a1793e442b20707fcc74e1aa85ce70b275a24acb
[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/futex.h>
41 #include <linux/compat.h>
42 #include <linux/kthread.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/blkdev.h>
62 #include <linux/fs_struct.h>
63 #include <linux/magic.h>
64 #include <linux/perf_event.h>
65 #include <linux/posix-timers.h>
66 #include <linux/user-return-notifier.h>
67 #include <linux/oom.h>
68 #include <linux/khugepaged.h>
69 #include <linux/signalfd.h>
70
71 #include <asm/pgtable.h>
72 #include <asm/pgalloc.h>
73 #include <asm/uaccess.h>
74 #include <asm/mmu_context.h>
75 #include <asm/cacheflush.h>
76 #include <asm/tlbflush.h>
77
78 #include <trace/events/sched.h>
79
80 #define CREATE_TRACE_POINTS
81 #include <trace/events/task.h>
82
83 /*
84  * Protected counters by write_lock_irq(&tasklist_lock)
85  */
86 unsigned long total_forks;      /* Handle normal Linux uptimes. */
87 int nr_threads;                 /* The idle threads do not count.. */
88
89 int max_threads;                /* tunable limit on nr_threads */
90
91 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
92
93 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
94
95 #ifdef CONFIG_PROVE_RCU
96 int lockdep_tasklist_lock_is_held(void)
97 {
98         return lockdep_is_held(&tasklist_lock);
99 }
100 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
101 #endif /* #ifdef CONFIG_PROVE_RCU */
102
103 int nr_processes(void)
104 {
105         int cpu;
106         int total = 0;
107
108         for_each_possible_cpu(cpu)
109                 total += per_cpu(process_counts, cpu);
110
111         return total;
112 }
113
114 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
115 static struct kmem_cache *task_struct_cachep;
116
117 static inline struct task_struct *alloc_task_struct_node(int node)
118 {
119         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
120 }
121
122 void __weak arch_release_task_struct(struct task_struct *tsk) { }
123
124 static inline void free_task_struct(struct task_struct *tsk)
125 {
126         arch_release_task_struct(tsk);
127         kmem_cache_free(task_struct_cachep, tsk);
128 }
129 #endif
130
131 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
132 void __weak arch_release_thread_info(struct thread_info *ti) { }
133
134 /*
135  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
136  * kmemcache based allocator.
137  */
138 # if THREAD_SIZE >= PAGE_SIZE
139 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
140                                                   int node)
141 {
142         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
143                                              THREAD_SIZE_ORDER);
144
145         return page ? page_address(page) : NULL;
146 }
147
148 static inline void free_thread_info(struct thread_info *ti)
149 {
150         arch_release_thread_info(ti);
151         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
152 }
153 # else
154 static struct kmem_cache *thread_info_cache;
155
156 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
157                                                   int node)
158 {
159         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
160 }
161
162 static void free_thread_info(struct thread_info *ti)
163 {
164         arch_release_thread_info(ti);
165         kmem_cache_free(thread_info_cache, ti);
166 }
167
168 void thread_info_cache_init(void)
169 {
170         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
171                                               THREAD_SIZE, 0, NULL);
172         BUG_ON(thread_info_cache == NULL);
173 }
174 # endif
175 #endif
176
177 /* SLAB cache for signal_struct structures (tsk->signal) */
178 static struct kmem_cache *signal_cachep;
179
180 /* SLAB cache for sighand_struct structures (tsk->sighand) */
181 struct kmem_cache *sighand_cachep;
182
183 /* SLAB cache for files_struct structures (tsk->files) */
184 struct kmem_cache *files_cachep;
185
186 /* SLAB cache for fs_struct structures (tsk->fs) */
187 struct kmem_cache *fs_cachep;
188
189 /* SLAB cache for vm_area_struct structures */
190 struct kmem_cache *vm_area_cachep;
191
192 /* SLAB cache for mm_struct structures (tsk->mm) */
193 static struct kmem_cache *mm_cachep;
194
195 static void account_kernel_stack(struct thread_info *ti, int account)
196 {
197         struct zone *zone = page_zone(virt_to_page(ti));
198
199         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
200 }
201
202 void free_task(struct task_struct *tsk)
203 {
204         account_kernel_stack(tsk->stack, -1);
205         free_thread_info(tsk->stack);
206         rt_mutex_debug_task_free(tsk);
207         ftrace_graph_exit_task(tsk);
208         free_task_struct(tsk);
209 }
210 EXPORT_SYMBOL(free_task);
211
212 static inline void free_signal_struct(struct signal_struct *sig)
213 {
214         taskstats_tgid_free(sig);
215         sched_autogroup_exit(sig);
216         kmem_cache_free(signal_cachep, sig);
217 }
218
219 static inline void put_signal_struct(struct signal_struct *sig)
220 {
221         if (atomic_dec_and_test(&sig->sigcnt))
222                 free_signal_struct(sig);
223 }
224
225 void __put_task_struct(struct task_struct *tsk)
226 {
227         WARN_ON(!tsk->exit_state);
228         WARN_ON(atomic_read(&tsk->usage));
229         WARN_ON(tsk == current);
230
231         security_task_free(tsk);
232         exit_creds(tsk);
233         delayacct_tsk_free(tsk);
234         put_signal_struct(tsk->signal);
235
236         if (!profile_handoff_task(tsk))
237                 free_task(tsk);
238 }
239 EXPORT_SYMBOL_GPL(__put_task_struct);
240
241 void __init __weak arch_task_cache_init(void) { }
242
243 void __init fork_init(unsigned long mempages)
244 {
245 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
246 #ifndef ARCH_MIN_TASKALIGN
247 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
248 #endif
249         /* create a slab on which task_structs can be allocated */
250         task_struct_cachep =
251                 kmem_cache_create("task_struct", sizeof(struct task_struct),
252                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
253 #endif
254
255         /* do the arch specific task caches init */
256         arch_task_cache_init();
257
258         /*
259          * The default maximum number of threads is set to a safe
260          * value: the thread structures can take up at most half
261          * of memory.
262          */
263         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
264
265         /*
266          * we need to allow at least 20 threads to boot a system
267          */
268         if (max_threads < 20)
269                 max_threads = 20;
270
271         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
272         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
273         init_task.signal->rlim[RLIMIT_SIGPENDING] =
274                 init_task.signal->rlim[RLIMIT_NPROC];
275 }
276
277 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
278                                                struct task_struct *src)
279 {
280         *dst = *src;
281         return 0;
282 }
283
284 static struct task_struct *dup_task_struct(struct task_struct *orig)
285 {
286         struct task_struct *tsk;
287         struct thread_info *ti;
288         unsigned long *stackend;
289         int node = tsk_fork_get_node(orig);
290         int err;
291
292         prepare_to_copy(orig);
293
294         tsk = alloc_task_struct_node(node);
295         if (!tsk)
296                 return NULL;
297
298         ti = alloc_thread_info_node(tsk, node);
299         if (!ti) {
300                 free_task_struct(tsk);
301                 return NULL;
302         }
303
304         err = arch_dup_task_struct(tsk, orig);
305         if (err)
306                 goto out;
307
308         tsk->stack = ti;
309
310         setup_thread_stack(tsk, orig);
311         clear_user_return_notifier(tsk);
312         clear_tsk_need_resched(tsk);
313         stackend = end_of_stack(tsk);
314         *stackend = STACK_END_MAGIC;    /* for overflow detection */
315
316 #ifdef CONFIG_CC_STACKPROTECTOR
317         tsk->stack_canary = get_random_int();
318 #endif
319
320         /*
321          * One for us, one for whoever does the "release_task()" (usually
322          * parent)
323          */
324         atomic_set(&tsk->usage, 2);
325 #ifdef CONFIG_BLK_DEV_IO_TRACE
326         tsk->btrace_seq = 0;
327 #endif
328         tsk->splice_pipe = NULL;
329
330         account_kernel_stack(ti, 1);
331
332         return tsk;
333
334 out:
335         free_thread_info(ti);
336         free_task_struct(tsk);
337         return NULL;
338 }
339
340 #ifdef CONFIG_MMU
341 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
342 {
343         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
344         struct rb_node **rb_link, *rb_parent;
345         int retval;
346         unsigned long charge;
347         struct mempolicy *pol;
348
349         down_write(&oldmm->mmap_sem);
350         flush_cache_dup_mm(oldmm);
351         /*
352          * Not linked in yet - no deadlock potential:
353          */
354         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
355
356         mm->locked_vm = 0;
357         mm->mmap = NULL;
358         mm->mmap_cache = NULL;
359         mm->free_area_cache = oldmm->mmap_base;
360         mm->cached_hole_size = ~0UL;
361         mm->map_count = 0;
362         cpumask_clear(mm_cpumask(mm));
363         mm->mm_rb = RB_ROOT;
364         rb_link = &mm->mm_rb.rb_node;
365         rb_parent = NULL;
366         pprev = &mm->mmap;
367         retval = ksm_fork(mm, oldmm);
368         if (retval)
369                 goto out;
370         retval = khugepaged_fork(mm, oldmm);
371         if (retval)
372                 goto out;
373
374         prev = NULL;
375         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
376                 struct file *file;
377
378                 if (mpnt->vm_flags & VM_DONTCOPY) {
379                         long pages = vma_pages(mpnt);
380                         mm->total_vm -= pages;
381                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
382                                                                 -pages);
383                         continue;
384                 }
385                 charge = 0;
386                 if (mpnt->vm_flags & VM_ACCOUNT) {
387                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
388                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
389                                 goto fail_nomem;
390                         charge = len;
391                 }
392                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
393                 if (!tmp)
394                         goto fail_nomem;
395                 *tmp = *mpnt;
396                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
397                 pol = mpol_dup(vma_policy(mpnt));
398                 retval = PTR_ERR(pol);
399                 if (IS_ERR(pol))
400                         goto fail_nomem_policy;
401                 vma_set_policy(tmp, pol);
402                 tmp->vm_mm = mm;
403                 if (anon_vma_fork(tmp, mpnt))
404                         goto fail_nomem_anon_vma_fork;
405                 tmp->vm_flags &= ~VM_LOCKED;
406                 tmp->vm_next = tmp->vm_prev = NULL;
407                 file = tmp->vm_file;
408                 if (file) {
409                         struct inode *inode = file->f_path.dentry->d_inode;
410                         struct address_space *mapping = file->f_mapping;
411
412                         get_file(file);
413                         if (tmp->vm_flags & VM_DENYWRITE)
414                                 atomic_dec(&inode->i_writecount);
415                         mutex_lock(&mapping->i_mmap_mutex);
416                         if (tmp->vm_flags & VM_SHARED)
417                                 mapping->i_mmap_writable++;
418                         flush_dcache_mmap_lock(mapping);
419                         /* insert tmp into the share list, just after mpnt */
420                         vma_prio_tree_add(tmp, mpnt);
421                         flush_dcache_mmap_unlock(mapping);
422                         mutex_unlock(&mapping->i_mmap_mutex);
423                 }
424
425                 /*
426                  * Clear hugetlb-related page reserves for children. This only
427                  * affects MAP_PRIVATE mappings. Faults generated by the child
428                  * are not guaranteed to succeed, even if read-only
429                  */
430                 if (is_vm_hugetlb_page(tmp))
431                         reset_vma_resv_huge_pages(tmp);
432
433                 /*
434                  * Link in the new vma and copy the page table entries.
435                  */
436                 *pprev = tmp;
437                 pprev = &tmp->vm_next;
438                 tmp->vm_prev = prev;
439                 prev = tmp;
440
441                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
442                 rb_link = &tmp->vm_rb.rb_right;
443                 rb_parent = &tmp->vm_rb;
444
445                 mm->map_count++;
446                 retval = copy_page_range(mm, oldmm, mpnt);
447
448                 if (tmp->vm_ops && tmp->vm_ops->open)
449                         tmp->vm_ops->open(tmp);
450
451                 if (retval)
452                         goto out;
453         }
454         /* a new mm has just been created */
455         arch_dup_mmap(oldmm, mm);
456         retval = 0;
457 out:
458         up_write(&mm->mmap_sem);
459         flush_tlb_mm(oldmm);
460         up_write(&oldmm->mmap_sem);
461         return retval;
462 fail_nomem_anon_vma_fork:
463         mpol_put(pol);
464 fail_nomem_policy:
465         kmem_cache_free(vm_area_cachep, tmp);
466 fail_nomem:
467         retval = -ENOMEM;
468         vm_unacct_memory(charge);
469         goto out;
470 }
471
472 static inline int mm_alloc_pgd(struct mm_struct *mm)
473 {
474         mm->pgd = pgd_alloc(mm);
475         if (unlikely(!mm->pgd))
476                 return -ENOMEM;
477         return 0;
478 }
479
480 static inline void mm_free_pgd(struct mm_struct *mm)
481 {
482         pgd_free(mm, mm->pgd);
483 }
484 #else
485 #define dup_mmap(mm, oldmm)     (0)
486 #define mm_alloc_pgd(mm)        (0)
487 #define mm_free_pgd(mm)
488 #endif /* CONFIG_MMU */
489
490 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
491
492 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
493 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
494
495 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
496
497 static int __init coredump_filter_setup(char *s)
498 {
499         default_dump_filter =
500                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
501                 MMF_DUMP_FILTER_MASK;
502         return 1;
503 }
504
505 __setup("coredump_filter=", coredump_filter_setup);
506
507 #include <linux/init_task.h>
508
509 static void mm_init_aio(struct mm_struct *mm)
510 {
511 #ifdef CONFIG_AIO
512         spin_lock_init(&mm->ioctx_lock);
513         INIT_HLIST_HEAD(&mm->ioctx_list);
514 #endif
515 }
516
517 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
518 {
519         atomic_set(&mm->mm_users, 1);
520         atomic_set(&mm->mm_count, 1);
521         init_rwsem(&mm->mmap_sem);
522         INIT_LIST_HEAD(&mm->mmlist);
523         mm->flags = (current->mm) ?
524                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
525         mm->core_state = NULL;
526         mm->nr_ptes = 0;
527         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
528         spin_lock_init(&mm->page_table_lock);
529         mm->free_area_cache = TASK_UNMAPPED_BASE;
530         mm->cached_hole_size = ~0UL;
531         mm_init_aio(mm);
532         mm_init_owner(mm, p);
533
534         if (likely(!mm_alloc_pgd(mm))) {
535                 mm->def_flags = 0;
536                 mmu_notifier_mm_init(mm);
537                 return mm;
538         }
539
540         free_mm(mm);
541         return NULL;
542 }
543
544 static void check_mm(struct mm_struct *mm)
545 {
546         int i;
547
548         for (i = 0; i < NR_MM_COUNTERS; i++) {
549                 long x = atomic_long_read(&mm->rss_stat.count[i]);
550
551                 if (unlikely(x))
552                         printk(KERN_ALERT "BUG: Bad rss-counter state "
553                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
554         }
555
556 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
557         VM_BUG_ON(mm->pmd_huge_pte);
558 #endif
559 }
560
561 /*
562  * Allocate and initialize an mm_struct.
563  */
564 struct mm_struct *mm_alloc(void)
565 {
566         struct mm_struct *mm;
567
568         mm = allocate_mm();
569         if (!mm)
570                 return NULL;
571
572         memset(mm, 0, sizeof(*mm));
573         mm_init_cpumask(mm);
574         return mm_init(mm, current);
575 }
576
577 /*
578  * Called when the last reference to the mm
579  * is dropped: either by a lazy thread or by
580  * mmput. Free the page directory and the mm.
581  */
582 void __mmdrop(struct mm_struct *mm)
583 {
584         BUG_ON(mm == &init_mm);
585         mm_free_pgd(mm);
586         destroy_context(mm);
587         mmu_notifier_mm_destroy(mm);
588         check_mm(mm);
589         free_mm(mm);
590 }
591 EXPORT_SYMBOL_GPL(__mmdrop);
592
593 /*
594  * Decrement the use count and release all resources for an mm.
595  */
596 void mmput(struct mm_struct *mm)
597 {
598         might_sleep();
599
600         if (atomic_dec_and_test(&mm->mm_users)) {
601                 exit_aio(mm);
602                 ksm_exit(mm);
603                 khugepaged_exit(mm); /* must run before exit_mmap */
604                 exit_mmap(mm);
605                 set_mm_exe_file(mm, NULL);
606                 if (!list_empty(&mm->mmlist)) {
607                         spin_lock(&mmlist_lock);
608                         list_del(&mm->mmlist);
609                         spin_unlock(&mmlist_lock);
610                 }
611                 put_swap_token(mm);
612                 if (mm->binfmt)
613                         module_put(mm->binfmt->module);
614                 mmdrop(mm);
615         }
616 }
617 EXPORT_SYMBOL_GPL(mmput);
618
619 /*
620  * We added or removed a vma mapping the executable. The vmas are only mapped
621  * during exec and are not mapped with the mmap system call.
622  * Callers must hold down_write() on the mm's mmap_sem for these
623  */
624 void added_exe_file_vma(struct mm_struct *mm)
625 {
626         mm->num_exe_file_vmas++;
627 }
628
629 void removed_exe_file_vma(struct mm_struct *mm)
630 {
631         mm->num_exe_file_vmas--;
632         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
633                 fput(mm->exe_file);
634                 mm->exe_file = NULL;
635         }
636
637 }
638
639 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
640 {
641         if (new_exe_file)
642                 get_file(new_exe_file);
643         if (mm->exe_file)
644                 fput(mm->exe_file);
645         mm->exe_file = new_exe_file;
646         mm->num_exe_file_vmas = 0;
647 }
648
649 struct file *get_mm_exe_file(struct mm_struct *mm)
650 {
651         struct file *exe_file;
652
653         /* We need mmap_sem to protect against races with removal of
654          * VM_EXECUTABLE vmas */
655         down_read(&mm->mmap_sem);
656         exe_file = mm->exe_file;
657         if (exe_file)
658                 get_file(exe_file);
659         up_read(&mm->mmap_sem);
660         return exe_file;
661 }
662
663 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
664 {
665         /* It's safe to write the exe_file pointer without exe_file_lock because
666          * this is called during fork when the task is not yet in /proc */
667         newmm->exe_file = get_mm_exe_file(oldmm);
668 }
669
670 /**
671  * get_task_mm - acquire a reference to the task's mm
672  *
673  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
674  * this kernel workthread has transiently adopted a user mm with use_mm,
675  * to do its AIO) is not set and if so returns a reference to it, after
676  * bumping up the use count.  User must release the mm via mmput()
677  * after use.  Typically used by /proc and ptrace.
678  */
679 struct mm_struct *get_task_mm(struct task_struct *task)
680 {
681         struct mm_struct *mm;
682
683         task_lock(task);
684         mm = task->mm;
685         if (mm) {
686                 if (task->flags & PF_KTHREAD)
687                         mm = NULL;
688                 else
689                         atomic_inc(&mm->mm_users);
690         }
691         task_unlock(task);
692         return mm;
693 }
694 EXPORT_SYMBOL_GPL(get_task_mm);
695
696 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
697 {
698         struct mm_struct *mm;
699         int err;
700
701         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
702         if (err)
703                 return ERR_PTR(err);
704
705         mm = get_task_mm(task);
706         if (mm && mm != current->mm &&
707                         !ptrace_may_access(task, mode)) {
708                 mmput(mm);
709                 mm = ERR_PTR(-EACCES);
710         }
711         mutex_unlock(&task->signal->cred_guard_mutex);
712
713         return mm;
714 }
715
716 static void complete_vfork_done(struct task_struct *tsk)
717 {
718         struct completion *vfork;
719
720         task_lock(tsk);
721         vfork = tsk->vfork_done;
722         if (likely(vfork)) {
723                 tsk->vfork_done = NULL;
724                 complete(vfork);
725         }
726         task_unlock(tsk);
727 }
728
729 static int wait_for_vfork_done(struct task_struct *child,
730                                 struct completion *vfork)
731 {
732         int killed;
733
734         freezer_do_not_count();
735         killed = wait_for_completion_killable(vfork);
736         freezer_count();
737
738         if (killed) {
739                 task_lock(child);
740                 child->vfork_done = NULL;
741                 task_unlock(child);
742         }
743
744         put_task_struct(child);
745         return killed;
746 }
747
748 /* Please note the differences between mmput and mm_release.
749  * mmput is called whenever we stop holding onto a mm_struct,
750  * error success whatever.
751  *
752  * mm_release is called after a mm_struct has been removed
753  * from the current process.
754  *
755  * This difference is important for error handling, when we
756  * only half set up a mm_struct for a new process and need to restore
757  * the old one.  Because we mmput the new mm_struct before
758  * restoring the old one. . .
759  * Eric Biederman 10 January 1998
760  */
761 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
762 {
763         /* Get rid of any futexes when releasing the mm */
764 #ifdef CONFIG_FUTEX
765         if (unlikely(tsk->robust_list)) {
766                 exit_robust_list(tsk);
767                 tsk->robust_list = NULL;
768         }
769 #ifdef CONFIG_COMPAT
770         if (unlikely(tsk->compat_robust_list)) {
771                 compat_exit_robust_list(tsk);
772                 tsk->compat_robust_list = NULL;
773         }
774 #endif
775         if (unlikely(!list_empty(&tsk->pi_state_list)))
776                 exit_pi_state_list(tsk);
777 #endif
778
779         /* Get rid of any cached register state */
780         deactivate_mm(tsk, mm);
781
782         if (tsk->vfork_done)
783                 complete_vfork_done(tsk);
784
785         /*
786          * If we're exiting normally, clear a user-space tid field if
787          * requested.  We leave this alone when dying by signal, to leave
788          * the value intact in a core dump, and to save the unnecessary
789          * trouble, say, a killed vfork parent shouldn't touch this mm.
790          * Userland only wants this done for a sys_exit.
791          */
792         if (tsk->clear_child_tid) {
793                 if (!(tsk->flags & PF_SIGNALED) &&
794                     atomic_read(&mm->mm_users) > 1) {
795                         /*
796                          * We don't check the error code - if userspace has
797                          * not set up a proper pointer then tough luck.
798                          */
799                         put_user(0, tsk->clear_child_tid);
800                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
801                                         1, NULL, NULL, 0);
802                 }
803                 tsk->clear_child_tid = NULL;
804         }
805 }
806
807 /*
808  * Allocate a new mm structure and copy contents from the
809  * mm structure of the passed in task structure.
810  */
811 struct mm_struct *dup_mm(struct task_struct *tsk)
812 {
813         struct mm_struct *mm, *oldmm = current->mm;
814         int err;
815
816         if (!oldmm)
817                 return NULL;
818
819         mm = allocate_mm();
820         if (!mm)
821                 goto fail_nomem;
822
823         memcpy(mm, oldmm, sizeof(*mm));
824         mm_init_cpumask(mm);
825
826         /* Initializing for Swap token stuff */
827         mm->token_priority = 0;
828         mm->last_interval = 0;
829
830 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
831         mm->pmd_huge_pte = NULL;
832 #endif
833
834         if (!mm_init(mm, tsk))
835                 goto fail_nomem;
836
837         if (init_new_context(tsk, mm))
838                 goto fail_nocontext;
839
840         dup_mm_exe_file(oldmm, mm);
841
842         err = dup_mmap(mm, oldmm);
843         if (err)
844                 goto free_pt;
845
846         mm->hiwater_rss = get_mm_rss(mm);
847         mm->hiwater_vm = mm->total_vm;
848
849         if (mm->binfmt && !try_module_get(mm->binfmt->module))
850                 goto free_pt;
851
852         return mm;
853
854 free_pt:
855         /* don't put binfmt in mmput, we haven't got module yet */
856         mm->binfmt = NULL;
857         mmput(mm);
858
859 fail_nomem:
860         return NULL;
861
862 fail_nocontext:
863         /*
864          * If init_new_context() failed, we cannot use mmput() to free the mm
865          * because it calls destroy_context()
866          */
867         mm_free_pgd(mm);
868         free_mm(mm);
869         return NULL;
870 }
871
872 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
873 {
874         struct mm_struct *mm, *oldmm;
875         int retval;
876
877         tsk->min_flt = tsk->maj_flt = 0;
878         tsk->nvcsw = tsk->nivcsw = 0;
879 #ifdef CONFIG_DETECT_HUNG_TASK
880         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
881 #endif
882
883         tsk->mm = NULL;
884         tsk->active_mm = NULL;
885
886         /*
887          * Are we cloning a kernel thread?
888          *
889          * We need to steal a active VM for that..
890          */
891         oldmm = current->mm;
892         if (!oldmm)
893                 return 0;
894
895         if (clone_flags & CLONE_VM) {
896                 atomic_inc(&oldmm->mm_users);
897                 mm = oldmm;
898                 goto good_mm;
899         }
900
901         retval = -ENOMEM;
902         mm = dup_mm(tsk);
903         if (!mm)
904                 goto fail_nomem;
905
906 good_mm:
907         /* Initializing for Swap token stuff */
908         mm->token_priority = 0;
909         mm->last_interval = 0;
910
911         tsk->mm = mm;
912         tsk->active_mm = mm;
913         return 0;
914
915 fail_nomem:
916         return retval;
917 }
918
919 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
920 {
921         struct fs_struct *fs = current->fs;
922         if (clone_flags & CLONE_FS) {
923                 /* tsk->fs is already what we want */
924                 spin_lock(&fs->lock);
925                 if (fs->in_exec) {
926                         spin_unlock(&fs->lock);
927                         return -EAGAIN;
928                 }
929                 fs->users++;
930                 spin_unlock(&fs->lock);
931                 return 0;
932         }
933         tsk->fs = copy_fs_struct(fs);
934         if (!tsk->fs)
935                 return -ENOMEM;
936         return 0;
937 }
938
939 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
940 {
941         struct files_struct *oldf, *newf;
942         int error = 0;
943
944         /*
945          * A background process may not have any files ...
946          */
947         oldf = current->files;
948         if (!oldf)
949                 goto out;
950
951         if (clone_flags & CLONE_FILES) {
952                 atomic_inc(&oldf->count);
953                 goto out;
954         }
955
956         newf = dup_fd(oldf, &error);
957         if (!newf)
958                 goto out;
959
960         tsk->files = newf;
961         error = 0;
962 out:
963         return error;
964 }
965
966 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
967 {
968 #ifdef CONFIG_BLOCK
969         struct io_context *ioc = current->io_context;
970         struct io_context *new_ioc;
971
972         if (!ioc)
973                 return 0;
974         /*
975          * Share io context with parent, if CLONE_IO is set
976          */
977         if (clone_flags & CLONE_IO) {
978                 tsk->io_context = ioc_task_link(ioc);
979                 if (unlikely(!tsk->io_context))
980                         return -ENOMEM;
981         } else if (ioprio_valid(ioc->ioprio)) {
982                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
983                 if (unlikely(!new_ioc))
984                         return -ENOMEM;
985
986                 new_ioc->ioprio = ioc->ioprio;
987                 put_io_context(new_ioc);
988         }
989 #endif
990         return 0;
991 }
992
993 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
994 {
995         struct sighand_struct *sig;
996
997         if (clone_flags & CLONE_SIGHAND) {
998                 atomic_inc(&current->sighand->count);
999                 return 0;
1000         }
1001         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1002         rcu_assign_pointer(tsk->sighand, sig);
1003         if (!sig)
1004                 return -ENOMEM;
1005         atomic_set(&sig->count, 1);
1006         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1007         return 0;
1008 }
1009
1010 void __cleanup_sighand(struct sighand_struct *sighand)
1011 {
1012         if (atomic_dec_and_test(&sighand->count)) {
1013                 signalfd_cleanup(sighand);
1014                 kmem_cache_free(sighand_cachep, sighand);
1015         }
1016 }
1017
1018
1019 /*
1020  * Initialize POSIX timer handling for a thread group.
1021  */
1022 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1023 {
1024         unsigned long cpu_limit;
1025
1026         /* Thread group counters. */
1027         thread_group_cputime_init(sig);
1028
1029         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1030         if (cpu_limit != RLIM_INFINITY) {
1031                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1032                 sig->cputimer.running = 1;
1033         }
1034
1035         /* The timer lists. */
1036         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1037         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1038         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1039 }
1040
1041 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1042 {
1043         struct signal_struct *sig;
1044
1045         if (clone_flags & CLONE_THREAD)
1046                 return 0;
1047
1048         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1049         tsk->signal = sig;
1050         if (!sig)
1051                 return -ENOMEM;
1052
1053         sig->nr_threads = 1;
1054         atomic_set(&sig->live, 1);
1055         atomic_set(&sig->sigcnt, 1);
1056         init_waitqueue_head(&sig->wait_chldexit);
1057         if (clone_flags & CLONE_NEWPID)
1058                 sig->flags |= SIGNAL_UNKILLABLE;
1059         sig->curr_target = tsk;
1060         init_sigpending(&sig->shared_pending);
1061         INIT_LIST_HEAD(&sig->posix_timers);
1062
1063         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1064         sig->real_timer.function = it_real_fn;
1065
1066         task_lock(current->group_leader);
1067         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1068         task_unlock(current->group_leader);
1069
1070         posix_cpu_timers_init_group(sig);
1071
1072         tty_audit_fork(sig);
1073         sched_autogroup_fork(sig);
1074
1075 #ifdef CONFIG_CGROUPS
1076         init_rwsem(&sig->group_rwsem);
1077 #endif
1078
1079         sig->oom_adj = current->signal->oom_adj;
1080         sig->oom_score_adj = current->signal->oom_score_adj;
1081         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1082
1083         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1084                                    current->signal->is_child_subreaper;
1085
1086         mutex_init(&sig->cred_guard_mutex);
1087
1088         return 0;
1089 }
1090
1091 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1092 {
1093         unsigned long new_flags = p->flags;
1094
1095         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1096         new_flags |= PF_FORKNOEXEC;
1097         p->flags = new_flags;
1098 }
1099
1100 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1101 {
1102         current->clear_child_tid = tidptr;
1103
1104         return task_pid_vnr(current);
1105 }
1106
1107 static void rt_mutex_init_task(struct task_struct *p)
1108 {
1109         raw_spin_lock_init(&p->pi_lock);
1110 #ifdef CONFIG_RT_MUTEXES
1111         plist_head_init(&p->pi_waiters);
1112         p->pi_blocked_on = NULL;
1113 #endif
1114 }
1115
1116 #ifdef CONFIG_MM_OWNER
1117 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1118 {
1119         mm->owner = p;
1120 }
1121 #endif /* CONFIG_MM_OWNER */
1122
1123 /*
1124  * Initialize POSIX timer handling for a single task.
1125  */
1126 static void posix_cpu_timers_init(struct task_struct *tsk)
1127 {
1128         tsk->cputime_expires.prof_exp = 0;
1129         tsk->cputime_expires.virt_exp = 0;
1130         tsk->cputime_expires.sched_exp = 0;
1131         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1132         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1133         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1134 }
1135
1136 /*
1137  * This creates a new process as a copy of the old one,
1138  * but does not actually start it yet.
1139  *
1140  * It copies the registers, and all the appropriate
1141  * parts of the process environment (as per the clone
1142  * flags). The actual kick-off is left to the caller.
1143  */
1144 static struct task_struct *copy_process(unsigned long clone_flags,
1145                                         unsigned long stack_start,
1146                                         struct pt_regs *regs,
1147                                         unsigned long stack_size,
1148                                         int __user *child_tidptr,
1149                                         struct pid *pid,
1150                                         int trace)
1151 {
1152         int retval;
1153         struct task_struct *p;
1154         int cgroup_callbacks_done = 0;
1155
1156         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1157                 return ERR_PTR(-EINVAL);
1158
1159         /*
1160          * Thread groups must share signals as well, and detached threads
1161          * can only be started up within the thread group.
1162          */
1163         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1164                 return ERR_PTR(-EINVAL);
1165
1166         /*
1167          * Shared signal handlers imply shared VM. By way of the above,
1168          * thread groups also imply shared VM. Blocking this case allows
1169          * for various simplifications in other code.
1170          */
1171         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1172                 return ERR_PTR(-EINVAL);
1173
1174         /*
1175          * Siblings of global init remain as zombies on exit since they are
1176          * not reaped by their parent (swapper). To solve this and to avoid
1177          * multi-rooted process trees, prevent global and container-inits
1178          * from creating siblings.
1179          */
1180         if ((clone_flags & CLONE_PARENT) &&
1181                                 current->signal->flags & SIGNAL_UNKILLABLE)
1182                 return ERR_PTR(-EINVAL);
1183
1184         retval = security_task_create(clone_flags);
1185         if (retval)
1186                 goto fork_out;
1187
1188         retval = -ENOMEM;
1189         p = dup_task_struct(current);
1190         if (!p)
1191                 goto fork_out;
1192
1193         ftrace_graph_init_task(p);
1194
1195         rt_mutex_init_task(p);
1196
1197 #ifdef CONFIG_PROVE_LOCKING
1198         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1199         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1200 #endif
1201         retval = -EAGAIN;
1202         if (atomic_read(&p->real_cred->user->processes) >=
1203                         task_rlimit(p, RLIMIT_NPROC)) {
1204                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1205                     p->real_cred->user != INIT_USER)
1206                         goto bad_fork_free;
1207         }
1208         current->flags &= ~PF_NPROC_EXCEEDED;
1209
1210         retval = copy_creds(p, clone_flags);
1211         if (retval < 0)
1212                 goto bad_fork_free;
1213
1214         /*
1215          * If multiple threads are within copy_process(), then this check
1216          * triggers too late. This doesn't hurt, the check is only there
1217          * to stop root fork bombs.
1218          */
1219         retval = -EAGAIN;
1220         if (nr_threads >= max_threads)
1221                 goto bad_fork_cleanup_count;
1222
1223         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1224                 goto bad_fork_cleanup_count;
1225
1226         p->did_exec = 0;
1227         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1228         copy_flags(clone_flags, p);
1229         INIT_LIST_HEAD(&p->children);
1230         INIT_LIST_HEAD(&p->sibling);
1231         rcu_copy_process(p);
1232         p->vfork_done = NULL;
1233         spin_lock_init(&p->alloc_lock);
1234
1235         init_sigpending(&p->pending);
1236
1237         p->utime = p->stime = p->gtime = 0;
1238         p->utimescaled = p->stimescaled = 0;
1239 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1240         p->prev_utime = p->prev_stime = 0;
1241 #endif
1242 #if defined(SPLIT_RSS_COUNTING)
1243         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1244 #endif
1245
1246         p->default_timer_slack_ns = current->timer_slack_ns;
1247
1248         task_io_accounting_init(&p->ioac);
1249         acct_clear_integrals(p);
1250
1251         posix_cpu_timers_init(p);
1252
1253         do_posix_clock_monotonic_gettime(&p->start_time);
1254         p->real_start_time = p->start_time;
1255         monotonic_to_bootbased(&p->real_start_time);
1256         p->io_context = NULL;
1257         p->audit_context = NULL;
1258         if (clone_flags & CLONE_THREAD)
1259                 threadgroup_change_begin(current);
1260         cgroup_fork(p);
1261 #ifdef CONFIG_NUMA
1262         p->mempolicy = mpol_dup(p->mempolicy);
1263         if (IS_ERR(p->mempolicy)) {
1264                 retval = PTR_ERR(p->mempolicy);
1265                 p->mempolicy = NULL;
1266                 goto bad_fork_cleanup_cgroup;
1267         }
1268         mpol_fix_fork_child_flag(p);
1269 #endif
1270 #ifdef CONFIG_CPUSETS
1271         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1272         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1273         seqcount_init(&p->mems_allowed_seq);
1274 #endif
1275 #ifdef CONFIG_TRACE_IRQFLAGS
1276         p->irq_events = 0;
1277 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1278         p->hardirqs_enabled = 1;
1279 #else
1280         p->hardirqs_enabled = 0;
1281 #endif
1282         p->hardirq_enable_ip = 0;
1283         p->hardirq_enable_event = 0;
1284         p->hardirq_disable_ip = _THIS_IP_;
1285         p->hardirq_disable_event = 0;
1286         p->softirqs_enabled = 1;
1287         p->softirq_enable_ip = _THIS_IP_;
1288         p->softirq_enable_event = 0;
1289         p->softirq_disable_ip = 0;
1290         p->softirq_disable_event = 0;
1291         p->hardirq_context = 0;
1292         p->softirq_context = 0;
1293 #endif
1294 #ifdef CONFIG_LOCKDEP
1295         p->lockdep_depth = 0; /* no locks held yet */
1296         p->curr_chain_key = 0;
1297         p->lockdep_recursion = 0;
1298 #endif
1299
1300 #ifdef CONFIG_DEBUG_MUTEXES
1301         p->blocked_on = NULL; /* not blocked yet */
1302 #endif
1303 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1304         p->memcg_batch.do_batch = 0;
1305         p->memcg_batch.memcg = NULL;
1306 #endif
1307
1308         /* Perform scheduler related setup. Assign this task to a CPU. */
1309         sched_fork(p);
1310
1311         retval = perf_event_init_task(p);
1312         if (retval)
1313                 goto bad_fork_cleanup_policy;
1314         retval = audit_alloc(p);
1315         if (retval)
1316                 goto bad_fork_cleanup_policy;
1317         /* copy all the process information */
1318         retval = copy_semundo(clone_flags, p);
1319         if (retval)
1320                 goto bad_fork_cleanup_audit;
1321         retval = copy_files(clone_flags, p);
1322         if (retval)
1323                 goto bad_fork_cleanup_semundo;
1324         retval = copy_fs(clone_flags, p);
1325         if (retval)
1326                 goto bad_fork_cleanup_files;
1327         retval = copy_sighand(clone_flags, p);
1328         if (retval)
1329                 goto bad_fork_cleanup_fs;
1330         retval = copy_signal(clone_flags, p);
1331         if (retval)
1332                 goto bad_fork_cleanup_sighand;
1333         retval = copy_mm(clone_flags, p);
1334         if (retval)
1335                 goto bad_fork_cleanup_signal;
1336         retval = copy_namespaces(clone_flags, p);
1337         if (retval)
1338                 goto bad_fork_cleanup_mm;
1339         retval = copy_io(clone_flags, p);
1340         if (retval)
1341                 goto bad_fork_cleanup_namespaces;
1342         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1343         if (retval)
1344                 goto bad_fork_cleanup_io;
1345
1346         if (pid != &init_struct_pid) {
1347                 retval = -ENOMEM;
1348                 pid = alloc_pid(p->nsproxy->pid_ns);
1349                 if (!pid)
1350                         goto bad_fork_cleanup_io;
1351         }
1352
1353         p->pid = pid_nr(pid);
1354         p->tgid = p->pid;
1355         if (clone_flags & CLONE_THREAD)
1356                 p->tgid = current->tgid;
1357
1358         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1359         /*
1360          * Clear TID on mm_release()?
1361          */
1362         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1363 #ifdef CONFIG_BLOCK
1364         p->plug = NULL;
1365 #endif
1366 #ifdef CONFIG_FUTEX
1367         p->robust_list = NULL;
1368 #ifdef CONFIG_COMPAT
1369         p->compat_robust_list = NULL;
1370 #endif
1371         INIT_LIST_HEAD(&p->pi_state_list);
1372         p->pi_state_cache = NULL;
1373 #endif
1374         /*
1375          * sigaltstack should be cleared when sharing the same VM
1376          */
1377         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1378                 p->sas_ss_sp = p->sas_ss_size = 0;
1379
1380         /*
1381          * Syscall tracing and stepping should be turned off in the
1382          * child regardless of CLONE_PTRACE.
1383          */
1384         user_disable_single_step(p);
1385         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1386 #ifdef TIF_SYSCALL_EMU
1387         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1388 #endif
1389         clear_all_latency_tracing(p);
1390
1391         /* ok, now we should be set up.. */
1392         if (clone_flags & CLONE_THREAD)
1393                 p->exit_signal = -1;
1394         else if (clone_flags & CLONE_PARENT)
1395                 p->exit_signal = current->group_leader->exit_signal;
1396         else
1397                 p->exit_signal = (clone_flags & CSIGNAL);
1398
1399         p->pdeath_signal = 0;
1400         p->exit_state = 0;
1401
1402         p->nr_dirtied = 0;
1403         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1404         p->dirty_paused_when = 0;
1405
1406         /*
1407          * Ok, make it visible to the rest of the system.
1408          * We dont wake it up yet.
1409          */
1410         p->group_leader = p;
1411         INIT_LIST_HEAD(&p->thread_group);
1412
1413         /* Now that the task is set up, run cgroup callbacks if
1414          * necessary. We need to run them before the task is visible
1415          * on the tasklist. */
1416         cgroup_fork_callbacks(p);
1417         cgroup_callbacks_done = 1;
1418
1419         /* Need tasklist lock for parent etc handling! */
1420         write_lock_irq(&tasklist_lock);
1421
1422         /* CLONE_PARENT re-uses the old parent */
1423         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1424                 p->real_parent = current->real_parent;
1425                 p->parent_exec_id = current->parent_exec_id;
1426         } else {
1427                 p->real_parent = current;
1428                 p->parent_exec_id = current->self_exec_id;
1429         }
1430
1431         spin_lock(&current->sighand->siglock);
1432
1433         /*
1434          * Process group and session signals need to be delivered to just the
1435          * parent before the fork or both the parent and the child after the
1436          * fork. Restart if a signal comes in before we add the new process to
1437          * it's process group.
1438          * A fatal signal pending means that current will exit, so the new
1439          * thread can't slip out of an OOM kill (or normal SIGKILL).
1440         */
1441         recalc_sigpending();
1442         if (signal_pending(current)) {
1443                 spin_unlock(&current->sighand->siglock);
1444                 write_unlock_irq(&tasklist_lock);
1445                 retval = -ERESTARTNOINTR;
1446                 goto bad_fork_free_pid;
1447         }
1448
1449         if (clone_flags & CLONE_THREAD) {
1450                 current->signal->nr_threads++;
1451                 atomic_inc(&current->signal->live);
1452                 atomic_inc(&current->signal->sigcnt);
1453                 p->group_leader = current->group_leader;
1454                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1455         }
1456
1457         if (likely(p->pid)) {
1458                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1459
1460                 if (thread_group_leader(p)) {
1461                         if (is_child_reaper(pid))
1462                                 p->nsproxy->pid_ns->child_reaper = p;
1463
1464                         p->signal->leader_pid = pid;
1465                         p->signal->tty = tty_kref_get(current->signal->tty);
1466                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1467                         attach_pid(p, PIDTYPE_SID, task_session(current));
1468                         list_add_tail(&p->sibling, &p->real_parent->children);
1469                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1470                         __this_cpu_inc(process_counts);
1471                 }
1472                 attach_pid(p, PIDTYPE_PID, pid);
1473                 nr_threads++;
1474         }
1475
1476         total_forks++;
1477         spin_unlock(&current->sighand->siglock);
1478         write_unlock_irq(&tasklist_lock);
1479         proc_fork_connector(p);
1480         cgroup_post_fork(p);
1481         if (clone_flags & CLONE_THREAD)
1482                 threadgroup_change_end(current);
1483         perf_event_fork(p);
1484
1485         trace_task_newtask(p, clone_flags);
1486
1487         return p;
1488
1489 bad_fork_free_pid:
1490         if (pid != &init_struct_pid)
1491                 free_pid(pid);
1492 bad_fork_cleanup_io:
1493         if (p->io_context)
1494                 exit_io_context(p);
1495 bad_fork_cleanup_namespaces:
1496         exit_task_namespaces(p);
1497 bad_fork_cleanup_mm:
1498         if (p->mm)
1499                 mmput(p->mm);
1500 bad_fork_cleanup_signal:
1501         if (!(clone_flags & CLONE_THREAD))
1502                 free_signal_struct(p->signal);
1503 bad_fork_cleanup_sighand:
1504         __cleanup_sighand(p->sighand);
1505 bad_fork_cleanup_fs:
1506         exit_fs(p); /* blocking */
1507 bad_fork_cleanup_files:
1508         exit_files(p); /* blocking */
1509 bad_fork_cleanup_semundo:
1510         exit_sem(p);
1511 bad_fork_cleanup_audit:
1512         audit_free(p);
1513 bad_fork_cleanup_policy:
1514         perf_event_free_task(p);
1515 #ifdef CONFIG_NUMA
1516         mpol_put(p->mempolicy);
1517 bad_fork_cleanup_cgroup:
1518 #endif
1519         if (clone_flags & CLONE_THREAD)
1520                 threadgroup_change_end(current);
1521         cgroup_exit(p, cgroup_callbacks_done);
1522         delayacct_tsk_free(p);
1523         module_put(task_thread_info(p)->exec_domain->module);
1524 bad_fork_cleanup_count:
1525         atomic_dec(&p->cred->user->processes);
1526         exit_creds(p);
1527 bad_fork_free:
1528         free_task(p);
1529 fork_out:
1530         return ERR_PTR(retval);
1531 }
1532
1533 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1534 {
1535         memset(regs, 0, sizeof(struct pt_regs));
1536         return regs;
1537 }
1538
1539 static inline void init_idle_pids(struct pid_link *links)
1540 {
1541         enum pid_type type;
1542
1543         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1544                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1545                 links[type].pid = &init_struct_pid;
1546         }
1547 }
1548
1549 struct task_struct * __cpuinit fork_idle(int cpu)
1550 {
1551         struct task_struct *task;
1552         struct pt_regs regs;
1553
1554         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1555                             &init_struct_pid, 0);
1556         if (!IS_ERR(task)) {
1557                 init_idle_pids(task->pids);
1558                 init_idle(task, cpu);
1559         }
1560
1561         return task;
1562 }
1563
1564 /*
1565  *  Ok, this is the main fork-routine.
1566  *
1567  * It copies the process, and if successful kick-starts
1568  * it and waits for it to finish using the VM if required.
1569  */
1570 long do_fork(unsigned long clone_flags,
1571               unsigned long stack_start,
1572               struct pt_regs *regs,
1573               unsigned long stack_size,
1574               int __user *parent_tidptr,
1575               int __user *child_tidptr)
1576 {
1577         struct task_struct *p;
1578         int trace = 0;
1579         long nr;
1580
1581         /*
1582          * Do some preliminary argument and permissions checking before we
1583          * actually start allocating stuff
1584          */
1585         if (clone_flags & CLONE_NEWUSER) {
1586                 if (clone_flags & CLONE_THREAD)
1587                         return -EINVAL;
1588                 /* hopefully this check will go away when userns support is
1589                  * complete
1590                  */
1591                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1592                                 !capable(CAP_SETGID))
1593                         return -EPERM;
1594         }
1595
1596         /*
1597          * Determine whether and which event to report to ptracer.  When
1598          * called from kernel_thread or CLONE_UNTRACED is explicitly
1599          * requested, no event is reported; otherwise, report if the event
1600          * for the type of forking is enabled.
1601          */
1602         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1603                 if (clone_flags & CLONE_VFORK)
1604                         trace = PTRACE_EVENT_VFORK;
1605                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1606                         trace = PTRACE_EVENT_CLONE;
1607                 else
1608                         trace = PTRACE_EVENT_FORK;
1609
1610                 if (likely(!ptrace_event_enabled(current, trace)))
1611                         trace = 0;
1612         }
1613
1614         p = copy_process(clone_flags, stack_start, regs, stack_size,
1615                          child_tidptr, NULL, trace);
1616         /*
1617          * Do this prior waking up the new thread - the thread pointer
1618          * might get invalid after that point, if the thread exits quickly.
1619          */
1620         if (!IS_ERR(p)) {
1621                 struct completion vfork;
1622
1623                 trace_sched_process_fork(current, p);
1624
1625                 nr = task_pid_vnr(p);
1626
1627                 if (clone_flags & CLONE_PARENT_SETTID)
1628                         put_user(nr, parent_tidptr);
1629
1630                 if (clone_flags & CLONE_VFORK) {
1631                         p->vfork_done = &vfork;
1632                         init_completion(&vfork);
1633                         get_task_struct(p);
1634                 }
1635
1636                 wake_up_new_task(p);
1637
1638                 /* forking complete and child started to run, tell ptracer */
1639                 if (unlikely(trace))
1640                         ptrace_event(trace, nr);
1641
1642                 if (clone_flags & CLONE_VFORK) {
1643                         if (!wait_for_vfork_done(p, &vfork))
1644                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1645                 }
1646         } else {
1647                 nr = PTR_ERR(p);
1648         }
1649         return nr;
1650 }
1651
1652 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1653 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1654 #endif
1655
1656 static void sighand_ctor(void *data)
1657 {
1658         struct sighand_struct *sighand = data;
1659
1660         spin_lock_init(&sighand->siglock);
1661         init_waitqueue_head(&sighand->signalfd_wqh);
1662 }
1663
1664 void __init proc_caches_init(void)
1665 {
1666         sighand_cachep = kmem_cache_create("sighand_cache",
1667                         sizeof(struct sighand_struct), 0,
1668                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1669                         SLAB_NOTRACK, sighand_ctor);
1670         signal_cachep = kmem_cache_create("signal_cache",
1671                         sizeof(struct signal_struct), 0,
1672                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1673         files_cachep = kmem_cache_create("files_cache",
1674                         sizeof(struct files_struct), 0,
1675                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1676         fs_cachep = kmem_cache_create("fs_cache",
1677                         sizeof(struct fs_struct), 0,
1678                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1679         /*
1680          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1681          * whole struct cpumask for the OFFSTACK case. We could change
1682          * this to *only* allocate as much of it as required by the
1683          * maximum number of CPU's we can ever have.  The cpumask_allocation
1684          * is at the end of the structure, exactly for that reason.
1685          */
1686         mm_cachep = kmem_cache_create("mm_struct",
1687                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1688                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1689         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1690         mmap_init();
1691         nsproxy_cache_init();
1692 }
1693
1694 /*
1695  * Check constraints on flags passed to the unshare system call.
1696  */
1697 static int check_unshare_flags(unsigned long unshare_flags)
1698 {
1699         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1700                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1701                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1702                 return -EINVAL;
1703         /*
1704          * Not implemented, but pretend it works if there is nothing to
1705          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1706          * needs to unshare vm.
1707          */
1708         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1709                 /* FIXME: get_task_mm() increments ->mm_users */
1710                 if (atomic_read(&current->mm->mm_users) > 1)
1711                         return -EINVAL;
1712         }
1713
1714         return 0;
1715 }
1716
1717 /*
1718  * Unshare the filesystem structure if it is being shared
1719  */
1720 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1721 {
1722         struct fs_struct *fs = current->fs;
1723
1724         if (!(unshare_flags & CLONE_FS) || !fs)
1725                 return 0;
1726
1727         /* don't need lock here; in the worst case we'll do useless copy */
1728         if (fs->users == 1)
1729                 return 0;
1730
1731         *new_fsp = copy_fs_struct(fs);
1732         if (!*new_fsp)
1733                 return -ENOMEM;
1734
1735         return 0;
1736 }
1737
1738 /*
1739  * Unshare file descriptor table if it is being shared
1740  */
1741 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1742 {
1743         struct files_struct *fd = current->files;
1744         int error = 0;
1745
1746         if ((unshare_flags & CLONE_FILES) &&
1747             (fd && atomic_read(&fd->count) > 1)) {
1748                 *new_fdp = dup_fd(fd, &error);
1749                 if (!*new_fdp)
1750                         return error;
1751         }
1752
1753         return 0;
1754 }
1755
1756 /*
1757  * unshare allows a process to 'unshare' part of the process
1758  * context which was originally shared using clone.  copy_*
1759  * functions used by do_fork() cannot be used here directly
1760  * because they modify an inactive task_struct that is being
1761  * constructed. Here we are modifying the current, active,
1762  * task_struct.
1763  */
1764 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1765 {
1766         struct fs_struct *fs, *new_fs = NULL;
1767         struct files_struct *fd, *new_fd = NULL;
1768         struct nsproxy *new_nsproxy = NULL;
1769         int do_sysvsem = 0;
1770         int err;
1771
1772         err = check_unshare_flags(unshare_flags);
1773         if (err)
1774                 goto bad_unshare_out;
1775
1776         /*
1777          * If unsharing namespace, must also unshare filesystem information.
1778          */
1779         if (unshare_flags & CLONE_NEWNS)
1780                 unshare_flags |= CLONE_FS;
1781         /*
1782          * CLONE_NEWIPC must also detach from the undolist: after switching
1783          * to a new ipc namespace, the semaphore arrays from the old
1784          * namespace are unreachable.
1785          */
1786         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1787                 do_sysvsem = 1;
1788         err = unshare_fs(unshare_flags, &new_fs);
1789         if (err)
1790                 goto bad_unshare_out;
1791         err = unshare_fd(unshare_flags, &new_fd);
1792         if (err)
1793                 goto bad_unshare_cleanup_fs;
1794         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1795         if (err)
1796                 goto bad_unshare_cleanup_fd;
1797
1798         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1799                 if (do_sysvsem) {
1800                         /*
1801                          * CLONE_SYSVSEM is equivalent to sys_exit().
1802                          */
1803                         exit_sem(current);
1804                 }
1805
1806                 if (new_nsproxy) {
1807                         switch_task_namespaces(current, new_nsproxy);
1808                         new_nsproxy = NULL;
1809                 }
1810
1811                 task_lock(current);
1812
1813                 if (new_fs) {
1814                         fs = current->fs;
1815                         spin_lock(&fs->lock);
1816                         current->fs = new_fs;
1817                         if (--fs->users)
1818                                 new_fs = NULL;
1819                         else
1820                                 new_fs = fs;
1821                         spin_unlock(&fs->lock);
1822                 }
1823
1824                 if (new_fd) {
1825                         fd = current->files;
1826                         current->files = new_fd;
1827                         new_fd = fd;
1828                 }
1829
1830                 task_unlock(current);
1831         }
1832
1833         if (new_nsproxy)
1834                 put_nsproxy(new_nsproxy);
1835
1836 bad_unshare_cleanup_fd:
1837         if (new_fd)
1838                 put_files_struct(new_fd);
1839
1840 bad_unshare_cleanup_fs:
1841         if (new_fs)
1842                 free_fs_struct(new_fs);
1843
1844 bad_unshare_out:
1845         return err;
1846 }
1847
1848 /*
1849  *      Helper to unshare the files of the current task.
1850  *      We don't want to expose copy_files internals to
1851  *      the exec layer of the kernel.
1852  */
1853
1854 int unshare_files(struct files_struct **displaced)
1855 {
1856         struct task_struct *task = current;
1857         struct files_struct *copy = NULL;
1858         int error;
1859
1860         error = unshare_fd(CLONE_FILES, &copy);
1861         if (error || !copy) {
1862                 *displaced = NULL;
1863                 return error;
1864         }
1865         *displaced = task->files;
1866         task_lock(task);
1867         task->files = copy;
1868         task_unlock(task);
1869         return 0;
1870 }