GFS2: Extend umount wait coverage to full glock lifetime
[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/perf_event.h>
37 #include <linux/highmem.h>
38 #include <linux/spinlock.h>
39 #include <linux/key.h>
40 #include <linux/personality.h>
41 #include <linux/binfmts.h>
42 #include <linux/utsname.h>
43 #include <linux/pid_namespace.h>
44 #include <linux/module.h>
45 #include <linux/namei.h>
46 #include <linux/proc_fs.h>
47 #include <linux/mount.h>
48 #include <linux/security.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 #include <linux/pipe_fs_i.h>
58
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
61 #include <asm/tlb.h>
62 #include "internal.h"
63
64 int core_uses_pid;
65 char core_pattern[CORENAME_MAX_SIZE] = "core";
66 unsigned int core_pipe_limit;
67 int suid_dumpable = 0;
68
69 /* The maximal length of core_pattern is also specified in sysctl.c */
70
71 static LIST_HEAD(formats);
72 static DEFINE_RWLOCK(binfmt_lock);
73
74 int __register_binfmt(struct linux_binfmt * fmt, int insert)
75 {
76         if (!fmt)
77                 return -EINVAL;
78         write_lock(&binfmt_lock);
79         insert ? list_add(&fmt->lh, &formats) :
80                  list_add_tail(&fmt->lh, &formats);
81         write_unlock(&binfmt_lock);
82         return 0;       
83 }
84
85 EXPORT_SYMBOL(__register_binfmt);
86
87 void unregister_binfmt(struct linux_binfmt * fmt)
88 {
89         write_lock(&binfmt_lock);
90         list_del(&fmt->lh);
91         write_unlock(&binfmt_lock);
92 }
93
94 EXPORT_SYMBOL(unregister_binfmt);
95
96 static inline void put_binfmt(struct linux_binfmt * fmt)
97 {
98         module_put(fmt->module);
99 }
100
101 /*
102  * Note that a shared library must be both readable and executable due to
103  * security reasons.
104  *
105  * Also note that we take the address to load from from the file itself.
106  */
107 SYSCALL_DEFINE1(uselib, const char __user *, library)
108 {
109         struct file *file;
110         char *tmp = getname(library);
111         int error = PTR_ERR(tmp);
112
113         if (IS_ERR(tmp))
114                 goto out;
115
116         file = do_filp_open(AT_FDCWD, tmp,
117                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
118                                 MAY_READ | MAY_EXEC | MAY_OPEN);
119         putname(tmp);
120         error = PTR_ERR(file);
121         if (IS_ERR(file))
122                 goto out;
123
124         error = -EINVAL;
125         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
126                 goto exit;
127
128         error = -EACCES;
129         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
130                 goto exit;
131
132         fsnotify_open(file->f_path.dentry);
133
134         error = -ENOEXEC;
135         if(file->f_op) {
136                 struct linux_binfmt * fmt;
137
138                 read_lock(&binfmt_lock);
139                 list_for_each_entry(fmt, &formats, lh) {
140                         if (!fmt->load_shlib)
141                                 continue;
142                         if (!try_module_get(fmt->module))
143                                 continue;
144                         read_unlock(&binfmt_lock);
145                         error = fmt->load_shlib(file);
146                         read_lock(&binfmt_lock);
147                         put_binfmt(fmt);
148                         if (error != -ENOEXEC)
149                                 break;
150                 }
151                 read_unlock(&binfmt_lock);
152         }
153 exit:
154         fput(file);
155 out:
156         return error;
157 }
158
159 #ifdef CONFIG_MMU
160
161 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
162                 int write)
163 {
164         struct page *page;
165         int ret;
166
167 #ifdef CONFIG_STACK_GROWSUP
168         if (write) {
169                 ret = expand_stack_downwards(bprm->vma, pos);
170                 if (ret < 0)
171                         return NULL;
172         }
173 #endif
174         ret = get_user_pages(current, bprm->mm, pos,
175                         1, write, 1, &page, NULL);
176         if (ret <= 0)
177                 return NULL;
178
179         if (write) {
180                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
181                 struct rlimit *rlim;
182
183                 /*
184                  * We've historically supported up to 32 pages (ARG_MAX)
185                  * of argument strings even with small stacks
186                  */
187                 if (size <= ARG_MAX)
188                         return page;
189
190                 /*
191                  * Limit to 1/4-th the stack size for the argv+env strings.
192                  * This ensures that:
193                  *  - the remaining binfmt code will not run out of stack space,
194                  *  - the program will have a reasonable amount of stack left
195                  *    to work from.
196                  */
197                 rlim = current->signal->rlim;
198                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
199                         put_page(page);
200                         return NULL;
201                 }
202         }
203
204         return page;
205 }
206
207 static void put_arg_page(struct page *page)
208 {
209         put_page(page);
210 }
211
212 static void free_arg_page(struct linux_binprm *bprm, int i)
213 {
214 }
215
216 static void free_arg_pages(struct linux_binprm *bprm)
217 {
218 }
219
220 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
221                 struct page *page)
222 {
223         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
224 }
225
226 static int __bprm_mm_init(struct linux_binprm *bprm)
227 {
228         int err;
229         struct vm_area_struct *vma = NULL;
230         struct mm_struct *mm = bprm->mm;
231
232         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
233         if (!vma)
234                 return -ENOMEM;
235
236         down_write(&mm->mmap_sem);
237         vma->vm_mm = mm;
238
239         /*
240          * Place the stack at the largest stack address the architecture
241          * supports. Later, we'll move this to an appropriate place. We don't
242          * use STACK_TOP because that can depend on attributes which aren't
243          * configured yet.
244          */
245         vma->vm_end = STACK_TOP_MAX;
246         vma->vm_start = vma->vm_end - PAGE_SIZE;
247         vma->vm_flags = VM_STACK_FLAGS;
248         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
249         err = insert_vm_struct(mm, vma);
250         if (err)
251                 goto err;
252
253         mm->stack_vm = mm->total_vm = 1;
254         up_write(&mm->mmap_sem);
255         bprm->p = vma->vm_end - sizeof(void *);
256         return 0;
257 err:
258         up_write(&mm->mmap_sem);
259         bprm->vma = NULL;
260         kmem_cache_free(vm_area_cachep, vma);
261         return err;
262 }
263
264 static bool valid_arg_len(struct linux_binprm *bprm, long len)
265 {
266         return len <= MAX_ARG_STRLEN;
267 }
268
269 #else
270
271 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
272                 int write)
273 {
274         struct page *page;
275
276         page = bprm->page[pos / PAGE_SIZE];
277         if (!page && write) {
278                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
279                 if (!page)
280                         return NULL;
281                 bprm->page[pos / PAGE_SIZE] = page;
282         }
283
284         return page;
285 }
286
287 static void put_arg_page(struct page *page)
288 {
289 }
290
291 static void free_arg_page(struct linux_binprm *bprm, int i)
292 {
293         if (bprm->page[i]) {
294                 __free_page(bprm->page[i]);
295                 bprm->page[i] = NULL;
296         }
297 }
298
299 static void free_arg_pages(struct linux_binprm *bprm)
300 {
301         int i;
302
303         for (i = 0; i < MAX_ARG_PAGES; i++)
304                 free_arg_page(bprm, i);
305 }
306
307 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
308                 struct page *page)
309 {
310 }
311
312 static int __bprm_mm_init(struct linux_binprm *bprm)
313 {
314         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
315         return 0;
316 }
317
318 static bool valid_arg_len(struct linux_binprm *bprm, long len)
319 {
320         return len <= bprm->p;
321 }
322
323 #endif /* CONFIG_MMU */
324
325 /*
326  * Create a new mm_struct and populate it with a temporary stack
327  * vm_area_struct.  We don't have enough context at this point to set the stack
328  * flags, permissions, and offset, so we use temporary values.  We'll update
329  * them later in setup_arg_pages().
330  */
331 int bprm_mm_init(struct linux_binprm *bprm)
332 {
333         int err;
334         struct mm_struct *mm = NULL;
335
336         bprm->mm = mm = mm_alloc();
337         err = -ENOMEM;
338         if (!mm)
339                 goto err;
340
341         err = init_new_context(current, mm);
342         if (err)
343                 goto err;
344
345         err = __bprm_mm_init(bprm);
346         if (err)
347                 goto err;
348
349         return 0;
350
351 err:
352         if (mm) {
353                 bprm->mm = NULL;
354                 mmdrop(mm);
355         }
356
357         return err;
358 }
359
360 /*
361  * count() counts the number of strings in array ARGV.
362  */
363 static int count(char __user * __user * argv, int max)
364 {
365         int i = 0;
366
367         if (argv != NULL) {
368                 for (;;) {
369                         char __user * p;
370
371                         if (get_user(p, argv))
372                                 return -EFAULT;
373                         if (!p)
374                                 break;
375                         argv++;
376                         if (i++ >= max)
377                                 return -E2BIG;
378                         cond_resched();
379                 }
380         }
381         return i;
382 }
383
384 /*
385  * 'copy_strings()' copies argument/environment strings from the old
386  * processes's memory to the new process's stack.  The call to get_user_pages()
387  * ensures the destination page is created and not swapped out.
388  */
389 static int copy_strings(int argc, char __user * __user * argv,
390                         struct linux_binprm *bprm)
391 {
392         struct page *kmapped_page = NULL;
393         char *kaddr = NULL;
394         unsigned long kpos = 0;
395         int ret;
396
397         while (argc-- > 0) {
398                 char __user *str;
399                 int len;
400                 unsigned long pos;
401
402                 if (get_user(str, argv+argc) ||
403                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
404                         ret = -EFAULT;
405                         goto out;
406                 }
407
408                 if (!valid_arg_len(bprm, len)) {
409                         ret = -E2BIG;
410                         goto out;
411                 }
412
413                 /* We're going to work our way backwords. */
414                 pos = bprm->p;
415                 str += len;
416                 bprm->p -= len;
417
418                 while (len > 0) {
419                         int offset, bytes_to_copy;
420
421                         offset = pos % PAGE_SIZE;
422                         if (offset == 0)
423                                 offset = PAGE_SIZE;
424
425                         bytes_to_copy = offset;
426                         if (bytes_to_copy > len)
427                                 bytes_to_copy = len;
428
429                         offset -= bytes_to_copy;
430                         pos -= bytes_to_copy;
431                         str -= bytes_to_copy;
432                         len -= bytes_to_copy;
433
434                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
435                                 struct page *page;
436
437                                 page = get_arg_page(bprm, pos, 1);
438                                 if (!page) {
439                                         ret = -E2BIG;
440                                         goto out;
441                                 }
442
443                                 if (kmapped_page) {
444                                         flush_kernel_dcache_page(kmapped_page);
445                                         kunmap(kmapped_page);
446                                         put_arg_page(kmapped_page);
447                                 }
448                                 kmapped_page = page;
449                                 kaddr = kmap(kmapped_page);
450                                 kpos = pos & PAGE_MASK;
451                                 flush_arg_page(bprm, kpos, kmapped_page);
452                         }
453                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
454                                 ret = -EFAULT;
455                                 goto out;
456                         }
457                 }
458         }
459         ret = 0;
460 out:
461         if (kmapped_page) {
462                 flush_kernel_dcache_page(kmapped_page);
463                 kunmap(kmapped_page);
464                 put_arg_page(kmapped_page);
465         }
466         return ret;
467 }
468
469 /*
470  * Like copy_strings, but get argv and its values from kernel memory.
471  */
472 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
473 {
474         int r;
475         mm_segment_t oldfs = get_fs();
476         set_fs(KERNEL_DS);
477         r = copy_strings(argc, (char __user * __user *)argv, bprm);
478         set_fs(oldfs);
479         return r;
480 }
481 EXPORT_SYMBOL(copy_strings_kernel);
482
483 #ifdef CONFIG_MMU
484
485 /*
486  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
487  * the binfmt code determines where the new stack should reside, we shift it to
488  * its final location.  The process proceeds as follows:
489  *
490  * 1) Use shift to calculate the new vma endpoints.
491  * 2) Extend vma to cover both the old and new ranges.  This ensures the
492  *    arguments passed to subsequent functions are consistent.
493  * 3) Move vma's page tables to the new range.
494  * 4) Free up any cleared pgd range.
495  * 5) Shrink the vma to cover only the new range.
496  */
497 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
498 {
499         struct mm_struct *mm = vma->vm_mm;
500         unsigned long old_start = vma->vm_start;
501         unsigned long old_end = vma->vm_end;
502         unsigned long length = old_end - old_start;
503         unsigned long new_start = old_start - shift;
504         unsigned long new_end = old_end - shift;
505         struct mmu_gather *tlb;
506
507         BUG_ON(new_start > new_end);
508
509         /*
510          * ensure there are no vmas between where we want to go
511          * and where we are
512          */
513         if (vma != find_vma(mm, new_start))
514                 return -EFAULT;
515
516         /*
517          * cover the whole range: [new_start, old_end)
518          */
519         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
520
521         /*
522          * move the page tables downwards, on failure we rely on
523          * process cleanup to remove whatever mess we made.
524          */
525         if (length != move_page_tables(vma, old_start,
526                                        vma, new_start, length))
527                 return -ENOMEM;
528
529         lru_add_drain();
530         tlb = tlb_gather_mmu(mm, 0);
531         if (new_end > old_start) {
532                 /*
533                  * when the old and new regions overlap clear from new_end.
534                  */
535                 free_pgd_range(tlb, new_end, old_end, new_end,
536                         vma->vm_next ? vma->vm_next->vm_start : 0);
537         } else {
538                 /*
539                  * otherwise, clean from old_start; this is done to not touch
540                  * the address space in [new_end, old_start) some architectures
541                  * have constraints on va-space that make this illegal (IA64) -
542                  * for the others its just a little faster.
543                  */
544                 free_pgd_range(tlb, old_start, old_end, new_end,
545                         vma->vm_next ? vma->vm_next->vm_start : 0);
546         }
547         tlb_finish_mmu(tlb, new_end, old_end);
548
549         /*
550          * shrink the vma to just the new range.
551          */
552         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
553
554         return 0;
555 }
556
557 #define EXTRA_STACK_VM_PAGES    20      /* random */
558
559 /*
560  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
561  * the stack is optionally relocated, and some extra space is added.
562  */
563 int setup_arg_pages(struct linux_binprm *bprm,
564                     unsigned long stack_top,
565                     int executable_stack)
566 {
567         unsigned long ret;
568         unsigned long stack_shift;
569         struct mm_struct *mm = current->mm;
570         struct vm_area_struct *vma = bprm->vma;
571         struct vm_area_struct *prev = NULL;
572         unsigned long vm_flags;
573         unsigned long stack_base;
574
575 #ifdef CONFIG_STACK_GROWSUP
576         /* Limit stack size to 1GB */
577         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
578         if (stack_base > (1 << 30))
579                 stack_base = 1 << 30;
580
581         /* Make sure we didn't let the argument array grow too large. */
582         if (vma->vm_end - vma->vm_start > stack_base)
583                 return -ENOMEM;
584
585         stack_base = PAGE_ALIGN(stack_top - stack_base);
586
587         stack_shift = vma->vm_start - stack_base;
588         mm->arg_start = bprm->p - stack_shift;
589         bprm->p = vma->vm_end - stack_shift;
590 #else
591         stack_top = arch_align_stack(stack_top);
592         stack_top = PAGE_ALIGN(stack_top);
593         stack_shift = vma->vm_end - stack_top;
594
595         bprm->p -= stack_shift;
596         mm->arg_start = bprm->p;
597 #endif
598
599         if (bprm->loader)
600                 bprm->loader -= stack_shift;
601         bprm->exec -= stack_shift;
602
603         down_write(&mm->mmap_sem);
604         vm_flags = VM_STACK_FLAGS;
605
606         /*
607          * Adjust stack execute permissions; explicitly enable for
608          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
609          * (arch default) otherwise.
610          */
611         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
612                 vm_flags |= VM_EXEC;
613         else if (executable_stack == EXSTACK_DISABLE_X)
614                 vm_flags &= ~VM_EXEC;
615         vm_flags |= mm->def_flags;
616
617         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
618                         vm_flags);
619         if (ret)
620                 goto out_unlock;
621         BUG_ON(prev != vma);
622
623         /* Move stack pages down in memory. */
624         if (stack_shift) {
625                 ret = shift_arg_pages(vma, stack_shift);
626                 if (ret)
627                         goto out_unlock;
628         }
629
630 #ifdef CONFIG_STACK_GROWSUP
631         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
632 #else
633         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
634 #endif
635         ret = expand_stack(vma, stack_base);
636         if (ret)
637                 ret = -EFAULT;
638
639 out_unlock:
640         up_write(&mm->mmap_sem);
641         return ret;
642 }
643 EXPORT_SYMBOL(setup_arg_pages);
644
645 #endif /* CONFIG_MMU */
646
647 struct file *open_exec(const char *name)
648 {
649         struct file *file;
650         int err;
651
652         file = do_filp_open(AT_FDCWD, name,
653                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
654                                 MAY_EXEC | MAY_OPEN);
655         if (IS_ERR(file))
656                 goto out;
657
658         err = -EACCES;
659         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
660                 goto exit;
661
662         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
663                 goto exit;
664
665         fsnotify_open(file->f_path.dentry);
666
667         err = deny_write_access(file);
668         if (err)
669                 goto exit;
670
671 out:
672         return file;
673
674 exit:
675         fput(file);
676         return ERR_PTR(err);
677 }
678 EXPORT_SYMBOL(open_exec);
679
680 int kernel_read(struct file *file, loff_t offset,
681                 char *addr, unsigned long count)
682 {
683         mm_segment_t old_fs;
684         loff_t pos = offset;
685         int result;
686
687         old_fs = get_fs();
688         set_fs(get_ds());
689         /* The cast to a user pointer is valid due to the set_fs() */
690         result = vfs_read(file, (void __user *)addr, count, &pos);
691         set_fs(old_fs);
692         return result;
693 }
694
695 EXPORT_SYMBOL(kernel_read);
696
697 static int exec_mmap(struct mm_struct *mm)
698 {
699         struct task_struct *tsk;
700         struct mm_struct * old_mm, *active_mm;
701
702         /* Notify parent that we're no longer interested in the old VM */
703         tsk = current;
704         old_mm = current->mm;
705         mm_release(tsk, old_mm);
706
707         if (old_mm) {
708                 /*
709                  * Make sure that if there is a core dump in progress
710                  * for the old mm, we get out and die instead of going
711                  * through with the exec.  We must hold mmap_sem around
712                  * checking core_state and changing tsk->mm.
713                  */
714                 down_read(&old_mm->mmap_sem);
715                 if (unlikely(old_mm->core_state)) {
716                         up_read(&old_mm->mmap_sem);
717                         return -EINTR;
718                 }
719         }
720         task_lock(tsk);
721         active_mm = tsk->active_mm;
722         tsk->mm = mm;
723         tsk->active_mm = mm;
724         activate_mm(active_mm, mm);
725         task_unlock(tsk);
726         arch_pick_mmap_layout(mm);
727         if (old_mm) {
728                 up_read(&old_mm->mmap_sem);
729                 BUG_ON(active_mm != old_mm);
730                 mm_update_next_owner(old_mm);
731                 mmput(old_mm);
732                 return 0;
733         }
734         mmdrop(active_mm);
735         return 0;
736 }
737
738 /*
739  * This function makes sure the current process has its own signal table,
740  * so that flush_signal_handlers can later reset the handlers without
741  * disturbing other processes.  (Other processes might share the signal
742  * table via the CLONE_SIGHAND option to clone().)
743  */
744 static int de_thread(struct task_struct *tsk)
745 {
746         struct signal_struct *sig = tsk->signal;
747         struct sighand_struct *oldsighand = tsk->sighand;
748         spinlock_t *lock = &oldsighand->siglock;
749         int count;
750
751         if (thread_group_empty(tsk))
752                 goto no_thread_group;
753
754         /*
755          * Kill all other threads in the thread group.
756          */
757         spin_lock_irq(lock);
758         if (signal_group_exit(sig)) {
759                 /*
760                  * Another group action in progress, just
761                  * return so that the signal is processed.
762                  */
763                 spin_unlock_irq(lock);
764                 return -EAGAIN;
765         }
766         sig->group_exit_task = tsk;
767         zap_other_threads(tsk);
768
769         /* Account for the thread group leader hanging around: */
770         count = thread_group_leader(tsk) ? 1 : 2;
771         sig->notify_count = count;
772         while (atomic_read(&sig->count) > count) {
773                 __set_current_state(TASK_UNINTERRUPTIBLE);
774                 spin_unlock_irq(lock);
775                 schedule();
776                 spin_lock_irq(lock);
777         }
778         spin_unlock_irq(lock);
779
780         /*
781          * At this point all other threads have exited, all we have to
782          * do is to wait for the thread group leader to become inactive,
783          * and to assume its PID:
784          */
785         if (!thread_group_leader(tsk)) {
786                 struct task_struct *leader = tsk->group_leader;
787
788                 sig->notify_count = -1; /* for exit_notify() */
789                 for (;;) {
790                         write_lock_irq(&tasklist_lock);
791                         if (likely(leader->exit_state))
792                                 break;
793                         __set_current_state(TASK_UNINTERRUPTIBLE);
794                         write_unlock_irq(&tasklist_lock);
795                         schedule();
796                 }
797
798                 /*
799                  * The only record we have of the real-time age of a
800                  * process, regardless of execs it's done, is start_time.
801                  * All the past CPU time is accumulated in signal_struct
802                  * from sister threads now dead.  But in this non-leader
803                  * exec, nothing survives from the original leader thread,
804                  * whose birth marks the true age of this process now.
805                  * When we take on its identity by switching to its PID, we
806                  * also take its birthdate (always earlier than our own).
807                  */
808                 tsk->start_time = leader->start_time;
809
810                 BUG_ON(!same_thread_group(leader, tsk));
811                 BUG_ON(has_group_leader_pid(tsk));
812                 /*
813                  * An exec() starts a new thread group with the
814                  * TGID of the previous thread group. Rehash the
815                  * two threads with a switched PID, and release
816                  * the former thread group leader:
817                  */
818
819                 /* Become a process group leader with the old leader's pid.
820                  * The old leader becomes a thread of the this thread group.
821                  * Note: The old leader also uses this pid until release_task
822                  *       is called.  Odd but simple and correct.
823                  */
824                 detach_pid(tsk, PIDTYPE_PID);
825                 tsk->pid = leader->pid;
826                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
827                 transfer_pid(leader, tsk, PIDTYPE_PGID);
828                 transfer_pid(leader, tsk, PIDTYPE_SID);
829
830                 list_replace_rcu(&leader->tasks, &tsk->tasks);
831                 list_replace_init(&leader->sibling, &tsk->sibling);
832
833                 tsk->group_leader = tsk;
834                 leader->group_leader = tsk;
835
836                 tsk->exit_signal = SIGCHLD;
837
838                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
839                 leader->exit_state = EXIT_DEAD;
840                 write_unlock_irq(&tasklist_lock);
841
842                 release_task(leader);
843         }
844
845         sig->group_exit_task = NULL;
846         sig->notify_count = 0;
847
848 no_thread_group:
849         if (current->mm)
850                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
851
852         exit_itimers(sig);
853         flush_itimer_signals();
854
855         if (atomic_read(&oldsighand->count) != 1) {
856                 struct sighand_struct *newsighand;
857                 /*
858                  * This ->sighand is shared with the CLONE_SIGHAND
859                  * but not CLONE_THREAD task, switch to the new one.
860                  */
861                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
862                 if (!newsighand)
863                         return -ENOMEM;
864
865                 atomic_set(&newsighand->count, 1);
866                 memcpy(newsighand->action, oldsighand->action,
867                        sizeof(newsighand->action));
868
869                 write_lock_irq(&tasklist_lock);
870                 spin_lock(&oldsighand->siglock);
871                 rcu_assign_pointer(tsk->sighand, newsighand);
872                 spin_unlock(&oldsighand->siglock);
873                 write_unlock_irq(&tasklist_lock);
874
875                 __cleanup_sighand(oldsighand);
876         }
877
878         BUG_ON(!thread_group_leader(tsk));
879         return 0;
880 }
881
882 /*
883  * These functions flushes out all traces of the currently running executable
884  * so that a new one can be started
885  */
886 static void flush_old_files(struct files_struct * files)
887 {
888         long j = -1;
889         struct fdtable *fdt;
890
891         spin_lock(&files->file_lock);
892         for (;;) {
893                 unsigned long set, i;
894
895                 j++;
896                 i = j * __NFDBITS;
897                 fdt = files_fdtable(files);
898                 if (i >= fdt->max_fds)
899                         break;
900                 set = fdt->close_on_exec->fds_bits[j];
901                 if (!set)
902                         continue;
903                 fdt->close_on_exec->fds_bits[j] = 0;
904                 spin_unlock(&files->file_lock);
905                 for ( ; set ; i++,set >>= 1) {
906                         if (set & 1) {
907                                 sys_close(i);
908                         }
909                 }
910                 spin_lock(&files->file_lock);
911
912         }
913         spin_unlock(&files->file_lock);
914 }
915
916 char *get_task_comm(char *buf, struct task_struct *tsk)
917 {
918         /* buf must be at least sizeof(tsk->comm) in size */
919         task_lock(tsk);
920         strncpy(buf, tsk->comm, sizeof(tsk->comm));
921         task_unlock(tsk);
922         return buf;
923 }
924
925 void set_task_comm(struct task_struct *tsk, char *buf)
926 {
927         task_lock(tsk);
928
929         /*
930          * Threads may access current->comm without holding
931          * the task lock, so write the string carefully.
932          * Readers without a lock may see incomplete new
933          * names but are safe from non-terminating string reads.
934          */
935         memset(tsk->comm, 0, TASK_COMM_LEN);
936         wmb();
937         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
938         task_unlock(tsk);
939         perf_event_comm(tsk);
940 }
941
942 int flush_old_exec(struct linux_binprm * bprm)
943 {
944         int retval;
945
946         /*
947          * Make sure we have a private signal table and that
948          * we are unassociated from the previous thread group.
949          */
950         retval = de_thread(current);
951         if (retval)
952                 goto out;
953
954         set_mm_exe_file(bprm->mm, bprm->file);
955
956         /*
957          * Release all of the old mmap stuff
958          */
959         retval = exec_mmap(bprm->mm);
960         if (retval)
961                 goto out;
962
963         bprm->mm = NULL;                /* We're using it now */
964
965         current->flags &= ~PF_RANDOMIZE;
966         flush_thread();
967         current->personality &= ~bprm->per_clear;
968
969         return 0;
970
971 out:
972         return retval;
973 }
974 EXPORT_SYMBOL(flush_old_exec);
975
976 void setup_new_exec(struct linux_binprm * bprm)
977 {
978         int i, ch;
979         char * name;
980         char tcomm[sizeof(current->comm)];
981
982         arch_pick_mmap_layout(current->mm);
983
984         /* This is the point of no return */
985         current->sas_ss_sp = current->sas_ss_size = 0;
986
987         if (current_euid() == current_uid() && current_egid() == current_gid())
988                 set_dumpable(current->mm, 1);
989         else
990                 set_dumpable(current->mm, suid_dumpable);
991
992         name = bprm->filename;
993
994         /* Copies the binary name from after last slash */
995         for (i=0; (ch = *(name++)) != '\0';) {
996                 if (ch == '/')
997                         i = 0; /* overwrite what we wrote */
998                 else
999                         if (i < (sizeof(tcomm) - 1))
1000                                 tcomm[i++] = ch;
1001         }
1002         tcomm[i] = '\0';
1003         set_task_comm(current, tcomm);
1004
1005         /* Set the new mm task size. We have to do that late because it may
1006          * depend on TIF_32BIT which is only updated in flush_thread() on
1007          * some architectures like powerpc
1008          */
1009         current->mm->task_size = TASK_SIZE;
1010
1011         /* install the new credentials */
1012         if (bprm->cred->uid != current_euid() ||
1013             bprm->cred->gid != current_egid()) {
1014                 current->pdeath_signal = 0;
1015         } else if (file_permission(bprm->file, MAY_READ) ||
1016                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1017                 set_dumpable(current->mm, suid_dumpable);
1018         }
1019
1020         /*
1021          * Flush performance counters when crossing a
1022          * security domain:
1023          */
1024         if (!get_dumpable(current->mm))
1025                 perf_event_exit_task(current);
1026
1027         /* An exec changes our domain. We are no longer part of the thread
1028            group */
1029
1030         current->self_exec_id++;
1031                         
1032         flush_signal_handlers(current, 0);
1033         flush_old_files(current->files);
1034 }
1035 EXPORT_SYMBOL(setup_new_exec);
1036
1037 /*
1038  * Prepare credentials and lock ->cred_guard_mutex.
1039  * install_exec_creds() commits the new creds and drops the lock.
1040  * Or, if exec fails before, free_bprm() should release ->cred and
1041  * and unlock.
1042  */
1043 int prepare_bprm_creds(struct linux_binprm *bprm)
1044 {
1045         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1046                 return -ERESTARTNOINTR;
1047
1048         bprm->cred = prepare_exec_creds();
1049         if (likely(bprm->cred))
1050                 return 0;
1051
1052         mutex_unlock(&current->cred_guard_mutex);
1053         return -ENOMEM;
1054 }
1055
1056 void free_bprm(struct linux_binprm *bprm)
1057 {
1058         free_arg_pages(bprm);
1059         if (bprm->cred) {
1060                 mutex_unlock(&current->cred_guard_mutex);
1061                 abort_creds(bprm->cred);
1062         }
1063         kfree(bprm);
1064 }
1065
1066 /*
1067  * install the new credentials for this executable
1068  */
1069 void install_exec_creds(struct linux_binprm *bprm)
1070 {
1071         security_bprm_committing_creds(bprm);
1072
1073         commit_creds(bprm->cred);
1074         bprm->cred = NULL;
1075         /*
1076          * cred_guard_mutex must be held at least to this point to prevent
1077          * ptrace_attach() from altering our determination of the task's
1078          * credentials; any time after this it may be unlocked.
1079          */
1080         security_bprm_committed_creds(bprm);
1081         mutex_unlock(&current->cred_guard_mutex);
1082 }
1083 EXPORT_SYMBOL(install_exec_creds);
1084
1085 /*
1086  * determine how safe it is to execute the proposed program
1087  * - the caller must hold current->cred_guard_mutex to protect against
1088  *   PTRACE_ATTACH
1089  */
1090 int check_unsafe_exec(struct linux_binprm *bprm)
1091 {
1092         struct task_struct *p = current, *t;
1093         unsigned n_fs;
1094         int res = 0;
1095
1096         bprm->unsafe = tracehook_unsafe_exec(p);
1097
1098         n_fs = 1;
1099         write_lock(&p->fs->lock);
1100         rcu_read_lock();
1101         for (t = next_thread(p); t != p; t = next_thread(t)) {
1102                 if (t->fs == p->fs)
1103                         n_fs++;
1104         }
1105         rcu_read_unlock();
1106
1107         if (p->fs->users > n_fs) {
1108                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1109         } else {
1110                 res = -EAGAIN;
1111                 if (!p->fs->in_exec) {
1112                         p->fs->in_exec = 1;
1113                         res = 1;
1114                 }
1115         }
1116         write_unlock(&p->fs->lock);
1117
1118         return res;
1119 }
1120
1121 /* 
1122  * Fill the binprm structure from the inode. 
1123  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1124  *
1125  * This may be called multiple times for binary chains (scripts for example).
1126  */
1127 int prepare_binprm(struct linux_binprm *bprm)
1128 {
1129         umode_t mode;
1130         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1131         int retval;
1132
1133         mode = inode->i_mode;
1134         if (bprm->file->f_op == NULL)
1135                 return -EACCES;
1136
1137         /* clear any previous set[ug]id data from a previous binary */
1138         bprm->cred->euid = current_euid();
1139         bprm->cred->egid = current_egid();
1140
1141         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1142                 /* Set-uid? */
1143                 if (mode & S_ISUID) {
1144                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1145                         bprm->cred->euid = inode->i_uid;
1146                 }
1147
1148                 /* Set-gid? */
1149                 /*
1150                  * If setgid is set but no group execute bit then this
1151                  * is a candidate for mandatory locking, not a setgid
1152                  * executable.
1153                  */
1154                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1155                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1156                         bprm->cred->egid = inode->i_gid;
1157                 }
1158         }
1159
1160         /* fill in binprm security blob */
1161         retval = security_bprm_set_creds(bprm);
1162         if (retval)
1163                 return retval;
1164         bprm->cred_prepared = 1;
1165
1166         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1167         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1168 }
1169
1170 EXPORT_SYMBOL(prepare_binprm);
1171
1172 /*
1173  * Arguments are '\0' separated strings found at the location bprm->p
1174  * points to; chop off the first by relocating brpm->p to right after
1175  * the first '\0' encountered.
1176  */
1177 int remove_arg_zero(struct linux_binprm *bprm)
1178 {
1179         int ret = 0;
1180         unsigned long offset;
1181         char *kaddr;
1182         struct page *page;
1183
1184         if (!bprm->argc)
1185                 return 0;
1186
1187         do {
1188                 offset = bprm->p & ~PAGE_MASK;
1189                 page = get_arg_page(bprm, bprm->p, 0);
1190                 if (!page) {
1191                         ret = -EFAULT;
1192                         goto out;
1193                 }
1194                 kaddr = kmap_atomic(page, KM_USER0);
1195
1196                 for (; offset < PAGE_SIZE && kaddr[offset];
1197                                 offset++, bprm->p++)
1198                         ;
1199
1200                 kunmap_atomic(kaddr, KM_USER0);
1201                 put_arg_page(page);
1202
1203                 if (offset == PAGE_SIZE)
1204                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1205         } while (offset == PAGE_SIZE);
1206
1207         bprm->p++;
1208         bprm->argc--;
1209         ret = 0;
1210
1211 out:
1212         return ret;
1213 }
1214 EXPORT_SYMBOL(remove_arg_zero);
1215
1216 /*
1217  * cycle the list of binary formats handler, until one recognizes the image
1218  */
1219 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1220 {
1221         unsigned int depth = bprm->recursion_depth;
1222         int try,retval;
1223         struct linux_binfmt *fmt;
1224
1225         retval = security_bprm_check(bprm);
1226         if (retval)
1227                 return retval;
1228
1229         /* kernel module loader fixup */
1230         /* so we don't try to load run modprobe in kernel space. */
1231         set_fs(USER_DS);
1232
1233         retval = audit_bprm(bprm);
1234         if (retval)
1235                 return retval;
1236
1237         retval = -ENOENT;
1238         for (try=0; try<2; try++) {
1239                 read_lock(&binfmt_lock);
1240                 list_for_each_entry(fmt, &formats, lh) {
1241                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1242                         if (!fn)
1243                                 continue;
1244                         if (!try_module_get(fmt->module))
1245                                 continue;
1246                         read_unlock(&binfmt_lock);
1247                         retval = fn(bprm, regs);
1248                         /*
1249                          * Restore the depth counter to its starting value
1250                          * in this call, so we don't have to rely on every
1251                          * load_binary function to restore it on return.
1252                          */
1253                         bprm->recursion_depth = depth;
1254                         if (retval >= 0) {
1255                                 if (depth == 0)
1256                                         tracehook_report_exec(fmt, bprm, regs);
1257                                 put_binfmt(fmt);
1258                                 allow_write_access(bprm->file);
1259                                 if (bprm->file)
1260                                         fput(bprm->file);
1261                                 bprm->file = NULL;
1262                                 current->did_exec = 1;
1263                                 proc_exec_connector(current);
1264                                 return retval;
1265                         }
1266                         read_lock(&binfmt_lock);
1267                         put_binfmt(fmt);
1268                         if (retval != -ENOEXEC || bprm->mm == NULL)
1269                                 break;
1270                         if (!bprm->file) {
1271                                 read_unlock(&binfmt_lock);
1272                                 return retval;
1273                         }
1274                 }
1275                 read_unlock(&binfmt_lock);
1276                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1277                         break;
1278 #ifdef CONFIG_MODULES
1279                 } else {
1280 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1281                         if (printable(bprm->buf[0]) &&
1282                             printable(bprm->buf[1]) &&
1283                             printable(bprm->buf[2]) &&
1284                             printable(bprm->buf[3]))
1285                                 break; /* -ENOEXEC */
1286                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1287 #endif
1288                 }
1289         }
1290         return retval;
1291 }
1292
1293 EXPORT_SYMBOL(search_binary_handler);
1294
1295 /*
1296  * sys_execve() executes a new program.
1297  */
1298 int do_execve(char * filename,
1299         char __user *__user *argv,
1300         char __user *__user *envp,
1301         struct pt_regs * regs)
1302 {
1303         struct linux_binprm *bprm;
1304         struct file *file;
1305         struct files_struct *displaced;
1306         bool clear_in_exec;
1307         int retval;
1308
1309         retval = unshare_files(&displaced);
1310         if (retval)
1311                 goto out_ret;
1312
1313         retval = -ENOMEM;
1314         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1315         if (!bprm)
1316                 goto out_files;
1317
1318         retval = prepare_bprm_creds(bprm);
1319         if (retval)
1320                 goto out_free;
1321
1322         retval = check_unsafe_exec(bprm);
1323         if (retval < 0)
1324                 goto out_free;
1325         clear_in_exec = retval;
1326         current->in_execve = 1;
1327
1328         file = open_exec(filename);
1329         retval = PTR_ERR(file);
1330         if (IS_ERR(file))
1331                 goto out_unmark;
1332
1333         sched_exec();
1334
1335         bprm->file = file;
1336         bprm->filename = filename;
1337         bprm->interp = filename;
1338
1339         retval = bprm_mm_init(bprm);
1340         if (retval)
1341                 goto out_file;
1342
1343         bprm->argc = count(argv, MAX_ARG_STRINGS);
1344         if ((retval = bprm->argc) < 0)
1345                 goto out;
1346
1347         bprm->envc = count(envp, MAX_ARG_STRINGS);
1348         if ((retval = bprm->envc) < 0)
1349                 goto out;
1350
1351         retval = prepare_binprm(bprm);
1352         if (retval < 0)
1353                 goto out;
1354
1355         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1356         if (retval < 0)
1357                 goto out;
1358
1359         bprm->exec = bprm->p;
1360         retval = copy_strings(bprm->envc, envp, bprm);
1361         if (retval < 0)
1362                 goto out;
1363
1364         retval = copy_strings(bprm->argc, argv, bprm);
1365         if (retval < 0)
1366                 goto out;
1367
1368         current->flags &= ~PF_KTHREAD;
1369         retval = search_binary_handler(bprm,regs);
1370         if (retval < 0)
1371                 goto out;
1372
1373         current->stack_start = current->mm->start_stack;
1374
1375         /* execve succeeded */
1376         current->fs->in_exec = 0;
1377         current->in_execve = 0;
1378         acct_update_integrals(current);
1379         free_bprm(bprm);
1380         if (displaced)
1381                 put_files_struct(displaced);
1382         return retval;
1383
1384 out:
1385         if (bprm->mm)
1386                 mmput (bprm->mm);
1387
1388 out_file:
1389         if (bprm->file) {
1390                 allow_write_access(bprm->file);
1391                 fput(bprm->file);
1392         }
1393
1394 out_unmark:
1395         if (clear_in_exec)
1396                 current->fs->in_exec = 0;
1397         current->in_execve = 0;
1398
1399 out_free:
1400         free_bprm(bprm);
1401
1402 out_files:
1403         if (displaced)
1404                 reset_files_struct(displaced);
1405 out_ret:
1406         return retval;
1407 }
1408
1409 void set_binfmt(struct linux_binfmt *new)
1410 {
1411         struct mm_struct *mm = current->mm;
1412
1413         if (mm->binfmt)
1414                 module_put(mm->binfmt->module);
1415
1416         mm->binfmt = new;
1417         if (new)
1418                 __module_get(new->module);
1419 }
1420
1421 EXPORT_SYMBOL(set_binfmt);
1422
1423 /* format_corename will inspect the pattern parameter, and output a
1424  * name into corename, which must have space for at least
1425  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1426  */
1427 static int format_corename(char *corename, long signr)
1428 {
1429         const struct cred *cred = current_cred();
1430         const char *pat_ptr = core_pattern;
1431         int ispipe = (*pat_ptr == '|');
1432         char *out_ptr = corename;
1433         char *const out_end = corename + CORENAME_MAX_SIZE;
1434         int rc;
1435         int pid_in_pattern = 0;
1436
1437         /* Repeat as long as we have more pattern to process and more output
1438            space */
1439         while (*pat_ptr) {
1440                 if (*pat_ptr != '%') {
1441                         if (out_ptr == out_end)
1442                                 goto out;
1443                         *out_ptr++ = *pat_ptr++;
1444                 } else {
1445                         switch (*++pat_ptr) {
1446                         case 0:
1447                                 goto out;
1448                         /* Double percent, output one percent */
1449                         case '%':
1450                                 if (out_ptr == out_end)
1451                                         goto out;
1452                                 *out_ptr++ = '%';
1453                                 break;
1454                         /* pid */
1455                         case 'p':
1456                                 pid_in_pattern = 1;
1457                                 rc = snprintf(out_ptr, out_end - out_ptr,
1458                                               "%d", task_tgid_vnr(current));
1459                                 if (rc > out_end - out_ptr)
1460                                         goto out;
1461                                 out_ptr += rc;
1462                                 break;
1463                         /* uid */
1464                         case 'u':
1465                                 rc = snprintf(out_ptr, out_end - out_ptr,
1466                                               "%d", cred->uid);
1467                                 if (rc > out_end - out_ptr)
1468                                         goto out;
1469                                 out_ptr += rc;
1470                                 break;
1471                         /* gid */
1472                         case 'g':
1473                                 rc = snprintf(out_ptr, out_end - out_ptr,
1474                                               "%d", cred->gid);
1475                                 if (rc > out_end - out_ptr)
1476                                         goto out;
1477                                 out_ptr += rc;
1478                                 break;
1479                         /* signal that caused the coredump */
1480                         case 's':
1481                                 rc = snprintf(out_ptr, out_end - out_ptr,
1482                                               "%ld", signr);
1483                                 if (rc > out_end - out_ptr)
1484                                         goto out;
1485                                 out_ptr += rc;
1486                                 break;
1487                         /* UNIX time of coredump */
1488                         case 't': {
1489                                 struct timeval tv;
1490                                 do_gettimeofday(&tv);
1491                                 rc = snprintf(out_ptr, out_end - out_ptr,
1492                                               "%lu", tv.tv_sec);
1493                                 if (rc > out_end - out_ptr)
1494                                         goto out;
1495                                 out_ptr += rc;
1496                                 break;
1497                         }
1498                         /* hostname */
1499                         case 'h':
1500                                 down_read(&uts_sem);
1501                                 rc = snprintf(out_ptr, out_end - out_ptr,
1502                                               "%s", utsname()->nodename);
1503                                 up_read(&uts_sem);
1504                                 if (rc > out_end - out_ptr)
1505                                         goto out;
1506                                 out_ptr += rc;
1507                                 break;
1508                         /* executable */
1509                         case 'e':
1510                                 rc = snprintf(out_ptr, out_end - out_ptr,
1511                                               "%s", current->comm);
1512                                 if (rc > out_end - out_ptr)
1513                                         goto out;
1514                                 out_ptr += rc;
1515                                 break;
1516                         /* core limit size */
1517                         case 'c':
1518                                 rc = snprintf(out_ptr, out_end - out_ptr,
1519                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1520                                 if (rc > out_end - out_ptr)
1521                                         goto out;
1522                                 out_ptr += rc;
1523                                 break;
1524                         default:
1525                                 break;
1526                         }
1527                         ++pat_ptr;
1528                 }
1529         }
1530         /* Backward compatibility with core_uses_pid:
1531          *
1532          * If core_pattern does not include a %p (as is the default)
1533          * and core_uses_pid is set, then .%pid will be appended to
1534          * the filename. Do not do this for piped commands. */
1535         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1536                 rc = snprintf(out_ptr, out_end - out_ptr,
1537                               ".%d", task_tgid_vnr(current));
1538                 if (rc > out_end - out_ptr)
1539                         goto out;
1540                 out_ptr += rc;
1541         }
1542 out:
1543         *out_ptr = 0;
1544         return ispipe;
1545 }
1546
1547 static int zap_process(struct task_struct *start)
1548 {
1549         struct task_struct *t;
1550         int nr = 0;
1551
1552         start->signal->flags = SIGNAL_GROUP_EXIT;
1553         start->signal->group_stop_count = 0;
1554
1555         t = start;
1556         do {
1557                 if (t != current && t->mm) {
1558                         sigaddset(&t->pending.signal, SIGKILL);
1559                         signal_wake_up(t, 1);
1560                         nr++;
1561                 }
1562         } while_each_thread(start, t);
1563
1564         return nr;
1565 }
1566
1567 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1568                                 struct core_state *core_state, int exit_code)
1569 {
1570         struct task_struct *g, *p;
1571         unsigned long flags;
1572         int nr = -EAGAIN;
1573
1574         spin_lock_irq(&tsk->sighand->siglock);
1575         if (!signal_group_exit(tsk->signal)) {
1576                 mm->core_state = core_state;
1577                 tsk->signal->group_exit_code = exit_code;
1578                 nr = zap_process(tsk);
1579         }
1580         spin_unlock_irq(&tsk->sighand->siglock);
1581         if (unlikely(nr < 0))
1582                 return nr;
1583
1584         if (atomic_read(&mm->mm_users) == nr + 1)
1585                 goto done;
1586         /*
1587          * We should find and kill all tasks which use this mm, and we should
1588          * count them correctly into ->nr_threads. We don't take tasklist
1589          * lock, but this is safe wrt:
1590          *
1591          * fork:
1592          *      None of sub-threads can fork after zap_process(leader). All
1593          *      processes which were created before this point should be
1594          *      visible to zap_threads() because copy_process() adds the new
1595          *      process to the tail of init_task.tasks list, and lock/unlock
1596          *      of ->siglock provides a memory barrier.
1597          *
1598          * do_exit:
1599          *      The caller holds mm->mmap_sem. This means that the task which
1600          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1601          *      its ->mm.
1602          *
1603          * de_thread:
1604          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1605          *      we must see either old or new leader, this does not matter.
1606          *      However, it can change p->sighand, so lock_task_sighand(p)
1607          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1608          *      it can't fail.
1609          *
1610          *      Note also that "g" can be the old leader with ->mm == NULL
1611          *      and already unhashed and thus removed from ->thread_group.
1612          *      This is OK, __unhash_process()->list_del_rcu() does not
1613          *      clear the ->next pointer, we will find the new leader via
1614          *      next_thread().
1615          */
1616         rcu_read_lock();
1617         for_each_process(g) {
1618                 if (g == tsk->group_leader)
1619                         continue;
1620                 if (g->flags & PF_KTHREAD)
1621                         continue;
1622                 p = g;
1623                 do {
1624                         if (p->mm) {
1625                                 if (unlikely(p->mm == mm)) {
1626                                         lock_task_sighand(p, &flags);
1627                                         nr += zap_process(p);
1628                                         unlock_task_sighand(p, &flags);
1629                                 }
1630                                 break;
1631                         }
1632                 } while_each_thread(g, p);
1633         }
1634         rcu_read_unlock();
1635 done:
1636         atomic_set(&core_state->nr_threads, nr);
1637         return nr;
1638 }
1639
1640 static int coredump_wait(int exit_code, struct core_state *core_state)
1641 {
1642         struct task_struct *tsk = current;
1643         struct mm_struct *mm = tsk->mm;
1644         struct completion *vfork_done;
1645         int core_waiters;
1646
1647         init_completion(&core_state->startup);
1648         core_state->dumper.task = tsk;
1649         core_state->dumper.next = NULL;
1650         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1651         up_write(&mm->mmap_sem);
1652
1653         if (unlikely(core_waiters < 0))
1654                 goto fail;
1655
1656         /*
1657          * Make sure nobody is waiting for us to release the VM,
1658          * otherwise we can deadlock when we wait on each other
1659          */
1660         vfork_done = tsk->vfork_done;
1661         if (vfork_done) {
1662                 tsk->vfork_done = NULL;
1663                 complete(vfork_done);
1664         }
1665
1666         if (core_waiters)
1667                 wait_for_completion(&core_state->startup);
1668 fail:
1669         return core_waiters;
1670 }
1671
1672 static void coredump_finish(struct mm_struct *mm)
1673 {
1674         struct core_thread *curr, *next;
1675         struct task_struct *task;
1676
1677         next = mm->core_state->dumper.next;
1678         while ((curr = next) != NULL) {
1679                 next = curr->next;
1680                 task = curr->task;
1681                 /*
1682                  * see exit_mm(), curr->task must not see
1683                  * ->task == NULL before we read ->next.
1684                  */
1685                 smp_mb();
1686                 curr->task = NULL;
1687                 wake_up_process(task);
1688         }
1689
1690         mm->core_state = NULL;
1691 }
1692
1693 /*
1694  * set_dumpable converts traditional three-value dumpable to two flags and
1695  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1696  * these bits are not changed atomically.  So get_dumpable can observe the
1697  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1698  * return either old dumpable or new one by paying attention to the order of
1699  * modifying the bits.
1700  *
1701  * dumpable |   mm->flags (binary)
1702  * old  new | initial interim  final
1703  * ---------+-----------------------
1704  *  0    1  |   00      01      01
1705  *  0    2  |   00      10(*)   11
1706  *  1    0  |   01      00      00
1707  *  1    2  |   01      11      11
1708  *  2    0  |   11      10(*)   00
1709  *  2    1  |   11      11      01
1710  *
1711  * (*) get_dumpable regards interim value of 10 as 11.
1712  */
1713 void set_dumpable(struct mm_struct *mm, int value)
1714 {
1715         switch (value) {
1716         case 0:
1717                 clear_bit(MMF_DUMPABLE, &mm->flags);
1718                 smp_wmb();
1719                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1720                 break;
1721         case 1:
1722                 set_bit(MMF_DUMPABLE, &mm->flags);
1723                 smp_wmb();
1724                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1725                 break;
1726         case 2:
1727                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1728                 smp_wmb();
1729                 set_bit(MMF_DUMPABLE, &mm->flags);
1730                 break;
1731         }
1732 }
1733
1734 int get_dumpable(struct mm_struct *mm)
1735 {
1736         int ret;
1737
1738         ret = mm->flags & 0x3;
1739         return (ret >= 2) ? 2 : ret;
1740 }
1741
1742 static void wait_for_dump_helpers(struct file *file)
1743 {
1744         struct pipe_inode_info *pipe;
1745
1746         pipe = file->f_path.dentry->d_inode->i_pipe;
1747
1748         pipe_lock(pipe);
1749         pipe->readers++;
1750         pipe->writers--;
1751
1752         while ((pipe->readers > 1) && (!signal_pending(current))) {
1753                 wake_up_interruptible_sync(&pipe->wait);
1754                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1755                 pipe_wait(pipe);
1756         }
1757
1758         pipe->readers--;
1759         pipe->writers++;
1760         pipe_unlock(pipe);
1761
1762 }
1763
1764
1765 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1766 {
1767         struct core_state core_state;
1768         char corename[CORENAME_MAX_SIZE + 1];
1769         struct mm_struct *mm = current->mm;
1770         struct linux_binfmt * binfmt;
1771         struct inode * inode;
1772         const struct cred *old_cred;
1773         struct cred *cred;
1774         int retval = 0;
1775         int flag = 0;
1776         int ispipe = 0;
1777         char **helper_argv = NULL;
1778         int helper_argc = 0;
1779         int dump_count = 0;
1780         static atomic_t core_dump_count = ATOMIC_INIT(0);
1781         struct coredump_params cprm = {
1782                 .signr = signr,
1783                 .regs = regs,
1784                 .limit = current->signal->rlim[RLIMIT_CORE].rlim_cur,
1785         };
1786
1787         audit_core_dumps(signr);
1788
1789         binfmt = mm->binfmt;
1790         if (!binfmt || !binfmt->core_dump)
1791                 goto fail;
1792
1793         cred = prepare_creds();
1794         if (!cred) {
1795                 retval = -ENOMEM;
1796                 goto fail;
1797         }
1798
1799         down_write(&mm->mmap_sem);
1800         /*
1801          * If another thread got here first, or we are not dumpable, bail out.
1802          */
1803         if (mm->core_state || !get_dumpable(mm)) {
1804                 up_write(&mm->mmap_sem);
1805                 put_cred(cred);
1806                 goto fail;
1807         }
1808
1809         /*
1810          *      We cannot trust fsuid as being the "true" uid of the
1811          *      process nor do we know its entire history. We only know it
1812          *      was tainted so we dump it as root in mode 2.
1813          */
1814         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1815                 flag = O_EXCL;          /* Stop rewrite attacks */
1816                 cred->fsuid = 0;        /* Dump root private */
1817         }
1818
1819         retval = coredump_wait(exit_code, &core_state);
1820         if (retval < 0) {
1821                 put_cred(cred);
1822                 goto fail;
1823         }
1824
1825         old_cred = override_creds(cred);
1826
1827         /*
1828          * Clear any false indication of pending signals that might
1829          * be seen by the filesystem code called to write the core file.
1830          */
1831         clear_thread_flag(TIF_SIGPENDING);
1832
1833         /*
1834          * lock_kernel() because format_corename() is controlled by sysctl, which
1835          * uses lock_kernel()
1836          */
1837         lock_kernel();
1838         ispipe = format_corename(corename, signr);
1839         unlock_kernel();
1840
1841         if ((!ispipe) && (cprm.limit < binfmt->min_coredump))
1842                 goto fail_unlock;
1843
1844         if (ispipe) {
1845                 if (cprm.limit == 0) {
1846                         /*
1847                          * Normally core limits are irrelevant to pipes, since
1848                          * we're not writing to the file system, but we use
1849                          * cprm.limit of 0 here as a speacial value. Any
1850                          * non-zero limit gets set to RLIM_INFINITY below, but
1851                          * a limit of 0 skips the dump.  This is a consistent
1852                          * way to catch recursive crashes.  We can still crash
1853                          * if the core_pattern binary sets RLIM_CORE =  !0
1854                          * but it runs as root, and can do lots of stupid things
1855                          * Note that we use task_tgid_vnr here to grab the pid
1856                          * of the process group leader.  That way we get the
1857                          * right pid if a thread in a multi-threaded
1858                          * core_pattern process dies.
1859                          */
1860                         printk(KERN_WARNING
1861                                 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1862                                 task_tgid_vnr(current), current->comm);
1863                         printk(KERN_WARNING "Aborting core\n");
1864                         goto fail_unlock;
1865                 }
1866
1867                 dump_count = atomic_inc_return(&core_dump_count);
1868                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1869                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1870                                task_tgid_vnr(current), current->comm);
1871                         printk(KERN_WARNING "Skipping core dump\n");
1872                         goto fail_dropcount;
1873                 }
1874
1875                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1876                 if (!helper_argv) {
1877                         printk(KERN_WARNING "%s failed to allocate memory\n",
1878                                __func__);
1879                         goto fail_dropcount;
1880                 }
1881
1882                 cprm.limit = RLIM_INFINITY;
1883
1884                 /* SIGPIPE can happen, but it's just never processed */
1885                 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1886                                 &cprm.file)) {
1887                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1888                                corename);
1889                         goto fail_dropcount;
1890                 }
1891         } else
1892                 cprm.file = filp_open(corename,
1893                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1894                                  0600);
1895         if (IS_ERR(cprm.file))
1896                 goto fail_dropcount;
1897         inode = cprm.file->f_path.dentry->d_inode;
1898         if (inode->i_nlink > 1)
1899                 goto close_fail;        /* multiple links - don't dump */
1900         if (!ispipe && d_unhashed(cprm.file->f_path.dentry))
1901                 goto close_fail;
1902
1903         /* AK: actually i see no reason to not allow this for named pipes etc.,
1904            but keep the previous behaviour for now. */
1905         if (!ispipe && !S_ISREG(inode->i_mode))
1906                 goto close_fail;
1907         /*
1908          * Dont allow local users get cute and trick others to coredump
1909          * into their pre-created files:
1910          */
1911         if (inode->i_uid != current_fsuid())
1912                 goto close_fail;
1913         if (!cprm.file->f_op)
1914                 goto close_fail;
1915         if (!cprm.file->f_op->write)
1916                 goto close_fail;
1917         if (!ispipe &&
1918             do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file) != 0)
1919                 goto close_fail;
1920
1921         retval = binfmt->core_dump(&cprm);
1922
1923         if (retval)
1924                 current->signal->group_exit_code |= 0x80;
1925 close_fail:
1926         if (ispipe && core_pipe_limit)
1927                 wait_for_dump_helpers(cprm.file);
1928         filp_close(cprm.file, NULL);
1929 fail_dropcount:
1930         if (dump_count)
1931                 atomic_dec(&core_dump_count);
1932 fail_unlock:
1933         if (helper_argv)
1934                 argv_free(helper_argv);
1935
1936         revert_creds(old_cred);
1937         put_cred(cred);
1938         coredump_finish(mm);
1939 fail:
1940         return;
1941 }