rcu: Use wrapper function instead of exporting tasklist_lock
[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/tracehook.h>
41 #include <linux/futex.h>
42 #include <linux/compat.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/proc_fs.h>
62 #include <linux/blkdev.h>
63 #include <linux/fs_struct.h>
64 #include <linux/magic.h>
65 #include <linux/perf_event.h>
66 #include <linux/posix-timers.h>
67 #include <linux/user-return-notifier.h>
68
69 #include <asm/pgtable.h>
70 #include <asm/pgalloc.h>
71 #include <asm/uaccess.h>
72 #include <asm/mmu_context.h>
73 #include <asm/cacheflush.h>
74 #include <asm/tlbflush.h>
75
76 #include <trace/events/sched.h>
77
78 /*
79  * Protected counters by write_lock_irq(&tasklist_lock)
80  */
81 unsigned long total_forks;      /* Handle normal Linux uptimes. */
82 int nr_threads;                 /* The idle threads do not count.. */
83
84 int max_threads;                /* tunable limit on nr_threads */
85
86 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
87
88 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
89
90 #ifdef CONFIG_PROVE_RCU
91 int lockdep_tasklist_lock_is_held(void)
92 {
93         return lockdep_is_held(&tasklist_lock);
94 }
95 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
96 #endif /* #ifdef CONFIG_PROVE_RCU */
97
98 int nr_processes(void)
99 {
100         int cpu;
101         int total = 0;
102
103         for_each_possible_cpu(cpu)
104                 total += per_cpu(process_counts, cpu);
105
106         return total;
107 }
108
109 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
110 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
111 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
112 static struct kmem_cache *task_struct_cachep;
113 #endif
114
115 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
116 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
117 {
118 #ifdef CONFIG_DEBUG_STACK_USAGE
119         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
120 #else
121         gfp_t mask = GFP_KERNEL;
122 #endif
123         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
124 }
125
126 static inline void free_thread_info(struct thread_info *ti)
127 {
128         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
129 }
130 #endif
131
132 /* SLAB cache for signal_struct structures (tsk->signal) */
133 static struct kmem_cache *signal_cachep;
134
135 /* SLAB cache for sighand_struct structures (tsk->sighand) */
136 struct kmem_cache *sighand_cachep;
137
138 /* SLAB cache for files_struct structures (tsk->files) */
139 struct kmem_cache *files_cachep;
140
141 /* SLAB cache for fs_struct structures (tsk->fs) */
142 struct kmem_cache *fs_cachep;
143
144 /* SLAB cache for vm_area_struct structures */
145 struct kmem_cache *vm_area_cachep;
146
147 /* SLAB cache for mm_struct structures (tsk->mm) */
148 static struct kmem_cache *mm_cachep;
149
150 static void account_kernel_stack(struct thread_info *ti, int account)
151 {
152         struct zone *zone = page_zone(virt_to_page(ti));
153
154         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
155 }
156
157 void free_task(struct task_struct *tsk)
158 {
159         prop_local_destroy_single(&tsk->dirties);
160         account_kernel_stack(tsk->stack, -1);
161         free_thread_info(tsk->stack);
162         rt_mutex_debug_task_free(tsk);
163         ftrace_graph_exit_task(tsk);
164         free_task_struct(tsk);
165 }
166 EXPORT_SYMBOL(free_task);
167
168 void __put_task_struct(struct task_struct *tsk)
169 {
170         WARN_ON(!tsk->exit_state);
171         WARN_ON(atomic_read(&tsk->usage));
172         WARN_ON(tsk == current);
173
174         exit_creds(tsk);
175         delayacct_tsk_free(tsk);
176
177         if (!profile_handoff_task(tsk))
178                 free_task(tsk);
179 }
180
181 /*
182  * macro override instead of weak attribute alias, to workaround
183  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
184  */
185 #ifndef arch_task_cache_init
186 #define arch_task_cache_init()
187 #endif
188
189 void __init fork_init(unsigned long mempages)
190 {
191 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
192 #ifndef ARCH_MIN_TASKALIGN
193 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
194 #endif
195         /* create a slab on which task_structs can be allocated */
196         task_struct_cachep =
197                 kmem_cache_create("task_struct", sizeof(struct task_struct),
198                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
199 #endif
200
201         /* do the arch specific task caches init */
202         arch_task_cache_init();
203
204         /*
205          * The default maximum number of threads is set to a safe
206          * value: the thread structures can take up at most half
207          * of memory.
208          */
209         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
210
211         /*
212          * we need to allow at least 20 threads to boot a system
213          */
214         if(max_threads < 20)
215                 max_threads = 20;
216
217         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
218         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
219         init_task.signal->rlim[RLIMIT_SIGPENDING] =
220                 init_task.signal->rlim[RLIMIT_NPROC];
221 }
222
223 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
224                                                struct task_struct *src)
225 {
226         *dst = *src;
227         return 0;
228 }
229
230 static struct task_struct *dup_task_struct(struct task_struct *orig)
231 {
232         struct task_struct *tsk;
233         struct thread_info *ti;
234         unsigned long *stackend;
235
236         int err;
237
238         prepare_to_copy(orig);
239
240         tsk = alloc_task_struct();
241         if (!tsk)
242                 return NULL;
243
244         ti = alloc_thread_info(tsk);
245         if (!ti) {
246                 free_task_struct(tsk);
247                 return NULL;
248         }
249
250         err = arch_dup_task_struct(tsk, orig);
251         if (err)
252                 goto out;
253
254         tsk->stack = ti;
255
256         err = prop_local_init_single(&tsk->dirties);
257         if (err)
258                 goto out;
259
260         setup_thread_stack(tsk, orig);
261         clear_user_return_notifier(tsk);
262         stackend = end_of_stack(tsk);
263         *stackend = STACK_END_MAGIC;    /* for overflow detection */
264
265 #ifdef CONFIG_CC_STACKPROTECTOR
266         tsk->stack_canary = get_random_int();
267 #endif
268
269         /* One for us, one for whoever does the "release_task()" (usually parent) */
270         atomic_set(&tsk->usage,2);
271         atomic_set(&tsk->fs_excl, 0);
272 #ifdef CONFIG_BLK_DEV_IO_TRACE
273         tsk->btrace_seq = 0;
274 #endif
275         tsk->splice_pipe = NULL;
276
277         account_kernel_stack(ti, 1);
278
279         return tsk;
280
281 out:
282         free_thread_info(ti);
283         free_task_struct(tsk);
284         return NULL;
285 }
286
287 #ifdef CONFIG_MMU
288 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
289 {
290         struct vm_area_struct *mpnt, *tmp, **pprev;
291         struct rb_node **rb_link, *rb_parent;
292         int retval;
293         unsigned long charge;
294         struct mempolicy *pol;
295
296         down_write(&oldmm->mmap_sem);
297         flush_cache_dup_mm(oldmm);
298         /*
299          * Not linked in yet - no deadlock potential:
300          */
301         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
302
303         mm->locked_vm = 0;
304         mm->mmap = NULL;
305         mm->mmap_cache = NULL;
306         mm->free_area_cache = oldmm->mmap_base;
307         mm->cached_hole_size = ~0UL;
308         mm->map_count = 0;
309         cpumask_clear(mm_cpumask(mm));
310         mm->mm_rb = RB_ROOT;
311         rb_link = &mm->mm_rb.rb_node;
312         rb_parent = NULL;
313         pprev = &mm->mmap;
314         retval = ksm_fork(mm, oldmm);
315         if (retval)
316                 goto out;
317
318         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
319                 struct file *file;
320
321                 if (mpnt->vm_flags & VM_DONTCOPY) {
322                         long pages = vma_pages(mpnt);
323                         mm->total_vm -= pages;
324                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
325                                                                 -pages);
326                         continue;
327                 }
328                 charge = 0;
329                 if (mpnt->vm_flags & VM_ACCOUNT) {
330                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
331                         if (security_vm_enough_memory(len))
332                                 goto fail_nomem;
333                         charge = len;
334                 }
335                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
336                 if (!tmp)
337                         goto fail_nomem;
338                 *tmp = *mpnt;
339                 pol = mpol_dup(vma_policy(mpnt));
340                 retval = PTR_ERR(pol);
341                 if (IS_ERR(pol))
342                         goto fail_nomem_policy;
343                 vma_set_policy(tmp, pol);
344                 tmp->vm_flags &= ~VM_LOCKED;
345                 tmp->vm_mm = mm;
346                 tmp->vm_next = NULL;
347                 anon_vma_link(tmp);
348                 file = tmp->vm_file;
349                 if (file) {
350                         struct inode *inode = file->f_path.dentry->d_inode;
351                         struct address_space *mapping = file->f_mapping;
352
353                         get_file(file);
354                         if (tmp->vm_flags & VM_DENYWRITE)
355                                 atomic_dec(&inode->i_writecount);
356                         spin_lock(&mapping->i_mmap_lock);
357                         if (tmp->vm_flags & VM_SHARED)
358                                 mapping->i_mmap_writable++;
359                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
360                         flush_dcache_mmap_lock(mapping);
361                         /* insert tmp into the share list, just after mpnt */
362                         vma_prio_tree_add(tmp, mpnt);
363                         flush_dcache_mmap_unlock(mapping);
364                         spin_unlock(&mapping->i_mmap_lock);
365                 }
366
367                 /*
368                  * Clear hugetlb-related page reserves for children. This only
369                  * affects MAP_PRIVATE mappings. Faults generated by the child
370                  * are not guaranteed to succeed, even if read-only
371                  */
372                 if (is_vm_hugetlb_page(tmp))
373                         reset_vma_resv_huge_pages(tmp);
374
375                 /*
376                  * Link in the new vma and copy the page table entries.
377                  */
378                 *pprev = tmp;
379                 pprev = &tmp->vm_next;
380
381                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
382                 rb_link = &tmp->vm_rb.rb_right;
383                 rb_parent = &tmp->vm_rb;
384
385                 mm->map_count++;
386                 retval = copy_page_range(mm, oldmm, mpnt);
387
388                 if (tmp->vm_ops && tmp->vm_ops->open)
389                         tmp->vm_ops->open(tmp);
390
391                 if (retval)
392                         goto out;
393         }
394         /* a new mm has just been created */
395         arch_dup_mmap(oldmm, mm);
396         retval = 0;
397 out:
398         up_write(&mm->mmap_sem);
399         flush_tlb_mm(oldmm);
400         up_write(&oldmm->mmap_sem);
401         return retval;
402 fail_nomem_policy:
403         kmem_cache_free(vm_area_cachep, tmp);
404 fail_nomem:
405         retval = -ENOMEM;
406         vm_unacct_memory(charge);
407         goto out;
408 }
409
410 static inline int mm_alloc_pgd(struct mm_struct * mm)
411 {
412         mm->pgd = pgd_alloc(mm);
413         if (unlikely(!mm->pgd))
414                 return -ENOMEM;
415         return 0;
416 }
417
418 static inline void mm_free_pgd(struct mm_struct * mm)
419 {
420         pgd_free(mm, mm->pgd);
421 }
422 #else
423 #define dup_mmap(mm, oldmm)     (0)
424 #define mm_alloc_pgd(mm)        (0)
425 #define mm_free_pgd(mm)
426 #endif /* CONFIG_MMU */
427
428 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
429
430 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
431 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
432
433 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
434
435 static int __init coredump_filter_setup(char *s)
436 {
437         default_dump_filter =
438                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
439                 MMF_DUMP_FILTER_MASK;
440         return 1;
441 }
442
443 __setup("coredump_filter=", coredump_filter_setup);
444
445 #include <linux/init_task.h>
446
447 static void mm_init_aio(struct mm_struct *mm)
448 {
449 #ifdef CONFIG_AIO
450         spin_lock_init(&mm->ioctx_lock);
451         INIT_HLIST_HEAD(&mm->ioctx_list);
452 #endif
453 }
454
455 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
456 {
457         atomic_set(&mm->mm_users, 1);
458         atomic_set(&mm->mm_count, 1);
459         init_rwsem(&mm->mmap_sem);
460         INIT_LIST_HEAD(&mm->mmlist);
461         mm->flags = (current->mm) ?
462                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
463         mm->core_state = NULL;
464         mm->nr_ptes = 0;
465         set_mm_counter(mm, file_rss, 0);
466         set_mm_counter(mm, anon_rss, 0);
467         spin_lock_init(&mm->page_table_lock);
468         mm->free_area_cache = TASK_UNMAPPED_BASE;
469         mm->cached_hole_size = ~0UL;
470         mm_init_aio(mm);
471         mm_init_owner(mm, p);
472
473         if (likely(!mm_alloc_pgd(mm))) {
474                 mm->def_flags = 0;
475                 mmu_notifier_mm_init(mm);
476                 return mm;
477         }
478
479         free_mm(mm);
480         return NULL;
481 }
482
483 /*
484  * Allocate and initialize an mm_struct.
485  */
486 struct mm_struct * mm_alloc(void)
487 {
488         struct mm_struct * mm;
489
490         mm = allocate_mm();
491         if (mm) {
492                 memset(mm, 0, sizeof(*mm));
493                 mm = mm_init(mm, current);
494         }
495         return mm;
496 }
497
498 /*
499  * Called when the last reference to the mm
500  * is dropped: either by a lazy thread or by
501  * mmput. Free the page directory and the mm.
502  */
503 void __mmdrop(struct mm_struct *mm)
504 {
505         BUG_ON(mm == &init_mm);
506         mm_free_pgd(mm);
507         destroy_context(mm);
508         mmu_notifier_mm_destroy(mm);
509         free_mm(mm);
510 }
511 EXPORT_SYMBOL_GPL(__mmdrop);
512
513 /*
514  * Decrement the use count and release all resources for an mm.
515  */
516 void mmput(struct mm_struct *mm)
517 {
518         might_sleep();
519
520         if (atomic_dec_and_test(&mm->mm_users)) {
521                 exit_aio(mm);
522                 ksm_exit(mm);
523                 exit_mmap(mm);
524                 set_mm_exe_file(mm, NULL);
525                 if (!list_empty(&mm->mmlist)) {
526                         spin_lock(&mmlist_lock);
527                         list_del(&mm->mmlist);
528                         spin_unlock(&mmlist_lock);
529                 }
530                 put_swap_token(mm);
531                 if (mm->binfmt)
532                         module_put(mm->binfmt->module);
533                 mmdrop(mm);
534         }
535 }
536 EXPORT_SYMBOL_GPL(mmput);
537
538 /**
539  * get_task_mm - acquire a reference to the task's mm
540  *
541  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
542  * this kernel workthread has transiently adopted a user mm with use_mm,
543  * to do its AIO) is not set and if so returns a reference to it, after
544  * bumping up the use count.  User must release the mm via mmput()
545  * after use.  Typically used by /proc and ptrace.
546  */
547 struct mm_struct *get_task_mm(struct task_struct *task)
548 {
549         struct mm_struct *mm;
550
551         task_lock(task);
552         mm = task->mm;
553         if (mm) {
554                 if (task->flags & PF_KTHREAD)
555                         mm = NULL;
556                 else
557                         atomic_inc(&mm->mm_users);
558         }
559         task_unlock(task);
560         return mm;
561 }
562 EXPORT_SYMBOL_GPL(get_task_mm);
563
564 /* Please note the differences between mmput and mm_release.
565  * mmput is called whenever we stop holding onto a mm_struct,
566  * error success whatever.
567  *
568  * mm_release is called after a mm_struct has been removed
569  * from the current process.
570  *
571  * This difference is important for error handling, when we
572  * only half set up a mm_struct for a new process and need to restore
573  * the old one.  Because we mmput the new mm_struct before
574  * restoring the old one. . .
575  * Eric Biederman 10 January 1998
576  */
577 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
578 {
579         struct completion *vfork_done = tsk->vfork_done;
580
581         /* Get rid of any futexes when releasing the mm */
582 #ifdef CONFIG_FUTEX
583         if (unlikely(tsk->robust_list)) {
584                 exit_robust_list(tsk);
585                 tsk->robust_list = NULL;
586         }
587 #ifdef CONFIG_COMPAT
588         if (unlikely(tsk->compat_robust_list)) {
589                 compat_exit_robust_list(tsk);
590                 tsk->compat_robust_list = NULL;
591         }
592 #endif
593         if (unlikely(!list_empty(&tsk->pi_state_list)))
594                 exit_pi_state_list(tsk);
595 #endif
596
597         /* Get rid of any cached register state */
598         deactivate_mm(tsk, mm);
599
600         /* notify parent sleeping on vfork() */
601         if (vfork_done) {
602                 tsk->vfork_done = NULL;
603                 complete(vfork_done);
604         }
605
606         /*
607          * If we're exiting normally, clear a user-space tid field if
608          * requested.  We leave this alone when dying by signal, to leave
609          * the value intact in a core dump, and to save the unnecessary
610          * trouble otherwise.  Userland only wants this done for a sys_exit.
611          */
612         if (tsk->clear_child_tid) {
613                 if (!(tsk->flags & PF_SIGNALED) &&
614                     atomic_read(&mm->mm_users) > 1) {
615                         /*
616                          * We don't check the error code - if userspace has
617                          * not set up a proper pointer then tough luck.
618                          */
619                         put_user(0, tsk->clear_child_tid);
620                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
621                                         1, NULL, NULL, 0);
622                 }
623                 tsk->clear_child_tid = NULL;
624         }
625 }
626
627 /*
628  * Allocate a new mm structure and copy contents from the
629  * mm structure of the passed in task structure.
630  */
631 struct mm_struct *dup_mm(struct task_struct *tsk)
632 {
633         struct mm_struct *mm, *oldmm = current->mm;
634         int err;
635
636         if (!oldmm)
637                 return NULL;
638
639         mm = allocate_mm();
640         if (!mm)
641                 goto fail_nomem;
642
643         memcpy(mm, oldmm, sizeof(*mm));
644
645         /* Initializing for Swap token stuff */
646         mm->token_priority = 0;
647         mm->last_interval = 0;
648
649         if (!mm_init(mm, tsk))
650                 goto fail_nomem;
651
652         if (init_new_context(tsk, mm))
653                 goto fail_nocontext;
654
655         dup_mm_exe_file(oldmm, mm);
656
657         err = dup_mmap(mm, oldmm);
658         if (err)
659                 goto free_pt;
660
661         mm->hiwater_rss = get_mm_rss(mm);
662         mm->hiwater_vm = mm->total_vm;
663
664         if (mm->binfmt && !try_module_get(mm->binfmt->module))
665                 goto free_pt;
666
667         return mm;
668
669 free_pt:
670         /* don't put binfmt in mmput, we haven't got module yet */
671         mm->binfmt = NULL;
672         mmput(mm);
673
674 fail_nomem:
675         return NULL;
676
677 fail_nocontext:
678         /*
679          * If init_new_context() failed, we cannot use mmput() to free the mm
680          * because it calls destroy_context()
681          */
682         mm_free_pgd(mm);
683         free_mm(mm);
684         return NULL;
685 }
686
687 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
688 {
689         struct mm_struct * mm, *oldmm;
690         int retval;
691
692         tsk->min_flt = tsk->maj_flt = 0;
693         tsk->nvcsw = tsk->nivcsw = 0;
694 #ifdef CONFIG_DETECT_HUNG_TASK
695         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
696 #endif
697
698         tsk->mm = NULL;
699         tsk->active_mm = NULL;
700
701         /*
702          * Are we cloning a kernel thread?
703          *
704          * We need to steal a active VM for that..
705          */
706         oldmm = current->mm;
707         if (!oldmm)
708                 return 0;
709
710         if (clone_flags & CLONE_VM) {
711                 atomic_inc(&oldmm->mm_users);
712                 mm = oldmm;
713                 goto good_mm;
714         }
715
716         retval = -ENOMEM;
717         mm = dup_mm(tsk);
718         if (!mm)
719                 goto fail_nomem;
720
721 good_mm:
722         /* Initializing for Swap token stuff */
723         mm->token_priority = 0;
724         mm->last_interval = 0;
725
726         tsk->mm = mm;
727         tsk->active_mm = mm;
728         return 0;
729
730 fail_nomem:
731         return retval;
732 }
733
734 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
735 {
736         struct fs_struct *fs = current->fs;
737         if (clone_flags & CLONE_FS) {
738                 /* tsk->fs is already what we want */
739                 write_lock(&fs->lock);
740                 if (fs->in_exec) {
741                         write_unlock(&fs->lock);
742                         return -EAGAIN;
743                 }
744                 fs->users++;
745                 write_unlock(&fs->lock);
746                 return 0;
747         }
748         tsk->fs = copy_fs_struct(fs);
749         if (!tsk->fs)
750                 return -ENOMEM;
751         return 0;
752 }
753
754 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
755 {
756         struct files_struct *oldf, *newf;
757         int error = 0;
758
759         /*
760          * A background process may not have any files ...
761          */
762         oldf = current->files;
763         if (!oldf)
764                 goto out;
765
766         if (clone_flags & CLONE_FILES) {
767                 atomic_inc(&oldf->count);
768                 goto out;
769         }
770
771         newf = dup_fd(oldf, &error);
772         if (!newf)
773                 goto out;
774
775         tsk->files = newf;
776         error = 0;
777 out:
778         return error;
779 }
780
781 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
782 {
783 #ifdef CONFIG_BLOCK
784         struct io_context *ioc = current->io_context;
785
786         if (!ioc)
787                 return 0;
788         /*
789          * Share io context with parent, if CLONE_IO is set
790          */
791         if (clone_flags & CLONE_IO) {
792                 tsk->io_context = ioc_task_link(ioc);
793                 if (unlikely(!tsk->io_context))
794                         return -ENOMEM;
795         } else if (ioprio_valid(ioc->ioprio)) {
796                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
797                 if (unlikely(!tsk->io_context))
798                         return -ENOMEM;
799
800                 tsk->io_context->ioprio = ioc->ioprio;
801         }
802 #endif
803         return 0;
804 }
805
806 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
807 {
808         struct sighand_struct *sig;
809
810         if (clone_flags & CLONE_SIGHAND) {
811                 atomic_inc(&current->sighand->count);
812                 return 0;
813         }
814         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
815         rcu_assign_pointer(tsk->sighand, sig);
816         if (!sig)
817                 return -ENOMEM;
818         atomic_set(&sig->count, 1);
819         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
820         return 0;
821 }
822
823 void __cleanup_sighand(struct sighand_struct *sighand)
824 {
825         if (atomic_dec_and_test(&sighand->count))
826                 kmem_cache_free(sighand_cachep, sighand);
827 }
828
829
830 /*
831  * Initialize POSIX timer handling for a thread group.
832  */
833 static void posix_cpu_timers_init_group(struct signal_struct *sig)
834 {
835         /* Thread group counters. */
836         thread_group_cputime_init(sig);
837
838         /* Expiration times and increments. */
839         sig->it[CPUCLOCK_PROF].expires = cputime_zero;
840         sig->it[CPUCLOCK_PROF].incr = cputime_zero;
841         sig->it[CPUCLOCK_VIRT].expires = cputime_zero;
842         sig->it[CPUCLOCK_VIRT].incr = cputime_zero;
843
844         /* Cached expiration times. */
845         sig->cputime_expires.prof_exp = cputime_zero;
846         sig->cputime_expires.virt_exp = cputime_zero;
847         sig->cputime_expires.sched_exp = 0;
848
849         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
850                 sig->cputime_expires.prof_exp =
851                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
852                 sig->cputimer.running = 1;
853         }
854
855         /* The timer lists. */
856         INIT_LIST_HEAD(&sig->cpu_timers[0]);
857         INIT_LIST_HEAD(&sig->cpu_timers[1]);
858         INIT_LIST_HEAD(&sig->cpu_timers[2]);
859 }
860
861 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
862 {
863         struct signal_struct *sig;
864
865         if (clone_flags & CLONE_THREAD)
866                 return 0;
867
868         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
869         tsk->signal = sig;
870         if (!sig)
871                 return -ENOMEM;
872
873         atomic_set(&sig->count, 1);
874         atomic_set(&sig->live, 1);
875         init_waitqueue_head(&sig->wait_chldexit);
876         sig->flags = 0;
877         if (clone_flags & CLONE_NEWPID)
878                 sig->flags |= SIGNAL_UNKILLABLE;
879         sig->group_exit_code = 0;
880         sig->group_exit_task = NULL;
881         sig->group_stop_count = 0;
882         sig->curr_target = tsk;
883         init_sigpending(&sig->shared_pending);
884         INIT_LIST_HEAD(&sig->posix_timers);
885
886         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
887         sig->it_real_incr.tv64 = 0;
888         sig->real_timer.function = it_real_fn;
889
890         sig->leader = 0;        /* session leadership doesn't inherit */
891         sig->tty_old_pgrp = NULL;
892         sig->tty = NULL;
893
894         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
895         sig->gtime = cputime_zero;
896         sig->cgtime = cputime_zero;
897 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
898         sig->prev_utime = sig->prev_stime = cputime_zero;
899 #endif
900         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
901         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
902         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
903         sig->maxrss = sig->cmaxrss = 0;
904         task_io_accounting_init(&sig->ioac);
905         sig->sum_sched_runtime = 0;
906         taskstats_tgid_init(sig);
907
908         task_lock(current->group_leader);
909         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
910         task_unlock(current->group_leader);
911
912         posix_cpu_timers_init_group(sig);
913
914         acct_init_pacct(&sig->pacct);
915
916         tty_audit_fork(sig);
917
918         sig->oom_adj = current->signal->oom_adj;
919
920         return 0;
921 }
922
923 void __cleanup_signal(struct signal_struct *sig)
924 {
925         thread_group_cputime_free(sig);
926         tty_kref_put(sig->tty);
927         kmem_cache_free(signal_cachep, sig);
928 }
929
930 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
931 {
932         unsigned long new_flags = p->flags;
933
934         new_flags &= ~PF_SUPERPRIV;
935         new_flags |= PF_FORKNOEXEC;
936         new_flags |= PF_STARTING;
937         p->flags = new_flags;
938         clear_freeze_flag(p);
939 }
940
941 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
942 {
943         current->clear_child_tid = tidptr;
944
945         return task_pid_vnr(current);
946 }
947
948 static void rt_mutex_init_task(struct task_struct *p)
949 {
950         raw_spin_lock_init(&p->pi_lock);
951 #ifdef CONFIG_RT_MUTEXES
952         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
953         p->pi_blocked_on = NULL;
954 #endif
955 }
956
957 #ifdef CONFIG_MM_OWNER
958 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
959 {
960         mm->owner = p;
961 }
962 #endif /* CONFIG_MM_OWNER */
963
964 /*
965  * Initialize POSIX timer handling for a single task.
966  */
967 static void posix_cpu_timers_init(struct task_struct *tsk)
968 {
969         tsk->cputime_expires.prof_exp = cputime_zero;
970         tsk->cputime_expires.virt_exp = cputime_zero;
971         tsk->cputime_expires.sched_exp = 0;
972         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
973         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
974         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
975 }
976
977 /*
978  * This creates a new process as a copy of the old one,
979  * but does not actually start it yet.
980  *
981  * It copies the registers, and all the appropriate
982  * parts of the process environment (as per the clone
983  * flags). The actual kick-off is left to the caller.
984  */
985 static struct task_struct *copy_process(unsigned long clone_flags,
986                                         unsigned long stack_start,
987                                         struct pt_regs *regs,
988                                         unsigned long stack_size,
989                                         int __user *child_tidptr,
990                                         struct pid *pid,
991                                         int trace)
992 {
993         int retval;
994         struct task_struct *p;
995         int cgroup_callbacks_done = 0;
996
997         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
998                 return ERR_PTR(-EINVAL);
999
1000         /*
1001          * Thread groups must share signals as well, and detached threads
1002          * can only be started up within the thread group.
1003          */
1004         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1005                 return ERR_PTR(-EINVAL);
1006
1007         /*
1008          * Shared signal handlers imply shared VM. By way of the above,
1009          * thread groups also imply shared VM. Blocking this case allows
1010          * for various simplifications in other code.
1011          */
1012         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1013                 return ERR_PTR(-EINVAL);
1014
1015         /*
1016          * Siblings of global init remain as zombies on exit since they are
1017          * not reaped by their parent (swapper). To solve this and to avoid
1018          * multi-rooted process trees, prevent global and container-inits
1019          * from creating siblings.
1020          */
1021         if ((clone_flags & CLONE_PARENT) &&
1022                                 current->signal->flags & SIGNAL_UNKILLABLE)
1023                 return ERR_PTR(-EINVAL);
1024
1025         retval = security_task_create(clone_flags);
1026         if (retval)
1027                 goto fork_out;
1028
1029         retval = -ENOMEM;
1030         p = dup_task_struct(current);
1031         if (!p)
1032                 goto fork_out;
1033
1034         ftrace_graph_init_task(p);
1035
1036         rt_mutex_init_task(p);
1037
1038 #ifdef CONFIG_PROVE_LOCKING
1039         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1040         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1041 #endif
1042         retval = -EAGAIN;
1043         if (atomic_read(&p->real_cred->user->processes) >=
1044                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1045                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1046                     p->real_cred->user != INIT_USER)
1047                         goto bad_fork_free;
1048         }
1049
1050         retval = copy_creds(p, clone_flags);
1051         if (retval < 0)
1052                 goto bad_fork_free;
1053
1054         /*
1055          * If multiple threads are within copy_process(), then this check
1056          * triggers too late. This doesn't hurt, the check is only there
1057          * to stop root fork bombs.
1058          */
1059         retval = -EAGAIN;
1060         if (nr_threads >= max_threads)
1061                 goto bad_fork_cleanup_count;
1062
1063         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1064                 goto bad_fork_cleanup_count;
1065
1066         p->did_exec = 0;
1067         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1068         copy_flags(clone_flags, p);
1069         INIT_LIST_HEAD(&p->children);
1070         INIT_LIST_HEAD(&p->sibling);
1071         rcu_copy_process(p);
1072         p->vfork_done = NULL;
1073         spin_lock_init(&p->alloc_lock);
1074
1075         init_sigpending(&p->pending);
1076
1077         p->utime = cputime_zero;
1078         p->stime = cputime_zero;
1079         p->gtime = cputime_zero;
1080         p->utimescaled = cputime_zero;
1081         p->stimescaled = cputime_zero;
1082 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1083         p->prev_utime = cputime_zero;
1084         p->prev_stime = cputime_zero;
1085 #endif
1086
1087         p->default_timer_slack_ns = current->timer_slack_ns;
1088
1089         task_io_accounting_init(&p->ioac);
1090         acct_clear_integrals(p);
1091
1092         posix_cpu_timers_init(p);
1093
1094         p->lock_depth = -1;             /* -1 = no lock */
1095         do_posix_clock_monotonic_gettime(&p->start_time);
1096         p->real_start_time = p->start_time;
1097         monotonic_to_bootbased(&p->real_start_time);
1098         p->io_context = NULL;
1099         p->audit_context = NULL;
1100         cgroup_fork(p);
1101 #ifdef CONFIG_NUMA
1102         p->mempolicy = mpol_dup(p->mempolicy);
1103         if (IS_ERR(p->mempolicy)) {
1104                 retval = PTR_ERR(p->mempolicy);
1105                 p->mempolicy = NULL;
1106                 goto bad_fork_cleanup_cgroup;
1107         }
1108         mpol_fix_fork_child_flag(p);
1109 #endif
1110 #ifdef CONFIG_TRACE_IRQFLAGS
1111         p->irq_events = 0;
1112 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1113         p->hardirqs_enabled = 1;
1114 #else
1115         p->hardirqs_enabled = 0;
1116 #endif
1117         p->hardirq_enable_ip = 0;
1118         p->hardirq_enable_event = 0;
1119         p->hardirq_disable_ip = _THIS_IP_;
1120         p->hardirq_disable_event = 0;
1121         p->softirqs_enabled = 1;
1122         p->softirq_enable_ip = _THIS_IP_;
1123         p->softirq_enable_event = 0;
1124         p->softirq_disable_ip = 0;
1125         p->softirq_disable_event = 0;
1126         p->hardirq_context = 0;
1127         p->softirq_context = 0;
1128 #endif
1129 #ifdef CONFIG_LOCKDEP
1130         p->lockdep_depth = 0; /* no locks held yet */
1131         p->curr_chain_key = 0;
1132         p->lockdep_recursion = 0;
1133 #endif
1134
1135 #ifdef CONFIG_DEBUG_MUTEXES
1136         p->blocked_on = NULL; /* not blocked yet */
1137 #endif
1138 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1139         p->memcg_batch.do_batch = 0;
1140         p->memcg_batch.memcg = NULL;
1141 #endif
1142
1143         p->bts = NULL;
1144
1145         p->stack_start = stack_start;
1146
1147         /* Perform scheduler related setup. Assign this task to a CPU. */
1148         sched_fork(p, clone_flags);
1149
1150         retval = perf_event_init_task(p);
1151         if (retval)
1152                 goto bad_fork_cleanup_policy;
1153
1154         if ((retval = audit_alloc(p)))
1155                 goto bad_fork_cleanup_policy;
1156         /* copy all the process information */
1157         if ((retval = copy_semundo(clone_flags, p)))
1158                 goto bad_fork_cleanup_audit;
1159         if ((retval = copy_files(clone_flags, p)))
1160                 goto bad_fork_cleanup_semundo;
1161         if ((retval = copy_fs(clone_flags, p)))
1162                 goto bad_fork_cleanup_files;
1163         if ((retval = copy_sighand(clone_flags, p)))
1164                 goto bad_fork_cleanup_fs;
1165         if ((retval = copy_signal(clone_flags, p)))
1166                 goto bad_fork_cleanup_sighand;
1167         if ((retval = copy_mm(clone_flags, p)))
1168                 goto bad_fork_cleanup_signal;
1169         if ((retval = copy_namespaces(clone_flags, p)))
1170                 goto bad_fork_cleanup_mm;
1171         if ((retval = copy_io(clone_flags, p)))
1172                 goto bad_fork_cleanup_namespaces;
1173         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1174         if (retval)
1175                 goto bad_fork_cleanup_io;
1176
1177         if (pid != &init_struct_pid) {
1178                 retval = -ENOMEM;
1179                 pid = alloc_pid(p->nsproxy->pid_ns);
1180                 if (!pid)
1181                         goto bad_fork_cleanup_io;
1182
1183                 if (clone_flags & CLONE_NEWPID) {
1184                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1185                         if (retval < 0)
1186                                 goto bad_fork_free_pid;
1187                 }
1188         }
1189
1190         p->pid = pid_nr(pid);
1191         p->tgid = p->pid;
1192         if (clone_flags & CLONE_THREAD)
1193                 p->tgid = current->tgid;
1194
1195         if (current->nsproxy != p->nsproxy) {
1196                 retval = ns_cgroup_clone(p, pid);
1197                 if (retval)
1198                         goto bad_fork_free_pid;
1199         }
1200
1201         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1202         /*
1203          * Clear TID on mm_release()?
1204          */
1205         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1206 #ifdef CONFIG_FUTEX
1207         p->robust_list = NULL;
1208 #ifdef CONFIG_COMPAT
1209         p->compat_robust_list = NULL;
1210 #endif
1211         INIT_LIST_HEAD(&p->pi_state_list);
1212         p->pi_state_cache = NULL;
1213 #endif
1214         /*
1215          * sigaltstack should be cleared when sharing the same VM
1216          */
1217         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1218                 p->sas_ss_sp = p->sas_ss_size = 0;
1219
1220         /*
1221          * Syscall tracing and stepping should be turned off in the
1222          * child regardless of CLONE_PTRACE.
1223          */
1224         user_disable_single_step(p);
1225         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1226 #ifdef TIF_SYSCALL_EMU
1227         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1228 #endif
1229         clear_all_latency_tracing(p);
1230
1231         /* ok, now we should be set up.. */
1232         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1233         p->pdeath_signal = 0;
1234         p->exit_state = 0;
1235
1236         /*
1237          * Ok, make it visible to the rest of the system.
1238          * We dont wake it up yet.
1239          */
1240         p->group_leader = p;
1241         INIT_LIST_HEAD(&p->thread_group);
1242
1243         /* Now that the task is set up, run cgroup callbacks if
1244          * necessary. We need to run them before the task is visible
1245          * on the tasklist. */
1246         cgroup_fork_callbacks(p);
1247         cgroup_callbacks_done = 1;
1248
1249         /* Need tasklist lock for parent etc handling! */
1250         write_lock_irq(&tasklist_lock);
1251
1252         /* CLONE_PARENT re-uses the old parent */
1253         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1254                 p->real_parent = current->real_parent;
1255                 p->parent_exec_id = current->parent_exec_id;
1256         } else {
1257                 p->real_parent = current;
1258                 p->parent_exec_id = current->self_exec_id;
1259         }
1260
1261         spin_lock(&current->sighand->siglock);
1262
1263         /*
1264          * Process group and session signals need to be delivered to just the
1265          * parent before the fork or both the parent and the child after the
1266          * fork. Restart if a signal comes in before we add the new process to
1267          * it's process group.
1268          * A fatal signal pending means that current will exit, so the new
1269          * thread can't slip out of an OOM kill (or normal SIGKILL).
1270          */
1271         recalc_sigpending();
1272         if (signal_pending(current)) {
1273                 spin_unlock(&current->sighand->siglock);
1274                 write_unlock_irq(&tasklist_lock);
1275                 retval = -ERESTARTNOINTR;
1276                 goto bad_fork_free_pid;
1277         }
1278
1279         if (clone_flags & CLONE_THREAD) {
1280                 atomic_inc(&current->signal->count);
1281                 atomic_inc(&current->signal->live);
1282                 p->group_leader = current->group_leader;
1283                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1284         }
1285
1286         if (likely(p->pid)) {
1287                 tracehook_finish_clone(p, clone_flags, trace);
1288
1289                 if (thread_group_leader(p)) {
1290                         if (clone_flags & CLONE_NEWPID)
1291                                 p->nsproxy->pid_ns->child_reaper = p;
1292
1293                         p->signal->leader_pid = pid;
1294                         tty_kref_put(p->signal->tty);
1295                         p->signal->tty = tty_kref_get(current->signal->tty);
1296                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1297                         attach_pid(p, PIDTYPE_SID, task_session(current));
1298                         list_add_tail(&p->sibling, &p->real_parent->children);
1299                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1300                         __get_cpu_var(process_counts)++;
1301                 }
1302                 attach_pid(p, PIDTYPE_PID, pid);
1303                 nr_threads++;
1304         }
1305
1306         total_forks++;
1307         spin_unlock(&current->sighand->siglock);
1308         write_unlock_irq(&tasklist_lock);
1309         proc_fork_connector(p);
1310         cgroup_post_fork(p);
1311         perf_event_fork(p);
1312         return p;
1313
1314 bad_fork_free_pid:
1315         if (pid != &init_struct_pid)
1316                 free_pid(pid);
1317 bad_fork_cleanup_io:
1318         if (p->io_context)
1319                 exit_io_context(p);
1320 bad_fork_cleanup_namespaces:
1321         exit_task_namespaces(p);
1322 bad_fork_cleanup_mm:
1323         if (p->mm)
1324                 mmput(p->mm);
1325 bad_fork_cleanup_signal:
1326         if (!(clone_flags & CLONE_THREAD))
1327                 __cleanup_signal(p->signal);
1328 bad_fork_cleanup_sighand:
1329         __cleanup_sighand(p->sighand);
1330 bad_fork_cleanup_fs:
1331         exit_fs(p); /* blocking */
1332 bad_fork_cleanup_files:
1333         exit_files(p); /* blocking */
1334 bad_fork_cleanup_semundo:
1335         exit_sem(p);
1336 bad_fork_cleanup_audit:
1337         audit_free(p);
1338 bad_fork_cleanup_policy:
1339         perf_event_free_task(p);
1340 #ifdef CONFIG_NUMA
1341         mpol_put(p->mempolicy);
1342 bad_fork_cleanup_cgroup:
1343 #endif
1344         cgroup_exit(p, cgroup_callbacks_done);
1345         delayacct_tsk_free(p);
1346         module_put(task_thread_info(p)->exec_domain->module);
1347 bad_fork_cleanup_count:
1348         atomic_dec(&p->cred->user->processes);
1349         exit_creds(p);
1350 bad_fork_free:
1351         free_task(p);
1352 fork_out:
1353         return ERR_PTR(retval);
1354 }
1355
1356 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1357 {
1358         memset(regs, 0, sizeof(struct pt_regs));
1359         return regs;
1360 }
1361
1362 struct task_struct * __cpuinit fork_idle(int cpu)
1363 {
1364         struct task_struct *task;
1365         struct pt_regs regs;
1366
1367         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1368                             &init_struct_pid, 0);
1369         if (!IS_ERR(task))
1370                 init_idle(task, cpu);
1371
1372         return task;
1373 }
1374
1375 /*
1376  *  Ok, this is the main fork-routine.
1377  *
1378  * It copies the process, and if successful kick-starts
1379  * it and waits for it to finish using the VM if required.
1380  */
1381 long do_fork(unsigned long clone_flags,
1382               unsigned long stack_start,
1383               struct pt_regs *regs,
1384               unsigned long stack_size,
1385               int __user *parent_tidptr,
1386               int __user *child_tidptr)
1387 {
1388         struct task_struct *p;
1389         int trace = 0;
1390         long nr;
1391
1392         /*
1393          * Do some preliminary argument and permissions checking before we
1394          * actually start allocating stuff
1395          */
1396         if (clone_flags & CLONE_NEWUSER) {
1397                 if (clone_flags & CLONE_THREAD)
1398                         return -EINVAL;
1399                 /* hopefully this check will go away when userns support is
1400                  * complete
1401                  */
1402                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1403                                 !capable(CAP_SETGID))
1404                         return -EPERM;
1405         }
1406
1407         /*
1408          * We hope to recycle these flags after 2.6.26
1409          */
1410         if (unlikely(clone_flags & CLONE_STOPPED)) {
1411                 static int __read_mostly count = 100;
1412
1413                 if (count > 0 && printk_ratelimit()) {
1414                         char comm[TASK_COMM_LEN];
1415
1416                         count--;
1417                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1418                                         "clone flags 0x%lx\n",
1419                                 get_task_comm(comm, current),
1420                                 clone_flags & CLONE_STOPPED);
1421                 }
1422         }
1423
1424         /*
1425          * When called from kernel_thread, don't do user tracing stuff.
1426          */
1427         if (likely(user_mode(regs)))
1428                 trace = tracehook_prepare_clone(clone_flags);
1429
1430         p = copy_process(clone_flags, stack_start, regs, stack_size,
1431                          child_tidptr, NULL, trace);
1432         /*
1433          * Do this prior waking up the new thread - the thread pointer
1434          * might get invalid after that point, if the thread exits quickly.
1435          */
1436         if (!IS_ERR(p)) {
1437                 struct completion vfork;
1438
1439                 trace_sched_process_fork(current, p);
1440
1441                 nr = task_pid_vnr(p);
1442
1443                 if (clone_flags & CLONE_PARENT_SETTID)
1444                         put_user(nr, parent_tidptr);
1445
1446                 if (clone_flags & CLONE_VFORK) {
1447                         p->vfork_done = &vfork;
1448                         init_completion(&vfork);
1449                 }
1450
1451                 audit_finish_fork(p);
1452                 tracehook_report_clone(regs, clone_flags, nr, p);
1453
1454                 /*
1455                  * We set PF_STARTING at creation in case tracing wants to
1456                  * use this to distinguish a fully live task from one that
1457                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1458                  * clear it and set the child going.
1459                  */
1460                 p->flags &= ~PF_STARTING;
1461
1462                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1463                         /*
1464                          * We'll start up with an immediate SIGSTOP.
1465                          */
1466                         sigaddset(&p->pending.signal, SIGSTOP);
1467                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1468                         __set_task_state(p, TASK_STOPPED);
1469                 } else {
1470                         wake_up_new_task(p, clone_flags);
1471                 }
1472
1473                 tracehook_report_clone_complete(trace, regs,
1474                                                 clone_flags, nr, p);
1475
1476                 if (clone_flags & CLONE_VFORK) {
1477                         freezer_do_not_count();
1478                         wait_for_completion(&vfork);
1479                         freezer_count();
1480                         tracehook_report_vfork_done(p, nr);
1481                 }
1482         } else {
1483                 nr = PTR_ERR(p);
1484         }
1485         return nr;
1486 }
1487
1488 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1489 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1490 #endif
1491
1492 static void sighand_ctor(void *data)
1493 {
1494         struct sighand_struct *sighand = data;
1495
1496         spin_lock_init(&sighand->siglock);
1497         init_waitqueue_head(&sighand->signalfd_wqh);
1498 }
1499
1500 void __init proc_caches_init(void)
1501 {
1502         sighand_cachep = kmem_cache_create("sighand_cache",
1503                         sizeof(struct sighand_struct), 0,
1504                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1505                         SLAB_NOTRACK, sighand_ctor);
1506         signal_cachep = kmem_cache_create("signal_cache",
1507                         sizeof(struct signal_struct), 0,
1508                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1509         files_cachep = kmem_cache_create("files_cache",
1510                         sizeof(struct files_struct), 0,
1511                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1512         fs_cachep = kmem_cache_create("fs_cache",
1513                         sizeof(struct fs_struct), 0,
1514                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1515         mm_cachep = kmem_cache_create("mm_struct",
1516                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1517                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1518         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1519         mmap_init();
1520 }
1521
1522 /*
1523  * Check constraints on flags passed to the unshare system call and
1524  * force unsharing of additional process context as appropriate.
1525  */
1526 static void check_unshare_flags(unsigned long *flags_ptr)
1527 {
1528         /*
1529          * If unsharing a thread from a thread group, must also
1530          * unshare vm.
1531          */
1532         if (*flags_ptr & CLONE_THREAD)
1533                 *flags_ptr |= CLONE_VM;
1534
1535         /*
1536          * If unsharing vm, must also unshare signal handlers.
1537          */
1538         if (*flags_ptr & CLONE_VM)
1539                 *flags_ptr |= CLONE_SIGHAND;
1540
1541         /*
1542          * If unsharing signal handlers and the task was created
1543          * using CLONE_THREAD, then must unshare the thread
1544          */
1545         if ((*flags_ptr & CLONE_SIGHAND) &&
1546             (atomic_read(&current->signal->count) > 1))
1547                 *flags_ptr |= CLONE_THREAD;
1548
1549         /*
1550          * If unsharing namespace, must also unshare filesystem information.
1551          */
1552         if (*flags_ptr & CLONE_NEWNS)
1553                 *flags_ptr |= CLONE_FS;
1554 }
1555
1556 /*
1557  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1558  */
1559 static int unshare_thread(unsigned long unshare_flags)
1560 {
1561         if (unshare_flags & CLONE_THREAD)
1562                 return -EINVAL;
1563
1564         return 0;
1565 }
1566
1567 /*
1568  * Unshare the filesystem structure if it is being shared
1569  */
1570 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1571 {
1572         struct fs_struct *fs = current->fs;
1573
1574         if (!(unshare_flags & CLONE_FS) || !fs)
1575                 return 0;
1576
1577         /* don't need lock here; in the worst case we'll do useless copy */
1578         if (fs->users == 1)
1579                 return 0;
1580
1581         *new_fsp = copy_fs_struct(fs);
1582         if (!*new_fsp)
1583                 return -ENOMEM;
1584
1585         return 0;
1586 }
1587
1588 /*
1589  * Unsharing of sighand is not supported yet
1590  */
1591 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1592 {
1593         struct sighand_struct *sigh = current->sighand;
1594
1595         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1596                 return -EINVAL;
1597         else
1598                 return 0;
1599 }
1600
1601 /*
1602  * Unshare vm if it is being shared
1603  */
1604 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1605 {
1606         struct mm_struct *mm = current->mm;
1607
1608         if ((unshare_flags & CLONE_VM) &&
1609             (mm && atomic_read(&mm->mm_users) > 1)) {
1610                 return -EINVAL;
1611         }
1612
1613         return 0;
1614 }
1615
1616 /*
1617  * Unshare file descriptor table if it is being shared
1618  */
1619 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1620 {
1621         struct files_struct *fd = current->files;
1622         int error = 0;
1623
1624         if ((unshare_flags & CLONE_FILES) &&
1625             (fd && atomic_read(&fd->count) > 1)) {
1626                 *new_fdp = dup_fd(fd, &error);
1627                 if (!*new_fdp)
1628                         return error;
1629         }
1630
1631         return 0;
1632 }
1633
1634 /*
1635  * unshare allows a process to 'unshare' part of the process
1636  * context which was originally shared using clone.  copy_*
1637  * functions used by do_fork() cannot be used here directly
1638  * because they modify an inactive task_struct that is being
1639  * constructed. Here we are modifying the current, active,
1640  * task_struct.
1641  */
1642 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1643 {
1644         int err = 0;
1645         struct fs_struct *fs, *new_fs = NULL;
1646         struct sighand_struct *new_sigh = NULL;
1647         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1648         struct files_struct *fd, *new_fd = NULL;
1649         struct nsproxy *new_nsproxy = NULL;
1650         int do_sysvsem = 0;
1651
1652         check_unshare_flags(&unshare_flags);
1653
1654         /* Return -EINVAL for all unsupported flags */
1655         err = -EINVAL;
1656         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1657                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1658                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1659                 goto bad_unshare_out;
1660
1661         /*
1662          * CLONE_NEWIPC must also detach from the undolist: after switching
1663          * to a new ipc namespace, the semaphore arrays from the old
1664          * namespace are unreachable.
1665          */
1666         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1667                 do_sysvsem = 1;
1668         if ((err = unshare_thread(unshare_flags)))
1669                 goto bad_unshare_out;
1670         if ((err = unshare_fs(unshare_flags, &new_fs)))
1671                 goto bad_unshare_cleanup_thread;
1672         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1673                 goto bad_unshare_cleanup_fs;
1674         if ((err = unshare_vm(unshare_flags, &new_mm)))
1675                 goto bad_unshare_cleanup_sigh;
1676         if ((err = unshare_fd(unshare_flags, &new_fd)))
1677                 goto bad_unshare_cleanup_vm;
1678         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1679                         new_fs)))
1680                 goto bad_unshare_cleanup_fd;
1681
1682         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1683                 if (do_sysvsem) {
1684                         /*
1685                          * CLONE_SYSVSEM is equivalent to sys_exit().
1686                          */
1687                         exit_sem(current);
1688                 }
1689
1690                 if (new_nsproxy) {
1691                         switch_task_namespaces(current, new_nsproxy);
1692                         new_nsproxy = NULL;
1693                 }
1694
1695                 task_lock(current);
1696
1697                 if (new_fs) {
1698                         fs = current->fs;
1699                         write_lock(&fs->lock);
1700                         current->fs = new_fs;
1701                         if (--fs->users)
1702                                 new_fs = NULL;
1703                         else
1704                                 new_fs = fs;
1705                         write_unlock(&fs->lock);
1706                 }
1707
1708                 if (new_mm) {
1709                         mm = current->mm;
1710                         active_mm = current->active_mm;
1711                         current->mm = new_mm;
1712                         current->active_mm = new_mm;
1713                         activate_mm(active_mm, new_mm);
1714                         new_mm = mm;
1715                 }
1716
1717                 if (new_fd) {
1718                         fd = current->files;
1719                         current->files = new_fd;
1720                         new_fd = fd;
1721                 }
1722
1723                 task_unlock(current);
1724         }
1725
1726         if (new_nsproxy)
1727                 put_nsproxy(new_nsproxy);
1728
1729 bad_unshare_cleanup_fd:
1730         if (new_fd)
1731                 put_files_struct(new_fd);
1732
1733 bad_unshare_cleanup_vm:
1734         if (new_mm)
1735                 mmput(new_mm);
1736
1737 bad_unshare_cleanup_sigh:
1738         if (new_sigh)
1739                 if (atomic_dec_and_test(&new_sigh->count))
1740                         kmem_cache_free(sighand_cachep, new_sigh);
1741
1742 bad_unshare_cleanup_fs:
1743         if (new_fs)
1744                 free_fs_struct(new_fs);
1745
1746 bad_unshare_cleanup_thread:
1747 bad_unshare_out:
1748         return err;
1749 }
1750
1751 /*
1752  *      Helper to unshare the files of the current task.
1753  *      We don't want to expose copy_files internals to
1754  *      the exec layer of the kernel.
1755  */
1756
1757 int unshare_files(struct files_struct **displaced)
1758 {
1759         struct task_struct *task = current;
1760         struct files_struct *copy = NULL;
1761         int error;
1762
1763         error = unshare_fd(CLONE_FILES, &copy);
1764         if (error || !copy) {
1765                 *displaced = NULL;
1766                 return error;
1767         }
1768         *displaced = task->files;
1769         task_lock(task);
1770         task->files = copy;
1771         task_unlock(task);
1772         return 0;
1773 }