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