Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm...
[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         uprobe_start_dup_mmap();
356         down_write(&oldmm->mmap_sem);
357         flush_cache_dup_mm(oldmm);
358         uprobe_dup_mmap(oldmm, mm);
359         /*
360          * Not linked in yet - no deadlock potential:
361          */
362         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
363
364         mm->locked_vm = 0;
365         mm->mmap = NULL;
366         mm->mmap_cache = NULL;
367         mm->free_area_cache = oldmm->mmap_base;
368         mm->cached_hole_size = ~0UL;
369         mm->map_count = 0;
370         cpumask_clear(mm_cpumask(mm));
371         mm->mm_rb = RB_ROOT;
372         rb_link = &mm->mm_rb.rb_node;
373         rb_parent = NULL;
374         pprev = &mm->mmap;
375         retval = ksm_fork(mm, oldmm);
376         if (retval)
377                 goto out;
378         retval = khugepaged_fork(mm, oldmm);
379         if (retval)
380                 goto out;
381
382         prev = NULL;
383         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
384                 struct file *file;
385
386                 if (mpnt->vm_flags & VM_DONTCOPY) {
387                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
388                                                         -vma_pages(mpnt));
389                         continue;
390                 }
391                 charge = 0;
392                 if (mpnt->vm_flags & VM_ACCOUNT) {
393                         unsigned long len = vma_pages(mpnt);
394
395                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
396                                 goto fail_nomem;
397                         charge = len;
398                 }
399                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
400                 if (!tmp)
401                         goto fail_nomem;
402                 *tmp = *mpnt;
403                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
404                 pol = mpol_dup(vma_policy(mpnt));
405                 retval = PTR_ERR(pol);
406                 if (IS_ERR(pol))
407                         goto fail_nomem_policy;
408                 vma_set_policy(tmp, pol);
409                 tmp->vm_mm = mm;
410                 if (anon_vma_fork(tmp, mpnt))
411                         goto fail_nomem_anon_vma_fork;
412                 tmp->vm_flags &= ~VM_LOCKED;
413                 tmp->vm_next = tmp->vm_prev = NULL;
414                 file = tmp->vm_file;
415                 if (file) {
416                         struct inode *inode = file->f_path.dentry->d_inode;
417                         struct address_space *mapping = file->f_mapping;
418
419                         get_file(file);
420                         if (tmp->vm_flags & VM_DENYWRITE)
421                                 atomic_dec(&inode->i_writecount);
422                         mutex_lock(&mapping->i_mmap_mutex);
423                         if (tmp->vm_flags & VM_SHARED)
424                                 mapping->i_mmap_writable++;
425                         flush_dcache_mmap_lock(mapping);
426                         /* insert tmp into the share list, just after mpnt */
427                         if (unlikely(tmp->vm_flags & VM_NONLINEAR))
428                                 vma_nonlinear_insert(tmp,
429                                                 &mapping->i_mmap_nonlinear);
430                         else
431                                 vma_interval_tree_insert_after(tmp, mpnt,
432                                                         &mapping->i_mmap);
433                         flush_dcache_mmap_unlock(mapping);
434                         mutex_unlock(&mapping->i_mmap_mutex);
435                 }
436
437                 /*
438                  * Clear hugetlb-related page reserves for children. This only
439                  * affects MAP_PRIVATE mappings. Faults generated by the child
440                  * are not guaranteed to succeed, even if read-only
441                  */
442                 if (is_vm_hugetlb_page(tmp))
443                         reset_vma_resv_huge_pages(tmp);
444
445                 /*
446                  * Link in the new vma and copy the page table entries.
447                  */
448                 *pprev = tmp;
449                 pprev = &tmp->vm_next;
450                 tmp->vm_prev = prev;
451                 prev = tmp;
452
453                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
454                 rb_link = &tmp->vm_rb.rb_right;
455                 rb_parent = &tmp->vm_rb;
456
457                 mm->map_count++;
458                 retval = copy_page_range(mm, oldmm, mpnt);
459
460                 if (tmp->vm_ops && tmp->vm_ops->open)
461                         tmp->vm_ops->open(tmp);
462
463                 if (retval)
464                         goto out;
465         }
466         /* a new mm has just been created */
467         arch_dup_mmap(oldmm, mm);
468         retval = 0;
469 out:
470         up_write(&mm->mmap_sem);
471         flush_tlb_mm(oldmm);
472         up_write(&oldmm->mmap_sem);
473         uprobe_end_dup_mmap();
474         return retval;
475 fail_nomem_anon_vma_fork:
476         mpol_put(pol);
477 fail_nomem_policy:
478         kmem_cache_free(vm_area_cachep, tmp);
479 fail_nomem:
480         retval = -ENOMEM;
481         vm_unacct_memory(charge);
482         goto out;
483 }
484
485 static inline int mm_alloc_pgd(struct mm_struct *mm)
486 {
487         mm->pgd = pgd_alloc(mm);
488         if (unlikely(!mm->pgd))
489                 return -ENOMEM;
490         return 0;
491 }
492
493 static inline void mm_free_pgd(struct mm_struct *mm)
494 {
495         pgd_free(mm, mm->pgd);
496 }
497 #else
498 #define dup_mmap(mm, oldmm)     (0)
499 #define mm_alloc_pgd(mm)        (0)
500 #define mm_free_pgd(mm)
501 #endif /* CONFIG_MMU */
502
503 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
504
505 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
506 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
507
508 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
509
510 static int __init coredump_filter_setup(char *s)
511 {
512         default_dump_filter =
513                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
514                 MMF_DUMP_FILTER_MASK;
515         return 1;
516 }
517
518 __setup("coredump_filter=", coredump_filter_setup);
519
520 #include <linux/init_task.h>
521
522 static void mm_init_aio(struct mm_struct *mm)
523 {
524 #ifdef CONFIG_AIO
525         spin_lock_init(&mm->ioctx_lock);
526         INIT_HLIST_HEAD(&mm->ioctx_list);
527 #endif
528 }
529
530 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
531 {
532         atomic_set(&mm->mm_users, 1);
533         atomic_set(&mm->mm_count, 1);
534         init_rwsem(&mm->mmap_sem);
535         INIT_LIST_HEAD(&mm->mmlist);
536         mm->flags = (current->mm) ?
537                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
538         mm->core_state = NULL;
539         mm->nr_ptes = 0;
540         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
541         spin_lock_init(&mm->page_table_lock);
542         mm->free_area_cache = TASK_UNMAPPED_BASE;
543         mm->cached_hole_size = ~0UL;
544         mm_init_aio(mm);
545         mm_init_owner(mm, p);
546
547         if (likely(!mm_alloc_pgd(mm))) {
548                 mm->def_flags = 0;
549                 mmu_notifier_mm_init(mm);
550                 return mm;
551         }
552
553         free_mm(mm);
554         return NULL;
555 }
556
557 static void check_mm(struct mm_struct *mm)
558 {
559         int i;
560
561         for (i = 0; i < NR_MM_COUNTERS; i++) {
562                 long x = atomic_long_read(&mm->rss_stat.count[i]);
563
564                 if (unlikely(x))
565                         printk(KERN_ALERT "BUG: Bad rss-counter state "
566                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
567         }
568
569 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
570         VM_BUG_ON(mm->pmd_huge_pte);
571 #endif
572 }
573
574 /*
575  * Allocate and initialize an mm_struct.
576  */
577 struct mm_struct *mm_alloc(void)
578 {
579         struct mm_struct *mm;
580
581         mm = allocate_mm();
582         if (!mm)
583                 return NULL;
584
585         memset(mm, 0, sizeof(*mm));
586         mm_init_cpumask(mm);
587         return mm_init(mm, current);
588 }
589
590 /*
591  * Called when the last reference to the mm
592  * is dropped: either by a lazy thread or by
593  * mmput. Free the page directory and the mm.
594  */
595 void __mmdrop(struct mm_struct *mm)
596 {
597         BUG_ON(mm == &init_mm);
598         mm_free_pgd(mm);
599         destroy_context(mm);
600         mmu_notifier_mm_destroy(mm);
601         check_mm(mm);
602         free_mm(mm);
603 }
604 EXPORT_SYMBOL_GPL(__mmdrop);
605
606 /*
607  * Decrement the use count and release all resources for an mm.
608  */
609 void mmput(struct mm_struct *mm)
610 {
611         might_sleep();
612
613         if (atomic_dec_and_test(&mm->mm_users)) {
614                 uprobe_clear_state(mm);
615                 exit_aio(mm);
616                 ksm_exit(mm);
617                 khugepaged_exit(mm); /* must run before exit_mmap */
618                 exit_mmap(mm);
619                 set_mm_exe_file(mm, NULL);
620                 if (!list_empty(&mm->mmlist)) {
621                         spin_lock(&mmlist_lock);
622                         list_del(&mm->mmlist);
623                         spin_unlock(&mmlist_lock);
624                 }
625                 if (mm->binfmt)
626                         module_put(mm->binfmt->module);
627                 mmdrop(mm);
628         }
629 }
630 EXPORT_SYMBOL_GPL(mmput);
631
632 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
633 {
634         if (new_exe_file)
635                 get_file(new_exe_file);
636         if (mm->exe_file)
637                 fput(mm->exe_file);
638         mm->exe_file = new_exe_file;
639 }
640
641 struct file *get_mm_exe_file(struct mm_struct *mm)
642 {
643         struct file *exe_file;
644
645         /* We need mmap_sem to protect against races with removal of exe_file */
646         down_read(&mm->mmap_sem);
647         exe_file = mm->exe_file;
648         if (exe_file)
649                 get_file(exe_file);
650         up_read(&mm->mmap_sem);
651         return exe_file;
652 }
653
654 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
655 {
656         /* It's safe to write the exe_file pointer without exe_file_lock because
657          * this is called during fork when the task is not yet in /proc */
658         newmm->exe_file = get_mm_exe_file(oldmm);
659 }
660
661 /**
662  * get_task_mm - acquire a reference to the task's mm
663  *
664  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
665  * this kernel workthread has transiently adopted a user mm with use_mm,
666  * to do its AIO) is not set and if so returns a reference to it, after
667  * bumping up the use count.  User must release the mm via mmput()
668  * after use.  Typically used by /proc and ptrace.
669  */
670 struct mm_struct *get_task_mm(struct task_struct *task)
671 {
672         struct mm_struct *mm;
673
674         task_lock(task);
675         mm = task->mm;
676         if (mm) {
677                 if (task->flags & PF_KTHREAD)
678                         mm = NULL;
679                 else
680                         atomic_inc(&mm->mm_users);
681         }
682         task_unlock(task);
683         return mm;
684 }
685 EXPORT_SYMBOL_GPL(get_task_mm);
686
687 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
688 {
689         struct mm_struct *mm;
690         int err;
691
692         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
693         if (err)
694                 return ERR_PTR(err);
695
696         mm = get_task_mm(task);
697         if (mm && mm != current->mm &&
698                         !ptrace_may_access(task, mode)) {
699                 mmput(mm);
700                 mm = ERR_PTR(-EACCES);
701         }
702         mutex_unlock(&task->signal->cred_guard_mutex);
703
704         return mm;
705 }
706
707 static void complete_vfork_done(struct task_struct *tsk)
708 {
709         struct completion *vfork;
710
711         task_lock(tsk);
712         vfork = tsk->vfork_done;
713         if (likely(vfork)) {
714                 tsk->vfork_done = NULL;
715                 complete(vfork);
716         }
717         task_unlock(tsk);
718 }
719
720 static int wait_for_vfork_done(struct task_struct *child,
721                                 struct completion *vfork)
722 {
723         int killed;
724
725         freezer_do_not_count();
726         killed = wait_for_completion_killable(vfork);
727         freezer_count();
728
729         if (killed) {
730                 task_lock(child);
731                 child->vfork_done = NULL;
732                 task_unlock(child);
733         }
734
735         put_task_struct(child);
736         return killed;
737 }
738
739 /* Please note the differences between mmput and mm_release.
740  * mmput is called whenever we stop holding onto a mm_struct,
741  * error success whatever.
742  *
743  * mm_release is called after a mm_struct has been removed
744  * from the current process.
745  *
746  * This difference is important for error handling, when we
747  * only half set up a mm_struct for a new process and need to restore
748  * the old one.  Because we mmput the new mm_struct before
749  * restoring the old one. . .
750  * Eric Biederman 10 January 1998
751  */
752 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
753 {
754         /* Get rid of any futexes when releasing the mm */
755 #ifdef CONFIG_FUTEX
756         if (unlikely(tsk->robust_list)) {
757                 exit_robust_list(tsk);
758                 tsk->robust_list = NULL;
759         }
760 #ifdef CONFIG_COMPAT
761         if (unlikely(tsk->compat_robust_list)) {
762                 compat_exit_robust_list(tsk);
763                 tsk->compat_robust_list = NULL;
764         }
765 #endif
766         if (unlikely(!list_empty(&tsk->pi_state_list)))
767                 exit_pi_state_list(tsk);
768 #endif
769
770         uprobe_free_utask(tsk);
771
772         /* Get rid of any cached register state */
773         deactivate_mm(tsk, mm);
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          * All done, finally we can wake up parent and return this mm to him.
798          * Also kthread_stop() uses this completion for synchronization.
799          */
800         if (tsk->vfork_done)
801                 complete_vfork_done(tsk);
802 }
803
804 /*
805  * Allocate a new mm structure and copy contents from the
806  * mm structure of the passed in task structure.
807  */
808 struct mm_struct *dup_mm(struct task_struct *tsk)
809 {
810         struct mm_struct *mm, *oldmm = current->mm;
811         int err;
812
813         if (!oldmm)
814                 return NULL;
815
816         mm = allocate_mm();
817         if (!mm)
818                 goto fail_nomem;
819
820         memcpy(mm, oldmm, sizeof(*mm));
821         mm_init_cpumask(mm);
822
823 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
824         mm->pmd_huge_pte = NULL;
825 #endif
826 #ifdef CONFIG_NUMA_BALANCING
827         mm->first_nid = NUMA_PTE_SCAN_INIT;
828 #endif
829         if (!mm_init(mm, tsk))
830                 goto fail_nomem;
831
832         if (init_new_context(tsk, mm))
833                 goto fail_nocontext;
834
835         dup_mm_exe_file(oldmm, mm);
836
837         err = dup_mmap(mm, oldmm);
838         if (err)
839                 goto free_pt;
840
841         mm->hiwater_rss = get_mm_rss(mm);
842         mm->hiwater_vm = mm->total_vm;
843
844         if (mm->binfmt && !try_module_get(mm->binfmt->module))
845                 goto free_pt;
846
847         return mm;
848
849 free_pt:
850         /* don't put binfmt in mmput, we haven't got module yet */
851         mm->binfmt = NULL;
852         mmput(mm);
853
854 fail_nomem:
855         return NULL;
856
857 fail_nocontext:
858         /*
859          * If init_new_context() failed, we cannot use mmput() to free the mm
860          * because it calls destroy_context()
861          */
862         mm_free_pgd(mm);
863         free_mm(mm);
864         return NULL;
865 }
866
867 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
868 {
869         struct mm_struct *mm, *oldmm;
870         int retval;
871
872         tsk->min_flt = tsk->maj_flt = 0;
873         tsk->nvcsw = tsk->nivcsw = 0;
874 #ifdef CONFIG_DETECT_HUNG_TASK
875         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
876 #endif
877
878         tsk->mm = NULL;
879         tsk->active_mm = NULL;
880
881         /*
882          * Are we cloning a kernel thread?
883          *
884          * We need to steal a active VM for that..
885          */
886         oldmm = current->mm;
887         if (!oldmm)
888                 return 0;
889
890         if (clone_flags & CLONE_VM) {
891                 atomic_inc(&oldmm->mm_users);
892                 mm = oldmm;
893                 goto good_mm;
894         }
895
896         retval = -ENOMEM;
897         mm = dup_mm(tsk);
898         if (!mm)
899                 goto fail_nomem;
900
901 good_mm:
902         tsk->mm = mm;
903         tsk->active_mm = mm;
904         return 0;
905
906 fail_nomem:
907         return retval;
908 }
909
910 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
911 {
912         struct fs_struct *fs = current->fs;
913         if (clone_flags & CLONE_FS) {
914                 /* tsk->fs is already what we want */
915                 spin_lock(&fs->lock);
916                 if (fs->in_exec) {
917                         spin_unlock(&fs->lock);
918                         return -EAGAIN;
919                 }
920                 fs->users++;
921                 spin_unlock(&fs->lock);
922                 return 0;
923         }
924         tsk->fs = copy_fs_struct(fs);
925         if (!tsk->fs)
926                 return -ENOMEM;
927         return 0;
928 }
929
930 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
931 {
932         struct files_struct *oldf, *newf;
933         int error = 0;
934
935         /*
936          * A background process may not have any files ...
937          */
938         oldf = current->files;
939         if (!oldf)
940                 goto out;
941
942         if (clone_flags & CLONE_FILES) {
943                 atomic_inc(&oldf->count);
944                 goto out;
945         }
946
947         newf = dup_fd(oldf, &error);
948         if (!newf)
949                 goto out;
950
951         tsk->files = newf;
952         error = 0;
953 out:
954         return error;
955 }
956
957 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
958 {
959 #ifdef CONFIG_BLOCK
960         struct io_context *ioc = current->io_context;
961         struct io_context *new_ioc;
962
963         if (!ioc)
964                 return 0;
965         /*
966          * Share io context with parent, if CLONE_IO is set
967          */
968         if (clone_flags & CLONE_IO) {
969                 ioc_task_link(ioc);
970                 tsk->io_context = ioc;
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         sig->curr_target = tsk;
1048         init_sigpending(&sig->shared_pending);
1049         INIT_LIST_HEAD(&sig->posix_timers);
1050
1051         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1052         sig->real_timer.function = it_real_fn;
1053
1054         task_lock(current->group_leader);
1055         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1056         task_unlock(current->group_leader);
1057
1058         posix_cpu_timers_init_group(sig);
1059
1060         tty_audit_fork(sig);
1061         sched_autogroup_fork(sig);
1062
1063 #ifdef CONFIG_CGROUPS
1064         init_rwsem(&sig->group_rwsem);
1065 #endif
1066
1067         sig->oom_score_adj = current->signal->oom_score_adj;
1068         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1069
1070         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1071                                    current->signal->is_child_subreaper;
1072
1073         mutex_init(&sig->cred_guard_mutex);
1074
1075         return 0;
1076 }
1077
1078 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1079 {
1080         unsigned long new_flags = p->flags;
1081
1082         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1083         new_flags |= PF_FORKNOEXEC;
1084         p->flags = new_flags;
1085 }
1086
1087 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1088 {
1089         current->clear_child_tid = tidptr;
1090
1091         return task_pid_vnr(current);
1092 }
1093
1094 static void rt_mutex_init_task(struct task_struct *p)
1095 {
1096         raw_spin_lock_init(&p->pi_lock);
1097 #ifdef CONFIG_RT_MUTEXES
1098         plist_head_init(&p->pi_waiters);
1099         p->pi_blocked_on = NULL;
1100 #endif
1101 }
1102
1103 #ifdef CONFIG_MM_OWNER
1104 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1105 {
1106         mm->owner = p;
1107 }
1108 #endif /* CONFIG_MM_OWNER */
1109
1110 /*
1111  * Initialize POSIX timer handling for a single task.
1112  */
1113 static void posix_cpu_timers_init(struct task_struct *tsk)
1114 {
1115         tsk->cputime_expires.prof_exp = 0;
1116         tsk->cputime_expires.virt_exp = 0;
1117         tsk->cputime_expires.sched_exp = 0;
1118         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1119         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1120         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1121 }
1122
1123 /*
1124  * This creates a new process as a copy of the old one,
1125  * but does not actually start it yet.
1126  *
1127  * It copies the registers, and all the appropriate
1128  * parts of the process environment (as per the clone
1129  * flags). The actual kick-off is left to the caller.
1130  */
1131 static struct task_struct *copy_process(unsigned long clone_flags,
1132                                         unsigned long stack_start,
1133                                         unsigned long stack_size,
1134                                         int __user *child_tidptr,
1135                                         struct pid *pid,
1136                                         int trace)
1137 {
1138         int retval;
1139         struct task_struct *p;
1140
1141         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1142                 return ERR_PTR(-EINVAL);
1143
1144         /*
1145          * Thread groups must share signals as well, and detached threads
1146          * can only be started up within the thread group.
1147          */
1148         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1149                 return ERR_PTR(-EINVAL);
1150
1151         /*
1152          * Shared signal handlers imply shared VM. By way of the above,
1153          * thread groups also imply shared VM. Blocking this case allows
1154          * for various simplifications in other code.
1155          */
1156         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1157                 return ERR_PTR(-EINVAL);
1158
1159         /*
1160          * Siblings of global init remain as zombies on exit since they are
1161          * not reaped by their parent (swapper). To solve this and to avoid
1162          * multi-rooted process trees, prevent global and container-inits
1163          * from creating siblings.
1164          */
1165         if ((clone_flags & CLONE_PARENT) &&
1166                                 current->signal->flags & SIGNAL_UNKILLABLE)
1167                 return ERR_PTR(-EINVAL);
1168
1169         retval = security_task_create(clone_flags);
1170         if (retval)
1171                 goto fork_out;
1172
1173         retval = -ENOMEM;
1174         p = dup_task_struct(current);
1175         if (!p)
1176                 goto fork_out;
1177
1178         ftrace_graph_init_task(p);
1179         get_seccomp_filter(p);
1180
1181         rt_mutex_init_task(p);
1182
1183 #ifdef CONFIG_PROVE_LOCKING
1184         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1185         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1186 #endif
1187         retval = -EAGAIN;
1188         if (atomic_read(&p->real_cred->user->processes) >=
1189                         task_rlimit(p, RLIMIT_NPROC)) {
1190                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1191                     p->real_cred->user != INIT_USER)
1192                         goto bad_fork_free;
1193         }
1194         current->flags &= ~PF_NPROC_EXCEEDED;
1195
1196         retval = copy_creds(p, clone_flags);
1197         if (retval < 0)
1198                 goto bad_fork_free;
1199
1200         /*
1201          * If multiple threads are within copy_process(), then this check
1202          * triggers too late. This doesn't hurt, the check is only there
1203          * to stop root fork bombs.
1204          */
1205         retval = -EAGAIN;
1206         if (nr_threads >= max_threads)
1207                 goto bad_fork_cleanup_count;
1208
1209         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1210                 goto bad_fork_cleanup_count;
1211
1212         p->did_exec = 0;
1213         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1214         copy_flags(clone_flags, p);
1215         INIT_LIST_HEAD(&p->children);
1216         INIT_LIST_HEAD(&p->sibling);
1217         rcu_copy_process(p);
1218         p->vfork_done = NULL;
1219         spin_lock_init(&p->alloc_lock);
1220
1221         init_sigpending(&p->pending);
1222
1223         p->utime = p->stime = p->gtime = 0;
1224         p->utimescaled = p->stimescaled = 0;
1225 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1226         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1227 #endif
1228 #if defined(SPLIT_RSS_COUNTING)
1229         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1230 #endif
1231
1232         p->default_timer_slack_ns = current->timer_slack_ns;
1233
1234         task_io_accounting_init(&p->ioac);
1235         acct_clear_integrals(p);
1236
1237         posix_cpu_timers_init(p);
1238
1239         do_posix_clock_monotonic_gettime(&p->start_time);
1240         p->real_start_time = p->start_time;
1241         monotonic_to_bootbased(&p->real_start_time);
1242         p->io_context = NULL;
1243         p->audit_context = NULL;
1244         if (clone_flags & CLONE_THREAD)
1245                 threadgroup_change_begin(current);
1246         cgroup_fork(p);
1247 #ifdef CONFIG_NUMA
1248         p->mempolicy = mpol_dup(p->mempolicy);
1249         if (IS_ERR(p->mempolicy)) {
1250                 retval = PTR_ERR(p->mempolicy);
1251                 p->mempolicy = NULL;
1252                 goto bad_fork_cleanup_cgroup;
1253         }
1254         mpol_fix_fork_child_flag(p);
1255 #endif
1256 #ifdef CONFIG_CPUSETS
1257         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1258         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1259         seqcount_init(&p->mems_allowed_seq);
1260 #endif
1261 #ifdef CONFIG_TRACE_IRQFLAGS
1262         p->irq_events = 0;
1263         p->hardirqs_enabled = 0;
1264         p->hardirq_enable_ip = 0;
1265         p->hardirq_enable_event = 0;
1266         p->hardirq_disable_ip = _THIS_IP_;
1267         p->hardirq_disable_event = 0;
1268         p->softirqs_enabled = 1;
1269         p->softirq_enable_ip = _THIS_IP_;
1270         p->softirq_enable_event = 0;
1271         p->softirq_disable_ip = 0;
1272         p->softirq_disable_event = 0;
1273         p->hardirq_context = 0;
1274         p->softirq_context = 0;
1275 #endif
1276 #ifdef CONFIG_LOCKDEP
1277         p->lockdep_depth = 0; /* no locks held yet */
1278         p->curr_chain_key = 0;
1279         p->lockdep_recursion = 0;
1280 #endif
1281
1282 #ifdef CONFIG_DEBUG_MUTEXES
1283         p->blocked_on = NULL; /* not blocked yet */
1284 #endif
1285 #ifdef CONFIG_MEMCG
1286         p->memcg_batch.do_batch = 0;
1287         p->memcg_batch.memcg = NULL;
1288 #endif
1289
1290         /* Perform scheduler related setup. Assign this task to a CPU. */
1291         sched_fork(p);
1292
1293         retval = perf_event_init_task(p);
1294         if (retval)
1295                 goto bad_fork_cleanup_policy;
1296         retval = audit_alloc(p);
1297         if (retval)
1298                 goto bad_fork_cleanup_policy;
1299         /* copy all the process information */
1300         retval = copy_semundo(clone_flags, p);
1301         if (retval)
1302                 goto bad_fork_cleanup_audit;
1303         retval = copy_files(clone_flags, p);
1304         if (retval)
1305                 goto bad_fork_cleanup_semundo;
1306         retval = copy_fs(clone_flags, p);
1307         if (retval)
1308                 goto bad_fork_cleanup_files;
1309         retval = copy_sighand(clone_flags, p);
1310         if (retval)
1311                 goto bad_fork_cleanup_fs;
1312         retval = copy_signal(clone_flags, p);
1313         if (retval)
1314                 goto bad_fork_cleanup_sighand;
1315         retval = copy_mm(clone_flags, p);
1316         if (retval)
1317                 goto bad_fork_cleanup_signal;
1318         retval = copy_namespaces(clone_flags, p);
1319         if (retval)
1320                 goto bad_fork_cleanup_mm;
1321         retval = copy_io(clone_flags, p);
1322         if (retval)
1323                 goto bad_fork_cleanup_namespaces;
1324         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1325         if (retval)
1326                 goto bad_fork_cleanup_io;
1327
1328         if (pid != &init_struct_pid) {
1329                 retval = -ENOMEM;
1330                 pid = alloc_pid(p->nsproxy->pid_ns);
1331                 if (!pid)
1332                         goto bad_fork_cleanup_io;
1333         }
1334
1335         p->pid = pid_nr(pid);
1336         p->tgid = p->pid;
1337         if (clone_flags & CLONE_THREAD)
1338                 p->tgid = current->tgid;
1339
1340         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1341         /*
1342          * Clear TID on mm_release()?
1343          */
1344         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1345 #ifdef CONFIG_BLOCK
1346         p->plug = NULL;
1347 #endif
1348 #ifdef CONFIG_FUTEX
1349         p->robust_list = NULL;
1350 #ifdef CONFIG_COMPAT
1351         p->compat_robust_list = NULL;
1352 #endif
1353         INIT_LIST_HEAD(&p->pi_state_list);
1354         p->pi_state_cache = NULL;
1355 #endif
1356         uprobe_copy_process(p);
1357         /*
1358          * sigaltstack should be cleared when sharing the same VM
1359          */
1360         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1361                 p->sas_ss_sp = p->sas_ss_size = 0;
1362
1363         /*
1364          * Syscall tracing and stepping should be turned off in the
1365          * child regardless of CLONE_PTRACE.
1366          */
1367         user_disable_single_step(p);
1368         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1369 #ifdef TIF_SYSCALL_EMU
1370         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1371 #endif
1372         clear_all_latency_tracing(p);
1373
1374         /* ok, now we should be set up.. */
1375         if (clone_flags & CLONE_THREAD)
1376                 p->exit_signal = -1;
1377         else if (clone_flags & CLONE_PARENT)
1378                 p->exit_signal = current->group_leader->exit_signal;
1379         else
1380                 p->exit_signal = (clone_flags & CSIGNAL);
1381
1382         p->pdeath_signal = 0;
1383         p->exit_state = 0;
1384
1385         p->nr_dirtied = 0;
1386         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1387         p->dirty_paused_when = 0;
1388
1389         /*
1390          * Ok, make it visible to the rest of the system.
1391          * We dont wake it up yet.
1392          */
1393         p->group_leader = p;
1394         INIT_LIST_HEAD(&p->thread_group);
1395         p->task_works = NULL;
1396
1397         /* Need tasklist lock for parent etc handling! */
1398         write_lock_irq(&tasklist_lock);
1399
1400         /* CLONE_PARENT re-uses the old parent */
1401         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1402                 p->real_parent = current->real_parent;
1403                 p->parent_exec_id = current->parent_exec_id;
1404         } else {
1405                 p->real_parent = current;
1406                 p->parent_exec_id = current->self_exec_id;
1407         }
1408
1409         spin_lock(&current->sighand->siglock);
1410
1411         /*
1412          * Process group and session signals need to be delivered to just the
1413          * parent before the fork or both the parent and the child after the
1414          * fork. Restart if a signal comes in before we add the new process to
1415          * it's process group.
1416          * A fatal signal pending means that current will exit, so the new
1417          * thread can't slip out of an OOM kill (or normal SIGKILL).
1418         */
1419         recalc_sigpending();
1420         if (signal_pending(current)) {
1421                 spin_unlock(&current->sighand->siglock);
1422                 write_unlock_irq(&tasklist_lock);
1423                 retval = -ERESTARTNOINTR;
1424                 goto bad_fork_free_pid;
1425         }
1426
1427         if (clone_flags & CLONE_THREAD) {
1428                 current->signal->nr_threads++;
1429                 atomic_inc(&current->signal->live);
1430                 atomic_inc(&current->signal->sigcnt);
1431                 p->group_leader = current->group_leader;
1432                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1433         }
1434
1435         if (likely(p->pid)) {
1436                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1437
1438                 if (thread_group_leader(p)) {
1439                         if (is_child_reaper(pid)) {
1440                                 ns_of_pid(pid)->child_reaper = p;
1441                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1442                         }
1443
1444                         p->signal->leader_pid = pid;
1445                         p->signal->tty = tty_kref_get(current->signal->tty);
1446                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1447                         attach_pid(p, PIDTYPE_SID, task_session(current));
1448                         list_add_tail(&p->sibling, &p->real_parent->children);
1449                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1450                         __this_cpu_inc(process_counts);
1451                 }
1452                 attach_pid(p, PIDTYPE_PID, pid);
1453                 nr_threads++;
1454         }
1455
1456         total_forks++;
1457         spin_unlock(&current->sighand->siglock);
1458         write_unlock_irq(&tasklist_lock);
1459         proc_fork_connector(p);
1460         cgroup_post_fork(p);
1461         if (clone_flags & CLONE_THREAD)
1462                 threadgroup_change_end(current);
1463         perf_event_fork(p);
1464
1465         trace_task_newtask(p, clone_flags);
1466
1467         return p;
1468
1469 bad_fork_free_pid:
1470         if (pid != &init_struct_pid)
1471                 free_pid(pid);
1472 bad_fork_cleanup_io:
1473         if (p->io_context)
1474                 exit_io_context(p);
1475 bad_fork_cleanup_namespaces:
1476         exit_task_namespaces(p);
1477 bad_fork_cleanup_mm:
1478         if (p->mm)
1479                 mmput(p->mm);
1480 bad_fork_cleanup_signal:
1481         if (!(clone_flags & CLONE_THREAD))
1482                 free_signal_struct(p->signal);
1483 bad_fork_cleanup_sighand:
1484         __cleanup_sighand(p->sighand);
1485 bad_fork_cleanup_fs:
1486         exit_fs(p); /* blocking */
1487 bad_fork_cleanup_files:
1488         exit_files(p); /* blocking */
1489 bad_fork_cleanup_semundo:
1490         exit_sem(p);
1491 bad_fork_cleanup_audit:
1492         audit_free(p);
1493 bad_fork_cleanup_policy:
1494         perf_event_free_task(p);
1495 #ifdef CONFIG_NUMA
1496         mpol_put(p->mempolicy);
1497 bad_fork_cleanup_cgroup:
1498 #endif
1499         if (clone_flags & CLONE_THREAD)
1500                 threadgroup_change_end(current);
1501         cgroup_exit(p, 0);
1502         delayacct_tsk_free(p);
1503         module_put(task_thread_info(p)->exec_domain->module);
1504 bad_fork_cleanup_count:
1505         atomic_dec(&p->cred->user->processes);
1506         exit_creds(p);
1507 bad_fork_free:
1508         free_task(p);
1509 fork_out:
1510         return ERR_PTR(retval);
1511 }
1512
1513 static inline void init_idle_pids(struct pid_link *links)
1514 {
1515         enum pid_type type;
1516
1517         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1518                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1519                 links[type].pid = &init_struct_pid;
1520         }
1521 }
1522
1523 struct task_struct * __cpuinit fork_idle(int cpu)
1524 {
1525         struct task_struct *task;
1526         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1527         if (!IS_ERR(task)) {
1528                 init_idle_pids(task->pids);
1529                 init_idle(task, cpu);
1530         }
1531
1532         return task;
1533 }
1534
1535 /*
1536  *  Ok, this is the main fork-routine.
1537  *
1538  * It copies the process, and if successful kick-starts
1539  * it and waits for it to finish using the VM if required.
1540  */
1541 long do_fork(unsigned long clone_flags,
1542               unsigned long stack_start,
1543               unsigned long stack_size,
1544               int __user *parent_tidptr,
1545               int __user *child_tidptr)
1546 {
1547         struct task_struct *p;
1548         int trace = 0;
1549         long nr;
1550
1551         /*
1552          * Do some preliminary argument and permissions checking before we
1553          * actually start allocating stuff
1554          */
1555         if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
1556                 if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
1557                         return -EINVAL;
1558         }
1559
1560         /*
1561          * Determine whether and which event to report to ptracer.  When
1562          * called from kernel_thread or CLONE_UNTRACED is explicitly
1563          * requested, no event is reported; otherwise, report if the event
1564          * for the type of forking is enabled.
1565          */
1566         if (!(clone_flags & CLONE_UNTRACED)) {
1567                 if (clone_flags & CLONE_VFORK)
1568                         trace = PTRACE_EVENT_VFORK;
1569                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1570                         trace = PTRACE_EVENT_CLONE;
1571                 else
1572                         trace = PTRACE_EVENT_FORK;
1573
1574                 if (likely(!ptrace_event_enabled(current, trace)))
1575                         trace = 0;
1576         }
1577
1578         p = copy_process(clone_flags, stack_start, stack_size,
1579                          child_tidptr, NULL, trace);
1580         /*
1581          * Do this prior waking up the new thread - the thread pointer
1582          * might get invalid after that point, if the thread exits quickly.
1583          */
1584         if (!IS_ERR(p)) {
1585                 struct completion vfork;
1586
1587                 trace_sched_process_fork(current, p);
1588
1589                 nr = task_pid_vnr(p);
1590
1591                 if (clone_flags & CLONE_PARENT_SETTID)
1592                         put_user(nr, parent_tidptr);
1593
1594                 if (clone_flags & CLONE_VFORK) {
1595                         p->vfork_done = &vfork;
1596                         init_completion(&vfork);
1597                         get_task_struct(p);
1598                 }
1599
1600                 wake_up_new_task(p);
1601
1602                 /* forking complete and child started to run, tell ptracer */
1603                 if (unlikely(trace))
1604                         ptrace_event(trace, nr);
1605
1606                 if (clone_flags & CLONE_VFORK) {
1607                         if (!wait_for_vfork_done(p, &vfork))
1608                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1609                 }
1610         } else {
1611                 nr = PTR_ERR(p);
1612         }
1613         return nr;
1614 }
1615
1616 #ifdef CONFIG_GENERIC_KERNEL_THREAD
1617 /*
1618  * Create a kernel thread.
1619  */
1620 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1621 {
1622         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1623                 (unsigned long)arg, NULL, NULL);
1624 }
1625 #endif
1626
1627 #ifdef __ARCH_WANT_SYS_FORK
1628 SYSCALL_DEFINE0(fork)
1629 {
1630 #ifdef CONFIG_MMU
1631         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1632 #else
1633         /* can not support in nommu mode */
1634         return(-EINVAL);
1635 #endif
1636 }
1637 #endif
1638
1639 #ifdef __ARCH_WANT_SYS_VFORK
1640 SYSCALL_DEFINE0(vfork)
1641 {
1642         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0, 
1643                         0, NULL, NULL);
1644 }
1645 #endif
1646
1647 #ifdef __ARCH_WANT_SYS_CLONE
1648 #ifdef CONFIG_CLONE_BACKWARDS
1649 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1650                  int __user *, parent_tidptr,
1651                  int, tls_val,
1652                  int __user *, child_tidptr)
1653 #elif defined(CONFIG_CLONE_BACKWARDS2)
1654 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1655                  int __user *, parent_tidptr,
1656                  int __user *, child_tidptr,
1657                  int, tls_val)
1658 #else
1659 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1660                  int __user *, parent_tidptr,
1661                  int __user *, child_tidptr,
1662                  int, tls_val)
1663 #endif
1664 {
1665         return do_fork(clone_flags, newsp, 0,
1666                 parent_tidptr, child_tidptr);
1667 }
1668 #endif
1669
1670 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1671 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1672 #endif
1673
1674 static void sighand_ctor(void *data)
1675 {
1676         struct sighand_struct *sighand = data;
1677
1678         spin_lock_init(&sighand->siglock);
1679         init_waitqueue_head(&sighand->signalfd_wqh);
1680 }
1681
1682 void __init proc_caches_init(void)
1683 {
1684         sighand_cachep = kmem_cache_create("sighand_cache",
1685                         sizeof(struct sighand_struct), 0,
1686                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1687                         SLAB_NOTRACK, sighand_ctor);
1688         signal_cachep = kmem_cache_create("signal_cache",
1689                         sizeof(struct signal_struct), 0,
1690                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1691         files_cachep = kmem_cache_create("files_cache",
1692                         sizeof(struct files_struct), 0,
1693                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1694         fs_cachep = kmem_cache_create("fs_cache",
1695                         sizeof(struct fs_struct), 0,
1696                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1697         /*
1698          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1699          * whole struct cpumask for the OFFSTACK case. We could change
1700          * this to *only* allocate as much of it as required by the
1701          * maximum number of CPU's we can ever have.  The cpumask_allocation
1702          * is at the end of the structure, exactly for that reason.
1703          */
1704         mm_cachep = kmem_cache_create("mm_struct",
1705                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1706                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1707         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1708         mmap_init();
1709         nsproxy_cache_init();
1710 }
1711
1712 /*
1713  * Check constraints on flags passed to the unshare system call.
1714  */
1715 static int check_unshare_flags(unsigned long unshare_flags)
1716 {
1717         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1718                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1719                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1720                                 CLONE_NEWUSER|CLONE_NEWPID))
1721                 return -EINVAL;
1722         /*
1723          * Not implemented, but pretend it works if there is nothing to
1724          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1725          * needs to unshare vm.
1726          */
1727         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1728                 /* FIXME: get_task_mm() increments ->mm_users */
1729                 if (atomic_read(&current->mm->mm_users) > 1)
1730                         return -EINVAL;
1731         }
1732
1733         return 0;
1734 }
1735
1736 /*
1737  * Unshare the filesystem structure if it is being shared
1738  */
1739 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1740 {
1741         struct fs_struct *fs = current->fs;
1742
1743         if (!(unshare_flags & CLONE_FS) || !fs)
1744                 return 0;
1745
1746         /* don't need lock here; in the worst case we'll do useless copy */
1747         if (fs->users == 1)
1748                 return 0;
1749
1750         *new_fsp = copy_fs_struct(fs);
1751         if (!*new_fsp)
1752                 return -ENOMEM;
1753
1754         return 0;
1755 }
1756
1757 /*
1758  * Unshare file descriptor table if it is being shared
1759  */
1760 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1761 {
1762         struct files_struct *fd = current->files;
1763         int error = 0;
1764
1765         if ((unshare_flags & CLONE_FILES) &&
1766             (fd && atomic_read(&fd->count) > 1)) {
1767                 *new_fdp = dup_fd(fd, &error);
1768                 if (!*new_fdp)
1769                         return error;
1770         }
1771
1772         return 0;
1773 }
1774
1775 /*
1776  * unshare allows a process to 'unshare' part of the process
1777  * context which was originally shared using clone.  copy_*
1778  * functions used by do_fork() cannot be used here directly
1779  * because they modify an inactive task_struct that is being
1780  * constructed. Here we are modifying the current, active,
1781  * task_struct.
1782  */
1783 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1784 {
1785         struct fs_struct *fs, *new_fs = NULL;
1786         struct files_struct *fd, *new_fd = NULL;
1787         struct cred *new_cred = NULL;
1788         struct nsproxy *new_nsproxy = NULL;
1789         int do_sysvsem = 0;
1790         int err;
1791
1792         /*
1793          * If unsharing a user namespace must also unshare the thread.
1794          */
1795         if (unshare_flags & CLONE_NEWUSER)
1796                 unshare_flags |= CLONE_THREAD;
1797         /*
1798          * If unsharing a pid namespace must also unshare the thread.
1799          */
1800         if (unshare_flags & CLONE_NEWPID)
1801                 unshare_flags |= CLONE_THREAD;
1802         /*
1803          * If unsharing a thread from a thread group, must also unshare vm.
1804          */
1805         if (unshare_flags & CLONE_THREAD)
1806                 unshare_flags |= CLONE_VM;
1807         /*
1808          * If unsharing vm, must also unshare signal handlers.
1809          */
1810         if (unshare_flags & CLONE_VM)
1811                 unshare_flags |= CLONE_SIGHAND;
1812         /*
1813          * If unsharing namespace, must also unshare filesystem information.
1814          */
1815         if (unshare_flags & CLONE_NEWNS)
1816                 unshare_flags |= CLONE_FS;
1817
1818         err = check_unshare_flags(unshare_flags);
1819         if (err)
1820                 goto bad_unshare_out;
1821         /*
1822          * CLONE_NEWIPC must also detach from the undolist: after switching
1823          * to a new ipc namespace, the semaphore arrays from the old
1824          * namespace are unreachable.
1825          */
1826         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1827                 do_sysvsem = 1;
1828         err = unshare_fs(unshare_flags, &new_fs);
1829         if (err)
1830                 goto bad_unshare_out;
1831         err = unshare_fd(unshare_flags, &new_fd);
1832         if (err)
1833                 goto bad_unshare_cleanup_fs;
1834         err = unshare_userns(unshare_flags, &new_cred);
1835         if (err)
1836                 goto bad_unshare_cleanup_fd;
1837         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1838                                          new_cred, new_fs);
1839         if (err)
1840                 goto bad_unshare_cleanup_cred;
1841
1842         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1843                 if (do_sysvsem) {
1844                         /*
1845                          * CLONE_SYSVSEM is equivalent to sys_exit().
1846                          */
1847                         exit_sem(current);
1848                 }
1849
1850                 if (new_nsproxy) {
1851                         switch_task_namespaces(current, new_nsproxy);
1852                         new_nsproxy = NULL;
1853                 }
1854
1855                 task_lock(current);
1856
1857                 if (new_fs) {
1858                         fs = current->fs;
1859                         spin_lock(&fs->lock);
1860                         current->fs = new_fs;
1861                         if (--fs->users)
1862                                 new_fs = NULL;
1863                         else
1864                                 new_fs = fs;
1865                         spin_unlock(&fs->lock);
1866                 }
1867
1868                 if (new_fd) {
1869                         fd = current->files;
1870                         current->files = new_fd;
1871                         new_fd = fd;
1872                 }
1873
1874                 task_unlock(current);
1875
1876                 if (new_cred) {
1877                         /* Install the new user namespace */
1878                         commit_creds(new_cred);
1879                         new_cred = NULL;
1880                 }
1881         }
1882
1883         if (new_nsproxy)
1884                 put_nsproxy(new_nsproxy);
1885
1886 bad_unshare_cleanup_cred:
1887         if (new_cred)
1888                 put_cred(new_cred);
1889 bad_unshare_cleanup_fd:
1890         if (new_fd)
1891                 put_files_struct(new_fd);
1892
1893 bad_unshare_cleanup_fs:
1894         if (new_fs)
1895                 free_fs_struct(new_fs);
1896
1897 bad_unshare_out:
1898         return err;
1899 }
1900
1901 /*
1902  *      Helper to unshare the files of the current task.
1903  *      We don't want to expose copy_files internals to
1904  *      the exec layer of the kernel.
1905  */
1906
1907 int unshare_files(struct files_struct **displaced)
1908 {
1909         struct task_struct *task = current;
1910         struct files_struct *copy = NULL;
1911         int error;
1912
1913         error = unshare_fd(CLONE_FILES, &copy);
1914         if (error || !copy) {
1915                 *displaced = NULL;
1916                 return error;
1917         }
1918         *displaced = task->files;
1919         task_lock(task);
1920         task->files = copy;
1921         task_unlock(task);
1922         return 0;
1923 }