Merge branch 'linus' into core/percpu
[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 <trace/sched.h>
64 #include <linux/magic.h>
65
66 #include <asm/pgtable.h>
67 #include <asm/pgalloc.h>
68 #include <asm/uaccess.h>
69 #include <asm/mmu_context.h>
70 #include <asm/cacheflush.h>
71 #include <asm/tlbflush.h>
72
73 /*
74  * Protected counters by write_lock_irq(&tasklist_lock)
75  */
76 unsigned long total_forks;      /* Handle normal Linux uptimes. */
77 int nr_threads;                 /* The idle threads do not count.. */
78
79 int max_threads;                /* tunable limit on nr_threads */
80
81 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
82
83 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
84
85 DEFINE_TRACE(sched_process_fork);
86
87 int nr_processes(void)
88 {
89         int cpu;
90         int total = 0;
91
92         for_each_online_cpu(cpu)
93                 total += per_cpu(process_counts, cpu);
94
95         return total;
96 }
97
98 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
99 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
100 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
101 static struct kmem_cache *task_struct_cachep;
102 #endif
103
104 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
105 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
106 {
107 #ifdef CONFIG_DEBUG_STACK_USAGE
108         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
109 #else
110         gfp_t mask = GFP_KERNEL;
111 #endif
112         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
113 }
114
115 static inline void free_thread_info(struct thread_info *ti)
116 {
117         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
118 }
119 #endif
120
121 /* SLAB cache for signal_struct structures (tsk->signal) */
122 static struct kmem_cache *signal_cachep;
123
124 /* SLAB cache for sighand_struct structures (tsk->sighand) */
125 struct kmem_cache *sighand_cachep;
126
127 /* SLAB cache for files_struct structures (tsk->files) */
128 struct kmem_cache *files_cachep;
129
130 /* SLAB cache for fs_struct structures (tsk->fs) */
131 struct kmem_cache *fs_cachep;
132
133 /* SLAB cache for vm_area_struct structures */
134 struct kmem_cache *vm_area_cachep;
135
136 /* SLAB cache for mm_struct structures (tsk->mm) */
137 static struct kmem_cache *mm_cachep;
138
139 void free_task(struct task_struct *tsk)
140 {
141         prop_local_destroy_single(&tsk->dirties);
142         free_thread_info(tsk->stack);
143         rt_mutex_debug_task_free(tsk);
144         ftrace_graph_exit_task(tsk);
145         free_task_struct(tsk);
146 }
147 EXPORT_SYMBOL(free_task);
148
149 void __put_task_struct(struct task_struct *tsk)
150 {
151         WARN_ON(!tsk->exit_state);
152         WARN_ON(atomic_read(&tsk->usage));
153         WARN_ON(tsk == current);
154
155         put_cred(tsk->real_cred);
156         put_cred(tsk->cred);
157         delayacct_tsk_free(tsk);
158
159         if (!profile_handoff_task(tsk))
160                 free_task(tsk);
161 }
162
163 /*
164  * macro override instead of weak attribute alias, to workaround
165  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
166  */
167 #ifndef arch_task_cache_init
168 #define arch_task_cache_init()
169 #endif
170
171 void __init fork_init(unsigned long mempages)
172 {
173 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
174 #ifndef ARCH_MIN_TASKALIGN
175 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
176 #endif
177         /* create a slab on which task_structs can be allocated */
178         task_struct_cachep =
179                 kmem_cache_create("task_struct", sizeof(struct task_struct),
180                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
181 #endif
182
183         /* do the arch specific task caches init */
184         arch_task_cache_init();
185
186         /*
187          * The default maximum number of threads is set to a safe
188          * value: the thread structures can take up at most half
189          * of memory.
190          */
191         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
192
193         /*
194          * we need to allow at least 20 threads to boot a system
195          */
196         if(max_threads < 20)
197                 max_threads = 20;
198
199         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
200         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
201         init_task.signal->rlim[RLIMIT_SIGPENDING] =
202                 init_task.signal->rlim[RLIMIT_NPROC];
203 }
204
205 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
206                                                struct task_struct *src)
207 {
208         *dst = *src;
209         return 0;
210 }
211
212 static struct task_struct *dup_task_struct(struct task_struct *orig)
213 {
214         struct task_struct *tsk;
215         struct thread_info *ti;
216         unsigned long *stackend;
217
218         int err;
219
220         prepare_to_copy(orig);
221
222         tsk = alloc_task_struct();
223         if (!tsk)
224                 return NULL;
225
226         ti = alloc_thread_info(tsk);
227         if (!ti) {
228                 free_task_struct(tsk);
229                 return NULL;
230         }
231
232         err = arch_dup_task_struct(tsk, orig);
233         if (err)
234                 goto out;
235
236         tsk->stack = ti;
237
238         err = prop_local_init_single(&tsk->dirties);
239         if (err)
240                 goto out;
241
242         setup_thread_stack(tsk, orig);
243         stackend = end_of_stack(tsk);
244         *stackend = STACK_END_MAGIC;    /* for overflow detection */
245
246 #ifdef CONFIG_CC_STACKPROTECTOR
247         tsk->stack_canary = get_random_int();
248 #endif
249
250         /* One for us, one for whoever does the "release_task()" (usually parent) */
251         atomic_set(&tsk->usage,2);
252         atomic_set(&tsk->fs_excl, 0);
253 #ifdef CONFIG_BLK_DEV_IO_TRACE
254         tsk->btrace_seq = 0;
255 #endif
256         tsk->splice_pipe = NULL;
257         return tsk;
258
259 out:
260         free_thread_info(ti);
261         free_task_struct(tsk);
262         return NULL;
263 }
264
265 #ifdef CONFIG_MMU
266 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
267 {
268         struct vm_area_struct *mpnt, *tmp, **pprev;
269         struct rb_node **rb_link, *rb_parent;
270         int retval;
271         unsigned long charge;
272         struct mempolicy *pol;
273
274         down_write(&oldmm->mmap_sem);
275         flush_cache_dup_mm(oldmm);
276         /*
277          * Not linked in yet - no deadlock potential:
278          */
279         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
280
281         mm->locked_vm = 0;
282         mm->mmap = NULL;
283         mm->mmap_cache = NULL;
284         mm->free_area_cache = oldmm->mmap_base;
285         mm->cached_hole_size = ~0UL;
286         mm->map_count = 0;
287         cpus_clear(mm->cpu_vm_mask);
288         mm->mm_rb = RB_ROOT;
289         rb_link = &mm->mm_rb.rb_node;
290         rb_parent = NULL;
291         pprev = &mm->mmap;
292
293         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
294                 struct file *file;
295
296                 if (mpnt->vm_flags & VM_DONTCOPY) {
297                         long pages = vma_pages(mpnt);
298                         mm->total_vm -= pages;
299                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
300                                                                 -pages);
301                         continue;
302                 }
303                 charge = 0;
304                 if (mpnt->vm_flags & VM_ACCOUNT) {
305                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
306                         if (security_vm_enough_memory(len))
307                                 goto fail_nomem;
308                         charge = len;
309                 }
310                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
311                 if (!tmp)
312                         goto fail_nomem;
313                 *tmp = *mpnt;
314                 pol = mpol_dup(vma_policy(mpnt));
315                 retval = PTR_ERR(pol);
316                 if (IS_ERR(pol))
317                         goto fail_nomem_policy;
318                 vma_set_policy(tmp, pol);
319                 tmp->vm_flags &= ~VM_LOCKED;
320                 tmp->vm_mm = mm;
321                 tmp->vm_next = NULL;
322                 anon_vma_link(tmp);
323                 file = tmp->vm_file;
324                 if (file) {
325                         struct inode *inode = file->f_path.dentry->d_inode;
326                         struct address_space *mapping = file->f_mapping;
327
328                         get_file(file);
329                         if (tmp->vm_flags & VM_DENYWRITE)
330                                 atomic_dec(&inode->i_writecount);
331                         spin_lock(&mapping->i_mmap_lock);
332                         if (tmp->vm_flags & VM_SHARED)
333                                 mapping->i_mmap_writable++;
334                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
335                         flush_dcache_mmap_lock(mapping);
336                         /* insert tmp into the share list, just after mpnt */
337                         vma_prio_tree_add(tmp, mpnt);
338                         flush_dcache_mmap_unlock(mapping);
339                         spin_unlock(&mapping->i_mmap_lock);
340                 }
341
342                 /*
343                  * Clear hugetlb-related page reserves for children. This only
344                  * affects MAP_PRIVATE mappings. Faults generated by the child
345                  * are not guaranteed to succeed, even if read-only
346                  */
347                 if (is_vm_hugetlb_page(tmp))
348                         reset_vma_resv_huge_pages(tmp);
349
350                 /*
351                  * Link in the new vma and copy the page table entries.
352                  */
353                 *pprev = tmp;
354                 pprev = &tmp->vm_next;
355
356                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
357                 rb_link = &tmp->vm_rb.rb_right;
358                 rb_parent = &tmp->vm_rb;
359
360                 mm->map_count++;
361                 retval = copy_page_range(mm, oldmm, mpnt);
362
363                 if (tmp->vm_ops && tmp->vm_ops->open)
364                         tmp->vm_ops->open(tmp);
365
366                 if (retval)
367                         goto out;
368         }
369         /* a new mm has just been created */
370         arch_dup_mmap(oldmm, mm);
371         retval = 0;
372 out:
373         up_write(&mm->mmap_sem);
374         flush_tlb_mm(oldmm);
375         up_write(&oldmm->mmap_sem);
376         return retval;
377 fail_nomem_policy:
378         kmem_cache_free(vm_area_cachep, tmp);
379 fail_nomem:
380         retval = -ENOMEM;
381         vm_unacct_memory(charge);
382         goto out;
383 }
384
385 static inline int mm_alloc_pgd(struct mm_struct * mm)
386 {
387         mm->pgd = pgd_alloc(mm);
388         if (unlikely(!mm->pgd))
389                 return -ENOMEM;
390         return 0;
391 }
392
393 static inline void mm_free_pgd(struct mm_struct * mm)
394 {
395         pgd_free(mm, mm->pgd);
396 }
397 #else
398 #define dup_mmap(mm, oldmm)     (0)
399 #define mm_alloc_pgd(mm)        (0)
400 #define mm_free_pgd(mm)
401 #endif /* CONFIG_MMU */
402
403 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
404
405 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
406 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
407
408 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
409
410 static int __init coredump_filter_setup(char *s)
411 {
412         default_dump_filter =
413                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
414                 MMF_DUMP_FILTER_MASK;
415         return 1;
416 }
417
418 __setup("coredump_filter=", coredump_filter_setup);
419
420 #include <linux/init_task.h>
421
422 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
423 {
424         atomic_set(&mm->mm_users, 1);
425         atomic_set(&mm->mm_count, 1);
426         init_rwsem(&mm->mmap_sem);
427         INIT_LIST_HEAD(&mm->mmlist);
428         mm->flags = (current->mm) ? current->mm->flags : default_dump_filter;
429         mm->core_state = NULL;
430         mm->nr_ptes = 0;
431         set_mm_counter(mm, file_rss, 0);
432         set_mm_counter(mm, anon_rss, 0);
433         spin_lock_init(&mm->page_table_lock);
434         spin_lock_init(&mm->ioctx_lock);
435         INIT_HLIST_HEAD(&mm->ioctx_list);
436         mm->free_area_cache = TASK_UNMAPPED_BASE;
437         mm->cached_hole_size = ~0UL;
438         mm_init_owner(mm, p);
439
440         if (likely(!mm_alloc_pgd(mm))) {
441                 mm->def_flags = 0;
442                 mmu_notifier_mm_init(mm);
443                 return mm;
444         }
445
446         free_mm(mm);
447         return NULL;
448 }
449
450 /*
451  * Allocate and initialize an mm_struct.
452  */
453 struct mm_struct * mm_alloc(void)
454 {
455         struct mm_struct * mm;
456
457         mm = allocate_mm();
458         if (mm) {
459                 memset(mm, 0, sizeof(*mm));
460                 mm = mm_init(mm, current);
461         }
462         return mm;
463 }
464
465 /*
466  * Called when the last reference to the mm
467  * is dropped: either by a lazy thread or by
468  * mmput. Free the page directory and the mm.
469  */
470 void __mmdrop(struct mm_struct *mm)
471 {
472         BUG_ON(mm == &init_mm);
473         mm_free_pgd(mm);
474         destroy_context(mm);
475         mmu_notifier_mm_destroy(mm);
476         free_mm(mm);
477 }
478 EXPORT_SYMBOL_GPL(__mmdrop);
479
480 /*
481  * Decrement the use count and release all resources for an mm.
482  */
483 void mmput(struct mm_struct *mm)
484 {
485         might_sleep();
486
487         if (atomic_dec_and_test(&mm->mm_users)) {
488                 exit_aio(mm);
489                 exit_mmap(mm);
490                 set_mm_exe_file(mm, NULL);
491                 if (!list_empty(&mm->mmlist)) {
492                         spin_lock(&mmlist_lock);
493                         list_del(&mm->mmlist);
494                         spin_unlock(&mmlist_lock);
495                 }
496                 put_swap_token(mm);
497                 mmdrop(mm);
498         }
499 }
500 EXPORT_SYMBOL_GPL(mmput);
501
502 /**
503  * get_task_mm - acquire a reference to the task's mm
504  *
505  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
506  * this kernel workthread has transiently adopted a user mm with use_mm,
507  * to do its AIO) is not set and if so returns a reference to it, after
508  * bumping up the use count.  User must release the mm via mmput()
509  * after use.  Typically used by /proc and ptrace.
510  */
511 struct mm_struct *get_task_mm(struct task_struct *task)
512 {
513         struct mm_struct *mm;
514
515         task_lock(task);
516         mm = task->mm;
517         if (mm) {
518                 if (task->flags & PF_KTHREAD)
519                         mm = NULL;
520                 else
521                         atomic_inc(&mm->mm_users);
522         }
523         task_unlock(task);
524         return mm;
525 }
526 EXPORT_SYMBOL_GPL(get_task_mm);
527
528 /* Please note the differences between mmput and mm_release.
529  * mmput is called whenever we stop holding onto a mm_struct,
530  * error success whatever.
531  *
532  * mm_release is called after a mm_struct has been removed
533  * from the current process.
534  *
535  * This difference is important for error handling, when we
536  * only half set up a mm_struct for a new process and need to restore
537  * the old one.  Because we mmput the new mm_struct before
538  * restoring the old one. . .
539  * Eric Biederman 10 January 1998
540  */
541 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
542 {
543         struct completion *vfork_done = tsk->vfork_done;
544
545         /* Get rid of any futexes when releasing the mm */
546 #ifdef CONFIG_FUTEX
547         if (unlikely(tsk->robust_list))
548                 exit_robust_list(tsk);
549 #ifdef CONFIG_COMPAT
550         if (unlikely(tsk->compat_robust_list))
551                 compat_exit_robust_list(tsk);
552 #endif
553 #endif
554
555         /* Get rid of any cached register state */
556         deactivate_mm(tsk, mm);
557
558         /* notify parent sleeping on vfork() */
559         if (vfork_done) {
560                 tsk->vfork_done = NULL;
561                 complete(vfork_done);
562         }
563
564         /*
565          * If we're exiting normally, clear a user-space tid field if
566          * requested.  We leave this alone when dying by signal, to leave
567          * the value intact in a core dump, and to save the unnecessary
568          * trouble otherwise.  Userland only wants this done for a sys_exit.
569          */
570         if (tsk->clear_child_tid
571             && !(tsk->flags & PF_SIGNALED)
572             && atomic_read(&mm->mm_users) > 1) {
573                 u32 __user * tidptr = tsk->clear_child_tid;
574                 tsk->clear_child_tid = NULL;
575
576                 /*
577                  * We don't check the error code - if userspace has
578                  * not set up a proper pointer then tough luck.
579                  */
580                 put_user(0, tidptr);
581                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
582         }
583 }
584
585 /*
586  * Allocate a new mm structure and copy contents from the
587  * mm structure of the passed in task structure.
588  */
589 struct mm_struct *dup_mm(struct task_struct *tsk)
590 {
591         struct mm_struct *mm, *oldmm = current->mm;
592         int err;
593
594         if (!oldmm)
595                 return NULL;
596
597         mm = allocate_mm();
598         if (!mm)
599                 goto fail_nomem;
600
601         memcpy(mm, oldmm, sizeof(*mm));
602
603         /* Initializing for Swap token stuff */
604         mm->token_priority = 0;
605         mm->last_interval = 0;
606
607         if (!mm_init(mm, tsk))
608                 goto fail_nomem;
609
610         if (init_new_context(tsk, mm))
611                 goto fail_nocontext;
612
613         dup_mm_exe_file(oldmm, mm);
614
615         err = dup_mmap(mm, oldmm);
616         if (err)
617                 goto free_pt;
618
619         mm->hiwater_rss = get_mm_rss(mm);
620         mm->hiwater_vm = mm->total_vm;
621
622         return mm;
623
624 free_pt:
625         mmput(mm);
626
627 fail_nomem:
628         return NULL;
629
630 fail_nocontext:
631         /*
632          * If init_new_context() failed, we cannot use mmput() to free the mm
633          * because it calls destroy_context()
634          */
635         mm_free_pgd(mm);
636         free_mm(mm);
637         return NULL;
638 }
639
640 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
641 {
642         struct mm_struct * mm, *oldmm;
643         int retval;
644
645         tsk->min_flt = tsk->maj_flt = 0;
646         tsk->nvcsw = tsk->nivcsw = 0;
647
648         tsk->mm = NULL;
649         tsk->active_mm = NULL;
650
651         /*
652          * Are we cloning a kernel thread?
653          *
654          * We need to steal a active VM for that..
655          */
656         oldmm = current->mm;
657         if (!oldmm)
658                 return 0;
659
660         if (clone_flags & CLONE_VM) {
661                 atomic_inc(&oldmm->mm_users);
662                 mm = oldmm;
663                 goto good_mm;
664         }
665
666         retval = -ENOMEM;
667         mm = dup_mm(tsk);
668         if (!mm)
669                 goto fail_nomem;
670
671 good_mm:
672         /* Initializing for Swap token stuff */
673         mm->token_priority = 0;
674         mm->last_interval = 0;
675
676         tsk->mm = mm;
677         tsk->active_mm = mm;
678         return 0;
679
680 fail_nomem:
681         return retval;
682 }
683
684 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
685 {
686         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
687         /* We don't need to lock fs - think why ;-) */
688         if (fs) {
689                 atomic_set(&fs->count, 1);
690                 rwlock_init(&fs->lock);
691                 fs->umask = old->umask;
692                 read_lock(&old->lock);
693                 fs->root = old->root;
694                 path_get(&old->root);
695                 fs->pwd = old->pwd;
696                 path_get(&old->pwd);
697                 read_unlock(&old->lock);
698         }
699         return fs;
700 }
701
702 struct fs_struct *copy_fs_struct(struct fs_struct *old)
703 {
704         return __copy_fs_struct(old);
705 }
706
707 EXPORT_SYMBOL_GPL(copy_fs_struct);
708
709 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
710 {
711         if (clone_flags & CLONE_FS) {
712                 atomic_inc(&current->fs->count);
713                 return 0;
714         }
715         tsk->fs = __copy_fs_struct(current->fs);
716         if (!tsk->fs)
717                 return -ENOMEM;
718         return 0;
719 }
720
721 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
722 {
723         struct files_struct *oldf, *newf;
724         int error = 0;
725
726         /*
727          * A background process may not have any files ...
728          */
729         oldf = current->files;
730         if (!oldf)
731                 goto out;
732
733         if (clone_flags & CLONE_FILES) {
734                 atomic_inc(&oldf->count);
735                 goto out;
736         }
737
738         newf = dup_fd(oldf, &error);
739         if (!newf)
740                 goto out;
741
742         tsk->files = newf;
743         error = 0;
744 out:
745         return error;
746 }
747
748 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
749 {
750 #ifdef CONFIG_BLOCK
751         struct io_context *ioc = current->io_context;
752
753         if (!ioc)
754                 return 0;
755         /*
756          * Share io context with parent, if CLONE_IO is set
757          */
758         if (clone_flags & CLONE_IO) {
759                 tsk->io_context = ioc_task_link(ioc);
760                 if (unlikely(!tsk->io_context))
761                         return -ENOMEM;
762         } else if (ioprio_valid(ioc->ioprio)) {
763                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
764                 if (unlikely(!tsk->io_context))
765                         return -ENOMEM;
766
767                 tsk->io_context->ioprio = ioc->ioprio;
768         }
769 #endif
770         return 0;
771 }
772
773 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
774 {
775         struct sighand_struct *sig;
776
777         if (clone_flags & CLONE_SIGHAND) {
778                 atomic_inc(&current->sighand->count);
779                 return 0;
780         }
781         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
782         rcu_assign_pointer(tsk->sighand, sig);
783         if (!sig)
784                 return -ENOMEM;
785         atomic_set(&sig->count, 1);
786         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
787         return 0;
788 }
789
790 void __cleanup_sighand(struct sighand_struct *sighand)
791 {
792         if (atomic_dec_and_test(&sighand->count))
793                 kmem_cache_free(sighand_cachep, sighand);
794 }
795
796
797 /*
798  * Initialize POSIX timer handling for a thread group.
799  */
800 static void posix_cpu_timers_init_group(struct signal_struct *sig)
801 {
802         /* Thread group counters. */
803         thread_group_cputime_init(sig);
804
805         /* Expiration times and increments. */
806         sig->it_virt_expires = cputime_zero;
807         sig->it_virt_incr = cputime_zero;
808         sig->it_prof_expires = cputime_zero;
809         sig->it_prof_incr = cputime_zero;
810
811         /* Cached expiration times. */
812         sig->cputime_expires.prof_exp = cputime_zero;
813         sig->cputime_expires.virt_exp = cputime_zero;
814         sig->cputime_expires.sched_exp = 0;
815
816         /* The timer lists. */
817         INIT_LIST_HEAD(&sig->cpu_timers[0]);
818         INIT_LIST_HEAD(&sig->cpu_timers[1]);
819         INIT_LIST_HEAD(&sig->cpu_timers[2]);
820 }
821
822 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
823 {
824         struct signal_struct *sig;
825
826         if (clone_flags & CLONE_THREAD) {
827                 atomic_inc(&current->signal->count);
828                 atomic_inc(&current->signal->live);
829                 return 0;
830         }
831         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
832
833         if (sig)
834                 posix_cpu_timers_init_group(sig);
835
836         tsk->signal = sig;
837         if (!sig)
838                 return -ENOMEM;
839
840         atomic_set(&sig->count, 1);
841         atomic_set(&sig->live, 1);
842         init_waitqueue_head(&sig->wait_chldexit);
843         sig->flags = 0;
844         sig->group_exit_code = 0;
845         sig->group_exit_task = NULL;
846         sig->group_stop_count = 0;
847         sig->curr_target = tsk;
848         init_sigpending(&sig->shared_pending);
849         INIT_LIST_HEAD(&sig->posix_timers);
850
851         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
852         sig->it_real_incr.tv64 = 0;
853         sig->real_timer.function = it_real_fn;
854
855         sig->leader = 0;        /* session leadership doesn't inherit */
856         sig->tty_old_pgrp = NULL;
857         sig->tty = NULL;
858
859         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
860         sig->gtime = cputime_zero;
861         sig->cgtime = cputime_zero;
862         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
863         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
864         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
865         task_io_accounting_init(&sig->ioac);
866         sig->sum_sched_runtime = 0;
867         taskstats_tgid_init(sig);
868
869         task_lock(current->group_leader);
870         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
871         task_unlock(current->group_leader);
872
873         acct_init_pacct(&sig->pacct);
874
875         tty_audit_fork(sig);
876
877         return 0;
878 }
879
880 void __cleanup_signal(struct signal_struct *sig)
881 {
882         thread_group_cputime_free(sig);
883         tty_kref_put(sig->tty);
884         kmem_cache_free(signal_cachep, sig);
885 }
886
887 static void cleanup_signal(struct task_struct *tsk)
888 {
889         struct signal_struct *sig = tsk->signal;
890
891         atomic_dec(&sig->live);
892
893         if (atomic_dec_and_test(&sig->count))
894                 __cleanup_signal(sig);
895 }
896
897 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
898 {
899         unsigned long new_flags = p->flags;
900
901         new_flags &= ~PF_SUPERPRIV;
902         new_flags |= PF_FORKNOEXEC;
903         new_flags |= PF_STARTING;
904         p->flags = new_flags;
905         clear_freeze_flag(p);
906 }
907
908 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
909 {
910         current->clear_child_tid = tidptr;
911
912         return task_pid_vnr(current);
913 }
914
915 static void rt_mutex_init_task(struct task_struct *p)
916 {
917         spin_lock_init(&p->pi_lock);
918 #ifdef CONFIG_RT_MUTEXES
919         plist_head_init(&p->pi_waiters, &p->pi_lock);
920         p->pi_blocked_on = NULL;
921 #endif
922 }
923
924 #ifdef CONFIG_MM_OWNER
925 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
926 {
927         mm->owner = p;
928 }
929 #endif /* CONFIG_MM_OWNER */
930
931 /*
932  * Initialize POSIX timer handling for a single task.
933  */
934 static void posix_cpu_timers_init(struct task_struct *tsk)
935 {
936         tsk->cputime_expires.prof_exp = cputime_zero;
937         tsk->cputime_expires.virt_exp = cputime_zero;
938         tsk->cputime_expires.sched_exp = 0;
939         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
940         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
941         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
942 }
943
944 /*
945  * This creates a new process as a copy of the old one,
946  * but does not actually start it yet.
947  *
948  * It copies the registers, and all the appropriate
949  * parts of the process environment (as per the clone
950  * flags). The actual kick-off is left to the caller.
951  */
952 static struct task_struct *copy_process(unsigned long clone_flags,
953                                         unsigned long stack_start,
954                                         struct pt_regs *regs,
955                                         unsigned long stack_size,
956                                         int __user *child_tidptr,
957                                         struct pid *pid,
958                                         int trace)
959 {
960         int retval;
961         struct task_struct *p;
962         int cgroup_callbacks_done = 0;
963
964         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
965                 return ERR_PTR(-EINVAL);
966
967         /*
968          * Thread groups must share signals as well, and detached threads
969          * can only be started up within the thread group.
970          */
971         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
972                 return ERR_PTR(-EINVAL);
973
974         /*
975          * Shared signal handlers imply shared VM. By way of the above,
976          * thread groups also imply shared VM. Blocking this case allows
977          * for various simplifications in other code.
978          */
979         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
980                 return ERR_PTR(-EINVAL);
981
982         retval = security_task_create(clone_flags);
983         if (retval)
984                 goto fork_out;
985
986         retval = -ENOMEM;
987         p = dup_task_struct(current);
988         if (!p)
989                 goto fork_out;
990
991         rt_mutex_init_task(p);
992
993 #ifdef CONFIG_PROVE_LOCKING
994         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
995         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
996 #endif
997         retval = -EAGAIN;
998         if (atomic_read(&p->real_cred->user->processes) >=
999                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1000                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1001                     p->real_cred->user != INIT_USER)
1002                         goto bad_fork_free;
1003         }
1004
1005         retval = copy_creds(p, clone_flags);
1006         if (retval < 0)
1007                 goto bad_fork_free;
1008
1009         /*
1010          * If multiple threads are within copy_process(), then this check
1011          * triggers too late. This doesn't hurt, the check is only there
1012          * to stop root fork bombs.
1013          */
1014         retval = -EAGAIN;
1015         if (nr_threads >= max_threads)
1016                 goto bad_fork_cleanup_count;
1017
1018         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1019                 goto bad_fork_cleanup_count;
1020
1021         if (p->binfmt && !try_module_get(p->binfmt->module))
1022                 goto bad_fork_cleanup_put_domain;
1023
1024         p->did_exec = 0;
1025         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1026         copy_flags(clone_flags, p);
1027         INIT_LIST_HEAD(&p->children);
1028         INIT_LIST_HEAD(&p->sibling);
1029 #ifdef CONFIG_PREEMPT_RCU
1030         p->rcu_read_lock_nesting = 0;
1031         p->rcu_flipctr_idx = 0;
1032 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1033         p->vfork_done = NULL;
1034         spin_lock_init(&p->alloc_lock);
1035
1036         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1037         init_sigpending(&p->pending);
1038
1039         p->utime = cputime_zero;
1040         p->stime = cputime_zero;
1041         p->gtime = cputime_zero;
1042         p->utimescaled = cputime_zero;
1043         p->stimescaled = cputime_zero;
1044         p->prev_utime = cputime_zero;
1045         p->prev_stime = cputime_zero;
1046
1047         p->default_timer_slack_ns = current->timer_slack_ns;
1048
1049 #ifdef CONFIG_DETECT_SOFTLOCKUP
1050         p->last_switch_count = 0;
1051         p->last_switch_timestamp = 0;
1052 #endif
1053
1054         task_io_accounting_init(&p->ioac);
1055         acct_clear_integrals(p);
1056
1057         posix_cpu_timers_init(p);
1058
1059         p->lock_depth = -1;             /* -1 = no lock */
1060         do_posix_clock_monotonic_gettime(&p->start_time);
1061         p->real_start_time = p->start_time;
1062         monotonic_to_bootbased(&p->real_start_time);
1063         p->io_context = NULL;
1064         p->audit_context = NULL;
1065         cgroup_fork(p);
1066 #ifdef CONFIG_NUMA
1067         p->mempolicy = mpol_dup(p->mempolicy);
1068         if (IS_ERR(p->mempolicy)) {
1069                 retval = PTR_ERR(p->mempolicy);
1070                 p->mempolicy = NULL;
1071                 goto bad_fork_cleanup_cgroup;
1072         }
1073         mpol_fix_fork_child_flag(p);
1074 #endif
1075 #ifdef CONFIG_TRACE_IRQFLAGS
1076         p->irq_events = 0;
1077 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1078         p->hardirqs_enabled = 1;
1079 #else
1080         p->hardirqs_enabled = 0;
1081 #endif
1082         p->hardirq_enable_ip = 0;
1083         p->hardirq_enable_event = 0;
1084         p->hardirq_disable_ip = _THIS_IP_;
1085         p->hardirq_disable_event = 0;
1086         p->softirqs_enabled = 1;
1087         p->softirq_enable_ip = _THIS_IP_;
1088         p->softirq_enable_event = 0;
1089         p->softirq_disable_ip = 0;
1090         p->softirq_disable_event = 0;
1091         p->hardirq_context = 0;
1092         p->softirq_context = 0;
1093 #endif
1094 #ifdef CONFIG_LOCKDEP
1095         p->lockdep_depth = 0; /* no locks held yet */
1096         p->curr_chain_key = 0;
1097         p->lockdep_recursion = 0;
1098 #endif
1099
1100 #ifdef CONFIG_DEBUG_MUTEXES
1101         p->blocked_on = NULL; /* not blocked yet */
1102 #endif
1103         if (unlikely(current->ptrace))
1104                 ptrace_fork(p, clone_flags);
1105
1106         /* Perform scheduler related setup. Assign this task to a CPU. */
1107         sched_fork(p, clone_flags);
1108
1109         if ((retval = audit_alloc(p)))
1110                 goto bad_fork_cleanup_policy;
1111         /* copy all the process information */
1112         if ((retval = copy_semundo(clone_flags, p)))
1113                 goto bad_fork_cleanup_audit;
1114         if ((retval = copy_files(clone_flags, p)))
1115                 goto bad_fork_cleanup_semundo;
1116         if ((retval = copy_fs(clone_flags, p)))
1117                 goto bad_fork_cleanup_files;
1118         if ((retval = copy_sighand(clone_flags, p)))
1119                 goto bad_fork_cleanup_fs;
1120         if ((retval = copy_signal(clone_flags, p)))
1121                 goto bad_fork_cleanup_sighand;
1122         if ((retval = copy_mm(clone_flags, p)))
1123                 goto bad_fork_cleanup_signal;
1124         if ((retval = copy_namespaces(clone_flags, p)))
1125                 goto bad_fork_cleanup_mm;
1126         if ((retval = copy_io(clone_flags, p)))
1127                 goto bad_fork_cleanup_namespaces;
1128         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1129         if (retval)
1130                 goto bad_fork_cleanup_io;
1131
1132         if (pid != &init_struct_pid) {
1133                 retval = -ENOMEM;
1134                 pid = alloc_pid(p->nsproxy->pid_ns);
1135                 if (!pid)
1136                         goto bad_fork_cleanup_io;
1137
1138                 if (clone_flags & CLONE_NEWPID) {
1139                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1140                         if (retval < 0)
1141                                 goto bad_fork_free_pid;
1142                 }
1143         }
1144
1145         ftrace_graph_init_task(p);
1146
1147         p->pid = pid_nr(pid);
1148         p->tgid = p->pid;
1149         if (clone_flags & CLONE_THREAD)
1150                 p->tgid = current->tgid;
1151
1152         if (current->nsproxy != p->nsproxy) {
1153                 retval = ns_cgroup_clone(p, pid);
1154                 if (retval)
1155                         goto bad_fork_free_graph;
1156         }
1157
1158         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1159         /*
1160          * Clear TID on mm_release()?
1161          */
1162         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1163 #ifdef CONFIG_FUTEX
1164         p->robust_list = NULL;
1165 #ifdef CONFIG_COMPAT
1166         p->compat_robust_list = NULL;
1167 #endif
1168         INIT_LIST_HEAD(&p->pi_state_list);
1169         p->pi_state_cache = NULL;
1170 #endif
1171         /*
1172          * sigaltstack should be cleared when sharing the same VM
1173          */
1174         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1175                 p->sas_ss_sp = p->sas_ss_size = 0;
1176
1177         /*
1178          * Syscall tracing should be turned off in the child regardless
1179          * of CLONE_PTRACE.
1180          */
1181         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1182 #ifdef TIF_SYSCALL_EMU
1183         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1184 #endif
1185         clear_all_latency_tracing(p);
1186
1187         /* ok, now we should be set up.. */
1188         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1189         p->pdeath_signal = 0;
1190         p->exit_state = 0;
1191
1192         /*
1193          * Ok, make it visible to the rest of the system.
1194          * We dont wake it up yet.
1195          */
1196         p->group_leader = p;
1197         INIT_LIST_HEAD(&p->thread_group);
1198
1199         /* Now that the task is set up, run cgroup callbacks if
1200          * necessary. We need to run them before the task is visible
1201          * on the tasklist. */
1202         cgroup_fork_callbacks(p);
1203         cgroup_callbacks_done = 1;
1204
1205         /* Need tasklist lock for parent etc handling! */
1206         write_lock_irq(&tasklist_lock);
1207
1208         /*
1209          * The task hasn't been attached yet, so its cpus_allowed mask will
1210          * not be changed, nor will its assigned CPU.
1211          *
1212          * The cpus_allowed mask of the parent may have changed after it was
1213          * copied first time - so re-copy it here, then check the child's CPU
1214          * to ensure it is on a valid CPU (and if not, just force it back to
1215          * parent's CPU). This avoids alot of nasty races.
1216          */
1217         p->cpus_allowed = current->cpus_allowed;
1218         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1219         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1220                         !cpu_online(task_cpu(p))))
1221                 set_task_cpu(p, smp_processor_id());
1222
1223         /* CLONE_PARENT re-uses the old parent */
1224         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1225                 p->real_parent = current->real_parent;
1226                 p->parent_exec_id = current->parent_exec_id;
1227         } else {
1228                 p->real_parent = current;
1229                 p->parent_exec_id = current->self_exec_id;
1230         }
1231
1232         spin_lock(&current->sighand->siglock);
1233
1234         /*
1235          * Process group and session signals need to be delivered to just the
1236          * parent before the fork or both the parent and the child after the
1237          * fork. Restart if a signal comes in before we add the new process to
1238          * it's process group.
1239          * A fatal signal pending means that current will exit, so the new
1240          * thread can't slip out of an OOM kill (or normal SIGKILL).
1241          */
1242         recalc_sigpending();
1243         if (signal_pending(current)) {
1244                 spin_unlock(&current->sighand->siglock);
1245                 write_unlock_irq(&tasklist_lock);
1246                 retval = -ERESTARTNOINTR;
1247                 goto bad_fork_free_graph;
1248         }
1249
1250         if (clone_flags & CLONE_THREAD) {
1251                 p->group_leader = current->group_leader;
1252                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1253         }
1254
1255         if (likely(p->pid)) {
1256                 list_add_tail(&p->sibling, &p->real_parent->children);
1257                 tracehook_finish_clone(p, clone_flags, trace);
1258
1259                 if (thread_group_leader(p)) {
1260                         if (clone_flags & CLONE_NEWPID)
1261                                 p->nsproxy->pid_ns->child_reaper = p;
1262
1263                         p->signal->leader_pid = pid;
1264                         tty_kref_put(p->signal->tty);
1265                         p->signal->tty = tty_kref_get(current->signal->tty);
1266                         set_task_pgrp(p, task_pgrp_nr(current));
1267                         set_task_session(p, task_session_nr(current));
1268                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1269                         attach_pid(p, PIDTYPE_SID, task_session(current));
1270                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1271                         __get_cpu_var(process_counts)++;
1272                 }
1273                 attach_pid(p, PIDTYPE_PID, pid);
1274                 nr_threads++;
1275         }
1276
1277         total_forks++;
1278         spin_unlock(&current->sighand->siglock);
1279         write_unlock_irq(&tasklist_lock);
1280         proc_fork_connector(p);
1281         cgroup_post_fork(p);
1282         return p;
1283
1284 bad_fork_free_graph:
1285         ftrace_graph_exit_task(p);
1286 bad_fork_free_pid:
1287         if (pid != &init_struct_pid)
1288                 free_pid(pid);
1289 bad_fork_cleanup_io:
1290         put_io_context(p->io_context);
1291 bad_fork_cleanup_namespaces:
1292         exit_task_namespaces(p);
1293 bad_fork_cleanup_mm:
1294         if (p->mm)
1295                 mmput(p->mm);
1296 bad_fork_cleanup_signal:
1297         cleanup_signal(p);
1298 bad_fork_cleanup_sighand:
1299         __cleanup_sighand(p->sighand);
1300 bad_fork_cleanup_fs:
1301         exit_fs(p); /* blocking */
1302 bad_fork_cleanup_files:
1303         exit_files(p); /* blocking */
1304 bad_fork_cleanup_semundo:
1305         exit_sem(p);
1306 bad_fork_cleanup_audit:
1307         audit_free(p);
1308 bad_fork_cleanup_policy:
1309 #ifdef CONFIG_NUMA
1310         mpol_put(p->mempolicy);
1311 bad_fork_cleanup_cgroup:
1312 #endif
1313         cgroup_exit(p, cgroup_callbacks_done);
1314         delayacct_tsk_free(p);
1315         if (p->binfmt)
1316                 module_put(p->binfmt->module);
1317 bad_fork_cleanup_put_domain:
1318         module_put(task_thread_info(p)->exec_domain->module);
1319 bad_fork_cleanup_count:
1320         atomic_dec(&p->cred->user->processes);
1321         put_cred(p->real_cred);
1322         put_cred(p->cred);
1323 bad_fork_free:
1324         free_task(p);
1325 fork_out:
1326         return ERR_PTR(retval);
1327 }
1328
1329 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1330 {
1331         memset(regs, 0, sizeof(struct pt_regs));
1332         return regs;
1333 }
1334
1335 struct task_struct * __cpuinit fork_idle(int cpu)
1336 {
1337         struct task_struct *task;
1338         struct pt_regs regs;
1339
1340         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1341                             &init_struct_pid, 0);
1342         if (!IS_ERR(task))
1343                 init_idle(task, cpu);
1344
1345         return task;
1346 }
1347
1348 /*
1349  *  Ok, this is the main fork-routine.
1350  *
1351  * It copies the process, and if successful kick-starts
1352  * it and waits for it to finish using the VM if required.
1353  */
1354 long do_fork(unsigned long clone_flags,
1355               unsigned long stack_start,
1356               struct pt_regs *regs,
1357               unsigned long stack_size,
1358               int __user *parent_tidptr,
1359               int __user *child_tidptr)
1360 {
1361         struct task_struct *p;
1362         int trace = 0;
1363         long nr;
1364
1365         /*
1366          * Do some preliminary argument and permissions checking before we
1367          * actually start allocating stuff
1368          */
1369         if (clone_flags & CLONE_NEWUSER) {
1370                 if (clone_flags & CLONE_THREAD)
1371                         return -EINVAL;
1372                 /* hopefully this check will go away when userns support is
1373                  * complete
1374                  */
1375                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1376                                 !capable(CAP_SETGID))
1377                         return -EPERM;
1378         }
1379
1380         /*
1381          * We hope to recycle these flags after 2.6.26
1382          */
1383         if (unlikely(clone_flags & CLONE_STOPPED)) {
1384                 static int __read_mostly count = 100;
1385
1386                 if (count > 0 && printk_ratelimit()) {
1387                         char comm[TASK_COMM_LEN];
1388
1389                         count--;
1390                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1391                                         "clone flags 0x%lx\n",
1392                                 get_task_comm(comm, current),
1393                                 clone_flags & CLONE_STOPPED);
1394                 }
1395         }
1396
1397         /*
1398          * When called from kernel_thread, don't do user tracing stuff.
1399          */
1400         if (likely(user_mode(regs)))
1401                 trace = tracehook_prepare_clone(clone_flags);
1402
1403         p = copy_process(clone_flags, stack_start, regs, stack_size,
1404                          child_tidptr, NULL, trace);
1405         /*
1406          * Do this prior waking up the new thread - the thread pointer
1407          * might get invalid after that point, if the thread exits quickly.
1408          */
1409         if (!IS_ERR(p)) {
1410                 struct completion vfork;
1411
1412                 trace_sched_process_fork(current, p);
1413
1414                 nr = task_pid_vnr(p);
1415
1416                 if (clone_flags & CLONE_PARENT_SETTID)
1417                         put_user(nr, parent_tidptr);
1418
1419                 if (clone_flags & CLONE_VFORK) {
1420                         p->vfork_done = &vfork;
1421                         init_completion(&vfork);
1422                 }
1423
1424                 audit_finish_fork(p);
1425                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1426
1427                 /*
1428                  * We set PF_STARTING at creation in case tracing wants to
1429                  * use this to distinguish a fully live task from one that
1430                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1431                  * clear it and set the child going.
1432                  */
1433                 p->flags &= ~PF_STARTING;
1434
1435                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1436                         /*
1437                          * We'll start up with an immediate SIGSTOP.
1438                          */
1439                         sigaddset(&p->pending.signal, SIGSTOP);
1440                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1441                         __set_task_state(p, TASK_STOPPED);
1442                 } else {
1443                         wake_up_new_task(p, clone_flags);
1444                 }
1445
1446                 tracehook_report_clone_complete(trace, regs,
1447                                                 clone_flags, nr, p);
1448
1449                 if (clone_flags & CLONE_VFORK) {
1450                         freezer_do_not_count();
1451                         wait_for_completion(&vfork);
1452                         freezer_count();
1453                         tracehook_report_vfork_done(p, nr);
1454                 }
1455         } else {
1456                 nr = PTR_ERR(p);
1457         }
1458         return nr;
1459 }
1460
1461 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1462 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1463 #endif
1464
1465 static void sighand_ctor(void *data)
1466 {
1467         struct sighand_struct *sighand = data;
1468
1469         spin_lock_init(&sighand->siglock);
1470         init_waitqueue_head(&sighand->signalfd_wqh);
1471 }
1472
1473 void __init proc_caches_init(void)
1474 {
1475         sighand_cachep = kmem_cache_create("sighand_cache",
1476                         sizeof(struct sighand_struct), 0,
1477                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1478                         sighand_ctor);
1479         signal_cachep = kmem_cache_create("signal_cache",
1480                         sizeof(struct signal_struct), 0,
1481                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1482         files_cachep = kmem_cache_create("files_cache",
1483                         sizeof(struct files_struct), 0,
1484                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1485         fs_cachep = kmem_cache_create("fs_cache",
1486                         sizeof(struct fs_struct), 0,
1487                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1488         mm_cachep = kmem_cache_create("mm_struct",
1489                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1490                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1491         mmap_init();
1492 }
1493
1494 /*
1495  * Check constraints on flags passed to the unshare system call and
1496  * force unsharing of additional process context as appropriate.
1497  */
1498 static void check_unshare_flags(unsigned long *flags_ptr)
1499 {
1500         /*
1501          * If unsharing a thread from a thread group, must also
1502          * unshare vm.
1503          */
1504         if (*flags_ptr & CLONE_THREAD)
1505                 *flags_ptr |= CLONE_VM;
1506
1507         /*
1508          * If unsharing vm, must also unshare signal handlers.
1509          */
1510         if (*flags_ptr & CLONE_VM)
1511                 *flags_ptr |= CLONE_SIGHAND;
1512
1513         /*
1514          * If unsharing signal handlers and the task was created
1515          * using CLONE_THREAD, then must unshare the thread
1516          */
1517         if ((*flags_ptr & CLONE_SIGHAND) &&
1518             (atomic_read(&current->signal->count) > 1))
1519                 *flags_ptr |= CLONE_THREAD;
1520
1521         /*
1522          * If unsharing namespace, must also unshare filesystem information.
1523          */
1524         if (*flags_ptr & CLONE_NEWNS)
1525                 *flags_ptr |= CLONE_FS;
1526 }
1527
1528 /*
1529  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1530  */
1531 static int unshare_thread(unsigned long unshare_flags)
1532 {
1533         if (unshare_flags & CLONE_THREAD)
1534                 return -EINVAL;
1535
1536         return 0;
1537 }
1538
1539 /*
1540  * Unshare the filesystem structure if it is being shared
1541  */
1542 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1543 {
1544         struct fs_struct *fs = current->fs;
1545
1546         if ((unshare_flags & CLONE_FS) &&
1547             (fs && atomic_read(&fs->count) > 1)) {
1548                 *new_fsp = __copy_fs_struct(current->fs);
1549                 if (!*new_fsp)
1550                         return -ENOMEM;
1551         }
1552
1553         return 0;
1554 }
1555
1556 /*
1557  * Unsharing of sighand is not supported yet
1558  */
1559 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1560 {
1561         struct sighand_struct *sigh = current->sighand;
1562
1563         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1564                 return -EINVAL;
1565         else
1566                 return 0;
1567 }
1568
1569 /*
1570  * Unshare vm if it is being shared
1571  */
1572 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1573 {
1574         struct mm_struct *mm = current->mm;
1575
1576         if ((unshare_flags & CLONE_VM) &&
1577             (mm && atomic_read(&mm->mm_users) > 1)) {
1578                 return -EINVAL;
1579         }
1580
1581         return 0;
1582 }
1583
1584 /*
1585  * Unshare file descriptor table if it is being shared
1586  */
1587 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1588 {
1589         struct files_struct *fd = current->files;
1590         int error = 0;
1591
1592         if ((unshare_flags & CLONE_FILES) &&
1593             (fd && atomic_read(&fd->count) > 1)) {
1594                 *new_fdp = dup_fd(fd, &error);
1595                 if (!*new_fdp)
1596                         return error;
1597         }
1598
1599         return 0;
1600 }
1601
1602 /*
1603  * unshare allows a process to 'unshare' part of the process
1604  * context which was originally shared using clone.  copy_*
1605  * functions used by do_fork() cannot be used here directly
1606  * because they modify an inactive task_struct that is being
1607  * constructed. Here we are modifying the current, active,
1608  * task_struct.
1609  */
1610 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1611 {
1612         int err = 0;
1613         struct fs_struct *fs, *new_fs = NULL;
1614         struct sighand_struct *new_sigh = NULL;
1615         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1616         struct files_struct *fd, *new_fd = NULL;
1617         struct nsproxy *new_nsproxy = NULL;
1618         int do_sysvsem = 0;
1619
1620         check_unshare_flags(&unshare_flags);
1621
1622         /* Return -EINVAL for all unsupported flags */
1623         err = -EINVAL;
1624         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1625                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1626                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1627                 goto bad_unshare_out;
1628
1629         /*
1630          * CLONE_NEWIPC must also detach from the undolist: after switching
1631          * to a new ipc namespace, the semaphore arrays from the old
1632          * namespace are unreachable.
1633          */
1634         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1635                 do_sysvsem = 1;
1636         if ((err = unshare_thread(unshare_flags)))
1637                 goto bad_unshare_out;
1638         if ((err = unshare_fs(unshare_flags, &new_fs)))
1639                 goto bad_unshare_cleanup_thread;
1640         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1641                 goto bad_unshare_cleanup_fs;
1642         if ((err = unshare_vm(unshare_flags, &new_mm)))
1643                 goto bad_unshare_cleanup_sigh;
1644         if ((err = unshare_fd(unshare_flags, &new_fd)))
1645                 goto bad_unshare_cleanup_vm;
1646         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1647                         new_fs)))
1648                 goto bad_unshare_cleanup_fd;
1649
1650         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1651                 if (do_sysvsem) {
1652                         /*
1653                          * CLONE_SYSVSEM is equivalent to sys_exit().
1654                          */
1655                         exit_sem(current);
1656                 }
1657
1658                 if (new_nsproxy) {
1659                         switch_task_namespaces(current, new_nsproxy);
1660                         new_nsproxy = NULL;
1661                 }
1662
1663                 task_lock(current);
1664
1665                 if (new_fs) {
1666                         fs = current->fs;
1667                         current->fs = new_fs;
1668                         new_fs = fs;
1669                 }
1670
1671                 if (new_mm) {
1672                         mm = current->mm;
1673                         active_mm = current->active_mm;
1674                         current->mm = new_mm;
1675                         current->active_mm = new_mm;
1676                         activate_mm(active_mm, new_mm);
1677                         new_mm = mm;
1678                 }
1679
1680                 if (new_fd) {
1681                         fd = current->files;
1682                         current->files = new_fd;
1683                         new_fd = fd;
1684                 }
1685
1686                 task_unlock(current);
1687         }
1688
1689         if (new_nsproxy)
1690                 put_nsproxy(new_nsproxy);
1691
1692 bad_unshare_cleanup_fd:
1693         if (new_fd)
1694                 put_files_struct(new_fd);
1695
1696 bad_unshare_cleanup_vm:
1697         if (new_mm)
1698                 mmput(new_mm);
1699
1700 bad_unshare_cleanup_sigh:
1701         if (new_sigh)
1702                 if (atomic_dec_and_test(&new_sigh->count))
1703                         kmem_cache_free(sighand_cachep, new_sigh);
1704
1705 bad_unshare_cleanup_fs:
1706         if (new_fs)
1707                 put_fs_struct(new_fs);
1708
1709 bad_unshare_cleanup_thread:
1710 bad_unshare_out:
1711         return err;
1712 }
1713
1714 /*
1715  *      Helper to unshare the files of the current task.
1716  *      We don't want to expose copy_files internals to
1717  *      the exec layer of the kernel.
1718  */
1719
1720 int unshare_files(struct files_struct **displaced)
1721 {
1722         struct task_struct *task = current;
1723         struct files_struct *copy = NULL;
1724         int error;
1725
1726         error = unshare_fd(CLONE_FILES, &copy);
1727         if (error || !copy) {
1728                 *displaced = NULL;
1729                 return error;
1730         }
1731         *displaced = task->files;
1732         task_lock(task);
1733         task->files = copy;
1734         task_unlock(task);
1735         return 0;
1736 }