Merge branch 'tracing-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[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/compat.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/profile.h>
52 #include <linux/rmap.h>
53 #include <linux/acct.h>
54 #include <linux/tsacct_kern.h>
55 #include <linux/cn_proc.h>
56 #include <linux/freezer.h>
57 #include <linux/delayacct.h>
58 #include <linux/taskstats_kern.h>
59 #include <linux/random.h>
60 #include <linux/tty.h>
61 #include <linux/proc_fs.h>
62 #include <linux/blkdev.h>
63 #include <trace/sched.h>
64
65 #include <asm/pgtable.h>
66 #include <asm/pgalloc.h>
67 #include <asm/uaccess.h>
68 #include <asm/mmu_context.h>
69 #include <asm/cacheflush.h>
70 #include <asm/tlbflush.h>
71
72 /*
73  * Protected counters by write_lock_irq(&tasklist_lock)
74  */
75 unsigned long total_forks;      /* Handle normal Linux uptimes. */
76 int nr_threads;                 /* The idle threads do not count.. */
77
78 int max_threads;                /* tunable limit on nr_threads */
79
80 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
81
82 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
83
84 DEFINE_TRACE(sched_process_fork);
85
86 int nr_processes(void)
87 {
88         int cpu;
89         int total = 0;
90
91         for_each_online_cpu(cpu)
92                 total += per_cpu(process_counts, cpu);
93
94         return total;
95 }
96
97 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
98 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
99 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
100 static struct kmem_cache *task_struct_cachep;
101 #endif
102
103 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
104 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
105 {
106 #ifdef CONFIG_DEBUG_STACK_USAGE
107         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
108 #else
109         gfp_t mask = GFP_KERNEL;
110 #endif
111         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
112 }
113
114 static inline void free_thread_info(struct thread_info *ti)
115 {
116         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
117 }
118 #endif
119
120 /* SLAB cache for signal_struct structures (tsk->signal) */
121 static struct kmem_cache *signal_cachep;
122
123 /* SLAB cache for sighand_struct structures (tsk->sighand) */
124 struct kmem_cache *sighand_cachep;
125
126 /* SLAB cache for files_struct structures (tsk->files) */
127 struct kmem_cache *files_cachep;
128
129 /* SLAB cache for fs_struct structures (tsk->fs) */
130 struct kmem_cache *fs_cachep;
131
132 /* SLAB cache for vm_area_struct structures */
133 struct kmem_cache *vm_area_cachep;
134
135 /* SLAB cache for mm_struct structures (tsk->mm) */
136 static struct kmem_cache *mm_cachep;
137
138 void free_task(struct task_struct *tsk)
139 {
140         prop_local_destroy_single(&tsk->dirties);
141         free_thread_info(tsk->stack);
142         rt_mutex_debug_task_free(tsk);
143         ftrace_graph_exit_task(tsk);
144         free_task_struct(tsk);
145 }
146 EXPORT_SYMBOL(free_task);
147
148 void __put_task_struct(struct task_struct *tsk)
149 {
150         WARN_ON(!tsk->exit_state);
151         WARN_ON(atomic_read(&tsk->usage));
152         WARN_ON(tsk == current);
153
154         put_cred(tsk->real_cred);
155         put_cred(tsk->cred);
156         delayacct_tsk_free(tsk);
157
158         if (!profile_handoff_task(tsk))
159                 free_task(tsk);
160 }
161
162 /*
163  * macro override instead of weak attribute alias, to workaround
164  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
165  */
166 #ifndef arch_task_cache_init
167 #define arch_task_cache_init()
168 #endif
169
170 void __init fork_init(unsigned long mempages)
171 {
172 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
173 #ifndef ARCH_MIN_TASKALIGN
174 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
175 #endif
176         /* create a slab on which task_structs can be allocated */
177         task_struct_cachep =
178                 kmem_cache_create("task_struct", sizeof(struct task_struct),
179                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
180 #endif
181
182         /* do the arch specific task caches init */
183         arch_task_cache_init();
184
185         /*
186          * The default maximum number of threads is set to a safe
187          * value: the thread structures can take up at most half
188          * of memory.
189          */
190         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
191
192         /*
193          * we need to allow at least 20 threads to boot a system
194          */
195         if(max_threads < 20)
196                 max_threads = 20;
197
198         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
199         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
200         init_task.signal->rlim[RLIMIT_SIGPENDING] =
201                 init_task.signal->rlim[RLIMIT_NPROC];
202 }
203
204 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
205                                                struct task_struct *src)
206 {
207         *dst = *src;
208         return 0;
209 }
210
211 static struct task_struct *dup_task_struct(struct task_struct *orig)
212 {
213         struct task_struct *tsk;
214         struct thread_info *ti;
215         int err;
216
217         prepare_to_copy(orig);
218
219         tsk = alloc_task_struct();
220         if (!tsk)
221                 return NULL;
222
223         ti = alloc_thread_info(tsk);
224         if (!ti) {
225                 free_task_struct(tsk);
226                 return NULL;
227         }
228
229         err = arch_dup_task_struct(tsk, orig);
230         if (err)
231                 goto out;
232
233         tsk->stack = ti;
234
235         err = prop_local_init_single(&tsk->dirties);
236         if (err)
237                 goto out;
238
239         setup_thread_stack(tsk, orig);
240
241 #ifdef CONFIG_CC_STACKPROTECTOR
242         tsk->stack_canary = get_random_int();
243 #endif
244
245         /* One for us, one for whoever does the "release_task()" (usually parent) */
246         atomic_set(&tsk->usage,2);
247         atomic_set(&tsk->fs_excl, 0);
248 #ifdef CONFIG_BLK_DEV_IO_TRACE
249         tsk->btrace_seq = 0;
250 #endif
251         tsk->splice_pipe = NULL;
252         return tsk;
253
254 out:
255         free_thread_info(ti);
256         free_task_struct(tsk);
257         return NULL;
258 }
259
260 #ifdef CONFIG_MMU
261 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
262 {
263         struct vm_area_struct *mpnt, *tmp, **pprev;
264         struct rb_node **rb_link, *rb_parent;
265         int retval;
266         unsigned long charge;
267         struct mempolicy *pol;
268
269         down_write(&oldmm->mmap_sem);
270         flush_cache_dup_mm(oldmm);
271         /*
272          * Not linked in yet - no deadlock potential:
273          */
274         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
275
276         mm->locked_vm = 0;
277         mm->mmap = NULL;
278         mm->mmap_cache = NULL;
279         mm->free_area_cache = oldmm->mmap_base;
280         mm->cached_hole_size = ~0UL;
281         mm->map_count = 0;
282         cpus_clear(mm->cpu_vm_mask);
283         mm->mm_rb = RB_ROOT;
284         rb_link = &mm->mm_rb.rb_node;
285         rb_parent = NULL;
286         pprev = &mm->mmap;
287
288         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
289                 struct file *file;
290
291                 if (mpnt->vm_flags & VM_DONTCOPY) {
292                         long pages = vma_pages(mpnt);
293                         mm->total_vm -= pages;
294                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
295                                                                 -pages);
296                         continue;
297                 }
298                 charge = 0;
299                 if (mpnt->vm_flags & VM_ACCOUNT) {
300                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
301                         if (security_vm_enough_memory(len))
302                                 goto fail_nomem;
303                         charge = len;
304                 }
305                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
306                 if (!tmp)
307                         goto fail_nomem;
308                 *tmp = *mpnt;
309                 pol = mpol_dup(vma_policy(mpnt));
310                 retval = PTR_ERR(pol);
311                 if (IS_ERR(pol))
312                         goto fail_nomem_policy;
313                 vma_set_policy(tmp, pol);
314                 tmp->vm_flags &= ~VM_LOCKED;
315                 tmp->vm_mm = mm;
316                 tmp->vm_next = NULL;
317                 anon_vma_link(tmp);
318                 file = tmp->vm_file;
319                 if (file) {
320                         struct inode *inode = file->f_path.dentry->d_inode;
321                         struct address_space *mapping = file->f_mapping;
322
323                         get_file(file);
324                         if (tmp->vm_flags & VM_DENYWRITE)
325                                 atomic_dec(&inode->i_writecount);
326                         spin_lock(&mapping->i_mmap_lock);
327                         if (tmp->vm_flags & VM_SHARED)
328                                 mapping->i_mmap_writable++;
329                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
330                         flush_dcache_mmap_lock(mapping);
331                         /* insert tmp into the share list, just after mpnt */
332                         vma_prio_tree_add(tmp, mpnt);
333                         flush_dcache_mmap_unlock(mapping);
334                         spin_unlock(&mapping->i_mmap_lock);
335                 }
336
337                 /*
338                  * Clear hugetlb-related page reserves for children. This only
339                  * affects MAP_PRIVATE mappings. Faults generated by the child
340                  * are not guaranteed to succeed, even if read-only
341                  */
342                 if (is_vm_hugetlb_page(tmp))
343                         reset_vma_resv_huge_pages(tmp);
344
345                 /*
346                  * Link in the new vma and copy the page table entries.
347                  */
348                 *pprev = tmp;
349                 pprev = &tmp->vm_next;
350
351                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
352                 rb_link = &tmp->vm_rb.rb_right;
353                 rb_parent = &tmp->vm_rb;
354
355                 mm->map_count++;
356                 retval = copy_page_range(mm, oldmm, mpnt);
357
358                 if (tmp->vm_ops && tmp->vm_ops->open)
359                         tmp->vm_ops->open(tmp);
360
361                 if (retval)
362                         goto out;
363         }
364         /* a new mm has just been created */
365         arch_dup_mmap(oldmm, mm);
366         retval = 0;
367 out:
368         up_write(&mm->mmap_sem);
369         flush_tlb_mm(oldmm);
370         up_write(&oldmm->mmap_sem);
371         return retval;
372 fail_nomem_policy:
373         kmem_cache_free(vm_area_cachep, tmp);
374 fail_nomem:
375         retval = -ENOMEM;
376         vm_unacct_memory(charge);
377         goto out;
378 }
379
380 static inline int mm_alloc_pgd(struct mm_struct * mm)
381 {
382         mm->pgd = pgd_alloc(mm);
383         if (unlikely(!mm->pgd))
384                 return -ENOMEM;
385         return 0;
386 }
387
388 static inline void mm_free_pgd(struct mm_struct * mm)
389 {
390         pgd_free(mm, mm->pgd);
391 }
392 #else
393 #define dup_mmap(mm, oldmm)     (0)
394 #define mm_alloc_pgd(mm)        (0)
395 #define mm_free_pgd(mm)
396 #endif /* CONFIG_MMU */
397
398 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
399
400 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
401 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
402
403 #include <linux/init_task.h>
404
405 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
406 {
407         atomic_set(&mm->mm_users, 1);
408         atomic_set(&mm->mm_count, 1);
409         init_rwsem(&mm->mmap_sem);
410         INIT_LIST_HEAD(&mm->mmlist);
411         mm->flags = (current->mm) ? current->mm->flags
412                                   : MMF_DUMP_FILTER_DEFAULT;
413         mm->core_state = NULL;
414         mm->nr_ptes = 0;
415         set_mm_counter(mm, file_rss, 0);
416         set_mm_counter(mm, anon_rss, 0);
417         spin_lock_init(&mm->page_table_lock);
418         rwlock_init(&mm->ioctx_list_lock);
419         mm->ioctx_list = NULL;
420         mm->free_area_cache = TASK_UNMAPPED_BASE;
421         mm->cached_hole_size = ~0UL;
422         mm_init_owner(mm, p);
423
424         if (likely(!mm_alloc_pgd(mm))) {
425                 mm->def_flags = 0;
426                 mmu_notifier_mm_init(mm);
427                 return mm;
428         }
429
430         free_mm(mm);
431         return NULL;
432 }
433
434 /*
435  * Allocate and initialize an mm_struct.
436  */
437 struct mm_struct * mm_alloc(void)
438 {
439         struct mm_struct * mm;
440
441         mm = allocate_mm();
442         if (mm) {
443                 memset(mm, 0, sizeof(*mm));
444                 mm = mm_init(mm, current);
445         }
446         return mm;
447 }
448
449 /*
450  * Called when the last reference to the mm
451  * is dropped: either by a lazy thread or by
452  * mmput. Free the page directory and the mm.
453  */
454 void __mmdrop(struct mm_struct *mm)
455 {
456         BUG_ON(mm == &init_mm);
457         mm_free_pgd(mm);
458         destroy_context(mm);
459         mmu_notifier_mm_destroy(mm);
460         free_mm(mm);
461 }
462 EXPORT_SYMBOL_GPL(__mmdrop);
463
464 /*
465  * Decrement the use count and release all resources for an mm.
466  */
467 void mmput(struct mm_struct *mm)
468 {
469         might_sleep();
470
471         if (atomic_dec_and_test(&mm->mm_users)) {
472                 exit_aio(mm);
473                 exit_mmap(mm);
474                 set_mm_exe_file(mm, NULL);
475                 if (!list_empty(&mm->mmlist)) {
476                         spin_lock(&mmlist_lock);
477                         list_del(&mm->mmlist);
478                         spin_unlock(&mmlist_lock);
479                 }
480                 put_swap_token(mm);
481                 mmdrop(mm);
482         }
483 }
484 EXPORT_SYMBOL_GPL(mmput);
485
486 /**
487  * get_task_mm - acquire a reference to the task's mm
488  *
489  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
490  * this kernel workthread has transiently adopted a user mm with use_mm,
491  * to do its AIO) is not set and if so returns a reference to it, after
492  * bumping up the use count.  User must release the mm via mmput()
493  * after use.  Typically used by /proc and ptrace.
494  */
495 struct mm_struct *get_task_mm(struct task_struct *task)
496 {
497         struct mm_struct *mm;
498
499         task_lock(task);
500         mm = task->mm;
501         if (mm) {
502                 if (task->flags & PF_KTHREAD)
503                         mm = NULL;
504                 else
505                         atomic_inc(&mm->mm_users);
506         }
507         task_unlock(task);
508         return mm;
509 }
510 EXPORT_SYMBOL_GPL(get_task_mm);
511
512 /* Please note the differences between mmput and mm_release.
513  * mmput is called whenever we stop holding onto a mm_struct,
514  * error success whatever.
515  *
516  * mm_release is called after a mm_struct has been removed
517  * from the current process.
518  *
519  * This difference is important for error handling, when we
520  * only half set up a mm_struct for a new process and need to restore
521  * the old one.  Because we mmput the new mm_struct before
522  * restoring the old one. . .
523  * Eric Biederman 10 January 1998
524  */
525 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
526 {
527         struct completion *vfork_done = tsk->vfork_done;
528
529         /* Get rid of any futexes when releasing the mm */
530 #ifdef CONFIG_FUTEX
531         if (unlikely(tsk->robust_list))
532                 exit_robust_list(tsk);
533 #ifdef CONFIG_COMPAT
534         if (unlikely(tsk->compat_robust_list))
535                 compat_exit_robust_list(tsk);
536 #endif
537 #endif
538
539         /* Get rid of any cached register state */
540         deactivate_mm(tsk, mm);
541
542         /* notify parent sleeping on vfork() */
543         if (vfork_done) {
544                 tsk->vfork_done = NULL;
545                 complete(vfork_done);
546         }
547
548         /*
549          * If we're exiting normally, clear a user-space tid field if
550          * requested.  We leave this alone when dying by signal, to leave
551          * the value intact in a core dump, and to save the unnecessary
552          * trouble otherwise.  Userland only wants this done for a sys_exit.
553          */
554         if (tsk->clear_child_tid
555             && !(tsk->flags & PF_SIGNALED)
556             && atomic_read(&mm->mm_users) > 1) {
557                 u32 __user * tidptr = tsk->clear_child_tid;
558                 tsk->clear_child_tid = NULL;
559
560                 /*
561                  * We don't check the error code - if userspace has
562                  * not set up a proper pointer then tough luck.
563                  */
564                 put_user(0, tidptr);
565                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
566         }
567 }
568
569 /*
570  * Allocate a new mm structure and copy contents from the
571  * mm structure of the passed in task structure.
572  */
573 struct mm_struct *dup_mm(struct task_struct *tsk)
574 {
575         struct mm_struct *mm, *oldmm = current->mm;
576         int err;
577
578         if (!oldmm)
579                 return NULL;
580
581         mm = allocate_mm();
582         if (!mm)
583                 goto fail_nomem;
584
585         memcpy(mm, oldmm, sizeof(*mm));
586
587         /* Initializing for Swap token stuff */
588         mm->token_priority = 0;
589         mm->last_interval = 0;
590
591         if (!mm_init(mm, tsk))
592                 goto fail_nomem;
593
594         if (init_new_context(tsk, mm))
595                 goto fail_nocontext;
596
597         dup_mm_exe_file(oldmm, mm);
598
599         err = dup_mmap(mm, oldmm);
600         if (err)
601                 goto free_pt;
602
603         mm->hiwater_rss = get_mm_rss(mm);
604         mm->hiwater_vm = mm->total_vm;
605
606         return mm;
607
608 free_pt:
609         mmput(mm);
610
611 fail_nomem:
612         return NULL;
613
614 fail_nocontext:
615         /*
616          * If init_new_context() failed, we cannot use mmput() to free the mm
617          * because it calls destroy_context()
618          */
619         mm_free_pgd(mm);
620         free_mm(mm);
621         return NULL;
622 }
623
624 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
625 {
626         struct mm_struct * mm, *oldmm;
627         int retval;
628
629         tsk->min_flt = tsk->maj_flt = 0;
630         tsk->nvcsw = tsk->nivcsw = 0;
631
632         tsk->mm = NULL;
633         tsk->active_mm = NULL;
634
635         /*
636          * Are we cloning a kernel thread?
637          *
638          * We need to steal a active VM for that..
639          */
640         oldmm = current->mm;
641         if (!oldmm)
642                 return 0;
643
644         if (clone_flags & CLONE_VM) {
645                 atomic_inc(&oldmm->mm_users);
646                 mm = oldmm;
647                 goto good_mm;
648         }
649
650         retval = -ENOMEM;
651         mm = dup_mm(tsk);
652         if (!mm)
653                 goto fail_nomem;
654
655 good_mm:
656         /* Initializing for Swap token stuff */
657         mm->token_priority = 0;
658         mm->last_interval = 0;
659
660         tsk->mm = mm;
661         tsk->active_mm = mm;
662         return 0;
663
664 fail_nomem:
665         return retval;
666 }
667
668 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
669 {
670         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
671         /* We don't need to lock fs - think why ;-) */
672         if (fs) {
673                 atomic_set(&fs->count, 1);
674                 rwlock_init(&fs->lock);
675                 fs->umask = old->umask;
676                 read_lock(&old->lock);
677                 fs->root = old->root;
678                 path_get(&old->root);
679                 fs->pwd = old->pwd;
680                 path_get(&old->pwd);
681                 read_unlock(&old->lock);
682         }
683         return fs;
684 }
685
686 struct fs_struct *copy_fs_struct(struct fs_struct *old)
687 {
688         return __copy_fs_struct(old);
689 }
690
691 EXPORT_SYMBOL_GPL(copy_fs_struct);
692
693 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
694 {
695         if (clone_flags & CLONE_FS) {
696                 atomic_inc(&current->fs->count);
697                 return 0;
698         }
699         tsk->fs = __copy_fs_struct(current->fs);
700         if (!tsk->fs)
701                 return -ENOMEM;
702         return 0;
703 }
704
705 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
706 {
707         struct files_struct *oldf, *newf;
708         int error = 0;
709
710         /*
711          * A background process may not have any files ...
712          */
713         oldf = current->files;
714         if (!oldf)
715                 goto out;
716
717         if (clone_flags & CLONE_FILES) {
718                 atomic_inc(&oldf->count);
719                 goto out;
720         }
721
722         newf = dup_fd(oldf, &error);
723         if (!newf)
724                 goto out;
725
726         tsk->files = newf;
727         error = 0;
728 out:
729         return error;
730 }
731
732 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
733 {
734 #ifdef CONFIG_BLOCK
735         struct io_context *ioc = current->io_context;
736
737         if (!ioc)
738                 return 0;
739         /*
740          * Share io context with parent, if CLONE_IO is set
741          */
742         if (clone_flags & CLONE_IO) {
743                 tsk->io_context = ioc_task_link(ioc);
744                 if (unlikely(!tsk->io_context))
745                         return -ENOMEM;
746         } else if (ioprio_valid(ioc->ioprio)) {
747                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
748                 if (unlikely(!tsk->io_context))
749                         return -ENOMEM;
750
751                 tsk->io_context->ioprio = ioc->ioprio;
752         }
753 #endif
754         return 0;
755 }
756
757 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
758 {
759         struct sighand_struct *sig;
760
761         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
762                 atomic_inc(&current->sighand->count);
763                 return 0;
764         }
765         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
766         rcu_assign_pointer(tsk->sighand, sig);
767         if (!sig)
768                 return -ENOMEM;
769         atomic_set(&sig->count, 1);
770         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
771         return 0;
772 }
773
774 void __cleanup_sighand(struct sighand_struct *sighand)
775 {
776         if (atomic_dec_and_test(&sighand->count))
777                 kmem_cache_free(sighand_cachep, sighand);
778 }
779
780
781 /*
782  * Initialize POSIX timer handling for a thread group.
783  */
784 static void posix_cpu_timers_init_group(struct signal_struct *sig)
785 {
786         /* Thread group counters. */
787         thread_group_cputime_init(sig);
788
789         /* Expiration times and increments. */
790         sig->it_virt_expires = cputime_zero;
791         sig->it_virt_incr = cputime_zero;
792         sig->it_prof_expires = cputime_zero;
793         sig->it_prof_incr = cputime_zero;
794
795         /* Cached expiration times. */
796         sig->cputime_expires.prof_exp = cputime_zero;
797         sig->cputime_expires.virt_exp = cputime_zero;
798         sig->cputime_expires.sched_exp = 0;
799
800         /* The timer lists. */
801         INIT_LIST_HEAD(&sig->cpu_timers[0]);
802         INIT_LIST_HEAD(&sig->cpu_timers[1]);
803         INIT_LIST_HEAD(&sig->cpu_timers[2]);
804 }
805
806 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
807 {
808         struct signal_struct *sig;
809         int ret;
810
811         if (clone_flags & CLONE_THREAD) {
812                 ret = thread_group_cputime_clone_thread(current);
813                 if (likely(!ret)) {
814                         atomic_inc(&current->signal->count);
815                         atomic_inc(&current->signal->live);
816                 }
817                 return ret;
818         }
819         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
820         tsk->signal = sig;
821         if (!sig)
822                 return -ENOMEM;
823
824         atomic_set(&sig->count, 1);
825         atomic_set(&sig->live, 1);
826         init_waitqueue_head(&sig->wait_chldexit);
827         sig->flags = 0;
828         sig->group_exit_code = 0;
829         sig->group_exit_task = NULL;
830         sig->group_stop_count = 0;
831         sig->curr_target = tsk;
832         init_sigpending(&sig->shared_pending);
833         INIT_LIST_HEAD(&sig->posix_timers);
834
835         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
836         sig->it_real_incr.tv64 = 0;
837         sig->real_timer.function = it_real_fn;
838
839         sig->leader = 0;        /* session leadership doesn't inherit */
840         sig->tty_old_pgrp = NULL;
841         sig->tty = NULL;
842
843         sig->cutime = sig->cstime = cputime_zero;
844         sig->gtime = cputime_zero;
845         sig->cgtime = cputime_zero;
846         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
847         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
848         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
849         task_io_accounting_init(&sig->ioac);
850         taskstats_tgid_init(sig);
851
852         task_lock(current->group_leader);
853         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
854         task_unlock(current->group_leader);
855
856         posix_cpu_timers_init_group(sig);
857
858         acct_init_pacct(&sig->pacct);
859
860         tty_audit_fork(sig);
861
862         return 0;
863 }
864
865 void __cleanup_signal(struct signal_struct *sig)
866 {
867         thread_group_cputime_free(sig);
868         tty_kref_put(sig->tty);
869         kmem_cache_free(signal_cachep, sig);
870 }
871
872 static void cleanup_signal(struct task_struct *tsk)
873 {
874         struct signal_struct *sig = tsk->signal;
875
876         atomic_dec(&sig->live);
877
878         if (atomic_dec_and_test(&sig->count))
879                 __cleanup_signal(sig);
880 }
881
882 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
883 {
884         unsigned long new_flags = p->flags;
885
886         new_flags &= ~PF_SUPERPRIV;
887         new_flags |= PF_FORKNOEXEC;
888         new_flags |= PF_STARTING;
889         p->flags = new_flags;
890         clear_freeze_flag(p);
891 }
892
893 asmlinkage long sys_set_tid_address(int __user *tidptr)
894 {
895         current->clear_child_tid = tidptr;
896
897         return task_pid_vnr(current);
898 }
899
900 static void rt_mutex_init_task(struct task_struct *p)
901 {
902         spin_lock_init(&p->pi_lock);
903 #ifdef CONFIG_RT_MUTEXES
904         plist_head_init(&p->pi_waiters, &p->pi_lock);
905         p->pi_blocked_on = NULL;
906 #endif
907 }
908
909 #ifdef CONFIG_MM_OWNER
910 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
911 {
912         mm->owner = p;
913 }
914 #endif /* CONFIG_MM_OWNER */
915
916 /*
917  * Initialize POSIX timer handling for a single task.
918  */
919 static void posix_cpu_timers_init(struct task_struct *tsk)
920 {
921         tsk->cputime_expires.prof_exp = cputime_zero;
922         tsk->cputime_expires.virt_exp = cputime_zero;
923         tsk->cputime_expires.sched_exp = 0;
924         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
925         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
926         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
927 }
928
929 /*
930  * This creates a new process as a copy of the old one,
931  * but does not actually start it yet.
932  *
933  * It copies the registers, and all the appropriate
934  * parts of the process environment (as per the clone
935  * flags). The actual kick-off is left to the caller.
936  */
937 static struct task_struct *copy_process(unsigned long clone_flags,
938                                         unsigned long stack_start,
939                                         struct pt_regs *regs,
940                                         unsigned long stack_size,
941                                         int __user *child_tidptr,
942                                         struct pid *pid,
943                                         int trace)
944 {
945         int retval;
946         struct task_struct *p;
947         int cgroup_callbacks_done = 0;
948
949         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
950                 return ERR_PTR(-EINVAL);
951
952         /*
953          * Thread groups must share signals as well, and detached threads
954          * can only be started up within the thread group.
955          */
956         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
957                 return ERR_PTR(-EINVAL);
958
959         /*
960          * Shared signal handlers imply shared VM. By way of the above,
961          * thread groups also imply shared VM. Blocking this case allows
962          * for various simplifications in other code.
963          */
964         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
965                 return ERR_PTR(-EINVAL);
966
967         retval = security_task_create(clone_flags);
968         if (retval)
969                 goto fork_out;
970
971         retval = -ENOMEM;
972         p = dup_task_struct(current);
973         if (!p)
974                 goto fork_out;
975
976         rt_mutex_init_task(p);
977
978 #ifdef CONFIG_PROVE_LOCKING
979         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
980         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
981 #endif
982         retval = -EAGAIN;
983         if (atomic_read(&p->real_cred->user->processes) >=
984                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
985                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
986                     p->real_cred->user != INIT_USER)
987                         goto bad_fork_free;
988         }
989
990         retval = copy_creds(p, clone_flags);
991         if (retval < 0)
992                 goto bad_fork_free;
993
994         /*
995          * If multiple threads are within copy_process(), then this check
996          * triggers too late. This doesn't hurt, the check is only there
997          * to stop root fork bombs.
998          */
999         if (nr_threads >= max_threads)
1000                 goto bad_fork_cleanup_count;
1001
1002         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1003                 goto bad_fork_cleanup_count;
1004
1005         if (p->binfmt && !try_module_get(p->binfmt->module))
1006                 goto bad_fork_cleanup_put_domain;
1007
1008         p->did_exec = 0;
1009         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1010         copy_flags(clone_flags, p);
1011         INIT_LIST_HEAD(&p->children);
1012         INIT_LIST_HEAD(&p->sibling);
1013 #ifdef CONFIG_PREEMPT_RCU
1014         p->rcu_read_lock_nesting = 0;
1015         p->rcu_flipctr_idx = 0;
1016 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1017         p->vfork_done = NULL;
1018         spin_lock_init(&p->alloc_lock);
1019
1020         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1021         init_sigpending(&p->pending);
1022
1023         p->utime = cputime_zero;
1024         p->stime = cputime_zero;
1025         p->gtime = cputime_zero;
1026         p->utimescaled = cputime_zero;
1027         p->stimescaled = cputime_zero;
1028         p->prev_utime = cputime_zero;
1029         p->prev_stime = cputime_zero;
1030
1031         p->default_timer_slack_ns = current->timer_slack_ns;
1032
1033 #ifdef CONFIG_DETECT_SOFTLOCKUP
1034         p->last_switch_count = 0;
1035         p->last_switch_timestamp = 0;
1036 #endif
1037
1038         task_io_accounting_init(&p->ioac);
1039         acct_clear_integrals(p);
1040
1041         posix_cpu_timers_init(p);
1042
1043         p->lock_depth = -1;             /* -1 = no lock */
1044         do_posix_clock_monotonic_gettime(&p->start_time);
1045         p->real_start_time = p->start_time;
1046         monotonic_to_bootbased(&p->real_start_time);
1047         p->io_context = NULL;
1048         p->audit_context = NULL;
1049         cgroup_fork(p);
1050 #ifdef CONFIG_NUMA
1051         p->mempolicy = mpol_dup(p->mempolicy);
1052         if (IS_ERR(p->mempolicy)) {
1053                 retval = PTR_ERR(p->mempolicy);
1054                 p->mempolicy = NULL;
1055                 goto bad_fork_cleanup_cgroup;
1056         }
1057         mpol_fix_fork_child_flag(p);
1058 #endif
1059 #ifdef CONFIG_TRACE_IRQFLAGS
1060         p->irq_events = 0;
1061 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1062         p->hardirqs_enabled = 1;
1063 #else
1064         p->hardirqs_enabled = 0;
1065 #endif
1066         p->hardirq_enable_ip = 0;
1067         p->hardirq_enable_event = 0;
1068         p->hardirq_disable_ip = _THIS_IP_;
1069         p->hardirq_disable_event = 0;
1070         p->softirqs_enabled = 1;
1071         p->softirq_enable_ip = _THIS_IP_;
1072         p->softirq_enable_event = 0;
1073         p->softirq_disable_ip = 0;
1074         p->softirq_disable_event = 0;
1075         p->hardirq_context = 0;
1076         p->softirq_context = 0;
1077 #endif
1078 #ifdef CONFIG_LOCKDEP
1079         p->lockdep_depth = 0; /* no locks held yet */
1080         p->curr_chain_key = 0;
1081         p->lockdep_recursion = 0;
1082 #endif
1083
1084 #ifdef CONFIG_DEBUG_MUTEXES
1085         p->blocked_on = NULL; /* not blocked yet */
1086 #endif
1087         if (unlikely(ptrace_reparented(current)))
1088                 ptrace_fork(p, clone_flags);
1089
1090         /* Perform scheduler related setup. Assign this task to a CPU. */
1091         sched_fork(p, clone_flags);
1092
1093         if ((retval = audit_alloc(p)))
1094                 goto bad_fork_cleanup_policy;
1095         /* copy all the process information */
1096         if ((retval = copy_semundo(clone_flags, p)))
1097                 goto bad_fork_cleanup_audit;
1098         if ((retval = copy_files(clone_flags, p)))
1099                 goto bad_fork_cleanup_semundo;
1100         if ((retval = copy_fs(clone_flags, p)))
1101                 goto bad_fork_cleanup_files;
1102         if ((retval = copy_sighand(clone_flags, p)))
1103                 goto bad_fork_cleanup_fs;
1104         if ((retval = copy_signal(clone_flags, p)))
1105                 goto bad_fork_cleanup_sighand;
1106         if ((retval = copy_mm(clone_flags, p)))
1107                 goto bad_fork_cleanup_signal;
1108         if ((retval = copy_namespaces(clone_flags, p)))
1109                 goto bad_fork_cleanup_mm;
1110         if ((retval = copy_io(clone_flags, p)))
1111                 goto bad_fork_cleanup_namespaces;
1112         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1113         if (retval)
1114                 goto bad_fork_cleanup_io;
1115
1116         if (pid != &init_struct_pid) {
1117                 retval = -ENOMEM;
1118                 pid = alloc_pid(task_active_pid_ns(p));
1119                 if (!pid)
1120                         goto bad_fork_cleanup_io;
1121
1122                 if (clone_flags & CLONE_NEWPID) {
1123                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1124                         if (retval < 0)
1125                                 goto bad_fork_free_pid;
1126                 }
1127         }
1128
1129         ftrace_graph_init_task(p);
1130
1131         p->pid = pid_nr(pid);
1132         p->tgid = p->pid;
1133         if (clone_flags & CLONE_THREAD)
1134                 p->tgid = current->tgid;
1135
1136         if (current->nsproxy != p->nsproxy) {
1137                 retval = ns_cgroup_clone(p, pid);
1138                 if (retval)
1139                         goto bad_fork_free_graph;
1140         }
1141
1142         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1143         /*
1144          * Clear TID on mm_release()?
1145          */
1146         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1147 #ifdef CONFIG_FUTEX
1148         p->robust_list = NULL;
1149 #ifdef CONFIG_COMPAT
1150         p->compat_robust_list = NULL;
1151 #endif
1152         INIT_LIST_HEAD(&p->pi_state_list);
1153         p->pi_state_cache = NULL;
1154 #endif
1155         /*
1156          * sigaltstack should be cleared when sharing the same VM
1157          */
1158         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1159                 p->sas_ss_sp = p->sas_ss_size = 0;
1160
1161         /*
1162          * Syscall tracing should be turned off in the child regardless
1163          * of CLONE_PTRACE.
1164          */
1165         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1166 #ifdef TIF_SYSCALL_EMU
1167         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1168 #endif
1169         clear_all_latency_tracing(p);
1170
1171         /* Our parent execution domain becomes current domain
1172            These must match for thread signalling to apply */
1173         p->parent_exec_id = p->self_exec_id;
1174
1175         /* ok, now we should be set up.. */
1176         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1177         p->pdeath_signal = 0;
1178         p->exit_state = 0;
1179
1180         /*
1181          * Ok, make it visible to the rest of the system.
1182          * We dont wake it up yet.
1183          */
1184         p->group_leader = p;
1185         INIT_LIST_HEAD(&p->thread_group);
1186
1187         /* Now that the task is set up, run cgroup callbacks if
1188          * necessary. We need to run them before the task is visible
1189          * on the tasklist. */
1190         cgroup_fork_callbacks(p);
1191         cgroup_callbacks_done = 1;
1192
1193         /* Need tasklist lock for parent etc handling! */
1194         write_lock_irq(&tasklist_lock);
1195
1196         /*
1197          * The task hasn't been attached yet, so its cpus_allowed mask will
1198          * not be changed, nor will its assigned CPU.
1199          *
1200          * The cpus_allowed mask of the parent may have changed after it was
1201          * copied first time - so re-copy it here, then check the child's CPU
1202          * to ensure it is on a valid CPU (and if not, just force it back to
1203          * parent's CPU). This avoids alot of nasty races.
1204          */
1205         p->cpus_allowed = current->cpus_allowed;
1206         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1207         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1208                         !cpu_online(task_cpu(p))))
1209                 set_task_cpu(p, smp_processor_id());
1210
1211         /* CLONE_PARENT re-uses the old parent */
1212         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1213                 p->real_parent = current->real_parent;
1214         else
1215                 p->real_parent = current;
1216
1217         spin_lock(&current->sighand->siglock);
1218
1219         /*
1220          * Process group and session signals need to be delivered to just the
1221          * parent before the fork or both the parent and the child after the
1222          * fork. Restart if a signal comes in before we add the new process to
1223          * it's process group.
1224          * A fatal signal pending means that current will exit, so the new
1225          * thread can't slip out of an OOM kill (or normal SIGKILL).
1226          */
1227         recalc_sigpending();
1228         if (signal_pending(current)) {
1229                 spin_unlock(&current->sighand->siglock);
1230                 write_unlock_irq(&tasklist_lock);
1231                 retval = -ERESTARTNOINTR;
1232                 goto bad_fork_free_graph;
1233         }
1234
1235         if (clone_flags & CLONE_THREAD) {
1236                 p->group_leader = current->group_leader;
1237                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1238         }
1239
1240         if (likely(p->pid)) {
1241                 list_add_tail(&p->sibling, &p->real_parent->children);
1242                 tracehook_finish_clone(p, clone_flags, trace);
1243
1244                 if (thread_group_leader(p)) {
1245                         if (clone_flags & CLONE_NEWPID)
1246                                 p->nsproxy->pid_ns->child_reaper = p;
1247
1248                         p->signal->leader_pid = pid;
1249                         tty_kref_put(p->signal->tty);
1250                         p->signal->tty = tty_kref_get(current->signal->tty);
1251                         set_task_pgrp(p, task_pgrp_nr(current));
1252                         set_task_session(p, task_session_nr(current));
1253                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1254                         attach_pid(p, PIDTYPE_SID, task_session(current));
1255                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1256                         __get_cpu_var(process_counts)++;
1257                 }
1258                 attach_pid(p, PIDTYPE_PID, pid);
1259                 nr_threads++;
1260         }
1261
1262         total_forks++;
1263         spin_unlock(&current->sighand->siglock);
1264         write_unlock_irq(&tasklist_lock);
1265         proc_fork_connector(p);
1266         cgroup_post_fork(p);
1267         return p;
1268
1269 bad_fork_free_graph:
1270         ftrace_graph_exit_task(p);
1271 bad_fork_free_pid:
1272         if (pid != &init_struct_pid)
1273                 free_pid(pid);
1274 bad_fork_cleanup_io:
1275         put_io_context(p->io_context);
1276 bad_fork_cleanup_namespaces:
1277         exit_task_namespaces(p);
1278 bad_fork_cleanup_mm:
1279         if (p->mm)
1280                 mmput(p->mm);
1281 bad_fork_cleanup_signal:
1282         cleanup_signal(p);
1283 bad_fork_cleanup_sighand:
1284         __cleanup_sighand(p->sighand);
1285 bad_fork_cleanup_fs:
1286         exit_fs(p); /* blocking */
1287 bad_fork_cleanup_files:
1288         exit_files(p); /* blocking */
1289 bad_fork_cleanup_semundo:
1290         exit_sem(p);
1291 bad_fork_cleanup_audit:
1292         audit_free(p);
1293 bad_fork_cleanup_policy:
1294 #ifdef CONFIG_NUMA
1295         mpol_put(p->mempolicy);
1296 bad_fork_cleanup_cgroup:
1297 #endif
1298         cgroup_exit(p, cgroup_callbacks_done);
1299         delayacct_tsk_free(p);
1300         if (p->binfmt)
1301                 module_put(p->binfmt->module);
1302 bad_fork_cleanup_put_domain:
1303         module_put(task_thread_info(p)->exec_domain->module);
1304 bad_fork_cleanup_count:
1305         atomic_dec(&p->cred->user->processes);
1306         put_cred(p->real_cred);
1307         put_cred(p->cred);
1308 bad_fork_free:
1309         free_task(p);
1310 fork_out:
1311         return ERR_PTR(retval);
1312 }
1313
1314 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1315 {
1316         memset(regs, 0, sizeof(struct pt_regs));
1317         return regs;
1318 }
1319
1320 struct task_struct * __cpuinit fork_idle(int cpu)
1321 {
1322         struct task_struct *task;
1323         struct pt_regs regs;
1324
1325         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1326                             &init_struct_pid, 0);
1327         if (!IS_ERR(task))
1328                 init_idle(task, cpu);
1329
1330         return task;
1331 }
1332
1333 /*
1334  *  Ok, this is the main fork-routine.
1335  *
1336  * It copies the process, and if successful kick-starts
1337  * it and waits for it to finish using the VM if required.
1338  */
1339 long do_fork(unsigned long clone_flags,
1340               unsigned long stack_start,
1341               struct pt_regs *regs,
1342               unsigned long stack_size,
1343               int __user *parent_tidptr,
1344               int __user *child_tidptr)
1345 {
1346         struct task_struct *p;
1347         int trace = 0;
1348         long nr;
1349
1350         /*
1351          * Do some preliminary argument and permissions checking before we
1352          * actually start allocating stuff
1353          */
1354         if (clone_flags & CLONE_NEWUSER) {
1355                 if (clone_flags & CLONE_THREAD)
1356                         return -EINVAL;
1357                 /* hopefully this check will go away when userns support is
1358                  * complete
1359                  */
1360                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1361                                 !capable(CAP_SETGID))
1362                         return -EPERM;
1363         }
1364
1365         /*
1366          * We hope to recycle these flags after 2.6.26
1367          */
1368         if (unlikely(clone_flags & CLONE_STOPPED)) {
1369                 static int __read_mostly count = 100;
1370
1371                 if (count > 0 && printk_ratelimit()) {
1372                         char comm[TASK_COMM_LEN];
1373
1374                         count--;
1375                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1376                                         "clone flags 0x%lx\n",
1377                                 get_task_comm(comm, current),
1378                                 clone_flags & CLONE_STOPPED);
1379                 }
1380         }
1381
1382         /*
1383          * When called from kernel_thread, don't do user tracing stuff.
1384          */
1385         if (likely(user_mode(regs)))
1386                 trace = tracehook_prepare_clone(clone_flags);
1387
1388         p = copy_process(clone_flags, stack_start, regs, stack_size,
1389                          child_tidptr, NULL, trace);
1390         /*
1391          * Do this prior waking up the new thread - the thread pointer
1392          * might get invalid after that point, if the thread exits quickly.
1393          */
1394         if (!IS_ERR(p)) {
1395                 struct completion vfork;
1396
1397                 trace_sched_process_fork(current, p);
1398
1399                 nr = task_pid_vnr(p);
1400
1401                 if (clone_flags & CLONE_PARENT_SETTID)
1402                         put_user(nr, parent_tidptr);
1403
1404                 if (clone_flags & CLONE_VFORK) {
1405                         p->vfork_done = &vfork;
1406                         init_completion(&vfork);
1407                 }
1408
1409                 audit_finish_fork(p);
1410                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1411
1412                 /*
1413                  * We set PF_STARTING at creation in case tracing wants to
1414                  * use this to distinguish a fully live task from one that
1415                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1416                  * clear it and set the child going.
1417                  */
1418                 p->flags &= ~PF_STARTING;
1419
1420                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1421                         /*
1422                          * We'll start up with an immediate SIGSTOP.
1423                          */
1424                         sigaddset(&p->pending.signal, SIGSTOP);
1425                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1426                         __set_task_state(p, TASK_STOPPED);
1427                 } else {
1428                         wake_up_new_task(p, clone_flags);
1429                 }
1430
1431                 tracehook_report_clone_complete(trace, regs,
1432                                                 clone_flags, nr, p);
1433
1434                 if (clone_flags & CLONE_VFORK) {
1435                         freezer_do_not_count();
1436                         wait_for_completion(&vfork);
1437                         freezer_count();
1438                         tracehook_report_vfork_done(p, nr);
1439                 }
1440         } else {
1441                 nr = PTR_ERR(p);
1442         }
1443         return nr;
1444 }
1445
1446 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1447 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1448 #endif
1449
1450 static void sighand_ctor(void *data)
1451 {
1452         struct sighand_struct *sighand = data;
1453
1454         spin_lock_init(&sighand->siglock);
1455         init_waitqueue_head(&sighand->signalfd_wqh);
1456 }
1457
1458 void __init proc_caches_init(void)
1459 {
1460         sighand_cachep = kmem_cache_create("sighand_cache",
1461                         sizeof(struct sighand_struct), 0,
1462                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1463                         sighand_ctor);
1464         signal_cachep = kmem_cache_create("signal_cache",
1465                         sizeof(struct signal_struct), 0,
1466                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1467         files_cachep = kmem_cache_create("files_cache",
1468                         sizeof(struct files_struct), 0,
1469                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1470         fs_cachep = kmem_cache_create("fs_cache",
1471                         sizeof(struct fs_struct), 0,
1472                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1473         vm_area_cachep = kmem_cache_create("vm_area_struct",
1474                         sizeof(struct vm_area_struct), 0,
1475                         SLAB_PANIC, NULL);
1476         mm_cachep = kmem_cache_create("mm_struct",
1477                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1478                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1479 }
1480
1481 /*
1482  * Check constraints on flags passed to the unshare system call and
1483  * force unsharing of additional process context as appropriate.
1484  */
1485 static void check_unshare_flags(unsigned long *flags_ptr)
1486 {
1487         /*
1488          * If unsharing a thread from a thread group, must also
1489          * unshare vm.
1490          */
1491         if (*flags_ptr & CLONE_THREAD)
1492                 *flags_ptr |= CLONE_VM;
1493
1494         /*
1495          * If unsharing vm, must also unshare signal handlers.
1496          */
1497         if (*flags_ptr & CLONE_VM)
1498                 *flags_ptr |= CLONE_SIGHAND;
1499
1500         /*
1501          * If unsharing signal handlers and the task was created
1502          * using CLONE_THREAD, then must unshare the thread
1503          */
1504         if ((*flags_ptr & CLONE_SIGHAND) &&
1505             (atomic_read(&current->signal->count) > 1))
1506                 *flags_ptr |= CLONE_THREAD;
1507
1508         /*
1509          * If unsharing namespace, must also unshare filesystem information.
1510          */
1511         if (*flags_ptr & CLONE_NEWNS)
1512                 *flags_ptr |= CLONE_FS;
1513 }
1514
1515 /*
1516  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1517  */
1518 static int unshare_thread(unsigned long unshare_flags)
1519 {
1520         if (unshare_flags & CLONE_THREAD)
1521                 return -EINVAL;
1522
1523         return 0;
1524 }
1525
1526 /*
1527  * Unshare the filesystem structure if it is being shared
1528  */
1529 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1530 {
1531         struct fs_struct *fs = current->fs;
1532
1533         if ((unshare_flags & CLONE_FS) &&
1534             (fs && atomic_read(&fs->count) > 1)) {
1535                 *new_fsp = __copy_fs_struct(current->fs);
1536                 if (!*new_fsp)
1537                         return -ENOMEM;
1538         }
1539
1540         return 0;
1541 }
1542
1543 /*
1544  * Unsharing of sighand is not supported yet
1545  */
1546 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1547 {
1548         struct sighand_struct *sigh = current->sighand;
1549
1550         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1551                 return -EINVAL;
1552         else
1553                 return 0;
1554 }
1555
1556 /*
1557  * Unshare vm if it is being shared
1558  */
1559 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1560 {
1561         struct mm_struct *mm = current->mm;
1562
1563         if ((unshare_flags & CLONE_VM) &&
1564             (mm && atomic_read(&mm->mm_users) > 1)) {
1565                 return -EINVAL;
1566         }
1567
1568         return 0;
1569 }
1570
1571 /*
1572  * Unshare file descriptor table if it is being shared
1573  */
1574 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1575 {
1576         struct files_struct *fd = current->files;
1577         int error = 0;
1578
1579         if ((unshare_flags & CLONE_FILES) &&
1580             (fd && atomic_read(&fd->count) > 1)) {
1581                 *new_fdp = dup_fd(fd, &error);
1582                 if (!*new_fdp)
1583                         return error;
1584         }
1585
1586         return 0;
1587 }
1588
1589 /*
1590  * unshare allows a process to 'unshare' part of the process
1591  * context which was originally shared using clone.  copy_*
1592  * functions used by do_fork() cannot be used here directly
1593  * because they modify an inactive task_struct that is being
1594  * constructed. Here we are modifying the current, active,
1595  * task_struct.
1596  */
1597 asmlinkage long sys_unshare(unsigned long unshare_flags)
1598 {
1599         int err = 0;
1600         struct fs_struct *fs, *new_fs = NULL;
1601         struct sighand_struct *new_sigh = NULL;
1602         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1603         struct files_struct *fd, *new_fd = NULL;
1604         struct nsproxy *new_nsproxy = NULL;
1605         int do_sysvsem = 0;
1606
1607         check_unshare_flags(&unshare_flags);
1608
1609         /* Return -EINVAL for all unsupported flags */
1610         err = -EINVAL;
1611         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1612                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1613                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1614                 goto bad_unshare_out;
1615
1616         /*
1617          * CLONE_NEWIPC must also detach from the undolist: after switching
1618          * to a new ipc namespace, the semaphore arrays from the old
1619          * namespace are unreachable.
1620          */
1621         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1622                 do_sysvsem = 1;
1623         if ((err = unshare_thread(unshare_flags)))
1624                 goto bad_unshare_out;
1625         if ((err = unshare_fs(unshare_flags, &new_fs)))
1626                 goto bad_unshare_cleanup_thread;
1627         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1628                 goto bad_unshare_cleanup_fs;
1629         if ((err = unshare_vm(unshare_flags, &new_mm)))
1630                 goto bad_unshare_cleanup_sigh;
1631         if ((err = unshare_fd(unshare_flags, &new_fd)))
1632                 goto bad_unshare_cleanup_vm;
1633         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1634                         new_fs)))
1635                 goto bad_unshare_cleanup_fd;
1636
1637         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1638                 if (do_sysvsem) {
1639                         /*
1640                          * CLONE_SYSVSEM is equivalent to sys_exit().
1641                          */
1642                         exit_sem(current);
1643                 }
1644
1645                 if (new_nsproxy) {
1646                         switch_task_namespaces(current, new_nsproxy);
1647                         new_nsproxy = NULL;
1648                 }
1649
1650                 task_lock(current);
1651
1652                 if (new_fs) {
1653                         fs = current->fs;
1654                         current->fs = new_fs;
1655                         new_fs = fs;
1656                 }
1657
1658                 if (new_mm) {
1659                         mm = current->mm;
1660                         active_mm = current->active_mm;
1661                         current->mm = new_mm;
1662                         current->active_mm = new_mm;
1663                         activate_mm(active_mm, new_mm);
1664                         new_mm = mm;
1665                 }
1666
1667                 if (new_fd) {
1668                         fd = current->files;
1669                         current->files = new_fd;
1670                         new_fd = fd;
1671                 }
1672
1673                 task_unlock(current);
1674         }
1675
1676         if (new_nsproxy)
1677                 put_nsproxy(new_nsproxy);
1678
1679 bad_unshare_cleanup_fd:
1680         if (new_fd)
1681                 put_files_struct(new_fd);
1682
1683 bad_unshare_cleanup_vm:
1684         if (new_mm)
1685                 mmput(new_mm);
1686
1687 bad_unshare_cleanup_sigh:
1688         if (new_sigh)
1689                 if (atomic_dec_and_test(&new_sigh->count))
1690                         kmem_cache_free(sighand_cachep, new_sigh);
1691
1692 bad_unshare_cleanup_fs:
1693         if (new_fs)
1694                 put_fs_struct(new_fs);
1695
1696 bad_unshare_cleanup_thread:
1697 bad_unshare_out:
1698         return err;
1699 }
1700
1701 /*
1702  *      Helper to unshare the files of the current task.
1703  *      We don't want to expose copy_files internals to
1704  *      the exec layer of the kernel.
1705  */
1706
1707 int unshare_files(struct files_struct **displaced)
1708 {
1709         struct task_struct *task = current;
1710         struct files_struct *copy = NULL;
1711         int error;
1712
1713         error = unshare_fd(CLONE_FILES, &copy);
1714         if (error || !copy) {
1715                 *displaced = NULL;
1716                 return error;
1717         }
1718         *displaced = task->files;
1719         task_lock(task);
1720         task->files = copy;
1721         task_unlock(task);
1722         return 0;
1723 }