Merge git://git.kernel.org/pub/scm/linux/kernel/git/x86/linux-2.6-x86
[linux-2.6] / 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/key.h>
26 #include <linux/binfmts.h>
27 #include <linux/mman.h>
28 #include <linux/fs.h>
29 #include <linux/nsproxy.h>
30 #include <linux/capability.h>
31 #include <linux/cpu.h>
32 #include <linux/cgroup.h>
33 #include <linux/security.h>
34 #include <linux/swap.h>
35 #include <linux/syscalls.h>
36 #include <linux/jiffies.h>
37 #include <linux/futex.h>
38 #include <linux/task_io_accounting_ops.h>
39 #include <linux/rcupdate.h>
40 #include <linux/ptrace.h>
41 #include <linux/mount.h>
42 #include <linux/audit.h>
43 #include <linux/memcontrol.h>
44 #include <linux/profile.h>
45 #include <linux/rmap.h>
46 #include <linux/acct.h>
47 #include <linux/tsacct_kern.h>
48 #include <linux/cn_proc.h>
49 #include <linux/freezer.h>
50 #include <linux/delayacct.h>
51 #include <linux/taskstats_kern.h>
52 #include <linux/random.h>
53 #include <linux/tty.h>
54 #include <linux/proc_fs.h>
55 #include <linux/blkdev.h>
56
57 #include <asm/pgtable.h>
58 #include <asm/pgalloc.h>
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
61 #include <asm/cacheflush.h>
62 #include <asm/tlbflush.h>
63
64 /*
65  * Protected counters by write_lock_irq(&tasklist_lock)
66  */
67 unsigned long total_forks;      /* Handle normal Linux uptimes. */
68 int nr_threads;                 /* The idle threads do not count.. */
69
70 int max_threads;                /* tunable limit on nr_threads */
71
72 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
73
74 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
75
76 int nr_processes(void)
77 {
78         int cpu;
79         int total = 0;
80
81         for_each_online_cpu(cpu)
82                 total += per_cpu(process_counts, cpu);
83
84         return total;
85 }
86
87 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
88 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
89 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
90 static struct kmem_cache *task_struct_cachep;
91 #endif
92
93 /* SLAB cache for signal_struct structures (tsk->signal) */
94 static struct kmem_cache *signal_cachep;
95
96 /* SLAB cache for sighand_struct structures (tsk->sighand) */
97 struct kmem_cache *sighand_cachep;
98
99 /* SLAB cache for files_struct structures (tsk->files) */
100 struct kmem_cache *files_cachep;
101
102 /* SLAB cache for fs_struct structures (tsk->fs) */
103 struct kmem_cache *fs_cachep;
104
105 /* SLAB cache for vm_area_struct structures */
106 struct kmem_cache *vm_area_cachep;
107
108 /* SLAB cache for mm_struct structures (tsk->mm) */
109 static struct kmem_cache *mm_cachep;
110
111 void free_task(struct task_struct *tsk)
112 {
113         prop_local_destroy_single(&tsk->dirties);
114         free_thread_info(tsk->stack);
115         rt_mutex_debug_task_free(tsk);
116         free_task_struct(tsk);
117 }
118 EXPORT_SYMBOL(free_task);
119
120 void __put_task_struct(struct task_struct *tsk)
121 {
122         WARN_ON(!tsk->exit_state);
123         WARN_ON(atomic_read(&tsk->usage));
124         WARN_ON(tsk == current);
125
126         security_task_free(tsk);
127         free_uid(tsk->user);
128         put_group_info(tsk->group_info);
129         delayacct_tsk_free(tsk);
130
131         if (!profile_handoff_task(tsk))
132                 free_task(tsk);
133 }
134
135 void __init fork_init(unsigned long mempages)
136 {
137 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
138 #ifndef ARCH_MIN_TASKALIGN
139 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
140 #endif
141         /* create a slab on which task_structs can be allocated */
142         task_struct_cachep =
143                 kmem_cache_create("task_struct", sizeof(struct task_struct),
144                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
145 #endif
146
147         /*
148          * The default maximum number of threads is set to a safe
149          * value: the thread structures can take up at most half
150          * of memory.
151          */
152         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
153
154         /*
155          * we need to allow at least 20 threads to boot a system
156          */
157         if(max_threads < 20)
158                 max_threads = 20;
159
160         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
161         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
162         init_task.signal->rlim[RLIMIT_SIGPENDING] =
163                 init_task.signal->rlim[RLIMIT_NPROC];
164 }
165
166 static struct task_struct *dup_task_struct(struct task_struct *orig)
167 {
168         struct task_struct *tsk;
169         struct thread_info *ti;
170         int err;
171
172         prepare_to_copy(orig);
173
174         tsk = alloc_task_struct();
175         if (!tsk)
176                 return NULL;
177
178         ti = alloc_thread_info(tsk);
179         if (!ti) {
180                 free_task_struct(tsk);
181                 return NULL;
182         }
183
184         *tsk = *orig;
185         tsk->stack = ti;
186
187         err = prop_local_init_single(&tsk->dirties);
188         if (err) {
189                 free_thread_info(ti);
190                 free_task_struct(tsk);
191                 return NULL;
192         }
193
194         setup_thread_stack(tsk, orig);
195
196 #ifdef CONFIG_CC_STACKPROTECTOR
197         tsk->stack_canary = get_random_int();
198 #endif
199
200         /* One for us, one for whoever does the "release_task()" (usually parent) */
201         atomic_set(&tsk->usage,2);
202         atomic_set(&tsk->fs_excl, 0);
203 #ifdef CONFIG_BLK_DEV_IO_TRACE
204         tsk->btrace_seq = 0;
205 #endif
206         tsk->splice_pipe = NULL;
207         return tsk;
208 }
209
210 #ifdef CONFIG_MMU
211 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
212 {
213         struct vm_area_struct *mpnt, *tmp, **pprev;
214         struct rb_node **rb_link, *rb_parent;
215         int retval;
216         unsigned long charge;
217         struct mempolicy *pol;
218
219         down_write(&oldmm->mmap_sem);
220         flush_cache_dup_mm(oldmm);
221         /*
222          * Not linked in yet - no deadlock potential:
223          */
224         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
225
226         mm->locked_vm = 0;
227         mm->mmap = NULL;
228         mm->mmap_cache = NULL;
229         mm->free_area_cache = oldmm->mmap_base;
230         mm->cached_hole_size = ~0UL;
231         mm->map_count = 0;
232         cpus_clear(mm->cpu_vm_mask);
233         mm->mm_rb = RB_ROOT;
234         rb_link = &mm->mm_rb.rb_node;
235         rb_parent = NULL;
236         pprev = &mm->mmap;
237
238         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
239                 struct file *file;
240
241                 if (mpnt->vm_flags & VM_DONTCOPY) {
242                         long pages = vma_pages(mpnt);
243                         mm->total_vm -= pages;
244                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
245                                                                 -pages);
246                         continue;
247                 }
248                 charge = 0;
249                 if (mpnt->vm_flags & VM_ACCOUNT) {
250                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
251                         if (security_vm_enough_memory(len))
252                                 goto fail_nomem;
253                         charge = len;
254                 }
255                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
256                 if (!tmp)
257                         goto fail_nomem;
258                 *tmp = *mpnt;
259                 pol = mpol_copy(vma_policy(mpnt));
260                 retval = PTR_ERR(pol);
261                 if (IS_ERR(pol))
262                         goto fail_nomem_policy;
263                 vma_set_policy(tmp, pol);
264                 tmp->vm_flags &= ~VM_LOCKED;
265                 tmp->vm_mm = mm;
266                 tmp->vm_next = NULL;
267                 anon_vma_link(tmp);
268                 file = tmp->vm_file;
269                 if (file) {
270                         struct inode *inode = file->f_path.dentry->d_inode;
271                         get_file(file);
272                         if (tmp->vm_flags & VM_DENYWRITE)
273                                 atomic_dec(&inode->i_writecount);
274
275                         /* insert tmp into the share list, just after mpnt */
276                         spin_lock(&file->f_mapping->i_mmap_lock);
277                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
278                         flush_dcache_mmap_lock(file->f_mapping);
279                         vma_prio_tree_add(tmp, mpnt);
280                         flush_dcache_mmap_unlock(file->f_mapping);
281                         spin_unlock(&file->f_mapping->i_mmap_lock);
282                 }
283
284                 /*
285                  * Link in the new vma and copy the page table entries.
286                  */
287                 *pprev = tmp;
288                 pprev = &tmp->vm_next;
289
290                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
291                 rb_link = &tmp->vm_rb.rb_right;
292                 rb_parent = &tmp->vm_rb;
293
294                 mm->map_count++;
295                 retval = copy_page_range(mm, oldmm, mpnt);
296
297                 if (tmp->vm_ops && tmp->vm_ops->open)
298                         tmp->vm_ops->open(tmp);
299
300                 if (retval)
301                         goto out;
302         }
303         /* a new mm has just been created */
304         arch_dup_mmap(oldmm, mm);
305         retval = 0;
306 out:
307         up_write(&mm->mmap_sem);
308         flush_tlb_mm(oldmm);
309         up_write(&oldmm->mmap_sem);
310         return retval;
311 fail_nomem_policy:
312         kmem_cache_free(vm_area_cachep, tmp);
313 fail_nomem:
314         retval = -ENOMEM;
315         vm_unacct_memory(charge);
316         goto out;
317 }
318
319 static inline int mm_alloc_pgd(struct mm_struct * mm)
320 {
321         mm->pgd = pgd_alloc(mm);
322         if (unlikely(!mm->pgd))
323                 return -ENOMEM;
324         return 0;
325 }
326
327 static inline void mm_free_pgd(struct mm_struct * mm)
328 {
329         pgd_free(mm, mm->pgd);
330 }
331 #else
332 #define dup_mmap(mm, oldmm)     (0)
333 #define mm_alloc_pgd(mm)        (0)
334 #define mm_free_pgd(mm)
335 #endif /* CONFIG_MMU */
336
337 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
338
339 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
340 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
341
342 #include <linux/init_task.h>
343
344 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
345 {
346         atomic_set(&mm->mm_users, 1);
347         atomic_set(&mm->mm_count, 1);
348         init_rwsem(&mm->mmap_sem);
349         INIT_LIST_HEAD(&mm->mmlist);
350         mm->flags = (current->mm) ? current->mm->flags
351                                   : MMF_DUMP_FILTER_DEFAULT;
352         mm->core_waiters = 0;
353         mm->nr_ptes = 0;
354         set_mm_counter(mm, file_rss, 0);
355         set_mm_counter(mm, anon_rss, 0);
356         spin_lock_init(&mm->page_table_lock);
357         rwlock_init(&mm->ioctx_list_lock);
358         mm->ioctx_list = NULL;
359         mm->free_area_cache = TASK_UNMAPPED_BASE;
360         mm->cached_hole_size = ~0UL;
361         mm_init_cgroup(mm, p);
362
363         if (likely(!mm_alloc_pgd(mm))) {
364                 mm->def_flags = 0;
365                 return mm;
366         }
367
368         mm_free_cgroup(mm);
369         free_mm(mm);
370         return NULL;
371 }
372
373 /*
374  * Allocate and initialize an mm_struct.
375  */
376 struct mm_struct * mm_alloc(void)
377 {
378         struct mm_struct * mm;
379
380         mm = allocate_mm();
381         if (mm) {
382                 memset(mm, 0, sizeof(*mm));
383                 mm = mm_init(mm, current);
384         }
385         return mm;
386 }
387
388 /*
389  * Called when the last reference to the mm
390  * is dropped: either by a lazy thread or by
391  * mmput. Free the page directory and the mm.
392  */
393 void __mmdrop(struct mm_struct *mm)
394 {
395         BUG_ON(mm == &init_mm);
396         mm_free_pgd(mm);
397         mm_free_cgroup(mm);
398         destroy_context(mm);
399         free_mm(mm);
400 }
401 EXPORT_SYMBOL_GPL(__mmdrop);
402
403 /*
404  * Decrement the use count and release all resources for an mm.
405  */
406 void mmput(struct mm_struct *mm)
407 {
408         might_sleep();
409
410         if (atomic_dec_and_test(&mm->mm_users)) {
411                 exit_aio(mm);
412                 exit_mmap(mm);
413                 if (!list_empty(&mm->mmlist)) {
414                         spin_lock(&mmlist_lock);
415                         list_del(&mm->mmlist);
416                         spin_unlock(&mmlist_lock);
417                 }
418                 put_swap_token(mm);
419                 mmdrop(mm);
420         }
421 }
422 EXPORT_SYMBOL_GPL(mmput);
423
424 /**
425  * get_task_mm - acquire a reference to the task's mm
426  *
427  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
428  * this kernel workthread has transiently adopted a user mm with use_mm,
429  * to do its AIO) is not set and if so returns a reference to it, after
430  * bumping up the use count.  User must release the mm via mmput()
431  * after use.  Typically used by /proc and ptrace.
432  */
433 struct mm_struct *get_task_mm(struct task_struct *task)
434 {
435         struct mm_struct *mm;
436
437         task_lock(task);
438         mm = task->mm;
439         if (mm) {
440                 if (task->flags & PF_BORROWED_MM)
441                         mm = NULL;
442                 else
443                         atomic_inc(&mm->mm_users);
444         }
445         task_unlock(task);
446         return mm;
447 }
448 EXPORT_SYMBOL_GPL(get_task_mm);
449
450 /* Please note the differences between mmput and mm_release.
451  * mmput is called whenever we stop holding onto a mm_struct,
452  * error success whatever.
453  *
454  * mm_release is called after a mm_struct has been removed
455  * from the current process.
456  *
457  * This difference is important for error handling, when we
458  * only half set up a mm_struct for a new process and need to restore
459  * the old one.  Because we mmput the new mm_struct before
460  * restoring the old one. . .
461  * Eric Biederman 10 January 1998
462  */
463 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
464 {
465         struct completion *vfork_done = tsk->vfork_done;
466
467         /* Get rid of any cached register state */
468         deactivate_mm(tsk, mm);
469
470         /* notify parent sleeping on vfork() */
471         if (vfork_done) {
472                 tsk->vfork_done = NULL;
473                 complete(vfork_done);
474         }
475
476         /*
477          * If we're exiting normally, clear a user-space tid field if
478          * requested.  We leave this alone when dying by signal, to leave
479          * the value intact in a core dump, and to save the unnecessary
480          * trouble otherwise.  Userland only wants this done for a sys_exit.
481          */
482         if (tsk->clear_child_tid
483             && !(tsk->flags & PF_SIGNALED)
484             && atomic_read(&mm->mm_users) > 1) {
485                 u32 __user * tidptr = tsk->clear_child_tid;
486                 tsk->clear_child_tid = NULL;
487
488                 /*
489                  * We don't check the error code - if userspace has
490                  * not set up a proper pointer then tough luck.
491                  */
492                 put_user(0, tidptr);
493                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
494         }
495 }
496
497 /*
498  * Allocate a new mm structure and copy contents from the
499  * mm structure of the passed in task structure.
500  */
501 static struct mm_struct *dup_mm(struct task_struct *tsk)
502 {
503         struct mm_struct *mm, *oldmm = current->mm;
504         int err;
505
506         if (!oldmm)
507                 return NULL;
508
509         mm = allocate_mm();
510         if (!mm)
511                 goto fail_nomem;
512
513         memcpy(mm, oldmm, sizeof(*mm));
514
515         /* Initializing for Swap token stuff */
516         mm->token_priority = 0;
517         mm->last_interval = 0;
518
519         if (!mm_init(mm, tsk))
520                 goto fail_nomem;
521
522         if (init_new_context(tsk, mm))
523                 goto fail_nocontext;
524
525         err = dup_mmap(mm, oldmm);
526         if (err)
527                 goto free_pt;
528
529         mm->hiwater_rss = get_mm_rss(mm);
530         mm->hiwater_vm = mm->total_vm;
531
532         return mm;
533
534 free_pt:
535         mmput(mm);
536
537 fail_nomem:
538         return NULL;
539
540 fail_nocontext:
541         /*
542          * If init_new_context() failed, we cannot use mmput() to free the mm
543          * because it calls destroy_context()
544          */
545         mm_free_pgd(mm);
546         free_mm(mm);
547         return NULL;
548 }
549
550 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
551 {
552         struct mm_struct * mm, *oldmm;
553         int retval;
554
555         tsk->min_flt = tsk->maj_flt = 0;
556         tsk->nvcsw = tsk->nivcsw = 0;
557
558         tsk->mm = NULL;
559         tsk->active_mm = NULL;
560
561         /*
562          * Are we cloning a kernel thread?
563          *
564          * We need to steal a active VM for that..
565          */
566         oldmm = current->mm;
567         if (!oldmm)
568                 return 0;
569
570         if (clone_flags & CLONE_VM) {
571                 atomic_inc(&oldmm->mm_users);
572                 mm = oldmm;
573                 goto good_mm;
574         }
575
576         retval = -ENOMEM;
577         mm = dup_mm(tsk);
578         if (!mm)
579                 goto fail_nomem;
580
581 good_mm:
582         /* Initializing for Swap token stuff */
583         mm->token_priority = 0;
584         mm->last_interval = 0;
585
586         tsk->mm = mm;
587         tsk->active_mm = mm;
588         return 0;
589
590 fail_nomem:
591         return retval;
592 }
593
594 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
595 {
596         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
597         /* We don't need to lock fs - think why ;-) */
598         if (fs) {
599                 atomic_set(&fs->count, 1);
600                 rwlock_init(&fs->lock);
601                 fs->umask = old->umask;
602                 read_lock(&old->lock);
603                 fs->root = old->root;
604                 path_get(&old->root);
605                 fs->pwd = old->pwd;
606                 path_get(&old->pwd);
607                 if (old->altroot.dentry) {
608                         fs->altroot = old->altroot;
609                         path_get(&old->altroot);
610                 } else {
611                         fs->altroot.mnt = NULL;
612                         fs->altroot.dentry = NULL;
613                 }
614                 read_unlock(&old->lock);
615         }
616         return fs;
617 }
618
619 struct fs_struct *copy_fs_struct(struct fs_struct *old)
620 {
621         return __copy_fs_struct(old);
622 }
623
624 EXPORT_SYMBOL_GPL(copy_fs_struct);
625
626 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
627 {
628         if (clone_flags & CLONE_FS) {
629                 atomic_inc(&current->fs->count);
630                 return 0;
631         }
632         tsk->fs = __copy_fs_struct(current->fs);
633         if (!tsk->fs)
634                 return -ENOMEM;
635         return 0;
636 }
637
638 static int count_open_files(struct fdtable *fdt)
639 {
640         int size = fdt->max_fds;
641         int i;
642
643         /* Find the last open fd */
644         for (i = size/(8*sizeof(long)); i > 0; ) {
645                 if (fdt->open_fds->fds_bits[--i])
646                         break;
647         }
648         i = (i+1) * 8 * sizeof(long);
649         return i;
650 }
651
652 static struct files_struct *alloc_files(void)
653 {
654         struct files_struct *newf;
655         struct fdtable *fdt;
656
657         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
658         if (!newf)
659                 goto out;
660
661         atomic_set(&newf->count, 1);
662
663         spin_lock_init(&newf->file_lock);
664         newf->next_fd = 0;
665         fdt = &newf->fdtab;
666         fdt->max_fds = NR_OPEN_DEFAULT;
667         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
668         fdt->open_fds = (fd_set *)&newf->open_fds_init;
669         fdt->fd = &newf->fd_array[0];
670         INIT_RCU_HEAD(&fdt->rcu);
671         fdt->next = NULL;
672         rcu_assign_pointer(newf->fdt, fdt);
673 out:
674         return newf;
675 }
676
677 /*
678  * Allocate a new files structure and copy contents from the
679  * passed in files structure.
680  * errorp will be valid only when the returned files_struct is NULL.
681  */
682 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
683 {
684         struct files_struct *newf;
685         struct file **old_fds, **new_fds;
686         int open_files, size, i;
687         struct fdtable *old_fdt, *new_fdt;
688
689         *errorp = -ENOMEM;
690         newf = alloc_files();
691         if (!newf)
692                 goto out;
693
694         spin_lock(&oldf->file_lock);
695         old_fdt = files_fdtable(oldf);
696         new_fdt = files_fdtable(newf);
697         open_files = count_open_files(old_fdt);
698
699         /*
700          * Check whether we need to allocate a larger fd array and fd set.
701          * Note: we're not a clone task, so the open count won't change.
702          */
703         if (open_files > new_fdt->max_fds) {
704                 new_fdt->max_fds = 0;
705                 spin_unlock(&oldf->file_lock);
706                 spin_lock(&newf->file_lock);
707                 *errorp = expand_files(newf, open_files-1);
708                 spin_unlock(&newf->file_lock);
709                 if (*errorp < 0)
710                         goto out_release;
711                 new_fdt = files_fdtable(newf);
712                 /*
713                  * Reacquire the oldf lock and a pointer to its fd table
714                  * who knows it may have a new bigger fd table. We need
715                  * the latest pointer.
716                  */
717                 spin_lock(&oldf->file_lock);
718                 old_fdt = files_fdtable(oldf);
719         }
720
721         old_fds = old_fdt->fd;
722         new_fds = new_fdt->fd;
723
724         memcpy(new_fdt->open_fds->fds_bits,
725                 old_fdt->open_fds->fds_bits, open_files/8);
726         memcpy(new_fdt->close_on_exec->fds_bits,
727                 old_fdt->close_on_exec->fds_bits, open_files/8);
728
729         for (i = open_files; i != 0; i--) {
730                 struct file *f = *old_fds++;
731                 if (f) {
732                         get_file(f);
733                 } else {
734                         /*
735                          * The fd may be claimed in the fd bitmap but not yet
736                          * instantiated in the files array if a sibling thread
737                          * is partway through open().  So make sure that this
738                          * fd is available to the new process.
739                          */
740                         FD_CLR(open_files - i, new_fdt->open_fds);
741                 }
742                 rcu_assign_pointer(*new_fds++, f);
743         }
744         spin_unlock(&oldf->file_lock);
745
746         /* compute the remainder to be cleared */
747         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
748
749         /* This is long word aligned thus could use a optimized version */
750         memset(new_fds, 0, size);
751
752         if (new_fdt->max_fds > open_files) {
753                 int left = (new_fdt->max_fds-open_files)/8;
754                 int start = open_files / (8 * sizeof(unsigned long));
755
756                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
757                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
758         }
759
760         return newf;
761
762 out_release:
763         kmem_cache_free(files_cachep, newf);
764 out:
765         return NULL;
766 }
767
768 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
769 {
770         struct files_struct *oldf, *newf;
771         int error = 0;
772
773         /*
774          * A background process may not have any files ...
775          */
776         oldf = current->files;
777         if (!oldf)
778                 goto out;
779
780         if (clone_flags & CLONE_FILES) {
781                 atomic_inc(&oldf->count);
782                 goto out;
783         }
784
785         /*
786          * Note: we may be using current for both targets (See exec.c)
787          * This works because we cache current->files (old) as oldf. Don't
788          * break this.
789          */
790         tsk->files = NULL;
791         newf = dup_fd(oldf, &error);
792         if (!newf)
793                 goto out;
794
795         tsk->files = newf;
796         error = 0;
797 out:
798         return error;
799 }
800
801 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
802 {
803 #ifdef CONFIG_BLOCK
804         struct io_context *ioc = current->io_context;
805
806         if (!ioc)
807                 return 0;
808         /*
809          * Share io context with parent, if CLONE_IO is set
810          */
811         if (clone_flags & CLONE_IO) {
812                 tsk->io_context = ioc_task_link(ioc);
813                 if (unlikely(!tsk->io_context))
814                         return -ENOMEM;
815         } else if (ioprio_valid(ioc->ioprio)) {
816                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
817                 if (unlikely(!tsk->io_context))
818                         return -ENOMEM;
819
820                 tsk->io_context->ioprio = ioc->ioprio;
821         }
822 #endif
823         return 0;
824 }
825
826 /*
827  *      Helper to unshare the files of the current task.
828  *      We don't want to expose copy_files internals to
829  *      the exec layer of the kernel.
830  */
831
832 int unshare_files(void)
833 {
834         struct files_struct *files  = current->files;
835         int rc;
836
837         BUG_ON(!files);
838
839         /* This can race but the race causes us to copy when we don't
840            need to and drop the copy */
841         if(atomic_read(&files->count) == 1)
842         {
843                 atomic_inc(&files->count);
844                 return 0;
845         }
846         rc = copy_files(0, current);
847         if(rc)
848                 current->files = files;
849         return rc;
850 }
851
852 EXPORT_SYMBOL(unshare_files);
853
854 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
855 {
856         struct sighand_struct *sig;
857
858         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
859                 atomic_inc(&current->sighand->count);
860                 return 0;
861         }
862         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
863         rcu_assign_pointer(tsk->sighand, sig);
864         if (!sig)
865                 return -ENOMEM;
866         atomic_set(&sig->count, 1);
867         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
868         return 0;
869 }
870
871 void __cleanup_sighand(struct sighand_struct *sighand)
872 {
873         if (atomic_dec_and_test(&sighand->count))
874                 kmem_cache_free(sighand_cachep, sighand);
875 }
876
877 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
878 {
879         struct signal_struct *sig;
880         int ret;
881
882         if (clone_flags & CLONE_THREAD) {
883                 atomic_inc(&current->signal->count);
884                 atomic_inc(&current->signal->live);
885                 return 0;
886         }
887         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
888         tsk->signal = sig;
889         if (!sig)
890                 return -ENOMEM;
891
892         ret = copy_thread_group_keys(tsk);
893         if (ret < 0) {
894                 kmem_cache_free(signal_cachep, sig);
895                 return ret;
896         }
897
898         atomic_set(&sig->count, 1);
899         atomic_set(&sig->live, 1);
900         init_waitqueue_head(&sig->wait_chldexit);
901         sig->flags = 0;
902         sig->group_exit_code = 0;
903         sig->group_exit_task = NULL;
904         sig->group_stop_count = 0;
905         sig->curr_target = NULL;
906         init_sigpending(&sig->shared_pending);
907         INIT_LIST_HEAD(&sig->posix_timers);
908
909         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
910         sig->it_real_incr.tv64 = 0;
911         sig->real_timer.function = it_real_fn;
912
913         sig->it_virt_expires = cputime_zero;
914         sig->it_virt_incr = cputime_zero;
915         sig->it_prof_expires = cputime_zero;
916         sig->it_prof_incr = cputime_zero;
917
918         sig->leader = 0;        /* session leadership doesn't inherit */
919         sig->tty_old_pgrp = NULL;
920
921         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
922         sig->gtime = cputime_zero;
923         sig->cgtime = cputime_zero;
924         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
925         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
926         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
927         sig->sum_sched_runtime = 0;
928         INIT_LIST_HEAD(&sig->cpu_timers[0]);
929         INIT_LIST_HEAD(&sig->cpu_timers[1]);
930         INIT_LIST_HEAD(&sig->cpu_timers[2]);
931         taskstats_tgid_init(sig);
932
933         task_lock(current->group_leader);
934         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
935         task_unlock(current->group_leader);
936
937         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
938                 /*
939                  * New sole thread in the process gets an expiry time
940                  * of the whole CPU time limit.
941                  */
942                 tsk->it_prof_expires =
943                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
944         }
945         acct_init_pacct(&sig->pacct);
946
947         tty_audit_fork(sig);
948
949         return 0;
950 }
951
952 void __cleanup_signal(struct signal_struct *sig)
953 {
954         exit_thread_group_keys(sig);
955         kmem_cache_free(signal_cachep, sig);
956 }
957
958 static void cleanup_signal(struct task_struct *tsk)
959 {
960         struct signal_struct *sig = tsk->signal;
961
962         atomic_dec(&sig->live);
963
964         if (atomic_dec_and_test(&sig->count))
965                 __cleanup_signal(sig);
966 }
967
968 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
969 {
970         unsigned long new_flags = p->flags;
971
972         new_flags &= ~PF_SUPERPRIV;
973         new_flags |= PF_FORKNOEXEC;
974         if (!(clone_flags & CLONE_PTRACE))
975                 p->ptrace = 0;
976         p->flags = new_flags;
977         clear_freeze_flag(p);
978 }
979
980 asmlinkage long sys_set_tid_address(int __user *tidptr)
981 {
982         current->clear_child_tid = tidptr;
983
984         return task_pid_vnr(current);
985 }
986
987 static void rt_mutex_init_task(struct task_struct *p)
988 {
989         spin_lock_init(&p->pi_lock);
990 #ifdef CONFIG_RT_MUTEXES
991         plist_head_init(&p->pi_waiters, &p->pi_lock);
992         p->pi_blocked_on = NULL;
993 #endif
994 }
995
996 /*
997  * This creates a new process as a copy of the old one,
998  * but does not actually start it yet.
999  *
1000  * It copies the registers, and all the appropriate
1001  * parts of the process environment (as per the clone
1002  * flags). The actual kick-off is left to the caller.
1003  */
1004 static struct task_struct *copy_process(unsigned long clone_flags,
1005                                         unsigned long stack_start,
1006                                         struct pt_regs *regs,
1007                                         unsigned long stack_size,
1008                                         int __user *child_tidptr,
1009                                         struct pid *pid)
1010 {
1011         int retval;
1012         struct task_struct *p;
1013         int cgroup_callbacks_done = 0;
1014
1015         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1016                 return ERR_PTR(-EINVAL);
1017
1018         /*
1019          * Thread groups must share signals as well, and detached threads
1020          * can only be started up within the thread group.
1021          */
1022         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1023                 return ERR_PTR(-EINVAL);
1024
1025         /*
1026          * Shared signal handlers imply shared VM. By way of the above,
1027          * thread groups also imply shared VM. Blocking this case allows
1028          * for various simplifications in other code.
1029          */
1030         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1031                 return ERR_PTR(-EINVAL);
1032
1033         retval = security_task_create(clone_flags);
1034         if (retval)
1035                 goto fork_out;
1036
1037         retval = -ENOMEM;
1038         p = dup_task_struct(current);
1039         if (!p)
1040                 goto fork_out;
1041
1042         rt_mutex_init_task(p);
1043
1044 #ifdef CONFIG_TRACE_IRQFLAGS
1045         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1046         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1047 #endif
1048         retval = -EAGAIN;
1049         if (atomic_read(&p->user->processes) >=
1050                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1051                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1052                     p->user != current->nsproxy->user_ns->root_user)
1053                         goto bad_fork_free;
1054         }
1055
1056         atomic_inc(&p->user->__count);
1057         atomic_inc(&p->user->processes);
1058         get_group_info(p->group_info);
1059
1060         /*
1061          * If multiple threads are within copy_process(), then this check
1062          * triggers too late. This doesn't hurt, the check is only there
1063          * to stop root fork bombs.
1064          */
1065         if (nr_threads >= max_threads)
1066                 goto bad_fork_cleanup_count;
1067
1068         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1069                 goto bad_fork_cleanup_count;
1070
1071         if (p->binfmt && !try_module_get(p->binfmt->module))
1072                 goto bad_fork_cleanup_put_domain;
1073
1074         p->did_exec = 0;
1075         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1076         copy_flags(clone_flags, p);
1077         INIT_LIST_HEAD(&p->children);
1078         INIT_LIST_HEAD(&p->sibling);
1079 #ifdef CONFIG_PREEMPT_RCU
1080         p->rcu_read_lock_nesting = 0;
1081         p->rcu_flipctr_idx = 0;
1082 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1083         p->vfork_done = NULL;
1084         spin_lock_init(&p->alloc_lock);
1085
1086         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1087         init_sigpending(&p->pending);
1088
1089         p->utime = cputime_zero;
1090         p->stime = cputime_zero;
1091         p->gtime = cputime_zero;
1092         p->utimescaled = cputime_zero;
1093         p->stimescaled = cputime_zero;
1094         p->prev_utime = cputime_zero;
1095         p->prev_stime = cputime_zero;
1096
1097 #ifdef CONFIG_DETECT_SOFTLOCKUP
1098         p->last_switch_count = 0;
1099         p->last_switch_timestamp = 0;
1100 #endif
1101
1102 #ifdef CONFIG_TASK_XACCT
1103         p->rchar = 0;           /* I/O counter: bytes read */
1104         p->wchar = 0;           /* I/O counter: bytes written */
1105         p->syscr = 0;           /* I/O counter: read syscalls */
1106         p->syscw = 0;           /* I/O counter: write syscalls */
1107 #endif
1108         task_io_accounting_init(p);
1109         acct_clear_integrals(p);
1110
1111         p->it_virt_expires = cputime_zero;
1112         p->it_prof_expires = cputime_zero;
1113         p->it_sched_expires = 0;
1114         INIT_LIST_HEAD(&p->cpu_timers[0]);
1115         INIT_LIST_HEAD(&p->cpu_timers[1]);
1116         INIT_LIST_HEAD(&p->cpu_timers[2]);
1117
1118         p->lock_depth = -1;             /* -1 = no lock */
1119         do_posix_clock_monotonic_gettime(&p->start_time);
1120         p->real_start_time = p->start_time;
1121         monotonic_to_bootbased(&p->real_start_time);
1122 #ifdef CONFIG_SECURITY
1123         p->security = NULL;
1124 #endif
1125         p->cap_bset = current->cap_bset;
1126         p->io_context = NULL;
1127         p->audit_context = NULL;
1128         cgroup_fork(p);
1129 #ifdef CONFIG_NUMA
1130         p->mempolicy = mpol_copy(p->mempolicy);
1131         if (IS_ERR(p->mempolicy)) {
1132                 retval = PTR_ERR(p->mempolicy);
1133                 p->mempolicy = NULL;
1134                 goto bad_fork_cleanup_cgroup;
1135         }
1136         mpol_fix_fork_child_flag(p);
1137 #endif
1138 #ifdef CONFIG_TRACE_IRQFLAGS
1139         p->irq_events = 0;
1140 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1141         p->hardirqs_enabled = 1;
1142 #else
1143         p->hardirqs_enabled = 0;
1144 #endif
1145         p->hardirq_enable_ip = 0;
1146         p->hardirq_enable_event = 0;
1147         p->hardirq_disable_ip = _THIS_IP_;
1148         p->hardirq_disable_event = 0;
1149         p->softirqs_enabled = 1;
1150         p->softirq_enable_ip = _THIS_IP_;
1151         p->softirq_enable_event = 0;
1152         p->softirq_disable_ip = 0;
1153         p->softirq_disable_event = 0;
1154         p->hardirq_context = 0;
1155         p->softirq_context = 0;
1156 #endif
1157 #ifdef CONFIG_LOCKDEP
1158         p->lockdep_depth = 0; /* no locks held yet */
1159         p->curr_chain_key = 0;
1160         p->lockdep_recursion = 0;
1161 #endif
1162
1163 #ifdef CONFIG_DEBUG_MUTEXES
1164         p->blocked_on = NULL; /* not blocked yet */
1165 #endif
1166
1167         /* Perform scheduler related setup. Assign this task to a CPU. */
1168         sched_fork(p, clone_flags);
1169
1170         if ((retval = security_task_alloc(p)))
1171                 goto bad_fork_cleanup_policy;
1172         if ((retval = audit_alloc(p)))
1173                 goto bad_fork_cleanup_security;
1174         /* copy all the process information */
1175         if ((retval = copy_semundo(clone_flags, p)))
1176                 goto bad_fork_cleanup_audit;
1177         if ((retval = copy_files(clone_flags, p)))
1178                 goto bad_fork_cleanup_semundo;
1179         if ((retval = copy_fs(clone_flags, p)))
1180                 goto bad_fork_cleanup_files;
1181         if ((retval = copy_sighand(clone_flags, p)))
1182                 goto bad_fork_cleanup_fs;
1183         if ((retval = copy_signal(clone_flags, p)))
1184                 goto bad_fork_cleanup_sighand;
1185         if ((retval = copy_mm(clone_flags, p)))
1186                 goto bad_fork_cleanup_signal;
1187         if ((retval = copy_keys(clone_flags, p)))
1188                 goto bad_fork_cleanup_mm;
1189         if ((retval = copy_namespaces(clone_flags, p)))
1190                 goto bad_fork_cleanup_keys;
1191         if ((retval = copy_io(clone_flags, p)))
1192                 goto bad_fork_cleanup_namespaces;
1193         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1194         if (retval)
1195                 goto bad_fork_cleanup_io;
1196
1197         if (pid != &init_struct_pid) {
1198                 retval = -ENOMEM;
1199                 pid = alloc_pid(task_active_pid_ns(p));
1200                 if (!pid)
1201                         goto bad_fork_cleanup_io;
1202
1203                 if (clone_flags & CLONE_NEWPID) {
1204                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1205                         if (retval < 0)
1206                                 goto bad_fork_free_pid;
1207                 }
1208         }
1209
1210         p->pid = pid_nr(pid);
1211         p->tgid = p->pid;
1212         if (clone_flags & CLONE_THREAD)
1213                 p->tgid = current->tgid;
1214
1215         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1216         /*
1217          * Clear TID on mm_release()?
1218          */
1219         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1220 #ifdef CONFIG_FUTEX
1221         p->robust_list = NULL;
1222 #ifdef CONFIG_COMPAT
1223         p->compat_robust_list = NULL;
1224 #endif
1225         INIT_LIST_HEAD(&p->pi_state_list);
1226         p->pi_state_cache = NULL;
1227 #endif
1228         /*
1229          * sigaltstack should be cleared when sharing the same VM
1230          */
1231         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1232                 p->sas_ss_sp = p->sas_ss_size = 0;
1233
1234         /*
1235          * Syscall tracing should be turned off in the child regardless
1236          * of CLONE_PTRACE.
1237          */
1238         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1239 #ifdef TIF_SYSCALL_EMU
1240         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1241 #endif
1242         clear_all_latency_tracing(p);
1243
1244         /* Our parent execution domain becomes current domain
1245            These must match for thread signalling to apply */
1246         p->parent_exec_id = p->self_exec_id;
1247
1248         /* ok, now we should be set up.. */
1249         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1250         p->pdeath_signal = 0;
1251         p->exit_state = 0;
1252
1253         /*
1254          * Ok, make it visible to the rest of the system.
1255          * We dont wake it up yet.
1256          */
1257         p->group_leader = p;
1258         INIT_LIST_HEAD(&p->thread_group);
1259         INIT_LIST_HEAD(&p->ptrace_children);
1260         INIT_LIST_HEAD(&p->ptrace_list);
1261
1262         /* Now that the task is set up, run cgroup callbacks if
1263          * necessary. We need to run them before the task is visible
1264          * on the tasklist. */
1265         cgroup_fork_callbacks(p);
1266         cgroup_callbacks_done = 1;
1267
1268         /* Need tasklist lock for parent etc handling! */
1269         write_lock_irq(&tasklist_lock);
1270
1271         /*
1272          * The task hasn't been attached yet, so its cpus_allowed mask will
1273          * not be changed, nor will its assigned CPU.
1274          *
1275          * The cpus_allowed mask of the parent may have changed after it was
1276          * copied first time - so re-copy it here, then check the child's CPU
1277          * to ensure it is on a valid CPU (and if not, just force it back to
1278          * parent's CPU). This avoids alot of nasty races.
1279          */
1280         p->cpus_allowed = current->cpus_allowed;
1281         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1282         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1283                         !cpu_online(task_cpu(p))))
1284                 set_task_cpu(p, smp_processor_id());
1285
1286         /* CLONE_PARENT re-uses the old parent */
1287         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1288                 p->real_parent = current->real_parent;
1289         else
1290                 p->real_parent = current;
1291         p->parent = p->real_parent;
1292
1293         spin_lock(&current->sighand->siglock);
1294
1295         /*
1296          * Process group and session signals need to be delivered to just the
1297          * parent before the fork or both the parent and the child after the
1298          * fork. Restart if a signal comes in before we add the new process to
1299          * it's process group.
1300          * A fatal signal pending means that current will exit, so the new
1301          * thread can't slip out of an OOM kill (or normal SIGKILL).
1302          */
1303         recalc_sigpending();
1304         if (signal_pending(current)) {
1305                 spin_unlock(&current->sighand->siglock);
1306                 write_unlock_irq(&tasklist_lock);
1307                 retval = -ERESTARTNOINTR;
1308                 goto bad_fork_free_pid;
1309         }
1310
1311         if (clone_flags & CLONE_THREAD) {
1312                 p->group_leader = current->group_leader;
1313                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1314
1315                 if (!cputime_eq(current->signal->it_virt_expires,
1316                                 cputime_zero) ||
1317                     !cputime_eq(current->signal->it_prof_expires,
1318                                 cputime_zero) ||
1319                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1320                     !list_empty(&current->signal->cpu_timers[0]) ||
1321                     !list_empty(&current->signal->cpu_timers[1]) ||
1322                     !list_empty(&current->signal->cpu_timers[2])) {
1323                         /*
1324                          * Have child wake up on its first tick to check
1325                          * for process CPU timers.
1326                          */
1327                         p->it_prof_expires = jiffies_to_cputime(1);
1328                 }
1329         }
1330
1331         if (likely(p->pid)) {
1332                 add_parent(p);
1333                 if (unlikely(p->ptrace & PT_PTRACED))
1334                         __ptrace_link(p, current->parent);
1335
1336                 if (thread_group_leader(p)) {
1337                         if (clone_flags & CLONE_NEWPID)
1338                                 p->nsproxy->pid_ns->child_reaper = p;
1339
1340                         p->signal->leader_pid = pid;
1341                         p->signal->tty = current->signal->tty;
1342                         set_task_pgrp(p, task_pgrp_nr(current));
1343                         set_task_session(p, task_session_nr(current));
1344                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1345                         attach_pid(p, PIDTYPE_SID, task_session(current));
1346                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1347                         __get_cpu_var(process_counts)++;
1348                 }
1349                 attach_pid(p, PIDTYPE_PID, pid);
1350                 nr_threads++;
1351         }
1352
1353         total_forks++;
1354         spin_unlock(&current->sighand->siglock);
1355         write_unlock_irq(&tasklist_lock);
1356         proc_fork_connector(p);
1357         cgroup_post_fork(p);
1358         return p;
1359
1360 bad_fork_free_pid:
1361         if (pid != &init_struct_pid)
1362                 free_pid(pid);
1363 bad_fork_cleanup_io:
1364         put_io_context(p->io_context);
1365 bad_fork_cleanup_namespaces:
1366         exit_task_namespaces(p);
1367 bad_fork_cleanup_keys:
1368         exit_keys(p);
1369 bad_fork_cleanup_mm:
1370         if (p->mm)
1371                 mmput(p->mm);
1372 bad_fork_cleanup_signal:
1373         cleanup_signal(p);
1374 bad_fork_cleanup_sighand:
1375         __cleanup_sighand(p->sighand);
1376 bad_fork_cleanup_fs:
1377         exit_fs(p); /* blocking */
1378 bad_fork_cleanup_files:
1379         exit_files(p); /* blocking */
1380 bad_fork_cleanup_semundo:
1381         exit_sem(p);
1382 bad_fork_cleanup_audit:
1383         audit_free(p);
1384 bad_fork_cleanup_security:
1385         security_task_free(p);
1386 bad_fork_cleanup_policy:
1387 #ifdef CONFIG_NUMA
1388         mpol_free(p->mempolicy);
1389 bad_fork_cleanup_cgroup:
1390 #endif
1391         cgroup_exit(p, cgroup_callbacks_done);
1392         delayacct_tsk_free(p);
1393         if (p->binfmt)
1394                 module_put(p->binfmt->module);
1395 bad_fork_cleanup_put_domain:
1396         module_put(task_thread_info(p)->exec_domain->module);
1397 bad_fork_cleanup_count:
1398         put_group_info(p->group_info);
1399         atomic_dec(&p->user->processes);
1400         free_uid(p->user);
1401 bad_fork_free:
1402         free_task(p);
1403 fork_out:
1404         return ERR_PTR(retval);
1405 }
1406
1407 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1408 {
1409         memset(regs, 0, sizeof(struct pt_regs));
1410         return regs;
1411 }
1412
1413 struct task_struct * __cpuinit fork_idle(int cpu)
1414 {
1415         struct task_struct *task;
1416         struct pt_regs regs;
1417
1418         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1419                                 &init_struct_pid);
1420         if (!IS_ERR(task))
1421                 init_idle(task, cpu);
1422
1423         return task;
1424 }
1425
1426 static int fork_traceflag(unsigned clone_flags)
1427 {
1428         if (clone_flags & CLONE_UNTRACED)
1429                 return 0;
1430         else if (clone_flags & CLONE_VFORK) {
1431                 if (current->ptrace & PT_TRACE_VFORK)
1432                         return PTRACE_EVENT_VFORK;
1433         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1434                 if (current->ptrace & PT_TRACE_CLONE)
1435                         return PTRACE_EVENT_CLONE;
1436         } else if (current->ptrace & PT_TRACE_FORK)
1437                 return PTRACE_EVENT_FORK;
1438
1439         return 0;
1440 }
1441
1442 /*
1443  *  Ok, this is the main fork-routine.
1444  *
1445  * It copies the process, and if successful kick-starts
1446  * it and waits for it to finish using the VM if required.
1447  */
1448 long do_fork(unsigned long clone_flags,
1449               unsigned long stack_start,
1450               struct pt_regs *regs,
1451               unsigned long stack_size,
1452               int __user *parent_tidptr,
1453               int __user *child_tidptr)
1454 {
1455         struct task_struct *p;
1456         int trace = 0;
1457         long nr;
1458
1459         /*
1460          * We hope to recycle these flags after 2.6.26
1461          */
1462         if (unlikely(clone_flags & CLONE_STOPPED)) {
1463                 static int __read_mostly count = 100;
1464
1465                 if (count > 0 && printk_ratelimit()) {
1466                         char comm[TASK_COMM_LEN];
1467
1468                         count--;
1469                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1470                                         "clone flags 0x%lx\n",
1471                                 get_task_comm(comm, current),
1472                                 clone_flags & CLONE_STOPPED);
1473                 }
1474         }
1475
1476         if (unlikely(current->ptrace)) {
1477                 trace = fork_traceflag (clone_flags);
1478                 if (trace)
1479                         clone_flags |= CLONE_PTRACE;
1480         }
1481
1482         p = copy_process(clone_flags, stack_start, regs, stack_size,
1483                         child_tidptr, NULL);
1484         /*
1485          * Do this prior waking up the new thread - the thread pointer
1486          * might get invalid after that point, if the thread exits quickly.
1487          */
1488         if (!IS_ERR(p)) {
1489                 struct completion vfork;
1490
1491                 nr = task_pid_vnr(p);
1492
1493                 if (clone_flags & CLONE_PARENT_SETTID)
1494                         put_user(nr, parent_tidptr);
1495
1496                 if (clone_flags & CLONE_VFORK) {
1497                         p->vfork_done = &vfork;
1498                         init_completion(&vfork);
1499                 }
1500
1501                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1502                         /*
1503                          * We'll start up with an immediate SIGSTOP.
1504                          */
1505                         sigaddset(&p->pending.signal, SIGSTOP);
1506                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1507                 }
1508
1509                 if (!(clone_flags & CLONE_STOPPED))
1510                         wake_up_new_task(p, clone_flags);
1511                 else
1512                         __set_task_state(p, TASK_STOPPED);
1513
1514                 if (unlikely (trace)) {
1515                         current->ptrace_message = nr;
1516                         ptrace_notify ((trace << 8) | SIGTRAP);
1517                 }
1518
1519                 if (clone_flags & CLONE_VFORK) {
1520                         freezer_do_not_count();
1521                         wait_for_completion(&vfork);
1522                         freezer_count();
1523                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1524                                 current->ptrace_message = nr;
1525                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1526                         }
1527                 }
1528         } else {
1529                 nr = PTR_ERR(p);
1530         }
1531         return nr;
1532 }
1533
1534 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1535 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1536 #endif
1537
1538 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1539 {
1540         struct sighand_struct *sighand = data;
1541
1542         spin_lock_init(&sighand->siglock);
1543         init_waitqueue_head(&sighand->signalfd_wqh);
1544 }
1545
1546 void __init proc_caches_init(void)
1547 {
1548         sighand_cachep = kmem_cache_create("sighand_cache",
1549                         sizeof(struct sighand_struct), 0,
1550                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1551                         sighand_ctor);
1552         signal_cachep = kmem_cache_create("signal_cache",
1553                         sizeof(struct signal_struct), 0,
1554                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1555         files_cachep = kmem_cache_create("files_cache",
1556                         sizeof(struct files_struct), 0,
1557                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1558         fs_cachep = kmem_cache_create("fs_cache",
1559                         sizeof(struct fs_struct), 0,
1560                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1561         vm_area_cachep = kmem_cache_create("vm_area_struct",
1562                         sizeof(struct vm_area_struct), 0,
1563                         SLAB_PANIC, NULL);
1564         mm_cachep = kmem_cache_create("mm_struct",
1565                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1566                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1567 }
1568
1569 /*
1570  * Check constraints on flags passed to the unshare system call and
1571  * force unsharing of additional process context as appropriate.
1572  */
1573 static void check_unshare_flags(unsigned long *flags_ptr)
1574 {
1575         /*
1576          * If unsharing a thread from a thread group, must also
1577          * unshare vm.
1578          */
1579         if (*flags_ptr & CLONE_THREAD)
1580                 *flags_ptr |= CLONE_VM;
1581
1582         /*
1583          * If unsharing vm, must also unshare signal handlers.
1584          */
1585         if (*flags_ptr & CLONE_VM)
1586                 *flags_ptr |= CLONE_SIGHAND;
1587
1588         /*
1589          * If unsharing signal handlers and the task was created
1590          * using CLONE_THREAD, then must unshare the thread
1591          */
1592         if ((*flags_ptr & CLONE_SIGHAND) &&
1593             (atomic_read(&current->signal->count) > 1))
1594                 *flags_ptr |= CLONE_THREAD;
1595
1596         /*
1597          * If unsharing namespace, must also unshare filesystem information.
1598          */
1599         if (*flags_ptr & CLONE_NEWNS)
1600                 *flags_ptr |= CLONE_FS;
1601 }
1602
1603 /*
1604  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1605  */
1606 static int unshare_thread(unsigned long unshare_flags)
1607 {
1608         if (unshare_flags & CLONE_THREAD)
1609                 return -EINVAL;
1610
1611         return 0;
1612 }
1613
1614 /*
1615  * Unshare the filesystem structure if it is being shared
1616  */
1617 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1618 {
1619         struct fs_struct *fs = current->fs;
1620
1621         if ((unshare_flags & CLONE_FS) &&
1622             (fs && atomic_read(&fs->count) > 1)) {
1623                 *new_fsp = __copy_fs_struct(current->fs);
1624                 if (!*new_fsp)
1625                         return -ENOMEM;
1626         }
1627
1628         return 0;
1629 }
1630
1631 /*
1632  * Unsharing of sighand is not supported yet
1633  */
1634 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1635 {
1636         struct sighand_struct *sigh = current->sighand;
1637
1638         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1639                 return -EINVAL;
1640         else
1641                 return 0;
1642 }
1643
1644 /*
1645  * Unshare vm if it is being shared
1646  */
1647 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1648 {
1649         struct mm_struct *mm = current->mm;
1650
1651         if ((unshare_flags & CLONE_VM) &&
1652             (mm && atomic_read(&mm->mm_users) > 1)) {
1653                 return -EINVAL;
1654         }
1655
1656         return 0;
1657 }
1658
1659 /*
1660  * Unshare file descriptor table if it is being shared
1661  */
1662 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1663 {
1664         struct files_struct *fd = current->files;
1665         int error = 0;
1666
1667         if ((unshare_flags & CLONE_FILES) &&
1668             (fd && atomic_read(&fd->count) > 1)) {
1669                 *new_fdp = dup_fd(fd, &error);
1670                 if (!*new_fdp)
1671                         return error;
1672         }
1673
1674         return 0;
1675 }
1676
1677 /*
1678  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1679  * supported yet
1680  */
1681 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1682 {
1683         if (unshare_flags & CLONE_SYSVSEM)
1684                 return -EINVAL;
1685
1686         return 0;
1687 }
1688
1689 /*
1690  * unshare allows a process to 'unshare' part of the process
1691  * context which was originally shared using clone.  copy_*
1692  * functions used by do_fork() cannot be used here directly
1693  * because they modify an inactive task_struct that is being
1694  * constructed. Here we are modifying the current, active,
1695  * task_struct.
1696  */
1697 asmlinkage long sys_unshare(unsigned long unshare_flags)
1698 {
1699         int err = 0;
1700         struct fs_struct *fs, *new_fs = NULL;
1701         struct sighand_struct *new_sigh = NULL;
1702         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1703         struct files_struct *fd, *new_fd = NULL;
1704         struct sem_undo_list *new_ulist = NULL;
1705         struct nsproxy *new_nsproxy = NULL;
1706
1707         check_unshare_flags(&unshare_flags);
1708
1709         /* Return -EINVAL for all unsupported flags */
1710         err = -EINVAL;
1711         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1712                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1713                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1714                                 CLONE_NEWNET))
1715                 goto bad_unshare_out;
1716
1717         if ((err = unshare_thread(unshare_flags)))
1718                 goto bad_unshare_out;
1719         if ((err = unshare_fs(unshare_flags, &new_fs)))
1720                 goto bad_unshare_cleanup_thread;
1721         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1722                 goto bad_unshare_cleanup_fs;
1723         if ((err = unshare_vm(unshare_flags, &new_mm)))
1724                 goto bad_unshare_cleanup_sigh;
1725         if ((err = unshare_fd(unshare_flags, &new_fd)))
1726                 goto bad_unshare_cleanup_vm;
1727         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1728                 goto bad_unshare_cleanup_fd;
1729         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1730                         new_fs)))
1731                 goto bad_unshare_cleanup_semundo;
1732
1733         if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
1734
1735                 if (new_nsproxy) {
1736                         switch_task_namespaces(current, new_nsproxy);
1737                         new_nsproxy = NULL;
1738                 }
1739
1740                 task_lock(current);
1741
1742                 if (new_fs) {
1743                         fs = current->fs;
1744                         current->fs = new_fs;
1745                         new_fs = fs;
1746                 }
1747
1748                 if (new_mm) {
1749                         mm = current->mm;
1750                         active_mm = current->active_mm;
1751                         current->mm = new_mm;
1752                         current->active_mm = new_mm;
1753                         activate_mm(active_mm, new_mm);
1754                         new_mm = mm;
1755                 }
1756
1757                 if (new_fd) {
1758                         fd = current->files;
1759                         current->files = new_fd;
1760                         new_fd = fd;
1761                 }
1762
1763                 task_unlock(current);
1764         }
1765
1766         if (new_nsproxy)
1767                 put_nsproxy(new_nsproxy);
1768
1769 bad_unshare_cleanup_semundo:
1770 bad_unshare_cleanup_fd:
1771         if (new_fd)
1772                 put_files_struct(new_fd);
1773
1774 bad_unshare_cleanup_vm:
1775         if (new_mm)
1776                 mmput(new_mm);
1777
1778 bad_unshare_cleanup_sigh:
1779         if (new_sigh)
1780                 if (atomic_dec_and_test(&new_sigh->count))
1781                         kmem_cache_free(sighand_cachep, new_sigh);
1782
1783 bad_unshare_cleanup_fs:
1784         if (new_fs)
1785                 put_fs_struct(new_fs);
1786
1787 bad_unshare_cleanup_thread:
1788 bad_unshare_out:
1789         return err;
1790 }