proc: turn signal_struct->count into "int nr_threads"
[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/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/tracehook.h>
41 #include <linux/futex.h>
42 #include <linux/compat.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/ftrace.h>
50 #include <linux/profile.h>
51 #include <linux/rmap.h>
52 #include <linux/ksm.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 <linux/fs_struct.h>
64 #include <linux/magic.h>
65 #include <linux/perf_event.h>
66 #include <linux/posix-timers.h>
67 #include <linux/user-return-notifier.h>
68
69 #include <asm/pgtable.h>
70 #include <asm/pgalloc.h>
71 #include <asm/uaccess.h>
72 #include <asm/mmu_context.h>
73 #include <asm/cacheflush.h>
74 #include <asm/tlbflush.h>
75
76 #include <trace/events/sched.h>
77
78 /*
79  * Protected counters by write_lock_irq(&tasklist_lock)
80  */
81 unsigned long total_forks;      /* Handle normal Linux uptimes. */
82 int nr_threads;                 /* The idle threads do not count.. */
83
84 int max_threads;                /* tunable limit on nr_threads */
85
86 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
87
88 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
89
90 #ifdef CONFIG_PROVE_RCU
91 int lockdep_tasklist_lock_is_held(void)
92 {
93         return lockdep_is_held(&tasklist_lock);
94 }
95 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
96 #endif /* #ifdef CONFIG_PROVE_RCU */
97
98 int nr_processes(void)
99 {
100         int cpu;
101         int total = 0;
102
103         for_each_possible_cpu(cpu)
104                 total += per_cpu(process_counts, cpu);
105
106         return total;
107 }
108
109 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
110 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
111 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
112 static struct kmem_cache *task_struct_cachep;
113 #endif
114
115 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
116 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
117 {
118 #ifdef CONFIG_DEBUG_STACK_USAGE
119         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
120 #else
121         gfp_t mask = GFP_KERNEL;
122 #endif
123         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
124 }
125
126 static inline void free_thread_info(struct thread_info *ti)
127 {
128         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
129 }
130 #endif
131
132 /* SLAB cache for signal_struct structures (tsk->signal) */
133 static struct kmem_cache *signal_cachep;
134
135 /* SLAB cache for sighand_struct structures (tsk->sighand) */
136 struct kmem_cache *sighand_cachep;
137
138 /* SLAB cache for files_struct structures (tsk->files) */
139 struct kmem_cache *files_cachep;
140
141 /* SLAB cache for fs_struct structures (tsk->fs) */
142 struct kmem_cache *fs_cachep;
143
144 /* SLAB cache for vm_area_struct structures */
145 struct kmem_cache *vm_area_cachep;
146
147 /* SLAB cache for mm_struct structures (tsk->mm) */
148 static struct kmem_cache *mm_cachep;
149
150 static void account_kernel_stack(struct thread_info *ti, int account)
151 {
152         struct zone *zone = page_zone(virt_to_page(ti));
153
154         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
155 }
156
157 void free_task(struct task_struct *tsk)
158 {
159         prop_local_destroy_single(&tsk->dirties);
160         account_kernel_stack(tsk->stack, -1);
161         free_thread_info(tsk->stack);
162         rt_mutex_debug_task_free(tsk);
163         ftrace_graph_exit_task(tsk);
164         free_task_struct(tsk);
165 }
166 EXPORT_SYMBOL(free_task);
167
168 static inline void free_signal_struct(struct signal_struct *sig)
169 {
170         taskstats_tgid_free(sig);
171         kmem_cache_free(signal_cachep, sig);
172 }
173
174 static inline void put_signal_struct(struct signal_struct *sig)
175 {
176         if (atomic_dec_and_test(&sig->sigcnt))
177                 free_signal_struct(sig);
178 }
179
180 void __put_task_struct(struct task_struct *tsk)
181 {
182         WARN_ON(!tsk->exit_state);
183         WARN_ON(atomic_read(&tsk->usage));
184         WARN_ON(tsk == current);
185
186         exit_creds(tsk);
187         delayacct_tsk_free(tsk);
188         put_signal_struct(tsk->signal);
189
190         if (!profile_handoff_task(tsk))
191                 free_task(tsk);
192 }
193
194 /*
195  * macro override instead of weak attribute alias, to workaround
196  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
197  */
198 #ifndef arch_task_cache_init
199 #define arch_task_cache_init()
200 #endif
201
202 void __init fork_init(unsigned long mempages)
203 {
204 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
205 #ifndef ARCH_MIN_TASKALIGN
206 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
207 #endif
208         /* create a slab on which task_structs can be allocated */
209         task_struct_cachep =
210                 kmem_cache_create("task_struct", sizeof(struct task_struct),
211                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
212 #endif
213
214         /* do the arch specific task caches init */
215         arch_task_cache_init();
216
217         /*
218          * The default maximum number of threads is set to a safe
219          * value: the thread structures can take up at most half
220          * of memory.
221          */
222         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
223
224         /*
225          * we need to allow at least 20 threads to boot a system
226          */
227         if(max_threads < 20)
228                 max_threads = 20;
229
230         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
231         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
232         init_task.signal->rlim[RLIMIT_SIGPENDING] =
233                 init_task.signal->rlim[RLIMIT_NPROC];
234 }
235
236 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
237                                                struct task_struct *src)
238 {
239         *dst = *src;
240         return 0;
241 }
242
243 static struct task_struct *dup_task_struct(struct task_struct *orig)
244 {
245         struct task_struct *tsk;
246         struct thread_info *ti;
247         unsigned long *stackend;
248
249         int err;
250
251         prepare_to_copy(orig);
252
253         tsk = alloc_task_struct();
254         if (!tsk)
255                 return NULL;
256
257         ti = alloc_thread_info(tsk);
258         if (!ti) {
259                 free_task_struct(tsk);
260                 return NULL;
261         }
262
263         err = arch_dup_task_struct(tsk, orig);
264         if (err)
265                 goto out;
266
267         tsk->stack = ti;
268
269         err = prop_local_init_single(&tsk->dirties);
270         if (err)
271                 goto out;
272
273         setup_thread_stack(tsk, orig);
274         clear_user_return_notifier(tsk);
275         stackend = end_of_stack(tsk);
276         *stackend = STACK_END_MAGIC;    /* for overflow detection */
277
278 #ifdef CONFIG_CC_STACKPROTECTOR
279         tsk->stack_canary = get_random_int();
280 #endif
281
282         /* One for us, one for whoever does the "release_task()" (usually parent) */
283         atomic_set(&tsk->usage,2);
284         atomic_set(&tsk->fs_excl, 0);
285 #ifdef CONFIG_BLK_DEV_IO_TRACE
286         tsk->btrace_seq = 0;
287 #endif
288         tsk->splice_pipe = NULL;
289
290         account_kernel_stack(ti, 1);
291
292         return tsk;
293
294 out:
295         free_thread_info(ti);
296         free_task_struct(tsk);
297         return NULL;
298 }
299
300 #ifdef CONFIG_MMU
301 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
302 {
303         struct vm_area_struct *mpnt, *tmp, **pprev;
304         struct rb_node **rb_link, *rb_parent;
305         int retval;
306         unsigned long charge;
307         struct mempolicy *pol;
308
309         down_write(&oldmm->mmap_sem);
310         flush_cache_dup_mm(oldmm);
311         /*
312          * Not linked in yet - no deadlock potential:
313          */
314         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
315
316         mm->locked_vm = 0;
317         mm->mmap = NULL;
318         mm->mmap_cache = NULL;
319         mm->free_area_cache = oldmm->mmap_base;
320         mm->cached_hole_size = ~0UL;
321         mm->map_count = 0;
322         cpumask_clear(mm_cpumask(mm));
323         mm->mm_rb = RB_ROOT;
324         rb_link = &mm->mm_rb.rb_node;
325         rb_parent = NULL;
326         pprev = &mm->mmap;
327         retval = ksm_fork(mm, oldmm);
328         if (retval)
329                 goto out;
330
331         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
332                 struct file *file;
333
334                 if (mpnt->vm_flags & VM_DONTCOPY) {
335                         long pages = vma_pages(mpnt);
336                         mm->total_vm -= pages;
337                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
338                                                                 -pages);
339                         continue;
340                 }
341                 charge = 0;
342                 if (mpnt->vm_flags & VM_ACCOUNT) {
343                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
344                         if (security_vm_enough_memory(len))
345                                 goto fail_nomem;
346                         charge = len;
347                 }
348                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
349                 if (!tmp)
350                         goto fail_nomem;
351                 *tmp = *mpnt;
352                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
353                 pol = mpol_dup(vma_policy(mpnt));
354                 retval = PTR_ERR(pol);
355                 if (IS_ERR(pol))
356                         goto fail_nomem_policy;
357                 vma_set_policy(tmp, pol);
358                 if (anon_vma_fork(tmp, mpnt))
359                         goto fail_nomem_anon_vma_fork;
360                 tmp->vm_flags &= ~VM_LOCKED;
361                 tmp->vm_mm = mm;
362                 tmp->vm_next = NULL;
363                 file = tmp->vm_file;
364                 if (file) {
365                         struct inode *inode = file->f_path.dentry->d_inode;
366                         struct address_space *mapping = file->f_mapping;
367
368                         get_file(file);
369                         if (tmp->vm_flags & VM_DENYWRITE)
370                                 atomic_dec(&inode->i_writecount);
371                         spin_lock(&mapping->i_mmap_lock);
372                         if (tmp->vm_flags & VM_SHARED)
373                                 mapping->i_mmap_writable++;
374                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
375                         flush_dcache_mmap_lock(mapping);
376                         /* insert tmp into the share list, just after mpnt */
377                         vma_prio_tree_add(tmp, mpnt);
378                         flush_dcache_mmap_unlock(mapping);
379                         spin_unlock(&mapping->i_mmap_lock);
380                 }
381
382                 /*
383                  * Clear hugetlb-related page reserves for children. This only
384                  * affects MAP_PRIVATE mappings. Faults generated by the child
385                  * are not guaranteed to succeed, even if read-only
386                  */
387                 if (is_vm_hugetlb_page(tmp))
388                         reset_vma_resv_huge_pages(tmp);
389
390                 /*
391                  * Link in the new vma and copy the page table entries.
392                  */
393                 *pprev = tmp;
394                 pprev = &tmp->vm_next;
395
396                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
397                 rb_link = &tmp->vm_rb.rb_right;
398                 rb_parent = &tmp->vm_rb;
399
400                 mm->map_count++;
401                 retval = copy_page_range(mm, oldmm, mpnt);
402
403                 if (tmp->vm_ops && tmp->vm_ops->open)
404                         tmp->vm_ops->open(tmp);
405
406                 if (retval)
407                         goto out;
408         }
409         /* a new mm has just been created */
410         arch_dup_mmap(oldmm, mm);
411         retval = 0;
412 out:
413         up_write(&mm->mmap_sem);
414         flush_tlb_mm(oldmm);
415         up_write(&oldmm->mmap_sem);
416         return retval;
417 fail_nomem_anon_vma_fork:
418         mpol_put(pol);
419 fail_nomem_policy:
420         kmem_cache_free(vm_area_cachep, tmp);
421 fail_nomem:
422         retval = -ENOMEM;
423         vm_unacct_memory(charge);
424         goto out;
425 }
426
427 static inline int mm_alloc_pgd(struct mm_struct * mm)
428 {
429         mm->pgd = pgd_alloc(mm);
430         if (unlikely(!mm->pgd))
431                 return -ENOMEM;
432         return 0;
433 }
434
435 static inline void mm_free_pgd(struct mm_struct * mm)
436 {
437         pgd_free(mm, mm->pgd);
438 }
439 #else
440 #define dup_mmap(mm, oldmm)     (0)
441 #define mm_alloc_pgd(mm)        (0)
442 #define mm_free_pgd(mm)
443 #endif /* CONFIG_MMU */
444
445 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
446
447 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
448 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
449
450 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
451
452 static int __init coredump_filter_setup(char *s)
453 {
454         default_dump_filter =
455                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
456                 MMF_DUMP_FILTER_MASK;
457         return 1;
458 }
459
460 __setup("coredump_filter=", coredump_filter_setup);
461
462 #include <linux/init_task.h>
463
464 static void mm_init_aio(struct mm_struct *mm)
465 {
466 #ifdef CONFIG_AIO
467         spin_lock_init(&mm->ioctx_lock);
468         INIT_HLIST_HEAD(&mm->ioctx_list);
469 #endif
470 }
471
472 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
473 {
474         atomic_set(&mm->mm_users, 1);
475         atomic_set(&mm->mm_count, 1);
476         init_rwsem(&mm->mmap_sem);
477         INIT_LIST_HEAD(&mm->mmlist);
478         mm->flags = (current->mm) ?
479                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
480         mm->core_state = NULL;
481         mm->nr_ptes = 0;
482         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
483         spin_lock_init(&mm->page_table_lock);
484         mm->free_area_cache = TASK_UNMAPPED_BASE;
485         mm->cached_hole_size = ~0UL;
486         mm_init_aio(mm);
487         mm_init_owner(mm, p);
488
489         if (likely(!mm_alloc_pgd(mm))) {
490                 mm->def_flags = 0;
491                 mmu_notifier_mm_init(mm);
492                 return mm;
493         }
494
495         free_mm(mm);
496         return NULL;
497 }
498
499 /*
500  * Allocate and initialize an mm_struct.
501  */
502 struct mm_struct * mm_alloc(void)
503 {
504         struct mm_struct * mm;
505
506         mm = allocate_mm();
507         if (mm) {
508                 memset(mm, 0, sizeof(*mm));
509                 mm = mm_init(mm, current);
510         }
511         return mm;
512 }
513
514 /*
515  * Called when the last reference to the mm
516  * is dropped: either by a lazy thread or by
517  * mmput. Free the page directory and the mm.
518  */
519 void __mmdrop(struct mm_struct *mm)
520 {
521         BUG_ON(mm == &init_mm);
522         mm_free_pgd(mm);
523         destroy_context(mm);
524         mmu_notifier_mm_destroy(mm);
525         free_mm(mm);
526 }
527 EXPORT_SYMBOL_GPL(__mmdrop);
528
529 /*
530  * Decrement the use count and release all resources for an mm.
531  */
532 void mmput(struct mm_struct *mm)
533 {
534         might_sleep();
535
536         if (atomic_dec_and_test(&mm->mm_users)) {
537                 exit_aio(mm);
538                 ksm_exit(mm);
539                 exit_mmap(mm);
540                 set_mm_exe_file(mm, NULL);
541                 if (!list_empty(&mm->mmlist)) {
542                         spin_lock(&mmlist_lock);
543                         list_del(&mm->mmlist);
544                         spin_unlock(&mmlist_lock);
545                 }
546                 put_swap_token(mm);
547                 if (mm->binfmt)
548                         module_put(mm->binfmt->module);
549                 mmdrop(mm);
550         }
551 }
552 EXPORT_SYMBOL_GPL(mmput);
553
554 /**
555  * get_task_mm - acquire a reference to the task's mm
556  *
557  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
558  * this kernel workthread has transiently adopted a user mm with use_mm,
559  * to do its AIO) is not set and if so returns a reference to it, after
560  * bumping up the use count.  User must release the mm via mmput()
561  * after use.  Typically used by /proc and ptrace.
562  */
563 struct mm_struct *get_task_mm(struct task_struct *task)
564 {
565         struct mm_struct *mm;
566
567         task_lock(task);
568         mm = task->mm;
569         if (mm) {
570                 if (task->flags & PF_KTHREAD)
571                         mm = NULL;
572                 else
573                         atomic_inc(&mm->mm_users);
574         }
575         task_unlock(task);
576         return mm;
577 }
578 EXPORT_SYMBOL_GPL(get_task_mm);
579
580 /* Please note the differences between mmput and mm_release.
581  * mmput is called whenever we stop holding onto a mm_struct,
582  * error success whatever.
583  *
584  * mm_release is called after a mm_struct has been removed
585  * from the current process.
586  *
587  * This difference is important for error handling, when we
588  * only half set up a mm_struct for a new process and need to restore
589  * the old one.  Because we mmput the new mm_struct before
590  * restoring the old one. . .
591  * Eric Biederman 10 January 1998
592  */
593 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
594 {
595         struct completion *vfork_done = tsk->vfork_done;
596
597         /* Get rid of any futexes when releasing the mm */
598 #ifdef CONFIG_FUTEX
599         if (unlikely(tsk->robust_list)) {
600                 exit_robust_list(tsk);
601                 tsk->robust_list = NULL;
602         }
603 #ifdef CONFIG_COMPAT
604         if (unlikely(tsk->compat_robust_list)) {
605                 compat_exit_robust_list(tsk);
606                 tsk->compat_robust_list = NULL;
607         }
608 #endif
609         if (unlikely(!list_empty(&tsk->pi_state_list)))
610                 exit_pi_state_list(tsk);
611 #endif
612
613         /* Get rid of any cached register state */
614         deactivate_mm(tsk, mm);
615
616         /* notify parent sleeping on vfork() */
617         if (vfork_done) {
618                 tsk->vfork_done = NULL;
619                 complete(vfork_done);
620         }
621
622         /*
623          * If we're exiting normally, clear a user-space tid field if
624          * requested.  We leave this alone when dying by signal, to leave
625          * the value intact in a core dump, and to save the unnecessary
626          * trouble otherwise.  Userland only wants this done for a sys_exit.
627          */
628         if (tsk->clear_child_tid) {
629                 if (!(tsk->flags & PF_SIGNALED) &&
630                     atomic_read(&mm->mm_users) > 1) {
631                         /*
632                          * We don't check the error code - if userspace has
633                          * not set up a proper pointer then tough luck.
634                          */
635                         put_user(0, tsk->clear_child_tid);
636                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
637                                         1, NULL, NULL, 0);
638                 }
639                 tsk->clear_child_tid = NULL;
640         }
641 }
642
643 /*
644  * Allocate a new mm structure and copy contents from the
645  * mm structure of the passed in task structure.
646  */
647 struct mm_struct *dup_mm(struct task_struct *tsk)
648 {
649         struct mm_struct *mm, *oldmm = current->mm;
650         int err;
651
652         if (!oldmm)
653                 return NULL;
654
655         mm = allocate_mm();
656         if (!mm)
657                 goto fail_nomem;
658
659         memcpy(mm, oldmm, sizeof(*mm));
660
661         /* Initializing for Swap token stuff */
662         mm->token_priority = 0;
663         mm->last_interval = 0;
664
665         if (!mm_init(mm, tsk))
666                 goto fail_nomem;
667
668         if (init_new_context(tsk, mm))
669                 goto fail_nocontext;
670
671         dup_mm_exe_file(oldmm, mm);
672
673         err = dup_mmap(mm, oldmm);
674         if (err)
675                 goto free_pt;
676
677         mm->hiwater_rss = get_mm_rss(mm);
678         mm->hiwater_vm = mm->total_vm;
679
680         if (mm->binfmt && !try_module_get(mm->binfmt->module))
681                 goto free_pt;
682
683         return mm;
684
685 free_pt:
686         /* don't put binfmt in mmput, we haven't got module yet */
687         mm->binfmt = NULL;
688         mmput(mm);
689
690 fail_nomem:
691         return NULL;
692
693 fail_nocontext:
694         /*
695          * If init_new_context() failed, we cannot use mmput() to free the mm
696          * because it calls destroy_context()
697          */
698         mm_free_pgd(mm);
699         free_mm(mm);
700         return NULL;
701 }
702
703 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
704 {
705         struct mm_struct * mm, *oldmm;
706         int retval;
707
708         tsk->min_flt = tsk->maj_flt = 0;
709         tsk->nvcsw = tsk->nivcsw = 0;
710 #ifdef CONFIG_DETECT_HUNG_TASK
711         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
712 #endif
713
714         tsk->mm = NULL;
715         tsk->active_mm = NULL;
716
717         /*
718          * Are we cloning a kernel thread?
719          *
720          * We need to steal a active VM for that..
721          */
722         oldmm = current->mm;
723         if (!oldmm)
724                 return 0;
725
726         if (clone_flags & CLONE_VM) {
727                 atomic_inc(&oldmm->mm_users);
728                 mm = oldmm;
729                 goto good_mm;
730         }
731
732         retval = -ENOMEM;
733         mm = dup_mm(tsk);
734         if (!mm)
735                 goto fail_nomem;
736
737 good_mm:
738         /* Initializing for Swap token stuff */
739         mm->token_priority = 0;
740         mm->last_interval = 0;
741
742         tsk->mm = mm;
743         tsk->active_mm = mm;
744         return 0;
745
746 fail_nomem:
747         return retval;
748 }
749
750 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
751 {
752         struct fs_struct *fs = current->fs;
753         if (clone_flags & CLONE_FS) {
754                 /* tsk->fs is already what we want */
755                 write_lock(&fs->lock);
756                 if (fs->in_exec) {
757                         write_unlock(&fs->lock);
758                         return -EAGAIN;
759                 }
760                 fs->users++;
761                 write_unlock(&fs->lock);
762                 return 0;
763         }
764         tsk->fs = copy_fs_struct(fs);
765         if (!tsk->fs)
766                 return -ENOMEM;
767         return 0;
768 }
769
770 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
771 {
772         struct files_struct *oldf, *newf;
773         int error = 0;
774
775         /*
776          * A background process may not have any files ...
777          */
778         oldf = current->files;
779         if (!oldf)
780                 goto out;
781
782         if (clone_flags & CLONE_FILES) {
783                 atomic_inc(&oldf->count);
784                 goto out;
785         }
786
787         newf = dup_fd(oldf, &error);
788         if (!newf)
789                 goto out;
790
791         tsk->files = newf;
792         error = 0;
793 out:
794         return error;
795 }
796
797 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
798 {
799 #ifdef CONFIG_BLOCK
800         struct io_context *ioc = current->io_context;
801
802         if (!ioc)
803                 return 0;
804         /*
805          * Share io context with parent, if CLONE_IO is set
806          */
807         if (clone_flags & CLONE_IO) {
808                 tsk->io_context = ioc_task_link(ioc);
809                 if (unlikely(!tsk->io_context))
810                         return -ENOMEM;
811         } else if (ioprio_valid(ioc->ioprio)) {
812                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
813                 if (unlikely(!tsk->io_context))
814                         return -ENOMEM;
815
816                 tsk->io_context->ioprio = ioc->ioprio;
817         }
818 #endif
819         return 0;
820 }
821
822 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
823 {
824         struct sighand_struct *sig;
825
826         if (clone_flags & CLONE_SIGHAND) {
827                 atomic_inc(&current->sighand->count);
828                 return 0;
829         }
830         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
831         rcu_assign_pointer(tsk->sighand, sig);
832         if (!sig)
833                 return -ENOMEM;
834         atomic_set(&sig->count, 1);
835         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
836         return 0;
837 }
838
839 void __cleanup_sighand(struct sighand_struct *sighand)
840 {
841         if (atomic_dec_and_test(&sighand->count))
842                 kmem_cache_free(sighand_cachep, sighand);
843 }
844
845
846 /*
847  * Initialize POSIX timer handling for a thread group.
848  */
849 static void posix_cpu_timers_init_group(struct signal_struct *sig)
850 {
851         unsigned long cpu_limit;
852
853         /* Thread group counters. */
854         thread_group_cputime_init(sig);
855
856         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
857         if (cpu_limit != RLIM_INFINITY) {
858                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
859                 sig->cputimer.running = 1;
860         }
861
862         /* The timer lists. */
863         INIT_LIST_HEAD(&sig->cpu_timers[0]);
864         INIT_LIST_HEAD(&sig->cpu_timers[1]);
865         INIT_LIST_HEAD(&sig->cpu_timers[2]);
866 }
867
868 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
869 {
870         struct signal_struct *sig;
871
872         if (clone_flags & CLONE_THREAD)
873                 return 0;
874
875         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
876         tsk->signal = sig;
877         if (!sig)
878                 return -ENOMEM;
879
880         sig->nr_threads = 1;
881         atomic_set(&sig->live, 1);
882         atomic_set(&sig->sigcnt, 1);
883         init_waitqueue_head(&sig->wait_chldexit);
884         if (clone_flags & CLONE_NEWPID)
885                 sig->flags |= SIGNAL_UNKILLABLE;
886         sig->curr_target = tsk;
887         init_sigpending(&sig->shared_pending);
888         INIT_LIST_HEAD(&sig->posix_timers);
889
890         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
891         sig->real_timer.function = it_real_fn;
892
893         task_lock(current->group_leader);
894         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
895         task_unlock(current->group_leader);
896
897         posix_cpu_timers_init_group(sig);
898
899         tty_audit_fork(sig);
900
901         sig->oom_adj = current->signal->oom_adj;
902
903         return 0;
904 }
905
906 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
907 {
908         unsigned long new_flags = p->flags;
909
910         new_flags &= ~PF_SUPERPRIV;
911         new_flags |= PF_FORKNOEXEC;
912         new_flags |= PF_STARTING;
913         p->flags = new_flags;
914         clear_freeze_flag(p);
915 }
916
917 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
918 {
919         current->clear_child_tid = tidptr;
920
921         return task_pid_vnr(current);
922 }
923
924 static void rt_mutex_init_task(struct task_struct *p)
925 {
926         raw_spin_lock_init(&p->pi_lock);
927 #ifdef CONFIG_RT_MUTEXES
928         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
929         p->pi_blocked_on = NULL;
930 #endif
931 }
932
933 #ifdef CONFIG_MM_OWNER
934 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
935 {
936         mm->owner = p;
937 }
938 #endif /* CONFIG_MM_OWNER */
939
940 /*
941  * Initialize POSIX timer handling for a single task.
942  */
943 static void posix_cpu_timers_init(struct task_struct *tsk)
944 {
945         tsk->cputime_expires.prof_exp = cputime_zero;
946         tsk->cputime_expires.virt_exp = cputime_zero;
947         tsk->cputime_expires.sched_exp = 0;
948         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
949         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
950         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
951 }
952
953 /*
954  * This creates a new process as a copy of the old one,
955  * but does not actually start it yet.
956  *
957  * It copies the registers, and all the appropriate
958  * parts of the process environment (as per the clone
959  * flags). The actual kick-off is left to the caller.
960  */
961 static struct task_struct *copy_process(unsigned long clone_flags,
962                                         unsigned long stack_start,
963                                         struct pt_regs *regs,
964                                         unsigned long stack_size,
965                                         int __user *child_tidptr,
966                                         struct pid *pid,
967                                         int trace)
968 {
969         int retval;
970         struct task_struct *p;
971         int cgroup_callbacks_done = 0;
972
973         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
974                 return ERR_PTR(-EINVAL);
975
976         /*
977          * Thread groups must share signals as well, and detached threads
978          * can only be started up within the thread group.
979          */
980         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
981                 return ERR_PTR(-EINVAL);
982
983         /*
984          * Shared signal handlers imply shared VM. By way of the above,
985          * thread groups also imply shared VM. Blocking this case allows
986          * for various simplifications in other code.
987          */
988         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
989                 return ERR_PTR(-EINVAL);
990
991         /*
992          * Siblings of global init remain as zombies on exit since they are
993          * not reaped by their parent (swapper). To solve this and to avoid
994          * multi-rooted process trees, prevent global and container-inits
995          * from creating siblings.
996          */
997         if ((clone_flags & CLONE_PARENT) &&
998                                 current->signal->flags & SIGNAL_UNKILLABLE)
999                 return ERR_PTR(-EINVAL);
1000
1001         retval = security_task_create(clone_flags);
1002         if (retval)
1003                 goto fork_out;
1004
1005         retval = -ENOMEM;
1006         p = dup_task_struct(current);
1007         if (!p)
1008                 goto fork_out;
1009
1010         ftrace_graph_init_task(p);
1011
1012         rt_mutex_init_task(p);
1013
1014 #ifdef CONFIG_PROVE_LOCKING
1015         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1016         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1017 #endif
1018         retval = -EAGAIN;
1019         if (atomic_read(&p->real_cred->user->processes) >=
1020                         task_rlimit(p, RLIMIT_NPROC)) {
1021                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1022                     p->real_cred->user != INIT_USER)
1023                         goto bad_fork_free;
1024         }
1025
1026         retval = copy_creds(p, clone_flags);
1027         if (retval < 0)
1028                 goto bad_fork_free;
1029
1030         /*
1031          * If multiple threads are within copy_process(), then this check
1032          * triggers too late. This doesn't hurt, the check is only there
1033          * to stop root fork bombs.
1034          */
1035         retval = -EAGAIN;
1036         if (nr_threads >= max_threads)
1037                 goto bad_fork_cleanup_count;
1038
1039         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1040                 goto bad_fork_cleanup_count;
1041
1042         p->did_exec = 0;
1043         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1044         copy_flags(clone_flags, p);
1045         INIT_LIST_HEAD(&p->children);
1046         INIT_LIST_HEAD(&p->sibling);
1047         rcu_copy_process(p);
1048         p->vfork_done = NULL;
1049         spin_lock_init(&p->alloc_lock);
1050
1051         init_sigpending(&p->pending);
1052
1053         p->utime = cputime_zero;
1054         p->stime = cputime_zero;
1055         p->gtime = cputime_zero;
1056         p->utimescaled = cputime_zero;
1057         p->stimescaled = cputime_zero;
1058 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1059         p->prev_utime = cputime_zero;
1060         p->prev_stime = cputime_zero;
1061 #endif
1062 #if defined(SPLIT_RSS_COUNTING)
1063         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1064 #endif
1065
1066         p->default_timer_slack_ns = current->timer_slack_ns;
1067
1068         task_io_accounting_init(&p->ioac);
1069         acct_clear_integrals(p);
1070
1071         posix_cpu_timers_init(p);
1072
1073         p->lock_depth = -1;             /* -1 = no lock */
1074         do_posix_clock_monotonic_gettime(&p->start_time);
1075         p->real_start_time = p->start_time;
1076         monotonic_to_bootbased(&p->real_start_time);
1077         p->io_context = NULL;
1078         p->audit_context = NULL;
1079         cgroup_fork(p);
1080 #ifdef CONFIG_NUMA
1081         p->mempolicy = mpol_dup(p->mempolicy);
1082         if (IS_ERR(p->mempolicy)) {
1083                 retval = PTR_ERR(p->mempolicy);
1084                 p->mempolicy = NULL;
1085                 goto bad_fork_cleanup_cgroup;
1086         }
1087         mpol_fix_fork_child_flag(p);
1088 #endif
1089 #ifdef CONFIG_CPUSETS
1090         p->cpuset_mem_spread_rotor = node_random(p->mems_allowed);
1091         p->cpuset_slab_spread_rotor = node_random(p->mems_allowed);
1092 #endif
1093 #ifdef CONFIG_TRACE_IRQFLAGS
1094         p->irq_events = 0;
1095 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1096         p->hardirqs_enabled = 1;
1097 #else
1098         p->hardirqs_enabled = 0;
1099 #endif
1100         p->hardirq_enable_ip = 0;
1101         p->hardirq_enable_event = 0;
1102         p->hardirq_disable_ip = _THIS_IP_;
1103         p->hardirq_disable_event = 0;
1104         p->softirqs_enabled = 1;
1105         p->softirq_enable_ip = _THIS_IP_;
1106         p->softirq_enable_event = 0;
1107         p->softirq_disable_ip = 0;
1108         p->softirq_disable_event = 0;
1109         p->hardirq_context = 0;
1110         p->softirq_context = 0;
1111 #endif
1112 #ifdef CONFIG_LOCKDEP
1113         p->lockdep_depth = 0; /* no locks held yet */
1114         p->curr_chain_key = 0;
1115         p->lockdep_recursion = 0;
1116 #endif
1117
1118 #ifdef CONFIG_DEBUG_MUTEXES
1119         p->blocked_on = NULL; /* not blocked yet */
1120 #endif
1121 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1122         p->memcg_batch.do_batch = 0;
1123         p->memcg_batch.memcg = NULL;
1124 #endif
1125
1126         /* Perform scheduler related setup. Assign this task to a CPU. */
1127         sched_fork(p, clone_flags);
1128
1129         retval = perf_event_init_task(p);
1130         if (retval)
1131                 goto bad_fork_cleanup_policy;
1132
1133         if ((retval = audit_alloc(p)))
1134                 goto bad_fork_cleanup_policy;
1135         /* copy all the process information */
1136         if ((retval = copy_semundo(clone_flags, p)))
1137                 goto bad_fork_cleanup_audit;
1138         if ((retval = copy_files(clone_flags, p)))
1139                 goto bad_fork_cleanup_semundo;
1140         if ((retval = copy_fs(clone_flags, p)))
1141                 goto bad_fork_cleanup_files;
1142         if ((retval = copy_sighand(clone_flags, p)))
1143                 goto bad_fork_cleanup_fs;
1144         if ((retval = copy_signal(clone_flags, p)))
1145                 goto bad_fork_cleanup_sighand;
1146         if ((retval = copy_mm(clone_flags, p)))
1147                 goto bad_fork_cleanup_signal;
1148         if ((retval = copy_namespaces(clone_flags, p)))
1149                 goto bad_fork_cleanup_mm;
1150         if ((retval = copy_io(clone_flags, p)))
1151                 goto bad_fork_cleanup_namespaces;
1152         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1153         if (retval)
1154                 goto bad_fork_cleanup_io;
1155
1156         if (pid != &init_struct_pid) {
1157                 retval = -ENOMEM;
1158                 pid = alloc_pid(p->nsproxy->pid_ns);
1159                 if (!pid)
1160                         goto bad_fork_cleanup_io;
1161
1162                 if (clone_flags & CLONE_NEWPID) {
1163                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1164                         if (retval < 0)
1165                                 goto bad_fork_free_pid;
1166                 }
1167         }
1168
1169         p->pid = pid_nr(pid);
1170         p->tgid = p->pid;
1171         if (clone_flags & CLONE_THREAD)
1172                 p->tgid = current->tgid;
1173
1174         if (current->nsproxy != p->nsproxy) {
1175                 retval = ns_cgroup_clone(p, pid);
1176                 if (retval)
1177                         goto bad_fork_free_pid;
1178         }
1179
1180         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1181         /*
1182          * Clear TID on mm_release()?
1183          */
1184         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1185 #ifdef CONFIG_FUTEX
1186         p->robust_list = NULL;
1187 #ifdef CONFIG_COMPAT
1188         p->compat_robust_list = NULL;
1189 #endif
1190         INIT_LIST_HEAD(&p->pi_state_list);
1191         p->pi_state_cache = NULL;
1192 #endif
1193         /*
1194          * sigaltstack should be cleared when sharing the same VM
1195          */
1196         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1197                 p->sas_ss_sp = p->sas_ss_size = 0;
1198
1199         /*
1200          * Syscall tracing and stepping should be turned off in the
1201          * child regardless of CLONE_PTRACE.
1202          */
1203         user_disable_single_step(p);
1204         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1205 #ifdef TIF_SYSCALL_EMU
1206         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1207 #endif
1208         clear_all_latency_tracing(p);
1209
1210         /* ok, now we should be set up.. */
1211         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1212         p->pdeath_signal = 0;
1213         p->exit_state = 0;
1214
1215         /*
1216          * Ok, make it visible to the rest of the system.
1217          * We dont wake it up yet.
1218          */
1219         p->group_leader = p;
1220         INIT_LIST_HEAD(&p->thread_group);
1221
1222         /* Now that the task is set up, run cgroup callbacks if
1223          * necessary. We need to run them before the task is visible
1224          * on the tasklist. */
1225         cgroup_fork_callbacks(p);
1226         cgroup_callbacks_done = 1;
1227
1228         /* Need tasklist lock for parent etc handling! */
1229         write_lock_irq(&tasklist_lock);
1230
1231         /* CLONE_PARENT re-uses the old parent */
1232         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1233                 p->real_parent = current->real_parent;
1234                 p->parent_exec_id = current->parent_exec_id;
1235         } else {
1236                 p->real_parent = current;
1237                 p->parent_exec_id = current->self_exec_id;
1238         }
1239
1240         spin_lock(&current->sighand->siglock);
1241
1242         /*
1243          * Process group and session signals need to be delivered to just the
1244          * parent before the fork or both the parent and the child after the
1245          * fork. Restart if a signal comes in before we add the new process to
1246          * it's process group.
1247          * A fatal signal pending means that current will exit, so the new
1248          * thread can't slip out of an OOM kill (or normal SIGKILL).
1249          */
1250         recalc_sigpending();
1251         if (signal_pending(current)) {
1252                 spin_unlock(&current->sighand->siglock);
1253                 write_unlock_irq(&tasklist_lock);
1254                 retval = -ERESTARTNOINTR;
1255                 goto bad_fork_free_pid;
1256         }
1257
1258         if (clone_flags & CLONE_THREAD) {
1259                 current->signal->nr_threads++;
1260                 atomic_inc(&current->signal->live);
1261                 atomic_inc(&current->signal->sigcnt);
1262                 p->group_leader = current->group_leader;
1263                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1264         }
1265
1266         if (likely(p->pid)) {
1267                 tracehook_finish_clone(p, clone_flags, trace);
1268
1269                 if (thread_group_leader(p)) {
1270                         if (clone_flags & CLONE_NEWPID)
1271                                 p->nsproxy->pid_ns->child_reaper = p;
1272
1273                         p->signal->leader_pid = pid;
1274                         p->signal->tty = tty_kref_get(current->signal->tty);
1275                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1276                         attach_pid(p, PIDTYPE_SID, task_session(current));
1277                         list_add_tail(&p->sibling, &p->real_parent->children);
1278                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1279                         __get_cpu_var(process_counts)++;
1280                 }
1281                 attach_pid(p, PIDTYPE_PID, pid);
1282                 nr_threads++;
1283         }
1284
1285         total_forks++;
1286         spin_unlock(&current->sighand->siglock);
1287         write_unlock_irq(&tasklist_lock);
1288         proc_fork_connector(p);
1289         cgroup_post_fork(p);
1290         perf_event_fork(p);
1291         return p;
1292
1293 bad_fork_free_pid:
1294         if (pid != &init_struct_pid)
1295                 free_pid(pid);
1296 bad_fork_cleanup_io:
1297         if (p->io_context)
1298                 exit_io_context(p);
1299 bad_fork_cleanup_namespaces:
1300         exit_task_namespaces(p);
1301 bad_fork_cleanup_mm:
1302         if (p->mm)
1303                 mmput(p->mm);
1304 bad_fork_cleanup_signal:
1305         if (!(clone_flags & CLONE_THREAD))
1306                 free_signal_struct(p->signal);
1307 bad_fork_cleanup_sighand:
1308         __cleanup_sighand(p->sighand);
1309 bad_fork_cleanup_fs:
1310         exit_fs(p); /* blocking */
1311 bad_fork_cleanup_files:
1312         exit_files(p); /* blocking */
1313 bad_fork_cleanup_semundo:
1314         exit_sem(p);
1315 bad_fork_cleanup_audit:
1316         audit_free(p);
1317 bad_fork_cleanup_policy:
1318         perf_event_free_task(p);
1319 #ifdef CONFIG_NUMA
1320         mpol_put(p->mempolicy);
1321 bad_fork_cleanup_cgroup:
1322 #endif
1323         cgroup_exit(p, cgroup_callbacks_done);
1324         delayacct_tsk_free(p);
1325         module_put(task_thread_info(p)->exec_domain->module);
1326 bad_fork_cleanup_count:
1327         atomic_dec(&p->cred->user->processes);
1328         exit_creds(p);
1329 bad_fork_free:
1330         free_task(p);
1331 fork_out:
1332         return ERR_PTR(retval);
1333 }
1334
1335 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1336 {
1337         memset(regs, 0, sizeof(struct pt_regs));
1338         return regs;
1339 }
1340
1341 struct task_struct * __cpuinit fork_idle(int cpu)
1342 {
1343         struct task_struct *task;
1344         struct pt_regs regs;
1345
1346         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1347                             &init_struct_pid, 0);
1348         if (!IS_ERR(task))
1349                 init_idle(task, cpu);
1350
1351         return task;
1352 }
1353
1354 /*
1355  *  Ok, this is the main fork-routine.
1356  *
1357  * It copies the process, and if successful kick-starts
1358  * it and waits for it to finish using the VM if required.
1359  */
1360 long do_fork(unsigned long clone_flags,
1361               unsigned long stack_start,
1362               struct pt_regs *regs,
1363               unsigned long stack_size,
1364               int __user *parent_tidptr,
1365               int __user *child_tidptr)
1366 {
1367         struct task_struct *p;
1368         int trace = 0;
1369         long nr;
1370
1371         /*
1372          * Do some preliminary argument and permissions checking before we
1373          * actually start allocating stuff
1374          */
1375         if (clone_flags & CLONE_NEWUSER) {
1376                 if (clone_flags & CLONE_THREAD)
1377                         return -EINVAL;
1378                 /* hopefully this check will go away when userns support is
1379                  * complete
1380                  */
1381                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1382                                 !capable(CAP_SETGID))
1383                         return -EPERM;
1384         }
1385
1386         /*
1387          * We hope to recycle these flags after 2.6.26
1388          */
1389         if (unlikely(clone_flags & CLONE_STOPPED)) {
1390                 static int __read_mostly count = 100;
1391
1392                 if (count > 0 && printk_ratelimit()) {
1393                         char comm[TASK_COMM_LEN];
1394
1395                         count--;
1396                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1397                                         "clone flags 0x%lx\n",
1398                                 get_task_comm(comm, current),
1399                                 clone_flags & CLONE_STOPPED);
1400                 }
1401         }
1402
1403         /*
1404          * When called from kernel_thread, don't do user tracing stuff.
1405          */
1406         if (likely(user_mode(regs)))
1407                 trace = tracehook_prepare_clone(clone_flags);
1408
1409         p = copy_process(clone_flags, stack_start, regs, stack_size,
1410                          child_tidptr, NULL, trace);
1411         /*
1412          * Do this prior waking up the new thread - the thread pointer
1413          * might get invalid after that point, if the thread exits quickly.
1414          */
1415         if (!IS_ERR(p)) {
1416                 struct completion vfork;
1417
1418                 trace_sched_process_fork(current, p);
1419
1420                 nr = task_pid_vnr(p);
1421
1422                 if (clone_flags & CLONE_PARENT_SETTID)
1423                         put_user(nr, parent_tidptr);
1424
1425                 if (clone_flags & CLONE_VFORK) {
1426                         p->vfork_done = &vfork;
1427                         init_completion(&vfork);
1428                 }
1429
1430                 audit_finish_fork(p);
1431                 tracehook_report_clone(regs, clone_flags, nr, p);
1432
1433                 /*
1434                  * We set PF_STARTING at creation in case tracing wants to
1435                  * use this to distinguish a fully live task from one that
1436                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1437                  * clear it and set the child going.
1438                  */
1439                 p->flags &= ~PF_STARTING;
1440
1441                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1442                         /*
1443                          * We'll start up with an immediate SIGSTOP.
1444                          */
1445                         sigaddset(&p->pending.signal, SIGSTOP);
1446                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1447                         __set_task_state(p, TASK_STOPPED);
1448                 } else {
1449                         wake_up_new_task(p, clone_flags);
1450                 }
1451
1452                 tracehook_report_clone_complete(trace, regs,
1453                                                 clone_flags, nr, p);
1454
1455                 if (clone_flags & CLONE_VFORK) {
1456                         freezer_do_not_count();
1457                         wait_for_completion(&vfork);
1458                         freezer_count();
1459                         tracehook_report_vfork_done(p, nr);
1460                 }
1461         } else {
1462                 nr = PTR_ERR(p);
1463         }
1464         return nr;
1465 }
1466
1467 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1468 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1469 #endif
1470
1471 static void sighand_ctor(void *data)
1472 {
1473         struct sighand_struct *sighand = data;
1474
1475         spin_lock_init(&sighand->siglock);
1476         init_waitqueue_head(&sighand->signalfd_wqh);
1477 }
1478
1479 void __init proc_caches_init(void)
1480 {
1481         sighand_cachep = kmem_cache_create("sighand_cache",
1482                         sizeof(struct sighand_struct), 0,
1483                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1484                         SLAB_NOTRACK, sighand_ctor);
1485         signal_cachep = kmem_cache_create("signal_cache",
1486                         sizeof(struct signal_struct), 0,
1487                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1488         files_cachep = kmem_cache_create("files_cache",
1489                         sizeof(struct files_struct), 0,
1490                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1491         fs_cachep = kmem_cache_create("fs_cache",
1492                         sizeof(struct fs_struct), 0,
1493                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1494         mm_cachep = kmem_cache_create("mm_struct",
1495                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1496                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1497         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1498         mmap_init();
1499 }
1500
1501 /*
1502  * Check constraints on flags passed to the unshare system call and
1503  * force unsharing of additional process context as appropriate.
1504  */
1505 static void check_unshare_flags(unsigned long *flags_ptr)
1506 {
1507         /*
1508          * If unsharing a thread from a thread group, must also
1509          * unshare vm.
1510          */
1511         if (*flags_ptr & CLONE_THREAD)
1512                 *flags_ptr |= CLONE_VM;
1513
1514         /*
1515          * If unsharing vm, must also unshare signal handlers.
1516          */
1517         if (*flags_ptr & CLONE_VM)
1518                 *flags_ptr |= CLONE_SIGHAND;
1519
1520         /*
1521          * If unsharing namespace, must also unshare filesystem information.
1522          */
1523         if (*flags_ptr & CLONE_NEWNS)
1524                 *flags_ptr |= CLONE_FS;
1525 }
1526
1527 /*
1528  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1529  */
1530 static int unshare_thread(unsigned long unshare_flags)
1531 {
1532         if (unshare_flags & CLONE_THREAD)
1533                 return -EINVAL;
1534
1535         return 0;
1536 }
1537
1538 /*
1539  * Unshare the filesystem structure if it is being shared
1540  */
1541 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1542 {
1543         struct fs_struct *fs = current->fs;
1544
1545         if (!(unshare_flags & CLONE_FS) || !fs)
1546                 return 0;
1547
1548         /* don't need lock here; in the worst case we'll do useless copy */
1549         if (fs->users == 1)
1550                 return 0;
1551
1552         *new_fsp = copy_fs_struct(fs);
1553         if (!*new_fsp)
1554                 return -ENOMEM;
1555
1556         return 0;
1557 }
1558
1559 /*
1560  * Unsharing of sighand is not supported yet
1561  */
1562 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1563 {
1564         struct sighand_struct *sigh = current->sighand;
1565
1566         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1567                 return -EINVAL;
1568         else
1569                 return 0;
1570 }
1571
1572 /*
1573  * Unshare vm if it is being shared
1574  */
1575 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1576 {
1577         struct mm_struct *mm = current->mm;
1578
1579         if ((unshare_flags & CLONE_VM) &&
1580             (mm && atomic_read(&mm->mm_users) > 1)) {
1581                 return -EINVAL;
1582         }
1583
1584         return 0;
1585 }
1586
1587 /*
1588  * Unshare file descriptor table if it is being shared
1589  */
1590 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1591 {
1592         struct files_struct *fd = current->files;
1593         int error = 0;
1594
1595         if ((unshare_flags & CLONE_FILES) &&
1596             (fd && atomic_read(&fd->count) > 1)) {
1597                 *new_fdp = dup_fd(fd, &error);
1598                 if (!*new_fdp)
1599                         return error;
1600         }
1601
1602         return 0;
1603 }
1604
1605 /*
1606  * unshare allows a process to 'unshare' part of the process
1607  * context which was originally shared using clone.  copy_*
1608  * functions used by do_fork() cannot be used here directly
1609  * because they modify an inactive task_struct that is being
1610  * constructed. Here we are modifying the current, active,
1611  * task_struct.
1612  */
1613 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1614 {
1615         int err = 0;
1616         struct fs_struct *fs, *new_fs = NULL;
1617         struct sighand_struct *new_sigh = NULL;
1618         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1619         struct files_struct *fd, *new_fd = NULL;
1620         struct nsproxy *new_nsproxy = NULL;
1621         int do_sysvsem = 0;
1622
1623         check_unshare_flags(&unshare_flags);
1624
1625         /* Return -EINVAL for all unsupported flags */
1626         err = -EINVAL;
1627         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1628                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1629                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1630                 goto bad_unshare_out;
1631
1632         /*
1633          * CLONE_NEWIPC must also detach from the undolist: after switching
1634          * to a new ipc namespace, the semaphore arrays from the old
1635          * namespace are unreachable.
1636          */
1637         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1638                 do_sysvsem = 1;
1639         if ((err = unshare_thread(unshare_flags)))
1640                 goto bad_unshare_out;
1641         if ((err = unshare_fs(unshare_flags, &new_fs)))
1642                 goto bad_unshare_cleanup_thread;
1643         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1644                 goto bad_unshare_cleanup_fs;
1645         if ((err = unshare_vm(unshare_flags, &new_mm)))
1646                 goto bad_unshare_cleanup_sigh;
1647         if ((err = unshare_fd(unshare_flags, &new_fd)))
1648                 goto bad_unshare_cleanup_vm;
1649         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1650                         new_fs)))
1651                 goto bad_unshare_cleanup_fd;
1652
1653         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1654                 if (do_sysvsem) {
1655                         /*
1656                          * CLONE_SYSVSEM is equivalent to sys_exit().
1657                          */
1658                         exit_sem(current);
1659                 }
1660
1661                 if (new_nsproxy) {
1662                         switch_task_namespaces(current, new_nsproxy);
1663                         new_nsproxy = NULL;
1664                 }
1665
1666                 task_lock(current);
1667
1668                 if (new_fs) {
1669                         fs = current->fs;
1670                         write_lock(&fs->lock);
1671                         current->fs = new_fs;
1672                         if (--fs->users)
1673                                 new_fs = NULL;
1674                         else
1675                                 new_fs = fs;
1676                         write_unlock(&fs->lock);
1677                 }
1678
1679                 if (new_mm) {
1680                         mm = current->mm;
1681                         active_mm = current->active_mm;
1682                         current->mm = new_mm;
1683                         current->active_mm = new_mm;
1684                         activate_mm(active_mm, new_mm);
1685                         new_mm = mm;
1686                 }
1687
1688                 if (new_fd) {
1689                         fd = current->files;
1690                         current->files = new_fd;
1691                         new_fd = fd;
1692                 }
1693
1694                 task_unlock(current);
1695         }
1696
1697         if (new_nsproxy)
1698                 put_nsproxy(new_nsproxy);
1699
1700 bad_unshare_cleanup_fd:
1701         if (new_fd)
1702                 put_files_struct(new_fd);
1703
1704 bad_unshare_cleanup_vm:
1705         if (new_mm)
1706                 mmput(new_mm);
1707
1708 bad_unshare_cleanup_sigh:
1709         if (new_sigh)
1710                 if (atomic_dec_and_test(&new_sigh->count))
1711                         kmem_cache_free(sighand_cachep, new_sigh);
1712
1713 bad_unshare_cleanup_fs:
1714         if (new_fs)
1715                 free_fs_struct(new_fs);
1716
1717 bad_unshare_cleanup_thread:
1718 bad_unshare_out:
1719         return err;
1720 }
1721
1722 /*
1723  *      Helper to unshare the files of the current task.
1724  *      We don't want to expose copy_files internals to
1725  *      the exec layer of the kernel.
1726  */
1727
1728 int unshare_files(struct files_struct **displaced)
1729 {
1730         struct task_struct *task = current;
1731         struct files_struct *copy = NULL;
1732         int error;
1733
1734         error = unshare_fd(CLONE_FILES, &copy);
1735         if (error || !copy) {
1736                 *displaced = NULL;
1737                 return error;
1738         }
1739         *displaced = task->files;
1740         task_lock(task);
1741         task->files = copy;
1742         task_unlock(task);
1743         return 0;
1744 }