ARM: tegra: cpuquiet: fix race condition
[linux-2.6.git] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/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/proc_fs.h>
51 #include <linux/profile.h>
52 #include <linux/rmap.h>
53 #include <linux/ksm.h>
54 #include <linux/acct.h>
55 #include <linux/tsacct_kern.h>
56 #include <linux/cn_proc.h>
57 #include <linux/freezer.h>
58 #include <linux/delayacct.h>
59 #include <linux/taskstats_kern.h>
60 #include <linux/random.h>
61 #include <linux/tty.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 #include <linux/oom.h>
69 #include <linux/khugepaged.h>
70 #include <linux/signalfd.h>
71
72 #include <asm/pgtable.h>
73 #include <asm/pgalloc.h>
74 #include <asm/uaccess.h>
75 #include <asm/mmu_context.h>
76 #include <asm/cacheflush.h>
77 #include <asm/tlbflush.h>
78
79 #include <trace/events/sched.h>
80
81 #define CREATE_TRACE_POINTS
82 #include <trace/events/task.h>
83
84 /*
85  * Protected counters by write_lock_irq(&tasklist_lock)
86  */
87 unsigned long total_forks;      /* Handle normal Linux uptimes. */
88 int nr_threads;                 /* The idle threads do not count.. */
89
90 int max_threads;                /* tunable limit on nr_threads */
91
92 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
93
94 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
95
96 #ifdef CONFIG_PROVE_RCU
97 int lockdep_tasklist_lock_is_held(void)
98 {
99         return lockdep_is_held(&tasklist_lock);
100 }
101 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
102 #endif /* #ifdef CONFIG_PROVE_RCU */
103
104 int nr_processes(void)
105 {
106         int cpu;
107         int total = 0;
108
109         for_each_possible_cpu(cpu)
110                 total += per_cpu(process_counts, cpu);
111
112         return total;
113 }
114
115 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
116 # define alloc_task_struct_node(node)           \
117                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
118 # define free_task_struct(tsk)                  \
119                 kmem_cache_free(task_struct_cachep, (tsk))
120 static struct kmem_cache *task_struct_cachep;
121 #endif
122
123 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
124 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
125                                                   int node)
126 {
127 #ifdef CONFIG_DEBUG_STACK_USAGE
128         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
129 #else
130         gfp_t mask = GFP_KERNEL;
131 #endif
132         struct page *page = alloc_pages_node(node, mask, THREAD_SIZE_ORDER);
133
134         return page ? page_address(page) : NULL;
135 }
136
137 static inline void free_thread_info(struct thread_info *ti)
138 {
139         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
140 }
141 #endif
142
143 /* SLAB cache for signal_struct structures (tsk->signal) */
144 static struct kmem_cache *signal_cachep;
145
146 /* SLAB cache for sighand_struct structures (tsk->sighand) */
147 struct kmem_cache *sighand_cachep;
148
149 /* SLAB cache for files_struct structures (tsk->files) */
150 struct kmem_cache *files_cachep;
151
152 /* SLAB cache for fs_struct structures (tsk->fs) */
153 struct kmem_cache *fs_cachep;
154
155 /* SLAB cache for vm_area_struct structures */
156 struct kmem_cache *vm_area_cachep;
157
158 /* SLAB cache for mm_struct structures (tsk->mm) */
159 static struct kmem_cache *mm_cachep;
160
161 /* Notifier list called when a task struct is freed */
162 static ATOMIC_NOTIFIER_HEAD(task_free_notifier);
163
164 static void account_kernel_stack(struct thread_info *ti, int account)
165 {
166         struct zone *zone = page_zone(virt_to_page(ti));
167
168         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
169 }
170
171 void free_task(struct task_struct *tsk)
172 {
173         account_kernel_stack(tsk->stack, -1);
174         free_thread_info(tsk->stack);
175         rt_mutex_debug_task_free(tsk);
176         ftrace_graph_exit_task(tsk);
177         free_task_struct(tsk);
178 }
179 EXPORT_SYMBOL(free_task);
180
181 static inline void free_signal_struct(struct signal_struct *sig)
182 {
183         taskstats_tgid_free(sig);
184         sched_autogroup_exit(sig);
185         kmem_cache_free(signal_cachep, sig);
186 }
187
188 static inline void put_signal_struct(struct signal_struct *sig)
189 {
190         if (atomic_dec_and_test(&sig->sigcnt))
191                 free_signal_struct(sig);
192 }
193
194 int task_free_register(struct notifier_block *n)
195 {
196         return atomic_notifier_chain_register(&task_free_notifier, n);
197 }
198 EXPORT_SYMBOL(task_free_register);
199
200 int task_free_unregister(struct notifier_block *n)
201 {
202         return atomic_notifier_chain_unregister(&task_free_notifier, n);
203 }
204 EXPORT_SYMBOL(task_free_unregister);
205
206 void __put_task_struct(struct task_struct *tsk)
207 {
208         WARN_ON(!tsk->exit_state);
209         WARN_ON(atomic_read(&tsk->usage));
210         WARN_ON(tsk == current);
211
212         security_task_free(tsk);
213         exit_creds(tsk);
214         delayacct_tsk_free(tsk);
215         put_signal_struct(tsk->signal);
216
217         atomic_notifier_call_chain(&task_free_notifier, 0, tsk);
218         if (!profile_handoff_task(tsk))
219                 free_task(tsk);
220 }
221 EXPORT_SYMBOL_GPL(__put_task_struct);
222
223 /*
224  * macro override instead of weak attribute alias, to workaround
225  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
226  */
227 #ifndef arch_task_cache_init
228 #define arch_task_cache_init()
229 #endif
230
231 void __init fork_init(unsigned long mempages)
232 {
233 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
234 #ifndef ARCH_MIN_TASKALIGN
235 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
236 #endif
237         /* create a slab on which task_structs can be allocated */
238         task_struct_cachep =
239                 kmem_cache_create("task_struct", sizeof(struct task_struct),
240                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
241 #endif
242
243         /* do the arch specific task caches init */
244         arch_task_cache_init();
245
246         /*
247          * The default maximum number of threads is set to a safe
248          * value: the thread structures can take up at most half
249          * of memory.
250          */
251         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
252
253         /*
254          * we need to allow at least 20 threads to boot a system
255          */
256         if (max_threads < 20)
257                 max_threads = 20;
258
259         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
260         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
261         init_task.signal->rlim[RLIMIT_SIGPENDING] =
262                 init_task.signal->rlim[RLIMIT_NPROC];
263 }
264
265 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
266                                                struct task_struct *src)
267 {
268         *dst = *src;
269         return 0;
270 }
271
272 static struct task_struct *dup_task_struct(struct task_struct *orig)
273 {
274         struct task_struct *tsk;
275         struct thread_info *ti;
276         unsigned long *stackend;
277         int node = tsk_fork_get_node(orig);
278         int err;
279
280         prepare_to_copy(orig);
281
282         tsk = alloc_task_struct_node(node);
283         if (!tsk)
284                 return NULL;
285
286         ti = alloc_thread_info_node(tsk, node);
287         if (!ti) {
288                 free_task_struct(tsk);
289                 return NULL;
290         }
291
292         err = arch_dup_task_struct(tsk, orig);
293         if (err)
294                 goto out;
295
296         tsk->stack = ti;
297
298         setup_thread_stack(tsk, orig);
299         clear_user_return_notifier(tsk);
300         clear_tsk_need_resched(tsk);
301         stackend = end_of_stack(tsk);
302         *stackend = STACK_END_MAGIC;    /* for overflow detection */
303
304 #ifdef CONFIG_CC_STACKPROTECTOR
305         tsk->stack_canary = get_random_int();
306 #endif
307
308         /*
309          * One for us, one for whoever does the "release_task()" (usually
310          * parent)
311          */
312         atomic_set(&tsk->usage, 2);
313 #ifdef CONFIG_BLK_DEV_IO_TRACE
314         tsk->btrace_seq = 0;
315 #endif
316         tsk->splice_pipe = NULL;
317
318         account_kernel_stack(ti, 1);
319
320         return tsk;
321
322 out:
323         free_thread_info(ti);
324         free_task_struct(tsk);
325         return NULL;
326 }
327
328 #ifdef CONFIG_MMU
329 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
330 {
331         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
332         struct rb_node **rb_link, *rb_parent;
333         int retval;
334         unsigned long charge;
335         struct mempolicy *pol;
336
337         down_write(&oldmm->mmap_sem);
338         flush_cache_dup_mm(oldmm);
339         /*
340          * Not linked in yet - no deadlock potential:
341          */
342         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
343
344         mm->locked_vm = 0;
345         mm->mmap = NULL;
346         mm->mmap_cache = NULL;
347         mm->free_area_cache = oldmm->mmap_base;
348         mm->cached_hole_size = ~0UL;
349         mm->map_count = 0;
350         cpumask_clear(mm_cpumask(mm));
351         mm->mm_rb = RB_ROOT;
352         rb_link = &mm->mm_rb.rb_node;
353         rb_parent = NULL;
354         pprev = &mm->mmap;
355         retval = ksm_fork(mm, oldmm);
356         if (retval)
357                 goto out;
358         retval = khugepaged_fork(mm, oldmm);
359         if (retval)
360                 goto out;
361
362         prev = NULL;
363         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
364                 struct file *file;
365
366                 if (mpnt->vm_flags & VM_DONTCOPY) {
367                         long pages = vma_pages(mpnt);
368                         mm->total_vm -= pages;
369                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
370                                                                 -pages);
371                         continue;
372                 }
373                 charge = 0;
374                 if (mpnt->vm_flags & VM_ACCOUNT) {
375                         unsigned long len;
376                         len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
377                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
378                                 goto fail_nomem;
379                         charge = len;
380                 }
381                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
382                 if (!tmp)
383                         goto fail_nomem;
384                 *tmp = *mpnt;
385                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
386                 pol = mpol_dup(vma_policy(mpnt));
387                 retval = PTR_ERR(pol);
388                 if (IS_ERR(pol))
389                         goto fail_nomem_policy;
390                 vma_set_policy(tmp, pol);
391                 tmp->vm_mm = mm;
392                 if (anon_vma_fork(tmp, mpnt))
393                         goto fail_nomem_anon_vma_fork;
394                 tmp->vm_flags &= ~VM_LOCKED;
395                 tmp->vm_next = tmp->vm_prev = NULL;
396                 file = tmp->vm_file;
397                 if (file) {
398                         struct inode *inode = file->f_path.dentry->d_inode;
399                         struct address_space *mapping = file->f_mapping;
400
401                         get_file(file);
402                         if (tmp->vm_flags & VM_DENYWRITE)
403                                 atomic_dec(&inode->i_writecount);
404                         mutex_lock(&mapping->i_mmap_mutex);
405                         if (tmp->vm_flags & VM_SHARED)
406                                 mapping->i_mmap_writable++;
407                         flush_dcache_mmap_lock(mapping);
408                         /* insert tmp into the share list, just after mpnt */
409                         vma_prio_tree_add(tmp, mpnt);
410                         flush_dcache_mmap_unlock(mapping);
411                         mutex_unlock(&mapping->i_mmap_mutex);
412                 }
413
414                 /*
415                  * Clear hugetlb-related page reserves for children. This only
416                  * affects MAP_PRIVATE mappings. Faults generated by the child
417                  * are not guaranteed to succeed, even if read-only
418                  */
419                 if (is_vm_hugetlb_page(tmp))
420                         reset_vma_resv_huge_pages(tmp);
421
422                 /*
423                  * Link in the new vma and copy the page table entries.
424                  */
425                 *pprev = tmp;
426                 pprev = &tmp->vm_next;
427                 tmp->vm_prev = prev;
428                 prev = tmp;
429
430                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
431                 rb_link = &tmp->vm_rb.rb_right;
432                 rb_parent = &tmp->vm_rb;
433
434                 mm->map_count++;
435                 retval = copy_page_range(mm, oldmm, mpnt);
436
437                 if (tmp->vm_ops && tmp->vm_ops->open)
438                         tmp->vm_ops->open(tmp);
439
440                 if (retval)
441                         goto out;
442         }
443         /* a new mm has just been created */
444         arch_dup_mmap(oldmm, mm);
445         retval = 0;
446 out:
447         up_write(&mm->mmap_sem);
448         flush_tlb_mm(oldmm);
449         up_write(&oldmm->mmap_sem);
450         return retval;
451 fail_nomem_anon_vma_fork:
452         mpol_put(pol);
453 fail_nomem_policy:
454         kmem_cache_free(vm_area_cachep, tmp);
455 fail_nomem:
456         retval = -ENOMEM;
457         vm_unacct_memory(charge);
458         goto out;
459 }
460
461 static inline int mm_alloc_pgd(struct mm_struct *mm)
462 {
463         mm->pgd = pgd_alloc(mm);
464         if (unlikely(!mm->pgd))
465                 return -ENOMEM;
466         return 0;
467 }
468
469 static inline void mm_free_pgd(struct mm_struct *mm)
470 {
471         pgd_free(mm, mm->pgd);
472 }
473 #else
474 #define dup_mmap(mm, oldmm)     (0)
475 #define mm_alloc_pgd(mm)        (0)
476 #define mm_free_pgd(mm)
477 #endif /* CONFIG_MMU */
478
479 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
480
481 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
482 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
483
484 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
485
486 static int __init coredump_filter_setup(char *s)
487 {
488         default_dump_filter =
489                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
490                 MMF_DUMP_FILTER_MASK;
491         return 1;
492 }
493
494 __setup("coredump_filter=", coredump_filter_setup);
495
496 #include <linux/init_task.h>
497
498 static void mm_init_aio(struct mm_struct *mm)
499 {
500 #ifdef CONFIG_AIO
501         spin_lock_init(&mm->ioctx_lock);
502         INIT_HLIST_HEAD(&mm->ioctx_list);
503 #endif
504 }
505
506 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
507 {
508         atomic_set(&mm->mm_users, 1);
509         atomic_set(&mm->mm_count, 1);
510         init_rwsem(&mm->mmap_sem);
511         INIT_LIST_HEAD(&mm->mmlist);
512         mm->flags = (current->mm) ?
513                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
514         mm->core_state = NULL;
515         mm->nr_ptes = 0;
516         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
517         spin_lock_init(&mm->page_table_lock);
518         mm->free_area_cache = TASK_UNMAPPED_BASE;
519         mm->cached_hole_size = ~0UL;
520         mm_init_aio(mm);
521         mm_init_owner(mm, p);
522
523         if (likely(!mm_alloc_pgd(mm))) {
524                 mm->def_flags = 0;
525                 mmu_notifier_mm_init(mm);
526                 return mm;
527         }
528
529         free_mm(mm);
530         return NULL;
531 }
532
533 static void check_mm(struct mm_struct *mm)
534 {
535         int i;
536
537         for (i = 0; i < NR_MM_COUNTERS; i++) {
538                 long x = atomic_long_read(&mm->rss_stat.count[i]);
539
540                 if (unlikely(x))
541                         printk(KERN_ALERT "BUG: Bad rss-counter state "
542                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
543         }
544
545 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
546         VM_BUG_ON(mm->pmd_huge_pte);
547 #endif
548 }
549
550 /*
551  * Allocate and initialize an mm_struct.
552  */
553 struct mm_struct *mm_alloc(void)
554 {
555         struct mm_struct *mm;
556
557         mm = allocate_mm();
558         if (!mm)
559                 return NULL;
560
561         memset(mm, 0, sizeof(*mm));
562         mm_init_cpumask(mm);
563         return mm_init(mm, current);
564 }
565
566 /*
567  * Called when the last reference to the mm
568  * is dropped: either by a lazy thread or by
569  * mmput. Free the page directory and the mm.
570  */
571 void __mmdrop(struct mm_struct *mm)
572 {
573         BUG_ON(mm == &init_mm);
574         mm_free_pgd(mm);
575         destroy_context(mm);
576         mmu_notifier_mm_destroy(mm);
577         check_mm(mm);
578         free_mm(mm);
579 }
580 EXPORT_SYMBOL_GPL(__mmdrop);
581
582 /*
583  * Decrement the use count and release all resources for an mm.
584  */
585 void mmput(struct mm_struct *mm)
586 {
587         might_sleep();
588
589         if (atomic_dec_and_test(&mm->mm_users)) {
590                 exit_aio(mm);
591                 ksm_exit(mm);
592                 khugepaged_exit(mm); /* must run before exit_mmap */
593                 exit_mmap(mm);
594                 set_mm_exe_file(mm, NULL);
595                 if (!list_empty(&mm->mmlist)) {
596                         spin_lock(&mmlist_lock);
597                         list_del(&mm->mmlist);
598                         spin_unlock(&mmlist_lock);
599                 }
600                 put_swap_token(mm);
601                 if (mm->binfmt)
602                         module_put(mm->binfmt->module);
603                 mmdrop(mm);
604         }
605 }
606 EXPORT_SYMBOL_GPL(mmput);
607
608 /*
609  * We added or removed a vma mapping the executable. The vmas are only mapped
610  * during exec and are not mapped with the mmap system call.
611  * Callers must hold down_write() on the mm's mmap_sem for these
612  */
613 void added_exe_file_vma(struct mm_struct *mm)
614 {
615         mm->num_exe_file_vmas++;
616 }
617
618 void removed_exe_file_vma(struct mm_struct *mm)
619 {
620         mm->num_exe_file_vmas--;
621         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
622                 fput(mm->exe_file);
623                 mm->exe_file = NULL;
624         }
625
626 }
627
628 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
629 {
630         if (new_exe_file)
631                 get_file(new_exe_file);
632         if (mm->exe_file)
633                 fput(mm->exe_file);
634         mm->exe_file = new_exe_file;
635         mm->num_exe_file_vmas = 0;
636 }
637
638 struct file *get_mm_exe_file(struct mm_struct *mm)
639 {
640         struct file *exe_file;
641
642         /* We need mmap_sem to protect against races with removal of
643          * VM_EXECUTABLE vmas */
644         down_read(&mm->mmap_sem);
645         exe_file = mm->exe_file;
646         if (exe_file)
647                 get_file(exe_file);
648         up_read(&mm->mmap_sem);
649         return exe_file;
650 }
651
652 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
653 {
654         /* It's safe to write the exe_file pointer without exe_file_lock because
655          * this is called during fork when the task is not yet in /proc */
656         newmm->exe_file = get_mm_exe_file(oldmm);
657 }
658
659 /**
660  * get_task_mm - acquire a reference to the task's mm
661  *
662  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
663  * this kernel workthread has transiently adopted a user mm with use_mm,
664  * to do its AIO) is not set and if so returns a reference to it, after
665  * bumping up the use count.  User must release the mm via mmput()
666  * after use.  Typically used by /proc and ptrace.
667  */
668 struct mm_struct *get_task_mm(struct task_struct *task)
669 {
670         struct mm_struct *mm;
671
672         task_lock(task);
673         mm = task->mm;
674         if (mm) {
675                 if (task->flags & PF_KTHREAD)
676                         mm = NULL;
677                 else
678                         atomic_inc(&mm->mm_users);
679         }
680         task_unlock(task);
681         return mm;
682 }
683 EXPORT_SYMBOL_GPL(get_task_mm);
684
685 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
686 {
687         struct mm_struct *mm;
688         int err;
689
690         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
691         if (err)
692                 return ERR_PTR(err);
693
694         mm = get_task_mm(task);
695         if (mm && mm != current->mm &&
696                         !ptrace_may_access(task, mode) &&
697                         !capable(CAP_SYS_RESOURCE)) {
698                 mmput(mm);
699                 mm = ERR_PTR(-EACCES);
700         }
701         mutex_unlock(&task->signal->cred_guard_mutex);
702
703         return mm;
704 }
705
706 static void complete_vfork_done(struct task_struct *tsk)
707 {
708         struct completion *vfork;
709
710         task_lock(tsk);
711         vfork = tsk->vfork_done;
712         if (likely(vfork)) {
713                 tsk->vfork_done = NULL;
714                 complete(vfork);
715         }
716         task_unlock(tsk);
717 }
718
719 static int wait_for_vfork_done(struct task_struct *child,
720                                 struct completion *vfork)
721 {
722         int killed;
723
724         freezer_do_not_count();
725         killed = wait_for_completion_killable(vfork);
726         freezer_count();
727
728         if (killed) {
729                 task_lock(child);
730                 child->vfork_done = NULL;
731                 task_unlock(child);
732         }
733
734         put_task_struct(child);
735         return killed;
736 }
737
738 /* Please note the differences between mmput and mm_release.
739  * mmput is called whenever we stop holding onto a mm_struct,
740  * error success whatever.
741  *
742  * mm_release is called after a mm_struct has been removed
743  * from the current process.
744  *
745  * This difference is important for error handling, when we
746  * only half set up a mm_struct for a new process and need to restore
747  * the old one.  Because we mmput the new mm_struct before
748  * restoring the old one. . .
749  * Eric Biederman 10 January 1998
750  */
751 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
752 {
753         /* Get rid of any futexes when releasing the mm */
754 #ifdef CONFIG_FUTEX
755         if (unlikely(tsk->robust_list)) {
756                 exit_robust_list(tsk);
757                 tsk->robust_list = NULL;
758         }
759 #ifdef CONFIG_COMPAT
760         if (unlikely(tsk->compat_robust_list)) {
761                 compat_exit_robust_list(tsk);
762                 tsk->compat_robust_list = NULL;
763         }
764 #endif
765         if (unlikely(!list_empty(&tsk->pi_state_list)))
766                 exit_pi_state_list(tsk);
767 #endif
768
769         /* Get rid of any cached register state */
770         deactivate_mm(tsk, mm);
771
772         if (tsk->vfork_done)
773                 complete_vfork_done(tsk);
774
775         /*
776          * If we're exiting normally, clear a user-space tid field if
777          * requested.  We leave this alone when dying by signal, to leave
778          * the value intact in a core dump, and to save the unnecessary
779          * trouble, say, a killed vfork parent shouldn't touch this mm.
780          * Userland only wants this done for a sys_exit.
781          */
782         if (tsk->clear_child_tid) {
783                 if (!(tsk->flags & PF_SIGNALED) &&
784                     atomic_read(&mm->mm_users) > 1) {
785                         /*
786                          * We don't check the error code - if userspace has
787                          * not set up a proper pointer then tough luck.
788                          */
789                         put_user(0, tsk->clear_child_tid);
790                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
791                                         1, NULL, NULL, 0);
792                 }
793                 tsk->clear_child_tid = NULL;
794         }
795 }
796
797 /*
798  * Allocate a new mm structure and copy contents from the
799  * mm structure of the passed in task structure.
800  */
801 struct mm_struct *dup_mm(struct task_struct *tsk)
802 {
803         struct mm_struct *mm, *oldmm = current->mm;
804         int err;
805
806         if (!oldmm)
807                 return NULL;
808
809         mm = allocate_mm();
810         if (!mm)
811                 goto fail_nomem;
812
813         memcpy(mm, oldmm, sizeof(*mm));
814         mm_init_cpumask(mm);
815
816         /* Initializing for Swap token stuff */
817         mm->token_priority = 0;
818         mm->last_interval = 0;
819
820 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
821         mm->pmd_huge_pte = NULL;
822 #endif
823
824         if (!mm_init(mm, tsk))
825                 goto fail_nomem;
826
827         if (init_new_context(tsk, mm))
828                 goto fail_nocontext;
829
830         dup_mm_exe_file(oldmm, mm);
831
832         err = dup_mmap(mm, oldmm);
833         if (err)
834                 goto free_pt;
835
836         mm->hiwater_rss = get_mm_rss(mm);
837         mm->hiwater_vm = mm->total_vm;
838
839         if (mm->binfmt && !try_module_get(mm->binfmt->module))
840                 goto free_pt;
841
842         return mm;
843
844 free_pt:
845         /* don't put binfmt in mmput, we haven't got module yet */
846         mm->binfmt = NULL;
847         mmput(mm);
848
849 fail_nomem:
850         return NULL;
851
852 fail_nocontext:
853         /*
854          * If init_new_context() failed, we cannot use mmput() to free the mm
855          * because it calls destroy_context()
856          */
857         mm_free_pgd(mm);
858         free_mm(mm);
859         return NULL;
860 }
861
862 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
863 {
864         struct mm_struct *mm, *oldmm;
865         int retval;
866
867         tsk->min_flt = tsk->maj_flt = 0;
868         tsk->nvcsw = tsk->nivcsw = 0;
869 #ifdef CONFIG_DETECT_HUNG_TASK
870         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
871 #endif
872
873         tsk->mm = NULL;
874         tsk->active_mm = NULL;
875
876         /*
877          * Are we cloning a kernel thread?
878          *
879          * We need to steal a active VM for that..
880          */
881         oldmm = current->mm;
882         if (!oldmm)
883                 return 0;
884
885         if (clone_flags & CLONE_VM) {
886                 atomic_inc(&oldmm->mm_users);
887                 mm = oldmm;
888                 goto good_mm;
889         }
890
891         retval = -ENOMEM;
892         mm = dup_mm(tsk);
893         if (!mm)
894                 goto fail_nomem;
895
896 good_mm:
897         /* Initializing for Swap token stuff */
898         mm->token_priority = 0;
899         mm->last_interval = 0;
900
901         tsk->mm = mm;
902         tsk->active_mm = mm;
903         return 0;
904
905 fail_nomem:
906         return retval;
907 }
908
909 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
910 {
911         struct fs_struct *fs = current->fs;
912         if (clone_flags & CLONE_FS) {
913                 /* tsk->fs is already what we want */
914                 spin_lock(&fs->lock);
915                 if (fs->in_exec) {
916                         spin_unlock(&fs->lock);
917                         return -EAGAIN;
918                 }
919                 fs->users++;
920                 spin_unlock(&fs->lock);
921                 return 0;
922         }
923         tsk->fs = copy_fs_struct(fs);
924         if (!tsk->fs)
925                 return -ENOMEM;
926         return 0;
927 }
928
929 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
930 {
931         struct files_struct *oldf, *newf;
932         int error = 0;
933
934         /*
935          * A background process may not have any files ...
936          */
937         oldf = current->files;
938         if (!oldf)
939                 goto out;
940
941         if (clone_flags & CLONE_FILES) {
942                 atomic_inc(&oldf->count);
943                 goto out;
944         }
945
946         newf = dup_fd(oldf, &error);
947         if (!newf)
948                 goto out;
949
950         tsk->files = newf;
951         error = 0;
952 out:
953         return error;
954 }
955
956 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
957 {
958 #ifdef CONFIG_BLOCK
959         struct io_context *ioc = current->io_context;
960         struct io_context *new_ioc;
961
962         if (!ioc)
963                 return 0;
964         /*
965          * Share io context with parent, if CLONE_IO is set
966          */
967         if (clone_flags & CLONE_IO) {
968                 tsk->io_context = ioc_task_link(ioc);
969                 if (unlikely(!tsk->io_context))
970                         return -ENOMEM;
971         } else if (ioprio_valid(ioc->ioprio)) {
972                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
973                 if (unlikely(!new_ioc))
974                         return -ENOMEM;
975
976                 new_ioc->ioprio = ioc->ioprio;
977                 put_io_context(new_ioc);
978         }
979 #endif
980         return 0;
981 }
982
983 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
984 {
985         struct sighand_struct *sig;
986
987         if (clone_flags & CLONE_SIGHAND) {
988                 atomic_inc(&current->sighand->count);
989                 return 0;
990         }
991         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
992         rcu_assign_pointer(tsk->sighand, sig);
993         if (!sig)
994                 return -ENOMEM;
995         atomic_set(&sig->count, 1);
996         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
997         return 0;
998 }
999
1000 void __cleanup_sighand(struct sighand_struct *sighand)
1001 {
1002         if (atomic_dec_and_test(&sighand->count)) {
1003                 signalfd_cleanup(sighand);
1004                 kmem_cache_free(sighand_cachep, sighand);
1005         }
1006 }
1007
1008
1009 /*
1010  * Initialize POSIX timer handling for a thread group.
1011  */
1012 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1013 {
1014         unsigned long cpu_limit;
1015
1016         /* Thread group counters. */
1017         thread_group_cputime_init(sig);
1018
1019         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1020         if (cpu_limit != RLIM_INFINITY) {
1021                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1022                 sig->cputimer.running = 1;
1023         }
1024
1025         /* The timer lists. */
1026         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1027         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1028         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1029 }
1030
1031 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1032 {
1033         struct signal_struct *sig;
1034
1035         if (clone_flags & CLONE_THREAD)
1036                 return 0;
1037
1038         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1039         tsk->signal = sig;
1040         if (!sig)
1041                 return -ENOMEM;
1042
1043         sig->nr_threads = 1;
1044         atomic_set(&sig->live, 1);
1045         atomic_set(&sig->sigcnt, 1);
1046         init_waitqueue_head(&sig->wait_chldexit);
1047         if (clone_flags & CLONE_NEWPID)
1048                 sig->flags |= SIGNAL_UNKILLABLE;
1049         sig->curr_target = tsk;
1050         init_sigpending(&sig->shared_pending);
1051         INIT_LIST_HEAD(&sig->posix_timers);
1052
1053         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1054         sig->real_timer.function = it_real_fn;
1055
1056         task_lock(current->group_leader);
1057         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1058         task_unlock(current->group_leader);
1059
1060         posix_cpu_timers_init_group(sig);
1061
1062         tty_audit_fork(sig);
1063         sched_autogroup_fork(sig);
1064
1065 #ifdef CONFIG_CGROUPS
1066         init_rwsem(&sig->group_rwsem);
1067 #endif
1068
1069         sig->oom_adj = current->signal->oom_adj;
1070         sig->oom_score_adj = current->signal->oom_score_adj;
1071         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1072
1073         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1074                                    current->signal->is_child_subreaper;
1075
1076         mutex_init(&sig->cred_guard_mutex);
1077
1078         return 0;
1079 }
1080
1081 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1082 {
1083         unsigned long new_flags = p->flags;
1084
1085         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1086         new_flags |= PF_FORKNOEXEC;
1087         p->flags = new_flags;
1088 }
1089
1090 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1091 {
1092         current->clear_child_tid = tidptr;
1093
1094         return task_pid_vnr(current);
1095 }
1096
1097 static void rt_mutex_init_task(struct task_struct *p)
1098 {
1099         raw_spin_lock_init(&p->pi_lock);
1100 #ifdef CONFIG_RT_MUTEXES
1101         plist_head_init(&p->pi_waiters);
1102         p->pi_blocked_on = NULL;
1103 #endif
1104 }
1105
1106 #ifdef CONFIG_MM_OWNER
1107 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1108 {
1109         mm->owner = p;
1110 }
1111 #endif /* CONFIG_MM_OWNER */
1112
1113 /*
1114  * Initialize POSIX timer handling for a single task.
1115  */
1116 static void posix_cpu_timers_init(struct task_struct *tsk)
1117 {
1118         tsk->cputime_expires.prof_exp = 0;
1119         tsk->cputime_expires.virt_exp = 0;
1120         tsk->cputime_expires.sched_exp = 0;
1121         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1122         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1123         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1124 }
1125
1126 /*
1127  * This creates a new process as a copy of the old one,
1128  * but does not actually start it yet.
1129  *
1130  * It copies the registers, and all the appropriate
1131  * parts of the process environment (as per the clone
1132  * flags). The actual kick-off is left to the caller.
1133  */
1134 static struct task_struct *copy_process(unsigned long clone_flags,
1135                                         unsigned long stack_start,
1136                                         struct pt_regs *regs,
1137                                         unsigned long stack_size,
1138                                         int __user *child_tidptr,
1139                                         struct pid *pid,
1140                                         int trace)
1141 {
1142         int retval;
1143         struct task_struct *p;
1144         int cgroup_callbacks_done = 0;
1145
1146         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1147                 return ERR_PTR(-EINVAL);
1148
1149         /*
1150          * Thread groups must share signals as well, and detached threads
1151          * can only be started up within the thread group.
1152          */
1153         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1154                 return ERR_PTR(-EINVAL);
1155
1156         /*
1157          * Shared signal handlers imply shared VM. By way of the above,
1158          * thread groups also imply shared VM. Blocking this case allows
1159          * for various simplifications in other code.
1160          */
1161         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1162                 return ERR_PTR(-EINVAL);
1163
1164         /*
1165          * Siblings of global init remain as zombies on exit since they are
1166          * not reaped by their parent (swapper). To solve this and to avoid
1167          * multi-rooted process trees, prevent global and container-inits
1168          * from creating siblings.
1169          */
1170         if ((clone_flags & CLONE_PARENT) &&
1171                                 current->signal->flags & SIGNAL_UNKILLABLE)
1172                 return ERR_PTR(-EINVAL);
1173
1174         retval = security_task_create(clone_flags);
1175         if (retval)
1176                 goto fork_out;
1177
1178         retval = -ENOMEM;
1179         p = dup_task_struct(current);
1180         if (!p)
1181                 goto fork_out;
1182
1183         ftrace_graph_init_task(p);
1184
1185         rt_mutex_init_task(p);
1186
1187 #ifdef CONFIG_PROVE_LOCKING
1188         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1189         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1190 #endif
1191         retval = -EAGAIN;
1192         if (atomic_read(&p->real_cred->user->processes) >=
1193                         task_rlimit(p, RLIMIT_NPROC)) {
1194                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1195                     p->real_cred->user != INIT_USER)
1196                         goto bad_fork_free;
1197         }
1198         current->flags &= ~PF_NPROC_EXCEEDED;
1199
1200         retval = copy_creds(p, clone_flags);
1201         if (retval < 0)
1202                 goto bad_fork_free;
1203
1204         /*
1205          * If multiple threads are within copy_process(), then this check
1206          * triggers too late. This doesn't hurt, the check is only there
1207          * to stop root fork bombs.
1208          */
1209         retval = -EAGAIN;
1210         if (nr_threads >= max_threads)
1211                 goto bad_fork_cleanup_count;
1212
1213         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1214                 goto bad_fork_cleanup_count;
1215
1216         p->did_exec = 0;
1217         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1218         copy_flags(clone_flags, p);
1219         INIT_LIST_HEAD(&p->children);
1220         INIT_LIST_HEAD(&p->sibling);
1221         rcu_copy_process(p);
1222         p->vfork_done = NULL;
1223         spin_lock_init(&p->alloc_lock);
1224
1225         init_sigpending(&p->pending);
1226
1227         p->utime = p->stime = p->gtime = 0;
1228         p->utimescaled = p->stimescaled = 0;
1229 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1230         p->prev_utime = p->prev_stime = 0;
1231 #endif
1232 #if defined(SPLIT_RSS_COUNTING)
1233         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1234 #endif
1235
1236         p->default_timer_slack_ns = current->timer_slack_ns;
1237
1238         task_io_accounting_init(&p->ioac);
1239         acct_clear_integrals(p);
1240
1241         posix_cpu_timers_init(p);
1242
1243         do_posix_clock_monotonic_gettime(&p->start_time);
1244         p->real_start_time = p->start_time;
1245         monotonic_to_bootbased(&p->real_start_time);
1246         p->io_context = NULL;
1247         p->audit_context = NULL;
1248         if (clone_flags & CLONE_THREAD)
1249                 threadgroup_change_begin(current);
1250         cgroup_fork(p);
1251 #ifdef CONFIG_NUMA
1252         p->mempolicy = mpol_dup(p->mempolicy);
1253         if (IS_ERR(p->mempolicy)) {
1254                 retval = PTR_ERR(p->mempolicy);
1255                 p->mempolicy = NULL;
1256                 goto bad_fork_cleanup_cgroup;
1257         }
1258         mpol_fix_fork_child_flag(p);
1259 #endif
1260 #ifdef CONFIG_CPUSETS
1261         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1262         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1263         seqcount_init(&p->mems_allowed_seq);
1264 #endif
1265 #ifdef CONFIG_TRACE_IRQFLAGS
1266         p->irq_events = 0;
1267 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1268         p->hardirqs_enabled = 1;
1269 #else
1270         p->hardirqs_enabled = 0;
1271 #endif
1272         p->hardirq_enable_ip = 0;
1273         p->hardirq_enable_event = 0;
1274         p->hardirq_disable_ip = _THIS_IP_;
1275         p->hardirq_disable_event = 0;
1276         p->softirqs_enabled = 1;
1277         p->softirq_enable_ip = _THIS_IP_;
1278         p->softirq_enable_event = 0;
1279         p->softirq_disable_ip = 0;
1280         p->softirq_disable_event = 0;
1281         p->hardirq_context = 0;
1282         p->softirq_context = 0;
1283 #endif
1284 #ifdef CONFIG_LOCKDEP
1285         p->lockdep_depth = 0; /* no locks held yet */
1286         p->curr_chain_key = 0;
1287         p->lockdep_recursion = 0;
1288 #endif
1289
1290 #ifdef CONFIG_DEBUG_MUTEXES
1291         p->blocked_on = NULL; /* not blocked yet */
1292 #endif
1293 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1294         p->memcg_batch.do_batch = 0;
1295         p->memcg_batch.memcg = NULL;
1296 #endif
1297
1298         /* Perform scheduler related setup. Assign this task to a CPU. */
1299         sched_fork(p);
1300
1301         retval = perf_event_init_task(p);
1302         if (retval)
1303                 goto bad_fork_cleanup_policy;
1304         retval = audit_alloc(p);
1305         if (retval)
1306                 goto bad_fork_cleanup_policy;
1307         /* copy all the process information */
1308         retval = copy_semundo(clone_flags, p);
1309         if (retval)
1310                 goto bad_fork_cleanup_audit;
1311         retval = copy_files(clone_flags, p);
1312         if (retval)
1313                 goto bad_fork_cleanup_semundo;
1314         retval = copy_fs(clone_flags, p);
1315         if (retval)
1316                 goto bad_fork_cleanup_files;
1317         retval = copy_sighand(clone_flags, p);
1318         if (retval)
1319                 goto bad_fork_cleanup_fs;
1320         retval = copy_signal(clone_flags, p);
1321         if (retval)
1322                 goto bad_fork_cleanup_sighand;
1323         retval = copy_mm(clone_flags, p);
1324         if (retval)
1325                 goto bad_fork_cleanup_signal;
1326         retval = copy_namespaces(clone_flags, p);
1327         if (retval)
1328                 goto bad_fork_cleanup_mm;
1329         retval = copy_io(clone_flags, p);
1330         if (retval)
1331                 goto bad_fork_cleanup_namespaces;
1332         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1333         if (retval)
1334                 goto bad_fork_cleanup_io;
1335
1336         if (pid != &init_struct_pid) {
1337                 retval = -ENOMEM;
1338                 pid = alloc_pid(p->nsproxy->pid_ns);
1339                 if (!pid)
1340                         goto bad_fork_cleanup_io;
1341         }
1342
1343         p->pid = pid_nr(pid);
1344         p->tgid = p->pid;
1345         if (clone_flags & CLONE_THREAD)
1346                 p->tgid = current->tgid;
1347
1348         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1349         /*
1350          * Clear TID on mm_release()?
1351          */
1352         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1353 #ifdef CONFIG_BLOCK
1354         p->plug = NULL;
1355 #endif
1356 #ifdef CONFIG_FUTEX
1357         p->robust_list = NULL;
1358 #ifdef CONFIG_COMPAT
1359         p->compat_robust_list = NULL;
1360 #endif
1361         INIT_LIST_HEAD(&p->pi_state_list);
1362         p->pi_state_cache = NULL;
1363 #endif
1364         /*
1365          * sigaltstack should be cleared when sharing the same VM
1366          */
1367         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1368                 p->sas_ss_sp = p->sas_ss_size = 0;
1369
1370         /*
1371          * Syscall tracing and stepping should be turned off in the
1372          * child regardless of CLONE_PTRACE.
1373          */
1374         user_disable_single_step(p);
1375         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1376 #ifdef TIF_SYSCALL_EMU
1377         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1378 #endif
1379         clear_all_latency_tracing(p);
1380
1381         /* ok, now we should be set up.. */
1382         if (clone_flags & CLONE_THREAD)
1383                 p->exit_signal = -1;
1384         else if (clone_flags & CLONE_PARENT)
1385                 p->exit_signal = current->group_leader->exit_signal;
1386         else
1387                 p->exit_signal = (clone_flags & CSIGNAL);
1388
1389         p->pdeath_signal = 0;
1390         p->exit_state = 0;
1391
1392         p->nr_dirtied = 0;
1393         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1394         p->dirty_paused_when = 0;
1395
1396         /*
1397          * Ok, make it visible to the rest of the system.
1398          * We dont wake it up yet.
1399          */
1400         p->group_leader = p;
1401         INIT_LIST_HEAD(&p->thread_group);
1402
1403         /* Now that the task is set up, run cgroup callbacks if
1404          * necessary. We need to run them before the task is visible
1405          * on the tasklist. */
1406         cgroup_fork_callbacks(p);
1407         cgroup_callbacks_done = 1;
1408
1409         /* Need tasklist lock for parent etc handling! */
1410         write_lock_irq(&tasklist_lock);
1411
1412         /* CLONE_PARENT re-uses the old parent */
1413         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1414                 p->real_parent = current->real_parent;
1415                 p->parent_exec_id = current->parent_exec_id;
1416         } else {
1417                 p->real_parent = current;
1418                 p->parent_exec_id = current->self_exec_id;
1419         }
1420
1421         spin_lock(&current->sighand->siglock);
1422
1423         /*
1424          * Process group and session signals need to be delivered to just the
1425          * parent before the fork or both the parent and the child after the
1426          * fork. Restart if a signal comes in before we add the new process to
1427          * it's process group.
1428          * A fatal signal pending means that current will exit, so the new
1429          * thread can't slip out of an OOM kill (or normal SIGKILL).
1430         */
1431         recalc_sigpending();
1432         if (signal_pending(current)) {
1433                 spin_unlock(&current->sighand->siglock);
1434                 write_unlock_irq(&tasklist_lock);
1435                 retval = -ERESTARTNOINTR;
1436                 goto bad_fork_free_pid;
1437         }
1438
1439         if (clone_flags & CLONE_THREAD) {
1440                 current->signal->nr_threads++;
1441                 atomic_inc(&current->signal->live);
1442                 atomic_inc(&current->signal->sigcnt);
1443                 p->group_leader = current->group_leader;
1444                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1445         }
1446
1447         if (likely(p->pid)) {
1448                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1449
1450                 if (thread_group_leader(p)) {
1451                         if (is_child_reaper(pid))
1452                                 p->nsproxy->pid_ns->child_reaper = p;
1453
1454                         p->signal->leader_pid = pid;
1455                         p->signal->tty = tty_kref_get(current->signal->tty);
1456                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1457                         attach_pid(p, PIDTYPE_SID, task_session(current));
1458                         list_add_tail(&p->sibling, &p->real_parent->children);
1459                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1460                         __this_cpu_inc(process_counts);
1461                 }
1462                 attach_pid(p, PIDTYPE_PID, pid);
1463                 nr_threads++;
1464         }
1465
1466         total_forks++;
1467         spin_unlock(&current->sighand->siglock);
1468         write_unlock_irq(&tasklist_lock);
1469         proc_fork_connector(p);
1470         cgroup_post_fork(p);
1471         if (clone_flags & CLONE_THREAD)
1472                 threadgroup_change_end(current);
1473         perf_event_fork(p);
1474
1475         trace_task_newtask(p, clone_flags);
1476
1477         return p;
1478
1479 bad_fork_free_pid:
1480         if (pid != &init_struct_pid)
1481                 free_pid(pid);
1482 bad_fork_cleanup_io:
1483         if (p->io_context)
1484                 exit_io_context(p);
1485 bad_fork_cleanup_namespaces:
1486         if (unlikely(clone_flags & CLONE_NEWPID))
1487                 pid_ns_release_proc(p->nsproxy->pid_ns);
1488         exit_task_namespaces(p);
1489 bad_fork_cleanup_mm:
1490         if (p->mm)
1491                 mmput(p->mm);
1492 bad_fork_cleanup_signal:
1493         if (!(clone_flags & CLONE_THREAD))
1494                 free_signal_struct(p->signal);
1495 bad_fork_cleanup_sighand:
1496         __cleanup_sighand(p->sighand);
1497 bad_fork_cleanup_fs:
1498         exit_fs(p); /* blocking */
1499 bad_fork_cleanup_files:
1500         exit_files(p); /* blocking */
1501 bad_fork_cleanup_semundo:
1502         exit_sem(p);
1503 bad_fork_cleanup_audit:
1504         audit_free(p);
1505 bad_fork_cleanup_policy:
1506         perf_event_free_task(p);
1507 #ifdef CONFIG_NUMA
1508         mpol_put(p->mempolicy);
1509 bad_fork_cleanup_cgroup:
1510 #endif
1511         if (clone_flags & CLONE_THREAD)
1512                 threadgroup_change_end(current);
1513         cgroup_exit(p, cgroup_callbacks_done);
1514         delayacct_tsk_free(p);
1515         module_put(task_thread_info(p)->exec_domain->module);
1516 bad_fork_cleanup_count:
1517         atomic_dec(&p->cred->user->processes);
1518         exit_creds(p);
1519 bad_fork_free:
1520         free_task(p);
1521 fork_out:
1522         return ERR_PTR(retval);
1523 }
1524
1525 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1526 {
1527         memset(regs, 0, sizeof(struct pt_regs));
1528         return regs;
1529 }
1530
1531 static inline void init_idle_pids(struct pid_link *links)
1532 {
1533         enum pid_type type;
1534
1535         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1536                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1537                 links[type].pid = &init_struct_pid;
1538         }
1539 }
1540
1541 struct task_struct * __cpuinit fork_idle(int cpu)
1542 {
1543         struct task_struct *task;
1544         struct pt_regs regs;
1545
1546         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1547                             &init_struct_pid, 0);
1548         if (!IS_ERR(task)) {
1549                 init_idle_pids(task->pids);
1550                 init_idle(task, cpu);
1551         }
1552
1553         return task;
1554 }
1555
1556 /*
1557  *  Ok, this is the main fork-routine.
1558  *
1559  * It copies the process, and if successful kick-starts
1560  * it and waits for it to finish using the VM if required.
1561  */
1562 long do_fork(unsigned long clone_flags,
1563               unsigned long stack_start,
1564               struct pt_regs *regs,
1565               unsigned long stack_size,
1566               int __user *parent_tidptr,
1567               int __user *child_tidptr)
1568 {
1569         struct task_struct *p;
1570         int trace = 0;
1571         long nr;
1572
1573         /*
1574          * Do some preliminary argument and permissions checking before we
1575          * actually start allocating stuff
1576          */
1577         if (clone_flags & CLONE_NEWUSER) {
1578                 if (clone_flags & CLONE_THREAD)
1579                         return -EINVAL;
1580                 /* hopefully this check will go away when userns support is
1581                  * complete
1582                  */
1583                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1584                                 !capable(CAP_SETGID))
1585                         return -EPERM;
1586         }
1587
1588         /*
1589          * Determine whether and which event to report to ptracer.  When
1590          * called from kernel_thread or CLONE_UNTRACED is explicitly
1591          * requested, no event is reported; otherwise, report if the event
1592          * for the type of forking is enabled.
1593          */
1594         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1595                 if (clone_flags & CLONE_VFORK)
1596                         trace = PTRACE_EVENT_VFORK;
1597                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1598                         trace = PTRACE_EVENT_CLONE;
1599                 else
1600                         trace = PTRACE_EVENT_FORK;
1601
1602                 if (likely(!ptrace_event_enabled(current, trace)))
1603                         trace = 0;
1604         }
1605
1606         p = copy_process(clone_flags, stack_start, regs, stack_size,
1607                          child_tidptr, NULL, trace);
1608         /*
1609          * Do this prior waking up the new thread - the thread pointer
1610          * might get invalid after that point, if the thread exits quickly.
1611          */
1612         if (!IS_ERR(p)) {
1613                 struct completion vfork;
1614
1615                 trace_sched_process_fork(current, p);
1616
1617                 nr = task_pid_vnr(p);
1618
1619                 if (clone_flags & CLONE_PARENT_SETTID)
1620                         put_user(nr, parent_tidptr);
1621
1622                 if (clone_flags & CLONE_VFORK) {
1623                         p->vfork_done = &vfork;
1624                         init_completion(&vfork);
1625                         get_task_struct(p);
1626                 }
1627
1628                 wake_up_new_task(p);
1629
1630                 /* forking complete and child started to run, tell ptracer */
1631                 if (unlikely(trace))
1632                         ptrace_event(trace, nr);
1633
1634                 if (clone_flags & CLONE_VFORK) {
1635                         if (!wait_for_vfork_done(p, &vfork))
1636                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1637                 }
1638         } else {
1639                 nr = PTR_ERR(p);
1640         }
1641         return nr;
1642 }
1643
1644 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1645 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1646 #endif
1647
1648 static void sighand_ctor(void *data)
1649 {
1650         struct sighand_struct *sighand = data;
1651
1652         spin_lock_init(&sighand->siglock);
1653         init_waitqueue_head(&sighand->signalfd_wqh);
1654 }
1655
1656 void __init proc_caches_init(void)
1657 {
1658         sighand_cachep = kmem_cache_create("sighand_cache",
1659                         sizeof(struct sighand_struct), 0,
1660                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1661                         SLAB_NOTRACK, sighand_ctor);
1662         signal_cachep = kmem_cache_create("signal_cache",
1663                         sizeof(struct signal_struct), 0,
1664                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1665         files_cachep = kmem_cache_create("files_cache",
1666                         sizeof(struct files_struct), 0,
1667                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1668         fs_cachep = kmem_cache_create("fs_cache",
1669                         sizeof(struct fs_struct), 0,
1670                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1671         /*
1672          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1673          * whole struct cpumask for the OFFSTACK case. We could change
1674          * this to *only* allocate as much of it as required by the
1675          * maximum number of CPU's we can ever have.  The cpumask_allocation
1676          * is at the end of the structure, exactly for that reason.
1677          */
1678         mm_cachep = kmem_cache_create("mm_struct",
1679                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1680                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1681         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1682         mmap_init();
1683         nsproxy_cache_init();
1684 }
1685
1686 /*
1687  * Check constraints on flags passed to the unshare system call.
1688  */
1689 static int check_unshare_flags(unsigned long unshare_flags)
1690 {
1691         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1692                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1693                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1694                 return -EINVAL;
1695         /*
1696          * Not implemented, but pretend it works if there is nothing to
1697          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1698          * needs to unshare vm.
1699          */
1700         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1701                 /* FIXME: get_task_mm() increments ->mm_users */
1702                 if (atomic_read(&current->mm->mm_users) > 1)
1703                         return -EINVAL;
1704         }
1705
1706         return 0;
1707 }
1708
1709 /*
1710  * Unshare the filesystem structure if it is being shared
1711  */
1712 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1713 {
1714         struct fs_struct *fs = current->fs;
1715
1716         if (!(unshare_flags & CLONE_FS) || !fs)
1717                 return 0;
1718
1719         /* don't need lock here; in the worst case we'll do useless copy */
1720         if (fs->users == 1)
1721                 return 0;
1722
1723         *new_fsp = copy_fs_struct(fs);
1724         if (!*new_fsp)
1725                 return -ENOMEM;
1726
1727         return 0;
1728 }
1729
1730 /*
1731  * Unshare file descriptor table if it is being shared
1732  */
1733 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1734 {
1735         struct files_struct *fd = current->files;
1736         int error = 0;
1737
1738         if ((unshare_flags & CLONE_FILES) &&
1739             (fd && atomic_read(&fd->count) > 1)) {
1740                 *new_fdp = dup_fd(fd, &error);
1741                 if (!*new_fdp)
1742                         return error;
1743         }
1744
1745         return 0;
1746 }
1747
1748 /*
1749  * unshare allows a process to 'unshare' part of the process
1750  * context which was originally shared using clone.  copy_*
1751  * functions used by do_fork() cannot be used here directly
1752  * because they modify an inactive task_struct that is being
1753  * constructed. Here we are modifying the current, active,
1754  * task_struct.
1755  */
1756 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1757 {
1758         struct fs_struct *fs, *new_fs = NULL;
1759         struct files_struct *fd, *new_fd = NULL;
1760         struct nsproxy *new_nsproxy = NULL;
1761         int do_sysvsem = 0;
1762         int err;
1763
1764         err = check_unshare_flags(unshare_flags);
1765         if (err)
1766                 goto bad_unshare_out;
1767
1768         /*
1769          * If unsharing namespace, must also unshare filesystem information.
1770          */
1771         if (unshare_flags & CLONE_NEWNS)
1772                 unshare_flags |= CLONE_FS;
1773         /*
1774          * CLONE_NEWIPC must also detach from the undolist: after switching
1775          * to a new ipc namespace, the semaphore arrays from the old
1776          * namespace are unreachable.
1777          */
1778         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1779                 do_sysvsem = 1;
1780         err = unshare_fs(unshare_flags, &new_fs);
1781         if (err)
1782                 goto bad_unshare_out;
1783         err = unshare_fd(unshare_flags, &new_fd);
1784         if (err)
1785                 goto bad_unshare_cleanup_fs;
1786         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1787         if (err)
1788                 goto bad_unshare_cleanup_fd;
1789
1790         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1791                 if (do_sysvsem) {
1792                         /*
1793                          * CLONE_SYSVSEM is equivalent to sys_exit().
1794                          */
1795                         exit_sem(current);
1796                 }
1797
1798                 if (new_nsproxy) {
1799                         switch_task_namespaces(current, new_nsproxy);
1800                         new_nsproxy = NULL;
1801                 }
1802
1803                 task_lock(current);
1804
1805                 if (new_fs) {
1806                         fs = current->fs;
1807                         spin_lock(&fs->lock);
1808                         current->fs = new_fs;
1809                         if (--fs->users)
1810                                 new_fs = NULL;
1811                         else
1812                                 new_fs = fs;
1813                         spin_unlock(&fs->lock);
1814                 }
1815
1816                 if (new_fd) {
1817                         fd = current->files;
1818                         current->files = new_fd;
1819                         new_fd = fd;
1820                 }
1821
1822                 task_unlock(current);
1823         }
1824
1825         if (new_nsproxy)
1826                 put_nsproxy(new_nsproxy);
1827
1828 bad_unshare_cleanup_fd:
1829         if (new_fd)
1830                 put_files_struct(new_fd);
1831
1832 bad_unshare_cleanup_fs:
1833         if (new_fs)
1834                 free_fs_struct(new_fs);
1835
1836 bad_unshare_out:
1837         return err;
1838 }
1839
1840 /*
1841  *      Helper to unshare the files of the current task.
1842  *      We don't want to expose copy_files internals to
1843  *      the exec layer of the kernel.
1844  */
1845
1846 int unshare_files(struct files_struct **displaced)
1847 {
1848         struct task_struct *task = current;
1849         struct files_struct *copy = NULL;
1850         int error;
1851
1852         error = unshare_fd(CLONE_FILES, &copy);
1853         if (error || !copy) {
1854                 *displaced = NULL;
1855                 return error;
1856         }
1857         *displaced = task->files;
1858         task_lock(task);
1859         task->files = copy;
1860         task_unlock(task);
1861         return 0;
1862 }