hrtimer: remove duplicate helper function
[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/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/fs.h>
30 #include <linux/nsproxy.h>
31 #include <linux/capability.h>
32 #include <linux/cpu.h>
33 #include <linux/cgroup.h>
34 #include <linux/security.h>
35 #include <linux/swap.h>
36 #include <linux/syscalls.h>
37 #include <linux/jiffies.h>
38 #include <linux/futex.h>
39 #include <linux/task_io_accounting_ops.h>
40 #include <linux/rcupdate.h>
41 #include <linux/ptrace.h>
42 #include <linux/mount.h>
43 #include <linux/audit.h>
44 #include <linux/memcontrol.h>
45 #include <linux/profile.h>
46 #include <linux/rmap.h>
47 #include <linux/acct.h>
48 #include <linux/tsacct_kern.h>
49 #include <linux/cn_proc.h>
50 #include <linux/freezer.h>
51 #include <linux/delayacct.h>
52 #include <linux/taskstats_kern.h>
53 #include <linux/random.h>
54 #include <linux/tty.h>
55 #include <linux/proc_fs.h>
56 #include <linux/blkdev.h>
57
58 #include <asm/pgtable.h>
59 #include <asm/pgalloc.h>
60 #include <asm/uaccess.h>
61 #include <asm/mmu_context.h>
62 #include <asm/cacheflush.h>
63 #include <asm/tlbflush.h>
64
65 /*
66  * Protected counters by write_lock_irq(&tasklist_lock)
67  */
68 unsigned long total_forks;      /* Handle normal Linux uptimes. */
69 int nr_threads;                 /* The idle threads do not count.. */
70
71 int max_threads;                /* tunable limit on nr_threads */
72
73 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
74
75 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
76
77 int nr_processes(void)
78 {
79         int cpu;
80         int total = 0;
81
82         for_each_online_cpu(cpu)
83                 total += per_cpu(process_counts, cpu);
84
85         return total;
86 }
87
88 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
89 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
90 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
91 static struct kmem_cache *task_struct_cachep;
92 #endif
93
94 /* SLAB cache for signal_struct structures (tsk->signal) */
95 static struct kmem_cache *signal_cachep;
96
97 /* SLAB cache for sighand_struct structures (tsk->sighand) */
98 struct kmem_cache *sighand_cachep;
99
100 /* SLAB cache for files_struct structures (tsk->files) */
101 struct kmem_cache *files_cachep;
102
103 /* SLAB cache for fs_struct structures (tsk->fs) */
104 struct kmem_cache *fs_cachep;
105
106 /* SLAB cache for vm_area_struct structures */
107 struct kmem_cache *vm_area_cachep;
108
109 /* SLAB cache for mm_struct structures (tsk->mm) */
110 static struct kmem_cache *mm_cachep;
111
112 void free_task(struct task_struct *tsk)
113 {
114         prop_local_destroy_single(&tsk->dirties);
115         free_thread_info(tsk->stack);
116         rt_mutex_debug_task_free(tsk);
117         free_task_struct(tsk);
118 }
119 EXPORT_SYMBOL(free_task);
120
121 void __put_task_struct(struct task_struct *tsk)
122 {
123         WARN_ON(!tsk->exit_state);
124         WARN_ON(atomic_read(&tsk->usage));
125         WARN_ON(tsk == current);
126
127         security_task_free(tsk);
128         free_uid(tsk->user);
129         put_group_info(tsk->group_info);
130         delayacct_tsk_free(tsk);
131
132         if (!profile_handoff_task(tsk))
133                 free_task(tsk);
134 }
135
136 /*
137  * macro override instead of weak attribute alias, to workaround
138  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
139  */
140 #ifndef arch_task_cache_init
141 #define arch_task_cache_init()
142 #endif
143
144 void __init fork_init(unsigned long mempages)
145 {
146 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
147 #ifndef ARCH_MIN_TASKALIGN
148 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
149 #endif
150         /* create a slab on which task_structs can be allocated */
151         task_struct_cachep =
152                 kmem_cache_create("task_struct", sizeof(struct task_struct),
153                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
154 #endif
155
156         /* do the arch specific task caches init */
157         arch_task_cache_init();
158
159         /*
160          * The default maximum number of threads is set to a safe
161          * value: the thread structures can take up at most half
162          * of memory.
163          */
164         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
165
166         /*
167          * we need to allow at least 20 threads to boot a system
168          */
169         if(max_threads < 20)
170                 max_threads = 20;
171
172         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
173         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
174         init_task.signal->rlim[RLIMIT_SIGPENDING] =
175                 init_task.signal->rlim[RLIMIT_NPROC];
176 }
177
178 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
179                                                struct task_struct *src)
180 {
181         *dst = *src;
182         return 0;
183 }
184
185 static struct task_struct *dup_task_struct(struct task_struct *orig)
186 {
187         struct task_struct *tsk;
188         struct thread_info *ti;
189         int err;
190
191         prepare_to_copy(orig);
192
193         tsk = alloc_task_struct();
194         if (!tsk)
195                 return NULL;
196
197         ti = alloc_thread_info(tsk);
198         if (!ti) {
199                 free_task_struct(tsk);
200                 return NULL;
201         }
202
203         err = arch_dup_task_struct(tsk, orig);
204         if (err)
205                 goto out;
206
207         tsk->stack = ti;
208
209         err = prop_local_init_single(&tsk->dirties);
210         if (err)
211                 goto out;
212
213         setup_thread_stack(tsk, orig);
214
215 #ifdef CONFIG_CC_STACKPROTECTOR
216         tsk->stack_canary = get_random_int();
217 #endif
218
219         /* One for us, one for whoever does the "release_task()" (usually parent) */
220         atomic_set(&tsk->usage,2);
221         atomic_set(&tsk->fs_excl, 0);
222 #ifdef CONFIG_BLK_DEV_IO_TRACE
223         tsk->btrace_seq = 0;
224 #endif
225         tsk->splice_pipe = NULL;
226         return tsk;
227
228 out:
229         free_thread_info(ti);
230         free_task_struct(tsk);
231         return NULL;
232 }
233
234 #ifdef CONFIG_MMU
235 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
236 {
237         struct vm_area_struct *mpnt, *tmp, **pprev;
238         struct rb_node **rb_link, *rb_parent;
239         int retval;
240         unsigned long charge;
241         struct mempolicy *pol;
242
243         down_write(&oldmm->mmap_sem);
244         flush_cache_dup_mm(oldmm);
245         /*
246          * Not linked in yet - no deadlock potential:
247          */
248         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
249
250         mm->locked_vm = 0;
251         mm->mmap = NULL;
252         mm->mmap_cache = NULL;
253         mm->free_area_cache = oldmm->mmap_base;
254         mm->cached_hole_size = ~0UL;
255         mm->map_count = 0;
256         cpus_clear(mm->cpu_vm_mask);
257         mm->mm_rb = RB_ROOT;
258         rb_link = &mm->mm_rb.rb_node;
259         rb_parent = NULL;
260         pprev = &mm->mmap;
261
262         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
263                 struct file *file;
264
265                 if (mpnt->vm_flags & VM_DONTCOPY) {
266                         long pages = vma_pages(mpnt);
267                         mm->total_vm -= pages;
268                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
269                                                                 -pages);
270                         continue;
271                 }
272                 charge = 0;
273                 if (mpnt->vm_flags & VM_ACCOUNT) {
274                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
275                         if (security_vm_enough_memory(len))
276                                 goto fail_nomem;
277                         charge = len;
278                 }
279                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
280                 if (!tmp)
281                         goto fail_nomem;
282                 *tmp = *mpnt;
283                 pol = mpol_dup(vma_policy(mpnt));
284                 retval = PTR_ERR(pol);
285                 if (IS_ERR(pol))
286                         goto fail_nomem_policy;
287                 vma_set_policy(tmp, pol);
288                 tmp->vm_flags &= ~VM_LOCKED;
289                 tmp->vm_mm = mm;
290                 tmp->vm_next = NULL;
291                 anon_vma_link(tmp);
292                 file = tmp->vm_file;
293                 if (file) {
294                         struct inode *inode = file->f_path.dentry->d_inode;
295                         get_file(file);
296                         if (tmp->vm_flags & VM_DENYWRITE)
297                                 atomic_dec(&inode->i_writecount);
298
299                         /* insert tmp into the share list, just after mpnt */
300                         spin_lock(&file->f_mapping->i_mmap_lock);
301                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
302                         flush_dcache_mmap_lock(file->f_mapping);
303                         vma_prio_tree_add(tmp, mpnt);
304                         flush_dcache_mmap_unlock(file->f_mapping);
305                         spin_unlock(&file->f_mapping->i_mmap_lock);
306                 }
307
308                 /*
309                  * Link in the new vma and copy the page table entries.
310                  */
311                 *pprev = tmp;
312                 pprev = &tmp->vm_next;
313
314                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
315                 rb_link = &tmp->vm_rb.rb_right;
316                 rb_parent = &tmp->vm_rb;
317
318                 mm->map_count++;
319                 retval = copy_page_range(mm, oldmm, mpnt);
320
321                 if (tmp->vm_ops && tmp->vm_ops->open)
322                         tmp->vm_ops->open(tmp);
323
324                 if (retval)
325                         goto out;
326         }
327         /* a new mm has just been created */
328         arch_dup_mmap(oldmm, mm);
329         retval = 0;
330 out:
331         up_write(&mm->mmap_sem);
332         flush_tlb_mm(oldmm);
333         up_write(&oldmm->mmap_sem);
334         return retval;
335 fail_nomem_policy:
336         kmem_cache_free(vm_area_cachep, tmp);
337 fail_nomem:
338         retval = -ENOMEM;
339         vm_unacct_memory(charge);
340         goto out;
341 }
342
343 static inline int mm_alloc_pgd(struct mm_struct * mm)
344 {
345         mm->pgd = pgd_alloc(mm);
346         if (unlikely(!mm->pgd))
347                 return -ENOMEM;
348         return 0;
349 }
350
351 static inline void mm_free_pgd(struct mm_struct * mm)
352 {
353         pgd_free(mm, mm->pgd);
354 }
355 #else
356 #define dup_mmap(mm, oldmm)     (0)
357 #define mm_alloc_pgd(mm)        (0)
358 #define mm_free_pgd(mm)
359 #endif /* CONFIG_MMU */
360
361 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
362
363 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
364 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
365
366 #include <linux/init_task.h>
367
368 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
369 {
370         atomic_set(&mm->mm_users, 1);
371         atomic_set(&mm->mm_count, 1);
372         init_rwsem(&mm->mmap_sem);
373         INIT_LIST_HEAD(&mm->mmlist);
374         mm->flags = (current->mm) ? current->mm->flags
375                                   : MMF_DUMP_FILTER_DEFAULT;
376         mm->core_waiters = 0;
377         mm->nr_ptes = 0;
378         set_mm_counter(mm, file_rss, 0);
379         set_mm_counter(mm, anon_rss, 0);
380         spin_lock_init(&mm->page_table_lock);
381         rwlock_init(&mm->ioctx_list_lock);
382         mm->ioctx_list = NULL;
383         mm->free_area_cache = TASK_UNMAPPED_BASE;
384         mm->cached_hole_size = ~0UL;
385         mm_init_owner(mm, p);
386
387         if (likely(!mm_alloc_pgd(mm))) {
388                 mm->def_flags = 0;
389                 return mm;
390         }
391
392         free_mm(mm);
393         return NULL;
394 }
395
396 /*
397  * Allocate and initialize an mm_struct.
398  */
399 struct mm_struct * mm_alloc(void)
400 {
401         struct mm_struct * mm;
402
403         mm = allocate_mm();
404         if (mm) {
405                 memset(mm, 0, sizeof(*mm));
406                 mm = mm_init(mm, current);
407         }
408         return mm;
409 }
410
411 /*
412  * Called when the last reference to the mm
413  * is dropped: either by a lazy thread or by
414  * mmput. Free the page directory and the mm.
415  */
416 void __mmdrop(struct mm_struct *mm)
417 {
418         BUG_ON(mm == &init_mm);
419         mm_free_pgd(mm);
420         destroy_context(mm);
421         free_mm(mm);
422 }
423 EXPORT_SYMBOL_GPL(__mmdrop);
424
425 /*
426  * Decrement the use count and release all resources for an mm.
427  */
428 void mmput(struct mm_struct *mm)
429 {
430         might_sleep();
431
432         if (atomic_dec_and_test(&mm->mm_users)) {
433                 exit_aio(mm);
434                 exit_mmap(mm);
435                 set_mm_exe_file(mm, NULL);
436                 if (!list_empty(&mm->mmlist)) {
437                         spin_lock(&mmlist_lock);
438                         list_del(&mm->mmlist);
439                         spin_unlock(&mmlist_lock);
440                 }
441                 put_swap_token(mm);
442                 mmdrop(mm);
443         }
444 }
445 EXPORT_SYMBOL_GPL(mmput);
446
447 /**
448  * get_task_mm - acquire a reference to the task's mm
449  *
450  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
451  * this kernel workthread has transiently adopted a user mm with use_mm,
452  * to do its AIO) is not set and if so returns a reference to it, after
453  * bumping up the use count.  User must release the mm via mmput()
454  * after use.  Typically used by /proc and ptrace.
455  */
456 struct mm_struct *get_task_mm(struct task_struct *task)
457 {
458         struct mm_struct *mm;
459
460         task_lock(task);
461         mm = task->mm;
462         if (mm) {
463                 if (task->flags & PF_BORROWED_MM)
464                         mm = NULL;
465                 else
466                         atomic_inc(&mm->mm_users);
467         }
468         task_unlock(task);
469         return mm;
470 }
471 EXPORT_SYMBOL_GPL(get_task_mm);
472
473 /* Please note the differences between mmput and mm_release.
474  * mmput is called whenever we stop holding onto a mm_struct,
475  * error success whatever.
476  *
477  * mm_release is called after a mm_struct has been removed
478  * from the current process.
479  *
480  * This difference is important for error handling, when we
481  * only half set up a mm_struct for a new process and need to restore
482  * the old one.  Because we mmput the new mm_struct before
483  * restoring the old one. . .
484  * Eric Biederman 10 January 1998
485  */
486 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
487 {
488         struct completion *vfork_done = tsk->vfork_done;
489
490         /* Get rid of any cached register state */
491         deactivate_mm(tsk, mm);
492
493         /* notify parent sleeping on vfork() */
494         if (vfork_done) {
495                 tsk->vfork_done = NULL;
496                 complete(vfork_done);
497         }
498
499         /*
500          * If we're exiting normally, clear a user-space tid field if
501          * requested.  We leave this alone when dying by signal, to leave
502          * the value intact in a core dump, and to save the unnecessary
503          * trouble otherwise.  Userland only wants this done for a sys_exit.
504          */
505         if (tsk->clear_child_tid
506             && !(tsk->flags & PF_SIGNALED)
507             && atomic_read(&mm->mm_users) > 1) {
508                 u32 __user * tidptr = tsk->clear_child_tid;
509                 tsk->clear_child_tid = NULL;
510
511                 /*
512                  * We don't check the error code - if userspace has
513                  * not set up a proper pointer then tough luck.
514                  */
515                 put_user(0, tidptr);
516                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
517         }
518 }
519
520 /*
521  * Allocate a new mm structure and copy contents from the
522  * mm structure of the passed in task structure.
523  */
524 struct mm_struct *dup_mm(struct task_struct *tsk)
525 {
526         struct mm_struct *mm, *oldmm = current->mm;
527         int err;
528
529         if (!oldmm)
530                 return NULL;
531
532         mm = allocate_mm();
533         if (!mm)
534                 goto fail_nomem;
535
536         memcpy(mm, oldmm, sizeof(*mm));
537
538         /* Initializing for Swap token stuff */
539         mm->token_priority = 0;
540         mm->last_interval = 0;
541
542         if (!mm_init(mm, tsk))
543                 goto fail_nomem;
544
545         if (init_new_context(tsk, mm))
546                 goto fail_nocontext;
547
548         dup_mm_exe_file(oldmm, mm);
549
550         err = dup_mmap(mm, oldmm);
551         if (err)
552                 goto free_pt;
553
554         mm->hiwater_rss = get_mm_rss(mm);
555         mm->hiwater_vm = mm->total_vm;
556
557         return mm;
558
559 free_pt:
560         mmput(mm);
561
562 fail_nomem:
563         return NULL;
564
565 fail_nocontext:
566         /*
567          * If init_new_context() failed, we cannot use mmput() to free the mm
568          * because it calls destroy_context()
569          */
570         mm_free_pgd(mm);
571         free_mm(mm);
572         return NULL;
573 }
574
575 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
576 {
577         struct mm_struct * mm, *oldmm;
578         int retval;
579
580         tsk->min_flt = tsk->maj_flt = 0;
581         tsk->nvcsw = tsk->nivcsw = 0;
582
583         tsk->mm = NULL;
584         tsk->active_mm = NULL;
585
586         /*
587          * Are we cloning a kernel thread?
588          *
589          * We need to steal a active VM for that..
590          */
591         oldmm = current->mm;
592         if (!oldmm)
593                 return 0;
594
595         if (clone_flags & CLONE_VM) {
596                 atomic_inc(&oldmm->mm_users);
597                 mm = oldmm;
598                 goto good_mm;
599         }
600
601         retval = -ENOMEM;
602         mm = dup_mm(tsk);
603         if (!mm)
604                 goto fail_nomem;
605
606 good_mm:
607         /* Initializing for Swap token stuff */
608         mm->token_priority = 0;
609         mm->last_interval = 0;
610
611         tsk->mm = mm;
612         tsk->active_mm = mm;
613         return 0;
614
615 fail_nomem:
616         return retval;
617 }
618
619 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
620 {
621         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
622         /* We don't need to lock fs - think why ;-) */
623         if (fs) {
624                 atomic_set(&fs->count, 1);
625                 rwlock_init(&fs->lock);
626                 fs->umask = old->umask;
627                 read_lock(&old->lock);
628                 fs->root = old->root;
629                 path_get(&old->root);
630                 fs->pwd = old->pwd;
631                 path_get(&old->pwd);
632                 if (old->altroot.dentry) {
633                         fs->altroot = old->altroot;
634                         path_get(&old->altroot);
635                 } else {
636                         fs->altroot.mnt = NULL;
637                         fs->altroot.dentry = NULL;
638                 }
639                 read_unlock(&old->lock);
640         }
641         return fs;
642 }
643
644 struct fs_struct *copy_fs_struct(struct fs_struct *old)
645 {
646         return __copy_fs_struct(old);
647 }
648
649 EXPORT_SYMBOL_GPL(copy_fs_struct);
650
651 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
652 {
653         if (clone_flags & CLONE_FS) {
654                 atomic_inc(&current->fs->count);
655                 return 0;
656         }
657         tsk->fs = __copy_fs_struct(current->fs);
658         if (!tsk->fs)
659                 return -ENOMEM;
660         return 0;
661 }
662
663 static int count_open_files(struct fdtable *fdt)
664 {
665         int size = fdt->max_fds;
666         int i;
667
668         /* Find the last open fd */
669         for (i = size/(8*sizeof(long)); i > 0; ) {
670                 if (fdt->open_fds->fds_bits[--i])
671                         break;
672         }
673         i = (i+1) * 8 * sizeof(long);
674         return i;
675 }
676
677 static struct files_struct *alloc_files(void)
678 {
679         struct files_struct *newf;
680         struct fdtable *fdt;
681
682         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
683         if (!newf)
684                 goto out;
685
686         atomic_set(&newf->count, 1);
687
688         spin_lock_init(&newf->file_lock);
689         newf->next_fd = 0;
690         fdt = &newf->fdtab;
691         fdt->max_fds = NR_OPEN_DEFAULT;
692         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
693         fdt->open_fds = (fd_set *)&newf->open_fds_init;
694         fdt->fd = &newf->fd_array[0];
695         INIT_RCU_HEAD(&fdt->rcu);
696         fdt->next = NULL;
697         rcu_assign_pointer(newf->fdt, fdt);
698 out:
699         return newf;
700 }
701
702 /*
703  * Allocate a new files structure and copy contents from the
704  * passed in files structure.
705  * errorp will be valid only when the returned files_struct is NULL.
706  */
707 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
708 {
709         struct files_struct *newf;
710         struct file **old_fds, **new_fds;
711         int open_files, size, i;
712         struct fdtable *old_fdt, *new_fdt;
713
714         *errorp = -ENOMEM;
715         newf = alloc_files();
716         if (!newf)
717                 goto out;
718
719         spin_lock(&oldf->file_lock);
720         old_fdt = files_fdtable(oldf);
721         new_fdt = files_fdtable(newf);
722         open_files = count_open_files(old_fdt);
723
724         /*
725          * Check whether we need to allocate a larger fd array and fd set.
726          * Note: we're not a clone task, so the open count won't change.
727          */
728         if (open_files > new_fdt->max_fds) {
729                 new_fdt->max_fds = 0;
730                 spin_unlock(&oldf->file_lock);
731                 spin_lock(&newf->file_lock);
732                 *errorp = expand_files(newf, open_files-1);
733                 spin_unlock(&newf->file_lock);
734                 if (*errorp < 0)
735                         goto out_release;
736                 new_fdt = files_fdtable(newf);
737                 /*
738                  * Reacquire the oldf lock and a pointer to its fd table
739                  * who knows it may have a new bigger fd table. We need
740                  * the latest pointer.
741                  */
742                 spin_lock(&oldf->file_lock);
743                 old_fdt = files_fdtable(oldf);
744         }
745
746         old_fds = old_fdt->fd;
747         new_fds = new_fdt->fd;
748
749         memcpy(new_fdt->open_fds->fds_bits,
750                 old_fdt->open_fds->fds_bits, open_files/8);
751         memcpy(new_fdt->close_on_exec->fds_bits,
752                 old_fdt->close_on_exec->fds_bits, open_files/8);
753
754         for (i = open_files; i != 0; i--) {
755                 struct file *f = *old_fds++;
756                 if (f) {
757                         get_file(f);
758                 } else {
759                         /*
760                          * The fd may be claimed in the fd bitmap but not yet
761                          * instantiated in the files array if a sibling thread
762                          * is partway through open().  So make sure that this
763                          * fd is available to the new process.
764                          */
765                         FD_CLR(open_files - i, new_fdt->open_fds);
766                 }
767                 rcu_assign_pointer(*new_fds++, f);
768         }
769         spin_unlock(&oldf->file_lock);
770
771         /* compute the remainder to be cleared */
772         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
773
774         /* This is long word aligned thus could use a optimized version */
775         memset(new_fds, 0, size);
776
777         if (new_fdt->max_fds > open_files) {
778                 int left = (new_fdt->max_fds-open_files)/8;
779                 int start = open_files / (8 * sizeof(unsigned long));
780
781                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
782                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
783         }
784
785         return newf;
786
787 out_release:
788         kmem_cache_free(files_cachep, newf);
789 out:
790         return NULL;
791 }
792
793 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
794 {
795         struct files_struct *oldf, *newf;
796         int error = 0;
797
798         /*
799          * A background process may not have any files ...
800          */
801         oldf = current->files;
802         if (!oldf)
803                 goto out;
804
805         if (clone_flags & CLONE_FILES) {
806                 atomic_inc(&oldf->count);
807                 goto out;
808         }
809
810         newf = dup_fd(oldf, &error);
811         if (!newf)
812                 goto out;
813
814         tsk->files = newf;
815         error = 0;
816 out:
817         return error;
818 }
819
820 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
821 {
822 #ifdef CONFIG_BLOCK
823         struct io_context *ioc = current->io_context;
824
825         if (!ioc)
826                 return 0;
827         /*
828          * Share io context with parent, if CLONE_IO is set
829          */
830         if (clone_flags & CLONE_IO) {
831                 tsk->io_context = ioc_task_link(ioc);
832                 if (unlikely(!tsk->io_context))
833                         return -ENOMEM;
834         } else if (ioprio_valid(ioc->ioprio)) {
835                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
836                 if (unlikely(!tsk->io_context))
837                         return -ENOMEM;
838
839                 tsk->io_context->ioprio = ioc->ioprio;
840         }
841 #endif
842         return 0;
843 }
844
845 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
846 {
847         struct sighand_struct *sig;
848
849         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
850                 atomic_inc(&current->sighand->count);
851                 return 0;
852         }
853         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
854         rcu_assign_pointer(tsk->sighand, sig);
855         if (!sig)
856                 return -ENOMEM;
857         atomic_set(&sig->count, 1);
858         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
859         return 0;
860 }
861
862 void __cleanup_sighand(struct sighand_struct *sighand)
863 {
864         if (atomic_dec_and_test(&sighand->count))
865                 kmem_cache_free(sighand_cachep, sighand);
866 }
867
868 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
869 {
870         struct signal_struct *sig;
871         int ret;
872
873         if (clone_flags & CLONE_THREAD) {
874                 atomic_inc(&current->signal->count);
875                 atomic_inc(&current->signal->live);
876                 return 0;
877         }
878         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
879         tsk->signal = sig;
880         if (!sig)
881                 return -ENOMEM;
882
883         ret = copy_thread_group_keys(tsk);
884         if (ret < 0) {
885                 kmem_cache_free(signal_cachep, sig);
886                 return ret;
887         }
888
889         atomic_set(&sig->count, 1);
890         atomic_set(&sig->live, 1);
891         init_waitqueue_head(&sig->wait_chldexit);
892         sig->flags = 0;
893         sig->group_exit_code = 0;
894         sig->group_exit_task = NULL;
895         sig->group_stop_count = 0;
896         sig->curr_target = tsk;
897         init_sigpending(&sig->shared_pending);
898         INIT_LIST_HEAD(&sig->posix_timers);
899
900         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
901         sig->it_real_incr.tv64 = 0;
902         sig->real_timer.function = it_real_fn;
903
904         sig->it_virt_expires = cputime_zero;
905         sig->it_virt_incr = cputime_zero;
906         sig->it_prof_expires = cputime_zero;
907         sig->it_prof_incr = cputime_zero;
908
909         sig->leader = 0;        /* session leadership doesn't inherit */
910         sig->tty_old_pgrp = NULL;
911
912         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
913         sig->gtime = cputime_zero;
914         sig->cgtime = cputime_zero;
915         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
916         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
917         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
918         sig->sum_sched_runtime = 0;
919         INIT_LIST_HEAD(&sig->cpu_timers[0]);
920         INIT_LIST_HEAD(&sig->cpu_timers[1]);
921         INIT_LIST_HEAD(&sig->cpu_timers[2]);
922         taskstats_tgid_init(sig);
923
924         task_lock(current->group_leader);
925         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
926         task_unlock(current->group_leader);
927
928         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
929                 /*
930                  * New sole thread in the process gets an expiry time
931                  * of the whole CPU time limit.
932                  */
933                 tsk->it_prof_expires =
934                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
935         }
936         acct_init_pacct(&sig->pacct);
937
938         tty_audit_fork(sig);
939
940         return 0;
941 }
942
943 void __cleanup_signal(struct signal_struct *sig)
944 {
945         exit_thread_group_keys(sig);
946         kmem_cache_free(signal_cachep, sig);
947 }
948
949 static void cleanup_signal(struct task_struct *tsk)
950 {
951         struct signal_struct *sig = tsk->signal;
952
953         atomic_dec(&sig->live);
954
955         if (atomic_dec_and_test(&sig->count))
956                 __cleanup_signal(sig);
957 }
958
959 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
960 {
961         unsigned long new_flags = p->flags;
962
963         new_flags &= ~PF_SUPERPRIV;
964         new_flags |= PF_FORKNOEXEC;
965         if (!(clone_flags & CLONE_PTRACE))
966                 p->ptrace = 0;
967         p->flags = new_flags;
968         clear_freeze_flag(p);
969 }
970
971 asmlinkage long sys_set_tid_address(int __user *tidptr)
972 {
973         current->clear_child_tid = tidptr;
974
975         return task_pid_vnr(current);
976 }
977
978 static void rt_mutex_init_task(struct task_struct *p)
979 {
980         spin_lock_init(&p->pi_lock);
981 #ifdef CONFIG_RT_MUTEXES
982         plist_head_init(&p->pi_waiters, &p->pi_lock);
983         p->pi_blocked_on = NULL;
984 #endif
985 }
986
987 #ifdef CONFIG_MM_OWNER
988 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
989 {
990         mm->owner = p;
991 }
992 #endif /* CONFIG_MM_OWNER */
993
994 /*
995  * This creates a new process as a copy of the old one,
996  * but does not actually start it yet.
997  *
998  * It copies the registers, and all the appropriate
999  * parts of the process environment (as per the clone
1000  * flags). The actual kick-off is left to the caller.
1001  */
1002 static struct task_struct *copy_process(unsigned long clone_flags,
1003                                         unsigned long stack_start,
1004                                         struct pt_regs *regs,
1005                                         unsigned long stack_size,
1006                                         int __user *child_tidptr,
1007                                         struct pid *pid)
1008 {
1009         int retval;
1010         struct task_struct *p;
1011         int cgroup_callbacks_done = 0;
1012
1013         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1014                 return ERR_PTR(-EINVAL);
1015
1016         /*
1017          * Thread groups must share signals as well, and detached threads
1018          * can only be started up within the thread group.
1019          */
1020         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1021                 return ERR_PTR(-EINVAL);
1022
1023         /*
1024          * Shared signal handlers imply shared VM. By way of the above,
1025          * thread groups also imply shared VM. Blocking this case allows
1026          * for various simplifications in other code.
1027          */
1028         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1029                 return ERR_PTR(-EINVAL);
1030
1031         retval = security_task_create(clone_flags);
1032         if (retval)
1033                 goto fork_out;
1034
1035         retval = -ENOMEM;
1036         p = dup_task_struct(current);
1037         if (!p)
1038                 goto fork_out;
1039
1040         rt_mutex_init_task(p);
1041
1042 #ifdef CONFIG_TRACE_IRQFLAGS
1043         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1044         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1045 #endif
1046         retval = -EAGAIN;
1047         if (atomic_read(&p->user->processes) >=
1048                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1049                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1050                     p->user != current->nsproxy->user_ns->root_user)
1051                         goto bad_fork_free;
1052         }
1053
1054         atomic_inc(&p->user->__count);
1055         atomic_inc(&p->user->processes);
1056         get_group_info(p->group_info);
1057
1058         /*
1059          * If multiple threads are within copy_process(), then this check
1060          * triggers too late. This doesn't hurt, the check is only there
1061          * to stop root fork bombs.
1062          */
1063         if (nr_threads >= max_threads)
1064                 goto bad_fork_cleanup_count;
1065
1066         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1067                 goto bad_fork_cleanup_count;
1068
1069         if (p->binfmt && !try_module_get(p->binfmt->module))
1070                 goto bad_fork_cleanup_put_domain;
1071
1072         p->did_exec = 0;
1073         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1074         copy_flags(clone_flags, p);
1075         INIT_LIST_HEAD(&p->children);
1076         INIT_LIST_HEAD(&p->sibling);
1077 #ifdef CONFIG_PREEMPT_RCU
1078         p->rcu_read_lock_nesting = 0;
1079         p->rcu_flipctr_idx = 0;
1080 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1081         p->vfork_done = NULL;
1082         spin_lock_init(&p->alloc_lock);
1083
1084         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1085         init_sigpending(&p->pending);
1086
1087         p->utime = cputime_zero;
1088         p->stime = cputime_zero;
1089         p->gtime = cputime_zero;
1090         p->utimescaled = cputime_zero;
1091         p->stimescaled = cputime_zero;
1092         p->prev_utime = cputime_zero;
1093         p->prev_stime = cputime_zero;
1094
1095 #ifdef CONFIG_DETECT_SOFTLOCKUP
1096         p->last_switch_count = 0;
1097         p->last_switch_timestamp = 0;
1098 #endif
1099
1100 #ifdef CONFIG_TASK_XACCT
1101         p->rchar = 0;           /* I/O counter: bytes read */
1102         p->wchar = 0;           /* I/O counter: bytes written */
1103         p->syscr = 0;           /* I/O counter: read syscalls */
1104         p->syscw = 0;           /* I/O counter: write syscalls */
1105 #endif
1106         task_io_accounting_init(p);
1107         acct_clear_integrals(p);
1108
1109         p->it_virt_expires = cputime_zero;
1110         p->it_prof_expires = cputime_zero;
1111         p->it_sched_expires = 0;
1112         INIT_LIST_HEAD(&p->cpu_timers[0]);
1113         INIT_LIST_HEAD(&p->cpu_timers[1]);
1114         INIT_LIST_HEAD(&p->cpu_timers[2]);
1115
1116         p->lock_depth = -1;             /* -1 = no lock */
1117         do_posix_clock_monotonic_gettime(&p->start_time);
1118         p->real_start_time = p->start_time;
1119         monotonic_to_bootbased(&p->real_start_time);
1120 #ifdef CONFIG_SECURITY
1121         p->security = NULL;
1122 #endif
1123         p->cap_bset = current->cap_bset;
1124         p->io_context = NULL;
1125         p->audit_context = NULL;
1126         cgroup_fork(p);
1127 #ifdef CONFIG_NUMA
1128         p->mempolicy = mpol_dup(p->mempolicy);
1129         if (IS_ERR(p->mempolicy)) {
1130                 retval = PTR_ERR(p->mempolicy);
1131                 p->mempolicy = NULL;
1132                 goto bad_fork_cleanup_cgroup;
1133         }
1134         mpol_fix_fork_child_flag(p);
1135 #endif
1136 #ifdef CONFIG_TRACE_IRQFLAGS
1137         p->irq_events = 0;
1138 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1139         p->hardirqs_enabled = 1;
1140 #else
1141         p->hardirqs_enabled = 0;
1142 #endif
1143         p->hardirq_enable_ip = 0;
1144         p->hardirq_enable_event = 0;
1145         p->hardirq_disable_ip = _THIS_IP_;
1146         p->hardirq_disable_event = 0;
1147         p->softirqs_enabled = 1;
1148         p->softirq_enable_ip = _THIS_IP_;
1149         p->softirq_enable_event = 0;
1150         p->softirq_disable_ip = 0;
1151         p->softirq_disable_event = 0;
1152         p->hardirq_context = 0;
1153         p->softirq_context = 0;
1154 #endif
1155 #ifdef CONFIG_LOCKDEP
1156         p->lockdep_depth = 0; /* no locks held yet */
1157         p->curr_chain_key = 0;
1158         p->lockdep_recursion = 0;
1159 #endif
1160
1161 #ifdef CONFIG_DEBUG_MUTEXES
1162         p->blocked_on = NULL; /* not blocked yet */
1163 #endif
1164
1165         /* Perform scheduler related setup. Assign this task to a CPU. */
1166         sched_fork(p, clone_flags);
1167
1168         if ((retval = security_task_alloc(p)))
1169                 goto bad_fork_cleanup_policy;
1170         if ((retval = audit_alloc(p)))
1171                 goto bad_fork_cleanup_security;
1172         /* copy all the process information */
1173         if ((retval = copy_semundo(clone_flags, p)))
1174                 goto bad_fork_cleanup_audit;
1175         if ((retval = copy_files(clone_flags, p)))
1176                 goto bad_fork_cleanup_semundo;
1177         if ((retval = copy_fs(clone_flags, p)))
1178                 goto bad_fork_cleanup_files;
1179         if ((retval = copy_sighand(clone_flags, p)))
1180                 goto bad_fork_cleanup_fs;
1181         if ((retval = copy_signal(clone_flags, p)))
1182                 goto bad_fork_cleanup_sighand;
1183         if ((retval = copy_mm(clone_flags, p)))
1184                 goto bad_fork_cleanup_signal;
1185         if ((retval = copy_keys(clone_flags, p)))
1186                 goto bad_fork_cleanup_mm;
1187         if ((retval = copy_namespaces(clone_flags, p)))
1188                 goto bad_fork_cleanup_keys;
1189         if ((retval = copy_io(clone_flags, p)))
1190                 goto bad_fork_cleanup_namespaces;
1191         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1192         if (retval)
1193                 goto bad_fork_cleanup_io;
1194
1195         if (pid != &init_struct_pid) {
1196                 retval = -ENOMEM;
1197                 pid = alloc_pid(task_active_pid_ns(p));
1198                 if (!pid)
1199                         goto bad_fork_cleanup_io;
1200
1201                 if (clone_flags & CLONE_NEWPID) {
1202                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1203                         if (retval < 0)
1204                                 goto bad_fork_free_pid;
1205                 }
1206         }
1207
1208         p->pid = pid_nr(pid);
1209         p->tgid = p->pid;
1210         if (clone_flags & CLONE_THREAD)
1211                 p->tgid = current->tgid;
1212
1213         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1214         /*
1215          * Clear TID on mm_release()?
1216          */
1217         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1218 #ifdef CONFIG_FUTEX
1219         p->robust_list = NULL;
1220 #ifdef CONFIG_COMPAT
1221         p->compat_robust_list = NULL;
1222 #endif
1223         INIT_LIST_HEAD(&p->pi_state_list);
1224         p->pi_state_cache = NULL;
1225 #endif
1226         /*
1227          * sigaltstack should be cleared when sharing the same VM
1228          */
1229         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1230                 p->sas_ss_sp = p->sas_ss_size = 0;
1231
1232         /*
1233          * Syscall tracing should be turned off in the child regardless
1234          * of CLONE_PTRACE.
1235          */
1236         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1237 #ifdef TIF_SYSCALL_EMU
1238         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1239 #endif
1240         clear_all_latency_tracing(p);
1241
1242         /* Our parent execution domain becomes current domain
1243            These must match for thread signalling to apply */
1244         p->parent_exec_id = p->self_exec_id;
1245
1246         /* ok, now we should be set up.. */
1247         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1248         p->pdeath_signal = 0;
1249         p->exit_state = 0;
1250
1251         /*
1252          * Ok, make it visible to the rest of the system.
1253          * We dont wake it up yet.
1254          */
1255         p->group_leader = p;
1256         INIT_LIST_HEAD(&p->thread_group);
1257         INIT_LIST_HEAD(&p->ptrace_children);
1258         INIT_LIST_HEAD(&p->ptrace_list);
1259
1260         /* Now that the task is set up, run cgroup callbacks if
1261          * necessary. We need to run them before the task is visible
1262          * on the tasklist. */
1263         cgroup_fork_callbacks(p);
1264         cgroup_callbacks_done = 1;
1265
1266         /* Need tasklist lock for parent etc handling! */
1267         write_lock_irq(&tasklist_lock);
1268
1269         /*
1270          * The task hasn't been attached yet, so its cpus_allowed mask will
1271          * not be changed, nor will its assigned CPU.
1272          *
1273          * The cpus_allowed mask of the parent may have changed after it was
1274          * copied first time - so re-copy it here, then check the child's CPU
1275          * to ensure it is on a valid CPU (and if not, just force it back to
1276          * parent's CPU). This avoids alot of nasty races.
1277          */
1278         p->cpus_allowed = current->cpus_allowed;
1279         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1280         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1281                         !cpu_online(task_cpu(p))))
1282                 set_task_cpu(p, smp_processor_id());
1283
1284         /* CLONE_PARENT re-uses the old parent */
1285         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1286                 p->real_parent = current->real_parent;
1287         else
1288                 p->real_parent = current;
1289         p->parent = p->real_parent;
1290
1291         spin_lock(&current->sighand->siglock);
1292
1293         /*
1294          * Process group and session signals need to be delivered to just the
1295          * parent before the fork or both the parent and the child after the
1296          * fork. Restart if a signal comes in before we add the new process to
1297          * it's process group.
1298          * A fatal signal pending means that current will exit, so the new
1299          * thread can't slip out of an OOM kill (or normal SIGKILL).
1300          */
1301         recalc_sigpending();
1302         if (signal_pending(current)) {
1303                 spin_unlock(&current->sighand->siglock);
1304                 write_unlock_irq(&tasklist_lock);
1305                 retval = -ERESTARTNOINTR;
1306                 goto bad_fork_free_pid;
1307         }
1308
1309         if (clone_flags & CLONE_THREAD) {
1310                 p->group_leader = current->group_leader;
1311                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1312
1313                 if (!cputime_eq(current->signal->it_virt_expires,
1314                                 cputime_zero) ||
1315                     !cputime_eq(current->signal->it_prof_expires,
1316                                 cputime_zero) ||
1317                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1318                     !list_empty(&current->signal->cpu_timers[0]) ||
1319                     !list_empty(&current->signal->cpu_timers[1]) ||
1320                     !list_empty(&current->signal->cpu_timers[2])) {
1321                         /*
1322                          * Have child wake up on its first tick to check
1323                          * for process CPU timers.
1324                          */
1325                         p->it_prof_expires = jiffies_to_cputime(1);
1326                 }
1327         }
1328
1329         if (likely(p->pid)) {
1330                 add_parent(p);
1331                 if (unlikely(p->ptrace & PT_PTRACED))
1332                         __ptrace_link(p, current->parent);
1333
1334                 if (thread_group_leader(p)) {
1335                         if (clone_flags & CLONE_NEWPID)
1336                                 p->nsproxy->pid_ns->child_reaper = p;
1337
1338                         p->signal->leader_pid = pid;
1339                         p->signal->tty = current->signal->tty;
1340                         set_task_pgrp(p, task_pgrp_nr(current));
1341                         set_task_session(p, task_session_nr(current));
1342                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1343                         attach_pid(p, PIDTYPE_SID, task_session(current));
1344                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1345                         __get_cpu_var(process_counts)++;
1346                 }
1347                 attach_pid(p, PIDTYPE_PID, pid);
1348                 nr_threads++;
1349         }
1350
1351         total_forks++;
1352         spin_unlock(&current->sighand->siglock);
1353         write_unlock_irq(&tasklist_lock);
1354         proc_fork_connector(p);
1355         cgroup_post_fork(p);
1356         return p;
1357
1358 bad_fork_free_pid:
1359         if (pid != &init_struct_pid)
1360                 free_pid(pid);
1361 bad_fork_cleanup_io:
1362         put_io_context(p->io_context);
1363 bad_fork_cleanup_namespaces:
1364         exit_task_namespaces(p);
1365 bad_fork_cleanup_keys:
1366         exit_keys(p);
1367 bad_fork_cleanup_mm:
1368         if (p->mm)
1369                 mmput(p->mm);
1370 bad_fork_cleanup_signal:
1371         cleanup_signal(p);
1372 bad_fork_cleanup_sighand:
1373         __cleanup_sighand(p->sighand);
1374 bad_fork_cleanup_fs:
1375         exit_fs(p); /* blocking */
1376 bad_fork_cleanup_files:
1377         exit_files(p); /* blocking */
1378 bad_fork_cleanup_semundo:
1379         exit_sem(p);
1380 bad_fork_cleanup_audit:
1381         audit_free(p);
1382 bad_fork_cleanup_security:
1383         security_task_free(p);
1384 bad_fork_cleanup_policy:
1385 #ifdef CONFIG_NUMA
1386         mpol_put(p->mempolicy);
1387 bad_fork_cleanup_cgroup:
1388 #endif
1389         cgroup_exit(p, cgroup_callbacks_done);
1390         delayacct_tsk_free(p);
1391         if (p->binfmt)
1392                 module_put(p->binfmt->module);
1393 bad_fork_cleanup_put_domain:
1394         module_put(task_thread_info(p)->exec_domain->module);
1395 bad_fork_cleanup_count:
1396         put_group_info(p->group_info);
1397         atomic_dec(&p->user->processes);
1398         free_uid(p->user);
1399 bad_fork_free:
1400         free_task(p);
1401 fork_out:
1402         return ERR_PTR(retval);
1403 }
1404
1405 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1406 {
1407         memset(regs, 0, sizeof(struct pt_regs));
1408         return regs;
1409 }
1410
1411 struct task_struct * __cpuinit fork_idle(int cpu)
1412 {
1413         struct task_struct *task;
1414         struct pt_regs regs;
1415
1416         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1417                                 &init_struct_pid);
1418         if (!IS_ERR(task))
1419                 init_idle(task, cpu);
1420
1421         return task;
1422 }
1423
1424 static int fork_traceflag(unsigned clone_flags)
1425 {
1426         if (clone_flags & CLONE_UNTRACED)
1427                 return 0;
1428         else if (clone_flags & CLONE_VFORK) {
1429                 if (current->ptrace & PT_TRACE_VFORK)
1430                         return PTRACE_EVENT_VFORK;
1431         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1432                 if (current->ptrace & PT_TRACE_CLONE)
1433                         return PTRACE_EVENT_CLONE;
1434         } else if (current->ptrace & PT_TRACE_FORK)
1435                 return PTRACE_EVENT_FORK;
1436
1437         return 0;
1438 }
1439
1440 /*
1441  *  Ok, this is the main fork-routine.
1442  *
1443  * It copies the process, and if successful kick-starts
1444  * it and waits for it to finish using the VM if required.
1445  */
1446 long do_fork(unsigned long clone_flags,
1447               unsigned long stack_start,
1448               struct pt_regs *regs,
1449               unsigned long stack_size,
1450               int __user *parent_tidptr,
1451               int __user *child_tidptr)
1452 {
1453         struct task_struct *p;
1454         int trace = 0;
1455         long nr;
1456
1457         /*
1458          * We hope to recycle these flags after 2.6.26
1459          */
1460         if (unlikely(clone_flags & CLONE_STOPPED)) {
1461                 static int __read_mostly count = 100;
1462
1463                 if (count > 0 && printk_ratelimit()) {
1464                         char comm[TASK_COMM_LEN];
1465
1466                         count--;
1467                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1468                                         "clone flags 0x%lx\n",
1469                                 get_task_comm(comm, current),
1470                                 clone_flags & CLONE_STOPPED);
1471                 }
1472         }
1473
1474         if (unlikely(current->ptrace)) {
1475                 trace = fork_traceflag (clone_flags);
1476                 if (trace)
1477                         clone_flags |= CLONE_PTRACE;
1478         }
1479
1480         p = copy_process(clone_flags, stack_start, regs, stack_size,
1481                         child_tidptr, NULL);
1482         /*
1483          * Do this prior waking up the new thread - the thread pointer
1484          * might get invalid after that point, if the thread exits quickly.
1485          */
1486         if (!IS_ERR(p)) {
1487                 struct completion vfork;
1488
1489                 nr = task_pid_vnr(p);
1490
1491                 if (clone_flags & CLONE_PARENT_SETTID)
1492                         put_user(nr, parent_tidptr);
1493
1494                 if (clone_flags & CLONE_VFORK) {
1495                         p->vfork_done = &vfork;
1496                         init_completion(&vfork);
1497                 }
1498
1499                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1500                         /*
1501                          * We'll start up with an immediate SIGSTOP.
1502                          */
1503                         sigaddset(&p->pending.signal, SIGSTOP);
1504                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1505                 }
1506
1507                 if (!(clone_flags & CLONE_STOPPED))
1508                         wake_up_new_task(p, clone_flags);
1509                 else
1510                         __set_task_state(p, TASK_STOPPED);
1511
1512                 if (unlikely (trace)) {
1513                         current->ptrace_message = nr;
1514                         ptrace_notify ((trace << 8) | SIGTRAP);
1515                 }
1516
1517                 if (clone_flags & CLONE_VFORK) {
1518                         freezer_do_not_count();
1519                         wait_for_completion(&vfork);
1520                         freezer_count();
1521                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1522                                 current->ptrace_message = nr;
1523                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1524                         }
1525                 }
1526         } else {
1527                 nr = PTR_ERR(p);
1528         }
1529         return nr;
1530 }
1531
1532 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1533 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1534 #endif
1535
1536 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1537 {
1538         struct sighand_struct *sighand = data;
1539
1540         spin_lock_init(&sighand->siglock);
1541         init_waitqueue_head(&sighand->signalfd_wqh);
1542 }
1543
1544 void __init proc_caches_init(void)
1545 {
1546         sighand_cachep = kmem_cache_create("sighand_cache",
1547                         sizeof(struct sighand_struct), 0,
1548                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1549                         sighand_ctor);
1550         signal_cachep = kmem_cache_create("signal_cache",
1551                         sizeof(struct signal_struct), 0,
1552                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1553         files_cachep = kmem_cache_create("files_cache",
1554                         sizeof(struct files_struct), 0,
1555                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1556         fs_cachep = kmem_cache_create("fs_cache",
1557                         sizeof(struct fs_struct), 0,
1558                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1559         vm_area_cachep = kmem_cache_create("vm_area_struct",
1560                         sizeof(struct vm_area_struct), 0,
1561                         SLAB_PANIC, NULL);
1562         mm_cachep = kmem_cache_create("mm_struct",
1563                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1564                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1565 }
1566
1567 /*
1568  * Check constraints on flags passed to the unshare system call and
1569  * force unsharing of additional process context as appropriate.
1570  */
1571 static void check_unshare_flags(unsigned long *flags_ptr)
1572 {
1573         /*
1574          * If unsharing a thread from a thread group, must also
1575          * unshare vm.
1576          */
1577         if (*flags_ptr & CLONE_THREAD)
1578                 *flags_ptr |= CLONE_VM;
1579
1580         /*
1581          * If unsharing vm, must also unshare signal handlers.
1582          */
1583         if (*flags_ptr & CLONE_VM)
1584                 *flags_ptr |= CLONE_SIGHAND;
1585
1586         /*
1587          * If unsharing signal handlers and the task was created
1588          * using CLONE_THREAD, then must unshare the thread
1589          */
1590         if ((*flags_ptr & CLONE_SIGHAND) &&
1591             (atomic_read(&current->signal->count) > 1))
1592                 *flags_ptr |= CLONE_THREAD;
1593
1594         /*
1595          * If unsharing namespace, must also unshare filesystem information.
1596          */
1597         if (*flags_ptr & CLONE_NEWNS)
1598                 *flags_ptr |= CLONE_FS;
1599 }
1600
1601 /*
1602  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1603  */
1604 static int unshare_thread(unsigned long unshare_flags)
1605 {
1606         if (unshare_flags & CLONE_THREAD)
1607                 return -EINVAL;
1608
1609         return 0;
1610 }
1611
1612 /*
1613  * Unshare the filesystem structure if it is being shared
1614  */
1615 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1616 {
1617         struct fs_struct *fs = current->fs;
1618
1619         if ((unshare_flags & CLONE_FS) &&
1620             (fs && atomic_read(&fs->count) > 1)) {
1621                 *new_fsp = __copy_fs_struct(current->fs);
1622                 if (!*new_fsp)
1623                         return -ENOMEM;
1624         }
1625
1626         return 0;
1627 }
1628
1629 /*
1630  * Unsharing of sighand is not supported yet
1631  */
1632 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1633 {
1634         struct sighand_struct *sigh = current->sighand;
1635
1636         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1637                 return -EINVAL;
1638         else
1639                 return 0;
1640 }
1641
1642 /*
1643  * Unshare vm if it is being shared
1644  */
1645 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1646 {
1647         struct mm_struct *mm = current->mm;
1648
1649         if ((unshare_flags & CLONE_VM) &&
1650             (mm && atomic_read(&mm->mm_users) > 1)) {
1651                 return -EINVAL;
1652         }
1653
1654         return 0;
1655 }
1656
1657 /*
1658  * Unshare file descriptor table if it is being shared
1659  */
1660 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1661 {
1662         struct files_struct *fd = current->files;
1663         int error = 0;
1664
1665         if ((unshare_flags & CLONE_FILES) &&
1666             (fd && atomic_read(&fd->count) > 1)) {
1667                 *new_fdp = dup_fd(fd, &error);
1668                 if (!*new_fdp)
1669                         return error;
1670         }
1671
1672         return 0;
1673 }
1674
1675 /*
1676  * unshare allows a process to 'unshare' part of the process
1677  * context which was originally shared using clone.  copy_*
1678  * functions used by do_fork() cannot be used here directly
1679  * because they modify an inactive task_struct that is being
1680  * constructed. Here we are modifying the current, active,
1681  * task_struct.
1682  */
1683 asmlinkage long sys_unshare(unsigned long unshare_flags)
1684 {
1685         int err = 0;
1686         struct fs_struct *fs, *new_fs = NULL;
1687         struct sighand_struct *new_sigh = NULL;
1688         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1689         struct files_struct *fd, *new_fd = NULL;
1690         struct nsproxy *new_nsproxy = NULL;
1691         int do_sysvsem = 0;
1692
1693         check_unshare_flags(&unshare_flags);
1694
1695         /* Return -EINVAL for all unsupported flags */
1696         err = -EINVAL;
1697         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1698                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1699                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1700                                 CLONE_NEWNET))
1701                 goto bad_unshare_out;
1702
1703         /*
1704          * CLONE_NEWIPC must also detach from the undolist: after switching
1705          * to a new ipc namespace, the semaphore arrays from the old
1706          * namespace are unreachable.
1707          */
1708         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1709                 do_sysvsem = 1;
1710         if ((err = unshare_thread(unshare_flags)))
1711                 goto bad_unshare_out;
1712         if ((err = unshare_fs(unshare_flags, &new_fs)))
1713                 goto bad_unshare_cleanup_thread;
1714         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1715                 goto bad_unshare_cleanup_fs;
1716         if ((err = unshare_vm(unshare_flags, &new_mm)))
1717                 goto bad_unshare_cleanup_sigh;
1718         if ((err = unshare_fd(unshare_flags, &new_fd)))
1719                 goto bad_unshare_cleanup_vm;
1720         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1721                         new_fs)))
1722                 goto bad_unshare_cleanup_fd;
1723
1724         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1725                 if (do_sysvsem) {
1726                         /*
1727                          * CLONE_SYSVSEM is equivalent to sys_exit().
1728                          */
1729                         exit_sem(current);
1730                 }
1731
1732                 if (new_nsproxy) {
1733                         switch_task_namespaces(current, new_nsproxy);
1734                         new_nsproxy = NULL;
1735                 }
1736
1737                 task_lock(current);
1738
1739                 if (new_fs) {
1740                         fs = current->fs;
1741                         current->fs = new_fs;
1742                         new_fs = fs;
1743                 }
1744
1745                 if (new_mm) {
1746                         mm = current->mm;
1747                         active_mm = current->active_mm;
1748                         current->mm = new_mm;
1749                         current->active_mm = new_mm;
1750                         activate_mm(active_mm, new_mm);
1751                         new_mm = mm;
1752                 }
1753
1754                 if (new_fd) {
1755                         fd = current->files;
1756                         current->files = new_fd;
1757                         new_fd = fd;
1758                 }
1759
1760                 task_unlock(current);
1761         }
1762
1763         if (new_nsproxy)
1764                 put_nsproxy(new_nsproxy);
1765
1766 bad_unshare_cleanup_fd:
1767         if (new_fd)
1768                 put_files_struct(new_fd);
1769
1770 bad_unshare_cleanup_vm:
1771         if (new_mm)
1772                 mmput(new_mm);
1773
1774 bad_unshare_cleanup_sigh:
1775         if (new_sigh)
1776                 if (atomic_dec_and_test(&new_sigh->count))
1777                         kmem_cache_free(sighand_cachep, new_sigh);
1778
1779 bad_unshare_cleanup_fs:
1780         if (new_fs)
1781                 put_fs_struct(new_fs);
1782
1783 bad_unshare_cleanup_thread:
1784 bad_unshare_out:
1785         return err;
1786 }
1787
1788 /*
1789  *      Helper to unshare the files of the current task.
1790  *      We don't want to expose copy_files internals to
1791  *      the exec layer of the kernel.
1792  */
1793
1794 int unshare_files(struct files_struct **displaced)
1795 {
1796         struct task_struct *task = current;
1797         struct files_struct *copy = NULL;
1798         int error;
1799
1800         error = unshare_fd(CLONE_FILES, &copy);
1801         if (error || !copy) {
1802                 *displaced = NULL;
1803                 return error;
1804         }
1805         *displaced = task->files;
1806         task_lock(task);
1807         task->files = copy;
1808         task_unlock(task);
1809         return 0;
1810 }