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