d1eb30e69cccc880f84e65514221e3a5f0e91245
[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         free_task_struct(tsk);
144 }
145 EXPORT_SYMBOL(free_task);
146
147 void __put_task_struct(struct task_struct *tsk)
148 {
149         WARN_ON(!tsk->exit_state);
150         WARN_ON(atomic_read(&tsk->usage));
151         WARN_ON(tsk == current);
152
153         security_task_free(tsk);
154         free_uid(tsk->user);
155         put_group_info(tsk->group_info);
156         delayacct_tsk_free(tsk);
157
158         if (!profile_handoff_task(tsk))
159                 free_task(tsk);
160 }
161
162 /*
163  * macro override instead of weak attribute alias, to workaround
164  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
165  */
166 #ifndef arch_task_cache_init
167 #define arch_task_cache_init()
168 #endif
169
170 void __init fork_init(unsigned long mempages)
171 {
172 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
173 #ifndef ARCH_MIN_TASKALIGN
174 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
175 #endif
176         /* create a slab on which task_structs can be allocated */
177         task_struct_cachep =
178                 kmem_cache_create("task_struct", sizeof(struct task_struct),
179                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
180 #endif
181
182         /* do the arch specific task caches init */
183         arch_task_cache_init();
184
185         /*
186          * The default maximum number of threads is set to a safe
187          * value: the thread structures can take up at most half
188          * of memory.
189          */
190         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
191
192         /*
193          * we need to allow at least 20 threads to boot a system
194          */
195         if(max_threads < 20)
196                 max_threads = 20;
197
198         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
199         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
200         init_task.signal->rlim[RLIMIT_SIGPENDING] =
201                 init_task.signal->rlim[RLIMIT_NPROC];
202 }
203
204 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
205                                                struct task_struct *src)
206 {
207         *dst = *src;
208         return 0;
209 }
210
211 static struct task_struct *dup_task_struct(struct task_struct *orig)
212 {
213         struct task_struct *tsk;
214         struct thread_info *ti;
215         int err;
216
217         prepare_to_copy(orig);
218
219         tsk = alloc_task_struct();
220         if (!tsk)
221                 return NULL;
222
223         ti = alloc_thread_info(tsk);
224         if (!ti) {
225                 free_task_struct(tsk);
226                 return NULL;
227         }
228
229         err = arch_dup_task_struct(tsk, orig);
230         if (err)
231                 goto out;
232
233         tsk->stack = ti;
234
235         err = prop_local_init_single(&tsk->dirties);
236         if (err)
237                 goto out;
238
239         setup_thread_stack(tsk, orig);
240
241 #ifdef CONFIG_CC_STACKPROTECTOR
242         tsk->stack_canary = get_random_int();
243 #endif
244
245         /* One for us, one for whoever does the "release_task()" (usually parent) */
246         atomic_set(&tsk->usage,2);
247         atomic_set(&tsk->fs_excl, 0);
248 #ifdef CONFIG_BLK_DEV_IO_TRACE
249         tsk->btrace_seq = 0;
250 #endif
251         tsk->splice_pipe = NULL;
252         return tsk;
253
254 out:
255         free_thread_info(ti);
256         free_task_struct(tsk);
257         return NULL;
258 }
259
260 #ifdef CONFIG_MMU
261 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
262 {
263         struct vm_area_struct *mpnt, *tmp, **pprev;
264         struct rb_node **rb_link, *rb_parent;
265         int retval;
266         unsigned long charge;
267         struct mempolicy *pol;
268
269         down_write(&oldmm->mmap_sem);
270         flush_cache_dup_mm(oldmm);
271         /*
272          * Not linked in yet - no deadlock potential:
273          */
274         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
275
276         mm->locked_vm = 0;
277         mm->mmap = NULL;
278         mm->mmap_cache = NULL;
279         mm->free_area_cache = oldmm->mmap_base;
280         mm->cached_hole_size = ~0UL;
281         mm->map_count = 0;
282         cpus_clear(mm->cpu_vm_mask);
283         mm->mm_rb = RB_ROOT;
284         rb_link = &mm->mm_rb.rb_node;
285         rb_parent = NULL;
286         pprev = &mm->mmap;
287
288         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
289                 struct file *file;
290
291                 if (mpnt->vm_flags & VM_DONTCOPY) {
292                         long pages = vma_pages(mpnt);
293                         mm->total_vm -= pages;
294                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
295                                                                 -pages);
296                         continue;
297                 }
298                 charge = 0;
299                 if (mpnt->vm_flags & VM_ACCOUNT) {
300                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
301                         if (security_vm_enough_memory(len))
302                                 goto fail_nomem;
303                         charge = len;
304                 }
305                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
306                 if (!tmp)
307                         goto fail_nomem;
308                 *tmp = *mpnt;
309                 pol = mpol_dup(vma_policy(mpnt));
310                 retval = PTR_ERR(pol);
311                 if (IS_ERR(pol))
312                         goto fail_nomem_policy;
313                 vma_set_policy(tmp, pol);
314                 tmp->vm_flags &= ~VM_LOCKED;
315                 tmp->vm_mm = mm;
316                 tmp->vm_next = NULL;
317                 anon_vma_link(tmp);
318                 file = tmp->vm_file;
319                 if (file) {
320                         struct inode *inode = file->f_path.dentry->d_inode;
321                         get_file(file);
322                         if (tmp->vm_flags & VM_DENYWRITE)
323                                 atomic_dec(&inode->i_writecount);
324
325                         /* insert tmp into the share list, just after mpnt */
326                         spin_lock(&file->f_mapping->i_mmap_lock);
327                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
328                         flush_dcache_mmap_lock(file->f_mapping);
329                         vma_prio_tree_add(tmp, mpnt);
330                         flush_dcache_mmap_unlock(file->f_mapping);
331                         spin_unlock(&file->f_mapping->i_mmap_lock);
332                 }
333
334                 /*
335                  * Clear hugetlb-related page reserves for children. This only
336                  * affects MAP_PRIVATE mappings. Faults generated by the child
337                  * are not guaranteed to succeed, even if read-only
338                  */
339                 if (is_vm_hugetlb_page(tmp))
340                         reset_vma_resv_huge_pages(tmp);
341
342                 /*
343                  * Link in the new vma and copy the page table entries.
344                  */
345                 *pprev = tmp;
346                 pprev = &tmp->vm_next;
347
348                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
349                 rb_link = &tmp->vm_rb.rb_right;
350                 rb_parent = &tmp->vm_rb;
351
352                 mm->map_count++;
353                 retval = copy_page_range(mm, oldmm, mpnt);
354
355                 if (tmp->vm_ops && tmp->vm_ops->open)
356                         tmp->vm_ops->open(tmp);
357
358                 if (retval)
359                         goto out;
360         }
361         /* a new mm has just been created */
362         arch_dup_mmap(oldmm, mm);
363         retval = 0;
364 out:
365         up_write(&mm->mmap_sem);
366         flush_tlb_mm(oldmm);
367         up_write(&oldmm->mmap_sem);
368         return retval;
369 fail_nomem_policy:
370         kmem_cache_free(vm_area_cachep, tmp);
371 fail_nomem:
372         retval = -ENOMEM;
373         vm_unacct_memory(charge);
374         goto out;
375 }
376
377 static inline int mm_alloc_pgd(struct mm_struct * mm)
378 {
379         mm->pgd = pgd_alloc(mm);
380         if (unlikely(!mm->pgd))
381                 return -ENOMEM;
382         return 0;
383 }
384
385 static inline void mm_free_pgd(struct mm_struct * mm)
386 {
387         pgd_free(mm, mm->pgd);
388 }
389 #else
390 #define dup_mmap(mm, oldmm)     (0)
391 #define mm_alloc_pgd(mm)        (0)
392 #define mm_free_pgd(mm)
393 #endif /* CONFIG_MMU */
394
395 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
396
397 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
398 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
399
400 #include <linux/init_task.h>
401
402 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
403 {
404         atomic_set(&mm->mm_users, 1);
405         atomic_set(&mm->mm_count, 1);
406         init_rwsem(&mm->mmap_sem);
407         INIT_LIST_HEAD(&mm->mmlist);
408         mm->flags = (current->mm) ? current->mm->flags
409                                   : MMF_DUMP_FILTER_DEFAULT;
410         mm->core_state = NULL;
411         mm->nr_ptes = 0;
412         set_mm_counter(mm, file_rss, 0);
413         set_mm_counter(mm, anon_rss, 0);
414         spin_lock_init(&mm->page_table_lock);
415         rwlock_init(&mm->ioctx_list_lock);
416         mm->ioctx_list = NULL;
417         mm->free_area_cache = TASK_UNMAPPED_BASE;
418         mm->cached_hole_size = ~0UL;
419         mm_init_owner(mm, p);
420
421         if (likely(!mm_alloc_pgd(mm))) {
422                 mm->def_flags = 0;
423                 mmu_notifier_mm_init(mm);
424                 return mm;
425         }
426
427         free_mm(mm);
428         return NULL;
429 }
430
431 /*
432  * Allocate and initialize an mm_struct.
433  */
434 struct mm_struct * mm_alloc(void)
435 {
436         struct mm_struct * mm;
437
438         mm = allocate_mm();
439         if (mm) {
440                 memset(mm, 0, sizeof(*mm));
441                 mm = mm_init(mm, current);
442         }
443         return mm;
444 }
445
446 /*
447  * Called when the last reference to the mm
448  * is dropped: either by a lazy thread or by
449  * mmput. Free the page directory and the mm.
450  */
451 void __mmdrop(struct mm_struct *mm)
452 {
453         BUG_ON(mm == &init_mm);
454         mm_free_pgd(mm);
455         destroy_context(mm);
456         mmu_notifier_mm_destroy(mm);
457         free_mm(mm);
458 }
459 EXPORT_SYMBOL_GPL(__mmdrop);
460
461 /*
462  * Decrement the use count and release all resources for an mm.
463  */
464 void mmput(struct mm_struct *mm)
465 {
466         might_sleep();
467
468         if (atomic_dec_and_test(&mm->mm_users)) {
469                 exit_aio(mm);
470                 exit_mmap(mm);
471                 set_mm_exe_file(mm, NULL);
472                 if (!list_empty(&mm->mmlist)) {
473                         spin_lock(&mmlist_lock);
474                         list_del(&mm->mmlist);
475                         spin_unlock(&mmlist_lock);
476                 }
477                 put_swap_token(mm);
478                 mmdrop(mm);
479         }
480 }
481 EXPORT_SYMBOL_GPL(mmput);
482
483 /**
484  * get_task_mm - acquire a reference to the task's mm
485  *
486  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
487  * this kernel workthread has transiently adopted a user mm with use_mm,
488  * to do its AIO) is not set and if so returns a reference to it, after
489  * bumping up the use count.  User must release the mm via mmput()
490  * after use.  Typically used by /proc and ptrace.
491  */
492 struct mm_struct *get_task_mm(struct task_struct *task)
493 {
494         struct mm_struct *mm;
495
496         task_lock(task);
497         mm = task->mm;
498         if (mm) {
499                 if (task->flags & PF_KTHREAD)
500                         mm = NULL;
501                 else
502                         atomic_inc(&mm->mm_users);
503         }
504         task_unlock(task);
505         return mm;
506 }
507 EXPORT_SYMBOL_GPL(get_task_mm);
508
509 /* Please note the differences between mmput and mm_release.
510  * mmput is called whenever we stop holding onto a mm_struct,
511  * error success whatever.
512  *
513  * mm_release is called after a mm_struct has been removed
514  * from the current process.
515  *
516  * This difference is important for error handling, when we
517  * only half set up a mm_struct for a new process and need to restore
518  * the old one.  Because we mmput the new mm_struct before
519  * restoring the old one. . .
520  * Eric Biederman 10 January 1998
521  */
522 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
523 {
524         struct completion *vfork_done = tsk->vfork_done;
525
526         /* Get rid of any futexes when releasing the mm */
527 #ifdef CONFIG_FUTEX
528         if (unlikely(tsk->robust_list))
529                 exit_robust_list(tsk);
530 #ifdef CONFIG_COMPAT
531         if (unlikely(tsk->compat_robust_list))
532                 compat_exit_robust_list(tsk);
533 #endif
534 #endif
535
536         /* Get rid of any cached register state */
537         deactivate_mm(tsk, mm);
538
539         /* notify parent sleeping on vfork() */
540         if (vfork_done) {
541                 tsk->vfork_done = NULL;
542                 complete(vfork_done);
543         }
544
545         /*
546          * If we're exiting normally, clear a user-space tid field if
547          * requested.  We leave this alone when dying by signal, to leave
548          * the value intact in a core dump, and to save the unnecessary
549          * trouble otherwise.  Userland only wants this done for a sys_exit.
550          */
551         if (tsk->clear_child_tid
552             && !(tsk->flags & PF_SIGNALED)
553             && atomic_read(&mm->mm_users) > 1) {
554                 u32 __user * tidptr = tsk->clear_child_tid;
555                 tsk->clear_child_tid = NULL;
556
557                 /*
558                  * We don't check the error code - if userspace has
559                  * not set up a proper pointer then tough luck.
560                  */
561                 put_user(0, tidptr);
562                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
563         }
564 }
565
566 /*
567  * Allocate a new mm structure and copy contents from the
568  * mm structure of the passed in task structure.
569  */
570 struct mm_struct *dup_mm(struct task_struct *tsk)
571 {
572         struct mm_struct *mm, *oldmm = current->mm;
573         int err;
574
575         if (!oldmm)
576                 return NULL;
577
578         mm = allocate_mm();
579         if (!mm)
580                 goto fail_nomem;
581
582         memcpy(mm, oldmm, sizeof(*mm));
583
584         /* Initializing for Swap token stuff */
585         mm->token_priority = 0;
586         mm->last_interval = 0;
587
588         if (!mm_init(mm, tsk))
589                 goto fail_nomem;
590
591         if (init_new_context(tsk, mm))
592                 goto fail_nocontext;
593
594         dup_mm_exe_file(oldmm, mm);
595
596         err = dup_mmap(mm, oldmm);
597         if (err)
598                 goto free_pt;
599
600         mm->hiwater_rss = get_mm_rss(mm);
601         mm->hiwater_vm = mm->total_vm;
602
603         return mm;
604
605 free_pt:
606         mmput(mm);
607
608 fail_nomem:
609         return NULL;
610
611 fail_nocontext:
612         /*
613          * If init_new_context() failed, we cannot use mmput() to free the mm
614          * because it calls destroy_context()
615          */
616         mm_free_pgd(mm);
617         free_mm(mm);
618         return NULL;
619 }
620
621 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
622 {
623         struct mm_struct * mm, *oldmm;
624         int retval;
625
626         tsk->min_flt = tsk->maj_flt = 0;
627         tsk->nvcsw = tsk->nivcsw = 0;
628
629         tsk->mm = NULL;
630         tsk->active_mm = NULL;
631
632         /*
633          * Are we cloning a kernel thread?
634          *
635          * We need to steal a active VM for that..
636          */
637         oldmm = current->mm;
638         if (!oldmm)
639                 return 0;
640
641         if (clone_flags & CLONE_VM) {
642                 atomic_inc(&oldmm->mm_users);
643                 mm = oldmm;
644                 goto good_mm;
645         }
646
647         retval = -ENOMEM;
648         mm = dup_mm(tsk);
649         if (!mm)
650                 goto fail_nomem;
651
652 good_mm:
653         /* Initializing for Swap token stuff */
654         mm->token_priority = 0;
655         mm->last_interval = 0;
656
657         tsk->mm = mm;
658         tsk->active_mm = mm;
659         return 0;
660
661 fail_nomem:
662         return retval;
663 }
664
665 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
666 {
667         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
668         /* We don't need to lock fs - think why ;-) */
669         if (fs) {
670                 atomic_set(&fs->count, 1);
671                 rwlock_init(&fs->lock);
672                 fs->umask = old->umask;
673                 read_lock(&old->lock);
674                 fs->root = old->root;
675                 path_get(&old->root);
676                 fs->pwd = old->pwd;
677                 path_get(&old->pwd);
678                 read_unlock(&old->lock);
679         }
680         return fs;
681 }
682
683 struct fs_struct *copy_fs_struct(struct fs_struct *old)
684 {
685         return __copy_fs_struct(old);
686 }
687
688 EXPORT_SYMBOL_GPL(copy_fs_struct);
689
690 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
691 {
692         if (clone_flags & CLONE_FS) {
693                 atomic_inc(&current->fs->count);
694                 return 0;
695         }
696         tsk->fs = __copy_fs_struct(current->fs);
697         if (!tsk->fs)
698                 return -ENOMEM;
699         return 0;
700 }
701
702 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
703 {
704         struct files_struct *oldf, *newf;
705         int error = 0;
706
707         /*
708          * A background process may not have any files ...
709          */
710         oldf = current->files;
711         if (!oldf)
712                 goto out;
713
714         if (clone_flags & CLONE_FILES) {
715                 atomic_inc(&oldf->count);
716                 goto out;
717         }
718
719         newf = dup_fd(oldf, &error);
720         if (!newf)
721                 goto out;
722
723         tsk->files = newf;
724         error = 0;
725 out:
726         return error;
727 }
728
729 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
730 {
731 #ifdef CONFIG_BLOCK
732         struct io_context *ioc = current->io_context;
733
734         if (!ioc)
735                 return 0;
736         /*
737          * Share io context with parent, if CLONE_IO is set
738          */
739         if (clone_flags & CLONE_IO) {
740                 tsk->io_context = ioc_task_link(ioc);
741                 if (unlikely(!tsk->io_context))
742                         return -ENOMEM;
743         } else if (ioprio_valid(ioc->ioprio)) {
744                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
745                 if (unlikely(!tsk->io_context))
746                         return -ENOMEM;
747
748                 tsk->io_context->ioprio = ioc->ioprio;
749         }
750 #endif
751         return 0;
752 }
753
754 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
755 {
756         struct sighand_struct *sig;
757
758         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
759                 atomic_inc(&current->sighand->count);
760                 return 0;
761         }
762         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
763         rcu_assign_pointer(tsk->sighand, sig);
764         if (!sig)
765                 return -ENOMEM;
766         atomic_set(&sig->count, 1);
767         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
768         return 0;
769 }
770
771 void __cleanup_sighand(struct sighand_struct *sighand)
772 {
773         if (atomic_dec_and_test(&sighand->count))
774                 kmem_cache_free(sighand_cachep, sighand);
775 }
776
777
778 /*
779  * Initialize POSIX timer handling for a thread group.
780  */
781 static void posix_cpu_timers_init_group(struct signal_struct *sig)
782 {
783         /* Thread group counters. */
784         thread_group_cputime_init(sig);
785
786         /* Expiration times and increments. */
787         sig->it_virt_expires = cputime_zero;
788         sig->it_virt_incr = cputime_zero;
789         sig->it_prof_expires = cputime_zero;
790         sig->it_prof_incr = cputime_zero;
791
792         /* Cached expiration times. */
793         sig->cputime_expires.prof_exp = cputime_zero;
794         sig->cputime_expires.virt_exp = cputime_zero;
795         sig->cputime_expires.sched_exp = 0;
796
797         /* The timer lists. */
798         INIT_LIST_HEAD(&sig->cpu_timers[0]);
799         INIT_LIST_HEAD(&sig->cpu_timers[1]);
800         INIT_LIST_HEAD(&sig->cpu_timers[2]);
801 }
802
803 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
804 {
805         struct signal_struct *sig;
806         int ret;
807
808         if (clone_flags & CLONE_THREAD) {
809                 ret = thread_group_cputime_clone_thread(current);
810                 if (likely(!ret)) {
811                         atomic_inc(&current->signal->count);
812                         atomic_inc(&current->signal->live);
813                 }
814                 return ret;
815         }
816         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
817         tsk->signal = sig;
818         if (!sig)
819                 return -ENOMEM;
820
821         ret = copy_thread_group_keys(tsk);
822         if (ret < 0) {
823                 kmem_cache_free(signal_cachep, sig);
824                 return ret;
825         }
826
827         atomic_set(&sig->count, 1);
828         atomic_set(&sig->live, 1);
829         init_waitqueue_head(&sig->wait_chldexit);
830         sig->flags = 0;
831         sig->group_exit_code = 0;
832         sig->group_exit_task = NULL;
833         sig->group_stop_count = 0;
834         sig->curr_target = tsk;
835         init_sigpending(&sig->shared_pending);
836         INIT_LIST_HEAD(&sig->posix_timers);
837
838         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
839         sig->it_real_incr.tv64 = 0;
840         sig->real_timer.function = it_real_fn;
841
842         sig->leader = 0;        /* session leadership doesn't inherit */
843         sig->tty_old_pgrp = NULL;
844         sig->tty = NULL;
845
846         sig->cutime = sig->cstime = cputime_zero;
847         sig->gtime = cputime_zero;
848         sig->cgtime = cputime_zero;
849         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
850         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
851         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
852         task_io_accounting_init(&sig->ioac);
853         taskstats_tgid_init(sig);
854
855         task_lock(current->group_leader);
856         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
857         task_unlock(current->group_leader);
858
859         posix_cpu_timers_init_group(sig);
860
861         acct_init_pacct(&sig->pacct);
862
863         tty_audit_fork(sig);
864
865         return 0;
866 }
867
868 void __cleanup_signal(struct signal_struct *sig)
869 {
870         thread_group_cputime_free(sig);
871         exit_thread_group_keys(sig);
872         tty_kref_put(sig->tty);
873         kmem_cache_free(signal_cachep, sig);
874 }
875
876 static void cleanup_signal(struct task_struct *tsk)
877 {
878         struct signal_struct *sig = tsk->signal;
879
880         atomic_dec(&sig->live);
881
882         if (atomic_dec_and_test(&sig->count))
883                 __cleanup_signal(sig);
884 }
885
886 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
887 {
888         unsigned long new_flags = p->flags;
889
890         new_flags &= ~PF_SUPERPRIV;
891         new_flags |= PF_FORKNOEXEC;
892         new_flags |= PF_STARTING;
893         p->flags = new_flags;
894         clear_freeze_flag(p);
895 }
896
897 asmlinkage long sys_set_tid_address(int __user *tidptr)
898 {
899         current->clear_child_tid = tidptr;
900
901         return task_pid_vnr(current);
902 }
903
904 static void rt_mutex_init_task(struct task_struct *p)
905 {
906         spin_lock_init(&p->pi_lock);
907 #ifdef CONFIG_RT_MUTEXES
908         plist_head_init(&p->pi_waiters, &p->pi_lock);
909         p->pi_blocked_on = NULL;
910 #endif
911 }
912
913 #ifdef CONFIG_MM_OWNER
914 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
915 {
916         mm->owner = p;
917 }
918 #endif /* CONFIG_MM_OWNER */
919
920 /*
921  * Initialize POSIX timer handling for a single task.
922  */
923 static void posix_cpu_timers_init(struct task_struct *tsk)
924 {
925         tsk->cputime_expires.prof_exp = cputime_zero;
926         tsk->cputime_expires.virt_exp = cputime_zero;
927         tsk->cputime_expires.sched_exp = 0;
928         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
929         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
930         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
931 }
932
933 /*
934  * This creates a new process as a copy of the old one,
935  * but does not actually start it yet.
936  *
937  * It copies the registers, and all the appropriate
938  * parts of the process environment (as per the clone
939  * flags). The actual kick-off is left to the caller.
940  */
941 static struct task_struct *copy_process(unsigned long clone_flags,
942                                         unsigned long stack_start,
943                                         struct pt_regs *regs,
944                                         unsigned long stack_size,
945                                         int __user *child_tidptr,
946                                         struct pid *pid,
947                                         int trace)
948 {
949         int retval;
950         struct task_struct *p;
951         int cgroup_callbacks_done = 0;
952
953         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
954                 return ERR_PTR(-EINVAL);
955
956         /*
957          * Thread groups must share signals as well, and detached threads
958          * can only be started up within the thread group.
959          */
960         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
961                 return ERR_PTR(-EINVAL);
962
963         /*
964          * Shared signal handlers imply shared VM. By way of the above,
965          * thread groups also imply shared VM. Blocking this case allows
966          * for various simplifications in other code.
967          */
968         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
969                 return ERR_PTR(-EINVAL);
970
971         retval = security_task_create(clone_flags);
972         if (retval)
973                 goto fork_out;
974
975         retval = -ENOMEM;
976         p = dup_task_struct(current);
977         if (!p)
978                 goto fork_out;
979
980         rt_mutex_init_task(p);
981
982 #ifdef CONFIG_PROVE_LOCKING
983         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
984         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
985 #endif
986         retval = -EAGAIN;
987         if (atomic_read(&p->user->processes) >=
988                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
989                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
990                     p->user != current->nsproxy->user_ns->root_user)
991                         goto bad_fork_free;
992         }
993
994         atomic_inc(&p->user->__count);
995         atomic_inc(&p->user->processes);
996         get_group_info(p->group_info);
997
998         /*
999          * If multiple threads are within copy_process(), then this check
1000          * triggers too late. This doesn't hurt, the check is only there
1001          * to stop root fork bombs.
1002          */
1003         if (nr_threads >= max_threads)
1004                 goto bad_fork_cleanup_count;
1005
1006         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1007                 goto bad_fork_cleanup_count;
1008
1009         if (p->binfmt && !try_module_get(p->binfmt->module))
1010                 goto bad_fork_cleanup_put_domain;
1011
1012         p->did_exec = 0;
1013         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1014         copy_flags(clone_flags, p);
1015         INIT_LIST_HEAD(&p->children);
1016         INIT_LIST_HEAD(&p->sibling);
1017 #ifdef CONFIG_PREEMPT_RCU
1018         p->rcu_read_lock_nesting = 0;
1019         p->rcu_flipctr_idx = 0;
1020 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1021         p->vfork_done = NULL;
1022         spin_lock_init(&p->alloc_lock);
1023
1024         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1025         init_sigpending(&p->pending);
1026
1027         p->utime = cputime_zero;
1028         p->stime = cputime_zero;
1029         p->gtime = cputime_zero;
1030         p->utimescaled = cputime_zero;
1031         p->stimescaled = cputime_zero;
1032         p->prev_utime = cputime_zero;
1033         p->prev_stime = cputime_zero;
1034
1035         p->default_timer_slack_ns = current->timer_slack_ns;
1036
1037 #ifdef CONFIG_DETECT_SOFTLOCKUP
1038         p->last_switch_count = 0;
1039         p->last_switch_timestamp = 0;
1040 #endif
1041
1042         task_io_accounting_init(&p->ioac);
1043         acct_clear_integrals(p);
1044
1045         posix_cpu_timers_init(p);
1046
1047         p->lock_depth = -1;             /* -1 = no lock */
1048         do_posix_clock_monotonic_gettime(&p->start_time);
1049         p->real_start_time = p->start_time;
1050         monotonic_to_bootbased(&p->real_start_time);
1051 #ifdef CONFIG_SECURITY
1052         p->security = NULL;
1053 #endif
1054         p->cap_bset = current->cap_bset;
1055         p->io_context = NULL;
1056         p->audit_context = NULL;
1057         cgroup_fork(p);
1058 #ifdef CONFIG_NUMA
1059         p->mempolicy = mpol_dup(p->mempolicy);
1060         if (IS_ERR(p->mempolicy)) {
1061                 retval = PTR_ERR(p->mempolicy);
1062                 p->mempolicy = NULL;
1063                 goto bad_fork_cleanup_cgroup;
1064         }
1065         mpol_fix_fork_child_flag(p);
1066 #endif
1067 #ifdef CONFIG_TRACE_IRQFLAGS
1068         p->irq_events = 0;
1069 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1070         p->hardirqs_enabled = 1;
1071 #else
1072         p->hardirqs_enabled = 0;
1073 #endif
1074         p->hardirq_enable_ip = 0;
1075         p->hardirq_enable_event = 0;
1076         p->hardirq_disable_ip = _THIS_IP_;
1077         p->hardirq_disable_event = 0;
1078         p->softirqs_enabled = 1;
1079         p->softirq_enable_ip = _THIS_IP_;
1080         p->softirq_enable_event = 0;
1081         p->softirq_disable_ip = 0;
1082         p->softirq_disable_event = 0;
1083         p->hardirq_context = 0;
1084         p->softirq_context = 0;
1085 #endif
1086 #ifdef CONFIG_LOCKDEP
1087         p->lockdep_depth = 0; /* no locks held yet */
1088         p->curr_chain_key = 0;
1089         p->lockdep_recursion = 0;
1090 #endif
1091
1092 #ifdef CONFIG_DEBUG_MUTEXES
1093         p->blocked_on = NULL; /* not blocked yet */
1094 #endif
1095
1096         /* Perform scheduler related setup. Assign this task to a CPU. */
1097         sched_fork(p, clone_flags);
1098
1099         if ((retval = security_task_alloc(p)))
1100                 goto bad_fork_cleanup_policy;
1101         if ((retval = audit_alloc(p)))
1102                 goto bad_fork_cleanup_security;
1103         /* copy all the process information */
1104         if ((retval = copy_semundo(clone_flags, p)))
1105                 goto bad_fork_cleanup_audit;
1106         if ((retval = copy_files(clone_flags, p)))
1107                 goto bad_fork_cleanup_semundo;
1108         if ((retval = copy_fs(clone_flags, p)))
1109                 goto bad_fork_cleanup_files;
1110         if ((retval = copy_sighand(clone_flags, p)))
1111                 goto bad_fork_cleanup_fs;
1112         if ((retval = copy_signal(clone_flags, p)))
1113                 goto bad_fork_cleanup_sighand;
1114         if ((retval = copy_mm(clone_flags, p)))
1115                 goto bad_fork_cleanup_signal;
1116         if ((retval = copy_keys(clone_flags, p)))
1117                 goto bad_fork_cleanup_mm;
1118         if ((retval = copy_namespaces(clone_flags, p)))
1119                 goto bad_fork_cleanup_keys;
1120         if ((retval = copy_io(clone_flags, p)))
1121                 goto bad_fork_cleanup_namespaces;
1122         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1123         if (retval)
1124                 goto bad_fork_cleanup_io;
1125
1126         if (pid != &init_struct_pid) {
1127                 retval = -ENOMEM;
1128                 pid = alloc_pid(task_active_pid_ns(p));
1129                 if (!pid)
1130                         goto bad_fork_cleanup_io;
1131
1132                 if (clone_flags & CLONE_NEWPID) {
1133                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1134                         if (retval < 0)
1135                                 goto bad_fork_free_pid;
1136                 }
1137         }
1138
1139         p->pid = pid_nr(pid);
1140         p->tgid = p->pid;
1141         if (clone_flags & CLONE_THREAD)
1142                 p->tgid = current->tgid;
1143
1144         if (current->nsproxy != p->nsproxy) {
1145                 retval = ns_cgroup_clone(p, pid);
1146                 if (retval)
1147                         goto bad_fork_free_pid;
1148         }
1149
1150         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1151         /*
1152          * Clear TID on mm_release()?
1153          */
1154         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1155 #ifdef CONFIG_FUTEX
1156         p->robust_list = NULL;
1157 #ifdef CONFIG_COMPAT
1158         p->compat_robust_list = NULL;
1159 #endif
1160         INIT_LIST_HEAD(&p->pi_state_list);
1161         p->pi_state_cache = NULL;
1162 #endif
1163         /*
1164          * sigaltstack should be cleared when sharing the same VM
1165          */
1166         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1167                 p->sas_ss_sp = p->sas_ss_size = 0;
1168
1169         /*
1170          * Syscall tracing should be turned off in the child regardless
1171          * of CLONE_PTRACE.
1172          */
1173         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1174 #ifdef TIF_SYSCALL_EMU
1175         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1176 #endif
1177         clear_all_latency_tracing(p);
1178
1179         /* Our parent execution domain becomes current domain
1180            These must match for thread signalling to apply */
1181         p->parent_exec_id = p->self_exec_id;
1182
1183         /* ok, now we should be set up.. */
1184         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1185         p->pdeath_signal = 0;
1186         p->exit_state = 0;
1187
1188         /*
1189          * Ok, make it visible to the rest of the system.
1190          * We dont wake it up yet.
1191          */
1192         p->group_leader = p;
1193         INIT_LIST_HEAD(&p->thread_group);
1194
1195         /* Now that the task is set up, run cgroup callbacks if
1196          * necessary. We need to run them before the task is visible
1197          * on the tasklist. */
1198         cgroup_fork_callbacks(p);
1199         cgroup_callbacks_done = 1;
1200
1201         /* Need tasklist lock for parent etc handling! */
1202         write_lock_irq(&tasklist_lock);
1203
1204         /*
1205          * The task hasn't been attached yet, so its cpus_allowed mask will
1206          * not be changed, nor will its assigned CPU.
1207          *
1208          * The cpus_allowed mask of the parent may have changed after it was
1209          * copied first time - so re-copy it here, then check the child's CPU
1210          * to ensure it is on a valid CPU (and if not, just force it back to
1211          * parent's CPU). This avoids alot of nasty races.
1212          */
1213         p->cpus_allowed = current->cpus_allowed;
1214         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1215         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1216                         !cpu_online(task_cpu(p))))
1217                 set_task_cpu(p, smp_processor_id());
1218
1219         /* CLONE_PARENT re-uses the old parent */
1220         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1221                 p->real_parent = current->real_parent;
1222         else
1223                 p->real_parent = current;
1224
1225         spin_lock(&current->sighand->siglock);
1226
1227         /*
1228          * Process group and session signals need to be delivered to just the
1229          * parent before the fork or both the parent and the child after the
1230          * fork. Restart if a signal comes in before we add the new process to
1231          * it's process group.
1232          * A fatal signal pending means that current will exit, so the new
1233          * thread can't slip out of an OOM kill (or normal SIGKILL).
1234          */
1235         recalc_sigpending();
1236         if (signal_pending(current)) {
1237                 spin_unlock(&current->sighand->siglock);
1238                 write_unlock_irq(&tasklist_lock);
1239                 retval = -ERESTARTNOINTR;
1240                 goto bad_fork_free_pid;
1241         }
1242
1243         if (clone_flags & CLONE_THREAD) {
1244                 p->group_leader = current->group_leader;
1245                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1246         }
1247
1248         if (likely(p->pid)) {
1249                 list_add_tail(&p->sibling, &p->real_parent->children);
1250                 tracehook_finish_clone(p, clone_flags, trace);
1251
1252                 if (thread_group_leader(p)) {
1253                         if (clone_flags & CLONE_NEWPID)
1254                                 p->nsproxy->pid_ns->child_reaper = p;
1255
1256                         p->signal->leader_pid = pid;
1257                         tty_kref_put(p->signal->tty);
1258                         p->signal->tty = tty_kref_get(current->signal->tty);
1259                         set_task_pgrp(p, task_pgrp_nr(current));
1260                         set_task_session(p, task_session_nr(current));
1261                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1262                         attach_pid(p, PIDTYPE_SID, task_session(current));
1263                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1264                         __get_cpu_var(process_counts)++;
1265                 }
1266                 attach_pid(p, PIDTYPE_PID, pid);
1267                 nr_threads++;
1268         }
1269
1270         total_forks++;
1271         spin_unlock(&current->sighand->siglock);
1272         write_unlock_irq(&tasklist_lock);
1273 #ifdef CONFIG_FUNCTION_RET_TRACER
1274         ftrace_retfunc_init_task(p);
1275 #endif
1276         proc_fork_connector(p);
1277         cgroup_post_fork(p);
1278         return p;
1279
1280 bad_fork_free_pid:
1281         if (pid != &init_struct_pid)
1282                 free_pid(pid);
1283 bad_fork_cleanup_io:
1284         put_io_context(p->io_context);
1285 bad_fork_cleanup_namespaces:
1286         exit_task_namespaces(p);
1287 bad_fork_cleanup_keys:
1288         exit_keys(p);
1289 bad_fork_cleanup_mm:
1290         if (p->mm)
1291                 mmput(p->mm);
1292 bad_fork_cleanup_signal:
1293         cleanup_signal(p);
1294 bad_fork_cleanup_sighand:
1295         __cleanup_sighand(p->sighand);
1296 bad_fork_cleanup_fs:
1297         exit_fs(p); /* blocking */
1298 bad_fork_cleanup_files:
1299         exit_files(p); /* blocking */
1300 bad_fork_cleanup_semundo:
1301         exit_sem(p);
1302 bad_fork_cleanup_audit:
1303         audit_free(p);
1304 bad_fork_cleanup_security:
1305         security_task_free(p);
1306 bad_fork_cleanup_policy:
1307 #ifdef CONFIG_NUMA
1308         mpol_put(p->mempolicy);
1309 bad_fork_cleanup_cgroup:
1310 #endif
1311         cgroup_exit(p, cgroup_callbacks_done);
1312         delayacct_tsk_free(p);
1313         if (p->binfmt)
1314                 module_put(p->binfmt->module);
1315 bad_fork_cleanup_put_domain:
1316         module_put(task_thread_info(p)->exec_domain->module);
1317 bad_fork_cleanup_count:
1318         put_group_info(p->group_info);
1319         atomic_dec(&p->user->processes);
1320         free_uid(p->user);
1321 bad_fork_free:
1322         free_task(p);
1323 fork_out:
1324         return ERR_PTR(retval);
1325 }
1326
1327 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1328 {
1329         memset(regs, 0, sizeof(struct pt_regs));
1330         return regs;
1331 }
1332
1333 struct task_struct * __cpuinit fork_idle(int cpu)
1334 {
1335         struct task_struct *task;
1336         struct pt_regs regs;
1337
1338         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1339                             &init_struct_pid, 0);
1340         if (!IS_ERR(task))
1341                 init_idle(task, cpu);
1342
1343         return task;
1344 }
1345
1346 /*
1347  *  Ok, this is the main fork-routine.
1348  *
1349  * It copies the process, and if successful kick-starts
1350  * it and waits for it to finish using the VM if required.
1351  */
1352 long do_fork(unsigned long clone_flags,
1353               unsigned long stack_start,
1354               struct pt_regs *regs,
1355               unsigned long stack_size,
1356               int __user *parent_tidptr,
1357               int __user *child_tidptr)
1358 {
1359         struct task_struct *p;
1360         int trace = 0;
1361         long nr;
1362
1363         /*
1364          * We hope to recycle these flags after 2.6.26
1365          */
1366         if (unlikely(clone_flags & CLONE_STOPPED)) {
1367                 static int __read_mostly count = 100;
1368
1369                 if (count > 0 && printk_ratelimit()) {
1370                         char comm[TASK_COMM_LEN];
1371
1372                         count--;
1373                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1374                                         "clone flags 0x%lx\n",
1375                                 get_task_comm(comm, current),
1376                                 clone_flags & CLONE_STOPPED);
1377                 }
1378         }
1379
1380         /*
1381          * When called from kernel_thread, don't do user tracing stuff.
1382          */
1383         if (likely(user_mode(regs)))
1384                 trace = tracehook_prepare_clone(clone_flags);
1385
1386         p = copy_process(clone_flags, stack_start, regs, stack_size,
1387                          child_tidptr, NULL, trace);
1388         /*
1389          * Do this prior waking up the new thread - the thread pointer
1390          * might get invalid after that point, if the thread exits quickly.
1391          */
1392         if (!IS_ERR(p)) {
1393                 struct completion vfork;
1394
1395                 trace_sched_process_fork(current, p);
1396
1397                 nr = task_pid_vnr(p);
1398
1399                 if (clone_flags & CLONE_PARENT_SETTID)
1400                         put_user(nr, parent_tidptr);
1401
1402                 if (clone_flags & CLONE_VFORK) {
1403                         p->vfork_done = &vfork;
1404                         init_completion(&vfork);
1405                 }
1406
1407                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1408
1409                 /*
1410                  * We set PF_STARTING at creation in case tracing wants to
1411                  * use this to distinguish a fully live task from one that
1412                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1413                  * clear it and set the child going.
1414                  */
1415                 p->flags &= ~PF_STARTING;
1416
1417                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1418                         /*
1419                          * We'll start up with an immediate SIGSTOP.
1420                          */
1421                         sigaddset(&p->pending.signal, SIGSTOP);
1422                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1423                         __set_task_state(p, TASK_STOPPED);
1424                 } else {
1425                         wake_up_new_task(p, clone_flags);
1426                 }
1427
1428                 tracehook_report_clone_complete(trace, regs,
1429                                                 clone_flags, nr, p);
1430
1431                 if (clone_flags & CLONE_VFORK) {
1432                         freezer_do_not_count();
1433                         wait_for_completion(&vfork);
1434                         freezer_count();
1435                         tracehook_report_vfork_done(p, nr);
1436                 }
1437         } else {
1438                 nr = PTR_ERR(p);
1439         }
1440         return nr;
1441 }
1442
1443 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1444 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1445 #endif
1446
1447 static void sighand_ctor(void *data)
1448 {
1449         struct sighand_struct *sighand = data;
1450
1451         spin_lock_init(&sighand->siglock);
1452         init_waitqueue_head(&sighand->signalfd_wqh);
1453 }
1454
1455 void __init proc_caches_init(void)
1456 {
1457         sighand_cachep = kmem_cache_create("sighand_cache",
1458                         sizeof(struct sighand_struct), 0,
1459                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1460                         sighand_ctor);
1461         signal_cachep = kmem_cache_create("signal_cache",
1462                         sizeof(struct signal_struct), 0,
1463                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1464         files_cachep = kmem_cache_create("files_cache",
1465                         sizeof(struct files_struct), 0,
1466                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1467         fs_cachep = kmem_cache_create("fs_cache",
1468                         sizeof(struct fs_struct), 0,
1469                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1470         vm_area_cachep = kmem_cache_create("vm_area_struct",
1471                         sizeof(struct vm_area_struct), 0,
1472                         SLAB_PANIC, NULL);
1473         mm_cachep = kmem_cache_create("mm_struct",
1474                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1475                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1476 }
1477
1478 /*
1479  * Check constraints on flags passed to the unshare system call and
1480  * force unsharing of additional process context as appropriate.
1481  */
1482 static void check_unshare_flags(unsigned long *flags_ptr)
1483 {
1484         /*
1485          * If unsharing a thread from a thread group, must also
1486          * unshare vm.
1487          */
1488         if (*flags_ptr & CLONE_THREAD)
1489                 *flags_ptr |= CLONE_VM;
1490
1491         /*
1492          * If unsharing vm, must also unshare signal handlers.
1493          */
1494         if (*flags_ptr & CLONE_VM)
1495                 *flags_ptr |= CLONE_SIGHAND;
1496
1497         /*
1498          * If unsharing signal handlers and the task was created
1499          * using CLONE_THREAD, then must unshare the thread
1500          */
1501         if ((*flags_ptr & CLONE_SIGHAND) &&
1502             (atomic_read(&current->signal->count) > 1))
1503                 *flags_ptr |= CLONE_THREAD;
1504
1505         /*
1506          * If unsharing namespace, must also unshare filesystem information.
1507          */
1508         if (*flags_ptr & CLONE_NEWNS)
1509                 *flags_ptr |= CLONE_FS;
1510 }
1511
1512 /*
1513  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1514  */
1515 static int unshare_thread(unsigned long unshare_flags)
1516 {
1517         if (unshare_flags & CLONE_THREAD)
1518                 return -EINVAL;
1519
1520         return 0;
1521 }
1522
1523 /*
1524  * Unshare the filesystem structure if it is being shared
1525  */
1526 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1527 {
1528         struct fs_struct *fs = current->fs;
1529
1530         if ((unshare_flags & CLONE_FS) &&
1531             (fs && atomic_read(&fs->count) > 1)) {
1532                 *new_fsp = __copy_fs_struct(current->fs);
1533                 if (!*new_fsp)
1534                         return -ENOMEM;
1535         }
1536
1537         return 0;
1538 }
1539
1540 /*
1541  * Unsharing of sighand is not supported yet
1542  */
1543 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1544 {
1545         struct sighand_struct *sigh = current->sighand;
1546
1547         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1548                 return -EINVAL;
1549         else
1550                 return 0;
1551 }
1552
1553 /*
1554  * Unshare vm if it is being shared
1555  */
1556 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1557 {
1558         struct mm_struct *mm = current->mm;
1559
1560         if ((unshare_flags & CLONE_VM) &&
1561             (mm && atomic_read(&mm->mm_users) > 1)) {
1562                 return -EINVAL;
1563         }
1564
1565         return 0;
1566 }
1567
1568 /*
1569  * Unshare file descriptor table if it is being shared
1570  */
1571 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1572 {
1573         struct files_struct *fd = current->files;
1574         int error = 0;
1575
1576         if ((unshare_flags & CLONE_FILES) &&
1577             (fd && atomic_read(&fd->count) > 1)) {
1578                 *new_fdp = dup_fd(fd, &error);
1579                 if (!*new_fdp)
1580                         return error;
1581         }
1582
1583         return 0;
1584 }
1585
1586 /*
1587  * unshare allows a process to 'unshare' part of the process
1588  * context which was originally shared using clone.  copy_*
1589  * functions used by do_fork() cannot be used here directly
1590  * because they modify an inactive task_struct that is being
1591  * constructed. Here we are modifying the current, active,
1592  * task_struct.
1593  */
1594 asmlinkage long sys_unshare(unsigned long unshare_flags)
1595 {
1596         int err = 0;
1597         struct fs_struct *fs, *new_fs = NULL;
1598         struct sighand_struct *new_sigh = NULL;
1599         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1600         struct files_struct *fd, *new_fd = NULL;
1601         struct nsproxy *new_nsproxy = NULL;
1602         int do_sysvsem = 0;
1603
1604         check_unshare_flags(&unshare_flags);
1605
1606         /* Return -EINVAL for all unsupported flags */
1607         err = -EINVAL;
1608         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1609                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1610                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1611                                 CLONE_NEWNET))
1612                 goto bad_unshare_out;
1613
1614         /*
1615          * CLONE_NEWIPC must also detach from the undolist: after switching
1616          * to a new ipc namespace, the semaphore arrays from the old
1617          * namespace are unreachable.
1618          */
1619         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1620                 do_sysvsem = 1;
1621         if ((err = unshare_thread(unshare_flags)))
1622                 goto bad_unshare_out;
1623         if ((err = unshare_fs(unshare_flags, &new_fs)))
1624                 goto bad_unshare_cleanup_thread;
1625         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1626                 goto bad_unshare_cleanup_fs;
1627         if ((err = unshare_vm(unshare_flags, &new_mm)))
1628                 goto bad_unshare_cleanup_sigh;
1629         if ((err = unshare_fd(unshare_flags, &new_fd)))
1630                 goto bad_unshare_cleanup_vm;
1631         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1632                         new_fs)))
1633                 goto bad_unshare_cleanup_fd;
1634
1635         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1636                 if (do_sysvsem) {
1637                         /*
1638                          * CLONE_SYSVSEM is equivalent to sys_exit().
1639                          */
1640                         exit_sem(current);
1641                 }
1642
1643                 if (new_nsproxy) {
1644                         switch_task_namespaces(current, new_nsproxy);
1645                         new_nsproxy = NULL;
1646                 }
1647
1648                 task_lock(current);
1649
1650                 if (new_fs) {
1651                         fs = current->fs;
1652                         current->fs = new_fs;
1653                         new_fs = fs;
1654                 }
1655
1656                 if (new_mm) {
1657                         mm = current->mm;
1658                         active_mm = current->active_mm;
1659                         current->mm = new_mm;
1660                         current->active_mm = new_mm;
1661                         activate_mm(active_mm, new_mm);
1662                         new_mm = mm;
1663                 }
1664
1665                 if (new_fd) {
1666                         fd = current->files;
1667                         current->files = new_fd;
1668                         new_fd = fd;
1669                 }
1670
1671                 task_unlock(current);
1672         }
1673
1674         if (new_nsproxy)
1675                 put_nsproxy(new_nsproxy);
1676
1677 bad_unshare_cleanup_fd:
1678         if (new_fd)
1679                 put_files_struct(new_fd);
1680
1681 bad_unshare_cleanup_vm:
1682         if (new_mm)
1683                 mmput(new_mm);
1684
1685 bad_unshare_cleanup_sigh:
1686         if (new_sigh)
1687                 if (atomic_dec_and_test(&new_sigh->count))
1688                         kmem_cache_free(sighand_cachep, new_sigh);
1689
1690 bad_unshare_cleanup_fs:
1691         if (new_fs)
1692                 put_fs_struct(new_fs);
1693
1694 bad_unshare_cleanup_thread:
1695 bad_unshare_out:
1696         return err;
1697 }
1698
1699 /*
1700  *      Helper to unshare the files of the current task.
1701  *      We don't want to expose copy_files internals to
1702  *      the exec layer of the kernel.
1703  */
1704
1705 int unshare_files(struct files_struct **displaced)
1706 {
1707         struct task_struct *task = current;
1708         struct files_struct *copy = NULL;
1709         int error;
1710
1711         error = unshare_fd(CLONE_FILES, &copy);
1712         if (error || !copy) {
1713                 *displaced = NULL;
1714                 return error;
1715         }
1716         *displaced = task->files;
1717         task_lock(task);
1718         task->files = copy;
1719         task_unlock(task);
1720         return 0;
1721 }