mm: kill vma flag VM_EXECUTABLE and mm->num_exe_file_vmas
[linux-3.10.git] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/seccomp.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/kthread.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/proc_fs.h>
52 #include <linux/profile.h>
53 #include <linux/rmap.h>
54 #include <linux/ksm.h>
55 #include <linux/acct.h>
56 #include <linux/tsacct_kern.h>
57 #include <linux/cn_proc.h>
58 #include <linux/freezer.h>
59 #include <linux/delayacct.h>
60 #include <linux/taskstats_kern.h>
61 #include <linux/random.h>
62 #include <linux/tty.h>
63 #include <linux/blkdev.h>
64 #include <linux/fs_struct.h>
65 #include <linux/magic.h>
66 #include <linux/perf_event.h>
67 #include <linux/posix-timers.h>
68 #include <linux/user-return-notifier.h>
69 #include <linux/oom.h>
70 #include <linux/khugepaged.h>
71 #include <linux/signalfd.h>
72 #include <linux/uprobes.h>
73
74 #include <asm/pgtable.h>
75 #include <asm/pgalloc.h>
76 #include <asm/uaccess.h>
77 #include <asm/mmu_context.h>
78 #include <asm/cacheflush.h>
79 #include <asm/tlbflush.h>
80
81 #include <trace/events/sched.h>
82
83 #define CREATE_TRACE_POINTS
84 #include <trace/events/task.h>
85
86 /*
87  * Protected counters by write_lock_irq(&tasklist_lock)
88  */
89 unsigned long total_forks;      /* Handle normal Linux uptimes. */
90 int nr_threads;                 /* The idle threads do not count.. */
91
92 int max_threads;                /* tunable limit on nr_threads */
93
94 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
95
96 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
97
98 #ifdef CONFIG_PROVE_RCU
99 int lockdep_tasklist_lock_is_held(void)
100 {
101         return lockdep_is_held(&tasklist_lock);
102 }
103 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
104 #endif /* #ifdef CONFIG_PROVE_RCU */
105
106 int nr_processes(void)
107 {
108         int cpu;
109         int total = 0;
110
111         for_each_possible_cpu(cpu)
112                 total += per_cpu(process_counts, cpu);
113
114         return total;
115 }
116
117 void __weak arch_release_task_struct(struct task_struct *tsk)
118 {
119 }
120
121 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
122 static struct kmem_cache *task_struct_cachep;
123
124 static inline struct task_struct *alloc_task_struct_node(int node)
125 {
126         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
127 }
128
129 static inline void free_task_struct(struct task_struct *tsk)
130 {
131         kmem_cache_free(task_struct_cachep, tsk);
132 }
133 #endif
134
135 void __weak arch_release_thread_info(struct thread_info *ti)
136 {
137 }
138
139 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
140
141 /*
142  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
143  * kmemcache based allocator.
144  */
145 # if THREAD_SIZE >= PAGE_SIZE
146 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
147                                                   int node)
148 {
149         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
150                                              THREAD_SIZE_ORDER);
151
152         return page ? page_address(page) : NULL;
153 }
154
155 static inline void free_thread_info(struct thread_info *ti)
156 {
157         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
158 }
159 # else
160 static struct kmem_cache *thread_info_cache;
161
162 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
163                                                   int node)
164 {
165         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
166 }
167
168 static void free_thread_info(struct thread_info *ti)
169 {
170         kmem_cache_free(thread_info_cache, ti);
171 }
172
173 void thread_info_cache_init(void)
174 {
175         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
176                                               THREAD_SIZE, 0, NULL);
177         BUG_ON(thread_info_cache == NULL);
178 }
179 # endif
180 #endif
181
182 /* SLAB cache for signal_struct structures (tsk->signal) */
183 static struct kmem_cache *signal_cachep;
184
185 /* SLAB cache for sighand_struct structures (tsk->sighand) */
186 struct kmem_cache *sighand_cachep;
187
188 /* SLAB cache for files_struct structures (tsk->files) */
189 struct kmem_cache *files_cachep;
190
191 /* SLAB cache for fs_struct structures (tsk->fs) */
192 struct kmem_cache *fs_cachep;
193
194 /* SLAB cache for vm_area_struct structures */
195 struct kmem_cache *vm_area_cachep;
196
197 /* SLAB cache for mm_struct structures (tsk->mm) */
198 static struct kmem_cache *mm_cachep;
199
200 static void account_kernel_stack(struct thread_info *ti, int account)
201 {
202         struct zone *zone = page_zone(virt_to_page(ti));
203
204         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
205 }
206
207 void free_task(struct task_struct *tsk)
208 {
209         account_kernel_stack(tsk->stack, -1);
210         arch_release_thread_info(tsk->stack);
211         free_thread_info(tsk->stack);
212         rt_mutex_debug_task_free(tsk);
213         ftrace_graph_exit_task(tsk);
214         put_seccomp_filter(tsk);
215         arch_release_task_struct(tsk);
216         free_task_struct(tsk);
217 }
218 EXPORT_SYMBOL(free_task);
219
220 static inline void free_signal_struct(struct signal_struct *sig)
221 {
222         taskstats_tgid_free(sig);
223         sched_autogroup_exit(sig);
224         kmem_cache_free(signal_cachep, sig);
225 }
226
227 static inline void put_signal_struct(struct signal_struct *sig)
228 {
229         if (atomic_dec_and_test(&sig->sigcnt))
230                 free_signal_struct(sig);
231 }
232
233 void __put_task_struct(struct task_struct *tsk)
234 {
235         WARN_ON(!tsk->exit_state);
236         WARN_ON(atomic_read(&tsk->usage));
237         WARN_ON(tsk == current);
238
239         security_task_free(tsk);
240         exit_creds(tsk);
241         delayacct_tsk_free(tsk);
242         put_signal_struct(tsk->signal);
243
244         if (!profile_handoff_task(tsk))
245                 free_task(tsk);
246 }
247 EXPORT_SYMBOL_GPL(__put_task_struct);
248
249 void __init __weak arch_task_cache_init(void) { }
250
251 void __init fork_init(unsigned long mempages)
252 {
253 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
254 #ifndef ARCH_MIN_TASKALIGN
255 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
256 #endif
257         /* create a slab on which task_structs can be allocated */
258         task_struct_cachep =
259                 kmem_cache_create("task_struct", sizeof(struct task_struct),
260                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
261 #endif
262
263         /* do the arch specific task caches init */
264         arch_task_cache_init();
265
266         /*
267          * The default maximum number of threads is set to a safe
268          * value: the thread structures can take up at most half
269          * of memory.
270          */
271         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
272
273         /*
274          * we need to allow at least 20 threads to boot a system
275          */
276         if (max_threads < 20)
277                 max_threads = 20;
278
279         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
280         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
281         init_task.signal->rlim[RLIMIT_SIGPENDING] =
282                 init_task.signal->rlim[RLIMIT_NPROC];
283 }
284
285 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
286                                                struct task_struct *src)
287 {
288         *dst = *src;
289         return 0;
290 }
291
292 static struct task_struct *dup_task_struct(struct task_struct *orig)
293 {
294         struct task_struct *tsk;
295         struct thread_info *ti;
296         unsigned long *stackend;
297         int node = tsk_fork_get_node(orig);
298         int err;
299
300         tsk = alloc_task_struct_node(node);
301         if (!tsk)
302                 return NULL;
303
304         ti = alloc_thread_info_node(tsk, node);
305         if (!ti)
306                 goto free_tsk;
307
308         err = arch_dup_task_struct(tsk, orig);
309         if (err)
310                 goto free_ti;
311
312         tsk->stack = ti;
313
314         setup_thread_stack(tsk, orig);
315         clear_user_return_notifier(tsk);
316         clear_tsk_need_resched(tsk);
317         stackend = end_of_stack(tsk);
318         *stackend = STACK_END_MAGIC;    /* for overflow detection */
319
320 #ifdef CONFIG_CC_STACKPROTECTOR
321         tsk->stack_canary = get_random_int();
322 #endif
323
324         /*
325          * One for us, one for whoever does the "release_task()" (usually
326          * parent)
327          */
328         atomic_set(&tsk->usage, 2);
329 #ifdef CONFIG_BLK_DEV_IO_TRACE
330         tsk->btrace_seq = 0;
331 #endif
332         tsk->splice_pipe = NULL;
333         tsk->task_frag.page = NULL;
334
335         account_kernel_stack(ti, 1);
336
337         return tsk;
338
339 free_ti:
340         free_thread_info(ti);
341 free_tsk:
342         free_task_struct(tsk);
343         return NULL;
344 }
345
346 #ifdef CONFIG_MMU
347 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
348 {
349         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
350         struct rb_node **rb_link, *rb_parent;
351         int retval;
352         unsigned long charge;
353         struct mempolicy *pol;
354
355         down_write(&oldmm->mmap_sem);
356         flush_cache_dup_mm(oldmm);
357         uprobe_dup_mmap(oldmm, mm);
358         /*
359          * Not linked in yet - no deadlock potential:
360          */
361         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
362
363         mm->locked_vm = 0;
364         mm->mmap = NULL;
365         mm->mmap_cache = NULL;
366         mm->free_area_cache = oldmm->mmap_base;
367         mm->cached_hole_size = ~0UL;
368         mm->map_count = 0;
369         cpumask_clear(mm_cpumask(mm));
370         mm->mm_rb = RB_ROOT;
371         rb_link = &mm->mm_rb.rb_node;
372         rb_parent = NULL;
373         pprev = &mm->mmap;
374         retval = ksm_fork(mm, oldmm);
375         if (retval)
376                 goto out;
377         retval = khugepaged_fork(mm, oldmm);
378         if (retval)
379                 goto out;
380
381         prev = NULL;
382         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
383                 struct file *file;
384
385                 if (mpnt->vm_flags & VM_DONTCOPY) {
386                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
387                                                         -vma_pages(mpnt));
388                         continue;
389                 }
390                 charge = 0;
391                 if (mpnt->vm_flags & VM_ACCOUNT) {
392                         unsigned long len = vma_pages(mpnt);
393
394                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
395                                 goto fail_nomem;
396                         charge = len;
397                 }
398                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
399                 if (!tmp)
400                         goto fail_nomem;
401                 *tmp = *mpnt;
402                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
403                 pol = mpol_dup(vma_policy(mpnt));
404                 retval = PTR_ERR(pol);
405                 if (IS_ERR(pol))
406                         goto fail_nomem_policy;
407                 vma_set_policy(tmp, pol);
408                 tmp->vm_mm = mm;
409                 if (anon_vma_fork(tmp, mpnt))
410                         goto fail_nomem_anon_vma_fork;
411                 tmp->vm_flags &= ~VM_LOCKED;
412                 tmp->vm_next = tmp->vm_prev = NULL;
413                 file = tmp->vm_file;
414                 if (file) {
415                         struct inode *inode = file->f_path.dentry->d_inode;
416                         struct address_space *mapping = file->f_mapping;
417
418                         get_file(file);
419                         if (tmp->vm_flags & VM_DENYWRITE)
420                                 atomic_dec(&inode->i_writecount);
421                         mutex_lock(&mapping->i_mmap_mutex);
422                         if (tmp->vm_flags & VM_SHARED)
423                                 mapping->i_mmap_writable++;
424                         flush_dcache_mmap_lock(mapping);
425                         /* insert tmp into the share list, just after mpnt */
426                         vma_prio_tree_add(tmp, mpnt);
427                         flush_dcache_mmap_unlock(mapping);
428                         mutex_unlock(&mapping->i_mmap_mutex);
429                 }
430
431                 /*
432                  * Clear hugetlb-related page reserves for children. This only
433                  * affects MAP_PRIVATE mappings. Faults generated by the child
434                  * are not guaranteed to succeed, even if read-only
435                  */
436                 if (is_vm_hugetlb_page(tmp))
437                         reset_vma_resv_huge_pages(tmp);
438
439                 /*
440                  * Link in the new vma and copy the page table entries.
441                  */
442                 *pprev = tmp;
443                 pprev = &tmp->vm_next;
444                 tmp->vm_prev = prev;
445                 prev = tmp;
446
447                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
448                 rb_link = &tmp->vm_rb.rb_right;
449                 rb_parent = &tmp->vm_rb;
450
451                 mm->map_count++;
452                 retval = copy_page_range(mm, oldmm, mpnt);
453
454                 if (tmp->vm_ops && tmp->vm_ops->open)
455                         tmp->vm_ops->open(tmp);
456
457                 if (retval)
458                         goto out;
459         }
460         /* a new mm has just been created */
461         arch_dup_mmap(oldmm, mm);
462         retval = 0;
463 out:
464         up_write(&mm->mmap_sem);
465         flush_tlb_mm(oldmm);
466         up_write(&oldmm->mmap_sem);
467         return retval;
468 fail_nomem_anon_vma_fork:
469         mpol_put(pol);
470 fail_nomem_policy:
471         kmem_cache_free(vm_area_cachep, tmp);
472 fail_nomem:
473         retval = -ENOMEM;
474         vm_unacct_memory(charge);
475         goto out;
476 }
477
478 static inline int mm_alloc_pgd(struct mm_struct *mm)
479 {
480         mm->pgd = pgd_alloc(mm);
481         if (unlikely(!mm->pgd))
482                 return -ENOMEM;
483         return 0;
484 }
485
486 static inline void mm_free_pgd(struct mm_struct *mm)
487 {
488         pgd_free(mm, mm->pgd);
489 }
490 #else
491 #define dup_mmap(mm, oldmm)     (0)
492 #define mm_alloc_pgd(mm)        (0)
493 #define mm_free_pgd(mm)
494 #endif /* CONFIG_MMU */
495
496 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
497
498 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
499 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
500
501 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
502
503 static int __init coredump_filter_setup(char *s)
504 {
505         default_dump_filter =
506                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
507                 MMF_DUMP_FILTER_MASK;
508         return 1;
509 }
510
511 __setup("coredump_filter=", coredump_filter_setup);
512
513 #include <linux/init_task.h>
514
515 static void mm_init_aio(struct mm_struct *mm)
516 {
517 #ifdef CONFIG_AIO
518         spin_lock_init(&mm->ioctx_lock);
519         INIT_HLIST_HEAD(&mm->ioctx_list);
520 #endif
521 }
522
523 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
524 {
525         atomic_set(&mm->mm_users, 1);
526         atomic_set(&mm->mm_count, 1);
527         init_rwsem(&mm->mmap_sem);
528         INIT_LIST_HEAD(&mm->mmlist);
529         mm->flags = (current->mm) ?
530                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
531         mm->core_state = NULL;
532         mm->nr_ptes = 0;
533         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
534         spin_lock_init(&mm->page_table_lock);
535         mm->free_area_cache = TASK_UNMAPPED_BASE;
536         mm->cached_hole_size = ~0UL;
537         mm_init_aio(mm);
538         mm_init_owner(mm, p);
539
540         if (likely(!mm_alloc_pgd(mm))) {
541                 mm->def_flags = 0;
542                 mmu_notifier_mm_init(mm);
543                 return mm;
544         }
545
546         free_mm(mm);
547         return NULL;
548 }
549
550 static void check_mm(struct mm_struct *mm)
551 {
552         int i;
553
554         for (i = 0; i < NR_MM_COUNTERS; i++) {
555                 long x = atomic_long_read(&mm->rss_stat.count[i]);
556
557                 if (unlikely(x))
558                         printk(KERN_ALERT "BUG: Bad rss-counter state "
559                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
560         }
561
562 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
563         VM_BUG_ON(mm->pmd_huge_pte);
564 #endif
565 }
566
567 /*
568  * Allocate and initialize an mm_struct.
569  */
570 struct mm_struct *mm_alloc(void)
571 {
572         struct mm_struct *mm;
573
574         mm = allocate_mm();
575         if (!mm)
576                 return NULL;
577
578         memset(mm, 0, sizeof(*mm));
579         mm_init_cpumask(mm);
580         return mm_init(mm, current);
581 }
582
583 /*
584  * Called when the last reference to the mm
585  * is dropped: either by a lazy thread or by
586  * mmput. Free the page directory and the mm.
587  */
588 void __mmdrop(struct mm_struct *mm)
589 {
590         BUG_ON(mm == &init_mm);
591         mm_free_pgd(mm);
592         destroy_context(mm);
593         mmu_notifier_mm_destroy(mm);
594         check_mm(mm);
595         free_mm(mm);
596 }
597 EXPORT_SYMBOL_GPL(__mmdrop);
598
599 /*
600  * Decrement the use count and release all resources for an mm.
601  */
602 void mmput(struct mm_struct *mm)
603 {
604         might_sleep();
605
606         if (atomic_dec_and_test(&mm->mm_users)) {
607                 uprobe_clear_state(mm);
608                 exit_aio(mm);
609                 ksm_exit(mm);
610                 khugepaged_exit(mm); /* must run before exit_mmap */
611                 exit_mmap(mm);
612                 set_mm_exe_file(mm, NULL);
613                 if (!list_empty(&mm->mmlist)) {
614                         spin_lock(&mmlist_lock);
615                         list_del(&mm->mmlist);
616                         spin_unlock(&mmlist_lock);
617                 }
618                 if (mm->binfmt)
619                         module_put(mm->binfmt->module);
620                 mmdrop(mm);
621         }
622 }
623 EXPORT_SYMBOL_GPL(mmput);
624
625 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
626 {
627         if (new_exe_file)
628                 get_file(new_exe_file);
629         if (mm->exe_file)
630                 fput(mm->exe_file);
631         mm->exe_file = new_exe_file;
632 }
633
634 struct file *get_mm_exe_file(struct mm_struct *mm)
635 {
636         struct file *exe_file;
637
638         /* We need mmap_sem to protect against races with removal of exe_file */
639         down_read(&mm->mmap_sem);
640         exe_file = mm->exe_file;
641         if (exe_file)
642                 get_file(exe_file);
643         up_read(&mm->mmap_sem);
644         return exe_file;
645 }
646
647 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
648 {
649         /* It's safe to write the exe_file pointer without exe_file_lock because
650          * this is called during fork when the task is not yet in /proc */
651         newmm->exe_file = get_mm_exe_file(oldmm);
652 }
653
654 /**
655  * get_task_mm - acquire a reference to the task's mm
656  *
657  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
658  * this kernel workthread has transiently adopted a user mm with use_mm,
659  * to do its AIO) is not set and if so returns a reference to it, after
660  * bumping up the use count.  User must release the mm via mmput()
661  * after use.  Typically used by /proc and ptrace.
662  */
663 struct mm_struct *get_task_mm(struct task_struct *task)
664 {
665         struct mm_struct *mm;
666
667         task_lock(task);
668         mm = task->mm;
669         if (mm) {
670                 if (task->flags & PF_KTHREAD)
671                         mm = NULL;
672                 else
673                         atomic_inc(&mm->mm_users);
674         }
675         task_unlock(task);
676         return mm;
677 }
678 EXPORT_SYMBOL_GPL(get_task_mm);
679
680 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
681 {
682         struct mm_struct *mm;
683         int err;
684
685         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
686         if (err)
687                 return ERR_PTR(err);
688
689         mm = get_task_mm(task);
690         if (mm && mm != current->mm &&
691                         !ptrace_may_access(task, mode)) {
692                 mmput(mm);
693                 mm = ERR_PTR(-EACCES);
694         }
695         mutex_unlock(&task->signal->cred_guard_mutex);
696
697         return mm;
698 }
699
700 static void complete_vfork_done(struct task_struct *tsk)
701 {
702         struct completion *vfork;
703
704         task_lock(tsk);
705         vfork = tsk->vfork_done;
706         if (likely(vfork)) {
707                 tsk->vfork_done = NULL;
708                 complete(vfork);
709         }
710         task_unlock(tsk);
711 }
712
713 static int wait_for_vfork_done(struct task_struct *child,
714                                 struct completion *vfork)
715 {
716         int killed;
717
718         freezer_do_not_count();
719         killed = wait_for_completion_killable(vfork);
720         freezer_count();
721
722         if (killed) {
723                 task_lock(child);
724                 child->vfork_done = NULL;
725                 task_unlock(child);
726         }
727
728         put_task_struct(child);
729         return killed;
730 }
731
732 /* Please note the differences between mmput and mm_release.
733  * mmput is called whenever we stop holding onto a mm_struct,
734  * error success whatever.
735  *
736  * mm_release is called after a mm_struct has been removed
737  * from the current process.
738  *
739  * This difference is important for error handling, when we
740  * only half set up a mm_struct for a new process and need to restore
741  * the old one.  Because we mmput the new mm_struct before
742  * restoring the old one. . .
743  * Eric Biederman 10 January 1998
744  */
745 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
746 {
747         /* Get rid of any futexes when releasing the mm */
748 #ifdef CONFIG_FUTEX
749         if (unlikely(tsk->robust_list)) {
750                 exit_robust_list(tsk);
751                 tsk->robust_list = NULL;
752         }
753 #ifdef CONFIG_COMPAT
754         if (unlikely(tsk->compat_robust_list)) {
755                 compat_exit_robust_list(tsk);
756                 tsk->compat_robust_list = NULL;
757         }
758 #endif
759         if (unlikely(!list_empty(&tsk->pi_state_list)))
760                 exit_pi_state_list(tsk);
761 #endif
762
763         uprobe_free_utask(tsk);
764
765         /* Get rid of any cached register state */
766         deactivate_mm(tsk, mm);
767
768         /*
769          * If we're exiting normally, clear a user-space tid field if
770          * requested.  We leave this alone when dying by signal, to leave
771          * the value intact in a core dump, and to save the unnecessary
772          * trouble, say, a killed vfork parent shouldn't touch this mm.
773          * Userland only wants this done for a sys_exit.
774          */
775         if (tsk->clear_child_tid) {
776                 if (!(tsk->flags & PF_SIGNALED) &&
777                     atomic_read(&mm->mm_users) > 1) {
778                         /*
779                          * We don't check the error code - if userspace has
780                          * not set up a proper pointer then tough luck.
781                          */
782                         put_user(0, tsk->clear_child_tid);
783                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
784                                         1, NULL, NULL, 0);
785                 }
786                 tsk->clear_child_tid = NULL;
787         }
788
789         /*
790          * All done, finally we can wake up parent and return this mm to him.
791          * Also kthread_stop() uses this completion for synchronization.
792          */
793         if (tsk->vfork_done)
794                 complete_vfork_done(tsk);
795 }
796
797 /*
798  * Allocate a new mm structure and copy contents from the
799  * mm structure of the passed in task structure.
800  */
801 struct mm_struct *dup_mm(struct task_struct *tsk)
802 {
803         struct mm_struct *mm, *oldmm = current->mm;
804         int err;
805
806         if (!oldmm)
807                 return NULL;
808
809         mm = allocate_mm();
810         if (!mm)
811                 goto fail_nomem;
812
813         memcpy(mm, oldmm, sizeof(*mm));
814         mm_init_cpumask(mm);
815
816 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
817         mm->pmd_huge_pte = NULL;
818 #endif
819         if (!mm_init(mm, tsk))
820                 goto fail_nomem;
821
822         if (init_new_context(tsk, mm))
823                 goto fail_nocontext;
824
825         dup_mm_exe_file(oldmm, mm);
826
827         err = dup_mmap(mm, oldmm);
828         if (err)
829                 goto free_pt;
830
831         mm->hiwater_rss = get_mm_rss(mm);
832         mm->hiwater_vm = mm->total_vm;
833
834         if (mm->binfmt && !try_module_get(mm->binfmt->module))
835                 goto free_pt;
836
837         return mm;
838
839 free_pt:
840         /* don't put binfmt in mmput, we haven't got module yet */
841         mm->binfmt = NULL;
842         mmput(mm);
843
844 fail_nomem:
845         return NULL;
846
847 fail_nocontext:
848         /*
849          * If init_new_context() failed, we cannot use mmput() to free the mm
850          * because it calls destroy_context()
851          */
852         mm_free_pgd(mm);
853         free_mm(mm);
854         return NULL;
855 }
856
857 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
858 {
859         struct mm_struct *mm, *oldmm;
860         int retval;
861
862         tsk->min_flt = tsk->maj_flt = 0;
863         tsk->nvcsw = tsk->nivcsw = 0;
864 #ifdef CONFIG_DETECT_HUNG_TASK
865         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
866 #endif
867
868         tsk->mm = NULL;
869         tsk->active_mm = NULL;
870
871         /*
872          * Are we cloning a kernel thread?
873          *
874          * We need to steal a active VM for that..
875          */
876         oldmm = current->mm;
877         if (!oldmm)
878                 return 0;
879
880         if (clone_flags & CLONE_VM) {
881                 atomic_inc(&oldmm->mm_users);
882                 mm = oldmm;
883                 goto good_mm;
884         }
885
886         retval = -ENOMEM;
887         mm = dup_mm(tsk);
888         if (!mm)
889                 goto fail_nomem;
890
891 good_mm:
892         tsk->mm = mm;
893         tsk->active_mm = mm;
894         return 0;
895
896 fail_nomem:
897         return retval;
898 }
899
900 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
901 {
902         struct fs_struct *fs = current->fs;
903         if (clone_flags & CLONE_FS) {
904                 /* tsk->fs is already what we want */
905                 spin_lock(&fs->lock);
906                 if (fs->in_exec) {
907                         spin_unlock(&fs->lock);
908                         return -EAGAIN;
909                 }
910                 fs->users++;
911                 spin_unlock(&fs->lock);
912                 return 0;
913         }
914         tsk->fs = copy_fs_struct(fs);
915         if (!tsk->fs)
916                 return -ENOMEM;
917         return 0;
918 }
919
920 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
921 {
922         struct files_struct *oldf, *newf;
923         int error = 0;
924
925         /*
926          * A background process may not have any files ...
927          */
928         oldf = current->files;
929         if (!oldf)
930                 goto out;
931
932         if (clone_flags & CLONE_FILES) {
933                 atomic_inc(&oldf->count);
934                 goto out;
935         }
936
937         newf = dup_fd(oldf, &error);
938         if (!newf)
939                 goto out;
940
941         tsk->files = newf;
942         error = 0;
943 out:
944         return error;
945 }
946
947 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
948 {
949 #ifdef CONFIG_BLOCK
950         struct io_context *ioc = current->io_context;
951         struct io_context *new_ioc;
952
953         if (!ioc)
954                 return 0;
955         /*
956          * Share io context with parent, if CLONE_IO is set
957          */
958         if (clone_flags & CLONE_IO) {
959                 ioc_task_link(ioc);
960                 tsk->io_context = ioc;
961         } else if (ioprio_valid(ioc->ioprio)) {
962                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
963                 if (unlikely(!new_ioc))
964                         return -ENOMEM;
965
966                 new_ioc->ioprio = ioc->ioprio;
967                 put_io_context(new_ioc);
968         }
969 #endif
970         return 0;
971 }
972
973 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
974 {
975         struct sighand_struct *sig;
976
977         if (clone_flags & CLONE_SIGHAND) {
978                 atomic_inc(&current->sighand->count);
979                 return 0;
980         }
981         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
982         rcu_assign_pointer(tsk->sighand, sig);
983         if (!sig)
984                 return -ENOMEM;
985         atomic_set(&sig->count, 1);
986         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
987         return 0;
988 }
989
990 void __cleanup_sighand(struct sighand_struct *sighand)
991 {
992         if (atomic_dec_and_test(&sighand->count)) {
993                 signalfd_cleanup(sighand);
994                 kmem_cache_free(sighand_cachep, sighand);
995         }
996 }
997
998
999 /*
1000  * Initialize POSIX timer handling for a thread group.
1001  */
1002 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1003 {
1004         unsigned long cpu_limit;
1005
1006         /* Thread group counters. */
1007         thread_group_cputime_init(sig);
1008
1009         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1010         if (cpu_limit != RLIM_INFINITY) {
1011                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1012                 sig->cputimer.running = 1;
1013         }
1014
1015         /* The timer lists. */
1016         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1017         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1018         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1019 }
1020
1021 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1022 {
1023         struct signal_struct *sig;
1024
1025         if (clone_flags & CLONE_THREAD)
1026                 return 0;
1027
1028         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1029         tsk->signal = sig;
1030         if (!sig)
1031                 return -ENOMEM;
1032
1033         sig->nr_threads = 1;
1034         atomic_set(&sig->live, 1);
1035         atomic_set(&sig->sigcnt, 1);
1036         init_waitqueue_head(&sig->wait_chldexit);
1037         if (clone_flags & CLONE_NEWPID)
1038                 sig->flags |= SIGNAL_UNKILLABLE;
1039         sig->curr_target = tsk;
1040         init_sigpending(&sig->shared_pending);
1041         INIT_LIST_HEAD(&sig->posix_timers);
1042
1043         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1044         sig->real_timer.function = it_real_fn;
1045
1046         task_lock(current->group_leader);
1047         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1048         task_unlock(current->group_leader);
1049
1050         posix_cpu_timers_init_group(sig);
1051
1052         tty_audit_fork(sig);
1053         sched_autogroup_fork(sig);
1054
1055 #ifdef CONFIG_CGROUPS
1056         init_rwsem(&sig->group_rwsem);
1057 #endif
1058
1059         sig->oom_adj = current->signal->oom_adj;
1060         sig->oom_score_adj = current->signal->oom_score_adj;
1061         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1062
1063         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1064                                    current->signal->is_child_subreaper;
1065
1066         mutex_init(&sig->cred_guard_mutex);
1067
1068         return 0;
1069 }
1070
1071 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1072 {
1073         unsigned long new_flags = p->flags;
1074
1075         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1076         new_flags |= PF_FORKNOEXEC;
1077         p->flags = new_flags;
1078 }
1079
1080 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1081 {
1082         current->clear_child_tid = tidptr;
1083
1084         return task_pid_vnr(current);
1085 }
1086
1087 static void rt_mutex_init_task(struct task_struct *p)
1088 {
1089         raw_spin_lock_init(&p->pi_lock);
1090 #ifdef CONFIG_RT_MUTEXES
1091         plist_head_init(&p->pi_waiters);
1092         p->pi_blocked_on = NULL;
1093 #endif
1094 }
1095
1096 #ifdef CONFIG_MM_OWNER
1097 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1098 {
1099         mm->owner = p;
1100 }
1101 #endif /* CONFIG_MM_OWNER */
1102
1103 /*
1104  * Initialize POSIX timer handling for a single task.
1105  */
1106 static void posix_cpu_timers_init(struct task_struct *tsk)
1107 {
1108         tsk->cputime_expires.prof_exp = 0;
1109         tsk->cputime_expires.virt_exp = 0;
1110         tsk->cputime_expires.sched_exp = 0;
1111         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1112         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1113         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1114 }
1115
1116 /*
1117  * This creates a new process as a copy of the old one,
1118  * but does not actually start it yet.
1119  *
1120  * It copies the registers, and all the appropriate
1121  * parts of the process environment (as per the clone
1122  * flags). The actual kick-off is left to the caller.
1123  */
1124 static struct task_struct *copy_process(unsigned long clone_flags,
1125                                         unsigned long stack_start,
1126                                         struct pt_regs *regs,
1127                                         unsigned long stack_size,
1128                                         int __user *child_tidptr,
1129                                         struct pid *pid,
1130                                         int trace)
1131 {
1132         int retval;
1133         struct task_struct *p;
1134         int cgroup_callbacks_done = 0;
1135
1136         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1137                 return ERR_PTR(-EINVAL);
1138
1139         /*
1140          * Thread groups must share signals as well, and detached threads
1141          * can only be started up within the thread group.
1142          */
1143         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1144                 return ERR_PTR(-EINVAL);
1145
1146         /*
1147          * Shared signal handlers imply shared VM. By way of the above,
1148          * thread groups also imply shared VM. Blocking this case allows
1149          * for various simplifications in other code.
1150          */
1151         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1152                 return ERR_PTR(-EINVAL);
1153
1154         /*
1155          * Siblings of global init remain as zombies on exit since they are
1156          * not reaped by their parent (swapper). To solve this and to avoid
1157          * multi-rooted process trees, prevent global and container-inits
1158          * from creating siblings.
1159          */
1160         if ((clone_flags & CLONE_PARENT) &&
1161                                 current->signal->flags & SIGNAL_UNKILLABLE)
1162                 return ERR_PTR(-EINVAL);
1163
1164         retval = security_task_create(clone_flags);
1165         if (retval)
1166                 goto fork_out;
1167
1168         retval = -ENOMEM;
1169         p = dup_task_struct(current);
1170         if (!p)
1171                 goto fork_out;
1172
1173         ftrace_graph_init_task(p);
1174         get_seccomp_filter(p);
1175
1176         rt_mutex_init_task(p);
1177
1178 #ifdef CONFIG_PROVE_LOCKING
1179         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1180         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1181 #endif
1182         retval = -EAGAIN;
1183         if (atomic_read(&p->real_cred->user->processes) >=
1184                         task_rlimit(p, RLIMIT_NPROC)) {
1185                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1186                     p->real_cred->user != INIT_USER)
1187                         goto bad_fork_free;
1188         }
1189         current->flags &= ~PF_NPROC_EXCEEDED;
1190
1191         retval = copy_creds(p, clone_flags);
1192         if (retval < 0)
1193                 goto bad_fork_free;
1194
1195         /*
1196          * If multiple threads are within copy_process(), then this check
1197          * triggers too late. This doesn't hurt, the check is only there
1198          * to stop root fork bombs.
1199          */
1200         retval = -EAGAIN;
1201         if (nr_threads >= max_threads)
1202                 goto bad_fork_cleanup_count;
1203
1204         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1205                 goto bad_fork_cleanup_count;
1206
1207         p->did_exec = 0;
1208         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1209         copy_flags(clone_flags, p);
1210         INIT_LIST_HEAD(&p->children);
1211         INIT_LIST_HEAD(&p->sibling);
1212         rcu_copy_process(p);
1213         p->vfork_done = NULL;
1214         spin_lock_init(&p->alloc_lock);
1215
1216         init_sigpending(&p->pending);
1217
1218         p->utime = p->stime = p->gtime = 0;
1219         p->utimescaled = p->stimescaled = 0;
1220 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1221         p->prev_utime = p->prev_stime = 0;
1222 #endif
1223 #if defined(SPLIT_RSS_COUNTING)
1224         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1225 #endif
1226
1227         p->default_timer_slack_ns = current->timer_slack_ns;
1228
1229         task_io_accounting_init(&p->ioac);
1230         acct_clear_integrals(p);
1231
1232         posix_cpu_timers_init(p);
1233
1234         do_posix_clock_monotonic_gettime(&p->start_time);
1235         p->real_start_time = p->start_time;
1236         monotonic_to_bootbased(&p->real_start_time);
1237         p->io_context = NULL;
1238         p->audit_context = NULL;
1239         if (clone_flags & CLONE_THREAD)
1240                 threadgroup_change_begin(current);
1241         cgroup_fork(p);
1242 #ifdef CONFIG_NUMA
1243         p->mempolicy = mpol_dup(p->mempolicy);
1244         if (IS_ERR(p->mempolicy)) {
1245                 retval = PTR_ERR(p->mempolicy);
1246                 p->mempolicy = NULL;
1247                 goto bad_fork_cleanup_cgroup;
1248         }
1249         mpol_fix_fork_child_flag(p);
1250 #endif
1251 #ifdef CONFIG_CPUSETS
1252         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1253         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1254         seqcount_init(&p->mems_allowed_seq);
1255 #endif
1256 #ifdef CONFIG_TRACE_IRQFLAGS
1257         p->irq_events = 0;
1258         p->hardirqs_enabled = 0;
1259         p->hardirq_enable_ip = 0;
1260         p->hardirq_enable_event = 0;
1261         p->hardirq_disable_ip = _THIS_IP_;
1262         p->hardirq_disable_event = 0;
1263         p->softirqs_enabled = 1;
1264         p->softirq_enable_ip = _THIS_IP_;
1265         p->softirq_enable_event = 0;
1266         p->softirq_disable_ip = 0;
1267         p->softirq_disable_event = 0;
1268         p->hardirq_context = 0;
1269         p->softirq_context = 0;
1270 #endif
1271 #ifdef CONFIG_LOCKDEP
1272         p->lockdep_depth = 0; /* no locks held yet */
1273         p->curr_chain_key = 0;
1274         p->lockdep_recursion = 0;
1275 #endif
1276
1277 #ifdef CONFIG_DEBUG_MUTEXES
1278         p->blocked_on = NULL; /* not blocked yet */
1279 #endif
1280 #ifdef CONFIG_MEMCG
1281         p->memcg_batch.do_batch = 0;
1282         p->memcg_batch.memcg = NULL;
1283 #endif
1284
1285         /* Perform scheduler related setup. Assign this task to a CPU. */
1286         sched_fork(p);
1287
1288         retval = perf_event_init_task(p);
1289         if (retval)
1290                 goto bad_fork_cleanup_policy;
1291         retval = audit_alloc(p);
1292         if (retval)
1293                 goto bad_fork_cleanup_policy;
1294         /* copy all the process information */
1295         retval = copy_semundo(clone_flags, p);
1296         if (retval)
1297                 goto bad_fork_cleanup_audit;
1298         retval = copy_files(clone_flags, p);
1299         if (retval)
1300                 goto bad_fork_cleanup_semundo;
1301         retval = copy_fs(clone_flags, p);
1302         if (retval)
1303                 goto bad_fork_cleanup_files;
1304         retval = copy_sighand(clone_flags, p);
1305         if (retval)
1306                 goto bad_fork_cleanup_fs;
1307         retval = copy_signal(clone_flags, p);
1308         if (retval)
1309                 goto bad_fork_cleanup_sighand;
1310         retval = copy_mm(clone_flags, p);
1311         if (retval)
1312                 goto bad_fork_cleanup_signal;
1313         retval = copy_namespaces(clone_flags, p);
1314         if (retval)
1315                 goto bad_fork_cleanup_mm;
1316         retval = copy_io(clone_flags, p);
1317         if (retval)
1318                 goto bad_fork_cleanup_namespaces;
1319         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1320         if (retval)
1321                 goto bad_fork_cleanup_io;
1322
1323         if (pid != &init_struct_pid) {
1324                 retval = -ENOMEM;
1325                 pid = alloc_pid(p->nsproxy->pid_ns);
1326                 if (!pid)
1327                         goto bad_fork_cleanup_io;
1328         }
1329
1330         p->pid = pid_nr(pid);
1331         p->tgid = p->pid;
1332         if (clone_flags & CLONE_THREAD)
1333                 p->tgid = current->tgid;
1334
1335         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1336         /*
1337          * Clear TID on mm_release()?
1338          */
1339         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1340 #ifdef CONFIG_BLOCK
1341         p->plug = NULL;
1342 #endif
1343 #ifdef CONFIG_FUTEX
1344         p->robust_list = NULL;
1345 #ifdef CONFIG_COMPAT
1346         p->compat_robust_list = NULL;
1347 #endif
1348         INIT_LIST_HEAD(&p->pi_state_list);
1349         p->pi_state_cache = NULL;
1350 #endif
1351         uprobe_copy_process(p);
1352         /*
1353          * sigaltstack should be cleared when sharing the same VM
1354          */
1355         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1356                 p->sas_ss_sp = p->sas_ss_size = 0;
1357
1358         /*
1359          * Syscall tracing and stepping should be turned off in the
1360          * child regardless of CLONE_PTRACE.
1361          */
1362         user_disable_single_step(p);
1363         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1364 #ifdef TIF_SYSCALL_EMU
1365         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1366 #endif
1367         clear_all_latency_tracing(p);
1368
1369         /* ok, now we should be set up.. */
1370         if (clone_flags & CLONE_THREAD)
1371                 p->exit_signal = -1;
1372         else if (clone_flags & CLONE_PARENT)
1373                 p->exit_signal = current->group_leader->exit_signal;
1374         else
1375                 p->exit_signal = (clone_flags & CSIGNAL);
1376
1377         p->pdeath_signal = 0;
1378         p->exit_state = 0;
1379
1380         p->nr_dirtied = 0;
1381         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1382         p->dirty_paused_when = 0;
1383
1384         /*
1385          * Ok, make it visible to the rest of the system.
1386          * We dont wake it up yet.
1387          */
1388         p->group_leader = p;
1389         INIT_LIST_HEAD(&p->thread_group);
1390         p->task_works = NULL;
1391
1392         /* Now that the task is set up, run cgroup callbacks if
1393          * necessary. We need to run them before the task is visible
1394          * on the tasklist. */
1395         cgroup_fork_callbacks(p);
1396         cgroup_callbacks_done = 1;
1397
1398         /* Need tasklist lock for parent etc handling! */
1399         write_lock_irq(&tasklist_lock);
1400
1401         /* CLONE_PARENT re-uses the old parent */
1402         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1403                 p->real_parent = current->real_parent;
1404                 p->parent_exec_id = current->parent_exec_id;
1405         } else {
1406                 p->real_parent = current;
1407                 p->parent_exec_id = current->self_exec_id;
1408         }
1409
1410         spin_lock(&current->sighand->siglock);
1411
1412         /*
1413          * Process group and session signals need to be delivered to just the
1414          * parent before the fork or both the parent and the child after the
1415          * fork. Restart if a signal comes in before we add the new process to
1416          * it's process group.
1417          * A fatal signal pending means that current will exit, so the new
1418          * thread can't slip out of an OOM kill (or normal SIGKILL).
1419         */
1420         recalc_sigpending();
1421         if (signal_pending(current)) {
1422                 spin_unlock(&current->sighand->siglock);
1423                 write_unlock_irq(&tasklist_lock);
1424                 retval = -ERESTARTNOINTR;
1425                 goto bad_fork_free_pid;
1426         }
1427
1428         if (clone_flags & CLONE_THREAD) {
1429                 current->signal->nr_threads++;
1430                 atomic_inc(&current->signal->live);
1431                 atomic_inc(&current->signal->sigcnt);
1432                 p->group_leader = current->group_leader;
1433                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1434         }
1435
1436         if (likely(p->pid)) {
1437                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1438
1439                 if (thread_group_leader(p)) {
1440                         if (is_child_reaper(pid))
1441                                 p->nsproxy->pid_ns->child_reaper = p;
1442
1443                         p->signal->leader_pid = pid;
1444                         p->signal->tty = tty_kref_get(current->signal->tty);
1445                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1446                         attach_pid(p, PIDTYPE_SID, task_session(current));
1447                         list_add_tail(&p->sibling, &p->real_parent->children);
1448                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1449                         __this_cpu_inc(process_counts);
1450                 }
1451                 attach_pid(p, PIDTYPE_PID, pid);
1452                 nr_threads++;
1453         }
1454
1455         total_forks++;
1456         spin_unlock(&current->sighand->siglock);
1457         write_unlock_irq(&tasklist_lock);
1458         proc_fork_connector(p);
1459         cgroup_post_fork(p);
1460         if (clone_flags & CLONE_THREAD)
1461                 threadgroup_change_end(current);
1462         perf_event_fork(p);
1463
1464         trace_task_newtask(p, clone_flags);
1465
1466         return p;
1467
1468 bad_fork_free_pid:
1469         if (pid != &init_struct_pid)
1470                 free_pid(pid);
1471 bad_fork_cleanup_io:
1472         if (p->io_context)
1473                 exit_io_context(p);
1474 bad_fork_cleanup_namespaces:
1475         if (unlikely(clone_flags & CLONE_NEWPID))
1476                 pid_ns_release_proc(p->nsproxy->pid_ns);
1477         exit_task_namespaces(p);
1478 bad_fork_cleanup_mm:
1479         if (p->mm)
1480                 mmput(p->mm);
1481 bad_fork_cleanup_signal:
1482         if (!(clone_flags & CLONE_THREAD))
1483                 free_signal_struct(p->signal);
1484 bad_fork_cleanup_sighand:
1485         __cleanup_sighand(p->sighand);
1486 bad_fork_cleanup_fs:
1487         exit_fs(p); /* blocking */
1488 bad_fork_cleanup_files:
1489         exit_files(p); /* blocking */
1490 bad_fork_cleanup_semundo:
1491         exit_sem(p);
1492 bad_fork_cleanup_audit:
1493         audit_free(p);
1494 bad_fork_cleanup_policy:
1495         perf_event_free_task(p);
1496 #ifdef CONFIG_NUMA
1497         mpol_put(p->mempolicy);
1498 bad_fork_cleanup_cgroup:
1499 #endif
1500         if (clone_flags & CLONE_THREAD)
1501                 threadgroup_change_end(current);
1502         cgroup_exit(p, cgroup_callbacks_done);
1503         delayacct_tsk_free(p);
1504         module_put(task_thread_info(p)->exec_domain->module);
1505 bad_fork_cleanup_count:
1506         atomic_dec(&p->cred->user->processes);
1507         exit_creds(p);
1508 bad_fork_free:
1509         free_task(p);
1510 fork_out:
1511         return ERR_PTR(retval);
1512 }
1513
1514 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1515 {
1516         memset(regs, 0, sizeof(struct pt_regs));
1517         return regs;
1518 }
1519
1520 static inline void init_idle_pids(struct pid_link *links)
1521 {
1522         enum pid_type type;
1523
1524         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1525                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1526                 links[type].pid = &init_struct_pid;
1527         }
1528 }
1529
1530 struct task_struct * __cpuinit fork_idle(int cpu)
1531 {
1532         struct task_struct *task;
1533         struct pt_regs regs;
1534
1535         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1536                             &init_struct_pid, 0);
1537         if (!IS_ERR(task)) {
1538                 init_idle_pids(task->pids);
1539                 init_idle(task, cpu);
1540         }
1541
1542         return task;
1543 }
1544
1545 /*
1546  *  Ok, this is the main fork-routine.
1547  *
1548  * It copies the process, and if successful kick-starts
1549  * it and waits for it to finish using the VM if required.
1550  */
1551 long do_fork(unsigned long clone_flags,
1552               unsigned long stack_start,
1553               struct pt_regs *regs,
1554               unsigned long stack_size,
1555               int __user *parent_tidptr,
1556               int __user *child_tidptr)
1557 {
1558         struct task_struct *p;
1559         int trace = 0;
1560         long nr;
1561
1562         /*
1563          * Do some preliminary argument and permissions checking before we
1564          * actually start allocating stuff
1565          */
1566         if (clone_flags & CLONE_NEWUSER) {
1567                 if (clone_flags & CLONE_THREAD)
1568                         return -EINVAL;
1569                 /* hopefully this check will go away when userns support is
1570                  * complete
1571                  */
1572                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1573                                 !capable(CAP_SETGID))
1574                         return -EPERM;
1575         }
1576
1577         /*
1578          * Determine whether and which event to report to ptracer.  When
1579          * called from kernel_thread or CLONE_UNTRACED is explicitly
1580          * requested, no event is reported; otherwise, report if the event
1581          * for the type of forking is enabled.
1582          */
1583         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1584                 if (clone_flags & CLONE_VFORK)
1585                         trace = PTRACE_EVENT_VFORK;
1586                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1587                         trace = PTRACE_EVENT_CLONE;
1588                 else
1589                         trace = PTRACE_EVENT_FORK;
1590
1591                 if (likely(!ptrace_event_enabled(current, trace)))
1592                         trace = 0;
1593         }
1594
1595         p = copy_process(clone_flags, stack_start, regs, stack_size,
1596                          child_tidptr, NULL, trace);
1597         /*
1598          * Do this prior waking up the new thread - the thread pointer
1599          * might get invalid after that point, if the thread exits quickly.
1600          */
1601         if (!IS_ERR(p)) {
1602                 struct completion vfork;
1603
1604                 trace_sched_process_fork(current, p);
1605
1606                 nr = task_pid_vnr(p);
1607
1608                 if (clone_flags & CLONE_PARENT_SETTID)
1609                         put_user(nr, parent_tidptr);
1610
1611                 if (clone_flags & CLONE_VFORK) {
1612                         p->vfork_done = &vfork;
1613                         init_completion(&vfork);
1614                         get_task_struct(p);
1615                 }
1616
1617                 wake_up_new_task(p);
1618
1619                 /* forking complete and child started to run, tell ptracer */
1620                 if (unlikely(trace))
1621                         ptrace_event(trace, nr);
1622
1623                 if (clone_flags & CLONE_VFORK) {
1624                         if (!wait_for_vfork_done(p, &vfork))
1625                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1626                 }
1627         } else {
1628                 nr = PTR_ERR(p);
1629         }
1630         return nr;
1631 }
1632
1633 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1634 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1635 #endif
1636
1637 static void sighand_ctor(void *data)
1638 {
1639         struct sighand_struct *sighand = data;
1640
1641         spin_lock_init(&sighand->siglock);
1642         init_waitqueue_head(&sighand->signalfd_wqh);
1643 }
1644
1645 void __init proc_caches_init(void)
1646 {
1647         sighand_cachep = kmem_cache_create("sighand_cache",
1648                         sizeof(struct sighand_struct), 0,
1649                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1650                         SLAB_NOTRACK, sighand_ctor);
1651         signal_cachep = kmem_cache_create("signal_cache",
1652                         sizeof(struct signal_struct), 0,
1653                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1654         files_cachep = kmem_cache_create("files_cache",
1655                         sizeof(struct files_struct), 0,
1656                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1657         fs_cachep = kmem_cache_create("fs_cache",
1658                         sizeof(struct fs_struct), 0,
1659                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1660         /*
1661          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1662          * whole struct cpumask for the OFFSTACK case. We could change
1663          * this to *only* allocate as much of it as required by the
1664          * maximum number of CPU's we can ever have.  The cpumask_allocation
1665          * is at the end of the structure, exactly for that reason.
1666          */
1667         mm_cachep = kmem_cache_create("mm_struct",
1668                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1669                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1670         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1671         mmap_init();
1672         nsproxy_cache_init();
1673 }
1674
1675 /*
1676  * Check constraints on flags passed to the unshare system call.
1677  */
1678 static int check_unshare_flags(unsigned long unshare_flags)
1679 {
1680         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1681                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1682                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1683                 return -EINVAL;
1684         /*
1685          * Not implemented, but pretend it works if there is nothing to
1686          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1687          * needs to unshare vm.
1688          */
1689         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1690                 /* FIXME: get_task_mm() increments ->mm_users */
1691                 if (atomic_read(&current->mm->mm_users) > 1)
1692                         return -EINVAL;
1693         }
1694
1695         return 0;
1696 }
1697
1698 /*
1699  * Unshare the filesystem structure if it is being shared
1700  */
1701 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1702 {
1703         struct fs_struct *fs = current->fs;
1704
1705         if (!(unshare_flags & CLONE_FS) || !fs)
1706                 return 0;
1707
1708         /* don't need lock here; in the worst case we'll do useless copy */
1709         if (fs->users == 1)
1710                 return 0;
1711
1712         *new_fsp = copy_fs_struct(fs);
1713         if (!*new_fsp)
1714                 return -ENOMEM;
1715
1716         return 0;
1717 }
1718
1719 /*
1720  * Unshare file descriptor table if it is being shared
1721  */
1722 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1723 {
1724         struct files_struct *fd = current->files;
1725         int error = 0;
1726
1727         if ((unshare_flags & CLONE_FILES) &&
1728             (fd && atomic_read(&fd->count) > 1)) {
1729                 *new_fdp = dup_fd(fd, &error);
1730                 if (!*new_fdp)
1731                         return error;
1732         }
1733
1734         return 0;
1735 }
1736
1737 /*
1738  * unshare allows a process to 'unshare' part of the process
1739  * context which was originally shared using clone.  copy_*
1740  * functions used by do_fork() cannot be used here directly
1741  * because they modify an inactive task_struct that is being
1742  * constructed. Here we are modifying the current, active,
1743  * task_struct.
1744  */
1745 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1746 {
1747         struct fs_struct *fs, *new_fs = NULL;
1748         struct files_struct *fd, *new_fd = NULL;
1749         struct nsproxy *new_nsproxy = NULL;
1750         int do_sysvsem = 0;
1751         int err;
1752
1753         err = check_unshare_flags(unshare_flags);
1754         if (err)
1755                 goto bad_unshare_out;
1756
1757         /*
1758          * If unsharing namespace, must also unshare filesystem information.
1759          */
1760         if (unshare_flags & CLONE_NEWNS)
1761                 unshare_flags |= CLONE_FS;
1762         /*
1763          * CLONE_NEWIPC must also detach from the undolist: after switching
1764          * to a new ipc namespace, the semaphore arrays from the old
1765          * namespace are unreachable.
1766          */
1767         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1768                 do_sysvsem = 1;
1769         err = unshare_fs(unshare_flags, &new_fs);
1770         if (err)
1771                 goto bad_unshare_out;
1772         err = unshare_fd(unshare_flags, &new_fd);
1773         if (err)
1774                 goto bad_unshare_cleanup_fs;
1775         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1776         if (err)
1777                 goto bad_unshare_cleanup_fd;
1778
1779         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1780                 if (do_sysvsem) {
1781                         /*
1782                          * CLONE_SYSVSEM is equivalent to sys_exit().
1783                          */
1784                         exit_sem(current);
1785                 }
1786
1787                 if (new_nsproxy) {
1788                         switch_task_namespaces(current, new_nsproxy);
1789                         new_nsproxy = NULL;
1790                 }
1791
1792                 task_lock(current);
1793
1794                 if (new_fs) {
1795                         fs = current->fs;
1796                         spin_lock(&fs->lock);
1797                         current->fs = new_fs;
1798                         if (--fs->users)
1799                                 new_fs = NULL;
1800                         else
1801                                 new_fs = fs;
1802                         spin_unlock(&fs->lock);
1803                 }
1804
1805                 if (new_fd) {
1806                         fd = current->files;
1807                         current->files = new_fd;
1808                         new_fd = fd;
1809                 }
1810
1811                 task_unlock(current);
1812         }
1813
1814         if (new_nsproxy)
1815                 put_nsproxy(new_nsproxy);
1816
1817 bad_unshare_cleanup_fd:
1818         if (new_fd)
1819                 put_files_struct(new_fd);
1820
1821 bad_unshare_cleanup_fs:
1822         if (new_fs)
1823                 free_fs_struct(new_fs);
1824
1825 bad_unshare_out:
1826         return err;
1827 }
1828
1829 /*
1830  *      Helper to unshare the files of the current task.
1831  *      We don't want to expose copy_files internals to
1832  *      the exec layer of the kernel.
1833  */
1834
1835 int unshare_files(struct files_struct **displaced)
1836 {
1837         struct task_struct *task = current;
1838         struct files_struct *copy = NULL;
1839         int error;
1840
1841         error = unshare_fd(CLONE_FILES, &copy);
1842         if (error || !copy) {
1843                 *displaced = NULL;
1844                 return error;
1845         }
1846         *displaced = task->files;
1847         task_lock(task);
1848         task->files = copy;
1849         task_unlock(task);
1850         return 0;
1851 }