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