kill the obsolete thread_group_cputime_free() helper
[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         kmem_cache_free(signal_cachep, sig);
171 }
172
173 static inline void put_signal_struct(struct signal_struct *sig)
174 {
175         if (atomic_dec_and_test(&sig->sigcnt))
176                 free_signal_struct(sig);
177 }
178
179 void __put_task_struct(struct task_struct *tsk)
180 {
181         WARN_ON(!tsk->exit_state);
182         WARN_ON(atomic_read(&tsk->usage));
183         WARN_ON(tsk == current);
184
185         exit_creds(tsk);
186         delayacct_tsk_free(tsk);
187         put_signal_struct(tsk->signal);
188
189         if (!profile_handoff_task(tsk))
190                 free_task(tsk);
191 }
192
193 /*
194  * macro override instead of weak attribute alias, to workaround
195  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
196  */
197 #ifndef arch_task_cache_init
198 #define arch_task_cache_init()
199 #endif
200
201 void __init fork_init(unsigned long mempages)
202 {
203 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
204 #ifndef ARCH_MIN_TASKALIGN
205 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
206 #endif
207         /* create a slab on which task_structs can be allocated */
208         task_struct_cachep =
209                 kmem_cache_create("task_struct", sizeof(struct task_struct),
210                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
211 #endif
212
213         /* do the arch specific task caches init */
214         arch_task_cache_init();
215
216         /*
217          * The default maximum number of threads is set to a safe
218          * value: the thread structures can take up at most half
219          * of memory.
220          */
221         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
222
223         /*
224          * we need to allow at least 20 threads to boot a system
225          */
226         if(max_threads < 20)
227                 max_threads = 20;
228
229         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
230         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
231         init_task.signal->rlim[RLIMIT_SIGPENDING] =
232                 init_task.signal->rlim[RLIMIT_NPROC];
233 }
234
235 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
236                                                struct task_struct *src)
237 {
238         *dst = *src;
239         return 0;
240 }
241
242 static struct task_struct *dup_task_struct(struct task_struct *orig)
243 {
244         struct task_struct *tsk;
245         struct thread_info *ti;
246         unsigned long *stackend;
247
248         int err;
249
250         prepare_to_copy(orig);
251
252         tsk = alloc_task_struct();
253         if (!tsk)
254                 return NULL;
255
256         ti = alloc_thread_info(tsk);
257         if (!ti) {
258                 free_task_struct(tsk);
259                 return NULL;
260         }
261
262         err = arch_dup_task_struct(tsk, orig);
263         if (err)
264                 goto out;
265
266         tsk->stack = ti;
267
268         err = prop_local_init_single(&tsk->dirties);
269         if (err)
270                 goto out;
271
272         setup_thread_stack(tsk, orig);
273         clear_user_return_notifier(tsk);
274         stackend = end_of_stack(tsk);
275         *stackend = STACK_END_MAGIC;    /* for overflow detection */
276
277 #ifdef CONFIG_CC_STACKPROTECTOR
278         tsk->stack_canary = get_random_int();
279 #endif
280
281         /* One for us, one for whoever does the "release_task()" (usually parent) */
282         atomic_set(&tsk->usage,2);
283         atomic_set(&tsk->fs_excl, 0);
284 #ifdef CONFIG_BLK_DEV_IO_TRACE
285         tsk->btrace_seq = 0;
286 #endif
287         tsk->splice_pipe = NULL;
288
289         account_kernel_stack(ti, 1);
290
291         return tsk;
292
293 out:
294         free_thread_info(ti);
295         free_task_struct(tsk);
296         return NULL;
297 }
298
299 #ifdef CONFIG_MMU
300 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
301 {
302         struct vm_area_struct *mpnt, *tmp, **pprev;
303         struct rb_node **rb_link, *rb_parent;
304         int retval;
305         unsigned long charge;
306         struct mempolicy *pol;
307
308         down_write(&oldmm->mmap_sem);
309         flush_cache_dup_mm(oldmm);
310         /*
311          * Not linked in yet - no deadlock potential:
312          */
313         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
314
315         mm->locked_vm = 0;
316         mm->mmap = NULL;
317         mm->mmap_cache = NULL;
318         mm->free_area_cache = oldmm->mmap_base;
319         mm->cached_hole_size = ~0UL;
320         mm->map_count = 0;
321         cpumask_clear(mm_cpumask(mm));
322         mm->mm_rb = RB_ROOT;
323         rb_link = &mm->mm_rb.rb_node;
324         rb_parent = NULL;
325         pprev = &mm->mmap;
326         retval = ksm_fork(mm, oldmm);
327         if (retval)
328                 goto out;
329
330         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
331                 struct file *file;
332
333                 if (mpnt->vm_flags & VM_DONTCOPY) {
334                         long pages = vma_pages(mpnt);
335                         mm->total_vm -= pages;
336                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
337                                                                 -pages);
338                         continue;
339                 }
340                 charge = 0;
341                 if (mpnt->vm_flags & VM_ACCOUNT) {
342                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
343                         if (security_vm_enough_memory(len))
344                                 goto fail_nomem;
345                         charge = len;
346                 }
347                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
348                 if (!tmp)
349                         goto fail_nomem;
350                 *tmp = *mpnt;
351                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
352                 pol = mpol_dup(vma_policy(mpnt));
353                 retval = PTR_ERR(pol);
354                 if (IS_ERR(pol))
355                         goto fail_nomem_policy;
356                 vma_set_policy(tmp, pol);
357                 if (anon_vma_fork(tmp, mpnt))
358                         goto fail_nomem_anon_vma_fork;
359                 tmp->vm_flags &= ~VM_LOCKED;
360                 tmp->vm_mm = mm;
361                 tmp->vm_next = NULL;
362                 file = tmp->vm_file;
363                 if (file) {
364                         struct inode *inode = file->f_path.dentry->d_inode;
365                         struct address_space *mapping = file->f_mapping;
366
367                         get_file(file);
368                         if (tmp->vm_flags & VM_DENYWRITE)
369                                 atomic_dec(&inode->i_writecount);
370                         spin_lock(&mapping->i_mmap_lock);
371                         if (tmp->vm_flags & VM_SHARED)
372                                 mapping->i_mmap_writable++;
373                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
374                         flush_dcache_mmap_lock(mapping);
375                         /* insert tmp into the share list, just after mpnt */
376                         vma_prio_tree_add(tmp, mpnt);
377                         flush_dcache_mmap_unlock(mapping);
378                         spin_unlock(&mapping->i_mmap_lock);
379                 }
380
381                 /*
382                  * Clear hugetlb-related page reserves for children. This only
383                  * affects MAP_PRIVATE mappings. Faults generated by the child
384                  * are not guaranteed to succeed, even if read-only
385                  */
386                 if (is_vm_hugetlb_page(tmp))
387                         reset_vma_resv_huge_pages(tmp);
388
389                 /*
390                  * Link in the new vma and copy the page table entries.
391                  */
392                 *pprev = tmp;
393                 pprev = &tmp->vm_next;
394
395                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
396                 rb_link = &tmp->vm_rb.rb_right;
397                 rb_parent = &tmp->vm_rb;
398
399                 mm->map_count++;
400                 retval = copy_page_range(mm, oldmm, mpnt);
401
402                 if (tmp->vm_ops && tmp->vm_ops->open)
403                         tmp->vm_ops->open(tmp);
404
405                 if (retval)
406                         goto out;
407         }
408         /* a new mm has just been created */
409         arch_dup_mmap(oldmm, mm);
410         retval = 0;
411 out:
412         up_write(&mm->mmap_sem);
413         flush_tlb_mm(oldmm);
414         up_write(&oldmm->mmap_sem);
415         return retval;
416 fail_nomem_anon_vma_fork:
417         mpol_put(pol);
418 fail_nomem_policy:
419         kmem_cache_free(vm_area_cachep, tmp);
420 fail_nomem:
421         retval = -ENOMEM;
422         vm_unacct_memory(charge);
423         goto out;
424 }
425
426 static inline int mm_alloc_pgd(struct mm_struct * mm)
427 {
428         mm->pgd = pgd_alloc(mm);
429         if (unlikely(!mm->pgd))
430                 return -ENOMEM;
431         return 0;
432 }
433
434 static inline void mm_free_pgd(struct mm_struct * mm)
435 {
436         pgd_free(mm, mm->pgd);
437 }
438 #else
439 #define dup_mmap(mm, oldmm)     (0)
440 #define mm_alloc_pgd(mm)        (0)
441 #define mm_free_pgd(mm)
442 #endif /* CONFIG_MMU */
443
444 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
445
446 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
447 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
448
449 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
450
451 static int __init coredump_filter_setup(char *s)
452 {
453         default_dump_filter =
454                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
455                 MMF_DUMP_FILTER_MASK;
456         return 1;
457 }
458
459 __setup("coredump_filter=", coredump_filter_setup);
460
461 #include <linux/init_task.h>
462
463 static void mm_init_aio(struct mm_struct *mm)
464 {
465 #ifdef CONFIG_AIO
466         spin_lock_init(&mm->ioctx_lock);
467         INIT_HLIST_HEAD(&mm->ioctx_list);
468 #endif
469 }
470
471 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
472 {
473         atomic_set(&mm->mm_users, 1);
474         atomic_set(&mm->mm_count, 1);
475         init_rwsem(&mm->mmap_sem);
476         INIT_LIST_HEAD(&mm->mmlist);
477         mm->flags = (current->mm) ?
478                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
479         mm->core_state = NULL;
480         mm->nr_ptes = 0;
481         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
482         spin_lock_init(&mm->page_table_lock);
483         mm->free_area_cache = TASK_UNMAPPED_BASE;
484         mm->cached_hole_size = ~0UL;
485         mm_init_aio(mm);
486         mm_init_owner(mm, p);
487
488         if (likely(!mm_alloc_pgd(mm))) {
489                 mm->def_flags = 0;
490                 mmu_notifier_mm_init(mm);
491                 return mm;
492         }
493
494         free_mm(mm);
495         return NULL;
496 }
497
498 /*
499  * Allocate and initialize an mm_struct.
500  */
501 struct mm_struct * mm_alloc(void)
502 {
503         struct mm_struct * mm;
504
505         mm = allocate_mm();
506         if (mm) {
507                 memset(mm, 0, sizeof(*mm));
508                 mm = mm_init(mm, current);
509         }
510         return mm;
511 }
512
513 /*
514  * Called when the last reference to the mm
515  * is dropped: either by a lazy thread or by
516  * mmput. Free the page directory and the mm.
517  */
518 void __mmdrop(struct mm_struct *mm)
519 {
520         BUG_ON(mm == &init_mm);
521         mm_free_pgd(mm);
522         destroy_context(mm);
523         mmu_notifier_mm_destroy(mm);
524         free_mm(mm);
525 }
526 EXPORT_SYMBOL_GPL(__mmdrop);
527
528 /*
529  * Decrement the use count and release all resources for an mm.
530  */
531 void mmput(struct mm_struct *mm)
532 {
533         might_sleep();
534
535         if (atomic_dec_and_test(&mm->mm_users)) {
536                 exit_aio(mm);
537                 ksm_exit(mm);
538                 exit_mmap(mm);
539                 set_mm_exe_file(mm, NULL);
540                 if (!list_empty(&mm->mmlist)) {
541                         spin_lock(&mmlist_lock);
542                         list_del(&mm->mmlist);
543                         spin_unlock(&mmlist_lock);
544                 }
545                 put_swap_token(mm);
546                 if (mm->binfmt)
547                         module_put(mm->binfmt->module);
548                 mmdrop(mm);
549         }
550 }
551 EXPORT_SYMBOL_GPL(mmput);
552
553 /**
554  * get_task_mm - acquire a reference to the task's mm
555  *
556  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
557  * this kernel workthread has transiently adopted a user mm with use_mm,
558  * to do its AIO) is not set and if so returns a reference to it, after
559  * bumping up the use count.  User must release the mm via mmput()
560  * after use.  Typically used by /proc and ptrace.
561  */
562 struct mm_struct *get_task_mm(struct task_struct *task)
563 {
564         struct mm_struct *mm;
565
566         task_lock(task);
567         mm = task->mm;
568         if (mm) {
569                 if (task->flags & PF_KTHREAD)
570                         mm = NULL;
571                 else
572                         atomic_inc(&mm->mm_users);
573         }
574         task_unlock(task);
575         return mm;
576 }
577 EXPORT_SYMBOL_GPL(get_task_mm);
578
579 /* Please note the differences between mmput and mm_release.
580  * mmput is called whenever we stop holding onto a mm_struct,
581  * error success whatever.
582  *
583  * mm_release is called after a mm_struct has been removed
584  * from the current process.
585  *
586  * This difference is important for error handling, when we
587  * only half set up a mm_struct for a new process and need to restore
588  * the old one.  Because we mmput the new mm_struct before
589  * restoring the old one. . .
590  * Eric Biederman 10 January 1998
591  */
592 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
593 {
594         struct completion *vfork_done = tsk->vfork_done;
595
596         /* Get rid of any futexes when releasing the mm */
597 #ifdef CONFIG_FUTEX
598         if (unlikely(tsk->robust_list)) {
599                 exit_robust_list(tsk);
600                 tsk->robust_list = NULL;
601         }
602 #ifdef CONFIG_COMPAT
603         if (unlikely(tsk->compat_robust_list)) {
604                 compat_exit_robust_list(tsk);
605                 tsk->compat_robust_list = NULL;
606         }
607 #endif
608         if (unlikely(!list_empty(&tsk->pi_state_list)))
609                 exit_pi_state_list(tsk);
610 #endif
611
612         /* Get rid of any cached register state */
613         deactivate_mm(tsk, mm);
614
615         /* notify parent sleeping on vfork() */
616         if (vfork_done) {
617                 tsk->vfork_done = NULL;
618                 complete(vfork_done);
619         }
620
621         /*
622          * If we're exiting normally, clear a user-space tid field if
623          * requested.  We leave this alone when dying by signal, to leave
624          * the value intact in a core dump, and to save the unnecessary
625          * trouble otherwise.  Userland only wants this done for a sys_exit.
626          */
627         if (tsk->clear_child_tid) {
628                 if (!(tsk->flags & PF_SIGNALED) &&
629                     atomic_read(&mm->mm_users) > 1) {
630                         /*
631                          * We don't check the error code - if userspace has
632                          * not set up a proper pointer then tough luck.
633                          */
634                         put_user(0, tsk->clear_child_tid);
635                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
636                                         1, NULL, NULL, 0);
637                 }
638                 tsk->clear_child_tid = NULL;
639         }
640 }
641
642 /*
643  * Allocate a new mm structure and copy contents from the
644  * mm structure of the passed in task structure.
645  */
646 struct mm_struct *dup_mm(struct task_struct *tsk)
647 {
648         struct mm_struct *mm, *oldmm = current->mm;
649         int err;
650
651         if (!oldmm)
652                 return NULL;
653
654         mm = allocate_mm();
655         if (!mm)
656                 goto fail_nomem;
657
658         memcpy(mm, oldmm, sizeof(*mm));
659
660         /* Initializing for Swap token stuff */
661         mm->token_priority = 0;
662         mm->last_interval = 0;
663
664         if (!mm_init(mm, tsk))
665                 goto fail_nomem;
666
667         if (init_new_context(tsk, mm))
668                 goto fail_nocontext;
669
670         dup_mm_exe_file(oldmm, mm);
671
672         err = dup_mmap(mm, oldmm);
673         if (err)
674                 goto free_pt;
675
676         mm->hiwater_rss = get_mm_rss(mm);
677         mm->hiwater_vm = mm->total_vm;
678
679         if (mm->binfmt && !try_module_get(mm->binfmt->module))
680                 goto free_pt;
681
682         return mm;
683
684 free_pt:
685         /* don't put binfmt in mmput, we haven't got module yet */
686         mm->binfmt = NULL;
687         mmput(mm);
688
689 fail_nomem:
690         return NULL;
691
692 fail_nocontext:
693         /*
694          * If init_new_context() failed, we cannot use mmput() to free the mm
695          * because it calls destroy_context()
696          */
697         mm_free_pgd(mm);
698         free_mm(mm);
699         return NULL;
700 }
701
702 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
703 {
704         struct mm_struct * mm, *oldmm;
705         int retval;
706
707         tsk->min_flt = tsk->maj_flt = 0;
708         tsk->nvcsw = tsk->nivcsw = 0;
709 #ifdef CONFIG_DETECT_HUNG_TASK
710         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
711 #endif
712
713         tsk->mm = NULL;
714         tsk->active_mm = NULL;
715
716         /*
717          * Are we cloning a kernel thread?
718          *
719          * We need to steal a active VM for that..
720          */
721         oldmm = current->mm;
722         if (!oldmm)
723                 return 0;
724
725         if (clone_flags & CLONE_VM) {
726                 atomic_inc(&oldmm->mm_users);
727                 mm = oldmm;
728                 goto good_mm;
729         }
730
731         retval = -ENOMEM;
732         mm = dup_mm(tsk);
733         if (!mm)
734                 goto fail_nomem;
735
736 good_mm:
737         /* Initializing for Swap token stuff */
738         mm->token_priority = 0;
739         mm->last_interval = 0;
740
741         tsk->mm = mm;
742         tsk->active_mm = mm;
743         return 0;
744
745 fail_nomem:
746         return retval;
747 }
748
749 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
750 {
751         struct fs_struct *fs = current->fs;
752         if (clone_flags & CLONE_FS) {
753                 /* tsk->fs is already what we want */
754                 write_lock(&fs->lock);
755                 if (fs->in_exec) {
756                         write_unlock(&fs->lock);
757                         return -EAGAIN;
758                 }
759                 fs->users++;
760                 write_unlock(&fs->lock);
761                 return 0;
762         }
763         tsk->fs = copy_fs_struct(fs);
764         if (!tsk->fs)
765                 return -ENOMEM;
766         return 0;
767 }
768
769 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
770 {
771         struct files_struct *oldf, *newf;
772         int error = 0;
773
774         /*
775          * A background process may not have any files ...
776          */
777         oldf = current->files;
778         if (!oldf)
779                 goto out;
780
781         if (clone_flags & CLONE_FILES) {
782                 atomic_inc(&oldf->count);
783                 goto out;
784         }
785
786         newf = dup_fd(oldf, &error);
787         if (!newf)
788                 goto out;
789
790         tsk->files = newf;
791         error = 0;
792 out:
793         return error;
794 }
795
796 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
797 {
798 #ifdef CONFIG_BLOCK
799         struct io_context *ioc = current->io_context;
800
801         if (!ioc)
802                 return 0;
803         /*
804          * Share io context with parent, if CLONE_IO is set
805          */
806         if (clone_flags & CLONE_IO) {
807                 tsk->io_context = ioc_task_link(ioc);
808                 if (unlikely(!tsk->io_context))
809                         return -ENOMEM;
810         } else if (ioprio_valid(ioc->ioprio)) {
811                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
812                 if (unlikely(!tsk->io_context))
813                         return -ENOMEM;
814
815                 tsk->io_context->ioprio = ioc->ioprio;
816         }
817 #endif
818         return 0;
819 }
820
821 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
822 {
823         struct sighand_struct *sig;
824
825         if (clone_flags & CLONE_SIGHAND) {
826                 atomic_inc(&current->sighand->count);
827                 return 0;
828         }
829         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
830         rcu_assign_pointer(tsk->sighand, sig);
831         if (!sig)
832                 return -ENOMEM;
833         atomic_set(&sig->count, 1);
834         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
835         return 0;
836 }
837
838 void __cleanup_sighand(struct sighand_struct *sighand)
839 {
840         if (atomic_dec_and_test(&sighand->count))
841                 kmem_cache_free(sighand_cachep, sighand);
842 }
843
844
845 /*
846  * Initialize POSIX timer handling for a thread group.
847  */
848 static void posix_cpu_timers_init_group(struct signal_struct *sig)
849 {
850         unsigned long cpu_limit;
851
852         /* Thread group counters. */
853         thread_group_cputime_init(sig);
854
855         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
856         if (cpu_limit != RLIM_INFINITY) {
857                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
858                 sig->cputimer.running = 1;
859         }
860
861         /* The timer lists. */
862         INIT_LIST_HEAD(&sig->cpu_timers[0]);
863         INIT_LIST_HEAD(&sig->cpu_timers[1]);
864         INIT_LIST_HEAD(&sig->cpu_timers[2]);
865 }
866
867 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
868 {
869         struct signal_struct *sig;
870
871         if (clone_flags & CLONE_THREAD)
872                 return 0;
873
874         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
875         tsk->signal = sig;
876         if (!sig)
877                 return -ENOMEM;
878
879         atomic_set(&sig->sigcnt, 1);
880         atomic_set(&sig->count, 1);
881         atomic_set(&sig->live, 1);
882         init_waitqueue_head(&sig->wait_chldexit);
883         if (clone_flags & CLONE_NEWPID)
884                 sig->flags |= SIGNAL_UNKILLABLE;
885         sig->curr_target = tsk;
886         init_sigpending(&sig->shared_pending);
887         INIT_LIST_HEAD(&sig->posix_timers);
888
889         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
890         sig->real_timer.function = it_real_fn;
891
892         task_lock(current->group_leader);
893         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
894         task_unlock(current->group_leader);
895
896         posix_cpu_timers_init_group(sig);
897
898         tty_audit_fork(sig);
899
900         sig->oom_adj = current->signal->oom_adj;
901
902         return 0;
903 }
904
905 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
906 {
907         unsigned long new_flags = p->flags;
908
909         new_flags &= ~PF_SUPERPRIV;
910         new_flags |= PF_FORKNOEXEC;
911         new_flags |= PF_STARTING;
912         p->flags = new_flags;
913         clear_freeze_flag(p);
914 }
915
916 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
917 {
918         current->clear_child_tid = tidptr;
919
920         return task_pid_vnr(current);
921 }
922
923 static void rt_mutex_init_task(struct task_struct *p)
924 {
925         raw_spin_lock_init(&p->pi_lock);
926 #ifdef CONFIG_RT_MUTEXES
927         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
928         p->pi_blocked_on = NULL;
929 #endif
930 }
931
932 #ifdef CONFIG_MM_OWNER
933 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
934 {
935         mm->owner = p;
936 }
937 #endif /* CONFIG_MM_OWNER */
938
939 /*
940  * Initialize POSIX timer handling for a single task.
941  */
942 static void posix_cpu_timers_init(struct task_struct *tsk)
943 {
944         tsk->cputime_expires.prof_exp = cputime_zero;
945         tsk->cputime_expires.virt_exp = cputime_zero;
946         tsk->cputime_expires.sched_exp = 0;
947         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
948         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
949         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
950 }
951
952 /*
953  * This creates a new process as a copy of the old one,
954  * but does not actually start it yet.
955  *
956  * It copies the registers, and all the appropriate
957  * parts of the process environment (as per the clone
958  * flags). The actual kick-off is left to the caller.
959  */
960 static struct task_struct *copy_process(unsigned long clone_flags,
961                                         unsigned long stack_start,
962                                         struct pt_regs *regs,
963                                         unsigned long stack_size,
964                                         int __user *child_tidptr,
965                                         struct pid *pid,
966                                         int trace)
967 {
968         int retval;
969         struct task_struct *p;
970         int cgroup_callbacks_done = 0;
971
972         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
973                 return ERR_PTR(-EINVAL);
974
975         /*
976          * Thread groups must share signals as well, and detached threads
977          * can only be started up within the thread group.
978          */
979         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
980                 return ERR_PTR(-EINVAL);
981
982         /*
983          * Shared signal handlers imply shared VM. By way of the above,
984          * thread groups also imply shared VM. Blocking this case allows
985          * for various simplifications in other code.
986          */
987         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
988                 return ERR_PTR(-EINVAL);
989
990         /*
991          * Siblings of global init remain as zombies on exit since they are
992          * not reaped by their parent (swapper). To solve this and to avoid
993          * multi-rooted process trees, prevent global and container-inits
994          * from creating siblings.
995          */
996         if ((clone_flags & CLONE_PARENT) &&
997                                 current->signal->flags & SIGNAL_UNKILLABLE)
998                 return ERR_PTR(-EINVAL);
999
1000         retval = security_task_create(clone_flags);
1001         if (retval)
1002                 goto fork_out;
1003
1004         retval = -ENOMEM;
1005         p = dup_task_struct(current);
1006         if (!p)
1007                 goto fork_out;
1008
1009         ftrace_graph_init_task(p);
1010
1011         rt_mutex_init_task(p);
1012
1013 #ifdef CONFIG_PROVE_LOCKING
1014         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1015         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1016 #endif
1017         retval = -EAGAIN;
1018         if (atomic_read(&p->real_cred->user->processes) >=
1019                         task_rlimit(p, RLIMIT_NPROC)) {
1020                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1021                     p->real_cred->user != INIT_USER)
1022                         goto bad_fork_free;
1023         }
1024
1025         retval = copy_creds(p, clone_flags);
1026         if (retval < 0)
1027                 goto bad_fork_free;
1028
1029         /*
1030          * If multiple threads are within copy_process(), then this check
1031          * triggers too late. This doesn't hurt, the check is only there
1032          * to stop root fork bombs.
1033          */
1034         retval = -EAGAIN;
1035         if (nr_threads >= max_threads)
1036                 goto bad_fork_cleanup_count;
1037
1038         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1039                 goto bad_fork_cleanup_count;
1040
1041         p->did_exec = 0;
1042         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1043         copy_flags(clone_flags, p);
1044         INIT_LIST_HEAD(&p->children);
1045         INIT_LIST_HEAD(&p->sibling);
1046         rcu_copy_process(p);
1047         p->vfork_done = NULL;
1048         spin_lock_init(&p->alloc_lock);
1049
1050         init_sigpending(&p->pending);
1051
1052         p->utime = cputime_zero;
1053         p->stime = cputime_zero;
1054         p->gtime = cputime_zero;
1055         p->utimescaled = cputime_zero;
1056         p->stimescaled = cputime_zero;
1057 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1058         p->prev_utime = cputime_zero;
1059         p->prev_stime = cputime_zero;
1060 #endif
1061 #if defined(SPLIT_RSS_COUNTING)
1062         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1063 #endif
1064
1065         p->default_timer_slack_ns = current->timer_slack_ns;
1066
1067         task_io_accounting_init(&p->ioac);
1068         acct_clear_integrals(p);
1069
1070         posix_cpu_timers_init(p);
1071
1072         p->lock_depth = -1;             /* -1 = no lock */
1073         do_posix_clock_monotonic_gettime(&p->start_time);
1074         p->real_start_time = p->start_time;
1075         monotonic_to_bootbased(&p->real_start_time);
1076         p->io_context = NULL;
1077         p->audit_context = NULL;
1078         cgroup_fork(p);
1079 #ifdef CONFIG_NUMA
1080         p->mempolicy = mpol_dup(p->mempolicy);
1081         if (IS_ERR(p->mempolicy)) {
1082                 retval = PTR_ERR(p->mempolicy);
1083                 p->mempolicy = NULL;
1084                 goto bad_fork_cleanup_cgroup;
1085         }
1086         mpol_fix_fork_child_flag(p);
1087 #endif
1088 #ifdef CONFIG_CPUSETS
1089         p->cpuset_mem_spread_rotor = node_random(p->mems_allowed);
1090         p->cpuset_slab_spread_rotor = node_random(p->mems_allowed);
1091 #endif
1092 #ifdef CONFIG_TRACE_IRQFLAGS
1093         p->irq_events = 0;
1094 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1095         p->hardirqs_enabled = 1;
1096 #else
1097         p->hardirqs_enabled = 0;
1098 #endif
1099         p->hardirq_enable_ip = 0;
1100         p->hardirq_enable_event = 0;
1101         p->hardirq_disable_ip = _THIS_IP_;
1102         p->hardirq_disable_event = 0;
1103         p->softirqs_enabled = 1;
1104         p->softirq_enable_ip = _THIS_IP_;
1105         p->softirq_enable_event = 0;
1106         p->softirq_disable_ip = 0;
1107         p->softirq_disable_event = 0;
1108         p->hardirq_context = 0;
1109         p->softirq_context = 0;
1110 #endif
1111 #ifdef CONFIG_LOCKDEP
1112         p->lockdep_depth = 0; /* no locks held yet */
1113         p->curr_chain_key = 0;
1114         p->lockdep_recursion = 0;
1115 #endif
1116
1117 #ifdef CONFIG_DEBUG_MUTEXES
1118         p->blocked_on = NULL; /* not blocked yet */
1119 #endif
1120 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1121         p->memcg_batch.do_batch = 0;
1122         p->memcg_batch.memcg = NULL;
1123 #endif
1124
1125         /* Perform scheduler related setup. Assign this task to a CPU. */
1126         sched_fork(p, clone_flags);
1127
1128         retval = perf_event_init_task(p);
1129         if (retval)
1130                 goto bad_fork_cleanup_policy;
1131
1132         if ((retval = audit_alloc(p)))
1133                 goto bad_fork_cleanup_policy;
1134         /* copy all the process information */
1135         if ((retval = copy_semundo(clone_flags, p)))
1136                 goto bad_fork_cleanup_audit;
1137         if ((retval = copy_files(clone_flags, p)))
1138                 goto bad_fork_cleanup_semundo;
1139         if ((retval = copy_fs(clone_flags, p)))
1140                 goto bad_fork_cleanup_files;
1141         if ((retval = copy_sighand(clone_flags, p)))
1142                 goto bad_fork_cleanup_fs;
1143         if ((retval = copy_signal(clone_flags, p)))
1144                 goto bad_fork_cleanup_sighand;
1145         if ((retval = copy_mm(clone_flags, p)))
1146                 goto bad_fork_cleanup_signal;
1147         if ((retval = copy_namespaces(clone_flags, p)))
1148                 goto bad_fork_cleanup_mm;
1149         if ((retval = copy_io(clone_flags, p)))
1150                 goto bad_fork_cleanup_namespaces;
1151         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1152         if (retval)
1153                 goto bad_fork_cleanup_io;
1154
1155         if (pid != &init_struct_pid) {
1156                 retval = -ENOMEM;
1157                 pid = alloc_pid(p->nsproxy->pid_ns);
1158                 if (!pid)
1159                         goto bad_fork_cleanup_io;
1160
1161                 if (clone_flags & CLONE_NEWPID) {
1162                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1163                         if (retval < 0)
1164                                 goto bad_fork_free_pid;
1165                 }
1166         }
1167
1168         p->pid = pid_nr(pid);
1169         p->tgid = p->pid;
1170         if (clone_flags & CLONE_THREAD)
1171                 p->tgid = current->tgid;
1172
1173         if (current->nsproxy != p->nsproxy) {
1174                 retval = ns_cgroup_clone(p, pid);
1175                 if (retval)
1176                         goto bad_fork_free_pid;
1177         }
1178
1179         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1180         /*
1181          * Clear TID on mm_release()?
1182          */
1183         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1184 #ifdef CONFIG_FUTEX
1185         p->robust_list = NULL;
1186 #ifdef CONFIG_COMPAT
1187         p->compat_robust_list = NULL;
1188 #endif
1189         INIT_LIST_HEAD(&p->pi_state_list);
1190         p->pi_state_cache = NULL;
1191 #endif
1192         /*
1193          * sigaltstack should be cleared when sharing the same VM
1194          */
1195         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1196                 p->sas_ss_sp = p->sas_ss_size = 0;
1197
1198         /*
1199          * Syscall tracing and stepping should be turned off in the
1200          * child regardless of CLONE_PTRACE.
1201          */
1202         user_disable_single_step(p);
1203         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1204 #ifdef TIF_SYSCALL_EMU
1205         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1206 #endif
1207         clear_all_latency_tracing(p);
1208
1209         /* ok, now we should be set up.. */
1210         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1211         p->pdeath_signal = 0;
1212         p->exit_state = 0;
1213
1214         /*
1215          * Ok, make it visible to the rest of the system.
1216          * We dont wake it up yet.
1217          */
1218         p->group_leader = p;
1219         INIT_LIST_HEAD(&p->thread_group);
1220
1221         /* Now that the task is set up, run cgroup callbacks if
1222          * necessary. We need to run them before the task is visible
1223          * on the tasklist. */
1224         cgroup_fork_callbacks(p);
1225         cgroup_callbacks_done = 1;
1226
1227         /* Need tasklist lock for parent etc handling! */
1228         write_lock_irq(&tasklist_lock);
1229
1230         /* CLONE_PARENT re-uses the old parent */
1231         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1232                 p->real_parent = current->real_parent;
1233                 p->parent_exec_id = current->parent_exec_id;
1234         } else {
1235                 p->real_parent = current;
1236                 p->parent_exec_id = current->self_exec_id;
1237         }
1238
1239         spin_lock(&current->sighand->siglock);
1240
1241         /*
1242          * Process group and session signals need to be delivered to just the
1243          * parent before the fork or both the parent and the child after the
1244          * fork. Restart if a signal comes in before we add the new process to
1245          * it's process group.
1246          * A fatal signal pending means that current will exit, so the new
1247          * thread can't slip out of an OOM kill (or normal SIGKILL).
1248          */
1249         recalc_sigpending();
1250         if (signal_pending(current)) {
1251                 spin_unlock(&current->sighand->siglock);
1252                 write_unlock_irq(&tasklist_lock);
1253                 retval = -ERESTARTNOINTR;
1254                 goto bad_fork_free_pid;
1255         }
1256
1257         if (clone_flags & CLONE_THREAD) {
1258                 atomic_inc(&current->signal->sigcnt);
1259                 atomic_inc(&current->signal->count);
1260                 atomic_inc(&current->signal->live);
1261                 p->group_leader = current->group_leader;
1262                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1263         }
1264
1265         if (likely(p->pid)) {
1266                 tracehook_finish_clone(p, clone_flags, trace);
1267
1268                 if (thread_group_leader(p)) {
1269                         if (clone_flags & CLONE_NEWPID)
1270                                 p->nsproxy->pid_ns->child_reaper = p;
1271
1272                         p->signal->leader_pid = pid;
1273                         p->signal->tty = tty_kref_get(current->signal->tty);
1274                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1275                         attach_pid(p, PIDTYPE_SID, task_session(current));
1276                         list_add_tail(&p->sibling, &p->real_parent->children);
1277                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1278                         __get_cpu_var(process_counts)++;
1279                 }
1280                 attach_pid(p, PIDTYPE_PID, pid);
1281                 nr_threads++;
1282         }
1283
1284         total_forks++;
1285         spin_unlock(&current->sighand->siglock);
1286         write_unlock_irq(&tasklist_lock);
1287         proc_fork_connector(p);
1288         cgroup_post_fork(p);
1289         perf_event_fork(p);
1290         return p;
1291
1292 bad_fork_free_pid:
1293         if (pid != &init_struct_pid)
1294                 free_pid(pid);
1295 bad_fork_cleanup_io:
1296         if (p->io_context)
1297                 exit_io_context(p);
1298 bad_fork_cleanup_namespaces:
1299         exit_task_namespaces(p);
1300 bad_fork_cleanup_mm:
1301         if (p->mm)
1302                 mmput(p->mm);
1303 bad_fork_cleanup_signal:
1304         if (!(clone_flags & CLONE_THREAD))
1305                 free_signal_struct(p->signal);
1306 bad_fork_cleanup_sighand:
1307         __cleanup_sighand(p->sighand);
1308 bad_fork_cleanup_fs:
1309         exit_fs(p); /* blocking */
1310 bad_fork_cleanup_files:
1311         exit_files(p); /* blocking */
1312 bad_fork_cleanup_semundo:
1313         exit_sem(p);
1314 bad_fork_cleanup_audit:
1315         audit_free(p);
1316 bad_fork_cleanup_policy:
1317         perf_event_free_task(p);
1318 #ifdef CONFIG_NUMA
1319         mpol_put(p->mempolicy);
1320 bad_fork_cleanup_cgroup:
1321 #endif
1322         cgroup_exit(p, cgroup_callbacks_done);
1323         delayacct_tsk_free(p);
1324         module_put(task_thread_info(p)->exec_domain->module);
1325 bad_fork_cleanup_count:
1326         atomic_dec(&p->cred->user->processes);
1327         exit_creds(p);
1328 bad_fork_free:
1329         free_task(p);
1330 fork_out:
1331         return ERR_PTR(retval);
1332 }
1333
1334 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1335 {
1336         memset(regs, 0, sizeof(struct pt_regs));
1337         return regs;
1338 }
1339
1340 struct task_struct * __cpuinit fork_idle(int cpu)
1341 {
1342         struct task_struct *task;
1343         struct pt_regs regs;
1344
1345         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1346                             &init_struct_pid, 0);
1347         if (!IS_ERR(task))
1348                 init_idle(task, cpu);
1349
1350         return task;
1351 }
1352
1353 /*
1354  *  Ok, this is the main fork-routine.
1355  *
1356  * It copies the process, and if successful kick-starts
1357  * it and waits for it to finish using the VM if required.
1358  */
1359 long do_fork(unsigned long clone_flags,
1360               unsigned long stack_start,
1361               struct pt_regs *regs,
1362               unsigned long stack_size,
1363               int __user *parent_tidptr,
1364               int __user *child_tidptr)
1365 {
1366         struct task_struct *p;
1367         int trace = 0;
1368         long nr;
1369
1370         /*
1371          * Do some preliminary argument and permissions checking before we
1372          * actually start allocating stuff
1373          */
1374         if (clone_flags & CLONE_NEWUSER) {
1375                 if (clone_flags & CLONE_THREAD)
1376                         return -EINVAL;
1377                 /* hopefully this check will go away when userns support is
1378                  * complete
1379                  */
1380                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1381                                 !capable(CAP_SETGID))
1382                         return -EPERM;
1383         }
1384
1385         /*
1386          * We hope to recycle these flags after 2.6.26
1387          */
1388         if (unlikely(clone_flags & CLONE_STOPPED)) {
1389                 static int __read_mostly count = 100;
1390
1391                 if (count > 0 && printk_ratelimit()) {
1392                         char comm[TASK_COMM_LEN];
1393
1394                         count--;
1395                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1396                                         "clone flags 0x%lx\n",
1397                                 get_task_comm(comm, current),
1398                                 clone_flags & CLONE_STOPPED);
1399                 }
1400         }
1401
1402         /*
1403          * When called from kernel_thread, don't do user tracing stuff.
1404          */
1405         if (likely(user_mode(regs)))
1406                 trace = tracehook_prepare_clone(clone_flags);
1407
1408         p = copy_process(clone_flags, stack_start, regs, stack_size,
1409                          child_tidptr, NULL, trace);
1410         /*
1411          * Do this prior waking up the new thread - the thread pointer
1412          * might get invalid after that point, if the thread exits quickly.
1413          */
1414         if (!IS_ERR(p)) {
1415                 struct completion vfork;
1416
1417                 trace_sched_process_fork(current, p);
1418
1419                 nr = task_pid_vnr(p);
1420
1421                 if (clone_flags & CLONE_PARENT_SETTID)
1422                         put_user(nr, parent_tidptr);
1423
1424                 if (clone_flags & CLONE_VFORK) {
1425                         p->vfork_done = &vfork;
1426                         init_completion(&vfork);
1427                 }
1428
1429                 audit_finish_fork(p);
1430                 tracehook_report_clone(regs, clone_flags, nr, p);
1431
1432                 /*
1433                  * We set PF_STARTING at creation in case tracing wants to
1434                  * use this to distinguish a fully live task from one that
1435                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1436                  * clear it and set the child going.
1437                  */
1438                 p->flags &= ~PF_STARTING;
1439
1440                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1441                         /*
1442                          * We'll start up with an immediate SIGSTOP.
1443                          */
1444                         sigaddset(&p->pending.signal, SIGSTOP);
1445                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1446                         __set_task_state(p, TASK_STOPPED);
1447                 } else {
1448                         wake_up_new_task(p, clone_flags);
1449                 }
1450
1451                 tracehook_report_clone_complete(trace, regs,
1452                                                 clone_flags, nr, p);
1453
1454                 if (clone_flags & CLONE_VFORK) {
1455                         freezer_do_not_count();
1456                         wait_for_completion(&vfork);
1457                         freezer_count();
1458                         tracehook_report_vfork_done(p, nr);
1459                 }
1460         } else {
1461                 nr = PTR_ERR(p);
1462         }
1463         return nr;
1464 }
1465
1466 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1467 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1468 #endif
1469
1470 static void sighand_ctor(void *data)
1471 {
1472         struct sighand_struct *sighand = data;
1473
1474         spin_lock_init(&sighand->siglock);
1475         init_waitqueue_head(&sighand->signalfd_wqh);
1476 }
1477
1478 void __init proc_caches_init(void)
1479 {
1480         sighand_cachep = kmem_cache_create("sighand_cache",
1481                         sizeof(struct sighand_struct), 0,
1482                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1483                         SLAB_NOTRACK, sighand_ctor);
1484         signal_cachep = kmem_cache_create("signal_cache",
1485                         sizeof(struct signal_struct), 0,
1486                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1487         files_cachep = kmem_cache_create("files_cache",
1488                         sizeof(struct files_struct), 0,
1489                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1490         fs_cachep = kmem_cache_create("fs_cache",
1491                         sizeof(struct fs_struct), 0,
1492                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1493         mm_cachep = kmem_cache_create("mm_struct",
1494                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1495                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1496         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1497         mmap_init();
1498 }
1499
1500 /*
1501  * Check constraints on flags passed to the unshare system call and
1502  * force unsharing of additional process context as appropriate.
1503  */
1504 static void check_unshare_flags(unsigned long *flags_ptr)
1505 {
1506         /*
1507          * If unsharing a thread from a thread group, must also
1508          * unshare vm.
1509          */
1510         if (*flags_ptr & CLONE_THREAD)
1511                 *flags_ptr |= CLONE_VM;
1512
1513         /*
1514          * If unsharing vm, must also unshare signal handlers.
1515          */
1516         if (*flags_ptr & CLONE_VM)
1517                 *flags_ptr |= CLONE_SIGHAND;
1518
1519         /*
1520          * If unsharing signal handlers and the task was created
1521          * using CLONE_THREAD, then must unshare the thread
1522          */
1523         if ((*flags_ptr & CLONE_SIGHAND) &&
1524             (atomic_read(&current->signal->count) > 1))
1525                 *flags_ptr |= CLONE_THREAD;
1526
1527         /*
1528          * If unsharing namespace, must also unshare filesystem information.
1529          */
1530         if (*flags_ptr & CLONE_NEWNS)
1531                 *flags_ptr |= CLONE_FS;
1532 }
1533
1534 /*
1535  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1536  */
1537 static int unshare_thread(unsigned long unshare_flags)
1538 {
1539         if (unshare_flags & CLONE_THREAD)
1540                 return -EINVAL;
1541
1542         return 0;
1543 }
1544
1545 /*
1546  * Unshare the filesystem structure if it is being shared
1547  */
1548 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1549 {
1550         struct fs_struct *fs = current->fs;
1551
1552         if (!(unshare_flags & CLONE_FS) || !fs)
1553                 return 0;
1554
1555         /* don't need lock here; in the worst case we'll do useless copy */
1556         if (fs->users == 1)
1557                 return 0;
1558
1559         *new_fsp = copy_fs_struct(fs);
1560         if (!*new_fsp)
1561                 return -ENOMEM;
1562
1563         return 0;
1564 }
1565
1566 /*
1567  * Unsharing of sighand is not supported yet
1568  */
1569 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1570 {
1571         struct sighand_struct *sigh = current->sighand;
1572
1573         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1574                 return -EINVAL;
1575         else
1576                 return 0;
1577 }
1578
1579 /*
1580  * Unshare vm if it is being shared
1581  */
1582 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1583 {
1584         struct mm_struct *mm = current->mm;
1585
1586         if ((unshare_flags & CLONE_VM) &&
1587             (mm && atomic_read(&mm->mm_users) > 1)) {
1588                 return -EINVAL;
1589         }
1590
1591         return 0;
1592 }
1593
1594 /*
1595  * Unshare file descriptor table if it is being shared
1596  */
1597 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1598 {
1599         struct files_struct *fd = current->files;
1600         int error = 0;
1601
1602         if ((unshare_flags & CLONE_FILES) &&
1603             (fd && atomic_read(&fd->count) > 1)) {
1604                 *new_fdp = dup_fd(fd, &error);
1605                 if (!*new_fdp)
1606                         return error;
1607         }
1608
1609         return 0;
1610 }
1611
1612 /*
1613  * unshare allows a process to 'unshare' part of the process
1614  * context which was originally shared using clone.  copy_*
1615  * functions used by do_fork() cannot be used here directly
1616  * because they modify an inactive task_struct that is being
1617  * constructed. Here we are modifying the current, active,
1618  * task_struct.
1619  */
1620 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1621 {
1622         int err = 0;
1623         struct fs_struct *fs, *new_fs = NULL;
1624         struct sighand_struct *new_sigh = NULL;
1625         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1626         struct files_struct *fd, *new_fd = NULL;
1627         struct nsproxy *new_nsproxy = NULL;
1628         int do_sysvsem = 0;
1629
1630         check_unshare_flags(&unshare_flags);
1631
1632         /* Return -EINVAL for all unsupported flags */
1633         err = -EINVAL;
1634         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1635                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1636                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1637                 goto bad_unshare_out;
1638
1639         /*
1640          * CLONE_NEWIPC must also detach from the undolist: after switching
1641          * to a new ipc namespace, the semaphore arrays from the old
1642          * namespace are unreachable.
1643          */
1644         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1645                 do_sysvsem = 1;
1646         if ((err = unshare_thread(unshare_flags)))
1647                 goto bad_unshare_out;
1648         if ((err = unshare_fs(unshare_flags, &new_fs)))
1649                 goto bad_unshare_cleanup_thread;
1650         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1651                 goto bad_unshare_cleanup_fs;
1652         if ((err = unshare_vm(unshare_flags, &new_mm)))
1653                 goto bad_unshare_cleanup_sigh;
1654         if ((err = unshare_fd(unshare_flags, &new_fd)))
1655                 goto bad_unshare_cleanup_vm;
1656         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1657                         new_fs)))
1658                 goto bad_unshare_cleanup_fd;
1659
1660         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1661                 if (do_sysvsem) {
1662                         /*
1663                          * CLONE_SYSVSEM is equivalent to sys_exit().
1664                          */
1665                         exit_sem(current);
1666                 }
1667
1668                 if (new_nsproxy) {
1669                         switch_task_namespaces(current, new_nsproxy);
1670                         new_nsproxy = NULL;
1671                 }
1672
1673                 task_lock(current);
1674
1675                 if (new_fs) {
1676                         fs = current->fs;
1677                         write_lock(&fs->lock);
1678                         current->fs = new_fs;
1679                         if (--fs->users)
1680                                 new_fs = NULL;
1681                         else
1682                                 new_fs = fs;
1683                         write_unlock(&fs->lock);
1684                 }
1685
1686                 if (new_mm) {
1687                         mm = current->mm;
1688                         active_mm = current->active_mm;
1689                         current->mm = new_mm;
1690                         current->active_mm = new_mm;
1691                         activate_mm(active_mm, new_mm);
1692                         new_mm = mm;
1693                 }
1694
1695                 if (new_fd) {
1696                         fd = current->files;
1697                         current->files = new_fd;
1698                         new_fd = fd;
1699                 }
1700
1701                 task_unlock(current);
1702         }
1703
1704         if (new_nsproxy)
1705                 put_nsproxy(new_nsproxy);
1706
1707 bad_unshare_cleanup_fd:
1708         if (new_fd)
1709                 put_files_struct(new_fd);
1710
1711 bad_unshare_cleanup_vm:
1712         if (new_mm)
1713                 mmput(new_mm);
1714
1715 bad_unshare_cleanup_sigh:
1716         if (new_sigh)
1717                 if (atomic_dec_and_test(&new_sigh->count))
1718                         kmem_cache_free(sighand_cachep, new_sigh);
1719
1720 bad_unshare_cleanup_fs:
1721         if (new_fs)
1722                 free_fs_struct(new_fs);
1723
1724 bad_unshare_cleanup_thread:
1725 bad_unshare_out:
1726         return err;
1727 }
1728
1729 /*
1730  *      Helper to unshare the files of the current task.
1731  *      We don't want to expose copy_files internals to
1732  *      the exec layer of the kernel.
1733  */
1734
1735 int unshare_files(struct files_struct **displaced)
1736 {
1737         struct task_struct *task = current;
1738         struct files_struct *copy = NULL;
1739         int error;
1740
1741         error = unshare_fd(CLONE_FILES, &copy);
1742         if (error || !copy) {
1743                 *displaced = NULL;
1744                 return error;
1745         }
1746         *displaced = task->files;
1747         task_lock(task);
1748         task->files = copy;
1749         task_unlock(task);
1750         return 0;
1751 }