Limit to 1/4-th the stack size for the argv+env strings.
203                  * This ensures that:
204                  *  - the remaining binfmt code will not run out of stack space,
205                  *  - the program will have a reasonable amount of stack left
206                  *    to work from.
207                  */
208                 rlim = current->signal->rlim;
209                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
210                         put_page(page);
211                         return NULL;
212                 }
213         }
214
215         return page;
216 }
217
218 static void put_arg_page(struct page *page)
219 {
220         put_page(page);
221 }
222
223 static void free_arg_page(struct linux_binprm *bprm, int i)
224 {
225 }
226
227 static void free_arg_pages(struct linux_binprm *bprm)
228 {
229 }
230
231 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
232                 struct page *page)
233 {
234         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
235 }
236
237 static int __bprm_mm_init(struct linux_binprm *bprm)
238 {
239         int err;
240         struct vm_area_struct *vma = NULL;
241         struct mm_struct *mm = bprm->mm;
242
243         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
244         if (!vma)
245                 return -ENOMEM;
246
247         down_write(&mm->mmap_sem);
248         vma->vm_mm = mm;
249
250         /*
251          * Place the stack at the largest stack address the architecture
252          * supports. Later, we'll move this to an appropriate place. We don't
253          * use STACK_TOP because that can depend on attributes which aren't
254          * configured yet.
255          */
256         vma->vm_end = STACK_TOP_MAX;
257         vma->vm_start = vma->vm_end - PAGE_SIZE;
258         vma->vm_flags = VM_STACK_FLAGS;
259         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
260         err = insert_vm_struct(mm, vma);
261         if (err)
262                 goto err;
263
264         mm->stack_vm = mm->total_vm = 1;
265         up_write(&mm->mmap_sem);
266         bprm->p = vma->vm_end - sizeof(void *);
267         return 0;
268 err:
269         up_write(&mm->mmap_sem);
270         bprm->vma = NULL;
271         kmem_cache_free(vm_area_cachep, vma);
272         return err;
273 }
274
275 static bool valid_arg_len(struct linux_binprm *bprm, long len)
276 {
277         return len <= MAX_ARG_STRLEN;
278 }
279
280 #else
281
282 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
283                 int write)
284 {
285         struct page *page;
286
287         page = bprm->page[pos / PAGE_SIZE];
288         if (!page && write) {
289                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
290                 if (!page)
291                         return NULL;
292                 bprm->page[pos / PAGE_SIZE] = page;
293         }
294
295         return page;
296 }
297
298 static void put_arg_page(struct page *page)
299 {
300 }
301
302 static void free_arg_page(struct linux_binprm *bprm, int i)
303 {
304         if (bprm->page[i]) {
305                 __free_page(bprm->page[i]);
306                 bprm->page[i] = NULL;
307         }
308 }
309
310 static void free_arg_pages(struct linux_binprm *bprm)
311 {
312         int i;
313
314         for (i = 0; i < MAX_ARG_PAGES; i++)
315                 free_arg_page(bprm, i);
316 }
317
318 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
319                 struct page *page)
320 {
321 }
322
323 static int __bprm_mm_init(struct linux_binprm *bprm)
324 {
325         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
326         return 0;
327 }
328
329 static bool valid_arg_len(struct linux_binprm *bprm, long len)
330 {
331         return len <= bprm->p;
332 }
333
334 #endif /* CONFIG_MMU */
335
336 /*
337  * Create a new mm_struct and populate it with a temporary stack
338  * vm_area_struct.  We don't have enough context at this point to set the stack
339  * flags, permissions, and offset, so we use temporary values.  We'll update
340  * them later in setup_arg_pages().
341  */
342 int bprm_mm_init(struct linux_binprm *bprm)
343 {
344         int err;
345         struct mm_struct *mm = NULL;
346
347         bprm->mm = mm = mm_alloc();
348         err = -ENOMEM;
349         if (!mm)
350                 goto err;
351
352         err = init_new_context(current, mm);
353         if (err)
354                 goto err;
355
356         err = __bprm_mm_init(bprm);
357         if (err)
358                 goto err;
359
360         return 0;
361
362 err:
363         if (mm) {
364                 bprm->mm = NULL;
365                 mmdrop(mm);
366         }
367
368         return err;
369 }
370
371 /*
372  * count() counts the number of strings in array ARGV.
373  */
374 static int count(char __user * __user * argv, int max)
375 {
376         int i = 0;
377
378         if (argv != NULL) {
379                 for (;;) {
380                         char __user * p;
381
382                         if (get_user(p, argv))
383                                 return -EFAULT;
384                         if (!p)
385                                 break;
386                         argv++;
387                         if (i++ >= max)
388                                 return -E2BIG;
389                         cond_resched();
390                 }
391         }
392         return i;
393 }
394
395 /*
396  * 'copy_strings()' copies argument/environment strings from the old
397  * processes's memory to the new process's stack.  The call to get_user_pages()
398  * ensures the destination page is created and not swapped out.
399  */
400 static int copy_strings(int argc, char __user * __user * argv,
401                         struct linux_binprm *bprm)
402 {
403         struct page *kmapped_page = NULL;
404         char *kaddr = NULL;
405         unsigned long kpos = 0;
406         int ret;
407
408         while (argc-- > 0) {
409                 char __user *str;
410                 int len;
411                 unsigned long pos;
412
413                 if (get_user(str, argv+argc) ||
414                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
415                         ret = -EFAULT;
416                         goto out;
417                 }
418
419                 if (!valid_arg_len(bprm, len)) {
420                         ret = -E2BIG;
421                         goto out;
422                 }
423
424                 /* We're going to work our way backwords. */
425                 pos = bprm->p;
426                 str += len;
427                 bprm->p -= len;
428
429                 while (len > 0) {
430                         int offset, bytes_to_copy;
431
432                         offset = pos % PAGE_SIZE;
433                         if (offset == 0)
434                                 offset = PAGE_SIZE;
435
436                         bytes_to_copy = offset;
437                         if (bytes_to_copy > len)
438                                 bytes_to_copy = len;
439
440                         offset -= bytes_to_copy;
441                         pos -= bytes_to_copy;
442                         str -= bytes_to_copy;
443                         len -= bytes_to_copy;
444
445                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
446                                 struct page *page;
447
448                                 page = get_arg_page(bprm, pos, 1);
449                                 if (!page) {
450                                         ret = -E2BIG;
451                                         goto out;
452                                 }
453
454                                 if (kmapped_page) {
455                                         flush_kernel_dcache_page(kmapped_page);
456                                         kunmap(kmapped_page);
457                                         put_arg_page(kmapped_page);
458                                 }
459                                 kmapped_page = page;
460                                 kaddr = kmap(kmapped_page);
461                                 kpos = pos & PAGE_MASK;
462                                 flush_arg_page(bprm, kpos, kmapped_page);
463                         }
464                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
465                                 ret = -EFAULT;
466                                 goto out;
467                         }
468                 }
469         }
470         ret = 0;
471 out:
472         if (kmapped_page) {
473                 flush_kernel_dcache_page(kmapped_page);
474                 kunmap(kmapped_page);
475                 put_arg_page(kmapped_page);
476         }
477         return ret;
478 }
479
480 /*
481  * Like copy_strings, but get argv and its values from kernel memory.
482  */
483 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
484 {
485         int r;
486         mm_segment_t oldfs = get_fs();
487         set_fs(KERNEL_DS);
488         r = copy_strings(argc, (char __user * __user *)argv, bprm);
489         set_fs(oldfs);
490         return r;
491 }
492 EXPORT_SYMBOL(copy_strings_kernel);
493
494 #ifdef CONFIG_MMU
495
496 /*
497  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
498  * the binfmt code determines where the new stack should reside, we shift it to
499  * its final location.  The process proceeds as follows:
500  *
501  * 1) Use shift to calculate the new vma endpoints.
502  * 2) Extend vma to cover both the old and new ranges.  This ensures the
503  *    arguments passed to subsequent functions are consistent.
504  * 3) Move vma's page tables to the new range.
505  * 4) Free up any cleared pgd range.
506  * 5) Shrink the vma to cover only the new range.
507  */
508 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
509 {
510         struct mm_struct *mm = vma->vm_mm;
511         unsigned long old_start = vma->vm_start;
512         unsigned long old_end = vma->vm_end;
513         unsigned long length = old_end - old_start;
514         unsigned long new_start = old_start - shift;
515         unsigned long new_end = old_end - shift;
516         struct mmu_gather *tlb;
517
518         BUG_ON(new_start > new_end);
519
520         /*
521          * ensure there are no vmas between where we want to go
522          * and where we are
523          */
524         if (vma != find_vma(mm, new_start))
525                 return -EFAULT;
526
527         /*
528          * cover the whole range: [new_start, old_end)
529          */
530         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
531
532         /*
533          * move the page tables downwards, on failure we rely on
534          * process cleanup to remove whatever mess we made.
535          */
536         if (length != move_page_tables(vma, old_start,
537                                        vma, new_start, length))
538                 return -ENOMEM;
539
540         lru_add_drain();
541         tlb = tlb_gather_mmu(mm, 0);
542         if (new_end > old_start) {
543                 /*
544                  * when the old and new regions overlap clear from new_end.
545                  */
546                 free_pgd_range(tlb, new_end, old_end, new_end,
547                         vma->vm_next ? vma->vm_next->vm_start : 0);
548         } else {
549                 /*
550                  * otherwise, clean from old_start; this is done to not touch
551                  * the address space in [new_end, old_start) some architectures
552                  * have constraints on va-space that make this illegal (IA64) -
553                  * for the others its just a little faster.
554                  */
555                 free_pgd_range(tlb, old_start, old_end, new_end,
556                         vma->vm_next ? vma->vm_next->vm_start : 0);
557         }
558         tlb_finish_mmu(tlb, new_end, old_end);
559
560         /*
561          * shrink the vma to just the new range.
562          */
563         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
564
565         return 0;
566 }
567
568 #define EXTRA_STACK_VM_PAGES    20      /* random */
569
570 /*
571  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
572  * the stack is optionally relocated, and some extra space is added.
573  */
574 int setup_arg_pages(struct linux_binprm *bprm,
575                     unsigned long stack_top,
576                     int executable_stack)
577 {
578         unsigned long ret;
579         unsigned long stack_shift;
580         struct mm_struct *mm = current->mm;
581         struct vm_area_struct *vma = bprm->vma;
582         struct vm_area_struct *prev = NULL;
583         unsigned long vm_flags;
584         unsigned long stack_base;
585
586 #ifdef CONFIG_STACK_GROWSUP
587         /* Limit stack size to 1GB */
588         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
589         if (stack_base > (1 << 30))
590                 stack_base = 1 << 30;
591
592         /* Make sure we didn't let the argument array grow too large. */
593         if (vma->vm_end - vma->vm_start > stack_base)
594                 return -ENOMEM;
595
596         stack_base = PAGE_ALIGN(stack_top - stack_base);
597
598         stack_shift = vma->vm_start - stack_base;
599         mm->arg_start = bprm->p - stack_shift;
600         bprm->p = vma->vm_end - stack_shift;
601 #else
602         stack_top = arch_align_stack(stack_top);
603         stack_top = PAGE_ALIGN(stack_top);
604         stack_shift = vma->vm_end - stack_top;
605
606         bprm->p -= stack_shift;
607         mm->arg_start = bprm->p;
608 #endif
609
610         if (bprm->loader)
611                 bprm->loader -= stack_shift;
612         bprm->exec -= stack_shift;
613
614         down_write(&mm->mmap_sem);
615         vm_flags = VM_STACK_FLAGS;
616
617         /*
618          * Adjust stack execute permissions; explicitly enable for
619          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
620          * (arch default) otherwise.
621          */
622         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
623                 vm_flags |= VM_EXEC;
624         else if (executable_stack == EXSTACK_DISABLE_X)
625                 vm_flags &= ~VM_EXEC;
626         vm_flags |= mm->def_flags;
627
628         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
629                         vm_flags);
630         if (ret)
631                 goto out_unlock;
632         BUG_ON(prev != vma);
633
634         /* Move stack pages down in memory. */
635         if (stack_shift) {
636                 ret = shift_arg_pages(vma, stack_shift);
637                 if (ret) {
638                         up_write(&mm->mmap_sem);
639                         return ret;
640                 }
641         }
642
643 #ifdef CONFIG_STACK_GROWSUP
644         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
645 #else
646         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
647 #endif
648         ret = expand_stack(vma, stack_base);
649         if (ret)
650                 ret = -EFAULT;
651
652 out_unlock:
653         up_write(&mm->mmap_sem);
654         return 0;
655 }
656 EXPORT_SYMBOL(setup_arg_pages);
657
658 #endif /* CONFIG_MMU */
659
660 struct file *open_exec(const char *name)
661 {
662         struct nameidata nd;
663         struct file *file;
664         int err;
665
666         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
667                                 FMODE_READ|FMODE_EXEC);
668         if (err)
669                 goto out;
670
671         err = -EACCES;
672         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
673                 goto out_path_put;
674
675         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
676                 goto out_path_put;
677
678         err = inode_permission(nd.path.dentry->d_inode, MAY_EXEC | MAY_OPEN);
679         if (err)
680                 goto out_path_put;
681         err = ima_path_check(&nd.path, MAY_EXEC | MAY_OPEN);
682         if (err)
683                 goto out_path_put;
684
685         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
686         if (IS_ERR(file))
687                 return file;
688
689         fsnotify_open(file->f_path.dentry);
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         int count;
777
778         if (thread_group_empty(tsk))
779                 goto no_thread_group;
780
781         /*
782          * Kill all other threads in the thread group.
783          */
784         spin_lock_irq(lock);
785         if (signal_group_exit(sig)) {
786                 /*
787                  * Another group action in progress, just
788                  * return so that the signal is processed.
789                  */
790                 spin_unlock_irq(lock);
791                 return -EAGAIN;
792         }
793         sig->group_exit_task = tsk;
794         zap_other_threads(tsk);
795
796         /* Account for the thread group leader hanging around: */
797         count = thread_group_leader(tsk) ? 1 : 2;
798         sig->notify_count = count;
799         while (atomic_read(&sig->count) > count) {
800                 __set_current_state(TASK_UNINTERRUPTIBLE);
801                 spin_unlock_irq(lock);
802                 schedule();
803                 spin_lock_irq(lock);
804         }
805         spin_unlock_irq(lock);
806
807         /*
808          * At this point all other threads have exited, all we have to
809          * do is to wait for the thread group leader to become inactive,
810          * and to assume its PID:
811          */
812         if (!thread_group_leader(tsk)) {
813                 struct task_struct *leader = tsk->group_leader;
814
815                 sig->notify_count = -1; /* for exit_notify() */
816                 for (;;) {
817                         write_lock_irq(&tasklist_lock);
818                         if (likely(leader->exit_state))
819                                 break;
820                         __set_current_state(TASK_UNINTERRUPTIBLE);
821                         write_unlock_irq(&tasklist_lock);
822                         schedule();
823                 }
824
825                 /*
826                  * The only record we have of the real-time age of a
827                  * process, regardless of execs it's done, is start_time.
828                  * All the past CPU time is accumulated in signal_struct
829                  * from sister threads now dead.  But in this non-leader
830                  * exec, nothing survives from the original leader thread,
831                  * whose birth marks the true age of this process now.
832                  * When we take on its identity by switching to its PID, we
833                  * also take its birthdate (always earlier than our own).
834                  */
835                 tsk->start_time = leader->start_time;
836
837                 BUG_ON(!same_thread_group(leader, tsk));
838                 BUG_ON(has_group_leader_pid(tsk));
839                 /*
840                  * An exec() starts a new thread group with the
841                  * TGID of the previous thread group. Rehash the
842                  * two threads with a switched PID, and release
843                  * the former thread group leader:
844                  */
845
846                 /* Become a process group leader with the old leader's pid.
847                  * The old leader becomes a thread of the this thread group.
848                  * Note: The old leader also uses this pid until release_task
849                  *       is called.  Odd but simple and correct.
850                  */
851                 detach_pid(tsk, PIDTYPE_PID);
852                 tsk->pid = leader->pid;
853                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
854                 transfer_pid(leader, tsk, PIDTYPE_PGID);
855                 transfer_pid(leader, tsk, PIDTYPE_SID);
856                 list_replace_rcu(&leader->tasks, &tsk->tasks);
857
858                 tsk->group_leader = tsk;
859                 leader->group_leader = tsk;
860
861                 tsk->exit_signal = SIGCHLD;
862
863                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
864                 leader->exit_state = EXIT_DEAD;
865                 write_unlock_irq(&tasklist_lock);
866
867                 release_task(leader);
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
877         if (atomic_read(&oldsighand->count) != 1) {
878                 struct sighand_struct *newsighand;
879                 /*
880                  * This ->sighand is shared with the CLONE_SIGHAND
881                  * but not CLONE_THREAD task, switch to the new one.
882                  */
883                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
884                 if (!newsighand)
885                         return -ENOMEM;
886
887                 atomic_set(&newsighand->count, 1);
888                 memcpy(newsighand->action, oldsighand->action,
889                        sizeof(newsighand->action));
890
891                 write_lock_irq(&tasklist_lock);
892                 spin_lock(&oldsighand->siglock);
893                 rcu_assign_pointer(tsk->sighand, newsighand);
894                 spin_unlock(&oldsighand->siglock);
895                 write_unlock_irq(&tasklist_lock);
896
897                 __cleanup_sighand(oldsighand);
898         }
899
900         BUG_ON(!thread_group_leader(tsk));
901         return 0;
902 }
903
904 /*
905  * These functions flushes out all traces of the currently running executable
906  * so that a new one can be started
907  */
908 static void flush_old_files(struct files_struct * files)
909 {
910         long j = -1;
911         struct fdtable *fdt;
912
913         spin_lock(&files->file_lock);
914         for (;;) {
915                 unsigned long set, i;
916
917                 j++;
918                 i = j * __NFDBITS;
919                 fdt = files_fdtable(files);
920                 if (i >= fdt->max_fds)
921                         break;
922                 set = fdt->close_on_exec->fds_bits[j];
923                 if (!set)
924                         continue;
925                 fdt->close_on_exec->fds_bits[j] = 0;
926                 spin_unlock(&files->file_lock);
927                 for ( ; set ; i++,set >>= 1) {
928                         if (set & 1) {
929                                 sys_close(i);
930                         }
931                 }
932                 spin_lock(&files->file_lock);
933
934         }
935         spin_unlock(&files->file_lock);
936 }
937
938 char *get_task_comm(char *buf, struct task_struct *tsk)
939 {
940         /* buf must be at least sizeof(tsk->comm) in size */
941         task_lock(tsk);
942         strncpy(buf, tsk->comm, sizeof(tsk->comm));
943         task_unlock(tsk);
944         return buf;
945 }
946
947 void set_task_comm(struct task_struct *tsk, char *buf)
948 {
949         task_lock(tsk);
950         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
951         task_unlock(tsk);
952 }
953
954 int flush_old_exec(struct linux_binprm * bprm)
955 {
956         char * name;
957         int i, ch, retval;
958         char tcomm[sizeof(current->comm)];
959
960         /*
961          * Make sure we have a private signal table and that
962          * we are unassociated from the previous thread group.
963          */
964         retval = de_thread(current);
965         if (retval)
966                 goto out;
967
968         set_mm_exe_file(bprm->mm, bprm->file);
969
970         /*
971          * Release all of the old mmap stuff
972          */
973         retval = exec_mmap(bprm->mm);
974         if (retval)
975                 goto out;
976
977         bprm->mm = NULL;                /* We're using it now */
978
979         /* This is the point of no return */
980         current->sas_ss_sp = current->sas_ss_size = 0;
981
982         if (current_euid() == current_uid() && current_egid() == current_gid())
983                 set_dumpable(current->mm, 1);
984         else
985                 set_dumpable(current->mm, suid_dumpable);
986
987         name = bprm->filename;
988
989         /* Copies the binary name from after last slash */
990         for (i=0; (ch = *(name++)) != '\0';) {
991                 if (ch == '/')
992                         i = 0; /* overwrite what we wrote */
993                 else
994                         if (i < (sizeof(tcomm) - 1))
995                                 tcomm[i++] = ch;
996         }
997         tcomm[i] = '\0';
998         set_task_comm(current, tcomm);
999
1000         current->flags &= ~PF_RANDOMIZE;
1001         flush_thread();
1002
1003         /* Set the new mm task size. We have to do that late because it may
1004          * depend on TIF_32BIT which is only updated in flush_thread() on
1005          * some architectures like powerpc
1006          */
1007         current->mm->task_size = TASK_SIZE;
1008
1009         /* install the new credentials */
1010         if (bprm->cred->uid != current_euid() ||
1011             bprm->cred->gid != current_egid()) {
1012                 current->pdeath_signal = 0;
1013         } else if (file_permission(bprm->file, MAY_READ) ||
1014                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1015                 set_dumpable(current->mm, suid_dumpable);
1016         }
1017
1018         current->personality &= ~bprm->per_clear;
1019
1020         /* An exec changes our domain. We are no longer part of the thread
1021            group */
1022
1023         current->self_exec_id++;
1024                         
1025         flush_signal_handlers(current, 0);
1026         flush_old_files(current->files);
1027
1028         return 0;
1029
1030 out:
1031         return retval;
1032 }
1033
1034 EXPORT_SYMBOL(flush_old_exec);
1035
1036 /*
1037  * install the new credentials for this executable
1038  */
1039 void install_exec_creds(struct linux_binprm *bprm)
1040 {
1041         security_bprm_committing_creds(bprm);
1042
1043         commit_creds(bprm->cred);
1044         bprm->cred = NULL;
1045
1046         /* cred_exec_mutex must be held at least to this point to prevent
1047          * ptrace_attach() from altering our determination of the task's
1048          * credentials; any time after this it may be unlocked */
1049
1050         security_bprm_committed_creds(bprm);
1051 }
1052 EXPORT_SYMBOL(install_exec_creds);
1053
1054 /*
1055  * determine how safe it is to execute the proposed program
1056  * - the caller must hold current->cred_exec_mutex to protect against
1057  *   PTRACE_ATTACH
1058  */
1059 int check_unsafe_exec(struct linux_binprm *bprm)
1060 {
1061         struct task_struct *p = current, *t;
1062         unsigned long flags;
1063         unsigned n_fs, n_sighand;
1064         int res = 0;
1065
1066         bprm->unsafe = tracehook_unsafe_exec(p);
1067
1068         n_fs = 1;
1069         n_sighand = 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                 n_sighand++;
1076         }
1077
1078         if (p->fs->users > n_fs ||
1079             atomic_read(&p->sighand->count) > n_sighand) {
1080                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1081         } else {
1082                 if (p->fs->in_exec)
1083                         res = -EAGAIN;
1084                 p->fs->in_exec = 1;
1085         }
1086
1087         unlock_task_sighand(p, &flags);
1088         write_unlock(&p->fs->lock);
1089
1090         return res;
1091 }
1092
1093 /* 
1094  * Fill the binprm structure from the inode. 
1095  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1096  *
1097  * This may be called multiple times for binary chains (scripts for example).
1098  */
1099 int prepare_binprm(struct linux_binprm *bprm)
1100 {
1101         umode_t mode;
1102         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1103         int retval;
1104
1105         mode = inode->i_mode;
1106         if (bprm->file->f_op == NULL)
1107                 return -EACCES;
1108
1109         /* clear any previous set[ug]id data from a previous binary */
1110         bprm->cred->euid = current_euid();
1111         bprm->cred->egid = current_egid();
1112
1113         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1114                 /* Set-uid? */
1115                 if (mode & S_ISUID) {
1116                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1117                         bprm->cred->euid = inode->i_uid;
1118                 }
1119
1120                 /* Set-gid? */
1121                 /*
1122                  * If setgid is set but no group execute bit then this
1123                  * is a candidate for mandatory locking, not a setgid
1124                  * executable.
1125                  */
1126                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1127                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1128                         bprm->cred->egid = inode->i_gid;
1129                 }
1130         }
1131
1132         /* fill in binprm security blob */
1133         retval = security_bprm_set_creds(bprm);
1134         if (retval)
1135                 return retval;
1136         bprm->cred_prepared = 1;
1137
1138         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1139         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1140 }
1141
1142 EXPORT_SYMBOL(prepare_binprm);
1143
1144 /*
1145  * Arguments are '\0' separated strings found at the location bprm->p
1146  * points to; chop off the first by relocating brpm->p to right after
1147  * the first '\0' encountered.
1148  */
1149 int remove_arg_zero(struct linux_binprm *bprm)
1150 {
1151         int ret = 0;
1152         unsigned long offset;
1153         char *kaddr;
1154         struct page *page;
1155
1156         if (!bprm->argc)
1157                 return 0;
1158
1159         do {
1160                 offset = bprm->p & ~PAGE_MASK;
1161                 page = get_arg_page(bprm, bprm->p, 0);
1162                 if (!page) {
1163                         ret = -EFAULT;
1164                         goto out;
1165                 }
1166                 kaddr = kmap_atomic(page, KM_USER0);
1167
1168                 for (; offset < PAGE_SIZE && kaddr[offset];
1169                                 offset++, bprm->p++)
1170                         ;
1171
1172                 kunmap_atomic(kaddr, KM_USER0);
1173                 put_arg_page(page);
1174
1175                 if (offset == PAGE_SIZE)
1176                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1177         } while (offset == PAGE_SIZE);
1178
1179         bprm->p++;
1180         bprm->argc--;
1181         ret = 0;
1182
1183 out:
1184         return ret;
1185 }
1186 EXPORT_SYMBOL(remove_arg_zero);
1187
1188 /*
1189  * cycle the list of binary formats handler, until one recognizes the image
1190  */
1191 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1192 {
1193         unsigned int depth = bprm->recursion_depth;
1194         int try,retval;
1195         struct linux_binfmt *fmt;
1196
1197         retval = security_bprm_check(bprm);
1198         if (retval)
1199                 return retval;
1200         retval = ima_bprm_check(bprm);
1201         if (retval)
1202                 return retval;
1203
1204         /* kernel module loader fixup */
1205         /* so we don't try to load run modprobe in kernel space. */
1206         set_fs(USER_DS);
1207
1208         retval = audit_bprm(bprm);
1209         if (retval)
1210                 return retval;
1211
1212         retval = -ENOENT;
1213         for (try=0; try<2; try++) {
1214                 read_lock(&binfmt_lock);
1215                 list_for_each_entry(fmt, &formats, lh) {
1216                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1217                         if (!fn)
1218                                 continue;
1219                         if (!try_module_get(fmt->module))
1220                                 continue;
1221                         read_unlock(&binfmt_lock);
1222                         retval = fn(bprm, regs);
1223                         /*
1224                          * Restore the depth counter to its starting value
1225                          * in this call, so we don't have to rely on every
1226                          * load_binary function to restore it on return.
1227                          */
1228                         bprm->recursion_depth = depth;
1229                         if (retval >= 0) {
1230                                 if (depth == 0)
1231                                         tracehook_report_exec(fmt, bprm, regs);
1232                                 put_binfmt(fmt);
1233                                 allow_write_access(bprm->file);
1234                                 if (bprm->file)
1235                                         fput(bprm->file);
1236                                 bprm->file = NULL;
1237                                 current->did_exec = 1;
1238                                 proc_exec_connector(current);
1239                                 return retval;
1240                         }
1241                         read_lock(&binfmt_lock);
1242                         put_binfmt(fmt);
1243                         if (retval != -ENOEXEC || bprm->mm == NULL)
1244                                 break;
1245                         if (!bprm->file) {
1246                                 read_unlock(&binfmt_lock);
1247                                 return retval;
1248                         }
1249                 }
1250                 read_unlock(&binfmt_lock);
1251                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1252                         break;
1253 #ifdef CONFIG_MODULES
1254                 } else {
1255 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1256                         if (printable(bprm->buf[0]) &&
1257                             printable(bprm->buf[1]) &&
1258                             printable(bprm->buf[2]) &&
1259                             printable(bprm->buf[3]))
1260                                 break; /* -ENOEXEC */
1261                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1262 #endif
1263                 }
1264         }
1265         return retval;
1266 }
1267
1268 EXPORT_SYMBOL(search_binary_handler);
1269
1270 void free_bprm(struct linux_binprm *bprm)
1271 {
1272         free_arg_pages(bprm);
1273         if (bprm->cred)
1274                 abort_creds(bprm->cred);
1275         kfree(bprm);
1276 }
1277
1278 /*
1279  * sys_execve() executes a new program.
1280  */
1281 int do_execve(char * filename,
1282         char __user *__user *argv,
1283         char __user *__user *envp,
1284         struct pt_regs * regs)
1285 {
1286         struct linux_binprm *bprm;
1287         struct file *file;
1288         struct files_struct *displaced;
1289         int retval;
1290
1291         retval = unshare_files(&displaced);
1292         if (retval)
1293                 goto out_ret;
1294
1295         retval = -ENOMEM;
1296         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1297         if (!bprm)
1298                 goto out_files;
1299
1300         retval = mutex_lock_interruptible(&current->cred_exec_mutex);
1301         if (retval < 0)
1302                 goto out_free;
1303         current->in_execve = 1;
1304
1305         retval = -ENOMEM;
1306         bprm->cred = prepare_exec_creds();
1307         if (!bprm->cred)
1308                 goto out_unlock;
1309
1310         retval = check_unsafe_exec(bprm);
1311         if (retval)
1312                 goto out_unlock;
1313
1314         file = open_exec(filename);
1315         retval = PTR_ERR(file);
1316         if (IS_ERR(file))
1317                 goto out_unmark;
1318
1319         sched_exec();
1320
1321         bprm->file = file;
1322         bprm->filename = filename;
1323         bprm->interp = filename;
1324
1325         retval = bprm_mm_init(bprm);
1326         if (retval)
1327                 goto out_file;
1328
1329         bprm->argc = count(argv, MAX_ARG_STRINGS);
1330         if ((retval = bprm->argc) < 0)
1331                 goto out;
1332
1333         bprm->envc = count(envp, MAX_ARG_STRINGS);
1334         if ((retval = bprm->envc) < 0)
1335                 goto out;
1336
1337         retval = prepare_binprm(bprm);
1338         if (retval < 0)
1339                 goto out;
1340
1341         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1342         if (retval < 0)
1343                 goto out;
1344
1345         bprm->exec = bprm->p;
1346         retval = copy_strings(bprm->envc, envp, bprm);
1347         if (retval < 0)
1348                 goto out;
1349
1350         retval = copy_strings(bprm->argc, argv, bprm);
1351         if (retval < 0)
1352                 goto out;
1353
1354         current->flags &= ~PF_KTHREAD;
1355         retval = search_binary_handler(bprm,regs);
1356         if (retval < 0)
1357                 goto out;
1358
1359         /* execve succeeded */
1360         write_lock(&current->fs->lock);
1361         current->fs->in_exec = 0;
1362         write_unlock(&current->fs->lock);
1363         current->in_execve = 0;
1364         mutex_unlock(&current->cred_exec_mutex);
1365         acct_update_integrals(current);
1366         free_bprm(bprm);
1367         if (displaced)
1368                 put_files_struct(displaced);
1369         return retval;
1370
1371 out:
1372         if (bprm->mm)
1373                 mmput (bprm->mm);
1374
1375 out_file:
1376         if (bprm->file) {
1377                 allow_write_access(bprm->file);
1378                 fput(bprm->file);
1379         }
1380
1381 out_unmark:
1382         write_lock(&current->fs->lock);
1383         current->fs->in_exec = 0;
1384         write_unlock(&current->fs->lock);
1385
1386 out_unlock:
1387         current->in_execve = 0;
1388         mutex_unlock(&current->cred_exec_mutex);
1389
1390 out_free:
1391         free_bprm(bprm);
1392
1393 out_files:
1394         if (displaced)
1395                 reset_files_struct(displaced);
1396 out_ret:
1397         return retval;
1398 }
1399
1400 int set_binfmt(struct linux_binfmt *new)
1401 {
1402         struct linux_binfmt *old = current->binfmt;
1403
1404         if (new) {
1405                 if (!try_module_get(new->module))
1406                         return -1;
1407         }
1408         current->binfmt = new;
1409         if (old)
1410                 module_put(old->module);
1411         return 0;
1412 }
1413
1414 EXPORT_SYMBOL(set_binfmt);
1415
1416 /* format_corename will inspect the pattern parameter, and output a
1417  * name into corename, which must have space for at least
1418  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1419  */
1420 static int format_corename(char *corename, long signr)
1421 {
1422         const struct cred *cred = current_cred();
1423         const char *pat_ptr = core_pattern;
1424         int ispipe = (*pat_ptr == '|');
1425         char *out_ptr = corename;
1426         char *const out_end = corename + CORENAME_MAX_SIZE;
1427         int rc;
1428         int pid_in_pattern = 0;
1429
1430         /* Repeat as long as we have more pattern to process and more output
1431            space */
1432         while (*pat_ptr) {
1433                 if (*pat_ptr != '%') {
1434                         if (out_ptr == out_end)
1435                                 goto out;
1436                         *out_ptr++ = *pat_ptr++;
1437                 } else {
1438                         switch (*++pat_ptr) {
1439                         case 0:
1440                                 goto out;
1441                         /* Double percent, output one percent */
1442                         case '%':
1443                                 if (out_ptr == out_end)
1444                                         goto out;
1445                                 *out_ptr++ = '%';
1446                                 break;
1447                         /* pid */
1448                         case 'p':
1449                                 pid_in_pattern = 1;
1450                                 rc = snprintf(out_ptr, out_end - out_ptr,
1451                                               "%d", task_tgid_vnr(current));
1452                                 if (rc > out_end - out_ptr)
1453                                         goto out;
1454                                 out_ptr += rc;
1455                                 break;
1456                         /* uid */
1457                         case 'u':
1458                                 rc = snprintf(out_ptr, out_end - out_ptr,
1459                                               "%d", cred->uid);
1460                                 if (rc > out_end - out_ptr)
1461                                         goto out;
1462                                 out_ptr += rc;
1463                                 break;
1464                         /* gid */
1465                         case 'g':
1466                                 rc = snprintf(out_ptr, out_end - out_ptr,
1467                                               "%d", cred->gid);
1468                                 if (rc > out_end - out_ptr)
1469                                         goto out;
1470                                 out_ptr += rc;
1471                                 break;
1472                         /* signal that caused the coredump */
1473                         case 's':
1474                                 rc = snprintf(out_ptr, out_end - out_ptr,
1475                                               "%ld", signr);
1476                                 if (rc > out_end - out_ptr)
1477                                         goto out;
1478                                 out_ptr += rc;
1479                                 break;
1480                         /* UNIX time of coredump */
1481                         case 't': {
1482                                 struct timeval tv;
1483                                 do_gettimeofday(&tv);
1484                                 rc = snprintf(out_ptr, out_end - out_ptr,
1485                                               "%lu", tv.tv_sec);
1486                                 if (rc > out_end - out_ptr)
1487                                         goto out;
1488                                 out_ptr += rc;
1489                                 break;
1490                         }
1491                         /* hostname */
1492                         case 'h':
1493                                 down_read(&uts_sem);
1494                                 rc = snprintf(out_ptr, out_end - out_ptr,
1495                                               "%s", utsname()->nodename);
1496                                 up_read(&uts_sem);
1497                                 if (rc > out_end - out_ptr)
1498                                         goto out;
1499                                 out_ptr += rc;
1500                                 break;
1501                         /* executable */
1502                         case 'e':
1503                                 rc = snprintf(out_ptr, out_end - out_ptr,
1504                                               "%s", current->comm);
1505                                 if (rc > out_end - out_ptr)
1506                                         goto out;
1507                                 out_ptr += rc;
1508                                 break;
1509                         /* core limit size */
1510                         case 'c':
1511                                 rc = snprintf(out_ptr, out_end - out_ptr,
1512                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1513                                 if (rc > out_end - out_ptr)
1514                                         goto out;
1515                                 out_ptr += rc;
1516                                 break;
1517                         default:
1518                                 break;
1519                         }
1520                         ++pat_ptr;
1521                 }
1522         }
1523         /* Backward compatibility with core_uses_pid:
1524          *
1525          * If core_pattern does not include a %p (as is the default)
1526          * and core_uses_pid is set, then .%pid will be appended to
1527          * the filename. Do not do this for piped commands. */
1528         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1529                 rc = snprintf(out_ptr, out_end - out_ptr,
1530                               ".%d", task_tgid_vnr(current));
1531                 if (rc > out_end - out_ptr)
1532                         goto out;
1533                 out_ptr += rc;
1534         }
1535 out:
1536         *out_ptr = 0;
1537         return ispipe;
1538 }
1539
1540 static int zap_process(struct task_struct *start)
1541 {
1542         struct task_struct *t;
1543         int nr = 0;
1544
1545         start->signal->flags = SIGNAL_GROUP_EXIT;
1546         start->signal->group_stop_count = 0;
1547
1548         t = start;
1549         do {
1550                 if (t != current && t->mm) {
1551                         sigaddset(&t->pending.signal, SIGKILL);
1552                         signal_wake_up(t, 1);
1553                         nr++;
1554                 }
1555         } while_each_thread(start, t);
1556
1557         return nr;
1558 }
1559
1560 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1561                                 struct core_state *core_state, int exit_code)
1562 {
1563         struct task_struct *g, *p;
1564         unsigned long flags;
1565         int nr = -EAGAIN;
1566
1567         spin_lock_irq(&tsk->sighand->siglock);
1568         if (!signal_group_exit(tsk->signal)) {
1569                 mm->core_state = core_state;
1570                 tsk->signal->group_exit_code = exit_code;
1571                 nr = zap_process(tsk);
1572         }
1573         spin_unlock_irq(&tsk->sighand->siglock);
1574         if (unlikely(nr < 0))
1575                 return nr;
1576
1577         if (atomic_read(&mm->mm_users) == nr + 1)
1578                 goto done;
1579         /*
1580          * We should find and kill all tasks which use this mm, and we should
1581          * count them correctly into ->nr_threads. We don't take tasklist
1582          * lock, but this is safe wrt:
1583          *
1584          * fork:
1585          *      None of sub-threads can fork after zap_process(leader). All
1586          *      processes which were created before this point should be
1587          *      visible to zap_threads() because copy_process() adds the new
1588          *      process to the tail of init_task.tasks list, and lock/unlock
1589          *      of ->siglock provides a memory barrier.
1590          *
1591          * do_exit:
1592          *      The caller holds mm->mmap_sem. This means that the task which
1593          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1594          *      its ->mm.
1595          *
1596          * de_thread:
1597          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1598          *      we must see either old or new leader, this does not matter.
1599          *      However, it can change p->sighand, so lock_task_sighand(p)
1600          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1601          *      it can't fail.
1602          *
1603          *      Note also that "g" can be the old leader with ->mm == NULL
1604          *      and already unhashed and thus removed from ->thread_group.
1605          *      This is OK, __unhash_process()->list_del_rcu() does not
1606          *      clear the ->next pointer, we will find the new leader via
1607          *      next_thread().
1608          */
1609         rcu_read_lock();
1610         for_each_process(g) {
1611                 if (g == tsk->group_leader)
1612                         continue;
1613                 if (g->flags & PF_KTHREAD)
1614                         continue;
1615                 p = g;
1616                 do {
1617                         if (p->mm) {
1618                                 if (unlikely(p->mm == mm)) {
1619                                         lock_task_sighand(p, &flags);
1620                                         nr += zap_process(p);
1621                                         unlock_task_sighand(p, &flags);
1622                                 }
1623                                 break;
1624                         }
1625                 } while_each_thread(g, p);
1626         }
1627         rcu_read_unlock();
1628 done:
1629         atomic_set(&core_state->nr_threads, nr);
1630         return nr;
1631 }
1632
1633 static int coredump_wait(int exit_code, struct core_state *core_state)
1634 {
1635         struct task_struct *tsk = current;
1636         struct mm_struct *mm = tsk->mm;
1637         struct completion *vfork_done;
1638         int core_waiters;
1639
1640         init_completion(&core_state->startup);
1641         core_state->dumper.task = tsk;
1642         core_state->dumper.next = NULL;
1643         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1644         up_write(&mm->mmap_sem);
1645
1646         if (unlikely(core_waiters < 0))
1647                 goto fail;
1648
1649         /*
1650          * Make sure nobody is waiting for us to release the VM,
1651          * otherwise we can deadlock when we wait on each other
1652          */
1653         vfork_done = tsk->vfork_done;
1654         if (vfork_done) {
1655                 tsk->vfork_done = NULL;
1656                 complete(vfork_done);
1657         }
1658
1659         if (core_waiters)
1660                 wait_for_completion(&core_state->startup);
1661 fail:
1662         return core_waiters;
1663 }
1664
1665 static void coredump_finish(struct mm_struct *mm)
1666 {
1667         struct core_thread *curr, *next;
1668         struct task_struct *task;
1669
1670         next = mm->core_state->dumper.next;
1671         while ((curr = next) != NULL) {
1672                 next = curr->next;
1673                 task = curr->task;
1674                 /*
1675                  * see exit_mm(), curr->task must not see
1676                  * ->task == NULL before we read ->next.
1677                  */
1678                 smp_mb();
1679                 curr->task = NULL;
1680                 wake_up_process(task);
1681         }
1682
1683         mm->core_state = NULL;
1684 }
1685
1686 /*
1687  * set_dumpable converts traditional three-value dumpable to two flags and
1688  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1689  * these bits are not changed atomically.  So get_dumpable can observe the
1690  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1691  * return either old dumpable or new one by paying attention to the order of
1692  * modifying the bits.
1693  *
1694  * dumpable |   mm->flags (binary)
1695  * old  new | initial interim  final
1696  * ---------+-----------------------
1697  *  0    1  |   00      01      01
1698  *  0    2  |   00      10(*)   11
1699  *  1    0  |   01      00      00
1700  *  1    2  |   01      11      11
1701  *  2    0  |   11      10(*)   00
1702  *  2    1  |   11      11      01
1703  *
1704  * (*) get_dumpable regards interim value of 10 as 11.
1705  */
1706 void set_dumpable(struct mm_struct *mm, int value)
1707 {
1708         switch (value) {
1709         case 0:
1710                 clear_bit(MMF_DUMPABLE, &mm->flags);
1711                 smp_wmb();
1712                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1713                 break;
1714         case 1:
1715                 set_bit(MMF_DUMPABLE, &mm->flags);
1716                 smp_wmb();
1717                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1718                 break;
1719         case 2:
1720                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1721                 smp_wmb();
1722                 set_bit(MMF_DUMPABLE, &mm->flags);
1723                 break;
1724         }
1725 }
1726
1727 int get_dumpable(struct mm_struct *mm)
1728 {
1729         int ret;
1730
1731         ret = mm->flags & 0x3;
1732         return (ret >= 2) ? 2 : ret;
1733 }
1734
1735 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1736 {
1737         struct core_state core_state;
1738         char corename[CORENAME_MAX_SIZE + 1];
1739         struct mm_struct *mm = current->mm;
1740         struct linux_binfmt * binfmt;
1741         struct inode * inode;
1742         struct file * file;
1743         const struct cred *old_cred;
1744         struct cred *cred;
1745         int retval = 0;
1746         int flag = 0;
1747         int ispipe = 0;
1748         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1749         char **helper_argv = NULL;
1750         int helper_argc = 0;
1751         char *delimit;
1752
1753         audit_core_dumps(signr);
1754
1755         binfmt = current->binfmt;
1756         if (!binfmt || !binfmt->core_dump)
1757                 goto fail;
1758
1759         cred = prepare_creds();
1760         if (!cred) {
1761                 retval = -ENOMEM;
1762                 goto fail;
1763         }
1764
1765         down_write(&mm->mmap_sem);
1766         /*
1767          * If another thread got here first, or we are not dumpable, bail out.
1768          */
1769         if (mm->core_state || !get_dumpable(mm)) {
1770                 up_write(&mm->mmap_sem);
1771                 put_cred(cred);
1772                 goto fail;
1773         }
1774
1775         /*
1776          *      We cannot trust fsuid as being the "true" uid of the
1777          *      process nor do we know its entire history. We only know it
1778          *      was tainted so we dump it as root in mode 2.
1779          */
1780         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1781                 flag = O_EXCL;          /* Stop rewrite attacks */
1782                 cred->fsuid = 0;        /* Dump root private */
1783         }
1784
1785         retval = coredump_wait(exit_code, &core_state);
1786         if (retval < 0) {
1787                 put_cred(cred);
1788                 goto fail;
1789         }
1790
1791         old_cred = override_creds(cred);
1792
1793         /*
1794          * Clear any false indication of pending signals that might
1795          * be seen by the filesystem code called to write the core file.
1796          */
1797         clear_thread_flag(TIF_SIGPENDING);
1798
1799         /*
1800          * lock_kernel() because format_corename() is controlled by sysctl, which
1801          * uses lock_kernel()
1802          */
1803         lock_kernel();
1804         ispipe = format_corename(corename, signr);
1805         unlock_kernel();
1806         /*
1807          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1808          * to a pipe.  Since we're not writing directly to the filesystem
1809          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1810          * created unless the pipe reader choses to write out the core file
1811          * at which point file size limits and permissions will be imposed
1812          * as it does with any other process
1813          */
1814         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1815                 goto fail_unlock;
1816
1817         if (ispipe) {
1818                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1819                 if (!helper_argv) {
1820                         printk(KERN_WARNING "%s failed to allocate memory\n",
1821                                __func__);
1822                         goto fail_unlock;
1823                 }
1824                 /* Terminate the string before the first option */
1825                 delimit = strchr(corename, ' ');
1826                 if (delimit)
1827                         *delimit = '\0';
1828                 delimit = strrchr(helper_argv[0], '/');
1829                 if (delimit)
1830                         delimit++;
1831                 else
1832                         delimit = helper_argv[0];
1833                 if (!strcmp(delimit, current->comm)) {
1834                         printk(KERN_NOTICE "Recursive core dump detected, "
1835                                         "aborting\n");
1836                         goto fail_unlock;
1837                 }
1838
1839                 core_limit = RLIM_INFINITY;
1840
1841                 /* SIGPIPE can happen, but it's just never processed */
1842                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1843                                 &file)) {
1844                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1845                                corename);
1846                         goto fail_unlock;
1847                 }
1848         } else
1849                 file = filp_open(corename,
1850                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1851                                  0600);
1852         if (IS_ERR(file))
1853                 goto fail_unlock;
1854         inode = file->f_path.dentry->d_inode;
1855         if (inode->i_nlink > 1)
1856                 goto close_fail;        /* multiple links - don't dump */
1857         if (!ispipe && d_unhashed(file->f_path.dentry))
1858                 goto close_fail;
1859
1860         /* AK: actually i see no reason to not allow this for named pipes etc.,
1861            but keep the previous behaviour for now. */
1862         if (!ispipe && !S_ISREG(inode->i_mode))
1863                 goto close_fail;
1864         /*
1865          * Dont allow local users get cute and trick others to coredump
1866          * into their pre-created files:
1867          */
1868         if (inode->i_uid != current_fsuid())
1869                 goto close_fail;
1870         if (!file->f_op)
1871                 goto close_fail;
1872         if (!file->f_op->write)
1873                 goto close_fail;
1874         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1875                 goto close_fail;
1876
1877         retval = binfmt->core_dump(signr, regs, file, core_limit);
1878
1879         if (retval)
1880                 current->signal->group_exit_code |= 0x80;
1881 close_fail:
1882         filp_close(file, NULL);
1883 fail_unlock:
1884         if (helper_argv)
1885                 argv_free(helper_argv);
1886
1887         revert_creds(old_cred);
1888         put_cred(cred);
1889         coredump_finish(mm);
1890 fail:
1891         return;
1892 }