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