856eac3ec52eb1a11fa6405d3e870da796c778ee
[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                 spin_lock(&fs->lock);
756                 if (fs->in_exec) {
757                         spin_unlock(&fs->lock);
758                         return -EAGAIN;
759                 }
760                 fs->users++;
761                 spin_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         sig->oom_score_adj = current->signal->oom_score_adj;
903
904         return 0;
905 }
906
907 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
908 {
909         unsigned long new_flags = p->flags;
910
911         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
912         new_flags |= PF_FORKNOEXEC;
913         new_flags |= PF_STARTING;
914         p->flags = new_flags;
915         clear_freeze_flag(p);
916 }
917
918 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
919 {
920         current->clear_child_tid = tidptr;
921
922         return task_pid_vnr(current);
923 }
924
925 static void rt_mutex_init_task(struct task_struct *p)
926 {
927         raw_spin_lock_init(&p->pi_lock);
928 #ifdef CONFIG_RT_MUTEXES
929         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
930         p->pi_blocked_on = NULL;
931 #endif
932 }
933
934 #ifdef CONFIG_MM_OWNER
935 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
936 {
937         mm->owner = p;
938 }
939 #endif /* CONFIG_MM_OWNER */
940
941 /*
942  * Initialize POSIX timer handling for a single task.
943  */
944 static void posix_cpu_timers_init(struct task_struct *tsk)
945 {
946         tsk->cputime_expires.prof_exp = cputime_zero;
947         tsk->cputime_expires.virt_exp = cputime_zero;
948         tsk->cputime_expires.sched_exp = 0;
949         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
950         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
951         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
952 }
953
954 /*
955  * This creates a new process as a copy of the old one,
956  * but does not actually start it yet.
957  *
958  * It copies the registers, and all the appropriate
959  * parts of the process environment (as per the clone
960  * flags). The actual kick-off is left to the caller.
961  */
962 static struct task_struct *copy_process(unsigned long clone_flags,
963                                         unsigned long stack_start,
964                                         struct pt_regs *regs,
965                                         unsigned long stack_size,
966                                         int __user *child_tidptr,
967                                         struct pid *pid,
968                                         int trace)
969 {
970         int retval;
971         struct task_struct *p;
972         int cgroup_callbacks_done = 0;
973
974         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
975                 return ERR_PTR(-EINVAL);
976
977         /*
978          * Thread groups must share signals as well, and detached threads
979          * can only be started up within the thread group.
980          */
981         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
982                 return ERR_PTR(-EINVAL);
983
984         /*
985          * Shared signal handlers imply shared VM. By way of the above,
986          * thread groups also imply shared VM. Blocking this case allows
987          * for various simplifications in other code.
988          */
989         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
990                 return ERR_PTR(-EINVAL);
991
992         /*
993          * Siblings of global init remain as zombies on exit since they are
994          * not reaped by their parent (swapper). To solve this and to avoid
995          * multi-rooted process trees, prevent global and container-inits
996          * from creating siblings.
997          */
998         if ((clone_flags & CLONE_PARENT) &&
999                                 current->signal->flags & SIGNAL_UNKILLABLE)
1000                 return ERR_PTR(-EINVAL);
1001
1002         retval = security_task_create(clone_flags);
1003         if (retval)
1004                 goto fork_out;
1005
1006         retval = -ENOMEM;
1007         p = dup_task_struct(current);
1008         if (!p)
1009                 goto fork_out;
1010
1011         ftrace_graph_init_task(p);
1012
1013         rt_mutex_init_task(p);
1014
1015 #ifdef CONFIG_PROVE_LOCKING
1016         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1017         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1018 #endif
1019         retval = -EAGAIN;
1020         if (atomic_read(&p->real_cred->user->processes) >=
1021                         task_rlimit(p, RLIMIT_NPROC)) {
1022                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1023                     p->real_cred->user != INIT_USER)
1024                         goto bad_fork_free;
1025         }
1026
1027         retval = copy_creds(p, clone_flags);
1028         if (retval < 0)
1029                 goto bad_fork_free;
1030
1031         /*
1032          * If multiple threads are within copy_process(), then this check
1033          * triggers too late. This doesn't hurt, the check is only there
1034          * to stop root fork bombs.
1035          */
1036         retval = -EAGAIN;
1037         if (nr_threads >= max_threads)
1038                 goto bad_fork_cleanup_count;
1039
1040         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1041                 goto bad_fork_cleanup_count;
1042
1043         p->did_exec = 0;
1044         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1045         copy_flags(clone_flags, p);
1046         INIT_LIST_HEAD(&p->children);
1047         INIT_LIST_HEAD(&p->sibling);
1048         rcu_copy_process(p);
1049         p->vfork_done = NULL;
1050         spin_lock_init(&p->alloc_lock);
1051
1052         init_sigpending(&p->pending);
1053
1054         p->utime = cputime_zero;
1055         p->stime = cputime_zero;
1056         p->gtime = cputime_zero;
1057         p->utimescaled = cputime_zero;
1058         p->stimescaled = cputime_zero;
1059 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1060         p->prev_utime = cputime_zero;
1061         p->prev_stime = cputime_zero;
1062 #endif
1063 #if defined(SPLIT_RSS_COUNTING)
1064         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1065 #endif
1066
1067         p->default_timer_slack_ns = current->timer_slack_ns;
1068
1069         task_io_accounting_init(&p->ioac);
1070         acct_clear_integrals(p);
1071
1072         posix_cpu_timers_init(p);
1073
1074         p->lock_depth = -1;             /* -1 = no lock */
1075         do_posix_clock_monotonic_gettime(&p->start_time);
1076         p->real_start_time = p->start_time;
1077         monotonic_to_bootbased(&p->real_start_time);
1078         p->io_context = NULL;
1079         p->audit_context = NULL;
1080         cgroup_fork(p);
1081 #ifdef CONFIG_NUMA
1082         p->mempolicy = mpol_dup(p->mempolicy);
1083         if (IS_ERR(p->mempolicy)) {
1084                 retval = PTR_ERR(p->mempolicy);
1085                 p->mempolicy = NULL;
1086                 goto bad_fork_cleanup_cgroup;
1087         }
1088         mpol_fix_fork_child_flag(p);
1089 #endif
1090 #ifdef CONFIG_TRACE_IRQFLAGS
1091         p->irq_events = 0;
1092 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1093         p->hardirqs_enabled = 1;
1094 #else
1095         p->hardirqs_enabled = 0;
1096 #endif
1097         p->hardirq_enable_ip = 0;
1098         p->hardirq_enable_event = 0;
1099         p->hardirq_disable_ip = _THIS_IP_;
1100         p->hardirq_disable_event = 0;
1101         p->softirqs_enabled = 1;
1102         p->softirq_enable_ip = _THIS_IP_;
1103         p->softirq_enable_event = 0;
1104         p->softirq_disable_ip = 0;
1105         p->softirq_disable_event = 0;
1106         p->hardirq_context = 0;
1107         p->softirq_context = 0;
1108 #endif
1109 #ifdef CONFIG_LOCKDEP
1110         p->lockdep_depth = 0; /* no locks held yet */
1111         p->curr_chain_key = 0;
1112         p->lockdep_recursion = 0;
1113 #endif
1114
1115 #ifdef CONFIG_DEBUG_MUTEXES
1116         p->blocked_on = NULL; /* not blocked yet */
1117 #endif
1118 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1119         p->memcg_batch.do_batch = 0;
1120         p->memcg_batch.memcg = NULL;
1121 #endif
1122
1123         /* Perform scheduler related setup. Assign this task to a CPU. */
1124         sched_fork(p, clone_flags);
1125
1126         retval = perf_event_init_task(p);
1127         if (retval)
1128                 goto bad_fork_cleanup_policy;
1129
1130         if ((retval = audit_alloc(p)))
1131                 goto bad_fork_cleanup_policy;
1132         /* copy all the process information */
1133         if ((retval = copy_semundo(clone_flags, p)))
1134                 goto bad_fork_cleanup_audit;
1135         if ((retval = copy_files(clone_flags, p)))
1136                 goto bad_fork_cleanup_semundo;
1137         if ((retval = copy_fs(clone_flags, p)))
1138                 goto bad_fork_cleanup_files;
1139         if ((retval = copy_sighand(clone_flags, p)))
1140                 goto bad_fork_cleanup_fs;
1141         if ((retval = copy_signal(clone_flags, p)))
1142                 goto bad_fork_cleanup_sighand;
1143         if ((retval = copy_mm(clone_flags, p)))
1144                 goto bad_fork_cleanup_signal;
1145         if ((retval = copy_namespaces(clone_flags, p)))
1146                 goto bad_fork_cleanup_mm;
1147         if ((retval = copy_io(clone_flags, p)))
1148                 goto bad_fork_cleanup_namespaces;
1149         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1150         if (retval)
1151                 goto bad_fork_cleanup_io;
1152
1153         if (pid != &init_struct_pid) {
1154                 retval = -ENOMEM;
1155                 pid = alloc_pid(p->nsproxy->pid_ns);
1156                 if (!pid)
1157                         goto bad_fork_cleanup_io;
1158
1159                 if (clone_flags & CLONE_NEWPID) {
1160                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1161                         if (retval < 0)
1162                                 goto bad_fork_free_pid;
1163                 }
1164         }
1165
1166         p->pid = pid_nr(pid);
1167         p->tgid = p->pid;
1168         if (clone_flags & CLONE_THREAD)
1169                 p->tgid = current->tgid;
1170
1171         if (current->nsproxy != p->nsproxy) {
1172                 retval = ns_cgroup_clone(p, pid);
1173                 if (retval)
1174                         goto bad_fork_free_pid;
1175         }
1176
1177         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1178         /*
1179          * Clear TID on mm_release()?
1180          */
1181         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1182 #ifdef CONFIG_FUTEX
1183         p->robust_list = NULL;
1184 #ifdef CONFIG_COMPAT
1185         p->compat_robust_list = NULL;
1186 #endif
1187         INIT_LIST_HEAD(&p->pi_state_list);
1188         p->pi_state_cache = NULL;
1189 #endif
1190         /*
1191          * sigaltstack should be cleared when sharing the same VM
1192          */
1193         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1194                 p->sas_ss_sp = p->sas_ss_size = 0;
1195
1196         /*
1197          * Syscall tracing and stepping should be turned off in the
1198          * child regardless of CLONE_PTRACE.
1199          */
1200         user_disable_single_step(p);
1201         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1202 #ifdef TIF_SYSCALL_EMU
1203         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1204 #endif
1205         clear_all_latency_tracing(p);
1206
1207         /* ok, now we should be set up.. */
1208         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1209         p->pdeath_signal = 0;
1210         p->exit_state = 0;
1211
1212         /*
1213          * Ok, make it visible to the rest of the system.
1214          * We dont wake it up yet.
1215          */
1216         p->group_leader = p;
1217         INIT_LIST_HEAD(&p->thread_group);
1218
1219         /* Now that the task is set up, run cgroup callbacks if
1220          * necessary. We need to run them before the task is visible
1221          * on the tasklist. */
1222         cgroup_fork_callbacks(p);
1223         cgroup_callbacks_done = 1;
1224
1225         /* Need tasklist lock for parent etc handling! */
1226         write_lock_irq(&tasklist_lock);
1227
1228         /* CLONE_PARENT re-uses the old parent */
1229         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1230                 p->real_parent = current->real_parent;
1231                 p->parent_exec_id = current->parent_exec_id;
1232         } else {
1233                 p->real_parent = current;
1234                 p->parent_exec_id = current->self_exec_id;
1235         }
1236
1237         spin_lock(&current->sighand->siglock);
1238
1239         /*
1240          * Process group and session signals need to be delivered to just the
1241          * parent before the fork or both the parent and the child after the
1242          * fork. Restart if a signal comes in before we add the new process to
1243          * it's process group.
1244          * A fatal signal pending means that current will exit, so the new
1245          * thread can't slip out of an OOM kill (or normal SIGKILL).
1246          */
1247         recalc_sigpending();
1248         if (signal_pending(current)) {
1249                 spin_unlock(&current->sighand->siglock);
1250                 write_unlock_irq(&tasklist_lock);
1251                 retval = -ERESTARTNOINTR;
1252                 goto bad_fork_free_pid;
1253         }
1254
1255         if (clone_flags & CLONE_THREAD) {
1256                 current->signal->nr_threads++;
1257                 atomic_inc(&current->signal->live);
1258                 atomic_inc(&current->signal->sigcnt);
1259                 p->group_leader = current->group_leader;
1260                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1261         }
1262
1263         if (likely(p->pid)) {
1264                 tracehook_finish_clone(p, clone_flags, trace);
1265
1266                 if (thread_group_leader(p)) {
1267                         if (clone_flags & CLONE_NEWPID)
1268                                 p->nsproxy->pid_ns->child_reaper = p;
1269
1270                         p->signal->leader_pid = pid;
1271                         p->signal->tty = tty_kref_get(current->signal->tty);
1272                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1273                         attach_pid(p, PIDTYPE_SID, task_session(current));
1274                         list_add_tail(&p->sibling, &p->real_parent->children);
1275                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1276                         __get_cpu_var(process_counts)++;
1277                 }
1278                 attach_pid(p, PIDTYPE_PID, pid);
1279                 nr_threads++;
1280         }
1281
1282         total_forks++;
1283         spin_unlock(&current->sighand->siglock);
1284         write_unlock_irq(&tasklist_lock);
1285         proc_fork_connector(p);
1286         cgroup_post_fork(p);
1287         perf_event_fork(p);
1288         return p;
1289
1290 bad_fork_free_pid:
1291         if (pid != &init_struct_pid)
1292                 free_pid(pid);
1293 bad_fork_cleanup_io:
1294         if (p->io_context)
1295                 exit_io_context(p);
1296 bad_fork_cleanup_namespaces:
1297         exit_task_namespaces(p);
1298 bad_fork_cleanup_mm:
1299         if (p->mm)
1300                 mmput(p->mm);
1301 bad_fork_cleanup_signal:
1302         if (!(clone_flags & CLONE_THREAD))
1303                 free_signal_struct(p->signal);
1304 bad_fork_cleanup_sighand:
1305         __cleanup_sighand(p->sighand);
1306 bad_fork_cleanup_fs:
1307         exit_fs(p); /* blocking */
1308 bad_fork_cleanup_files:
1309         exit_files(p); /* blocking */
1310 bad_fork_cleanup_semundo:
1311         exit_sem(p);
1312 bad_fork_cleanup_audit:
1313         audit_free(p);
1314 bad_fork_cleanup_policy:
1315         perf_event_free_task(p);
1316 #ifdef CONFIG_NUMA
1317         mpol_put(p->mempolicy);
1318 bad_fork_cleanup_cgroup:
1319 #endif
1320         cgroup_exit(p, cgroup_callbacks_done);
1321         delayacct_tsk_free(p);
1322         module_put(task_thread_info(p)->exec_domain->module);
1323 bad_fork_cleanup_count:
1324         atomic_dec(&p->cred->user->processes);
1325         exit_creds(p);
1326 bad_fork_free:
1327         free_task(p);
1328 fork_out:
1329         return ERR_PTR(retval);
1330 }
1331
1332 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1333 {
1334         memset(regs, 0, sizeof(struct pt_regs));
1335         return regs;
1336 }
1337
1338 static inline void init_idle_pids(struct pid_link *links)
1339 {
1340         enum pid_type type;
1341
1342         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1343                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1344                 links[type].pid = &init_struct_pid;
1345         }
1346 }
1347
1348 struct task_struct * __cpuinit fork_idle(int cpu)
1349 {
1350         struct task_struct *task;
1351         struct pt_regs regs;
1352
1353         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1354                             &init_struct_pid, 0);
1355         if (!IS_ERR(task)) {
1356                 init_idle_pids(task->pids);
1357                 init_idle(task, cpu);
1358         }
1359
1360         return task;
1361 }
1362
1363 /*
1364  *  Ok, this is the main fork-routine.
1365  *
1366  * It copies the process, and if successful kick-starts
1367  * it and waits for it to finish using the VM if required.
1368  */
1369 long do_fork(unsigned long clone_flags,
1370               unsigned long stack_start,
1371               struct pt_regs *regs,
1372               unsigned long stack_size,
1373               int __user *parent_tidptr,
1374               int __user *child_tidptr)
1375 {
1376         struct task_struct *p;
1377         int trace = 0;
1378         long nr;
1379
1380         /*
1381          * Do some preliminary argument and permissions checking before we
1382          * actually start allocating stuff
1383          */
1384         if (clone_flags & CLONE_NEWUSER) {
1385                 if (clone_flags & CLONE_THREAD)
1386                         return -EINVAL;
1387                 /* hopefully this check will go away when userns support is
1388                  * complete
1389                  */
1390                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1391                                 !capable(CAP_SETGID))
1392                         return -EPERM;
1393         }
1394
1395         /*
1396          * We hope to recycle these flags after 2.6.26
1397          */
1398         if (unlikely(clone_flags & CLONE_STOPPED)) {
1399                 static int __read_mostly count = 100;
1400
1401                 if (count > 0 && printk_ratelimit()) {
1402                         char comm[TASK_COMM_LEN];
1403
1404                         count--;
1405                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1406                                         "clone flags 0x%lx\n",
1407                                 get_task_comm(comm, current),
1408                                 clone_flags & CLONE_STOPPED);
1409                 }
1410         }
1411
1412         /*
1413          * When called from kernel_thread, don't do user tracing stuff.
1414          */
1415         if (likely(user_mode(regs)))
1416                 trace = tracehook_prepare_clone(clone_flags);
1417
1418         p = copy_process(clone_flags, stack_start, regs, stack_size,
1419                          child_tidptr, NULL, trace);
1420         /*
1421          * Do this prior waking up the new thread - the thread pointer
1422          * might get invalid after that point, if the thread exits quickly.
1423          */
1424         if (!IS_ERR(p)) {
1425                 struct completion vfork;
1426
1427                 trace_sched_process_fork(current, p);
1428
1429                 nr = task_pid_vnr(p);
1430
1431                 if (clone_flags & CLONE_PARENT_SETTID)
1432                         put_user(nr, parent_tidptr);
1433
1434                 if (clone_flags & CLONE_VFORK) {
1435                         p->vfork_done = &vfork;
1436                         init_completion(&vfork);
1437                 }
1438
1439                 audit_finish_fork(p);
1440                 tracehook_report_clone(regs, clone_flags, nr, p);
1441
1442                 /*
1443                  * We set PF_STARTING at creation in case tracing wants to
1444                  * use this to distinguish a fully live task from one that
1445                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1446                  * clear it and set the child going.
1447                  */
1448                 p->flags &= ~PF_STARTING;
1449
1450                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1451                         /*
1452                          * We'll start up with an immediate SIGSTOP.
1453                          */
1454                         sigaddset(&p->pending.signal, SIGSTOP);
1455                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1456                         __set_task_state(p, TASK_STOPPED);
1457                 } else {
1458                         wake_up_new_task(p, clone_flags);
1459                 }
1460
1461                 tracehook_report_clone_complete(trace, regs,
1462                                                 clone_flags, nr, p);
1463
1464                 if (clone_flags & CLONE_VFORK) {
1465                         freezer_do_not_count();
1466                         wait_for_completion(&vfork);
1467                         freezer_count();
1468                         tracehook_report_vfork_done(p, nr);
1469                 }
1470         } else {
1471                 nr = PTR_ERR(p);
1472         }
1473         return nr;
1474 }
1475
1476 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1477 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1478 #endif
1479
1480 static void sighand_ctor(void *data)
1481 {
1482         struct sighand_struct *sighand = data;
1483
1484         spin_lock_init(&sighand->siglock);
1485         init_waitqueue_head(&sighand->signalfd_wqh);
1486 }
1487
1488 void __init proc_caches_init(void)
1489 {
1490         sighand_cachep = kmem_cache_create("sighand_cache",
1491                         sizeof(struct sighand_struct), 0,
1492                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1493                         SLAB_NOTRACK, sighand_ctor);
1494         signal_cachep = kmem_cache_create("signal_cache",
1495                         sizeof(struct signal_struct), 0,
1496                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1497         files_cachep = kmem_cache_create("files_cache",
1498                         sizeof(struct files_struct), 0,
1499                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1500         fs_cachep = kmem_cache_create("fs_cache",
1501                         sizeof(struct fs_struct), 0,
1502                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1503         mm_cachep = kmem_cache_create("mm_struct",
1504                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1505                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1506         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1507         mmap_init();
1508 }
1509
1510 /*
1511  * Check constraints on flags passed to the unshare system call and
1512  * force unsharing of additional process context as appropriate.
1513  */
1514 static void check_unshare_flags(unsigned long *flags_ptr)
1515 {
1516         /*
1517          * If unsharing a thread from a thread group, must also
1518          * unshare vm.
1519          */
1520         if (*flags_ptr & CLONE_THREAD)
1521                 *flags_ptr |= CLONE_VM;
1522
1523         /*
1524          * If unsharing vm, must also unshare signal handlers.
1525          */
1526         if (*flags_ptr & CLONE_VM)
1527                 *flags_ptr |= CLONE_SIGHAND;
1528
1529         /*
1530          * If unsharing namespace, must also unshare filesystem information.
1531          */
1532         if (*flags_ptr & CLONE_NEWNS)
1533                 *flags_ptr |= CLONE_FS;
1534 }
1535
1536 /*
1537  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1538  */
1539 static int unshare_thread(unsigned long unshare_flags)
1540 {
1541         if (unshare_flags & CLONE_THREAD)
1542                 return -EINVAL;
1543
1544         return 0;
1545 }
1546
1547 /*
1548  * Unshare the filesystem structure if it is being shared
1549  */
1550 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1551 {
1552         struct fs_struct *fs = current->fs;
1553
1554         if (!(unshare_flags & CLONE_FS) || !fs)
1555                 return 0;
1556
1557         /* don't need lock here; in the worst case we'll do useless copy */
1558         if (fs->users == 1)
1559                 return 0;
1560
1561         *new_fsp = copy_fs_struct(fs);
1562         if (!*new_fsp)
1563                 return -ENOMEM;
1564
1565         return 0;
1566 }
1567
1568 /*
1569  * Unsharing of sighand is not supported yet
1570  */
1571 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1572 {
1573         struct sighand_struct *sigh = current->sighand;
1574
1575         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1576                 return -EINVAL;
1577         else
1578                 return 0;
1579 }
1580
1581 /*
1582  * Unshare vm if it is being shared
1583  */
1584 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1585 {
1586         struct mm_struct *mm = current->mm;
1587
1588         if ((unshare_flags & CLONE_VM) &&
1589             (mm && atomic_read(&mm->mm_users) > 1)) {
1590                 return -EINVAL;
1591         }
1592
1593         return 0;
1594 }
1595
1596 /*
1597  * Unshare file descriptor table if it is being shared
1598  */
1599 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1600 {
1601         struct files_struct *fd = current->files;
1602         int error = 0;
1603
1604         if ((unshare_flags & CLONE_FILES) &&
1605             (fd && atomic_read(&fd->count) > 1)) {
1606                 *new_fdp = dup_fd(fd, &error);
1607                 if (!*new_fdp)
1608                         return error;
1609         }
1610
1611         return 0;
1612 }
1613
1614 /*
1615  * unshare allows a process to 'unshare' part of the process
1616  * context which was originally shared using clone.  copy_*
1617  * functions used by do_fork() cannot be used here directly
1618  * because they modify an inactive task_struct that is being
1619  * constructed. Here we are modifying the current, active,
1620  * task_struct.
1621  */
1622 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1623 {
1624         int err = 0;
1625         struct fs_struct *fs, *new_fs = NULL;
1626         struct sighand_struct *new_sigh = NULL;
1627         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1628         struct files_struct *fd, *new_fd = NULL;
1629         struct nsproxy *new_nsproxy = NULL;
1630         int do_sysvsem = 0;
1631
1632         check_unshare_flags(&unshare_flags);
1633
1634         /* Return -EINVAL for all unsupported flags */
1635         err = -EINVAL;
1636         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1637                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1638                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1639                 goto bad_unshare_out;
1640
1641         /*
1642          * CLONE_NEWIPC must also detach from the undolist: after switching
1643          * to a new ipc namespace, the semaphore arrays from the old
1644          * namespace are unreachable.
1645          */
1646         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1647                 do_sysvsem = 1;
1648         if ((err = unshare_thread(unshare_flags)))
1649                 goto bad_unshare_out;
1650         if ((err = unshare_fs(unshare_flags, &new_fs)))
1651                 goto bad_unshare_cleanup_thread;
1652         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1653                 goto bad_unshare_cleanup_fs;
1654         if ((err = unshare_vm(unshare_flags, &new_mm)))
1655                 goto bad_unshare_cleanup_sigh;
1656         if ((err = unshare_fd(unshare_flags, &new_fd)))
1657                 goto bad_unshare_cleanup_vm;
1658         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1659                         new_fs)))
1660                 goto bad_unshare_cleanup_fd;
1661
1662         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1663                 if (do_sysvsem) {
1664                         /*
1665                          * CLONE_SYSVSEM is equivalent to sys_exit().
1666                          */
1667                         exit_sem(current);
1668                 }
1669
1670                 if (new_nsproxy) {
1671                         switch_task_namespaces(current, new_nsproxy);
1672                         new_nsproxy = NULL;
1673                 }
1674
1675                 task_lock(current);
1676
1677                 if (new_fs) {
1678                         fs = current->fs;
1679                         spin_lock(&fs->lock);
1680                         current->fs = new_fs;
1681                         if (--fs->users)
1682                                 new_fs = NULL;
1683                         else
1684                                 new_fs = fs;
1685                         spin_unlock(&fs->lock);
1686                 }
1687
1688                 if (new_mm) {
1689                         mm = current->mm;
1690                         active_mm = current->active_mm;
1691                         current->mm = new_mm;
1692                         current->active_mm = new_mm;
1693                         activate_mm(active_mm, new_mm);
1694                         new_mm = mm;
1695                 }
1696
1697                 if (new_fd) {
1698                         fd = current->files;
1699                         current->files = new_fd;
1700                         new_fd = fd;
1701                 }
1702
1703                 task_unlock(current);
1704         }
1705
1706         if (new_nsproxy)
1707                 put_nsproxy(new_nsproxy);
1708
1709 bad_unshare_cleanup_fd:
1710         if (new_fd)
1711                 put_files_struct(new_fd);
1712
1713 bad_unshare_cleanup_vm:
1714         if (new_mm)
1715                 mmput(new_mm);
1716
1717 bad_unshare_cleanup_sigh:
1718         if (new_sigh)
1719                 if (atomic_dec_and_test(&new_sigh->count))
1720                         kmem_cache_free(sighand_cachep, new_sigh);
1721
1722 bad_unshare_cleanup_fs:
1723         if (new_fs)
1724                 free_fs_struct(new_fs);
1725
1726 bad_unshare_cleanup_thread:
1727 bad_unshare_out:
1728         return err;
1729 }
1730
1731 /*
1732  *      Helper to unshare the files of the current task.
1733  *      We don't want to expose copy_files internals to
1734  *      the exec layer of the kernel.
1735  */
1736
1737 int unshare_files(struct files_struct **displaced)
1738 {
1739         struct task_struct *task = current;
1740         struct files_struct *copy = NULL;
1741         int error;
1742
1743         error = unshare_fd(CLONE_FILES, &copy);
1744         if (error || !copy) {
1745                 *displaced = NULL;
1746                 return error;
1747         }
1748         *displaced = task->files;
1749         task_lock(task);
1750         task->files = copy;
1751         task_unlock(task);
1752         return 0;
1753 }