17bbf093356d085123b83fa93c05a89e380fb53e
[linux-3.10.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 EXPORT_SYMBOL_GPL(tasklist_lock);
90
91 int nr_processes(void)
92 {
93         int cpu;
94         int total = 0;
95
96         for_each_possible_cpu(cpu)
97                 total += per_cpu(process_counts, cpu);
98
99         return total;
100 }
101
102 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
103 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
104 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
105 static struct kmem_cache *task_struct_cachep;
106 #endif
107
108 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
109 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
110 {
111 #ifdef CONFIG_DEBUG_STACK_USAGE
112         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
113 #else
114         gfp_t mask = GFP_KERNEL;
115 #endif
116         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
117 }
118
119 static inline void free_thread_info(struct thread_info *ti)
120 {
121         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
122 }
123 #endif
124
125 /* SLAB cache for signal_struct structures (tsk->signal) */
126 static struct kmem_cache *signal_cachep;
127
128 /* SLAB cache for sighand_struct structures (tsk->sighand) */
129 struct kmem_cache *sighand_cachep;
130
131 /* SLAB cache for files_struct structures (tsk->files) */
132 struct kmem_cache *files_cachep;
133
134 /* SLAB cache for fs_struct structures (tsk->fs) */
135 struct kmem_cache *fs_cachep;
136
137 /* SLAB cache for vm_area_struct structures */
138 struct kmem_cache *vm_area_cachep;
139
140 /* SLAB cache for mm_struct structures (tsk->mm) */
141 static struct kmem_cache *mm_cachep;
142
143 static void account_kernel_stack(struct thread_info *ti, int account)
144 {
145         struct zone *zone = page_zone(virt_to_page(ti));
146
147         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
148 }
149
150 void free_task(struct task_struct *tsk)
151 {
152         prop_local_destroy_single(&tsk->dirties);
153         account_kernel_stack(tsk->stack, -1);
154         free_thread_info(tsk->stack);
155         rt_mutex_debug_task_free(tsk);
156         ftrace_graph_exit_task(tsk);
157         free_task_struct(tsk);
158 }
159 EXPORT_SYMBOL(free_task);
160
161 void __put_task_struct(struct task_struct *tsk)
162 {
163         WARN_ON(!tsk->exit_state);
164         WARN_ON(atomic_read(&tsk->usage));
165         WARN_ON(tsk == current);
166
167         exit_creds(tsk);
168         delayacct_tsk_free(tsk);
169
170         if (!profile_handoff_task(tsk))
171                 free_task(tsk);
172 }
173
174 /*
175  * macro override instead of weak attribute alias, to workaround
176  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
177  */
178 #ifndef arch_task_cache_init
179 #define arch_task_cache_init()
180 #endif
181
182 void __init fork_init(unsigned long mempages)
183 {
184 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
185 #ifndef ARCH_MIN_TASKALIGN
186 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
187 #endif
188         /* create a slab on which task_structs can be allocated */
189         task_struct_cachep =
190                 kmem_cache_create("task_struct", sizeof(struct task_struct),
191                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
192 #endif
193
194         /* do the arch specific task caches init */
195         arch_task_cache_init();
196
197         /*
198          * The default maximum number of threads is set to a safe
199          * value: the thread structures can take up at most half
200          * of memory.
201          */
202         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
203
204         /*
205          * we need to allow at least 20 threads to boot a system
206          */
207         if(max_threads < 20)
208                 max_threads = 20;
209
210         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
211         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
212         init_task.signal->rlim[RLIMIT_SIGPENDING] =
213                 init_task.signal->rlim[RLIMIT_NPROC];
214 }
215
216 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
217                                                struct task_struct *src)
218 {
219         *dst = *src;
220         return 0;
221 }
222
223 static struct task_struct *dup_task_struct(struct task_struct *orig)
224 {
225         struct task_struct *tsk;
226         struct thread_info *ti;
227         unsigned long *stackend;
228
229         int err;
230
231         prepare_to_copy(orig);
232
233         tsk = alloc_task_struct();
234         if (!tsk)
235                 return NULL;
236
237         ti = alloc_thread_info(tsk);
238         if (!ti) {
239                 free_task_struct(tsk);
240                 return NULL;
241         }
242
243         err = arch_dup_task_struct(tsk, orig);
244         if (err)
245                 goto out;
246
247         tsk->stack = ti;
248
249         err = prop_local_init_single(&tsk->dirties);
250         if (err)
251                 goto out;
252
253         setup_thread_stack(tsk, orig);
254         clear_user_return_notifier(tsk);
255         stackend = end_of_stack(tsk);
256         *stackend = STACK_END_MAGIC;    /* for overflow detection */
257
258 #ifdef CONFIG_CC_STACKPROTECTOR
259         tsk->stack_canary = get_random_int();
260 #endif
261
262         /* One for us, one for whoever does the "release_task()" (usually parent) */
263         atomic_set(&tsk->usage,2);
264         atomic_set(&tsk->fs_excl, 0);
265 #ifdef CONFIG_BLK_DEV_IO_TRACE
266         tsk->btrace_seq = 0;
267 #endif
268         tsk->splice_pipe = NULL;
269
270         account_kernel_stack(ti, 1);
271
272         return tsk;
273
274 out:
275         free_thread_info(ti);
276         free_task_struct(tsk);
277         return NULL;
278 }
279
280 #ifdef CONFIG_MMU
281 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
282 {
283         struct vm_area_struct *mpnt, *tmp, **pprev;
284         struct rb_node **rb_link, *rb_parent;
285         int retval;
286         unsigned long charge;
287         struct mempolicy *pol;
288
289         down_write(&oldmm->mmap_sem);
290         flush_cache_dup_mm(oldmm);
291         /*
292          * Not linked in yet - no deadlock potential:
293          */
294         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
295
296         mm->locked_vm = 0;
297         mm->mmap = NULL;
298         mm->mmap_cache = NULL;
299         mm->free_area_cache = oldmm->mmap_base;
300         mm->cached_hole_size = ~0UL;
301         mm->map_count = 0;
302         cpumask_clear(mm_cpumask(mm));
303         mm->mm_rb = RB_ROOT;
304         rb_link = &mm->mm_rb.rb_node;
305         rb_parent = NULL;
306         pprev = &mm->mmap;
307         retval = ksm_fork(mm, oldmm);
308         if (retval)
309                 goto out;
310
311         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
312                 struct file *file;
313
314                 if (mpnt->vm_flags & VM_DONTCOPY) {
315                         long pages = vma_pages(mpnt);
316                         mm->total_vm -= pages;
317                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
318                                                                 -pages);
319                         continue;
320                 }
321                 charge = 0;
322                 if (mpnt->vm_flags & VM_ACCOUNT) {
323                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
324                         if (security_vm_enough_memory(len))
325                                 goto fail_nomem;
326                         charge = len;
327                 }
328                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
329                 if (!tmp)
330                         goto fail_nomem;
331                 *tmp = *mpnt;
332                 pol = mpol_dup(vma_policy(mpnt));
333                 retval = PTR_ERR(pol);
334                 if (IS_ERR(pol))
335                         goto fail_nomem_policy;
336                 vma_set_policy(tmp, pol);
337                 tmp->vm_flags &= ~VM_LOCKED;
338                 tmp->vm_mm = mm;
339                 tmp->vm_next = NULL;
340                 anon_vma_link(tmp);
341                 file = tmp->vm_file;
342                 if (file) {
343                         struct inode *inode = file->f_path.dentry->d_inode;
344                         struct address_space *mapping = file->f_mapping;
345
346                         get_file(file);
347                         if (tmp->vm_flags & VM_DENYWRITE)
348                                 atomic_dec(&inode->i_writecount);
349                         spin_lock(&mapping->i_mmap_lock);
350                         if (tmp->vm_flags & VM_SHARED)
351                                 mapping->i_mmap_writable++;
352                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
353                         flush_dcache_mmap_lock(mapping);
354                         /* insert tmp into the share list, just after mpnt */
355                         vma_prio_tree_add(tmp, mpnt);
356                         flush_dcache_mmap_unlock(mapping);
357                         spin_unlock(&mapping->i_mmap_lock);
358                 }
359
360                 /*
361                  * Clear hugetlb-related page reserves for children. This only
362                  * affects MAP_PRIVATE mappings. Faults generated by the child
363                  * are not guaranteed to succeed, even if read-only
364                  */
365                 if (is_vm_hugetlb_page(tmp))
366                         reset_vma_resv_huge_pages(tmp);
367
368                 /*
369                  * Link in the new vma and copy the page table entries.
370                  */
371                 *pprev = tmp;
372                 pprev = &tmp->vm_next;
373
374                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
375                 rb_link = &tmp->vm_rb.rb_right;
376                 rb_parent = &tmp->vm_rb;
377
378                 mm->map_count++;
379                 retval = copy_page_range(mm, oldmm, mpnt);
380
381                 if (tmp->vm_ops && tmp->vm_ops->open)
382                         tmp->vm_ops->open(tmp);
383
384                 if (retval)
385                         goto out;
386         }
387         /* a new mm has just been created */
388         arch_dup_mmap(oldmm, mm);
389         retval = 0;
390 out:
391         up_write(&mm->mmap_sem);
392         flush_tlb_mm(oldmm);
393         up_write(&oldmm->mmap_sem);
394         return retval;
395 fail_nomem_policy:
396         kmem_cache_free(vm_area_cachep, tmp);
397 fail_nomem:
398         retval = -ENOMEM;
399         vm_unacct_memory(charge);
400         goto out;
401 }
402
403 static inline int mm_alloc_pgd(struct mm_struct * mm)
404 {
405         mm->pgd = pgd_alloc(mm);
406         if (unlikely(!mm->pgd))
407                 return -ENOMEM;
408         return 0;
409 }
410
411 static inline void mm_free_pgd(struct mm_struct * mm)
412 {
413         pgd_free(mm, mm->pgd);
414 }
415 #else
416 #define dup_mmap(mm, oldmm)     (0)
417 #define mm_alloc_pgd(mm)        (0)
418 #define mm_free_pgd(mm)
419 #endif /* CONFIG_MMU */
420
421 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
422
423 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
424 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
425
426 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
427
428 static int __init coredump_filter_setup(char *s)
429 {
430         default_dump_filter =
431                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
432                 MMF_DUMP_FILTER_MASK;
433         return 1;
434 }
435
436 __setup("coredump_filter=", coredump_filter_setup);
437
438 #include <linux/init_task.h>
439
440 static void mm_init_aio(struct mm_struct *mm)
441 {
442 #ifdef CONFIG_AIO
443         spin_lock_init(&mm->ioctx_lock);
444         INIT_HLIST_HEAD(&mm->ioctx_list);
445 #endif
446 }
447
448 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
449 {
450         atomic_set(&mm->mm_users, 1);
451         atomic_set(&mm->mm_count, 1);
452         init_rwsem(&mm->mmap_sem);
453         INIT_LIST_HEAD(&mm->mmlist);
454         mm->flags = (current->mm) ?
455                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
456         mm->core_state = NULL;
457         mm->nr_ptes = 0;
458         set_mm_counter(mm, file_rss, 0);
459         set_mm_counter(mm, anon_rss, 0);
460         spin_lock_init(&mm->page_table_lock);
461         mm->free_area_cache = TASK_UNMAPPED_BASE;
462         mm->cached_hole_size = ~0UL;
463         mm_init_aio(mm);
464         mm_init_owner(mm, p);
465
466         if (likely(!mm_alloc_pgd(mm))) {
467                 mm->def_flags = 0;
468                 mmu_notifier_mm_init(mm);
469                 return mm;
470         }
471
472         free_mm(mm);
473         return NULL;
474 }
475
476 /*
477  * Allocate and initialize an mm_struct.
478  */
479 struct mm_struct * mm_alloc(void)
480 {
481         struct mm_struct * mm;
482
483         mm = allocate_mm();
484         if (mm) {
485                 memset(mm, 0, sizeof(*mm));
486                 mm = mm_init(mm, current);
487         }
488         return mm;
489 }
490
491 /*
492  * Called when the last reference to the mm
493  * is dropped: either by a lazy thread or by
494  * mmput. Free the page directory and the mm.
495  */
496 void __mmdrop(struct mm_struct *mm)
497 {
498         BUG_ON(mm == &init_mm);
499         mm_free_pgd(mm);
500         destroy_context(mm);
501         mmu_notifier_mm_destroy(mm);
502         free_mm(mm);
503 }
504 EXPORT_SYMBOL_GPL(__mmdrop);
505
506 /*
507  * Decrement the use count and release all resources for an mm.
508  */
509 void mmput(struct mm_struct *mm)
510 {
511         might_sleep();
512
513         if (atomic_dec_and_test(&mm->mm_users)) {
514                 exit_aio(mm);
515                 ksm_exit(mm);
516                 exit_mmap(mm);
517                 set_mm_exe_file(mm, NULL);
518                 if (!list_empty(&mm->mmlist)) {
519                         spin_lock(&mmlist_lock);
520                         list_del(&mm->mmlist);
521                         spin_unlock(&mmlist_lock);
522                 }
523                 put_swap_token(mm);
524                 if (mm->binfmt)
525                         module_put(mm->binfmt->module);
526                 mmdrop(mm);
527         }
528 }
529 EXPORT_SYMBOL_GPL(mmput);
530
531 /**
532  * get_task_mm - acquire a reference to the task's mm
533  *
534  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
535  * this kernel workthread has transiently adopted a user mm with use_mm,
536  * to do its AIO) is not set and if so returns a reference to it, after
537  * bumping up the use count.  User must release the mm via mmput()
538  * after use.  Typically used by /proc and ptrace.
539  */
540 struct mm_struct *get_task_mm(struct task_struct *task)
541 {
542         struct mm_struct *mm;
543
544         task_lock(task);
545         mm = task->mm;
546         if (mm) {
547                 if (task->flags & PF_KTHREAD)
548                         mm = NULL;
549                 else
550                         atomic_inc(&mm->mm_users);
551         }
552         task_unlock(task);
553         return mm;
554 }
555 EXPORT_SYMBOL_GPL(get_task_mm);
556
557 /* Please note the differences between mmput and mm_release.
558  * mmput is called whenever we stop holding onto a mm_struct,
559  * error success whatever.
560  *
561  * mm_release is called after a mm_struct has been removed
562  * from the current process.
563  *
564  * This difference is important for error handling, when we
565  * only half set up a mm_struct for a new process and need to restore
566  * the old one.  Because we mmput the new mm_struct before
567  * restoring the old one. . .
568  * Eric Biederman 10 January 1998
569  */
570 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
571 {
572         struct completion *vfork_done = tsk->vfork_done;
573
574         /* Get rid of any futexes when releasing the mm */
575 #ifdef CONFIG_FUTEX
576         if (unlikely(tsk->robust_list)) {
577                 exit_robust_list(tsk);
578                 tsk->robust_list = NULL;
579         }
580 #ifdef CONFIG_COMPAT
581         if (unlikely(tsk->compat_robust_list)) {
582                 compat_exit_robust_list(tsk);
583                 tsk->compat_robust_list = NULL;
584         }
585 #endif
586         if (unlikely(!list_empty(&tsk->pi_state_list)))
587                 exit_pi_state_list(tsk);
588 #endif
589
590         /* Get rid of any cached register state */
591         deactivate_mm(tsk, mm);
592
593         /* notify parent sleeping on vfork() */
594         if (vfork_done) {
595                 tsk->vfork_done = NULL;
596                 complete(vfork_done);
597         }
598
599         /*
600          * If we're exiting normally, clear a user-space tid field if
601          * requested.  We leave this alone when dying by signal, to leave
602          * the value intact in a core dump, and to save the unnecessary
603          * trouble otherwise.  Userland only wants this done for a sys_exit.
604          */
605         if (tsk->clear_child_tid) {
606                 if (!(tsk->flags & PF_SIGNALED) &&
607                     atomic_read(&mm->mm_users) > 1) {
608                         /*
609                          * We don't check the error code - if userspace has
610                          * not set up a proper pointer then tough luck.
611                          */
612                         put_user(0, tsk->clear_child_tid);
613                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
614                                         1, NULL, NULL, 0);
615                 }
616                 tsk->clear_child_tid = NULL;
617         }
618 }
619
620 /*
621  * Allocate a new mm structure and copy contents from the
622  * mm structure of the passed in task structure.
623  */
624 struct mm_struct *dup_mm(struct task_struct *tsk)
625 {
626         struct mm_struct *mm, *oldmm = current->mm;
627         int err;
628
629         if (!oldmm)
630                 return NULL;
631
632         mm = allocate_mm();
633         if (!mm)
634                 goto fail_nomem;
635
636         memcpy(mm, oldmm, sizeof(*mm));
637
638         /* Initializing for Swap token stuff */
639         mm->token_priority = 0;
640         mm->last_interval = 0;
641
642         if (!mm_init(mm, tsk))
643                 goto fail_nomem;
644
645         if (init_new_context(tsk, mm))
646                 goto fail_nocontext;
647
648         dup_mm_exe_file(oldmm, mm);
649
650         err = dup_mmap(mm, oldmm);
651         if (err)
652                 goto free_pt;
653
654         mm->hiwater_rss = get_mm_rss(mm);
655         mm->hiwater_vm = mm->total_vm;
656
657         if (mm->binfmt && !try_module_get(mm->binfmt->module))
658                 goto free_pt;
659
660         return mm;
661
662 free_pt:
663         /* don't put binfmt in mmput, we haven't got module yet */
664         mm->binfmt = NULL;
665         mmput(mm);
666
667 fail_nomem:
668         return NULL;
669
670 fail_nocontext:
671         /*
672          * If init_new_context() failed, we cannot use mmput() to free the mm
673          * because it calls destroy_context()
674          */
675         mm_free_pgd(mm);
676         free_mm(mm);
677         return NULL;
678 }
679
680 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
681 {
682         struct mm_struct * mm, *oldmm;
683         int retval;
684
685         tsk->min_flt = tsk->maj_flt = 0;
686         tsk->nvcsw = tsk->nivcsw = 0;
687 #ifdef CONFIG_DETECT_HUNG_TASK
688         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
689 #endif
690
691         tsk->mm = NULL;
692         tsk->active_mm = NULL;
693
694         /*
695          * Are we cloning a kernel thread?
696          *
697          * We need to steal a active VM for that..
698          */
699         oldmm = current->mm;
700         if (!oldmm)
701                 return 0;
702
703         if (clone_flags & CLONE_VM) {
704                 atomic_inc(&oldmm->mm_users);
705                 mm = oldmm;
706                 goto good_mm;
707         }
708
709         retval = -ENOMEM;
710         mm = dup_mm(tsk);
711         if (!mm)
712                 goto fail_nomem;
713
714 good_mm:
715         /* Initializing for Swap token stuff */
716         mm->token_priority = 0;
717         mm->last_interval = 0;
718
719         tsk->mm = mm;
720         tsk->active_mm = mm;
721         return 0;
722
723 fail_nomem:
724         return retval;
725 }
726
727 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
728 {
729         struct fs_struct *fs = current->fs;
730         if (clone_flags & CLONE_FS) {
731                 /* tsk->fs is already what we want */
732                 write_lock(&fs->lock);
733                 if (fs->in_exec) {
734                         write_unlock(&fs->lock);
735                         return -EAGAIN;
736                 }
737                 fs->users++;
738                 write_unlock(&fs->lock);
739                 return 0;
740         }
741         tsk->fs = copy_fs_struct(fs);
742         if (!tsk->fs)
743                 return -ENOMEM;
744         return 0;
745 }
746
747 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
748 {
749         struct files_struct *oldf, *newf;
750         int error = 0;
751
752         /*
753          * A background process may not have any files ...
754          */
755         oldf = current->files;
756         if (!oldf)
757                 goto out;
758
759         if (clone_flags & CLONE_FILES) {
760                 atomic_inc(&oldf->count);
761                 goto out;
762         }
763
764         newf = dup_fd(oldf, &error);
765         if (!newf)
766                 goto out;
767
768         tsk->files = newf;
769         error = 0;
770 out:
771         return error;
772 }
773
774 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
775 {
776 #ifdef CONFIG_BLOCK
777         struct io_context *ioc = current->io_context;
778
779         if (!ioc)
780                 return 0;
781         /*
782          * Share io context with parent, if CLONE_IO is set
783          */
784         if (clone_flags & CLONE_IO) {
785                 tsk->io_context = ioc_task_link(ioc);
786                 if (unlikely(!tsk->io_context))
787                         return -ENOMEM;
788         } else if (ioprio_valid(ioc->ioprio)) {
789                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
790                 if (unlikely(!tsk->io_context))
791                         return -ENOMEM;
792
793                 tsk->io_context->ioprio = ioc->ioprio;
794         }
795 #endif
796         return 0;
797 }
798
799 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
800 {
801         struct sighand_struct *sig;
802
803         if (clone_flags & CLONE_SIGHAND) {
804                 atomic_inc(&current->sighand->count);
805                 return 0;
806         }
807         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
808         rcu_assign_pointer(tsk->sighand, sig);
809         if (!sig)
810                 return -ENOMEM;
811         atomic_set(&sig->count, 1);
812         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
813         return 0;
814 }
815
816 void __cleanup_sighand(struct sighand_struct *sighand)
817 {
818         if (atomic_dec_and_test(&sighand->count))
819                 kmem_cache_free(sighand_cachep, sighand);
820 }
821
822
823 /*
824  * Initialize POSIX timer handling for a thread group.
825  */
826 static void posix_cpu_timers_init_group(struct signal_struct *sig)
827 {
828         /* Thread group counters. */
829         thread_group_cputime_init(sig);
830
831         /* Expiration times and increments. */
832         sig->it[CPUCLOCK_PROF].expires = cputime_zero;
833         sig->it[CPUCLOCK_PROF].incr = cputime_zero;
834         sig->it[CPUCLOCK_VIRT].expires = cputime_zero;
835         sig->it[CPUCLOCK_VIRT].incr = cputime_zero;
836
837         /* Cached expiration times. */
838         sig->cputime_expires.prof_exp = cputime_zero;
839         sig->cputime_expires.virt_exp = cputime_zero;
840         sig->cputime_expires.sched_exp = 0;
841
842         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
843                 sig->cputime_expires.prof_exp =
844                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
845                 sig->cputimer.running = 1;
846         }
847
848         /* The timer lists. */
849         INIT_LIST_HEAD(&sig->cpu_timers[0]);
850         INIT_LIST_HEAD(&sig->cpu_timers[1]);
851         INIT_LIST_HEAD(&sig->cpu_timers[2]);
852 }
853
854 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
855 {
856         struct signal_struct *sig;
857
858         if (clone_flags & CLONE_THREAD)
859                 return 0;
860
861         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
862         tsk->signal = sig;
863         if (!sig)
864                 return -ENOMEM;
865
866         atomic_set(&sig->count, 1);
867         atomic_set(&sig->live, 1);
868         init_waitqueue_head(&sig->wait_chldexit);
869         sig->flags = 0;
870         if (clone_flags & CLONE_NEWPID)
871                 sig->flags |= SIGNAL_UNKILLABLE;
872         sig->group_exit_code = 0;
873         sig->group_exit_task = NULL;
874         sig->group_stop_count = 0;
875         sig->curr_target = tsk;
876         init_sigpending(&sig->shared_pending);
877         INIT_LIST_HEAD(&sig->posix_timers);
878
879         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
880         sig->it_real_incr.tv64 = 0;
881         sig->real_timer.function = it_real_fn;
882
883         sig->leader = 0;        /* session leadership doesn't inherit */
884         sig->tty_old_pgrp = NULL;
885         sig->tty = NULL;
886
887         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
888         sig->gtime = cputime_zero;
889         sig->cgtime = cputime_zero;
890 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
891         sig->prev_utime = sig->prev_stime = cputime_zero;
892 #endif
893         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
894         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
895         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
896         sig->maxrss = sig->cmaxrss = 0;
897         task_io_accounting_init(&sig->ioac);
898         sig->sum_sched_runtime = 0;
899         taskstats_tgid_init(sig);
900
901         task_lock(current->group_leader);
902         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
903         task_unlock(current->group_leader);
904
905         posix_cpu_timers_init_group(sig);
906
907         acct_init_pacct(&sig->pacct);
908
909         tty_audit_fork(sig);
910
911         sig->oom_adj = current->signal->oom_adj;
912
913         return 0;
914 }
915
916 void __cleanup_signal(struct signal_struct *sig)
917 {
918         thread_group_cputime_free(sig);
919         tty_kref_put(sig->tty);
920         kmem_cache_free(signal_cachep, sig);
921 }
922
923 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
924 {
925         unsigned long new_flags = p->flags;
926
927         new_flags &= ~PF_SUPERPRIV;
928         new_flags |= PF_FORKNOEXEC;
929         new_flags |= PF_STARTING;
930         p->flags = new_flags;
931         clear_freeze_flag(p);
932 }
933
934 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
935 {
936         current->clear_child_tid = tidptr;
937
938         return task_pid_vnr(current);
939 }
940
941 static void rt_mutex_init_task(struct task_struct *p)
942 {
943         raw_spin_lock_init(&p->pi_lock);
944 #ifdef CONFIG_RT_MUTEXES
945         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
946         p->pi_blocked_on = NULL;
947 #endif
948 }
949
950 #ifdef CONFIG_MM_OWNER
951 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
952 {
953         mm->owner = p;
954 }
955 #endif /* CONFIG_MM_OWNER */
956
957 /*
958  * Initialize POSIX timer handling for a single task.
959  */
960 static void posix_cpu_timers_init(struct task_struct *tsk)
961 {
962         tsk->cputime_expires.prof_exp = cputime_zero;
963         tsk->cputime_expires.virt_exp = cputime_zero;
964         tsk->cputime_expires.sched_exp = 0;
965         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
966         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
967         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
968 }
969
970 /*
971  * This creates a new process as a copy of the old one,
972  * but does not actually start it yet.
973  *
974  * It copies the registers, and all the appropriate
975  * parts of the process environment (as per the clone
976  * flags). The actual kick-off is left to the caller.
977  */
978 static struct task_struct *copy_process(unsigned long clone_flags,
979                                         unsigned long stack_start,
980                                         struct pt_regs *regs,
981                                         unsigned long stack_size,
982                                         int __user *child_tidptr,
983                                         struct pid *pid,
984                                         int trace)
985 {
986         int retval;
987         struct task_struct *p;
988         int cgroup_callbacks_done = 0;
989
990         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
991                 return ERR_PTR(-EINVAL);
992
993         /*
994          * Thread groups must share signals as well, and detached threads
995          * can only be started up within the thread group.
996          */
997         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
998                 return ERR_PTR(-EINVAL);
999
1000         /*
1001          * Shared signal handlers imply shared VM. By way of the above,
1002          * thread groups also imply shared VM. Blocking this case allows
1003          * for various simplifications in other code.
1004          */
1005         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1006                 return ERR_PTR(-EINVAL);
1007
1008         /*
1009          * Siblings of global init remain as zombies on exit since they are
1010          * not reaped by their parent (swapper). To solve this and to avoid
1011          * multi-rooted process trees, prevent global and container-inits
1012          * from creating siblings.
1013          */
1014         if ((clone_flags & CLONE_PARENT) &&
1015                                 current->signal->flags & SIGNAL_UNKILLABLE)
1016                 return ERR_PTR(-EINVAL);
1017
1018         retval = security_task_create(clone_flags);
1019         if (retval)
1020                 goto fork_out;
1021
1022         retval = -ENOMEM;
1023         p = dup_task_struct(current);
1024         if (!p)
1025                 goto fork_out;
1026
1027         ftrace_graph_init_task(p);
1028
1029         rt_mutex_init_task(p);
1030
1031 #ifdef CONFIG_PROVE_LOCKING
1032         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1033         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1034 #endif
1035         retval = -EAGAIN;
1036         if (atomic_read(&p->real_cred->user->processes) >=
1037                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1038                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1039                     p->real_cred->user != INIT_USER)
1040                         goto bad_fork_free;
1041         }
1042
1043         retval = copy_creds(p, clone_flags);
1044         if (retval < 0)
1045                 goto bad_fork_free;
1046
1047         /*
1048          * If multiple threads are within copy_process(), then this check
1049          * triggers too late. This doesn't hurt, the check is only there
1050          * to stop root fork bombs.
1051          */
1052         retval = -EAGAIN;
1053         if (nr_threads >= max_threads)
1054                 goto bad_fork_cleanup_count;
1055
1056         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1057                 goto bad_fork_cleanup_count;
1058
1059         p->did_exec = 0;
1060         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1061         copy_flags(clone_flags, p);
1062         INIT_LIST_HEAD(&p->children);
1063         INIT_LIST_HEAD(&p->sibling);
1064         rcu_copy_process(p);
1065         p->vfork_done = NULL;
1066         spin_lock_init(&p->alloc_lock);
1067
1068         init_sigpending(&p->pending);
1069
1070         p->utime = cputime_zero;
1071         p->stime = cputime_zero;
1072         p->gtime = cputime_zero;
1073         p->utimescaled = cputime_zero;
1074         p->stimescaled = cputime_zero;
1075 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1076         p->prev_utime = cputime_zero;
1077         p->prev_stime = cputime_zero;
1078 #endif
1079
1080         p->default_timer_slack_ns = current->timer_slack_ns;
1081
1082         task_io_accounting_init(&p->ioac);
1083         acct_clear_integrals(p);
1084
1085         posix_cpu_timers_init(p);
1086
1087         p->lock_depth = -1;             /* -1 = no lock */
1088         do_posix_clock_monotonic_gettime(&p->start_time);
1089         p->real_start_time = p->start_time;
1090         monotonic_to_bootbased(&p->real_start_time);
1091         p->io_context = NULL;
1092         p->audit_context = NULL;
1093         cgroup_fork(p);
1094 #ifdef CONFIG_NUMA
1095         p->mempolicy = mpol_dup(p->mempolicy);
1096         if (IS_ERR(p->mempolicy)) {
1097                 retval = PTR_ERR(p->mempolicy);
1098                 p->mempolicy = NULL;
1099                 goto bad_fork_cleanup_cgroup;
1100         }
1101         mpol_fix_fork_child_flag(p);
1102 #endif
1103 #ifdef CONFIG_TRACE_IRQFLAGS
1104         p->irq_events = 0;
1105 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1106         p->hardirqs_enabled = 1;
1107 #else
1108         p->hardirqs_enabled = 0;
1109 #endif
1110         p->hardirq_enable_ip = 0;
1111         p->hardirq_enable_event = 0;
1112         p->hardirq_disable_ip = _THIS_IP_;
1113         p->hardirq_disable_event = 0;
1114         p->softirqs_enabled = 1;
1115         p->softirq_enable_ip = _THIS_IP_;
1116         p->softirq_enable_event = 0;
1117         p->softirq_disable_ip = 0;
1118         p->softirq_disable_event = 0;
1119         p->hardirq_context = 0;
1120         p->softirq_context = 0;
1121 #endif
1122 #ifdef CONFIG_LOCKDEP
1123         p->lockdep_depth = 0; /* no locks held yet */
1124         p->curr_chain_key = 0;
1125         p->lockdep_recursion = 0;
1126 #endif
1127
1128 #ifdef CONFIG_DEBUG_MUTEXES
1129         p->blocked_on = NULL; /* not blocked yet */
1130 #endif
1131 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1132         p->memcg_batch.do_batch = 0;
1133         p->memcg_batch.memcg = NULL;
1134 #endif
1135
1136         p->bts = NULL;
1137
1138         p->stack_start = stack_start;
1139
1140         /* Perform scheduler related setup. Assign this task to a CPU. */
1141         sched_fork(p, clone_flags);
1142
1143         retval = perf_event_init_task(p);
1144         if (retval)
1145                 goto bad_fork_cleanup_policy;
1146
1147         if ((retval = audit_alloc(p)))
1148                 goto bad_fork_cleanup_policy;
1149         /* copy all the process information */
1150         if ((retval = copy_semundo(clone_flags, p)))
1151                 goto bad_fork_cleanup_audit;
1152         if ((retval = copy_files(clone_flags, p)))
1153                 goto bad_fork_cleanup_semundo;
1154         if ((retval = copy_fs(clone_flags, p)))
1155                 goto bad_fork_cleanup_files;
1156         if ((retval = copy_sighand(clone_flags, p)))
1157                 goto bad_fork_cleanup_fs;
1158         if ((retval = copy_signal(clone_flags, p)))
1159                 goto bad_fork_cleanup_sighand;
1160         if ((retval = copy_mm(clone_flags, p)))
1161                 goto bad_fork_cleanup_signal;
1162         if ((retval = copy_namespaces(clone_flags, p)))
1163                 goto bad_fork_cleanup_mm;
1164         if ((retval = copy_io(clone_flags, p)))
1165                 goto bad_fork_cleanup_namespaces;
1166         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1167         if (retval)
1168                 goto bad_fork_cleanup_io;
1169
1170         if (pid != &init_struct_pid) {
1171                 retval = -ENOMEM;
1172                 pid = alloc_pid(p->nsproxy->pid_ns);
1173                 if (!pid)
1174                         goto bad_fork_cleanup_io;
1175
1176                 if (clone_flags & CLONE_NEWPID) {
1177                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1178                         if (retval < 0)
1179                                 goto bad_fork_free_pid;
1180                 }
1181         }
1182
1183         p->pid = pid_nr(pid);
1184         p->tgid = p->pid;
1185         if (clone_flags & CLONE_THREAD)
1186                 p->tgid = current->tgid;
1187
1188         if (current->nsproxy != p->nsproxy) {
1189                 retval = ns_cgroup_clone(p, pid);
1190                 if (retval)
1191                         goto bad_fork_free_pid;
1192         }
1193
1194         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1195         /*
1196          * Clear TID on mm_release()?
1197          */
1198         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1199 #ifdef CONFIG_FUTEX
1200         p->robust_list = NULL;
1201 #ifdef CONFIG_COMPAT
1202         p->compat_robust_list = NULL;
1203 #endif
1204         INIT_LIST_HEAD(&p->pi_state_list);
1205         p->pi_state_cache = NULL;
1206 #endif
1207         /*
1208          * sigaltstack should be cleared when sharing the same VM
1209          */
1210         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1211                 p->sas_ss_sp = p->sas_ss_size = 0;
1212
1213         /*
1214          * Syscall tracing and stepping should be turned off in the
1215          * child regardless of CLONE_PTRACE.
1216          */
1217         user_disable_single_step(p);
1218         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1219 #ifdef TIF_SYSCALL_EMU
1220         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1221 #endif
1222         clear_all_latency_tracing(p);
1223
1224         /* ok, now we should be set up.. */
1225         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1226         p->pdeath_signal = 0;
1227         p->exit_state = 0;
1228
1229         /*
1230          * Ok, make it visible to the rest of the system.
1231          * We dont wake it up yet.
1232          */
1233         p->group_leader = p;
1234         INIT_LIST_HEAD(&p->thread_group);
1235
1236         /* Now that the task is set up, run cgroup callbacks if
1237          * necessary. We need to run them before the task is visible
1238          * on the tasklist. */
1239         cgroup_fork_callbacks(p);
1240         cgroup_callbacks_done = 1;
1241
1242         /* Need tasklist lock for parent etc handling! */
1243         write_lock_irq(&tasklist_lock);
1244
1245         /* CLONE_PARENT re-uses the old parent */
1246         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1247                 p->real_parent = current->real_parent;
1248                 p->parent_exec_id = current->parent_exec_id;
1249         } else {
1250                 p->real_parent = current;
1251                 p->parent_exec_id = current->self_exec_id;
1252         }
1253
1254         spin_lock(&current->sighand->siglock);
1255
1256         /*
1257          * Process group and session signals need to be delivered to just the
1258          * parent before the fork or both the parent and the child after the
1259          * fork. Restart if a signal comes in before we add the new process to
1260          * it's process group.
1261          * A fatal signal pending means that current will exit, so the new
1262          * thread can't slip out of an OOM kill (or normal SIGKILL).
1263          */
1264         recalc_sigpending();
1265         if (signal_pending(current)) {
1266                 spin_unlock(&current->sighand->siglock);
1267                 write_unlock_irq(&tasklist_lock);
1268                 retval = -ERESTARTNOINTR;
1269                 goto bad_fork_free_pid;
1270         }
1271
1272         if (clone_flags & CLONE_THREAD) {
1273                 atomic_inc(&current->signal->count);
1274                 atomic_inc(&current->signal->live);
1275                 p->group_leader = current->group_leader;
1276                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1277         }
1278
1279         if (likely(p->pid)) {
1280                 tracehook_finish_clone(p, clone_flags, trace);
1281
1282                 if (thread_group_leader(p)) {
1283                         if (clone_flags & CLONE_NEWPID)
1284                                 p->nsproxy->pid_ns->child_reaper = p;
1285
1286                         p->signal->leader_pid = pid;
1287                         tty_kref_put(p->signal->tty);
1288                         p->signal->tty = tty_kref_get(current->signal->tty);
1289                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1290                         attach_pid(p, PIDTYPE_SID, task_session(current));
1291                         list_add_tail(&p->sibling, &p->real_parent->children);
1292                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1293                         __get_cpu_var(process_counts)++;
1294                 }
1295                 attach_pid(p, PIDTYPE_PID, pid);
1296                 nr_threads++;
1297         }
1298
1299         total_forks++;
1300         spin_unlock(&current->sighand->siglock);
1301         write_unlock_irq(&tasklist_lock);
1302         proc_fork_connector(p);
1303         cgroup_post_fork(p);
1304         perf_event_fork(p);
1305         return p;
1306
1307 bad_fork_free_pid:
1308         if (pid != &init_struct_pid)
1309                 free_pid(pid);
1310 bad_fork_cleanup_io:
1311         if (p->io_context)
1312                 exit_io_context(p);
1313 bad_fork_cleanup_namespaces:
1314         exit_task_namespaces(p);
1315 bad_fork_cleanup_mm:
1316         if (p->mm)
1317                 mmput(p->mm);
1318 bad_fork_cleanup_signal:
1319         if (!(clone_flags & CLONE_THREAD))
1320                 __cleanup_signal(p->signal);
1321 bad_fork_cleanup_sighand:
1322         __cleanup_sighand(p->sighand);
1323 bad_fork_cleanup_fs:
1324         exit_fs(p); /* blocking */
1325 bad_fork_cleanup_files:
1326         exit_files(p); /* blocking */
1327 bad_fork_cleanup_semundo:
1328         exit_sem(p);
1329 bad_fork_cleanup_audit:
1330         audit_free(p);
1331 bad_fork_cleanup_policy:
1332         perf_event_free_task(p);
1333 #ifdef CONFIG_NUMA
1334         mpol_put(p->mempolicy);
1335 bad_fork_cleanup_cgroup:
1336 #endif
1337         cgroup_exit(p, cgroup_callbacks_done);
1338         delayacct_tsk_free(p);
1339         module_put(task_thread_info(p)->exec_domain->module);
1340 bad_fork_cleanup_count:
1341         atomic_dec(&p->cred->user->processes);
1342         exit_creds(p);
1343 bad_fork_free:
1344         free_task(p);
1345 fork_out:
1346         return ERR_PTR(retval);
1347 }
1348
1349 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1350 {
1351         memset(regs, 0, sizeof(struct pt_regs));
1352         return regs;
1353 }
1354
1355 struct task_struct * __cpuinit fork_idle(int cpu)
1356 {
1357         struct task_struct *task;
1358         struct pt_regs regs;
1359
1360         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1361                             &init_struct_pid, 0);
1362         if (!IS_ERR(task))
1363                 init_idle(task, cpu);
1364
1365         return task;
1366 }
1367
1368 /*
1369  *  Ok, this is the main fork-routine.
1370  *
1371  * It copies the process, and if successful kick-starts
1372  * it and waits for it to finish using the VM if required.
1373  */
1374 long do_fork(unsigned long clone_flags,
1375               unsigned long stack_start,
1376               struct pt_regs *regs,
1377               unsigned long stack_size,
1378               int __user *parent_tidptr,
1379               int __user *child_tidptr)
1380 {
1381         struct task_struct *p;
1382         int trace = 0;
1383         long nr;
1384
1385         /*
1386          * Do some preliminary argument and permissions checking before we
1387          * actually start allocating stuff
1388          */
1389         if (clone_flags & CLONE_NEWUSER) {
1390                 if (clone_flags & CLONE_THREAD)
1391                         return -EINVAL;
1392                 /* hopefully this check will go away when userns support is
1393                  * complete
1394                  */
1395                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1396                                 !capable(CAP_SETGID))
1397                         return -EPERM;
1398         }
1399
1400         /*
1401          * We hope to recycle these flags after 2.6.26
1402          */
1403         if (unlikely(clone_flags & CLONE_STOPPED)) {
1404                 static int __read_mostly count = 100;
1405
1406                 if (count > 0 && printk_ratelimit()) {
1407                         char comm[TASK_COMM_LEN];
1408
1409                         count--;
1410                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1411                                         "clone flags 0x%lx\n",
1412                                 get_task_comm(comm, current),
1413                                 clone_flags & CLONE_STOPPED);
1414                 }
1415         }
1416
1417         /*
1418          * When called from kernel_thread, don't do user tracing stuff.
1419          */
1420         if (likely(user_mode(regs)))
1421                 trace = tracehook_prepare_clone(clone_flags);
1422
1423         p = copy_process(clone_flags, stack_start, regs, stack_size,
1424                          child_tidptr, NULL, trace);
1425         /*
1426          * Do this prior waking up the new thread - the thread pointer
1427          * might get invalid after that point, if the thread exits quickly.
1428          */
1429         if (!IS_ERR(p)) {
1430                 struct completion vfork;
1431
1432                 trace_sched_process_fork(current, p);
1433
1434                 nr = task_pid_vnr(p);
1435
1436                 if (clone_flags & CLONE_PARENT_SETTID)
1437                         put_user(nr, parent_tidptr);
1438
1439                 if (clone_flags & CLONE_VFORK) {
1440                         p->vfork_done = &vfork;
1441                         init_completion(&vfork);
1442                 }
1443
1444                 audit_finish_fork(p);
1445                 tracehook_report_clone(regs, clone_flags, nr, p);
1446
1447                 /*
1448                  * We set PF_STARTING at creation in case tracing wants to
1449                  * use this to distinguish a fully live task from one that
1450                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1451                  * clear it and set the child going.
1452                  */
1453                 p->flags &= ~PF_STARTING;
1454
1455                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1456                         /*
1457                          * We'll start up with an immediate SIGSTOP.
1458                          */
1459                         sigaddset(&p->pending.signal, SIGSTOP);
1460                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1461                         __set_task_state(p, TASK_STOPPED);
1462                 } else {
1463                         wake_up_new_task(p, clone_flags);
1464                 }
1465
1466                 tracehook_report_clone_complete(trace, regs,
1467                                                 clone_flags, nr, p);
1468
1469                 if (clone_flags & CLONE_VFORK) {
1470                         freezer_do_not_count();
1471                         wait_for_completion(&vfork);
1472                         freezer_count();
1473                         tracehook_report_vfork_done(p, nr);
1474                 }
1475         } else {
1476                 nr = PTR_ERR(p);
1477         }
1478         return nr;
1479 }
1480
1481 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1482 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1483 #endif
1484
1485 static void sighand_ctor(void *data)
1486 {
1487         struct sighand_struct *sighand = data;
1488
1489         spin_lock_init(&sighand->siglock);
1490         init_waitqueue_head(&sighand->signalfd_wqh);
1491 }
1492
1493 void __init proc_caches_init(void)
1494 {
1495         sighand_cachep = kmem_cache_create("sighand_cache",
1496                         sizeof(struct sighand_struct), 0,
1497                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1498                         SLAB_NOTRACK, sighand_ctor);
1499         signal_cachep = kmem_cache_create("signal_cache",
1500                         sizeof(struct signal_struct), 0,
1501                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1502         files_cachep = kmem_cache_create("files_cache",
1503                         sizeof(struct files_struct), 0,
1504                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1505         fs_cachep = kmem_cache_create("fs_cache",
1506                         sizeof(struct fs_struct), 0,
1507                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1508         mm_cachep = kmem_cache_create("mm_struct",
1509                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1510                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1511         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1512         mmap_init();
1513 }
1514
1515 /*
1516  * Check constraints on flags passed to the unshare system call and
1517  * force unsharing of additional process context as appropriate.
1518  */
1519 static void check_unshare_flags(unsigned long *flags_ptr)
1520 {
1521         /*
1522          * If unsharing a thread from a thread group, must also
1523          * unshare vm.
1524          */
1525         if (*flags_ptr & CLONE_THREAD)
1526                 *flags_ptr |= CLONE_VM;
1527
1528         /*
1529          * If unsharing vm, must also unshare signal handlers.
1530          */
1531         if (*flags_ptr & CLONE_VM)
1532                 *flags_ptr |= CLONE_SIGHAND;
1533
1534         /*
1535          * If unsharing signal handlers and the task was created
1536          * using CLONE_THREAD, then must unshare the thread
1537          */
1538         if ((*flags_ptr & CLONE_SIGHAND) &&
1539             (atomic_read(&current->signal->count) > 1))
1540                 *flags_ptr |= CLONE_THREAD;
1541
1542         /*
1543          * If unsharing namespace, must also unshare filesystem information.
1544          */
1545         if (*flags_ptr & CLONE_NEWNS)
1546                 *flags_ptr |= CLONE_FS;
1547 }
1548
1549 /*
1550  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1551  */
1552 static int unshare_thread(unsigned long unshare_flags)
1553 {
1554         if (unshare_flags & CLONE_THREAD)
1555                 return -EINVAL;
1556
1557         return 0;
1558 }
1559
1560 /*
1561  * Unshare the filesystem structure if it is being shared
1562  */
1563 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1564 {
1565         struct fs_struct *fs = current->fs;
1566
1567         if (!(unshare_flags & CLONE_FS) || !fs)
1568                 return 0;
1569
1570         /* don't need lock here; in the worst case we'll do useless copy */
1571         if (fs->users == 1)
1572                 return 0;
1573
1574         *new_fsp = copy_fs_struct(fs);
1575         if (!*new_fsp)
1576                 return -ENOMEM;
1577
1578         return 0;
1579 }
1580
1581 /*
1582  * Unsharing of sighand is not supported yet
1583  */
1584 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1585 {
1586         struct sighand_struct *sigh = current->sighand;
1587
1588         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1589                 return -EINVAL;
1590         else
1591                 return 0;
1592 }
1593
1594 /*
1595  * Unshare vm if it is being shared
1596  */
1597 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1598 {
1599         struct mm_struct *mm = current->mm;
1600
1601         if ((unshare_flags & CLONE_VM) &&
1602             (mm && atomic_read(&mm->mm_users) > 1)) {
1603                 return -EINVAL;
1604         }
1605
1606         return 0;
1607 }
1608
1609 /*
1610  * Unshare file descriptor table if it is being shared
1611  */
1612 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1613 {
1614         struct files_struct *fd = current->files;
1615         int error = 0;
1616
1617         if ((unshare_flags & CLONE_FILES) &&
1618             (fd && atomic_read(&fd->count) > 1)) {
1619                 *new_fdp = dup_fd(fd, &error);
1620                 if (!*new_fdp)
1621                         return error;
1622         }
1623
1624         return 0;
1625 }
1626
1627 /*
1628  * unshare allows a process to 'unshare' part of the process
1629  * context which was originally shared using clone.  copy_*
1630  * functions used by do_fork() cannot be used here directly
1631  * because they modify an inactive task_struct that is being
1632  * constructed. Here we are modifying the current, active,
1633  * task_struct.
1634  */
1635 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1636 {
1637         int err = 0;
1638         struct fs_struct *fs, *new_fs = NULL;
1639         struct sighand_struct *new_sigh = NULL;
1640         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1641         struct files_struct *fd, *new_fd = NULL;
1642         struct nsproxy *new_nsproxy = NULL;
1643         int do_sysvsem = 0;
1644
1645         check_unshare_flags(&unshare_flags);
1646
1647         /* Return -EINVAL for all unsupported flags */
1648         err = -EINVAL;
1649         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1650                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1651                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1652                 goto bad_unshare_out;
1653
1654         /*
1655          * CLONE_NEWIPC must also detach from the undolist: after switching
1656          * to a new ipc namespace, the semaphore arrays from the old
1657          * namespace are unreachable.
1658          */
1659         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1660                 do_sysvsem = 1;
1661         if ((err = unshare_thread(unshare_flags)))
1662                 goto bad_unshare_out;
1663         if ((err = unshare_fs(unshare_flags, &new_fs)))
1664                 goto bad_unshare_cleanup_thread;
1665         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1666                 goto bad_unshare_cleanup_fs;
1667         if ((err = unshare_vm(unshare_flags, &new_mm)))
1668                 goto bad_unshare_cleanup_sigh;
1669         if ((err = unshare_fd(unshare_flags, &new_fd)))
1670                 goto bad_unshare_cleanup_vm;
1671         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1672                         new_fs)))
1673                 goto bad_unshare_cleanup_fd;
1674
1675         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1676                 if (do_sysvsem) {
1677                         /*
1678                          * CLONE_SYSVSEM is equivalent to sys_exit().
1679                          */
1680                         exit_sem(current);
1681                 }
1682
1683                 if (new_nsproxy) {
1684                         switch_task_namespaces(current, new_nsproxy);
1685                         new_nsproxy = NULL;
1686                 }
1687
1688                 task_lock(current);
1689
1690                 if (new_fs) {
1691                         fs = current->fs;
1692                         write_lock(&fs->lock);
1693                         current->fs = new_fs;
1694                         if (--fs->users)
1695                                 new_fs = NULL;
1696                         else
1697                                 new_fs = fs;
1698                         write_unlock(&fs->lock);
1699                 }
1700
1701                 if (new_mm) {
1702                         mm = current->mm;
1703                         active_mm = current->active_mm;
1704                         current->mm = new_mm;
1705                         current->active_mm = new_mm;
1706                         activate_mm(active_mm, new_mm);
1707                         new_mm = mm;
1708                 }
1709
1710                 if (new_fd) {
1711                         fd = current->files;
1712                         current->files = new_fd;
1713                         new_fd = fd;
1714                 }
1715
1716                 task_unlock(current);
1717         }
1718
1719         if (new_nsproxy)
1720                 put_nsproxy(new_nsproxy);
1721
1722 bad_unshare_cleanup_fd:
1723         if (new_fd)
1724                 put_files_struct(new_fd);
1725
1726 bad_unshare_cleanup_vm:
1727         if (new_mm)
1728                 mmput(new_mm);
1729
1730 bad_unshare_cleanup_sigh:
1731         if (new_sigh)
1732                 if (atomic_dec_and_test(&new_sigh->count))
1733                         kmem_cache_free(sighand_cachep, new_sigh);
1734
1735 bad_unshare_cleanup_fs:
1736         if (new_fs)
1737                 free_fs_struct(new_fs);
1738
1739 bad_unshare_cleanup_thread:
1740 bad_unshare_out:
1741         return err;
1742 }
1743
1744 /*
1745  *      Helper to unshare the files of the current task.
1746  *      We don't want to expose copy_files internals to
1747  *      the exec layer of the kernel.
1748  */
1749
1750 int unshare_files(struct files_struct **displaced)
1751 {
1752         struct task_struct *task = current;
1753         struct files_struct *copy = NULL;
1754         int error;
1755
1756         error = unshare_fd(CLONE_FILES, &copy);
1757         if (error || !copy) {
1758                 *displaced = NULL;
1759                 return error;
1760         }
1761         *displaced = task->files;
1762         task_lock(task);
1763         task->files = copy;
1764         task_unlock(task);
1765         return 0;
1766 }