Merge branch 'sched/core' into cpus4096
[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
65 #include <asm/pgtable.h>
66 #include <asm/pgalloc.h>
67 #include <asm/uaccess.h>
68 #include <asm/mmu_context.h>
69 #include <asm/cacheflush.h>
70 #include <asm/tlbflush.h>
71
72 /*
73  * Protected counters by write_lock_irq(&tasklist_lock)
74  */
75 unsigned long total_forks;      /* Handle normal Linux uptimes. */
76 int nr_threads;                 /* The idle threads do not count.. */
77
78 int max_threads;                /* tunable limit on nr_threads */
79
80 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
81
82 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
83
84 DEFINE_TRACE(sched_process_fork);
85
86 int nr_processes(void)
87 {
88         int cpu;
89         int total = 0;
90
91         for_each_online_cpu(cpu)
92                 total += per_cpu(process_counts, cpu);
93
94         return total;
95 }
96
97 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
98 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
99 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
100 static struct kmem_cache *task_struct_cachep;
101 #endif
102
103 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
104 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
105 {
106 #ifdef CONFIG_DEBUG_STACK_USAGE
107         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
108 #else
109         gfp_t mask = GFP_KERNEL;
110 #endif
111         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
112 }
113
114 static inline void free_thread_info(struct thread_info *ti)
115 {
116         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
117 }
118 #endif
119
120 /* SLAB cache for signal_struct structures (tsk->signal) */
121 static struct kmem_cache *signal_cachep;
122
123 /* SLAB cache for sighand_struct structures (tsk->sighand) */
124 struct kmem_cache *sighand_cachep;
125
126 /* SLAB cache for files_struct structures (tsk->files) */
127 struct kmem_cache *files_cachep;
128
129 /* SLAB cache for fs_struct structures (tsk->fs) */
130 struct kmem_cache *fs_cachep;
131
132 /* SLAB cache for vm_area_struct structures */
133 struct kmem_cache *vm_area_cachep;
134
135 /* SLAB cache for mm_struct structures (tsk->mm) */
136 static struct kmem_cache *mm_cachep;
137
138 void free_task(struct task_struct *tsk)
139 {
140         prop_local_destroy_single(&tsk->dirties);
141         free_thread_info(tsk->stack);
142         rt_mutex_debug_task_free(tsk);
143         ftrace_graph_exit_task(tsk);
144         free_task_struct(tsk);
145 }
146 EXPORT_SYMBOL(free_task);
147
148 void __put_task_struct(struct task_struct *tsk)
149 {
150         WARN_ON(!tsk->exit_state);
151         WARN_ON(atomic_read(&tsk->usage));
152         WARN_ON(tsk == current);
153
154         security_task_free(tsk);
155         free_uid(tsk->user);
156         put_group_info(tsk->group_info);
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         int err;
217
218         prepare_to_copy(orig);
219
220         tsk = alloc_task_struct();
221         if (!tsk)
222                 return NULL;
223
224         ti = alloc_thread_info(tsk);
225         if (!ti) {
226                 free_task_struct(tsk);
227                 return NULL;
228         }
229
230         err = arch_dup_task_struct(tsk, orig);
231         if (err)
232                 goto out;
233
234         tsk->stack = ti;
235
236         err = prop_local_init_single(&tsk->dirties);
237         if (err)
238                 goto out;
239
240         setup_thread_stack(tsk, orig);
241
242 #ifdef CONFIG_CC_STACKPROTECTOR
243         tsk->stack_canary = get_random_int();
244 #endif
245
246         /* One for us, one for whoever does the "release_task()" (usually parent) */
247         atomic_set(&tsk->usage,2);
248         atomic_set(&tsk->fs_excl, 0);
249 #ifdef CONFIG_BLK_DEV_IO_TRACE
250         tsk->btrace_seq = 0;
251 #endif
252         tsk->splice_pipe = NULL;
253         return tsk;
254
255 out:
256         free_thread_info(ti);
257         free_task_struct(tsk);
258         return NULL;
259 }
260
261 #ifdef CONFIG_MMU
262 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
263 {
264         struct vm_area_struct *mpnt, *tmp, **pprev;
265         struct rb_node **rb_link, *rb_parent;
266         int retval;
267         unsigned long charge;
268         struct mempolicy *pol;
269
270         down_write(&oldmm->mmap_sem);
271         flush_cache_dup_mm(oldmm);
272         /*
273          * Not linked in yet - no deadlock potential:
274          */
275         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
276
277         mm->locked_vm = 0;
278         mm->mmap = NULL;
279         mm->mmap_cache = NULL;
280         mm->free_area_cache = oldmm->mmap_base;
281         mm->cached_hole_size = ~0UL;
282         mm->map_count = 0;
283         cpus_clear(mm->cpu_vm_mask);
284         mm->mm_rb = RB_ROOT;
285         rb_link = &mm->mm_rb.rb_node;
286         rb_parent = NULL;
287         pprev = &mm->mmap;
288
289         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
290                 struct file *file;
291
292                 if (mpnt->vm_flags & VM_DONTCOPY) {
293                         long pages = vma_pages(mpnt);
294                         mm->total_vm -= pages;
295                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
296                                                                 -pages);
297                         continue;
298                 }
299                 charge = 0;
300                 if (mpnt->vm_flags & VM_ACCOUNT) {
301                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
302                         if (security_vm_enough_memory(len))
303                                 goto fail_nomem;
304                         charge = len;
305                 }
306                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
307                 if (!tmp)
308                         goto fail_nomem;
309                 *tmp = *mpnt;
310                 pol = mpol_dup(vma_policy(mpnt));
311                 retval = PTR_ERR(pol);
312                 if (IS_ERR(pol))
313                         goto fail_nomem_policy;
314                 vma_set_policy(tmp, pol);
315                 tmp->vm_flags &= ~VM_LOCKED;
316                 tmp->vm_mm = mm;
317                 tmp->vm_next = NULL;
318                 anon_vma_link(tmp);
319                 file = tmp->vm_file;
320                 if (file) {
321                         struct inode *inode = file->f_path.dentry->d_inode;
322                         struct address_space *mapping = file->f_mapping;
323
324                         get_file(file);
325                         if (tmp->vm_flags & VM_DENYWRITE)
326                                 atomic_dec(&inode->i_writecount);
327                         spin_lock(&mapping->i_mmap_lock);
328                         if (tmp->vm_flags & VM_SHARED)
329                                 mapping->i_mmap_writable++;
330                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
331                         flush_dcache_mmap_lock(mapping);
332                         /* insert tmp into the share list, just after mpnt */
333                         vma_prio_tree_add(tmp, mpnt);
334                         flush_dcache_mmap_unlock(mapping);
335                         spin_unlock(&mapping->i_mmap_lock);
336                 }
337
338                 /*
339                  * Clear hugetlb-related page reserves for children. This only
340                  * affects MAP_PRIVATE mappings. Faults generated by the child
341                  * are not guaranteed to succeed, even if read-only
342                  */
343                 if (is_vm_hugetlb_page(tmp))
344                         reset_vma_resv_huge_pages(tmp);
345
346                 /*
347                  * Link in the new vma and copy the page table entries.
348                  */
349                 *pprev = tmp;
350                 pprev = &tmp->vm_next;
351
352                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
353                 rb_link = &tmp->vm_rb.rb_right;
354                 rb_parent = &tmp->vm_rb;
355
356                 mm->map_count++;
357                 retval = copy_page_range(mm, oldmm, mpnt);
358
359                 if (tmp->vm_ops && tmp->vm_ops->open)
360                         tmp->vm_ops->open(tmp);
361
362                 if (retval)
363                         goto out;
364         }
365         /* a new mm has just been created */
366         arch_dup_mmap(oldmm, mm);
367         retval = 0;
368 out:
369         up_write(&mm->mmap_sem);
370         flush_tlb_mm(oldmm);
371         up_write(&oldmm->mmap_sem);
372         return retval;
373 fail_nomem_policy:
374         kmem_cache_free(vm_area_cachep, tmp);
375 fail_nomem:
376         retval = -ENOMEM;
377         vm_unacct_memory(charge);
378         goto out;
379 }
380
381 static inline int mm_alloc_pgd(struct mm_struct * mm)
382 {
383         mm->pgd = pgd_alloc(mm);
384         if (unlikely(!mm->pgd))
385                 return -ENOMEM;
386         return 0;
387 }
388
389 static inline void mm_free_pgd(struct mm_struct * mm)
390 {
391         pgd_free(mm, mm->pgd);
392 }
393 #else
394 #define dup_mmap(mm, oldmm)     (0)
395 #define mm_alloc_pgd(mm)        (0)
396 #define mm_free_pgd(mm)
397 #endif /* CONFIG_MMU */
398
399 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
400
401 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
402 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
403
404 #include <linux/init_task.h>
405
406 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
407 {
408         atomic_set(&mm->mm_users, 1);
409         atomic_set(&mm->mm_count, 1);
410         init_rwsem(&mm->mmap_sem);
411         INIT_LIST_HEAD(&mm->mmlist);
412         mm->flags = (current->mm) ? current->mm->flags
413                                   : MMF_DUMP_FILTER_DEFAULT;
414         mm->core_state = NULL;
415         mm->nr_ptes = 0;
416         set_mm_counter(mm, file_rss, 0);
417         set_mm_counter(mm, anon_rss, 0);
418         spin_lock_init(&mm->page_table_lock);
419         rwlock_init(&mm->ioctx_list_lock);
420         mm->ioctx_list = NULL;
421         mm->free_area_cache = TASK_UNMAPPED_BASE;
422         mm->cached_hole_size = ~0UL;
423         mm_init_owner(mm, p);
424
425         if (likely(!mm_alloc_pgd(mm))) {
426                 mm->def_flags = 0;
427                 mmu_notifier_mm_init(mm);
428                 return mm;
429         }
430
431         free_mm(mm);
432         return NULL;
433 }
434
435 /*
436  * Allocate and initialize an mm_struct.
437  */
438 struct mm_struct * mm_alloc(void)
439 {
440         struct mm_struct * mm;
441
442         mm = allocate_mm();
443         if (mm) {
444                 memset(mm, 0, sizeof(*mm));
445                 mm = mm_init(mm, current);
446         }
447         return mm;
448 }
449
450 /*
451  * Called when the last reference to the mm
452  * is dropped: either by a lazy thread or by
453  * mmput. Free the page directory and the mm.
454  */
455 void __mmdrop(struct mm_struct *mm)
456 {
457         BUG_ON(mm == &init_mm);
458         mm_free_pgd(mm);
459         destroy_context(mm);
460         mmu_notifier_mm_destroy(mm);
461         free_mm(mm);
462 }
463 EXPORT_SYMBOL_GPL(__mmdrop);
464
465 /*
466  * Decrement the use count and release all resources for an mm.
467  */
468 void mmput(struct mm_struct *mm)
469 {
470         might_sleep();
471
472         if (atomic_dec_and_test(&mm->mm_users)) {
473                 exit_aio(mm);
474                 exit_mmap(mm);
475                 set_mm_exe_file(mm, NULL);
476                 if (!list_empty(&mm->mmlist)) {
477                         spin_lock(&mmlist_lock);
478                         list_del(&mm->mmlist);
479                         spin_unlock(&mmlist_lock);
480                 }
481                 put_swap_token(mm);
482                 mmdrop(mm);
483         }
484 }
485 EXPORT_SYMBOL_GPL(mmput);
486
487 /**
488  * get_task_mm - acquire a reference to the task's mm
489  *
490  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
491  * this kernel workthread has transiently adopted a user mm with use_mm,
492  * to do its AIO) is not set and if so returns a reference to it, after
493  * bumping up the use count.  User must release the mm via mmput()
494  * after use.  Typically used by /proc and ptrace.
495  */
496 struct mm_struct *get_task_mm(struct task_struct *task)
497 {
498         struct mm_struct *mm;
499
500         task_lock(task);
501         mm = task->mm;
502         if (mm) {
503                 if (task->flags & PF_KTHREAD)
504                         mm = NULL;
505                 else
506                         atomic_inc(&mm->mm_users);
507         }
508         task_unlock(task);
509         return mm;
510 }
511 EXPORT_SYMBOL_GPL(get_task_mm);
512
513 /* Please note the differences between mmput and mm_release.
514  * mmput is called whenever we stop holding onto a mm_struct,
515  * error success whatever.
516  *
517  * mm_release is called after a mm_struct has been removed
518  * from the current process.
519  *
520  * This difference is important for error handling, when we
521  * only half set up a mm_struct for a new process and need to restore
522  * the old one.  Because we mmput the new mm_struct before
523  * restoring the old one. . .
524  * Eric Biederman 10 January 1998
525  */
526 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
527 {
528         struct completion *vfork_done = tsk->vfork_done;
529
530         /* Get rid of any futexes when releasing the mm */
531 #ifdef CONFIG_FUTEX
532         if (unlikely(tsk->robust_list))
533                 exit_robust_list(tsk);
534 #ifdef CONFIG_COMPAT
535         if (unlikely(tsk->compat_robust_list))
536                 compat_exit_robust_list(tsk);
537 #endif
538 #endif
539
540         /* Get rid of any cached register state */
541         deactivate_mm(tsk, mm);
542
543         /* notify parent sleeping on vfork() */
544         if (vfork_done) {
545                 tsk->vfork_done = NULL;
546                 complete(vfork_done);
547         }
548
549         /*
550          * If we're exiting normally, clear a user-space tid field if
551          * requested.  We leave this alone when dying by signal, to leave
552          * the value intact in a core dump, and to save the unnecessary
553          * trouble otherwise.  Userland only wants this done for a sys_exit.
554          */
555         if (tsk->clear_child_tid
556             && !(tsk->flags & PF_SIGNALED)
557             && atomic_read(&mm->mm_users) > 1) {
558                 u32 __user * tidptr = tsk->clear_child_tid;
559                 tsk->clear_child_tid = NULL;
560
561                 /*
562                  * We don't check the error code - if userspace has
563                  * not set up a proper pointer then tough luck.
564                  */
565                 put_user(0, tidptr);
566                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
567         }
568 }
569
570 /*
571  * Allocate a new mm structure and copy contents from the
572  * mm structure of the passed in task structure.
573  */
574 struct mm_struct *dup_mm(struct task_struct *tsk)
575 {
576         struct mm_struct *mm, *oldmm = current->mm;
577         int err;
578
579         if (!oldmm)
580                 return NULL;
581
582         mm = allocate_mm();
583         if (!mm)
584                 goto fail_nomem;
585
586         memcpy(mm, oldmm, sizeof(*mm));
587
588         /* Initializing for Swap token stuff */
589         mm->token_priority = 0;
590         mm->last_interval = 0;
591
592         if (!mm_init(mm, tsk))
593                 goto fail_nomem;
594
595         if (init_new_context(tsk, mm))
596                 goto fail_nocontext;
597
598         dup_mm_exe_file(oldmm, mm);
599
600         err = dup_mmap(mm, oldmm);
601         if (err)
602                 goto free_pt;
603
604         mm->hiwater_rss = get_mm_rss(mm);
605         mm->hiwater_vm = mm->total_vm;
606
607         return mm;
608
609 free_pt:
610         mmput(mm);
611
612 fail_nomem:
613         return NULL;
614
615 fail_nocontext:
616         /*
617          * If init_new_context() failed, we cannot use mmput() to free the mm
618          * because it calls destroy_context()
619          */
620         mm_free_pgd(mm);
621         free_mm(mm);
622         return NULL;
623 }
624
625 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
626 {
627         struct mm_struct * mm, *oldmm;
628         int retval;
629
630         tsk->min_flt = tsk->maj_flt = 0;
631         tsk->nvcsw = tsk->nivcsw = 0;
632
633         tsk->mm = NULL;
634         tsk->active_mm = NULL;
635
636         /*
637          * Are we cloning a kernel thread?
638          *
639          * We need to steal a active VM for that..
640          */
641         oldmm = current->mm;
642         if (!oldmm)
643                 return 0;
644
645         if (clone_flags & CLONE_VM) {
646                 atomic_inc(&oldmm->mm_users);
647                 mm = oldmm;
648                 goto good_mm;
649         }
650
651         retval = -ENOMEM;
652         mm = dup_mm(tsk);
653         if (!mm)
654                 goto fail_nomem;
655
656 good_mm:
657         /* Initializing for Swap token stuff */
658         mm->token_priority = 0;
659         mm->last_interval = 0;
660
661         tsk->mm = mm;
662         tsk->active_mm = mm;
663         return 0;
664
665 fail_nomem:
666         return retval;
667 }
668
669 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
670 {
671         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
672         /* We don't need to lock fs - think why ;-) */
673         if (fs) {
674                 atomic_set(&fs->count, 1);
675                 rwlock_init(&fs->lock);
676                 fs->umask = old->umask;
677                 read_lock(&old->lock);
678                 fs->root = old->root;
679                 path_get(&old->root);
680                 fs->pwd = old->pwd;
681                 path_get(&old->pwd);
682                 read_unlock(&old->lock);
683         }
684         return fs;
685 }
686
687 struct fs_struct *copy_fs_struct(struct fs_struct *old)
688 {
689         return __copy_fs_struct(old);
690 }
691
692 EXPORT_SYMBOL_GPL(copy_fs_struct);
693
694 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
695 {
696         if (clone_flags & CLONE_FS) {
697                 atomic_inc(&current->fs->count);
698                 return 0;
699         }
700         tsk->fs = __copy_fs_struct(current->fs);
701         if (!tsk->fs)
702                 return -ENOMEM;
703         return 0;
704 }
705
706 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
707 {
708         struct files_struct *oldf, *newf;
709         int error = 0;
710
711         /*
712          * A background process may not have any files ...
713          */
714         oldf = current->files;
715         if (!oldf)
716                 goto out;
717
718         if (clone_flags & CLONE_FILES) {
719                 atomic_inc(&oldf->count);
720                 goto out;
721         }
722
723         newf = dup_fd(oldf, &error);
724         if (!newf)
725                 goto out;
726
727         tsk->files = newf;
728         error = 0;
729 out:
730         return error;
731 }
732
733 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
734 {
735 #ifdef CONFIG_BLOCK
736         struct io_context *ioc = current->io_context;
737
738         if (!ioc)
739                 return 0;
740         /*
741          * Share io context with parent, if CLONE_IO is set
742          */
743         if (clone_flags & CLONE_IO) {
744                 tsk->io_context = ioc_task_link(ioc);
745                 if (unlikely(!tsk->io_context))
746                         return -ENOMEM;
747         } else if (ioprio_valid(ioc->ioprio)) {
748                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
749                 if (unlikely(!tsk->io_context))
750                         return -ENOMEM;
751
752                 tsk->io_context->ioprio = ioc->ioprio;
753         }
754 #endif
755         return 0;
756 }
757
758 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
759 {
760         struct sighand_struct *sig;
761
762         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
763                 atomic_inc(&current->sighand->count);
764                 return 0;
765         }
766         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
767         rcu_assign_pointer(tsk->sighand, sig);
768         if (!sig)
769                 return -ENOMEM;
770         atomic_set(&sig->count, 1);
771         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
772         return 0;
773 }
774
775 void __cleanup_sighand(struct sighand_struct *sighand)
776 {
777         if (atomic_dec_and_test(&sighand->count))
778                 kmem_cache_free(sighand_cachep, sighand);
779 }
780
781
782 /*
783  * Initialize POSIX timer handling for a thread group.
784  */
785 static void posix_cpu_timers_init_group(struct signal_struct *sig)
786 {
787         /* Thread group counters. */
788         thread_group_cputime_init(sig);
789
790         /* Expiration times and increments. */
791         sig->it_virt_expires = cputime_zero;
792         sig->it_virt_incr = cputime_zero;
793         sig->it_prof_expires = cputime_zero;
794         sig->it_prof_incr = cputime_zero;
795
796         /* Cached expiration times. */
797         sig->cputime_expires.prof_exp = cputime_zero;
798         sig->cputime_expires.virt_exp = cputime_zero;
799         sig->cputime_expires.sched_exp = 0;
800
801         /* The timer lists. */
802         INIT_LIST_HEAD(&sig->cpu_timers[0]);
803         INIT_LIST_HEAD(&sig->cpu_timers[1]);
804         INIT_LIST_HEAD(&sig->cpu_timers[2]);
805 }
806
807 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
808 {
809         struct signal_struct *sig;
810         int ret;
811
812         if (clone_flags & CLONE_THREAD) {
813                 ret = thread_group_cputime_clone_thread(current);
814                 if (likely(!ret)) {
815                         atomic_inc(&current->signal->count);
816                         atomic_inc(&current->signal->live);
817                 }
818                 return ret;
819         }
820         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
821         tsk->signal = sig;
822         if (!sig)
823                 return -ENOMEM;
824
825         ret = copy_thread_group_keys(tsk);
826         if (ret < 0) {
827                 kmem_cache_free(signal_cachep, sig);
828                 return ret;
829         }
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         sig->group_exit_code = 0;
836         sig->group_exit_task = NULL;
837         sig->group_stop_count = 0;
838         sig->curr_target = tsk;
839         init_sigpending(&sig->shared_pending);
840         INIT_LIST_HEAD(&sig->posix_timers);
841
842         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
843         sig->it_real_incr.tv64 = 0;
844         sig->real_timer.function = it_real_fn;
845
846         sig->leader = 0;        /* session leadership doesn't inherit */
847         sig->tty_old_pgrp = NULL;
848         sig->tty = NULL;
849
850         sig->cutime = sig->cstime = cputime_zero;
851         sig->gtime = cputime_zero;
852         sig->cgtime = cputime_zero;
853         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
854         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
855         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
856         task_io_accounting_init(&sig->ioac);
857         taskstats_tgid_init(sig);
858
859         task_lock(current->group_leader);
860         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
861         task_unlock(current->group_leader);
862
863         posix_cpu_timers_init_group(sig);
864
865         acct_init_pacct(&sig->pacct);
866
867         tty_audit_fork(sig);
868
869         return 0;
870 }
871
872 void __cleanup_signal(struct signal_struct *sig)
873 {
874         thread_group_cputime_free(sig);
875         exit_thread_group_keys(sig);
876         tty_kref_put(sig->tty);
877         kmem_cache_free(signal_cachep, sig);
878 }
879
880 static void cleanup_signal(struct task_struct *tsk)
881 {
882         struct signal_struct *sig = tsk->signal;
883
884         atomic_dec(&sig->live);
885
886         if (atomic_dec_and_test(&sig->count))
887                 __cleanup_signal(sig);
888 }
889
890 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
891 {
892         unsigned long new_flags = p->flags;
893
894         new_flags &= ~PF_SUPERPRIV;
895         new_flags |= PF_FORKNOEXEC;
896         new_flags |= PF_STARTING;
897         p->flags = new_flags;
898         clear_freeze_flag(p);
899 }
900
901 asmlinkage long sys_set_tid_address(int __user *tidptr)
902 {
903         current->clear_child_tid = tidptr;
904
905         return task_pid_vnr(current);
906 }
907
908 static void rt_mutex_init_task(struct task_struct *p)
909 {
910         spin_lock_init(&p->pi_lock);
911 #ifdef CONFIG_RT_MUTEXES
912         plist_head_init(&p->pi_waiters, &p->pi_lock);
913         p->pi_blocked_on = NULL;
914 #endif
915 }
916
917 #ifdef CONFIG_MM_OWNER
918 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
919 {
920         mm->owner = p;
921 }
922 #endif /* CONFIG_MM_OWNER */
923
924 /*
925  * Initialize POSIX timer handling for a single task.
926  */
927 static void posix_cpu_timers_init(struct task_struct *tsk)
928 {
929         tsk->cputime_expires.prof_exp = cputime_zero;
930         tsk->cputime_expires.virt_exp = cputime_zero;
931         tsk->cputime_expires.sched_exp = 0;
932         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
933         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
934         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
935 }
936
937 /*
938  * This creates a new process as a copy of the old one,
939  * but does not actually start it yet.
940  *
941  * It copies the registers, and all the appropriate
942  * parts of the process environment (as per the clone
943  * flags). The actual kick-off is left to the caller.
944  */
945 static struct task_struct *copy_process(unsigned long clone_flags,
946                                         unsigned long stack_start,
947                                         struct pt_regs *regs,
948                                         unsigned long stack_size,
949                                         int __user *child_tidptr,
950                                         struct pid *pid,
951                                         int trace)
952 {
953         int retval;
954         struct task_struct *p;
955         int cgroup_callbacks_done = 0;
956
957         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
958                 return ERR_PTR(-EINVAL);
959
960         /*
961          * Thread groups must share signals as well, and detached threads
962          * can only be started up within the thread group.
963          */
964         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
965                 return ERR_PTR(-EINVAL);
966
967         /*
968          * Shared signal handlers imply shared VM. By way of the above,
969          * thread groups also imply shared VM. Blocking this case allows
970          * for various simplifications in other code.
971          */
972         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
973                 return ERR_PTR(-EINVAL);
974
975         retval = security_task_create(clone_flags);
976         if (retval)
977                 goto fork_out;
978
979         retval = -ENOMEM;
980         p = dup_task_struct(current);
981         if (!p)
982                 goto fork_out;
983
984         rt_mutex_init_task(p);
985
986 #ifdef CONFIG_PROVE_LOCKING
987         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
988         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
989 #endif
990         retval = -EAGAIN;
991         if (atomic_read(&p->user->processes) >=
992                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
993                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
994                     p->user != current->nsproxy->user_ns->root_user)
995                         goto bad_fork_free;
996         }
997
998         atomic_inc(&p->user->__count);
999         atomic_inc(&p->user->processes);
1000         get_group_info(p->group_info);
1001
1002         /*
1003          * If multiple threads are within copy_process(), then this check
1004          * triggers too late. This doesn't hurt, the check is only there
1005          * to stop root fork bombs.
1006          */
1007         if (nr_threads >= max_threads)
1008                 goto bad_fork_cleanup_count;
1009
1010         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1011                 goto bad_fork_cleanup_count;
1012
1013         if (p->binfmt && !try_module_get(p->binfmt->module))
1014                 goto bad_fork_cleanup_put_domain;
1015
1016         p->did_exec = 0;
1017         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1018         copy_flags(clone_flags, p);
1019         INIT_LIST_HEAD(&p->children);
1020         INIT_LIST_HEAD(&p->sibling);
1021 #ifdef CONFIG_PREEMPT_RCU
1022         p->rcu_read_lock_nesting = 0;
1023         p->rcu_flipctr_idx = 0;
1024 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1025         p->vfork_done = NULL;
1026         spin_lock_init(&p->alloc_lock);
1027
1028         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1029         init_sigpending(&p->pending);
1030
1031         p->utime = cputime_zero;
1032         p->stime = cputime_zero;
1033         p->gtime = cputime_zero;
1034         p->utimescaled = cputime_zero;
1035         p->stimescaled = cputime_zero;
1036         p->prev_utime = cputime_zero;
1037         p->prev_stime = cputime_zero;
1038
1039         p->default_timer_slack_ns = current->timer_slack_ns;
1040
1041 #ifdef CONFIG_DETECT_SOFTLOCKUP
1042         p->last_switch_count = 0;
1043         p->last_switch_timestamp = 0;
1044 #endif
1045
1046         task_io_accounting_init(&p->ioac);
1047         acct_clear_integrals(p);
1048
1049         posix_cpu_timers_init(p);
1050
1051         p->lock_depth = -1;             /* -1 = no lock */
1052         do_posix_clock_monotonic_gettime(&p->start_time);
1053         p->real_start_time = p->start_time;
1054         monotonic_to_bootbased(&p->real_start_time);
1055 #ifdef CONFIG_SECURITY
1056         p->security = NULL;
1057 #endif
1058         p->cap_bset = current->cap_bset;
1059         p->io_context = NULL;
1060         p->audit_context = NULL;
1061         cgroup_fork(p);
1062 #ifdef CONFIG_NUMA
1063         p->mempolicy = mpol_dup(p->mempolicy);
1064         if (IS_ERR(p->mempolicy)) {
1065                 retval = PTR_ERR(p->mempolicy);
1066                 p->mempolicy = NULL;
1067                 goto bad_fork_cleanup_cgroup;
1068         }
1069         mpol_fix_fork_child_flag(p);
1070 #endif
1071 #ifdef CONFIG_TRACE_IRQFLAGS
1072         p->irq_events = 0;
1073 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1074         p->hardirqs_enabled = 1;
1075 #else
1076         p->hardirqs_enabled = 0;
1077 #endif
1078         p->hardirq_enable_ip = 0;
1079         p->hardirq_enable_event = 0;
1080         p->hardirq_disable_ip = _THIS_IP_;
1081         p->hardirq_disable_event = 0;
1082         p->softirqs_enabled = 1;
1083         p->softirq_enable_ip = _THIS_IP_;
1084         p->softirq_enable_event = 0;
1085         p->softirq_disable_ip = 0;
1086         p->softirq_disable_event = 0;
1087         p->hardirq_context = 0;
1088         p->softirq_context = 0;
1089 #endif
1090 #ifdef CONFIG_LOCKDEP
1091         p->lockdep_depth = 0; /* no locks held yet */
1092         p->curr_chain_key = 0;
1093         p->lockdep_recursion = 0;
1094 #endif
1095
1096 #ifdef CONFIG_DEBUG_MUTEXES
1097         p->blocked_on = NULL; /* not blocked yet */
1098 #endif
1099
1100         /* Perform scheduler related setup. Assign this task to a CPU. */
1101         sched_fork(p, clone_flags);
1102
1103         if ((retval = security_task_alloc(p)))
1104                 goto bad_fork_cleanup_policy;
1105         if ((retval = audit_alloc(p)))
1106                 goto bad_fork_cleanup_security;
1107         /* copy all the process information */
1108         if ((retval = copy_semundo(clone_flags, p)))
1109                 goto bad_fork_cleanup_audit;
1110         if ((retval = copy_files(clone_flags, p)))
1111                 goto bad_fork_cleanup_semundo;
1112         if ((retval = copy_fs(clone_flags, p)))
1113                 goto bad_fork_cleanup_files;
1114         if ((retval = copy_sighand(clone_flags, p)))
1115                 goto bad_fork_cleanup_fs;
1116         if ((retval = copy_signal(clone_flags, p)))
1117                 goto bad_fork_cleanup_sighand;
1118         if ((retval = copy_mm(clone_flags, p)))
1119                 goto bad_fork_cleanup_signal;
1120         if ((retval = copy_keys(clone_flags, p)))
1121                 goto bad_fork_cleanup_mm;
1122         if ((retval = copy_namespaces(clone_flags, p)))
1123                 goto bad_fork_cleanup_keys;
1124         if ((retval = copy_io(clone_flags, p)))
1125                 goto bad_fork_cleanup_namespaces;
1126         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1127         if (retval)
1128                 goto bad_fork_cleanup_io;
1129
1130         if (pid != &init_struct_pid) {
1131                 retval = -ENOMEM;
1132                 pid = alloc_pid(task_active_pid_ns(p));
1133                 if (!pid)
1134                         goto bad_fork_cleanup_io;
1135
1136                 if (clone_flags & CLONE_NEWPID) {
1137                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1138                         if (retval < 0)
1139                                 goto bad_fork_free_pid;
1140                 }
1141         }
1142
1143         ftrace_graph_init_task(p);
1144
1145         p->pid = pid_nr(pid);
1146         p->tgid = p->pid;
1147         if (clone_flags & CLONE_THREAD)
1148                 p->tgid = current->tgid;
1149
1150         if (current->nsproxy != p->nsproxy) {
1151                 retval = ns_cgroup_clone(p, pid);
1152                 if (retval)
1153                         goto bad_fork_free_graph;
1154         }
1155
1156         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1157         /*
1158          * Clear TID on mm_release()?
1159          */
1160         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1161 #ifdef CONFIG_FUTEX
1162         p->robust_list = NULL;
1163 #ifdef CONFIG_COMPAT
1164         p->compat_robust_list = NULL;
1165 #endif
1166         INIT_LIST_HEAD(&p->pi_state_list);
1167         p->pi_state_cache = NULL;
1168 #endif
1169         /*
1170          * sigaltstack should be cleared when sharing the same VM
1171          */
1172         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1173                 p->sas_ss_sp = p->sas_ss_size = 0;
1174
1175         /*
1176          * Syscall tracing should be turned off in the child regardless
1177          * of CLONE_PTRACE.
1178          */
1179         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1180 #ifdef TIF_SYSCALL_EMU
1181         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1182 #endif
1183         clear_all_latency_tracing(p);
1184
1185         /* Our parent execution domain becomes current domain
1186            These must match for thread signalling to apply */
1187         p->parent_exec_id = p->self_exec_id;
1188
1189         /* ok, now we should be set up.. */
1190         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1191         p->pdeath_signal = 0;
1192         p->exit_state = 0;
1193
1194         /*
1195          * Ok, make it visible to the rest of the system.
1196          * We dont wake it up yet.
1197          */
1198         p->group_leader = p;
1199         INIT_LIST_HEAD(&p->thread_group);
1200
1201         /* Now that the task is set up, run cgroup callbacks if
1202          * necessary. We need to run them before the task is visible
1203          * on the tasklist. */
1204         cgroup_fork_callbacks(p);
1205         cgroup_callbacks_done = 1;
1206
1207         /* Need tasklist lock for parent etc handling! */
1208         write_lock_irq(&tasklist_lock);
1209
1210         /*
1211          * The task hasn't been attached yet, so its cpus_allowed mask will
1212          * not be changed, nor will its assigned CPU.
1213          *
1214          * The cpus_allowed mask of the parent may have changed after it was
1215          * copied first time - so re-copy it here, then check the child's CPU
1216          * to ensure it is on a valid CPU (and if not, just force it back to
1217          * parent's CPU). This avoids alot of nasty races.
1218          */
1219         p->cpus_allowed = current->cpus_allowed;
1220         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1221         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1222                         !cpu_online(task_cpu(p))))
1223                 set_task_cpu(p, smp_processor_id());
1224
1225         /* CLONE_PARENT re-uses the old parent */
1226         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1227                 p->real_parent = current->real_parent;
1228         else
1229                 p->real_parent = current;
1230
1231         spin_lock(&current->sighand->siglock);
1232
1233         /*
1234          * Process group and session signals need to be delivered to just the
1235          * parent before the fork or both the parent and the child after the
1236          * fork. Restart if a signal comes in before we add the new process to
1237          * it's process group.
1238          * A fatal signal pending means that current will exit, so the new
1239          * thread can't slip out of an OOM kill (or normal SIGKILL).
1240          */
1241         recalc_sigpending();
1242         if (signal_pending(current)) {
1243                 spin_unlock(&current->sighand->siglock);
1244                 write_unlock_irq(&tasklist_lock);
1245                 retval = -ERESTARTNOINTR;
1246                 goto bad_fork_free_graph;
1247         }
1248
1249         if (clone_flags & CLONE_THREAD) {
1250                 p->group_leader = current->group_leader;
1251                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1252         }
1253
1254         if (likely(p->pid)) {
1255                 list_add_tail(&p->sibling, &p->real_parent->children);
1256                 tracehook_finish_clone(p, clone_flags, trace);
1257
1258                 if (thread_group_leader(p)) {
1259                         if (clone_flags & CLONE_NEWPID)
1260                                 p->nsproxy->pid_ns->child_reaper = p;
1261
1262                         p->signal->leader_pid = pid;
1263                         tty_kref_put(p->signal->tty);
1264                         p->signal->tty = tty_kref_get(current->signal->tty);
1265                         set_task_pgrp(p, task_pgrp_nr(current));
1266                         set_task_session(p, task_session_nr(current));
1267                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1268                         attach_pid(p, PIDTYPE_SID, task_session(current));
1269                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1270                         __get_cpu_var(process_counts)++;
1271                 }
1272                 attach_pid(p, PIDTYPE_PID, pid);
1273                 nr_threads++;
1274         }
1275
1276         total_forks++;
1277         spin_unlock(&current->sighand->siglock);
1278         write_unlock_irq(&tasklist_lock);
1279         proc_fork_connector(p);
1280         cgroup_post_fork(p);
1281         return p;
1282
1283 bad_fork_free_graph:
1284         ftrace_graph_exit_task(p);
1285 bad_fork_free_pid:
1286         if (pid != &init_struct_pid)
1287                 free_pid(pid);
1288 bad_fork_cleanup_io:
1289         put_io_context(p->io_context);
1290 bad_fork_cleanup_namespaces:
1291         exit_task_namespaces(p);
1292 bad_fork_cleanup_keys:
1293         exit_keys(p);
1294 bad_fork_cleanup_mm:
1295         if (p->mm)
1296                 mmput(p->mm);
1297 bad_fork_cleanup_signal:
1298         cleanup_signal(p);
1299 bad_fork_cleanup_sighand:
1300         __cleanup_sighand(p->sighand);
1301 bad_fork_cleanup_fs:
1302         exit_fs(p); /* blocking */
1303 bad_fork_cleanup_files:
1304         exit_files(p); /* blocking */
1305 bad_fork_cleanup_semundo:
1306         exit_sem(p);
1307 bad_fork_cleanup_audit:
1308         audit_free(p);
1309 bad_fork_cleanup_security:
1310         security_task_free(p);
1311 bad_fork_cleanup_policy:
1312 #ifdef CONFIG_NUMA
1313         mpol_put(p->mempolicy);
1314 bad_fork_cleanup_cgroup:
1315 #endif
1316         cgroup_exit(p, cgroup_callbacks_done);
1317         delayacct_tsk_free(p);
1318         if (p->binfmt)
1319                 module_put(p->binfmt->module);
1320 bad_fork_cleanup_put_domain:
1321         module_put(task_thread_info(p)->exec_domain->module);
1322 bad_fork_cleanup_count:
1323         put_group_info(p->group_info);
1324         atomic_dec(&p->user->processes);
1325         free_uid(p->user);
1326 bad_fork_free:
1327         free_task(p);
1328 fork_out:
1329         return ERR_PTR(retval);
1330 }
1331
1332 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1333 {
1334         memset(regs, 0, sizeof(struct pt_regs));
1335         return regs;
1336 }
1337
1338 struct task_struct * __cpuinit fork_idle(int cpu)
1339 {
1340         struct task_struct *task;
1341         struct pt_regs regs;
1342
1343         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1344                             &init_struct_pid, 0);
1345         if (!IS_ERR(task))
1346                 init_idle(task, cpu);
1347
1348         return task;
1349 }
1350
1351 /*
1352  *  Ok, this is the main fork-routine.
1353  *
1354  * It copies the process, and if successful kick-starts
1355  * it and waits for it to finish using the VM if required.
1356  */
1357 long do_fork(unsigned long clone_flags,
1358               unsigned long stack_start,
1359               struct pt_regs *regs,
1360               unsigned long stack_size,
1361               int __user *parent_tidptr,
1362               int __user *child_tidptr)
1363 {
1364         struct task_struct *p;
1365         int trace = 0;
1366         long nr;
1367
1368         /*
1369          * We hope to recycle these flags after 2.6.26
1370          */
1371         if (unlikely(clone_flags & CLONE_STOPPED)) {
1372                 static int __read_mostly count = 100;
1373
1374                 if (count > 0 && printk_ratelimit()) {
1375                         char comm[TASK_COMM_LEN];
1376
1377                         count--;
1378                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1379                                         "clone flags 0x%lx\n",
1380                                 get_task_comm(comm, current),
1381                                 clone_flags & CLONE_STOPPED);
1382                 }
1383         }
1384
1385         /*
1386          * When called from kernel_thread, don't do user tracing stuff.
1387          */
1388         if (likely(user_mode(regs)))
1389                 trace = tracehook_prepare_clone(clone_flags);
1390
1391         p = copy_process(clone_flags, stack_start, regs, stack_size,
1392                          child_tidptr, NULL, trace);
1393         /*
1394          * Do this prior waking up the new thread - the thread pointer
1395          * might get invalid after that point, if the thread exits quickly.
1396          */
1397         if (!IS_ERR(p)) {
1398                 struct completion vfork;
1399
1400                 trace_sched_process_fork(current, p);
1401
1402                 nr = task_pid_vnr(p);
1403
1404                 if (clone_flags & CLONE_PARENT_SETTID)
1405                         put_user(nr, parent_tidptr);
1406
1407                 if (clone_flags & CLONE_VFORK) {
1408                         p->vfork_done = &vfork;
1409                         init_completion(&vfork);
1410                 }
1411
1412                 audit_finish_fork(p);
1413                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1414
1415                 /*
1416                  * We set PF_STARTING at creation in case tracing wants to
1417                  * use this to distinguish a fully live task from one that
1418                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1419                  * clear it and set the child going.
1420                  */
1421                 p->flags &= ~PF_STARTING;
1422
1423                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1424                         /*
1425                          * We'll start up with an immediate SIGSTOP.
1426                          */
1427                         sigaddset(&p->pending.signal, SIGSTOP);
1428                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1429                         __set_task_state(p, TASK_STOPPED);
1430                 } else {
1431                         wake_up_new_task(p, clone_flags);
1432                 }
1433
1434                 tracehook_report_clone_complete(trace, regs,
1435                                                 clone_flags, nr, p);
1436
1437                 if (clone_flags & CLONE_VFORK) {
1438                         freezer_do_not_count();
1439                         wait_for_completion(&vfork);
1440                         freezer_count();
1441                         tracehook_report_vfork_done(p, nr);
1442                 }
1443         } else {
1444                 nr = PTR_ERR(p);
1445         }
1446         return nr;
1447 }
1448
1449 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1450 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1451 #endif
1452
1453 static void sighand_ctor(void *data)
1454 {
1455         struct sighand_struct *sighand = data;
1456
1457         spin_lock_init(&sighand->siglock);
1458         init_waitqueue_head(&sighand->signalfd_wqh);
1459 }
1460
1461 void __init proc_caches_init(void)
1462 {
1463         sighand_cachep = kmem_cache_create("sighand_cache",
1464                         sizeof(struct sighand_struct), 0,
1465                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1466                         sighand_ctor);
1467         signal_cachep = kmem_cache_create("signal_cache",
1468                         sizeof(struct signal_struct), 0,
1469                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1470         files_cachep = kmem_cache_create("files_cache",
1471                         sizeof(struct files_struct), 0,
1472                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1473         fs_cachep = kmem_cache_create("fs_cache",
1474                         sizeof(struct fs_struct), 0,
1475                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1476         vm_area_cachep = kmem_cache_create("vm_area_struct",
1477                         sizeof(struct vm_area_struct), 0,
1478                         SLAB_PANIC, NULL);
1479         mm_cachep = kmem_cache_create("mm_struct",
1480                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1481                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
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) &&
1537             (fs && atomic_read(&fs->count) > 1)) {
1538                 *new_fsp = __copy_fs_struct(current->fs);
1539                 if (!*new_fsp)
1540                         return -ENOMEM;
1541         }
1542
1543         return 0;
1544 }
1545
1546 /*
1547  * Unsharing of sighand is not supported yet
1548  */
1549 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1550 {
1551         struct sighand_struct *sigh = current->sighand;
1552
1553         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1554                 return -EINVAL;
1555         else
1556                 return 0;
1557 }
1558
1559 /*
1560  * Unshare vm if it is being shared
1561  */
1562 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1563 {
1564         struct mm_struct *mm = current->mm;
1565
1566         if ((unshare_flags & CLONE_VM) &&
1567             (mm && atomic_read(&mm->mm_users) > 1)) {
1568                 return -EINVAL;
1569         }
1570
1571         return 0;
1572 }
1573
1574 /*
1575  * Unshare file descriptor table if it is being shared
1576  */
1577 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1578 {
1579         struct files_struct *fd = current->files;
1580         int error = 0;
1581
1582         if ((unshare_flags & CLONE_FILES) &&
1583             (fd && atomic_read(&fd->count) > 1)) {
1584                 *new_fdp = dup_fd(fd, &error);
1585                 if (!*new_fdp)
1586                         return error;
1587         }
1588
1589         return 0;
1590 }
1591
1592 /*
1593  * unshare allows a process to 'unshare' part of the process
1594  * context which was originally shared using clone.  copy_*
1595  * functions used by do_fork() cannot be used here directly
1596  * because they modify an inactive task_struct that is being
1597  * constructed. Here we are modifying the current, active,
1598  * task_struct.
1599  */
1600 asmlinkage long sys_unshare(unsigned long unshare_flags)
1601 {
1602         int err = 0;
1603         struct fs_struct *fs, *new_fs = NULL;
1604         struct sighand_struct *new_sigh = NULL;
1605         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1606         struct files_struct *fd, *new_fd = NULL;
1607         struct nsproxy *new_nsproxy = NULL;
1608         int do_sysvsem = 0;
1609
1610         check_unshare_flags(&unshare_flags);
1611
1612         /* Return -EINVAL for all unsupported flags */
1613         err = -EINVAL;
1614         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1615                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1616                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1617                                 CLONE_NEWNET))
1618                 goto bad_unshare_out;
1619
1620         /*
1621          * CLONE_NEWIPC must also detach from the undolist: after switching
1622          * to a new ipc namespace, the semaphore arrays from the old
1623          * namespace are unreachable.
1624          */
1625         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1626                 do_sysvsem = 1;
1627         if ((err = unshare_thread(unshare_flags)))
1628                 goto bad_unshare_out;
1629         if ((err = unshare_fs(unshare_flags, &new_fs)))
1630                 goto bad_unshare_cleanup_thread;
1631         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1632                 goto bad_unshare_cleanup_fs;
1633         if ((err = unshare_vm(unshare_flags, &new_mm)))
1634                 goto bad_unshare_cleanup_sigh;
1635         if ((err = unshare_fd(unshare_flags, &new_fd)))
1636                 goto bad_unshare_cleanup_vm;
1637         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1638                         new_fs)))
1639                 goto bad_unshare_cleanup_fd;
1640
1641         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1642                 if (do_sysvsem) {
1643                         /*
1644                          * CLONE_SYSVSEM is equivalent to sys_exit().
1645                          */
1646                         exit_sem(current);
1647                 }
1648
1649                 if (new_nsproxy) {
1650                         switch_task_namespaces(current, new_nsproxy);
1651                         new_nsproxy = NULL;
1652                 }
1653
1654                 task_lock(current);
1655
1656                 if (new_fs) {
1657                         fs = current->fs;
1658                         current->fs = new_fs;
1659                         new_fs = fs;
1660                 }
1661
1662                 if (new_mm) {
1663                         mm = current->mm;
1664                         active_mm = current->active_mm;
1665                         current->mm = new_mm;
1666                         current->active_mm = new_mm;
1667                         activate_mm(active_mm, new_mm);
1668                         new_mm = mm;
1669                 }
1670
1671                 if (new_fd) {
1672                         fd = current->files;
1673                         current->files = new_fd;
1674                         new_fd = fd;
1675                 }
1676
1677                 task_unlock(current);
1678         }
1679
1680         if (new_nsproxy)
1681                 put_nsproxy(new_nsproxy);
1682
1683 bad_unshare_cleanup_fd:
1684         if (new_fd)
1685                 put_files_struct(new_fd);
1686
1687 bad_unshare_cleanup_vm:
1688         if (new_mm)
1689                 mmput(new_mm);
1690
1691 bad_unshare_cleanup_sigh:
1692         if (new_sigh)
1693                 if (atomic_dec_and_test(&new_sigh->count))
1694                         kmem_cache_free(sighand_cachep, new_sigh);
1695
1696 bad_unshare_cleanup_fs:
1697         if (new_fs)
1698                 put_fs_struct(new_fs);
1699
1700 bad_unshare_cleanup_thread:
1701 bad_unshare_out:
1702         return err;
1703 }
1704
1705 /*
1706  *      Helper to unshare the files of the current task.
1707  *      We don't want to expose copy_files internals to
1708  *      the exec layer of the kernel.
1709  */
1710
1711 int unshare_files(struct files_struct **displaced)
1712 {
1713         struct task_struct *task = current;
1714         struct files_struct *copy = NULL;
1715         int error;
1716
1717         error = unshare_fd(CLONE_FILES, &copy);
1718         if (error || !copy) {
1719                 *displaced = NULL;
1720                 return error;
1721         }
1722         *displaced = task->files;
1723         task_lock(task);
1724         task->files = copy;
1725         task_unlock(task);
1726         return 0;
1727 }