tracepoints: add DECLARE_TRACE() and DEFINE_TRACE()
[linux-2.6.git] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/mnt_namespace.h>
21 #include <linux/personality.h>
22 #include <linux/mempolicy.h>
23 #include <linux/sem.h>
24 #include <linux/file.h>
25 #include <linux/fdtable.h>
26 #include <linux/iocontext.h>
27 #include <linux/key.h>
28 #include <linux/binfmts.h>
29 #include <linux/mman.h>
30 #include <linux/mmu_notifier.h>
31 #include <linux/fs.h>
32 #include <linux/nsproxy.h>
33 #include <linux/capability.h>
34 #include <linux/cpu.h>
35 #include <linux/cgroup.h>
36 #include <linux/security.h>
37 #include <linux/hugetlb.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/tracehook.h>
42 #include <linux/futex.h>
43 #include <linux/task_io_accounting_ops.h>
44 #include <linux/rcupdate.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/audit.h>
48 #include <linux/memcontrol.h>
49 #include <linux/profile.h>
50 #include <linux/rmap.h>
51 #include <linux/acct.h>
52 #include <linux/tsacct_kern.h>
53 #include <linux/cn_proc.h>
54 #include <linux/freezer.h>
55 #include <linux/delayacct.h>
56 #include <linux/taskstats_kern.h>
57 #include <linux/random.h>
58 #include <linux/tty.h>
59 #include <linux/proc_fs.h>
60 #include <linux/blkdev.h>
61 #include <trace/sched.h>
62
63 #include <asm/pgtable.h>
64 #include <asm/pgalloc.h>
65 #include <asm/uaccess.h>
66 #include <asm/mmu_context.h>
67 #include <asm/cacheflush.h>
68 #include <asm/tlbflush.h>
69
70 /*
71  * Protected counters by write_lock_irq(&tasklist_lock)
72  */
73 unsigned long total_forks;      /* Handle normal Linux uptimes. */
74 int nr_threads;                 /* The idle threads do not count.. */
75
76 int max_threads;                /* tunable limit on nr_threads */
77
78 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
79
80 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
81
82 DEFINE_TRACE(sched_process_fork);
83
84 int nr_processes(void)
85 {
86         int cpu;
87         int total = 0;
88
89         for_each_online_cpu(cpu)
90                 total += per_cpu(process_counts, cpu);
91
92         return total;
93 }
94
95 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
96 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
97 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
98 static struct kmem_cache *task_struct_cachep;
99 #endif
100
101 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
102 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
103 {
104 #ifdef CONFIG_DEBUG_STACK_USAGE
105         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
106 #else
107         gfp_t mask = GFP_KERNEL;
108 #endif
109         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
110 }
111
112 static inline void free_thread_info(struct thread_info *ti)
113 {
114         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
115 }
116 #endif
117
118 /* SLAB cache for signal_struct structures (tsk->signal) */
119 static struct kmem_cache *signal_cachep;
120
121 /* SLAB cache for sighand_struct structures (tsk->sighand) */
122 struct kmem_cache *sighand_cachep;
123
124 /* SLAB cache for files_struct structures (tsk->files) */
125 struct kmem_cache *files_cachep;
126
127 /* SLAB cache for fs_struct structures (tsk->fs) */
128 struct kmem_cache *fs_cachep;
129
130 /* SLAB cache for vm_area_struct structures */
131 struct kmem_cache *vm_area_cachep;
132
133 /* SLAB cache for mm_struct structures (tsk->mm) */
134 static struct kmem_cache *mm_cachep;
135
136 void free_task(struct task_struct *tsk)
137 {
138         prop_local_destroy_single(&tsk->dirties);
139         free_thread_info(tsk->stack);
140         rt_mutex_debug_task_free(tsk);
141         free_task_struct(tsk);
142 }
143 EXPORT_SYMBOL(free_task);
144
145 void __put_task_struct(struct task_struct *tsk)
146 {
147         WARN_ON(!tsk->exit_state);
148         WARN_ON(atomic_read(&tsk->usage));
149         WARN_ON(tsk == current);
150
151         security_task_free(tsk);
152         free_uid(tsk->user);
153         put_group_info(tsk->group_info);
154         delayacct_tsk_free(tsk);
155
156         if (!profile_handoff_task(tsk))
157                 free_task(tsk);
158 }
159
160 /*
161  * macro override instead of weak attribute alias, to workaround
162  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
163  */
164 #ifndef arch_task_cache_init
165 #define arch_task_cache_init()
166 #endif
167
168 void __init fork_init(unsigned long mempages)
169 {
170 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
171 #ifndef ARCH_MIN_TASKALIGN
172 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
173 #endif
174         /* create a slab on which task_structs can be allocated */
175         task_struct_cachep =
176                 kmem_cache_create("task_struct", sizeof(struct task_struct),
177                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
178 #endif
179
180         /* do the arch specific task caches init */
181         arch_task_cache_init();
182
183         /*
184          * The default maximum number of threads is set to a safe
185          * value: the thread structures can take up at most half
186          * of memory.
187          */
188         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
189
190         /*
191          * we need to allow at least 20 threads to boot a system
192          */
193         if(max_threads < 20)
194                 max_threads = 20;
195
196         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
197         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
198         init_task.signal->rlim[RLIMIT_SIGPENDING] =
199                 init_task.signal->rlim[RLIMIT_NPROC];
200 }
201
202 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
203                                                struct task_struct *src)
204 {
205         *dst = *src;
206         return 0;
207 }
208
209 static struct task_struct *dup_task_struct(struct task_struct *orig)
210 {
211         struct task_struct *tsk;
212         struct thread_info *ti;
213         int err;
214
215         prepare_to_copy(orig);
216
217         tsk = alloc_task_struct();
218         if (!tsk)
219                 return NULL;
220
221         ti = alloc_thread_info(tsk);
222         if (!ti) {
223                 free_task_struct(tsk);
224                 return NULL;
225         }
226
227         err = arch_dup_task_struct(tsk, orig);
228         if (err)
229                 goto out;
230
231         tsk->stack = ti;
232
233         err = prop_local_init_single(&tsk->dirties);
234         if (err)
235                 goto out;
236
237         setup_thread_stack(tsk, orig);
238
239 #ifdef CONFIG_CC_STACKPROTECTOR
240         tsk->stack_canary = get_random_int();
241 #endif
242
243         /* One for us, one for whoever does the "release_task()" (usually parent) */
244         atomic_set(&tsk->usage,2);
245         atomic_set(&tsk->fs_excl, 0);
246 #ifdef CONFIG_BLK_DEV_IO_TRACE
247         tsk->btrace_seq = 0;
248 #endif
249         tsk->splice_pipe = NULL;
250         return tsk;
251
252 out:
253         free_thread_info(ti);
254         free_task_struct(tsk);
255         return NULL;
256 }
257
258 #ifdef CONFIG_MMU
259 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
260 {
261         struct vm_area_struct *mpnt, *tmp, **pprev;
262         struct rb_node **rb_link, *rb_parent;
263         int retval;
264         unsigned long charge;
265         struct mempolicy *pol;
266
267         down_write(&oldmm->mmap_sem);
268         flush_cache_dup_mm(oldmm);
269         /*
270          * Not linked in yet - no deadlock potential:
271          */
272         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
273
274         mm->locked_vm = 0;
275         mm->mmap = NULL;
276         mm->mmap_cache = NULL;
277         mm->free_area_cache = oldmm->mmap_base;
278         mm->cached_hole_size = ~0UL;
279         mm->map_count = 0;
280         cpus_clear(mm->cpu_vm_mask);
281         mm->mm_rb = RB_ROOT;
282         rb_link = &mm->mm_rb.rb_node;
283         rb_parent = NULL;
284         pprev = &mm->mmap;
285
286         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
287                 struct file *file;
288
289                 if (mpnt->vm_flags & VM_DONTCOPY) {
290                         long pages = vma_pages(mpnt);
291                         mm->total_vm -= pages;
292                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
293                                                                 -pages);
294                         continue;
295                 }
296                 charge = 0;
297                 if (mpnt->vm_flags & VM_ACCOUNT) {
298                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
299                         if (security_vm_enough_memory(len))
300                                 goto fail_nomem;
301                         charge = len;
302                 }
303                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
304                 if (!tmp)
305                         goto fail_nomem;
306                 *tmp = *mpnt;
307                 pol = mpol_dup(vma_policy(mpnt));
308                 retval = PTR_ERR(pol);
309                 if (IS_ERR(pol))
310                         goto fail_nomem_policy;
311                 vma_set_policy(tmp, pol);
312                 tmp->vm_flags &= ~VM_LOCKED;
313                 tmp->vm_mm = mm;
314                 tmp->vm_next = NULL;
315                 anon_vma_link(tmp);
316                 file = tmp->vm_file;
317                 if (file) {
318                         struct inode *inode = file->f_path.dentry->d_inode;
319                         get_file(file);
320                         if (tmp->vm_flags & VM_DENYWRITE)
321                                 atomic_dec(&inode->i_writecount);
322
323                         /* insert tmp into the share list, just after mpnt */
324                         spin_lock(&file->f_mapping->i_mmap_lock);
325                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
326                         flush_dcache_mmap_lock(file->f_mapping);
327                         vma_prio_tree_add(tmp, mpnt);
328                         flush_dcache_mmap_unlock(file->f_mapping);
329                         spin_unlock(&file->f_mapping->i_mmap_lock);
330                 }
331
332                 /*
333                  * Clear hugetlb-related page reserves for children. This only
334                  * affects MAP_PRIVATE mappings. Faults generated by the child
335                  * are not guaranteed to succeed, even if read-only
336                  */
337                 if (is_vm_hugetlb_page(tmp))
338                         reset_vma_resv_huge_pages(tmp);
339
340                 /*
341                  * Link in the new vma and copy the page table entries.
342                  */
343                 *pprev = tmp;
344                 pprev = &tmp->vm_next;
345
346                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
347                 rb_link = &tmp->vm_rb.rb_right;
348                 rb_parent = &tmp->vm_rb;
349
350                 mm->map_count++;
351                 retval = copy_page_range(mm, oldmm, mpnt);
352
353                 if (tmp->vm_ops && tmp->vm_ops->open)
354                         tmp->vm_ops->open(tmp);
355
356                 if (retval)
357                         goto out;
358         }
359         /* a new mm has just been created */
360         arch_dup_mmap(oldmm, mm);
361         retval = 0;
362 out:
363         up_write(&mm->mmap_sem);
364         flush_tlb_mm(oldmm);
365         up_write(&oldmm->mmap_sem);
366         return retval;
367 fail_nomem_policy:
368         kmem_cache_free(vm_area_cachep, tmp);
369 fail_nomem:
370         retval = -ENOMEM;
371         vm_unacct_memory(charge);
372         goto out;
373 }
374
375 static inline int mm_alloc_pgd(struct mm_struct * mm)
376 {
377         mm->pgd = pgd_alloc(mm);
378         if (unlikely(!mm->pgd))
379                 return -ENOMEM;
380         return 0;
381 }
382
383 static inline void mm_free_pgd(struct mm_struct * mm)
384 {
385         pgd_free(mm, mm->pgd);
386 }
387 #else
388 #define dup_mmap(mm, oldmm)     (0)
389 #define mm_alloc_pgd(mm)        (0)
390 #define mm_free_pgd(mm)
391 #endif /* CONFIG_MMU */
392
393 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
394
395 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
396 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
397
398 #include <linux/init_task.h>
399
400 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
401 {
402         atomic_set(&mm->mm_users, 1);
403         atomic_set(&mm->mm_count, 1);
404         init_rwsem(&mm->mmap_sem);
405         INIT_LIST_HEAD(&mm->mmlist);
406         mm->flags = (current->mm) ? current->mm->flags
407                                   : MMF_DUMP_FILTER_DEFAULT;
408         mm->core_state = NULL;
409         mm->nr_ptes = 0;
410         set_mm_counter(mm, file_rss, 0);
411         set_mm_counter(mm, anon_rss, 0);
412         spin_lock_init(&mm->page_table_lock);
413         rwlock_init(&mm->ioctx_list_lock);
414         mm->ioctx_list = NULL;
415         mm->free_area_cache = TASK_UNMAPPED_BASE;
416         mm->cached_hole_size = ~0UL;
417         mm_init_owner(mm, p);
418
419         if (likely(!mm_alloc_pgd(mm))) {
420                 mm->def_flags = 0;
421                 mmu_notifier_mm_init(mm);
422                 return mm;
423         }
424
425         free_mm(mm);
426         return NULL;
427 }
428
429 /*
430  * Allocate and initialize an mm_struct.
431  */
432 struct mm_struct * mm_alloc(void)
433 {
434         struct mm_struct * mm;
435
436         mm = allocate_mm();
437         if (mm) {
438                 memset(mm, 0, sizeof(*mm));
439                 mm = mm_init(mm, current);
440         }
441         return mm;
442 }
443
444 /*
445  * Called when the last reference to the mm
446  * is dropped: either by a lazy thread or by
447  * mmput. Free the page directory and the mm.
448  */
449 void __mmdrop(struct mm_struct *mm)
450 {
451         BUG_ON(mm == &init_mm);
452         mm_free_pgd(mm);
453         destroy_context(mm);
454         mmu_notifier_mm_destroy(mm);
455         free_mm(mm);
456 }
457 EXPORT_SYMBOL_GPL(__mmdrop);
458
459 /*
460  * Decrement the use count and release all resources for an mm.
461  */
462 void mmput(struct mm_struct *mm)
463 {
464         might_sleep();
465
466         if (atomic_dec_and_test(&mm->mm_users)) {
467                 exit_aio(mm);
468                 exit_mmap(mm);
469                 set_mm_exe_file(mm, NULL);
470                 if (!list_empty(&mm->mmlist)) {
471                         spin_lock(&mmlist_lock);
472                         list_del(&mm->mmlist);
473                         spin_unlock(&mmlist_lock);
474                 }
475                 put_swap_token(mm);
476                 mmdrop(mm);
477         }
478 }
479 EXPORT_SYMBOL_GPL(mmput);
480
481 /**
482  * get_task_mm - acquire a reference to the task's mm
483  *
484  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
485  * this kernel workthread has transiently adopted a user mm with use_mm,
486  * to do its AIO) is not set and if so returns a reference to it, after
487  * bumping up the use count.  User must release the mm via mmput()
488  * after use.  Typically used by /proc and ptrace.
489  */
490 struct mm_struct *get_task_mm(struct task_struct *task)
491 {
492         struct mm_struct *mm;
493
494         task_lock(task);
495         mm = task->mm;
496         if (mm) {
497                 if (task->flags & PF_KTHREAD)
498                         mm = NULL;
499                 else
500                         atomic_inc(&mm->mm_users);
501         }
502         task_unlock(task);
503         return mm;
504 }
505 EXPORT_SYMBOL_GPL(get_task_mm);
506
507 /* Please note the differences between mmput and mm_release.
508  * mmput is called whenever we stop holding onto a mm_struct,
509  * error success whatever.
510  *
511  * mm_release is called after a mm_struct has been removed
512  * from the current process.
513  *
514  * This difference is important for error handling, when we
515  * only half set up a mm_struct for a new process and need to restore
516  * the old one.  Because we mmput the new mm_struct before
517  * restoring the old one. . .
518  * Eric Biederman 10 January 1998
519  */
520 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
521 {
522         struct completion *vfork_done = tsk->vfork_done;
523
524         /* Get rid of any cached register state */
525         deactivate_mm(tsk, mm);
526
527         /* notify parent sleeping on vfork() */
528         if (vfork_done) {
529                 tsk->vfork_done = NULL;
530                 complete(vfork_done);
531         }
532
533         /*
534          * If we're exiting normally, clear a user-space tid field if
535          * requested.  We leave this alone when dying by signal, to leave
536          * the value intact in a core dump, and to save the unnecessary
537          * trouble otherwise.  Userland only wants this done for a sys_exit.
538          */
539         if (tsk->clear_child_tid
540             && !(tsk->flags & PF_SIGNALED)
541             && atomic_read(&mm->mm_users) > 1) {
542                 u32 __user * tidptr = tsk->clear_child_tid;
543                 tsk->clear_child_tid = NULL;
544
545                 /*
546                  * We don't check the error code - if userspace has
547                  * not set up a proper pointer then tough luck.
548                  */
549                 put_user(0, tidptr);
550                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
551         }
552 }
553
554 /*
555  * Allocate a new mm structure and copy contents from the
556  * mm structure of the passed in task structure.
557  */
558 struct mm_struct *dup_mm(struct task_struct *tsk)
559 {
560         struct mm_struct *mm, *oldmm = current->mm;
561         int err;
562
563         if (!oldmm)
564                 return NULL;
565
566         mm = allocate_mm();
567         if (!mm)
568                 goto fail_nomem;
569
570         memcpy(mm, oldmm, sizeof(*mm));
571
572         /* Initializing for Swap token stuff */
573         mm->token_priority = 0;
574         mm->last_interval = 0;
575
576         if (!mm_init(mm, tsk))
577                 goto fail_nomem;
578
579         if (init_new_context(tsk, mm))
580                 goto fail_nocontext;
581
582         dup_mm_exe_file(oldmm, mm);
583
584         err = dup_mmap(mm, oldmm);
585         if (err)
586                 goto free_pt;
587
588         mm->hiwater_rss = get_mm_rss(mm);
589         mm->hiwater_vm = mm->total_vm;
590
591         return mm;
592
593 free_pt:
594         mmput(mm);
595
596 fail_nomem:
597         return NULL;
598
599 fail_nocontext:
600         /*
601          * If init_new_context() failed, we cannot use mmput() to free the mm
602          * because it calls destroy_context()
603          */
604         mm_free_pgd(mm);
605         free_mm(mm);
606         return NULL;
607 }
608
609 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
610 {
611         struct mm_struct * mm, *oldmm;
612         int retval;
613
614         tsk->min_flt = tsk->maj_flt = 0;
615         tsk->nvcsw = tsk->nivcsw = 0;
616
617         tsk->mm = NULL;
618         tsk->active_mm = NULL;
619
620         /*
621          * Are we cloning a kernel thread?
622          *
623          * We need to steal a active VM for that..
624          */
625         oldmm = current->mm;
626         if (!oldmm)
627                 return 0;
628
629         if (clone_flags & CLONE_VM) {
630                 atomic_inc(&oldmm->mm_users);
631                 mm = oldmm;
632                 goto good_mm;
633         }
634
635         retval = -ENOMEM;
636         mm = dup_mm(tsk);
637         if (!mm)
638                 goto fail_nomem;
639
640 good_mm:
641         /* Initializing for Swap token stuff */
642         mm->token_priority = 0;
643         mm->last_interval = 0;
644
645         tsk->mm = mm;
646         tsk->active_mm = mm;
647         return 0;
648
649 fail_nomem:
650         return retval;
651 }
652
653 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
654 {
655         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
656         /* We don't need to lock fs - think why ;-) */
657         if (fs) {
658                 atomic_set(&fs->count, 1);
659                 rwlock_init(&fs->lock);
660                 fs->umask = old->umask;
661                 read_lock(&old->lock);
662                 fs->root = old->root;
663                 path_get(&old->root);
664                 fs->pwd = old->pwd;
665                 path_get(&old->pwd);
666                 read_unlock(&old->lock);
667         }
668         return fs;
669 }
670
671 struct fs_struct *copy_fs_struct(struct fs_struct *old)
672 {
673         return __copy_fs_struct(old);
674 }
675
676 EXPORT_SYMBOL_GPL(copy_fs_struct);
677
678 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
679 {
680         if (clone_flags & CLONE_FS) {
681                 atomic_inc(&current->fs->count);
682                 return 0;
683         }
684         tsk->fs = __copy_fs_struct(current->fs);
685         if (!tsk->fs)
686                 return -ENOMEM;
687         return 0;
688 }
689
690 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
691 {
692         struct files_struct *oldf, *newf;
693         int error = 0;
694
695         /*
696          * A background process may not have any files ...
697          */
698         oldf = current->files;
699         if (!oldf)
700                 goto out;
701
702         if (clone_flags & CLONE_FILES) {
703                 atomic_inc(&oldf->count);
704                 goto out;
705         }
706
707         newf = dup_fd(oldf, &error);
708         if (!newf)
709                 goto out;
710
711         tsk->files = newf;
712         error = 0;
713 out:
714         return error;
715 }
716
717 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
718 {
719 #ifdef CONFIG_BLOCK
720         struct io_context *ioc = current->io_context;
721
722         if (!ioc)
723                 return 0;
724         /*
725          * Share io context with parent, if CLONE_IO is set
726          */
727         if (clone_flags & CLONE_IO) {
728                 tsk->io_context = ioc_task_link(ioc);
729                 if (unlikely(!tsk->io_context))
730                         return -ENOMEM;
731         } else if (ioprio_valid(ioc->ioprio)) {
732                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
733                 if (unlikely(!tsk->io_context))
734                         return -ENOMEM;
735
736                 tsk->io_context->ioprio = ioc->ioprio;
737         }
738 #endif
739         return 0;
740 }
741
742 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
743 {
744         struct sighand_struct *sig;
745
746         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
747                 atomic_inc(&current->sighand->count);
748                 return 0;
749         }
750         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
751         rcu_assign_pointer(tsk->sighand, sig);
752         if (!sig)
753                 return -ENOMEM;
754         atomic_set(&sig->count, 1);
755         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
756         return 0;
757 }
758
759 void __cleanup_sighand(struct sighand_struct *sighand)
760 {
761         if (atomic_dec_and_test(&sighand->count))
762                 kmem_cache_free(sighand_cachep, sighand);
763 }
764
765
766 /*
767  * Initialize POSIX timer handling for a thread group.
768  */
769 static void posix_cpu_timers_init_group(struct signal_struct *sig)
770 {
771         /* Thread group counters. */
772         thread_group_cputime_init(sig);
773
774         /* Expiration times and increments. */
775         sig->it_virt_expires = cputime_zero;
776         sig->it_virt_incr = cputime_zero;
777         sig->it_prof_expires = cputime_zero;
778         sig->it_prof_incr = cputime_zero;
779
780         /* Cached expiration times. */
781         sig->cputime_expires.prof_exp = cputime_zero;
782         sig->cputime_expires.virt_exp = cputime_zero;
783         sig->cputime_expires.sched_exp = 0;
784
785         /* The timer lists. */
786         INIT_LIST_HEAD(&sig->cpu_timers[0]);
787         INIT_LIST_HEAD(&sig->cpu_timers[1]);
788         INIT_LIST_HEAD(&sig->cpu_timers[2]);
789 }
790
791 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
792 {
793         struct signal_struct *sig;
794         int ret;
795
796         if (clone_flags & CLONE_THREAD) {
797                 ret = thread_group_cputime_clone_thread(current);
798                 if (likely(!ret)) {
799                         atomic_inc(&current->signal->count);
800                         atomic_inc(&current->signal->live);
801                 }
802                 return ret;
803         }
804         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
805         tsk->signal = sig;
806         if (!sig)
807                 return -ENOMEM;
808
809         ret = copy_thread_group_keys(tsk);
810         if (ret < 0) {
811                 kmem_cache_free(signal_cachep, sig);
812                 return ret;
813         }
814
815         atomic_set(&sig->count, 1);
816         atomic_set(&sig->live, 1);
817         init_waitqueue_head(&sig->wait_chldexit);
818         sig->flags = 0;
819         sig->group_exit_code = 0;
820         sig->group_exit_task = NULL;
821         sig->group_stop_count = 0;
822         sig->curr_target = tsk;
823         init_sigpending(&sig->shared_pending);
824         INIT_LIST_HEAD(&sig->posix_timers);
825
826         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
827         sig->it_real_incr.tv64 = 0;
828         sig->real_timer.function = it_real_fn;
829
830         sig->leader = 0;        /* session leadership doesn't inherit */
831         sig->tty_old_pgrp = NULL;
832         sig->tty = NULL;
833
834         sig->cutime = sig->cstime = cputime_zero;
835         sig->gtime = cputime_zero;
836         sig->cgtime = cputime_zero;
837         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
838         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
839         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
840         task_io_accounting_init(&sig->ioac);
841         taskstats_tgid_init(sig);
842
843         task_lock(current->group_leader);
844         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
845         task_unlock(current->group_leader);
846
847         posix_cpu_timers_init_group(sig);
848
849         acct_init_pacct(&sig->pacct);
850
851         tty_audit_fork(sig);
852
853         return 0;
854 }
855
856 void __cleanup_signal(struct signal_struct *sig)
857 {
858         thread_group_cputime_free(sig);
859         exit_thread_group_keys(sig);
860         tty_kref_put(sig->tty);
861         kmem_cache_free(signal_cachep, sig);
862 }
863
864 static void cleanup_signal(struct task_struct *tsk)
865 {
866         struct signal_struct *sig = tsk->signal;
867
868         atomic_dec(&sig->live);
869
870         if (atomic_dec_and_test(&sig->count))
871                 __cleanup_signal(sig);
872 }
873
874 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
875 {
876         unsigned long new_flags = p->flags;
877
878         new_flags &= ~PF_SUPERPRIV;
879         new_flags |= PF_FORKNOEXEC;
880         new_flags |= PF_STARTING;
881         p->flags = new_flags;
882         clear_freeze_flag(p);
883 }
884
885 asmlinkage long sys_set_tid_address(int __user *tidptr)
886 {
887         current->clear_child_tid = tidptr;
888
889         return task_pid_vnr(current);
890 }
891
892 static void rt_mutex_init_task(struct task_struct *p)
893 {
894         spin_lock_init(&p->pi_lock);
895 #ifdef CONFIG_RT_MUTEXES
896         plist_head_init(&p->pi_waiters, &p->pi_lock);
897         p->pi_blocked_on = NULL;
898 #endif
899 }
900
901 #ifdef CONFIG_MM_OWNER
902 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
903 {
904         mm->owner = p;
905 }
906 #endif /* CONFIG_MM_OWNER */
907
908 /*
909  * Initialize POSIX timer handling for a single task.
910  */
911 static void posix_cpu_timers_init(struct task_struct *tsk)
912 {
913         tsk->cputime_expires.prof_exp = cputime_zero;
914         tsk->cputime_expires.virt_exp = cputime_zero;
915         tsk->cputime_expires.sched_exp = 0;
916         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
917         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
918         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
919 }
920
921 /*
922  * This creates a new process as a copy of the old one,
923  * but does not actually start it yet.
924  *
925  * It copies the registers, and all the appropriate
926  * parts of the process environment (as per the clone
927  * flags). The actual kick-off is left to the caller.
928  */
929 static struct task_struct *copy_process(unsigned long clone_flags,
930                                         unsigned long stack_start,
931                                         struct pt_regs *regs,
932                                         unsigned long stack_size,
933                                         int __user *child_tidptr,
934                                         struct pid *pid,
935                                         int trace)
936 {
937         int retval;
938         struct task_struct *p;
939         int cgroup_callbacks_done = 0;
940
941         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
942                 return ERR_PTR(-EINVAL);
943
944         /*
945          * Thread groups must share signals as well, and detached threads
946          * can only be started up within the thread group.
947          */
948         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
949                 return ERR_PTR(-EINVAL);
950
951         /*
952          * Shared signal handlers imply shared VM. By way of the above,
953          * thread groups also imply shared VM. Blocking this case allows
954          * for various simplifications in other code.
955          */
956         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
957                 return ERR_PTR(-EINVAL);
958
959         retval = security_task_create(clone_flags);
960         if (retval)
961                 goto fork_out;
962
963         retval = -ENOMEM;
964         p = dup_task_struct(current);
965         if (!p)
966                 goto fork_out;
967
968         rt_mutex_init_task(p);
969
970 #ifdef CONFIG_PROVE_LOCKING
971         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
972         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
973 #endif
974         retval = -EAGAIN;
975         if (atomic_read(&p->user->processes) >=
976                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
977                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
978                     p->user != current->nsproxy->user_ns->root_user)
979                         goto bad_fork_free;
980         }
981
982         atomic_inc(&p->user->__count);
983         atomic_inc(&p->user->processes);
984         get_group_info(p->group_info);
985
986         /*
987          * If multiple threads are within copy_process(), then this check
988          * triggers too late. This doesn't hurt, the check is only there
989          * to stop root fork bombs.
990          */
991         if (nr_threads >= max_threads)
992                 goto bad_fork_cleanup_count;
993
994         if (!try_module_get(task_thread_info(p)->exec_domain->module))
995                 goto bad_fork_cleanup_count;
996
997         if (p->binfmt && !try_module_get(p->binfmt->module))
998                 goto bad_fork_cleanup_put_domain;
999
1000         p->did_exec = 0;
1001         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1002         copy_flags(clone_flags, p);
1003         INIT_LIST_HEAD(&p->children);
1004         INIT_LIST_HEAD(&p->sibling);
1005 #ifdef CONFIG_PREEMPT_RCU
1006         p->rcu_read_lock_nesting = 0;
1007         p->rcu_flipctr_idx = 0;
1008 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1009         p->vfork_done = NULL;
1010         spin_lock_init(&p->alloc_lock);
1011
1012         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1013         init_sigpending(&p->pending);
1014
1015         p->utime = cputime_zero;
1016         p->stime = cputime_zero;
1017         p->gtime = cputime_zero;
1018         p->utimescaled = cputime_zero;
1019         p->stimescaled = cputime_zero;
1020         p->prev_utime = cputime_zero;
1021         p->prev_stime = cputime_zero;
1022
1023         p->default_timer_slack_ns = current->timer_slack_ns;
1024
1025 #ifdef CONFIG_DETECT_SOFTLOCKUP
1026         p->last_switch_count = 0;
1027         p->last_switch_timestamp = 0;
1028 #endif
1029
1030         task_io_accounting_init(&p->ioac);
1031         acct_clear_integrals(p);
1032
1033         posix_cpu_timers_init(p);
1034
1035         p->lock_depth = -1;             /* -1 = no lock */
1036         do_posix_clock_monotonic_gettime(&p->start_time);
1037         p->real_start_time = p->start_time;
1038         monotonic_to_bootbased(&p->real_start_time);
1039 #ifdef CONFIG_SECURITY
1040         p->security = NULL;
1041 #endif
1042         p->cap_bset = current->cap_bset;
1043         p->io_context = NULL;
1044         p->audit_context = NULL;
1045         cgroup_fork(p);
1046 #ifdef CONFIG_NUMA
1047         p->mempolicy = mpol_dup(p->mempolicy);
1048         if (IS_ERR(p->mempolicy)) {
1049                 retval = PTR_ERR(p->mempolicy);
1050                 p->mempolicy = NULL;
1051                 goto bad_fork_cleanup_cgroup;
1052         }
1053         mpol_fix_fork_child_flag(p);
1054 #endif
1055 #ifdef CONFIG_TRACE_IRQFLAGS
1056         p->irq_events = 0;
1057 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1058         p->hardirqs_enabled = 1;
1059 #else
1060         p->hardirqs_enabled = 0;
1061 #endif
1062         p->hardirq_enable_ip = 0;
1063         p->hardirq_enable_event = 0;
1064         p->hardirq_disable_ip = _THIS_IP_;
1065         p->hardirq_disable_event = 0;
1066         p->softirqs_enabled = 1;
1067         p->softirq_enable_ip = _THIS_IP_;
1068         p->softirq_enable_event = 0;
1069         p->softirq_disable_ip = 0;
1070         p->softirq_disable_event = 0;
1071         p->hardirq_context = 0;
1072         p->softirq_context = 0;
1073 #endif
1074 #ifdef CONFIG_LOCKDEP
1075         p->lockdep_depth = 0; /* no locks held yet */
1076         p->curr_chain_key = 0;
1077         p->lockdep_recursion = 0;
1078 #endif
1079
1080 #ifdef CONFIG_DEBUG_MUTEXES
1081         p->blocked_on = NULL; /* not blocked yet */
1082 #endif
1083
1084         /* Perform scheduler related setup. Assign this task to a CPU. */
1085         sched_fork(p, clone_flags);
1086
1087         if ((retval = security_task_alloc(p)))
1088                 goto bad_fork_cleanup_policy;
1089         if ((retval = audit_alloc(p)))
1090                 goto bad_fork_cleanup_security;
1091         /* copy all the process information */
1092         if ((retval = copy_semundo(clone_flags, p)))
1093                 goto bad_fork_cleanup_audit;
1094         if ((retval = copy_files(clone_flags, p)))
1095                 goto bad_fork_cleanup_semundo;
1096         if ((retval = copy_fs(clone_flags, p)))
1097                 goto bad_fork_cleanup_files;
1098         if ((retval = copy_sighand(clone_flags, p)))
1099                 goto bad_fork_cleanup_fs;
1100         if ((retval = copy_signal(clone_flags, p)))
1101                 goto bad_fork_cleanup_sighand;
1102         if ((retval = copy_mm(clone_flags, p)))
1103                 goto bad_fork_cleanup_signal;
1104         if ((retval = copy_keys(clone_flags, p)))
1105                 goto bad_fork_cleanup_mm;
1106         if ((retval = copy_namespaces(clone_flags, p)))
1107                 goto bad_fork_cleanup_keys;
1108         if ((retval = copy_io(clone_flags, p)))
1109                 goto bad_fork_cleanup_namespaces;
1110         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1111         if (retval)
1112                 goto bad_fork_cleanup_io;
1113
1114         if (pid != &init_struct_pid) {
1115                 retval = -ENOMEM;
1116                 pid = alloc_pid(task_active_pid_ns(p));
1117                 if (!pid)
1118                         goto bad_fork_cleanup_io;
1119
1120                 if (clone_flags & CLONE_NEWPID) {
1121                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1122                         if (retval < 0)
1123                                 goto bad_fork_free_pid;
1124                 }
1125         }
1126
1127         p->pid = pid_nr(pid);
1128         p->tgid = p->pid;
1129         if (clone_flags & CLONE_THREAD)
1130                 p->tgid = current->tgid;
1131
1132         if (current->nsproxy != p->nsproxy) {
1133                 retval = ns_cgroup_clone(p, pid);
1134                 if (retval)
1135                         goto bad_fork_free_pid;
1136         }
1137
1138         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1139         /*
1140          * Clear TID on mm_release()?
1141          */
1142         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1143 #ifdef CONFIG_FUTEX
1144         p->robust_list = NULL;
1145 #ifdef CONFIG_COMPAT
1146         p->compat_robust_list = NULL;
1147 #endif
1148         INIT_LIST_HEAD(&p->pi_state_list);
1149         p->pi_state_cache = NULL;
1150 #endif
1151         /*
1152          * sigaltstack should be cleared when sharing the same VM
1153          */
1154         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1155                 p->sas_ss_sp = p->sas_ss_size = 0;
1156
1157         /*
1158          * Syscall tracing should be turned off in the child regardless
1159          * of CLONE_PTRACE.
1160          */
1161         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1162 #ifdef TIF_SYSCALL_EMU
1163         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1164 #endif
1165         clear_all_latency_tracing(p);
1166
1167         /* Our parent execution domain becomes current domain
1168            These must match for thread signalling to apply */
1169         p->parent_exec_id = p->self_exec_id;
1170
1171         /* ok, now we should be set up.. */
1172         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1173         p->pdeath_signal = 0;
1174         p->exit_state = 0;
1175
1176         /*
1177          * Ok, make it visible to the rest of the system.
1178          * We dont wake it up yet.
1179          */
1180         p->group_leader = p;
1181         INIT_LIST_HEAD(&p->thread_group);
1182
1183         /* Now that the task is set up, run cgroup callbacks if
1184          * necessary. We need to run them before the task is visible
1185          * on the tasklist. */
1186         cgroup_fork_callbacks(p);
1187         cgroup_callbacks_done = 1;
1188
1189         /* Need tasklist lock for parent etc handling! */
1190         write_lock_irq(&tasklist_lock);
1191
1192         /*
1193          * The task hasn't been attached yet, so its cpus_allowed mask will
1194          * not be changed, nor will its assigned CPU.
1195          *
1196          * The cpus_allowed mask of the parent may have changed after it was
1197          * copied first time - so re-copy it here, then check the child's CPU
1198          * to ensure it is on a valid CPU (and if not, just force it back to
1199          * parent's CPU). This avoids alot of nasty races.
1200          */
1201         p->cpus_allowed = current->cpus_allowed;
1202         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1203         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1204                         !cpu_online(task_cpu(p))))
1205                 set_task_cpu(p, smp_processor_id());
1206
1207         /* CLONE_PARENT re-uses the old parent */
1208         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1209                 p->real_parent = current->real_parent;
1210         else
1211                 p->real_parent = current;
1212
1213         spin_lock(&current->sighand->siglock);
1214
1215         /*
1216          * Process group and session signals need to be delivered to just the
1217          * parent before the fork or both the parent and the child after the
1218          * fork. Restart if a signal comes in before we add the new process to
1219          * it's process group.
1220          * A fatal signal pending means that current will exit, so the new
1221          * thread can't slip out of an OOM kill (or normal SIGKILL).
1222          */
1223         recalc_sigpending();
1224         if (signal_pending(current)) {
1225                 spin_unlock(&current->sighand->siglock);
1226                 write_unlock_irq(&tasklist_lock);
1227                 retval = -ERESTARTNOINTR;
1228                 goto bad_fork_free_pid;
1229         }
1230
1231         if (clone_flags & CLONE_THREAD) {
1232                 p->group_leader = current->group_leader;
1233                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1234         }
1235
1236         if (likely(p->pid)) {
1237                 list_add_tail(&p->sibling, &p->real_parent->children);
1238                 tracehook_finish_clone(p, clone_flags, trace);
1239
1240                 if (thread_group_leader(p)) {
1241                         if (clone_flags & CLONE_NEWPID)
1242                                 p->nsproxy->pid_ns->child_reaper = p;
1243
1244                         p->signal->leader_pid = pid;
1245                         tty_kref_put(p->signal->tty);
1246                         p->signal->tty = tty_kref_get(current->signal->tty);
1247                         set_task_pgrp(p, task_pgrp_nr(current));
1248                         set_task_session(p, task_session_nr(current));
1249                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1250                         attach_pid(p, PIDTYPE_SID, task_session(current));
1251                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1252                         __get_cpu_var(process_counts)++;
1253                 }
1254                 attach_pid(p, PIDTYPE_PID, pid);
1255                 nr_threads++;
1256         }
1257
1258         total_forks++;
1259         spin_unlock(&current->sighand->siglock);
1260         write_unlock_irq(&tasklist_lock);
1261         proc_fork_connector(p);
1262         cgroup_post_fork(p);
1263         return p;
1264
1265 bad_fork_free_pid:
1266         if (pid != &init_struct_pid)
1267                 free_pid(pid);
1268 bad_fork_cleanup_io:
1269         put_io_context(p->io_context);
1270 bad_fork_cleanup_namespaces:
1271         exit_task_namespaces(p);
1272 bad_fork_cleanup_keys:
1273         exit_keys(p);
1274 bad_fork_cleanup_mm:
1275         if (p->mm)
1276                 mmput(p->mm);
1277 bad_fork_cleanup_signal:
1278         cleanup_signal(p);
1279 bad_fork_cleanup_sighand:
1280         __cleanup_sighand(p->sighand);
1281 bad_fork_cleanup_fs:
1282         exit_fs(p); /* blocking */
1283 bad_fork_cleanup_files:
1284         exit_files(p); /* blocking */
1285 bad_fork_cleanup_semundo:
1286         exit_sem(p);
1287 bad_fork_cleanup_audit:
1288         audit_free(p);
1289 bad_fork_cleanup_security:
1290         security_task_free(p);
1291 bad_fork_cleanup_policy:
1292 #ifdef CONFIG_NUMA
1293         mpol_put(p->mempolicy);
1294 bad_fork_cleanup_cgroup:
1295 #endif
1296         cgroup_exit(p, cgroup_callbacks_done);
1297         delayacct_tsk_free(p);
1298         if (p->binfmt)
1299                 module_put(p->binfmt->module);
1300 bad_fork_cleanup_put_domain:
1301         module_put(task_thread_info(p)->exec_domain->module);
1302 bad_fork_cleanup_count:
1303         put_group_info(p->group_info);
1304         atomic_dec(&p->user->processes);
1305         free_uid(p->user);
1306 bad_fork_free:
1307         free_task(p);
1308 fork_out:
1309         return ERR_PTR(retval);
1310 }
1311
1312 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1313 {
1314         memset(regs, 0, sizeof(struct pt_regs));
1315         return regs;
1316 }
1317
1318 struct task_struct * __cpuinit fork_idle(int cpu)
1319 {
1320         struct task_struct *task;
1321         struct pt_regs regs;
1322
1323         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1324                             &init_struct_pid, 0);
1325         if (!IS_ERR(task))
1326                 init_idle(task, cpu);
1327
1328         return task;
1329 }
1330
1331 /*
1332  *  Ok, this is the main fork-routine.
1333  *
1334  * It copies the process, and if successful kick-starts
1335  * it and waits for it to finish using the VM if required.
1336  */
1337 long do_fork(unsigned long clone_flags,
1338               unsigned long stack_start,
1339               struct pt_regs *regs,
1340               unsigned long stack_size,
1341               int __user *parent_tidptr,
1342               int __user *child_tidptr)
1343 {
1344         struct task_struct *p;
1345         int trace = 0;
1346         long nr;
1347
1348         /*
1349          * We hope to recycle these flags after 2.6.26
1350          */
1351         if (unlikely(clone_flags & CLONE_STOPPED)) {
1352                 static int __read_mostly count = 100;
1353
1354                 if (count > 0 && printk_ratelimit()) {
1355                         char comm[TASK_COMM_LEN];
1356
1357                         count--;
1358                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1359                                         "clone flags 0x%lx\n",
1360                                 get_task_comm(comm, current),
1361                                 clone_flags & CLONE_STOPPED);
1362                 }
1363         }
1364
1365         /*
1366          * When called from kernel_thread, don't do user tracing stuff.
1367          */
1368         if (likely(user_mode(regs)))
1369                 trace = tracehook_prepare_clone(clone_flags);
1370
1371         p = copy_process(clone_flags, stack_start, regs, stack_size,
1372                          child_tidptr, NULL, trace);
1373         /*
1374          * Do this prior waking up the new thread - the thread pointer
1375          * might get invalid after that point, if the thread exits quickly.
1376          */
1377         if (!IS_ERR(p)) {
1378                 struct completion vfork;
1379
1380                 trace_sched_process_fork(current, p);
1381
1382                 nr = task_pid_vnr(p);
1383
1384                 if (clone_flags & CLONE_PARENT_SETTID)
1385                         put_user(nr, parent_tidptr);
1386
1387                 if (clone_flags & CLONE_VFORK) {
1388                         p->vfork_done = &vfork;
1389                         init_completion(&vfork);
1390                 }
1391
1392                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1393
1394                 /*
1395                  * We set PF_STARTING at creation in case tracing wants to
1396                  * use this to distinguish a fully live task from one that
1397                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1398                  * clear it and set the child going.
1399                  */
1400                 p->flags &= ~PF_STARTING;
1401
1402                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1403                         /*
1404                          * We'll start up with an immediate SIGSTOP.
1405                          */
1406                         sigaddset(&p->pending.signal, SIGSTOP);
1407                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1408                         __set_task_state(p, TASK_STOPPED);
1409                 } else {
1410                         wake_up_new_task(p, clone_flags);
1411                 }
1412
1413                 tracehook_report_clone_complete(trace, regs,
1414                                                 clone_flags, nr, p);
1415
1416                 if (clone_flags & CLONE_VFORK) {
1417                         freezer_do_not_count();
1418                         wait_for_completion(&vfork);
1419                         freezer_count();
1420                         tracehook_report_vfork_done(p, nr);
1421                 }
1422         } else {
1423                 nr = PTR_ERR(p);
1424         }
1425         return nr;
1426 }
1427
1428 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1429 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1430 #endif
1431
1432 static void sighand_ctor(void *data)
1433 {
1434         struct sighand_struct *sighand = data;
1435
1436         spin_lock_init(&sighand->siglock);
1437         init_waitqueue_head(&sighand->signalfd_wqh);
1438 }
1439
1440 void __init proc_caches_init(void)
1441 {
1442         sighand_cachep = kmem_cache_create("sighand_cache",
1443                         sizeof(struct sighand_struct), 0,
1444                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1445                         sighand_ctor);
1446         signal_cachep = kmem_cache_create("signal_cache",
1447                         sizeof(struct signal_struct), 0,
1448                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1449         files_cachep = kmem_cache_create("files_cache",
1450                         sizeof(struct files_struct), 0,
1451                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1452         fs_cachep = kmem_cache_create("fs_cache",
1453                         sizeof(struct fs_struct), 0,
1454                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1455         vm_area_cachep = kmem_cache_create("vm_area_struct",
1456                         sizeof(struct vm_area_struct), 0,
1457                         SLAB_PANIC, NULL);
1458         mm_cachep = kmem_cache_create("mm_struct",
1459                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1460                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1461 }
1462
1463 /*
1464  * Check constraints on flags passed to the unshare system call and
1465  * force unsharing of additional process context as appropriate.
1466  */
1467 static void check_unshare_flags(unsigned long *flags_ptr)
1468 {
1469         /*
1470          * If unsharing a thread from a thread group, must also
1471          * unshare vm.
1472          */
1473         if (*flags_ptr & CLONE_THREAD)
1474                 *flags_ptr |= CLONE_VM;
1475
1476         /*
1477          * If unsharing vm, must also unshare signal handlers.
1478          */
1479         if (*flags_ptr & CLONE_VM)
1480                 *flags_ptr |= CLONE_SIGHAND;
1481
1482         /*
1483          * If unsharing signal handlers and the task was created
1484          * using CLONE_THREAD, then must unshare the thread
1485          */
1486         if ((*flags_ptr & CLONE_SIGHAND) &&
1487             (atomic_read(&current->signal->count) > 1))
1488                 *flags_ptr |= CLONE_THREAD;
1489
1490         /*
1491          * If unsharing namespace, must also unshare filesystem information.
1492          */
1493         if (*flags_ptr & CLONE_NEWNS)
1494                 *flags_ptr |= CLONE_FS;
1495 }
1496
1497 /*
1498  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1499  */
1500 static int unshare_thread(unsigned long unshare_flags)
1501 {
1502         if (unshare_flags & CLONE_THREAD)
1503                 return -EINVAL;
1504
1505         return 0;
1506 }
1507
1508 /*
1509  * Unshare the filesystem structure if it is being shared
1510  */
1511 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1512 {
1513         struct fs_struct *fs = current->fs;
1514
1515         if ((unshare_flags & CLONE_FS) &&
1516             (fs && atomic_read(&fs->count) > 1)) {
1517                 *new_fsp = __copy_fs_struct(current->fs);
1518                 if (!*new_fsp)
1519                         return -ENOMEM;
1520         }
1521
1522         return 0;
1523 }
1524
1525 /*
1526  * Unsharing of sighand is not supported yet
1527  */
1528 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1529 {
1530         struct sighand_struct *sigh = current->sighand;
1531
1532         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1533                 return -EINVAL;
1534         else
1535                 return 0;
1536 }
1537
1538 /*
1539  * Unshare vm if it is being shared
1540  */
1541 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1542 {
1543         struct mm_struct *mm = current->mm;
1544
1545         if ((unshare_flags & CLONE_VM) &&
1546             (mm && atomic_read(&mm->mm_users) > 1)) {
1547                 return -EINVAL;
1548         }
1549
1550         return 0;
1551 }
1552
1553 /*
1554  * Unshare file descriptor table if it is being shared
1555  */
1556 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1557 {
1558         struct files_struct *fd = current->files;
1559         int error = 0;
1560
1561         if ((unshare_flags & CLONE_FILES) &&
1562             (fd && atomic_read(&fd->count) > 1)) {
1563                 *new_fdp = dup_fd(fd, &error);
1564                 if (!*new_fdp)
1565                         return error;
1566         }
1567
1568         return 0;
1569 }
1570
1571 /*
1572  * unshare allows a process to 'unshare' part of the process
1573  * context which was originally shared using clone.  copy_*
1574  * functions used by do_fork() cannot be used here directly
1575  * because they modify an inactive task_struct that is being
1576  * constructed. Here we are modifying the current, active,
1577  * task_struct.
1578  */
1579 asmlinkage long sys_unshare(unsigned long unshare_flags)
1580 {
1581         int err = 0;
1582         struct fs_struct *fs, *new_fs = NULL;
1583         struct sighand_struct *new_sigh = NULL;
1584         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1585         struct files_struct *fd, *new_fd = NULL;
1586         struct nsproxy *new_nsproxy = NULL;
1587         int do_sysvsem = 0;
1588
1589         check_unshare_flags(&unshare_flags);
1590
1591         /* Return -EINVAL for all unsupported flags */
1592         err = -EINVAL;
1593         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1594                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1595                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1596                                 CLONE_NEWNET))
1597                 goto bad_unshare_out;
1598
1599         /*
1600          * CLONE_NEWIPC must also detach from the undolist: after switching
1601          * to a new ipc namespace, the semaphore arrays from the old
1602          * namespace are unreachable.
1603          */
1604         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1605                 do_sysvsem = 1;
1606         if ((err = unshare_thread(unshare_flags)))
1607                 goto bad_unshare_out;
1608         if ((err = unshare_fs(unshare_flags, &new_fs)))
1609                 goto bad_unshare_cleanup_thread;
1610         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1611                 goto bad_unshare_cleanup_fs;
1612         if ((err = unshare_vm(unshare_flags, &new_mm)))
1613                 goto bad_unshare_cleanup_sigh;
1614         if ((err = unshare_fd(unshare_flags, &new_fd)))
1615                 goto bad_unshare_cleanup_vm;
1616         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1617                         new_fs)))
1618                 goto bad_unshare_cleanup_fd;
1619
1620         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1621                 if (do_sysvsem) {
1622                         /*
1623                          * CLONE_SYSVSEM is equivalent to sys_exit().
1624                          */
1625                         exit_sem(current);
1626                 }
1627
1628                 if (new_nsproxy) {
1629                         switch_task_namespaces(current, new_nsproxy);
1630                         new_nsproxy = NULL;
1631                 }
1632
1633                 task_lock(current);
1634
1635                 if (new_fs) {
1636                         fs = current->fs;
1637                         current->fs = new_fs;
1638                         new_fs = fs;
1639                 }
1640
1641                 if (new_mm) {
1642                         mm = current->mm;
1643                         active_mm = current->active_mm;
1644                         current->mm = new_mm;
1645                         current->active_mm = new_mm;
1646                         activate_mm(active_mm, new_mm);
1647                         new_mm = mm;
1648                 }
1649
1650                 if (new_fd) {
1651                         fd = current->files;
1652                         current->files = new_fd;
1653                         new_fd = fd;
1654                 }
1655
1656                 task_unlock(current);
1657         }
1658
1659         if (new_nsproxy)
1660                 put_nsproxy(new_nsproxy);
1661
1662 bad_unshare_cleanup_fd:
1663         if (new_fd)
1664                 put_files_struct(new_fd);
1665
1666 bad_unshare_cleanup_vm:
1667         if (new_mm)
1668                 mmput(new_mm);
1669
1670 bad_unshare_cleanup_sigh:
1671         if (new_sigh)
1672                 if (atomic_dec_and_test(&new_sigh->count))
1673                         kmem_cache_free(sighand_cachep, new_sigh);
1674
1675 bad_unshare_cleanup_fs:
1676         if (new_fs)
1677                 put_fs_struct(new_fs);
1678
1679 bad_unshare_cleanup_thread:
1680 bad_unshare_out:
1681         return err;
1682 }
1683
1684 /*
1685  *      Helper to unshare the files of the current task.
1686  *      We don't want to expose copy_files internals to
1687  *      the exec layer of the kernel.
1688  */
1689
1690 int unshare_files(struct files_struct **displaced)
1691 {
1692         struct task_struct *task = current;
1693         struct files_struct *copy = NULL;
1694         int error;
1695
1696         error = unshare_fd(CLONE_FILES, &copy);
1697         if (error || !copy) {
1698                 *displaced = NULL;
1699                 return error;
1700         }
1701         *displaced = task->files;
1702         task_lock(task);
1703         task->files = copy;
1704         task_unlock(task);
1705         return 0;
1706 }