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