perf_counter: Move child perfcounter init to after scheduler init
[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/mnt_namespace.h>
21 #include <linux/personality.h>
22 #include <linux/mempolicy.h>
23 #include <linux/sem.h>
24 #include <linux/file.h>
25 #include <linux/fdtable.h>
26 #include <linux/iocontext.h>
27 #include <linux/key.h>
28 #include <linux/binfmts.h>
29 #include <linux/mman.h>
30 #include <linux/mmu_notifier.h>
31 #include <linux/fs.h>
32 #include <linux/nsproxy.h>
33 #include <linux/capability.h>
34 #include <linux/cpu.h>
35 #include <linux/cgroup.h>
36 #include <linux/security.h>
37 #include <linux/hugetlb.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/tracehook.h>
42 #include <linux/futex.h>
43 #include <linux/compat.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/profile.h>
52 #include <linux/rmap.h>
53 #include <linux/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 <trace/sched.h>
65 #include <linux/magic.h>
66 #include <linux/perf_counter.h>
67
68 #include <asm/pgtable.h>
69 #include <asm/pgalloc.h>
70 #include <asm/uaccess.h>
71 #include <asm/mmu_context.h>
72 #include <asm/cacheflush.h>
73 #include <asm/tlbflush.h>
74
75 /*
76  * Protected counters by write_lock_irq(&tasklist_lock)
77  */
78 unsigned long total_forks;      /* Handle normal Linux uptimes. */
79 int nr_threads;                 /* The idle threads do not count.. */
80
81 int max_threads;                /* tunable limit on nr_threads */
82
83 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
84
85 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
86
87 DEFINE_TRACE(sched_process_fork);
88
89 int nr_processes(void)
90 {
91         int cpu;
92         int total = 0;
93
94         for_each_online_cpu(cpu)
95                 total += per_cpu(process_counts, cpu);
96
97         return total;
98 }
99
100 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
101 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
102 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
103 static struct kmem_cache *task_struct_cachep;
104 #endif
105
106 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
107 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
108 {
109 #ifdef CONFIG_DEBUG_STACK_USAGE
110         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
111 #else
112         gfp_t mask = GFP_KERNEL;
113 #endif
114         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
115 }
116
117 static inline void free_thread_info(struct thread_info *ti)
118 {
119         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
120 }
121 #endif
122
123 /* SLAB cache for signal_struct structures (tsk->signal) */
124 static struct kmem_cache *signal_cachep;
125
126 /* SLAB cache for sighand_struct structures (tsk->sighand) */
127 struct kmem_cache *sighand_cachep;
128
129 /* SLAB cache for files_struct structures (tsk->files) */
130 struct kmem_cache *files_cachep;
131
132 /* SLAB cache for fs_struct structures (tsk->fs) */
133 struct kmem_cache *fs_cachep;
134
135 /* SLAB cache for vm_area_struct structures */
136 struct kmem_cache *vm_area_cachep;
137
138 /* SLAB cache for mm_struct structures (tsk->mm) */
139 static struct kmem_cache *mm_cachep;
140
141 void free_task(struct task_struct *tsk)
142 {
143         prop_local_destroy_single(&tsk->dirties);
144         free_thread_info(tsk->stack);
145         rt_mutex_debug_task_free(tsk);
146         ftrace_graph_exit_task(tsk);
147         free_task_struct(tsk);
148 }
149 EXPORT_SYMBOL(free_task);
150
151 void __put_task_struct(struct task_struct *tsk)
152 {
153         WARN_ON(!tsk->exit_state);
154         WARN_ON(atomic_read(&tsk->usage));
155         WARN_ON(tsk == current);
156
157         put_cred(tsk->real_cred);
158         put_cred(tsk->cred);
159         delayacct_tsk_free(tsk);
160
161         if (!profile_handoff_task(tsk))
162                 free_task(tsk);
163 }
164
165 /*
166  * macro override instead of weak attribute alias, to workaround
167  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
168  */
169 #ifndef arch_task_cache_init
170 #define arch_task_cache_init()
171 #endif
172
173 void __init fork_init(unsigned long mempages)
174 {
175 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
176 #ifndef ARCH_MIN_TASKALIGN
177 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
178 #endif
179         /* create a slab on which task_structs can be allocated */
180         task_struct_cachep =
181                 kmem_cache_create("task_struct", sizeof(struct task_struct),
182                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
183 #endif
184
185         /* do the arch specific task caches init */
186         arch_task_cache_init();
187
188         /*
189          * The default maximum number of threads is set to a safe
190          * value: the thread structures can take up at most half
191          * of memory.
192          */
193         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
194
195         /*
196          * we need to allow at least 20 threads to boot a system
197          */
198         if(max_threads < 20)
199                 max_threads = 20;
200
201         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
202         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
203         init_task.signal->rlim[RLIMIT_SIGPENDING] =
204                 init_task.signal->rlim[RLIMIT_NPROC];
205 }
206
207 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
208                                                struct task_struct *src)
209 {
210         *dst = *src;
211         return 0;
212 }
213
214 static struct task_struct *dup_task_struct(struct task_struct *orig)
215 {
216         struct task_struct *tsk;
217         struct thread_info *ti;
218         unsigned long *stackend;
219
220         int err;
221
222         prepare_to_copy(orig);
223
224         tsk = alloc_task_struct();
225         if (!tsk)
226                 return NULL;
227
228         ti = alloc_thread_info(tsk);
229         if (!ti) {
230                 free_task_struct(tsk);
231                 return NULL;
232         }
233
234         err = arch_dup_task_struct(tsk, orig);
235         if (err)
236                 goto out;
237
238         tsk->stack = ti;
239
240         err = prop_local_init_single(&tsk->dirties);
241         if (err)
242                 goto out;
243
244         setup_thread_stack(tsk, orig);
245         stackend = end_of_stack(tsk);
246         *stackend = STACK_END_MAGIC;    /* for overflow detection */
247
248 #ifdef CONFIG_CC_STACKPROTECTOR
249         tsk->stack_canary = get_random_int();
250 #endif
251
252         /* One for us, one for whoever does the "release_task()" (usually parent) */
253         atomic_set(&tsk->usage,2);
254         atomic_set(&tsk->fs_excl, 0);
255 #ifdef CONFIG_BLK_DEV_IO_TRACE
256         tsk->btrace_seq = 0;
257 #endif
258         tsk->splice_pipe = NULL;
259         return tsk;
260
261 out:
262         free_thread_info(ti);
263         free_task_struct(tsk);
264         return NULL;
265 }
266
267 #ifdef CONFIG_MMU
268 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
269 {
270         struct vm_area_struct *mpnt, *tmp, **pprev;
271         struct rb_node **rb_link, *rb_parent;
272         int retval;
273         unsigned long charge;
274         struct mempolicy *pol;
275
276         down_write(&oldmm->mmap_sem);
277         flush_cache_dup_mm(oldmm);
278         /*
279          * Not linked in yet - no deadlock potential:
280          */
281         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
282
283         mm->locked_vm = 0;
284         mm->mmap = NULL;
285         mm->mmap_cache = NULL;
286         mm->free_area_cache = oldmm->mmap_base;
287         mm->cached_hole_size = ~0UL;
288         mm->map_count = 0;
289         cpumask_clear(mm_cpumask(mm));
290         mm->mm_rb = RB_ROOT;
291         rb_link = &mm->mm_rb.rb_node;
292         rb_parent = NULL;
293         pprev = &mm->mmap;
294
295         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
296                 struct file *file;
297
298                 if (mpnt->vm_flags & VM_DONTCOPY) {
299                         long pages = vma_pages(mpnt);
300                         mm->total_vm -= pages;
301                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
302                                                                 -pages);
303                         continue;
304                 }
305                 charge = 0;
306                 if (mpnt->vm_flags & VM_ACCOUNT) {
307                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
308                         if (security_vm_enough_memory(len))
309                                 goto fail_nomem;
310                         charge = len;
311                 }
312                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
313                 if (!tmp)
314                         goto fail_nomem;
315                 *tmp = *mpnt;
316                 pol = mpol_dup(vma_policy(mpnt));
317                 retval = PTR_ERR(pol);
318                 if (IS_ERR(pol))
319                         goto fail_nomem_policy;
320                 vma_set_policy(tmp, pol);
321                 tmp->vm_flags &= ~VM_LOCKED;
322                 tmp->vm_mm = mm;
323                 tmp->vm_next = NULL;
324                 anon_vma_link(tmp);
325                 file = tmp->vm_file;
326                 if (file) {
327                         struct inode *inode = file->f_path.dentry->d_inode;
328                         struct address_space *mapping = file->f_mapping;
329
330                         get_file(file);
331                         if (tmp->vm_flags & VM_DENYWRITE)
332                                 atomic_dec(&inode->i_writecount);
333                         spin_lock(&mapping->i_mmap_lock);
334                         if (tmp->vm_flags & VM_SHARED)
335                                 mapping->i_mmap_writable++;
336                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
337                         flush_dcache_mmap_lock(mapping);
338                         /* insert tmp into the share list, just after mpnt */
339                         vma_prio_tree_add(tmp, mpnt);
340                         flush_dcache_mmap_unlock(mapping);
341                         spin_unlock(&mapping->i_mmap_lock);
342                 }
343
344                 /*
345                  * Clear hugetlb-related page reserves for children. This only
346                  * affects MAP_PRIVATE mappings. Faults generated by the child
347                  * are not guaranteed to succeed, even if read-only
348                  */
349                 if (is_vm_hugetlb_page(tmp))
350                         reset_vma_resv_huge_pages(tmp);
351
352                 /*
353                  * Link in the new vma and copy the page table entries.
354                  */
355                 *pprev = tmp;
356                 pprev = &tmp->vm_next;
357
358                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
359                 rb_link = &tmp->vm_rb.rb_right;
360                 rb_parent = &tmp->vm_rb;
361
362                 mm->map_count++;
363                 retval = copy_page_range(mm, oldmm, mpnt);
364
365                 if (tmp->vm_ops && tmp->vm_ops->open)
366                         tmp->vm_ops->open(tmp);
367
368                 if (retval)
369                         goto out;
370         }
371         /* a new mm has just been created */
372         arch_dup_mmap(oldmm, mm);
373         retval = 0;
374 out:
375         up_write(&mm->mmap_sem);
376         flush_tlb_mm(oldmm);
377         up_write(&oldmm->mmap_sem);
378         return retval;
379 fail_nomem_policy:
380         kmem_cache_free(vm_area_cachep, tmp);
381 fail_nomem:
382         retval = -ENOMEM;
383         vm_unacct_memory(charge);
384         goto out;
385 }
386
387 static inline int mm_alloc_pgd(struct mm_struct * mm)
388 {
389         mm->pgd = pgd_alloc(mm);
390         if (unlikely(!mm->pgd))
391                 return -ENOMEM;
392         return 0;
393 }
394
395 static inline void mm_free_pgd(struct mm_struct * mm)
396 {
397         pgd_free(mm, mm->pgd);
398 }
399 #else
400 #define dup_mmap(mm, oldmm)     (0)
401 #define mm_alloc_pgd(mm)        (0)
402 #define mm_free_pgd(mm)
403 #endif /* CONFIG_MMU */
404
405 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
406
407 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
408 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
409
410 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
411
412 static int __init coredump_filter_setup(char *s)
413 {
414         default_dump_filter =
415                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
416                 MMF_DUMP_FILTER_MASK;
417         return 1;
418 }
419
420 __setup("coredump_filter=", coredump_filter_setup);
421
422 #include <linux/init_task.h>
423
424 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
425 {
426         atomic_set(&mm->mm_users, 1);
427         atomic_set(&mm->mm_count, 1);
428         init_rwsem(&mm->mmap_sem);
429         INIT_LIST_HEAD(&mm->mmlist);
430         mm->flags = (current->mm) ? current->mm->flags : default_dump_filter;
431         mm->core_state = NULL;
432         mm->nr_ptes = 0;
433         set_mm_counter(mm, file_rss, 0);
434         set_mm_counter(mm, anon_rss, 0);
435         spin_lock_init(&mm->page_table_lock);
436         spin_lock_init(&mm->ioctx_lock);
437         INIT_HLIST_HEAD(&mm->ioctx_list);
438         mm->free_area_cache = TASK_UNMAPPED_BASE;
439         mm->cached_hole_size = ~0UL;
440         mm_init_owner(mm, p);
441
442         if (likely(!mm_alloc_pgd(mm))) {
443                 mm->def_flags = 0;
444                 mmu_notifier_mm_init(mm);
445                 return mm;
446         }
447
448         free_mm(mm);
449         return NULL;
450 }
451
452 /*
453  * Allocate and initialize an mm_struct.
454  */
455 struct mm_struct * mm_alloc(void)
456 {
457         struct mm_struct * mm;
458
459         mm = allocate_mm();
460         if (mm) {
461                 memset(mm, 0, sizeof(*mm));
462                 mm = mm_init(mm, current);
463         }
464         return mm;
465 }
466
467 /*
468  * Called when the last reference to the mm
469  * is dropped: either by a lazy thread or by
470  * mmput. Free the page directory and the mm.
471  */
472 void __mmdrop(struct mm_struct *mm)
473 {
474         BUG_ON(mm == &init_mm);
475         mm_free_pgd(mm);
476         destroy_context(mm);
477         mmu_notifier_mm_destroy(mm);
478         free_mm(mm);
479 }
480 EXPORT_SYMBOL_GPL(__mmdrop);
481
482 /*
483  * Decrement the use count and release all resources for an mm.
484  */
485 void mmput(struct mm_struct *mm)
486 {
487         might_sleep();
488
489         if (atomic_dec_and_test(&mm->mm_users)) {
490                 exit_aio(mm);
491                 exit_mmap(mm);
492                 set_mm_exe_file(mm, NULL);
493                 if (!list_empty(&mm->mmlist)) {
494                         spin_lock(&mmlist_lock);
495                         list_del(&mm->mmlist);
496                         spin_unlock(&mmlist_lock);
497                 }
498                 put_swap_token(mm);
499                 mmdrop(mm);
500         }
501 }
502 EXPORT_SYMBOL_GPL(mmput);
503
504 /**
505  * get_task_mm - acquire a reference to the task's mm
506  *
507  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
508  * this kernel workthread has transiently adopted a user mm with use_mm,
509  * to do its AIO) is not set and if so returns a reference to it, after
510  * bumping up the use count.  User must release the mm via mmput()
511  * after use.  Typically used by /proc and ptrace.
512  */
513 struct mm_struct *get_task_mm(struct task_struct *task)
514 {
515         struct mm_struct *mm;
516
517         task_lock(task);
518         mm = task->mm;
519         if (mm) {
520                 if (task->flags & PF_KTHREAD)
521                         mm = NULL;
522                 else
523                         atomic_inc(&mm->mm_users);
524         }
525         task_unlock(task);
526         return mm;
527 }
528 EXPORT_SYMBOL_GPL(get_task_mm);
529
530 /* Please note the differences between mmput and mm_release.
531  * mmput is called whenever we stop holding onto a mm_struct,
532  * error success whatever.
533  *
534  * mm_release is called after a mm_struct has been removed
535  * from the current process.
536  *
537  * This difference is important for error handling, when we
538  * only half set up a mm_struct for a new process and need to restore
539  * the old one.  Because we mmput the new mm_struct before
540  * restoring the old one. . .
541  * Eric Biederman 10 January 1998
542  */
543 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
544 {
545         struct completion *vfork_done = tsk->vfork_done;
546
547         /* Get rid of any futexes when releasing the mm */
548 #ifdef CONFIG_FUTEX
549         if (unlikely(tsk->robust_list))
550                 exit_robust_list(tsk);
551 #ifdef CONFIG_COMPAT
552         if (unlikely(tsk->compat_robust_list))
553                 compat_exit_robust_list(tsk);
554 #endif
555 #endif
556
557         /* Get rid of any cached register state */
558         deactivate_mm(tsk, mm);
559
560         /* notify parent sleeping on vfork() */
561         if (vfork_done) {
562                 tsk->vfork_done = NULL;
563                 complete(vfork_done);
564         }
565
566         /*
567          * If we're exiting normally, clear a user-space tid field if
568          * requested.  We leave this alone when dying by signal, to leave
569          * the value intact in a core dump, and to save the unnecessary
570          * trouble otherwise.  Userland only wants this done for a sys_exit.
571          */
572         if (tsk->clear_child_tid
573             && !(tsk->flags & PF_SIGNALED)
574             && atomic_read(&mm->mm_users) > 1) {
575                 u32 __user * tidptr = tsk->clear_child_tid;
576                 tsk->clear_child_tid = NULL;
577
578                 /*
579                  * We don't check the error code - if userspace has
580                  * not set up a proper pointer then tough luck.
581                  */
582                 put_user(0, tidptr);
583                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
584         }
585 }
586
587 /*
588  * Allocate a new mm structure and copy contents from the
589  * mm structure of the passed in task structure.
590  */
591 struct mm_struct *dup_mm(struct task_struct *tsk)
592 {
593         struct mm_struct *mm, *oldmm = current->mm;
594         int err;
595
596         if (!oldmm)
597                 return NULL;
598
599         mm = allocate_mm();
600         if (!mm)
601                 goto fail_nomem;
602
603         memcpy(mm, oldmm, sizeof(*mm));
604
605         /* Initializing for Swap token stuff */
606         mm->token_priority = 0;
607         mm->last_interval = 0;
608
609         if (!mm_init(mm, tsk))
610                 goto fail_nomem;
611
612         if (init_new_context(tsk, mm))
613                 goto fail_nocontext;
614
615         dup_mm_exe_file(oldmm, mm);
616
617         err = dup_mmap(mm, oldmm);
618         if (err)
619                 goto free_pt;
620
621         mm->hiwater_rss = get_mm_rss(mm);
622         mm->hiwater_vm = mm->total_vm;
623
624         return mm;
625
626 free_pt:
627         mmput(mm);
628
629 fail_nomem:
630         return NULL;
631
632 fail_nocontext:
633         /*
634          * If init_new_context() failed, we cannot use mmput() to free the mm
635          * because it calls destroy_context()
636          */
637         mm_free_pgd(mm);
638         free_mm(mm);
639         return NULL;
640 }
641
642 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
643 {
644         struct mm_struct * mm, *oldmm;
645         int retval;
646
647         tsk->min_flt = tsk->maj_flt = 0;
648         tsk->nvcsw = tsk->nivcsw = 0;
649 #ifdef CONFIG_DETECT_HUNG_TASK
650         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
651 #endif
652
653         tsk->mm = NULL;
654         tsk->active_mm = NULL;
655
656         /*
657          * Are we cloning a kernel thread?
658          *
659          * We need to steal a active VM for that..
660          */
661         oldmm = current->mm;
662         if (!oldmm)
663                 return 0;
664
665         if (clone_flags & CLONE_VM) {
666                 atomic_inc(&oldmm->mm_users);
667                 mm = oldmm;
668                 goto good_mm;
669         }
670
671         retval = -ENOMEM;
672         mm = dup_mm(tsk);
673         if (!mm)
674                 goto fail_nomem;
675
676 good_mm:
677         /* Initializing for Swap token stuff */
678         mm->token_priority = 0;
679         mm->last_interval = 0;
680
681         tsk->mm = mm;
682         tsk->active_mm = mm;
683         return 0;
684
685 fail_nomem:
686         return retval;
687 }
688
689 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
690 {
691         struct fs_struct *fs = current->fs;
692         if (clone_flags & CLONE_FS) {
693                 /* tsk->fs is already what we want */
694                 write_lock(&fs->lock);
695                 if (fs->in_exec) {
696                         write_unlock(&fs->lock);
697                         return -EAGAIN;
698                 }
699                 fs->users++;
700                 write_unlock(&fs->lock);
701                 return 0;
702         }
703         tsk->fs = copy_fs_struct(fs);
704         if (!tsk->fs)
705                 return -ENOMEM;
706         return 0;
707 }
708
709 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
710 {
711         struct files_struct *oldf, *newf;
712         int error = 0;
713
714         /*
715          * A background process may not have any files ...
716          */
717         oldf = current->files;
718         if (!oldf)
719                 goto out;
720
721         if (clone_flags & CLONE_FILES) {
722                 atomic_inc(&oldf->count);
723                 goto out;
724         }
725
726         newf = dup_fd(oldf, &error);
727         if (!newf)
728                 goto out;
729
730         tsk->files = newf;
731         error = 0;
732 out:
733         return error;
734 }
735
736 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
737 {
738 #ifdef CONFIG_BLOCK
739         struct io_context *ioc = current->io_context;
740
741         if (!ioc)
742                 return 0;
743         /*
744          * Share io context with parent, if CLONE_IO is set
745          */
746         if (clone_flags & CLONE_IO) {
747                 tsk->io_context = ioc_task_link(ioc);
748                 if (unlikely(!tsk->io_context))
749                         return -ENOMEM;
750         } else if (ioprio_valid(ioc->ioprio)) {
751                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
752                 if (unlikely(!tsk->io_context))
753                         return -ENOMEM;
754
755                 tsk->io_context->ioprio = ioc->ioprio;
756         }
757 #endif
758         return 0;
759 }
760
761 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
762 {
763         struct sighand_struct *sig;
764
765         if (clone_flags & CLONE_SIGHAND) {
766                 atomic_inc(&current->sighand->count);
767                 return 0;
768         }
769         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
770         rcu_assign_pointer(tsk->sighand, sig);
771         if (!sig)
772                 return -ENOMEM;
773         atomic_set(&sig->count, 1);
774         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
775         return 0;
776 }
777
778 void __cleanup_sighand(struct sighand_struct *sighand)
779 {
780         if (atomic_dec_and_test(&sighand->count))
781                 kmem_cache_free(sighand_cachep, sighand);
782 }
783
784
785 /*
786  * Initialize POSIX timer handling for a thread group.
787  */
788 static void posix_cpu_timers_init_group(struct signal_struct *sig)
789 {
790         /* Thread group counters. */
791         thread_group_cputime_init(sig);
792
793         /* Expiration times and increments. */
794         sig->it_virt_expires = cputime_zero;
795         sig->it_virt_incr = cputime_zero;
796         sig->it_prof_expires = cputime_zero;
797         sig->it_prof_incr = cputime_zero;
798
799         /* Cached expiration times. */
800         sig->cputime_expires.prof_exp = cputime_zero;
801         sig->cputime_expires.virt_exp = cputime_zero;
802         sig->cputime_expires.sched_exp = 0;
803
804         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
805                 sig->cputime_expires.prof_exp =
806                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
807                 sig->cputimer.running = 1;
808         }
809
810         /* The timer lists. */
811         INIT_LIST_HEAD(&sig->cpu_timers[0]);
812         INIT_LIST_HEAD(&sig->cpu_timers[1]);
813         INIT_LIST_HEAD(&sig->cpu_timers[2]);
814 }
815
816 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
817 {
818         struct signal_struct *sig;
819
820         if (clone_flags & CLONE_THREAD) {
821                 atomic_inc(&current->signal->count);
822                 atomic_inc(&current->signal->live);
823                 return 0;
824         }
825
826         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
827         tsk->signal = sig;
828         if (!sig)
829                 return -ENOMEM;
830
831         atomic_set(&sig->count, 1);
832         atomic_set(&sig->live, 1);
833         init_waitqueue_head(&sig->wait_chldexit);
834         sig->flags = 0;
835         if (clone_flags & CLONE_NEWPID)
836                 sig->flags |= SIGNAL_UNKILLABLE;
837         sig->group_exit_code = 0;
838         sig->group_exit_task = NULL;
839         sig->group_stop_count = 0;
840         sig->curr_target = tsk;
841         init_sigpending(&sig->shared_pending);
842         INIT_LIST_HEAD(&sig->posix_timers);
843
844         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
845         sig->it_real_incr.tv64 = 0;
846         sig->real_timer.function = it_real_fn;
847
848         sig->leader = 0;        /* session leadership doesn't inherit */
849         sig->tty_old_pgrp = NULL;
850         sig->tty = NULL;
851
852         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
853         sig->gtime = cputime_zero;
854         sig->cgtime = cputime_zero;
855         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
856         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
857         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
858         task_io_accounting_init(&sig->ioac);
859         sig->sum_sched_runtime = 0;
860         taskstats_tgid_init(sig);
861
862         task_lock(current->group_leader);
863         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
864         task_unlock(current->group_leader);
865
866         posix_cpu_timers_init_group(sig);
867
868         acct_init_pacct(&sig->pacct);
869
870         tty_audit_fork(sig);
871
872         return 0;
873 }
874
875 void __cleanup_signal(struct signal_struct *sig)
876 {
877         thread_group_cputime_free(sig);
878         tty_kref_put(sig->tty);
879         kmem_cache_free(signal_cachep, sig);
880 }
881
882 static void cleanup_signal(struct task_struct *tsk)
883 {
884         struct signal_struct *sig = tsk->signal;
885
886         atomic_dec(&sig->live);
887
888         if (atomic_dec_and_test(&sig->count))
889                 __cleanup_signal(sig);
890 }
891
892 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
893 {
894         unsigned long new_flags = p->flags;
895
896         new_flags &= ~PF_SUPERPRIV;
897         new_flags |= PF_FORKNOEXEC;
898         new_flags |= PF_STARTING;
899         p->flags = new_flags;
900         clear_freeze_flag(p);
901 }
902
903 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
904 {
905         current->clear_child_tid = tidptr;
906
907         return task_pid_vnr(current);
908 }
909
910 static void rt_mutex_init_task(struct task_struct *p)
911 {
912         spin_lock_init(&p->pi_lock);
913 #ifdef CONFIG_RT_MUTEXES
914         plist_head_init(&p->pi_waiters, &p->pi_lock);
915         p->pi_blocked_on = NULL;
916 #endif
917 }
918
919 #ifdef CONFIG_MM_OWNER
920 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
921 {
922         mm->owner = p;
923 }
924 #endif /* CONFIG_MM_OWNER */
925
926 /*
927  * Initialize POSIX timer handling for a single task.
928  */
929 static void posix_cpu_timers_init(struct task_struct *tsk)
930 {
931         tsk->cputime_expires.prof_exp = cputime_zero;
932         tsk->cputime_expires.virt_exp = cputime_zero;
933         tsk->cputime_expires.sched_exp = 0;
934         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
935         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
936         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
937 }
938
939 /*
940  * This creates a new process as a copy of the old one,
941  * but does not actually start it yet.
942  *
943  * It copies the registers, and all the appropriate
944  * parts of the process environment (as per the clone
945  * flags). The actual kick-off is left to the caller.
946  */
947 static struct task_struct *copy_process(unsigned long clone_flags,
948                                         unsigned long stack_start,
949                                         struct pt_regs *regs,
950                                         unsigned long stack_size,
951                                         int __user *child_tidptr,
952                                         struct pid *pid,
953                                         int trace)
954 {
955         int retval;
956         struct task_struct *p;
957         int cgroup_callbacks_done = 0;
958
959         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
960                 return ERR_PTR(-EINVAL);
961
962         /*
963          * Thread groups must share signals as well, and detached threads
964          * can only be started up within the thread group.
965          */
966         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
967                 return ERR_PTR(-EINVAL);
968
969         /*
970          * Shared signal handlers imply shared VM. By way of the above,
971          * thread groups also imply shared VM. Blocking this case allows
972          * for various simplifications in other code.
973          */
974         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
975                 return ERR_PTR(-EINVAL);
976
977         retval = security_task_create(clone_flags);
978         if (retval)
979                 goto fork_out;
980
981         retval = -ENOMEM;
982         p = dup_task_struct(current);
983         if (!p)
984                 goto fork_out;
985
986         rt_mutex_init_task(p);
987
988 #ifdef CONFIG_PROVE_LOCKING
989         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
990         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
991 #endif
992         retval = -EAGAIN;
993         if (atomic_read(&p->real_cred->user->processes) >=
994                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
995                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
996                     p->real_cred->user != INIT_USER)
997                         goto bad_fork_free;
998         }
999
1000         retval = copy_creds(p, clone_flags);
1001         if (retval < 0)
1002                 goto bad_fork_free;
1003
1004         /*
1005          * If multiple threads are within copy_process(), then this check
1006          * triggers too late. This doesn't hurt, the check is only there
1007          * to stop root fork bombs.
1008          */
1009         retval = -EAGAIN;
1010         if (nr_threads >= max_threads)
1011                 goto bad_fork_cleanup_count;
1012
1013         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1014                 goto bad_fork_cleanup_count;
1015
1016         if (p->binfmt && !try_module_get(p->binfmt->module))
1017                 goto bad_fork_cleanup_put_domain;
1018
1019         p->did_exec = 0;
1020         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1021         copy_flags(clone_flags, p);
1022         INIT_LIST_HEAD(&p->children);
1023         INIT_LIST_HEAD(&p->sibling);
1024 #ifdef CONFIG_PREEMPT_RCU
1025         p->rcu_read_lock_nesting = 0;
1026         p->rcu_flipctr_idx = 0;
1027 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1028         p->vfork_done = NULL;
1029         spin_lock_init(&p->alloc_lock);
1030
1031         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1032         init_sigpending(&p->pending);
1033
1034         p->utime = cputime_zero;
1035         p->stime = cputime_zero;
1036         p->gtime = cputime_zero;
1037         p->utimescaled = cputime_zero;
1038         p->stimescaled = cputime_zero;
1039         p->prev_utime = cputime_zero;
1040         p->prev_stime = cputime_zero;
1041
1042         p->default_timer_slack_ns = current->timer_slack_ns;
1043
1044         task_io_accounting_init(&p->ioac);
1045         acct_clear_integrals(p);
1046
1047         posix_cpu_timers_init(p);
1048
1049         p->lock_depth = -1;             /* -1 = no lock */
1050         do_posix_clock_monotonic_gettime(&p->start_time);
1051         p->real_start_time = p->start_time;
1052         monotonic_to_bootbased(&p->real_start_time);
1053         p->io_context = NULL;
1054         p->audit_context = NULL;
1055         cgroup_fork(p);
1056 #ifdef CONFIG_NUMA
1057         p->mempolicy = mpol_dup(p->mempolicy);
1058         if (IS_ERR(p->mempolicy)) {
1059                 retval = PTR_ERR(p->mempolicy);
1060                 p->mempolicy = NULL;
1061                 goto bad_fork_cleanup_cgroup;
1062         }
1063         mpol_fix_fork_child_flag(p);
1064 #endif
1065 #ifdef CONFIG_TRACE_IRQFLAGS
1066         p->irq_events = 0;
1067 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1068         p->hardirqs_enabled = 1;
1069 #else
1070         p->hardirqs_enabled = 0;
1071 #endif
1072         p->hardirq_enable_ip = 0;
1073         p->hardirq_enable_event = 0;
1074         p->hardirq_disable_ip = _THIS_IP_;
1075         p->hardirq_disable_event = 0;
1076         p->softirqs_enabled = 1;
1077         p->softirq_enable_ip = _THIS_IP_;
1078         p->softirq_enable_event = 0;
1079         p->softirq_disable_ip = 0;
1080         p->softirq_disable_event = 0;
1081         p->hardirq_context = 0;
1082         p->softirq_context = 0;
1083 #endif
1084 #ifdef CONFIG_LOCKDEP
1085         p->lockdep_depth = 0; /* no locks held yet */
1086         p->curr_chain_key = 0;
1087         p->lockdep_recursion = 0;
1088 #endif
1089
1090 #ifdef CONFIG_DEBUG_MUTEXES
1091         p->blocked_on = NULL; /* not blocked yet */
1092 #endif
1093         if (unlikely(current->ptrace))
1094                 ptrace_fork(p, clone_flags);
1095
1096         /* Perform scheduler related setup. Assign this task to a CPU. */
1097         sched_fork(p, clone_flags);
1098         perf_counter_init_task(p);
1099
1100         if ((retval = audit_alloc(p)))
1101                 goto bad_fork_cleanup_policy;
1102         /* copy all the process information */
1103         if ((retval = copy_semundo(clone_flags, p)))
1104                 goto bad_fork_cleanup_audit;
1105         if ((retval = copy_files(clone_flags, p)))
1106                 goto bad_fork_cleanup_semundo;
1107         if ((retval = copy_fs(clone_flags, p)))
1108                 goto bad_fork_cleanup_files;
1109         if ((retval = copy_sighand(clone_flags, p)))
1110                 goto bad_fork_cleanup_fs;
1111         if ((retval = copy_signal(clone_flags, p)))
1112                 goto bad_fork_cleanup_sighand;
1113         if ((retval = copy_mm(clone_flags, p)))
1114                 goto bad_fork_cleanup_signal;
1115         if ((retval = copy_namespaces(clone_flags, p)))
1116                 goto bad_fork_cleanup_mm;
1117         if ((retval = copy_io(clone_flags, p)))
1118                 goto bad_fork_cleanup_namespaces;
1119         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1120         if (retval)
1121                 goto bad_fork_cleanup_io;
1122
1123         if (pid != &init_struct_pid) {
1124                 retval = -ENOMEM;
1125                 pid = alloc_pid(p->nsproxy->pid_ns);
1126                 if (!pid)
1127                         goto bad_fork_cleanup_io;
1128
1129                 if (clone_flags & CLONE_NEWPID) {
1130                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1131                         if (retval < 0)
1132                                 goto bad_fork_free_pid;
1133                 }
1134         }
1135
1136         ftrace_graph_init_task(p);
1137
1138         p->pid = pid_nr(pid);
1139         p->tgid = p->pid;
1140         if (clone_flags & CLONE_THREAD)
1141                 p->tgid = current->tgid;
1142
1143         if (current->nsproxy != p->nsproxy) {
1144                 retval = ns_cgroup_clone(p, pid);
1145                 if (retval)
1146                         goto bad_fork_free_graph;
1147         }
1148
1149         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1150         /*
1151          * Clear TID on mm_release()?
1152          */
1153         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1154 #ifdef CONFIG_FUTEX
1155         p->robust_list = NULL;
1156 #ifdef CONFIG_COMPAT
1157         p->compat_robust_list = NULL;
1158 #endif
1159         INIT_LIST_HEAD(&p->pi_state_list);
1160         p->pi_state_cache = NULL;
1161 #endif
1162         /*
1163          * sigaltstack should be cleared when sharing the same VM
1164          */
1165         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1166                 p->sas_ss_sp = p->sas_ss_size = 0;
1167
1168         /*
1169          * Syscall tracing should be turned off in the child regardless
1170          * of CLONE_PTRACE.
1171          */
1172         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1173 #ifdef TIF_SYSCALL_EMU
1174         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1175 #endif
1176         clear_all_latency_tracing(p);
1177
1178         /* ok, now we should be set up.. */
1179         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1180         p->pdeath_signal = 0;
1181         p->exit_state = 0;
1182
1183         /*
1184          * Ok, make it visible to the rest of the system.
1185          * We dont wake it up yet.
1186          */
1187         p->group_leader = p;
1188         INIT_LIST_HEAD(&p->thread_group);
1189
1190         /* Now that the task is set up, run cgroup callbacks if
1191          * necessary. We need to run them before the task is visible
1192          * on the tasklist. */
1193         cgroup_fork_callbacks(p);
1194         cgroup_callbacks_done = 1;
1195
1196         /* Need tasklist lock for parent etc handling! */
1197         write_lock_irq(&tasklist_lock);
1198
1199         /*
1200          * The task hasn't been attached yet, so its cpus_allowed mask will
1201          * not be changed, nor will its assigned CPU.
1202          *
1203          * The cpus_allowed mask of the parent may have changed after it was
1204          * copied first time - so re-copy it here, then check the child's CPU
1205          * to ensure it is on a valid CPU (and if not, just force it back to
1206          * parent's CPU). This avoids alot of nasty races.
1207          */
1208         p->cpus_allowed = current->cpus_allowed;
1209         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1210         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1211                         !cpu_online(task_cpu(p))))
1212                 set_task_cpu(p, smp_processor_id());
1213
1214         /* CLONE_PARENT re-uses the old parent */
1215         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1216                 p->real_parent = current->real_parent;
1217                 p->parent_exec_id = current->parent_exec_id;
1218         } else {
1219                 p->real_parent = current;
1220                 p->parent_exec_id = current->self_exec_id;
1221         }
1222
1223         spin_lock(&current->sighand->siglock);
1224
1225         /*
1226          * Process group and session signals need to be delivered to just the
1227          * parent before the fork or both the parent and the child after the
1228          * fork. Restart if a signal comes in before we add the new process to
1229          * it's process group.
1230          * A fatal signal pending means that current will exit, so the new
1231          * thread can't slip out of an OOM kill (or normal SIGKILL).
1232          */
1233         recalc_sigpending();
1234         if (signal_pending(current)) {
1235                 spin_unlock(&current->sighand->siglock);
1236                 write_unlock_irq(&tasklist_lock);
1237                 retval = -ERESTARTNOINTR;
1238                 goto bad_fork_free_graph;
1239         }
1240
1241         if (clone_flags & CLONE_THREAD) {
1242                 p->group_leader = current->group_leader;
1243                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1244         }
1245
1246         if (likely(p->pid)) {
1247                 list_add_tail(&p->sibling, &p->real_parent->children);
1248                 tracehook_finish_clone(p, clone_flags, trace);
1249
1250                 if (thread_group_leader(p)) {
1251                         if (clone_flags & CLONE_NEWPID)
1252                                 p->nsproxy->pid_ns->child_reaper = p;
1253
1254                         p->signal->leader_pid = pid;
1255                         tty_kref_put(p->signal->tty);
1256                         p->signal->tty = tty_kref_get(current->signal->tty);
1257                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1258                         attach_pid(p, PIDTYPE_SID, task_session(current));
1259                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1260                         __get_cpu_var(process_counts)++;
1261                 }
1262                 attach_pid(p, PIDTYPE_PID, pid);
1263                 nr_threads++;
1264         }
1265
1266         total_forks++;
1267         spin_unlock(&current->sighand->siglock);
1268         write_unlock_irq(&tasklist_lock);
1269         proc_fork_connector(p);
1270         cgroup_post_fork(p);
1271         return p;
1272
1273 bad_fork_free_graph:
1274         ftrace_graph_exit_task(p);
1275 bad_fork_free_pid:
1276         if (pid != &init_struct_pid)
1277                 free_pid(pid);
1278 bad_fork_cleanup_io:
1279         put_io_context(p->io_context);
1280 bad_fork_cleanup_namespaces:
1281         exit_task_namespaces(p);
1282 bad_fork_cleanup_mm:
1283         if (p->mm)
1284                 mmput(p->mm);
1285 bad_fork_cleanup_signal:
1286         cleanup_signal(p);
1287 bad_fork_cleanup_sighand:
1288         __cleanup_sighand(p->sighand);
1289 bad_fork_cleanup_fs:
1290         exit_fs(p); /* blocking */
1291 bad_fork_cleanup_files:
1292         exit_files(p); /* blocking */
1293 bad_fork_cleanup_semundo:
1294         exit_sem(p);
1295 bad_fork_cleanup_audit:
1296         audit_free(p);
1297 bad_fork_cleanup_policy:
1298 #ifdef CONFIG_NUMA
1299         mpol_put(p->mempolicy);
1300 bad_fork_cleanup_cgroup:
1301 #endif
1302         cgroup_exit(p, cgroup_callbacks_done);
1303         delayacct_tsk_free(p);
1304         if (p->binfmt)
1305                 module_put(p->binfmt->module);
1306 bad_fork_cleanup_put_domain:
1307         module_put(task_thread_info(p)->exec_domain->module);
1308 bad_fork_cleanup_count:
1309         atomic_dec(&p->cred->user->processes);
1310         put_cred(p->real_cred);
1311         put_cred(p->cred);
1312 bad_fork_free:
1313         free_task(p);
1314 fork_out:
1315         return ERR_PTR(retval);
1316 }
1317
1318 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1319 {
1320         memset(regs, 0, sizeof(struct pt_regs));
1321         return regs;
1322 }
1323
1324 struct task_struct * __cpuinit fork_idle(int cpu)
1325 {
1326         struct task_struct *task;
1327         struct pt_regs regs;
1328
1329         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1330                             &init_struct_pid, 0);
1331         if (!IS_ERR(task))
1332                 init_idle(task, cpu);
1333
1334         return task;
1335 }
1336
1337 /*
1338  *  Ok, this is the main fork-routine.
1339  *
1340  * It copies the process, and if successful kick-starts
1341  * it and waits for it to finish using the VM if required.
1342  */
1343 long do_fork(unsigned long clone_flags,
1344               unsigned long stack_start,
1345               struct pt_regs *regs,
1346               unsigned long stack_size,
1347               int __user *parent_tidptr,
1348               int __user *child_tidptr)
1349 {
1350         struct task_struct *p;
1351         int trace = 0;
1352         long nr;
1353
1354         /*
1355          * Do some preliminary argument and permissions checking before we
1356          * actually start allocating stuff
1357          */
1358         if (clone_flags & CLONE_NEWUSER) {
1359                 if (clone_flags & CLONE_THREAD)
1360                         return -EINVAL;
1361                 /* hopefully this check will go away when userns support is
1362                  * complete
1363                  */
1364                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1365                                 !capable(CAP_SETGID))
1366                         return -EPERM;
1367         }
1368
1369         /*
1370          * We hope to recycle these flags after 2.6.26
1371          */
1372         if (unlikely(clone_flags & CLONE_STOPPED)) {
1373                 static int __read_mostly count = 100;
1374
1375                 if (count > 0 && printk_ratelimit()) {
1376                         char comm[TASK_COMM_LEN];
1377
1378                         count--;
1379                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1380                                         "clone flags 0x%lx\n",
1381                                 get_task_comm(comm, current),
1382                                 clone_flags & CLONE_STOPPED);
1383                 }
1384         }
1385
1386         /*
1387          * When called from kernel_thread, don't do user tracing stuff.
1388          */
1389         if (likely(user_mode(regs)))
1390                 trace = tracehook_prepare_clone(clone_flags);
1391
1392         p = copy_process(clone_flags, stack_start, regs, stack_size,
1393                          child_tidptr, NULL, trace);
1394         /*
1395          * Do this prior waking up the new thread - the thread pointer
1396          * might get invalid after that point, if the thread exits quickly.
1397          */
1398         if (!IS_ERR(p)) {
1399                 struct completion vfork;
1400
1401                 trace_sched_process_fork(current, p);
1402
1403                 nr = task_pid_vnr(p);
1404
1405                 if (clone_flags & CLONE_PARENT_SETTID)
1406                         put_user(nr, parent_tidptr);
1407
1408                 if (clone_flags & CLONE_VFORK) {
1409                         p->vfork_done = &vfork;
1410                         init_completion(&vfork);
1411                 }
1412
1413                 audit_finish_fork(p);
1414                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1415
1416                 /*
1417                  * We set PF_STARTING at creation in case tracing wants to
1418                  * use this to distinguish a fully live task from one that
1419                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1420                  * clear it and set the child going.
1421                  */
1422                 p->flags &= ~PF_STARTING;
1423
1424                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1425                         /*
1426                          * We'll start up with an immediate SIGSTOP.
1427                          */
1428                         sigaddset(&p->pending.signal, SIGSTOP);
1429                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1430                         __set_task_state(p, TASK_STOPPED);
1431                 } else {
1432                         wake_up_new_task(p, clone_flags);
1433                 }
1434
1435                 tracehook_report_clone_complete(trace, regs,
1436                                                 clone_flags, nr, p);
1437
1438                 if (clone_flags & CLONE_VFORK) {
1439                         freezer_do_not_count();
1440                         wait_for_completion(&vfork);
1441                         freezer_count();
1442                         tracehook_report_vfork_done(p, nr);
1443                 }
1444         } else {
1445                 nr = PTR_ERR(p);
1446         }
1447         return nr;
1448 }
1449
1450 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1451 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1452 #endif
1453
1454 static void sighand_ctor(void *data)
1455 {
1456         struct sighand_struct *sighand = data;
1457
1458         spin_lock_init(&sighand->siglock);
1459         init_waitqueue_head(&sighand->signalfd_wqh);
1460 }
1461
1462 void __init proc_caches_init(void)
1463 {
1464         sighand_cachep = kmem_cache_create("sighand_cache",
1465                         sizeof(struct sighand_struct), 0,
1466                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1467                         sighand_ctor);
1468         signal_cachep = kmem_cache_create("signal_cache",
1469                         sizeof(struct signal_struct), 0,
1470                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1471         files_cachep = kmem_cache_create("files_cache",
1472                         sizeof(struct files_struct), 0,
1473                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1474         fs_cachep = kmem_cache_create("fs_cache",
1475                         sizeof(struct fs_struct), 0,
1476                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1477         mm_cachep = kmem_cache_create("mm_struct",
1478                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1479                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1480         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1481         mmap_init();
1482 }
1483
1484 /*
1485  * Check constraints on flags passed to the unshare system call and
1486  * force unsharing of additional process context as appropriate.
1487  */
1488 static void check_unshare_flags(unsigned long *flags_ptr)
1489 {
1490         /*
1491          * If unsharing a thread from a thread group, must also
1492          * unshare vm.
1493          */
1494         if (*flags_ptr & CLONE_THREAD)
1495                 *flags_ptr |= CLONE_VM;
1496
1497         /*
1498          * If unsharing vm, must also unshare signal handlers.
1499          */
1500         if (*flags_ptr & CLONE_VM)
1501                 *flags_ptr |= CLONE_SIGHAND;
1502
1503         /*
1504          * If unsharing signal handlers and the task was created
1505          * using CLONE_THREAD, then must unshare the thread
1506          */
1507         if ((*flags_ptr & CLONE_SIGHAND) &&
1508             (atomic_read(&current->signal->count) > 1))
1509                 *flags_ptr |= CLONE_THREAD;
1510
1511         /*
1512          * If unsharing namespace, must also unshare filesystem information.
1513          */
1514         if (*flags_ptr & CLONE_NEWNS)
1515                 *flags_ptr |= CLONE_FS;
1516 }
1517
1518 /*
1519  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1520  */
1521 static int unshare_thread(unsigned long unshare_flags)
1522 {
1523         if (unshare_flags & CLONE_THREAD)
1524                 return -EINVAL;
1525
1526         return 0;
1527 }
1528
1529 /*
1530  * Unshare the filesystem structure if it is being shared
1531  */
1532 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1533 {
1534         struct fs_struct *fs = current->fs;
1535
1536         if (!(unshare_flags & CLONE_FS) || !fs)
1537                 return 0;
1538
1539         /* don't need lock here; in the worst case we'll do useless copy */
1540         if (fs->users == 1)
1541                 return 0;
1542
1543         *new_fsp = copy_fs_struct(fs);
1544         if (!*new_fsp)
1545                 return -ENOMEM;
1546
1547         return 0;
1548 }
1549
1550 /*
1551  * Unsharing of sighand is not supported yet
1552  */
1553 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1554 {
1555         struct sighand_struct *sigh = current->sighand;
1556
1557         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1558                 return -EINVAL;
1559         else
1560                 return 0;
1561 }
1562
1563 /*
1564  * Unshare vm if it is being shared
1565  */
1566 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1567 {
1568         struct mm_struct *mm = current->mm;
1569
1570         if ((unshare_flags & CLONE_VM) &&
1571             (mm && atomic_read(&mm->mm_users) > 1)) {
1572                 return -EINVAL;
1573         }
1574
1575         return 0;
1576 }
1577
1578 /*
1579  * Unshare file descriptor table if it is being shared
1580  */
1581 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1582 {
1583         struct files_struct *fd = current->files;
1584         int error = 0;
1585
1586         if ((unshare_flags & CLONE_FILES) &&
1587             (fd && atomic_read(&fd->count) > 1)) {
1588                 *new_fdp = dup_fd(fd, &error);
1589                 if (!*new_fdp)
1590                         return error;
1591         }
1592
1593         return 0;
1594 }
1595
1596 /*
1597  * unshare allows a process to 'unshare' part of the process
1598  * context which was originally shared using clone.  copy_*
1599  * functions used by do_fork() cannot be used here directly
1600  * because they modify an inactive task_struct that is being
1601  * constructed. Here we are modifying the current, active,
1602  * task_struct.
1603  */
1604 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1605 {
1606         int err = 0;
1607         struct fs_struct *fs, *new_fs = NULL;
1608         struct sighand_struct *new_sigh = NULL;
1609         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1610         struct files_struct *fd, *new_fd = NULL;
1611         struct nsproxy *new_nsproxy = NULL;
1612         int do_sysvsem = 0;
1613
1614         check_unshare_flags(&unshare_flags);
1615
1616         /* Return -EINVAL for all unsupported flags */
1617         err = -EINVAL;
1618         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1619                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1620                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1621                 goto bad_unshare_out;
1622
1623         /*
1624          * CLONE_NEWIPC must also detach from the undolist: after switching
1625          * to a new ipc namespace, the semaphore arrays from the old
1626          * namespace are unreachable.
1627          */
1628         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1629                 do_sysvsem = 1;
1630         if ((err = unshare_thread(unshare_flags)))
1631                 goto bad_unshare_out;
1632         if ((err = unshare_fs(unshare_flags, &new_fs)))
1633                 goto bad_unshare_cleanup_thread;
1634         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1635                 goto bad_unshare_cleanup_fs;
1636         if ((err = unshare_vm(unshare_flags, &new_mm)))
1637                 goto bad_unshare_cleanup_sigh;
1638         if ((err = unshare_fd(unshare_flags, &new_fd)))
1639                 goto bad_unshare_cleanup_vm;
1640         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1641                         new_fs)))
1642                 goto bad_unshare_cleanup_fd;
1643
1644         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1645                 if (do_sysvsem) {
1646                         /*
1647                          * CLONE_SYSVSEM is equivalent to sys_exit().
1648                          */
1649                         exit_sem(current);
1650                 }
1651
1652                 if (new_nsproxy) {
1653                         switch_task_namespaces(current, new_nsproxy);
1654                         new_nsproxy = NULL;
1655                 }
1656
1657                 task_lock(current);
1658
1659                 if (new_fs) {
1660                         fs = current->fs;
1661                         write_lock(&fs->lock);
1662                         current->fs = new_fs;
1663                         if (--fs->users)
1664                                 new_fs = NULL;
1665                         else
1666                                 new_fs = fs;
1667                         write_unlock(&fs->lock);
1668                 }
1669
1670                 if (new_mm) {
1671                         mm = current->mm;
1672                         active_mm = current->active_mm;
1673                         current->mm = new_mm;
1674                         current->active_mm = new_mm;
1675                         activate_mm(active_mm, new_mm);
1676                         new_mm = mm;
1677                 }
1678
1679                 if (new_fd) {
1680                         fd = current->files;
1681                         current->files = new_fd;
1682                         new_fd = fd;
1683                 }
1684
1685                 task_unlock(current);
1686         }
1687
1688         if (new_nsproxy)
1689                 put_nsproxy(new_nsproxy);
1690
1691 bad_unshare_cleanup_fd:
1692         if (new_fd)
1693                 put_files_struct(new_fd);
1694
1695 bad_unshare_cleanup_vm:
1696         if (new_mm)
1697                 mmput(new_mm);
1698
1699 bad_unshare_cleanup_sigh:
1700         if (new_sigh)
1701                 if (atomic_dec_and_test(&new_sigh->count))
1702                         kmem_cache_free(sighand_cachep, new_sigh);
1703
1704 bad_unshare_cleanup_fs:
1705         if (new_fs)
1706                 free_fs_struct(new_fs);
1707
1708 bad_unshare_cleanup_thread:
1709 bad_unshare_out:
1710         return err;
1711 }
1712
1713 /*
1714  *      Helper to unshare the files of the current task.
1715  *      We don't want to expose copy_files internals to
1716  *      the exec layer of the kernel.
1717  */
1718
1719 int unshare_files(struct files_struct **displaced)
1720 {
1721         struct task_struct *task = current;
1722         struct files_struct *copy = NULL;
1723         int error;
1724
1725         error = unshare_fd(CLONE_FILES, &copy);
1726         if (error || !copy) {
1727                 *displaced = NULL;
1728                 return error;
1729         }
1730         *displaced = task->files;
1731         task_lock(task);
1732         task->files = copy;
1733         task_unlock(task);
1734         return 0;
1735 }