930c51865ab44a4b75e219acf13a8354ff1410a6
[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 #include <linux/proc_fs.h>
54
55 #include <asm/pgtable.h>
56 #include <asm/pgalloc.h>
57 #include <asm/uaccess.h>
58 #include <asm/mmu_context.h>
59 #include <asm/cacheflush.h>
60 #include <asm/tlbflush.h>
61
62 /*
63  * Protected counters by write_lock_irq(&tasklist_lock)
64  */
65 unsigned long total_forks;      /* Handle normal Linux uptimes. */
66 int nr_threads;                 /* The idle threads do not count.. */
67
68 int max_threads;                /* tunable limit on nr_threads */
69
70 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
71
72 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
73
74 int nr_processes(void)
75 {
76         int cpu;
77         int total = 0;
78
79         for_each_online_cpu(cpu)
80                 total += per_cpu(process_counts, cpu);
81
82         return total;
83 }
84
85 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
86 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
87 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
88 static struct kmem_cache *task_struct_cachep;
89 #endif
90
91 /* SLAB cache for signal_struct structures (tsk->signal) */
92 static struct kmem_cache *signal_cachep;
93
94 /* SLAB cache for sighand_struct structures (tsk->sighand) */
95 struct kmem_cache *sighand_cachep;
96
97 /* SLAB cache for files_struct structures (tsk->files) */
98 struct kmem_cache *files_cachep;
99
100 /* SLAB cache for fs_struct structures (tsk->fs) */
101 struct kmem_cache *fs_cachep;
102
103 /* SLAB cache for vm_area_struct structures */
104 struct kmem_cache *vm_area_cachep;
105
106 /* SLAB cache for mm_struct structures (tsk->mm) */
107 static struct kmem_cache *mm_cachep;
108
109 void free_task(struct task_struct *tsk)
110 {
111         prop_local_destroy_single(&tsk->dirties);
112         free_thread_info(tsk->stack);
113         rt_mutex_debug_task_free(tsk);
114         free_task_struct(tsk);
115 }
116 EXPORT_SYMBOL(free_task);
117
118 void __put_task_struct(struct task_struct *tsk)
119 {
120         WARN_ON(!tsk->exit_state);
121         WARN_ON(atomic_read(&tsk->usage));
122         WARN_ON(tsk == current);
123
124         security_task_free(tsk);
125         free_uid(tsk->user);
126         put_group_info(tsk->group_info);
127         delayacct_tsk_free(tsk);
128
129         if (!profile_handoff_task(tsk))
130                 free_task(tsk);
131 }
132
133 void __init fork_init(unsigned long mempages)
134 {
135 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
136 #ifndef ARCH_MIN_TASKALIGN
137 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
138 #endif
139         /* create a slab on which task_structs can be allocated */
140         task_struct_cachep =
141                 kmem_cache_create("task_struct", sizeof(struct task_struct),
142                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
143 #endif
144
145         /*
146          * The default maximum number of threads is set to a safe
147          * value: the thread structures can take up at most half
148          * of memory.
149          */
150         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
151
152         /*
153          * we need to allow at least 20 threads to boot a system
154          */
155         if(max_threads < 20)
156                 max_threads = 20;
157
158         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
159         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
160         init_task.signal->rlim[RLIMIT_SIGPENDING] =
161                 init_task.signal->rlim[RLIMIT_NPROC];
162 }
163
164 static struct task_struct *dup_task_struct(struct task_struct *orig)
165 {
166         struct task_struct *tsk;
167         struct thread_info *ti;
168         int err;
169
170         prepare_to_copy(orig);
171
172         tsk = alloc_task_struct();
173         if (!tsk)
174                 return NULL;
175
176         ti = alloc_thread_info(tsk);
177         if (!ti) {
178                 free_task_struct(tsk);
179                 return NULL;
180         }
181
182         *tsk = *orig;
183         tsk->stack = ti;
184
185         err = prop_local_init_single(&tsk->dirties);
186         if (err) {
187                 free_thread_info(ti);
188                 free_task_struct(tsk);
189                 return NULL;
190         }
191
192         setup_thread_stack(tsk, orig);
193
194 #ifdef CONFIG_CC_STACKPROTECTOR
195         tsk->stack_canary = get_random_int();
196 #endif
197
198         /* One for us, one for whoever does the "release_task()" (usually parent) */
199         atomic_set(&tsk->usage,2);
200         atomic_set(&tsk->fs_excl, 0);
201 #ifdef CONFIG_BLK_DEV_IO_TRACE
202         tsk->btrace_seq = 0;
203 #endif
204         tsk->splice_pipe = NULL;
205         return tsk;
206 }
207
208 #ifdef CONFIG_MMU
209 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
210 {
211         struct vm_area_struct *mpnt, *tmp, **pprev;
212         struct rb_node **rb_link, *rb_parent;
213         int retval;
214         unsigned long charge;
215         struct mempolicy *pol;
216
217         down_write(&oldmm->mmap_sem);
218         flush_cache_dup_mm(oldmm);
219         /*
220          * Not linked in yet - no deadlock potential:
221          */
222         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
223
224         mm->locked_vm = 0;
225         mm->mmap = NULL;
226         mm->mmap_cache = NULL;
227         mm->free_area_cache = oldmm->mmap_base;
228         mm->cached_hole_size = ~0UL;
229         mm->map_count = 0;
230         cpus_clear(mm->cpu_vm_mask);
231         mm->mm_rb = RB_ROOT;
232         rb_link = &mm->mm_rb.rb_node;
233         rb_parent = NULL;
234         pprev = &mm->mmap;
235
236         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
237                 struct file *file;
238
239                 if (mpnt->vm_flags & VM_DONTCOPY) {
240                         long pages = vma_pages(mpnt);
241                         mm->total_vm -= pages;
242                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
243                                                                 -pages);
244                         continue;
245                 }
246                 charge = 0;
247                 if (mpnt->vm_flags & VM_ACCOUNT) {
248                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
249                         if (security_vm_enough_memory(len))
250                                 goto fail_nomem;
251                         charge = len;
252                 }
253                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
254                 if (!tmp)
255                         goto fail_nomem;
256                 *tmp = *mpnt;
257                 pol = mpol_copy(vma_policy(mpnt));
258                 retval = PTR_ERR(pol);
259                 if (IS_ERR(pol))
260                         goto fail_nomem_policy;
261                 vma_set_policy(tmp, pol);
262                 tmp->vm_flags &= ~VM_LOCKED;
263                 tmp->vm_mm = mm;
264                 tmp->vm_next = NULL;
265                 anon_vma_link(tmp);
266                 file = tmp->vm_file;
267                 if (file) {
268                         struct inode *inode = file->f_path.dentry->d_inode;
269                         get_file(file);
270                         if (tmp->vm_flags & VM_DENYWRITE)
271                                 atomic_dec(&inode->i_writecount);
272
273                         /* insert tmp into the share list, just after mpnt */
274                         spin_lock(&file->f_mapping->i_mmap_lock);
275                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
276                         flush_dcache_mmap_lock(file->f_mapping);
277                         vma_prio_tree_add(tmp, mpnt);
278                         flush_dcache_mmap_unlock(file->f_mapping);
279                         spin_unlock(&file->f_mapping->i_mmap_lock);
280                 }
281
282                 /*
283                  * Link in the new vma and copy the page table entries.
284                  */
285                 *pprev = tmp;
286                 pprev = &tmp->vm_next;
287
288                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
289                 rb_link = &tmp->vm_rb.rb_right;
290                 rb_parent = &tmp->vm_rb;
291
292                 mm->map_count++;
293                 retval = copy_page_range(mm, oldmm, mpnt);
294
295                 if (tmp->vm_ops && tmp->vm_ops->open)
296                         tmp->vm_ops->open(tmp);
297
298                 if (retval)
299                         goto out;
300         }
301         /* a new mm has just been created */
302         arch_dup_mmap(oldmm, mm);
303         retval = 0;
304 out:
305         up_write(&mm->mmap_sem);
306         flush_tlb_mm(oldmm);
307         up_write(&oldmm->mmap_sem);
308         return retval;
309 fail_nomem_policy:
310         kmem_cache_free(vm_area_cachep, tmp);
311 fail_nomem:
312         retval = -ENOMEM;
313         vm_unacct_memory(charge);
314         goto out;
315 }
316
317 static inline int mm_alloc_pgd(struct mm_struct * mm)
318 {
319         mm->pgd = pgd_alloc(mm);
320         if (unlikely(!mm->pgd))
321                 return -ENOMEM;
322         return 0;
323 }
324
325 static inline void mm_free_pgd(struct mm_struct * mm)
326 {
327         pgd_free(mm->pgd);
328 }
329 #else
330 #define dup_mmap(mm, oldmm)     (0)
331 #define mm_alloc_pgd(mm)        (0)
332 #define mm_free_pgd(mm)
333 #endif /* CONFIG_MMU */
334
335 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
336
337 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
338 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
339
340 #include <linux/init_task.h>
341
342 static struct mm_struct * mm_init(struct mm_struct * mm)
343 {
344         atomic_set(&mm->mm_users, 1);
345         atomic_set(&mm->mm_count, 1);
346         init_rwsem(&mm->mmap_sem);
347         INIT_LIST_HEAD(&mm->mmlist);
348         mm->flags = (current->mm) ? current->mm->flags
349                                   : MMF_DUMP_FILTER_DEFAULT;
350         mm->core_waiters = 0;
351         mm->nr_ptes = 0;
352         set_mm_counter(mm, file_rss, 0);
353         set_mm_counter(mm, anon_rss, 0);
354         spin_lock_init(&mm->page_table_lock);
355         rwlock_init(&mm->ioctx_list_lock);
356         mm->ioctx_list = NULL;
357         mm->free_area_cache = TASK_UNMAPPED_BASE;
358         mm->cached_hole_size = ~0UL;
359
360         if (likely(!mm_alloc_pgd(mm))) {
361                 mm->def_flags = 0;
362                 return mm;
363         }
364         free_mm(mm);
365         return NULL;
366 }
367
368 /*
369  * Allocate and initialize an mm_struct.
370  */
371 struct mm_struct * mm_alloc(void)
372 {
373         struct mm_struct * mm;
374
375         mm = allocate_mm();
376         if (mm) {
377                 memset(mm, 0, sizeof(*mm));
378                 mm = mm_init(mm);
379         }
380         return mm;
381 }
382
383 /*
384  * Called when the last reference to the mm
385  * is dropped: either by a lazy thread or by
386  * mmput. Free the page directory and the mm.
387  */
388 void fastcall __mmdrop(struct mm_struct *mm)
389 {
390         BUG_ON(mm == &init_mm);
391         mm_free_pgd(mm);
392         destroy_context(mm);
393         free_mm(mm);
394 }
395
396 /*
397  * Decrement the use count and release all resources for an mm.
398  */
399 void mmput(struct mm_struct *mm)
400 {
401         might_sleep();
402
403         if (atomic_dec_and_test(&mm->mm_users)) {
404                 exit_aio(mm);
405                 exit_mmap(mm);
406                 if (!list_empty(&mm->mmlist)) {
407                         spin_lock(&mmlist_lock);
408                         list_del(&mm->mmlist);
409                         spin_unlock(&mmlist_lock);
410                 }
411                 put_swap_token(mm);
412                 mmdrop(mm);
413         }
414 }
415 EXPORT_SYMBOL_GPL(mmput);
416
417 /**
418  * get_task_mm - acquire a reference to the task's mm
419  *
420  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
421  * this kernel workthread has transiently adopted a user mm with use_mm,
422  * to do its AIO) is not set and if so returns a reference to it, after
423  * bumping up the use count.  User must release the mm via mmput()
424  * after use.  Typically used by /proc and ptrace.
425  */
426 struct mm_struct *get_task_mm(struct task_struct *task)
427 {
428         struct mm_struct *mm;
429
430         task_lock(task);
431         mm = task->mm;
432         if (mm) {
433                 if (task->flags & PF_BORROWED_MM)
434                         mm = NULL;
435                 else
436                         atomic_inc(&mm->mm_users);
437         }
438         task_unlock(task);
439         return mm;
440 }
441 EXPORT_SYMBOL_GPL(get_task_mm);
442
443 /* Please note the differences between mmput and mm_release.
444  * mmput is called whenever we stop holding onto a mm_struct,
445  * error success whatever.
446  *
447  * mm_release is called after a mm_struct has been removed
448  * from the current process.
449  *
450  * This difference is important for error handling, when we
451  * only half set up a mm_struct for a new process and need to restore
452  * the old one.  Because we mmput the new mm_struct before
453  * restoring the old one. . .
454  * Eric Biederman 10 January 1998
455  */
456 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
457 {
458         struct completion *vfork_done = tsk->vfork_done;
459
460         /* Get rid of any cached register state */
461         deactivate_mm(tsk, mm);
462
463         /* notify parent sleeping on vfork() */
464         if (vfork_done) {
465                 tsk->vfork_done = NULL;
466                 complete(vfork_done);
467         }
468
469         /*
470          * If we're exiting normally, clear a user-space tid field if
471          * requested.  We leave this alone when dying by signal, to leave
472          * the value intact in a core dump, and to save the unnecessary
473          * trouble otherwise.  Userland only wants this done for a sys_exit.
474          */
475         if (tsk->clear_child_tid
476             && !(tsk->flags & PF_SIGNALED)
477             && atomic_read(&mm->mm_users) > 1) {
478                 u32 __user * tidptr = tsk->clear_child_tid;
479                 tsk->clear_child_tid = NULL;
480
481                 /*
482                  * We don't check the error code - if userspace has
483                  * not set up a proper pointer then tough luck.
484                  */
485                 put_user(0, tidptr);
486                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
487         }
488 }
489
490 /*
491  * Allocate a new mm structure and copy contents from the
492  * mm structure of the passed in task structure.
493  */
494 static struct mm_struct *dup_mm(struct task_struct *tsk)
495 {
496         struct mm_struct *mm, *oldmm = current->mm;
497         int err;
498
499         if (!oldmm)
500                 return NULL;
501
502         mm = allocate_mm();
503         if (!mm)
504                 goto fail_nomem;
505
506         memcpy(mm, oldmm, sizeof(*mm));
507
508         /* Initializing for Swap token stuff */
509         mm->token_priority = 0;
510         mm->last_interval = 0;
511
512         if (!mm_init(mm))
513                 goto fail_nomem;
514
515         if (init_new_context(tsk, mm))
516                 goto fail_nocontext;
517
518         err = dup_mmap(mm, oldmm);
519         if (err)
520                 goto free_pt;
521
522         mm->hiwater_rss = get_mm_rss(mm);
523         mm->hiwater_vm = mm->total_vm;
524
525         return mm;
526
527 free_pt:
528         mmput(mm);
529
530 fail_nomem:
531         return NULL;
532
533 fail_nocontext:
534         /*
535          * If init_new_context() failed, we cannot use mmput() to free the mm
536          * because it calls destroy_context()
537          */
538         mm_free_pgd(mm);
539         free_mm(mm);
540         return NULL;
541 }
542
543 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
544 {
545         struct mm_struct * mm, *oldmm;
546         int retval;
547
548         tsk->min_flt = tsk->maj_flt = 0;
549         tsk->nvcsw = tsk->nivcsw = 0;
550
551         tsk->mm = NULL;
552         tsk->active_mm = NULL;
553
554         /*
555          * Are we cloning a kernel thread?
556          *
557          * We need to steal a active VM for that..
558          */
559         oldmm = current->mm;
560         if (!oldmm)
561                 return 0;
562
563         if (clone_flags & CLONE_VM) {
564                 atomic_inc(&oldmm->mm_users);
565                 mm = oldmm;
566                 goto good_mm;
567         }
568
569         retval = -ENOMEM;
570         mm = dup_mm(tsk);
571         if (!mm)
572                 goto fail_nomem;
573
574 good_mm:
575         /* Initializing for Swap token stuff */
576         mm->token_priority = 0;
577         mm->last_interval = 0;
578
579         tsk->mm = mm;
580         tsk->active_mm = mm;
581         return 0;
582
583 fail_nomem:
584         return retval;
585 }
586
587 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
588 {
589         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
590         /* We don't need to lock fs - think why ;-) */
591         if (fs) {
592                 atomic_set(&fs->count, 1);
593                 rwlock_init(&fs->lock);
594                 fs->umask = old->umask;
595                 read_lock(&old->lock);
596                 fs->rootmnt = mntget(old->rootmnt);
597                 fs->root = dget(old->root);
598                 fs->pwdmnt = mntget(old->pwdmnt);
599                 fs->pwd = dget(old->pwd);
600                 if (old->altroot) {
601                         fs->altrootmnt = mntget(old->altrootmnt);
602                         fs->altroot = dget(old->altroot);
603                 } else {
604                         fs->altrootmnt = NULL;
605                         fs->altroot = NULL;
606                 }
607                 read_unlock(&old->lock);
608         }
609         return fs;
610 }
611
612 struct fs_struct *copy_fs_struct(struct fs_struct *old)
613 {
614         return __copy_fs_struct(old);
615 }
616
617 EXPORT_SYMBOL_GPL(copy_fs_struct);
618
619 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
620 {
621         if (clone_flags & CLONE_FS) {
622                 atomic_inc(&current->fs->count);
623                 return 0;
624         }
625         tsk->fs = __copy_fs_struct(current->fs);
626         if (!tsk->fs)
627                 return -ENOMEM;
628         return 0;
629 }
630
631 static int count_open_files(struct fdtable *fdt)
632 {
633         int size = fdt->max_fds;
634         int i;
635
636         /* Find the last open fd */
637         for (i = size/(8*sizeof(long)); i > 0; ) {
638                 if (fdt->open_fds->fds_bits[--i])
639                         break;
640         }
641         i = (i+1) * 8 * sizeof(long);
642         return i;
643 }
644
645 static struct files_struct *alloc_files(void)
646 {
647         struct files_struct *newf;
648         struct fdtable *fdt;
649
650         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
651         if (!newf)
652                 goto out;
653
654         atomic_set(&newf->count, 1);
655
656         spin_lock_init(&newf->file_lock);
657         newf->next_fd = 0;
658         fdt = &newf->fdtab;
659         fdt->max_fds = NR_OPEN_DEFAULT;
660         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
661         fdt->open_fds = (fd_set *)&newf->open_fds_init;
662         fdt->fd = &newf->fd_array[0];
663         INIT_RCU_HEAD(&fdt->rcu);
664         fdt->next = NULL;
665         rcu_assign_pointer(newf->fdt, fdt);
666 out:
667         return newf;
668 }
669
670 /*
671  * Allocate a new files structure and copy contents from the
672  * passed in files structure.
673  * errorp will be valid only when the returned files_struct is NULL.
674  */
675 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
676 {
677         struct files_struct *newf;
678         struct file **old_fds, **new_fds;
679         int open_files, size, i;
680         struct fdtable *old_fdt, *new_fdt;
681
682         *errorp = -ENOMEM;
683         newf = alloc_files();
684         if (!newf)
685                 goto out;
686
687         spin_lock(&oldf->file_lock);
688         old_fdt = files_fdtable(oldf);
689         new_fdt = files_fdtable(newf);
690         open_files = count_open_files(old_fdt);
691
692         /*
693          * Check whether we need to allocate a larger fd array and fd set.
694          * Note: we're not a clone task, so the open count won't change.
695          */
696         if (open_files > new_fdt->max_fds) {
697                 new_fdt->max_fds = 0;
698                 spin_unlock(&oldf->file_lock);
699                 spin_lock(&newf->file_lock);
700                 *errorp = expand_files(newf, open_files-1);
701                 spin_unlock(&newf->file_lock);
702                 if (*errorp < 0)
703                         goto out_release;
704                 new_fdt = files_fdtable(newf);
705                 /*
706                  * Reacquire the oldf lock and a pointer to its fd table
707                  * who knows it may have a new bigger fd table. We need
708                  * the latest pointer.
709                  */
710                 spin_lock(&oldf->file_lock);
711                 old_fdt = files_fdtable(oldf);
712         }
713
714         old_fds = old_fdt->fd;
715         new_fds = new_fdt->fd;
716
717         memcpy(new_fdt->open_fds->fds_bits,
718                 old_fdt->open_fds->fds_bits, open_files/8);
719         memcpy(new_fdt->close_on_exec->fds_bits,
720                 old_fdt->close_on_exec->fds_bits, open_files/8);
721
722         for (i = open_files; i != 0; i--) {
723                 struct file *f = *old_fds++;
724                 if (f) {
725                         get_file(f);
726                 } else {
727                         /*
728                          * The fd may be claimed in the fd bitmap but not yet
729                          * instantiated in the files array if a sibling thread
730                          * is partway through open().  So make sure that this
731                          * fd is available to the new process.
732                          */
733                         FD_CLR(open_files - i, new_fdt->open_fds);
734                 }
735                 rcu_assign_pointer(*new_fds++, f);
736         }
737         spin_unlock(&oldf->file_lock);
738
739         /* compute the remainder to be cleared */
740         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
741
742         /* This is long word aligned thus could use a optimized version */
743         memset(new_fds, 0, size);
744
745         if (new_fdt->max_fds > open_files) {
746                 int left = (new_fdt->max_fds-open_files)/8;
747                 int start = open_files / (8 * sizeof(unsigned long));
748
749                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
750                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
751         }
752
753         return newf;
754
755 out_release:
756         kmem_cache_free(files_cachep, newf);
757 out:
758         return NULL;
759 }
760
761 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
762 {
763         struct files_struct *oldf, *newf;
764         int error = 0;
765
766         /*
767          * A background process may not have any files ...
768          */
769         oldf = current->files;
770         if (!oldf)
771                 goto out;
772
773         if (clone_flags & CLONE_FILES) {
774                 atomic_inc(&oldf->count);
775                 goto out;
776         }
777
778         /*
779          * Note: we may be using current for both targets (See exec.c)
780          * This works because we cache current->files (old) as oldf. Don't
781          * break this.
782          */
783         tsk->files = NULL;
784         newf = dup_fd(oldf, &error);
785         if (!newf)
786                 goto out;
787
788         tsk->files = newf;
789         error = 0;
790 out:
791         return error;
792 }
793
794 /*
795  *      Helper to unshare the files of the current task.
796  *      We don't want to expose copy_files internals to
797  *      the exec layer of the kernel.
798  */
799
800 int unshare_files(void)
801 {
802         struct files_struct *files  = current->files;
803         int rc;
804
805         BUG_ON(!files);
806
807         /* This can race but the race causes us to copy when we don't
808            need to and drop the copy */
809         if(atomic_read(&files->count) == 1)
810         {
811                 atomic_inc(&files->count);
812                 return 0;
813         }
814         rc = copy_files(0, current);
815         if(rc)
816                 current->files = files;
817         return rc;
818 }
819
820 EXPORT_SYMBOL(unshare_files);
821
822 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
823 {
824         struct sighand_struct *sig;
825
826         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
827                 atomic_inc(&current->sighand->count);
828                 return 0;
829         }
830         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
831         rcu_assign_pointer(tsk->sighand, sig);
832         if (!sig)
833                 return -ENOMEM;
834         atomic_set(&sig->count, 1);
835         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
836         return 0;
837 }
838
839 void __cleanup_sighand(struct sighand_struct *sighand)
840 {
841         if (atomic_dec_and_test(&sighand->count))
842                 kmem_cache_free(sighand_cachep, sighand);
843 }
844
845 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
846 {
847         struct signal_struct *sig;
848         int ret;
849
850         if (clone_flags & CLONE_THREAD) {
851                 atomic_inc(&current->signal->count);
852                 atomic_inc(&current->signal->live);
853                 return 0;
854         }
855         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
856         tsk->signal = sig;
857         if (!sig)
858                 return -ENOMEM;
859
860         ret = copy_thread_group_keys(tsk);
861         if (ret < 0) {
862                 kmem_cache_free(signal_cachep, sig);
863                 return ret;
864         }
865
866         atomic_set(&sig->count, 1);
867         atomic_set(&sig->live, 1);
868         init_waitqueue_head(&sig->wait_chldexit);
869         sig->flags = 0;
870         sig->group_exit_code = 0;
871         sig->group_exit_task = NULL;
872         sig->group_stop_count = 0;
873         sig->curr_target = NULL;
874         init_sigpending(&sig->shared_pending);
875         INIT_LIST_HEAD(&sig->posix_timers);
876
877         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
878         sig->it_real_incr.tv64 = 0;
879         sig->real_timer.function = it_real_fn;
880         sig->tsk = tsk;
881
882         sig->it_virt_expires = cputime_zero;
883         sig->it_virt_incr = cputime_zero;
884         sig->it_prof_expires = cputime_zero;
885         sig->it_prof_incr = cputime_zero;
886
887         sig->leader = 0;        /* session leadership doesn't inherit */
888         sig->tty_old_pgrp = NULL;
889
890         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
891         sig->gtime = cputime_zero;
892         sig->cgtime = cputime_zero;
893         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
894         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
895         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
896         sig->sum_sched_runtime = 0;
897         INIT_LIST_HEAD(&sig->cpu_timers[0]);
898         INIT_LIST_HEAD(&sig->cpu_timers[1]);
899         INIT_LIST_HEAD(&sig->cpu_timers[2]);
900         taskstats_tgid_init(sig);
901
902         task_lock(current->group_leader);
903         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
904         task_unlock(current->group_leader);
905
906         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
907                 /*
908                  * New sole thread in the process gets an expiry time
909                  * of the whole CPU time limit.
910                  */
911                 tsk->it_prof_expires =
912                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
913         }
914         acct_init_pacct(&sig->pacct);
915
916         tty_audit_fork(sig);
917
918         return 0;
919 }
920
921 void __cleanup_signal(struct signal_struct *sig)
922 {
923         exit_thread_group_keys(sig);
924         kmem_cache_free(signal_cachep, sig);
925 }
926
927 static void cleanup_signal(struct task_struct *tsk)
928 {
929         struct signal_struct *sig = tsk->signal;
930
931         atomic_dec(&sig->live);
932
933         if (atomic_dec_and_test(&sig->count))
934                 __cleanup_signal(sig);
935 }
936
937 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
938 {
939         unsigned long new_flags = p->flags;
940
941         new_flags &= ~PF_SUPERPRIV;
942         new_flags |= PF_FORKNOEXEC;
943         if (!(clone_flags & CLONE_PTRACE))
944                 p->ptrace = 0;
945         p->flags = new_flags;
946         clear_freeze_flag(p);
947 }
948
949 asmlinkage long sys_set_tid_address(int __user *tidptr)
950 {
951         current->clear_child_tid = tidptr;
952
953         return task_pid_vnr(current);
954 }
955
956 static void rt_mutex_init_task(struct task_struct *p)
957 {
958         spin_lock_init(&p->pi_lock);
959 #ifdef CONFIG_RT_MUTEXES
960         plist_head_init(&p->pi_waiters, &p->pi_lock);
961         p->pi_blocked_on = NULL;
962 #endif
963 }
964
965 /*
966  * This creates a new process as a copy of the old one,
967  * but does not actually start it yet.
968  *
969  * It copies the registers, and all the appropriate
970  * parts of the process environment (as per the clone
971  * flags). The actual kick-off is left to the caller.
972  */
973 static struct task_struct *copy_process(unsigned long clone_flags,
974                                         unsigned long stack_start,
975                                         struct pt_regs *regs,
976                                         unsigned long stack_size,
977                                         int __user *child_tidptr,
978                                         struct pid *pid)
979 {
980         int retval;
981         struct task_struct *p;
982         int cgroup_callbacks_done = 0;
983
984         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
985                 return ERR_PTR(-EINVAL);
986
987         /*
988          * Thread groups must share signals as well, and detached threads
989          * can only be started up within the thread group.
990          */
991         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
992                 return ERR_PTR(-EINVAL);
993
994         /*
995          * Shared signal handlers imply shared VM. By way of the above,
996          * thread groups also imply shared VM. Blocking this case allows
997          * for various simplifications in other code.
998          */
999         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1000                 return ERR_PTR(-EINVAL);
1001
1002         retval = security_task_create(clone_flags);
1003         if (retval)
1004                 goto fork_out;
1005
1006         retval = -ENOMEM;
1007         p = dup_task_struct(current);
1008         if (!p)
1009                 goto fork_out;
1010
1011         rt_mutex_init_task(p);
1012
1013 #ifdef CONFIG_TRACE_IRQFLAGS
1014         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1015         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1016 #endif
1017         retval = -EAGAIN;
1018         if (atomic_read(&p->user->processes) >=
1019                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1020                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1021                     p->user != current->nsproxy->user_ns->root_user)
1022                         goto bad_fork_free;
1023         }
1024
1025         atomic_inc(&p->user->__count);
1026         atomic_inc(&p->user->processes);
1027         get_group_info(p->group_info);
1028
1029         /*
1030          * If multiple threads are within copy_process(), then this check
1031          * triggers too late. This doesn't hurt, the check is only there
1032          * to stop root fork bombs.
1033          */
1034         if (nr_threads >= max_threads)
1035                 goto bad_fork_cleanup_count;
1036
1037         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1038                 goto bad_fork_cleanup_count;
1039
1040         if (p->binfmt && !try_module_get(p->binfmt->module))
1041                 goto bad_fork_cleanup_put_domain;
1042
1043         p->did_exec = 0;
1044         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1045         copy_flags(clone_flags, p);
1046         INIT_LIST_HEAD(&p->children);
1047         INIT_LIST_HEAD(&p->sibling);
1048         p->vfork_done = NULL;
1049         spin_lock_init(&p->alloc_lock);
1050
1051         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1052         init_sigpending(&p->pending);
1053
1054         p->utime = cputime_zero;
1055         p->stime = cputime_zero;
1056         p->gtime = cputime_zero;
1057         p->utimescaled = cputime_zero;
1058         p->stimescaled = cputime_zero;
1059         p->prev_utime = cputime_zero;
1060         p->prev_stime = cputime_zero;
1061
1062 #ifdef CONFIG_DETECT_SOFTLOCKUP
1063         p->last_switch_count = 0;
1064         p->last_switch_timestamp = 0;
1065 #endif
1066
1067 #ifdef CONFIG_TASK_XACCT
1068         p->rchar = 0;           /* I/O counter: bytes read */
1069         p->wchar = 0;           /* I/O counter: bytes written */
1070         p->syscr = 0;           /* I/O counter: read syscalls */
1071         p->syscw = 0;           /* I/O counter: write syscalls */
1072 #endif
1073         task_io_accounting_init(p);
1074         acct_clear_integrals(p);
1075
1076         p->it_virt_expires = cputime_zero;
1077         p->it_prof_expires = cputime_zero;
1078         p->it_sched_expires = 0;
1079         INIT_LIST_HEAD(&p->cpu_timers[0]);
1080         INIT_LIST_HEAD(&p->cpu_timers[1]);
1081         INIT_LIST_HEAD(&p->cpu_timers[2]);
1082
1083         p->lock_depth = -1;             /* -1 = no lock */
1084         do_posix_clock_monotonic_gettime(&p->start_time);
1085         p->real_start_time = p->start_time;
1086         monotonic_to_bootbased(&p->real_start_time);
1087 #ifdef CONFIG_SECURITY
1088         p->security = NULL;
1089 #endif
1090         p->io_context = NULL;
1091         p->audit_context = NULL;
1092         cgroup_fork(p);
1093 #ifdef CONFIG_NUMA
1094         p->mempolicy = mpol_copy(p->mempolicy);
1095         if (IS_ERR(p->mempolicy)) {
1096                 retval = PTR_ERR(p->mempolicy);
1097                 p->mempolicy = NULL;
1098                 goto bad_fork_cleanup_cgroup;
1099         }
1100         mpol_fix_fork_child_flag(p);
1101 #endif
1102 #ifdef CONFIG_TRACE_IRQFLAGS
1103         p->irq_events = 0;
1104 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1105         p->hardirqs_enabled = 1;
1106 #else
1107         p->hardirqs_enabled = 0;
1108 #endif
1109         p->hardirq_enable_ip = 0;
1110         p->hardirq_enable_event = 0;
1111         p->hardirq_disable_ip = _THIS_IP_;
1112         p->hardirq_disable_event = 0;
1113         p->softirqs_enabled = 1;
1114         p->softirq_enable_ip = _THIS_IP_;
1115         p->softirq_enable_event = 0;
1116         p->softirq_disable_ip = 0;
1117         p->softirq_disable_event = 0;
1118         p->hardirq_context = 0;
1119         p->softirq_context = 0;
1120 #endif
1121 #ifdef CONFIG_LOCKDEP
1122         p->lockdep_depth = 0; /* no locks held yet */
1123         p->curr_chain_key = 0;
1124         p->lockdep_recursion = 0;
1125 #endif
1126
1127 #ifdef CONFIG_DEBUG_MUTEXES
1128         p->blocked_on = NULL; /* not blocked yet */
1129 #endif
1130
1131         /* Perform scheduler related setup. Assign this task to a CPU. */
1132         sched_fork(p, clone_flags);
1133
1134         if ((retval = security_task_alloc(p)))
1135                 goto bad_fork_cleanup_policy;
1136         if ((retval = audit_alloc(p)))
1137                 goto bad_fork_cleanup_security;
1138         /* copy all the process information */
1139         if ((retval = copy_semundo(clone_flags, p)))
1140                 goto bad_fork_cleanup_audit;
1141         if ((retval = copy_files(clone_flags, p)))
1142                 goto bad_fork_cleanup_semundo;
1143         if ((retval = copy_fs(clone_flags, p)))
1144                 goto bad_fork_cleanup_files;
1145         if ((retval = copy_sighand(clone_flags, p)))
1146                 goto bad_fork_cleanup_fs;
1147         if ((retval = copy_signal(clone_flags, p)))
1148                 goto bad_fork_cleanup_sighand;
1149         if ((retval = copy_mm(clone_flags, p)))
1150                 goto bad_fork_cleanup_signal;
1151         if ((retval = copy_keys(clone_flags, p)))
1152                 goto bad_fork_cleanup_mm;
1153         if ((retval = copy_namespaces(clone_flags, p)))
1154                 goto bad_fork_cleanup_keys;
1155         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1156         if (retval)
1157                 goto bad_fork_cleanup_namespaces;
1158
1159         if (pid != &init_struct_pid) {
1160                 retval = -ENOMEM;
1161                 pid = alloc_pid(task_active_pid_ns(p));
1162                 if (!pid)
1163                         goto bad_fork_cleanup_namespaces;
1164
1165                 if (clone_flags & CLONE_NEWPID) {
1166                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1167                         if (retval < 0)
1168                                 goto bad_fork_free_pid;
1169                 }
1170         }
1171
1172         p->pid = pid_nr(pid);
1173         p->tgid = p->pid;
1174         if (clone_flags & CLONE_THREAD)
1175                 p->tgid = current->tgid;
1176
1177         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1178         /*
1179          * Clear TID on mm_release()?
1180          */
1181         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1182 #ifdef CONFIG_FUTEX
1183         p->robust_list = NULL;
1184 #ifdef CONFIG_COMPAT
1185         p->compat_robust_list = NULL;
1186 #endif
1187         INIT_LIST_HEAD(&p->pi_state_list);
1188         p->pi_state_cache = NULL;
1189 #endif
1190         /*
1191          * sigaltstack should be cleared when sharing the same VM
1192          */
1193         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1194                 p->sas_ss_sp = p->sas_ss_size = 0;
1195
1196         /*
1197          * Syscall tracing should be turned off in the child regardless
1198          * of CLONE_PTRACE.
1199          */
1200         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1201 #ifdef TIF_SYSCALL_EMU
1202         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1203 #endif
1204
1205         /* Our parent execution domain becomes current domain
1206            These must match for thread signalling to apply */
1207         p->parent_exec_id = p->self_exec_id;
1208
1209         /* ok, now we should be set up.. */
1210         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1211         p->pdeath_signal = 0;
1212         p->exit_state = 0;
1213
1214         /*
1215          * Ok, make it visible to the rest of the system.
1216          * We dont wake it up yet.
1217          */
1218         p->group_leader = p;
1219         INIT_LIST_HEAD(&p->thread_group);
1220         INIT_LIST_HEAD(&p->ptrace_children);
1221         INIT_LIST_HEAD(&p->ptrace_list);
1222
1223         /* Now that the task is set up, run cgroup callbacks if
1224          * necessary. We need to run them before the task is visible
1225          * on the tasklist. */
1226         cgroup_fork_callbacks(p);
1227         cgroup_callbacks_done = 1;
1228
1229         /* Need tasklist lock for parent etc handling! */
1230         write_lock_irq(&tasklist_lock);
1231
1232         /* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
1233         p->ioprio = current->ioprio;
1234
1235         /*
1236          * The task hasn't been attached yet, so its cpus_allowed mask will
1237          * not be changed, nor will its assigned CPU.
1238          *
1239          * The cpus_allowed mask of the parent may have changed after it was
1240          * copied first time - so re-copy it here, then check the child's CPU
1241          * to ensure it is on a valid CPU (and if not, just force it back to
1242          * parent's CPU). This avoids alot of nasty races.
1243          */
1244         p->cpus_allowed = current->cpus_allowed;
1245         p->nr_cpus_allowed = current->nr_cpus_allowed;
1246         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1247                         !cpu_online(task_cpu(p))))
1248                 set_task_cpu(p, smp_processor_id());
1249
1250         /* CLONE_PARENT re-uses the old parent */
1251         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1252                 p->real_parent = current->real_parent;
1253         else
1254                 p->real_parent = current;
1255         p->parent = p->real_parent;
1256
1257         spin_lock(&current->sighand->siglock);
1258
1259         /*
1260          * Process group and session signals need to be delivered to just the
1261          * parent before the fork or both the parent and the child after the
1262          * fork. Restart if a signal comes in before we add the new process to
1263          * it's process group.
1264          * A fatal signal pending means that current will exit, so the new
1265          * thread can't slip out of an OOM kill (or normal SIGKILL).
1266          */
1267         recalc_sigpending();
1268         if (signal_pending(current)) {
1269                 spin_unlock(&current->sighand->siglock);
1270                 write_unlock_irq(&tasklist_lock);
1271                 retval = -ERESTARTNOINTR;
1272                 goto bad_fork_free_pid;
1273         }
1274
1275         if (clone_flags & CLONE_THREAD) {
1276                 p->group_leader = current->group_leader;
1277                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1278
1279                 if (!cputime_eq(current->signal->it_virt_expires,
1280                                 cputime_zero) ||
1281                     !cputime_eq(current->signal->it_prof_expires,
1282                                 cputime_zero) ||
1283                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1284                     !list_empty(&current->signal->cpu_timers[0]) ||
1285                     !list_empty(&current->signal->cpu_timers[1]) ||
1286                     !list_empty(&current->signal->cpu_timers[2])) {
1287                         /*
1288                          * Have child wake up on its first tick to check
1289                          * for process CPU timers.
1290                          */
1291                         p->it_prof_expires = jiffies_to_cputime(1);
1292                 }
1293         }
1294
1295         if (likely(p->pid)) {
1296                 add_parent(p);
1297                 if (unlikely(p->ptrace & PT_PTRACED))
1298                         __ptrace_link(p, current->parent);
1299
1300                 if (thread_group_leader(p)) {
1301                         if (clone_flags & CLONE_NEWPID)
1302                                 p->nsproxy->pid_ns->child_reaper = p;
1303
1304                         p->signal->tty = current->signal->tty;
1305                         set_task_pgrp(p, task_pgrp_nr(current));
1306                         set_task_session(p, task_session_nr(current));
1307                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1308                         attach_pid(p, PIDTYPE_SID, task_session(current));
1309                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1310                         __get_cpu_var(process_counts)++;
1311                 }
1312                 attach_pid(p, PIDTYPE_PID, pid);
1313                 nr_threads++;
1314         }
1315
1316         total_forks++;
1317         spin_unlock(&current->sighand->siglock);
1318         write_unlock_irq(&tasklist_lock);
1319         proc_fork_connector(p);
1320         cgroup_post_fork(p);
1321         return p;
1322
1323 bad_fork_free_pid:
1324         if (pid != &init_struct_pid)
1325                 free_pid(pid);
1326 bad_fork_cleanup_namespaces:
1327         exit_task_namespaces(p);
1328 bad_fork_cleanup_keys:
1329         exit_keys(p);
1330 bad_fork_cleanup_mm:
1331         if (p->mm)
1332                 mmput(p->mm);
1333 bad_fork_cleanup_signal:
1334         cleanup_signal(p);
1335 bad_fork_cleanup_sighand:
1336         __cleanup_sighand(p->sighand);
1337 bad_fork_cleanup_fs:
1338         exit_fs(p); /* blocking */
1339 bad_fork_cleanup_files:
1340         exit_files(p); /* blocking */
1341 bad_fork_cleanup_semundo:
1342         exit_sem(p);
1343 bad_fork_cleanup_audit:
1344         audit_free(p);
1345 bad_fork_cleanup_security:
1346         security_task_free(p);
1347 bad_fork_cleanup_policy:
1348 #ifdef CONFIG_NUMA
1349         mpol_free(p->mempolicy);
1350 bad_fork_cleanup_cgroup:
1351 #endif
1352         cgroup_exit(p, cgroup_callbacks_done);
1353         delayacct_tsk_free(p);
1354         if (p->binfmt)
1355                 module_put(p->binfmt->module);
1356 bad_fork_cleanup_put_domain:
1357         module_put(task_thread_info(p)->exec_domain->module);
1358 bad_fork_cleanup_count:
1359         put_group_info(p->group_info);
1360         atomic_dec(&p->user->processes);
1361         free_uid(p->user);
1362 bad_fork_free:
1363         free_task(p);
1364 fork_out:
1365         return ERR_PTR(retval);
1366 }
1367
1368 noinline struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1369 {
1370         memset(regs, 0, sizeof(struct pt_regs));
1371         return regs;
1372 }
1373
1374 struct task_struct * __cpuinit fork_idle(int cpu)
1375 {
1376         struct task_struct *task;
1377         struct pt_regs regs;
1378
1379         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1380                                 &init_struct_pid);
1381         if (!IS_ERR(task))
1382                 init_idle(task, cpu);
1383
1384         return task;
1385 }
1386
1387 static int fork_traceflag(unsigned clone_flags)
1388 {
1389         if (clone_flags & CLONE_UNTRACED)
1390                 return 0;
1391         else if (clone_flags & CLONE_VFORK) {
1392                 if (current->ptrace & PT_TRACE_VFORK)
1393                         return PTRACE_EVENT_VFORK;
1394         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1395                 if (current->ptrace & PT_TRACE_CLONE)
1396                         return PTRACE_EVENT_CLONE;
1397         } else if (current->ptrace & PT_TRACE_FORK)
1398                 return PTRACE_EVENT_FORK;
1399
1400         return 0;
1401 }
1402
1403 /*
1404  *  Ok, this is the main fork-routine.
1405  *
1406  * It copies the process, and if successful kick-starts
1407  * it and waits for it to finish using the VM if required.
1408  */
1409 long do_fork(unsigned long clone_flags,
1410               unsigned long stack_start,
1411               struct pt_regs *regs,
1412               unsigned long stack_size,
1413               int __user *parent_tidptr,
1414               int __user *child_tidptr)
1415 {
1416         struct task_struct *p;
1417         int trace = 0;
1418         long nr;
1419
1420         if (unlikely(current->ptrace)) {
1421                 trace = fork_traceflag (clone_flags);
1422                 if (trace)
1423                         clone_flags |= CLONE_PTRACE;
1424         }
1425
1426         p = copy_process(clone_flags, stack_start, regs, stack_size,
1427                         child_tidptr, NULL);
1428         /*
1429          * Do this prior waking up the new thread - the thread pointer
1430          * might get invalid after that point, if the thread exits quickly.
1431          */
1432         if (!IS_ERR(p)) {
1433                 struct completion vfork;
1434
1435                 /*
1436                  * this is enough to call pid_nr_ns here, but this if
1437                  * improves optimisation of regular fork()
1438                  */
1439                 nr = (clone_flags & CLONE_NEWPID) ?
1440                         task_pid_nr_ns(p, current->nsproxy->pid_ns) :
1441                                 task_pid_vnr(p);
1442
1443                 if (clone_flags & CLONE_PARENT_SETTID)
1444                         put_user(nr, parent_tidptr);
1445
1446                 if (clone_flags & CLONE_VFORK) {
1447                         p->vfork_done = &vfork;
1448                         init_completion(&vfork);
1449                 }
1450
1451                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1452                         /*
1453                          * We'll start up with an immediate SIGSTOP.
1454                          */
1455                         sigaddset(&p->pending.signal, SIGSTOP);
1456                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1457                 }
1458
1459                 if (!(clone_flags & CLONE_STOPPED))
1460                         wake_up_new_task(p, clone_flags);
1461                 else
1462                         p->state = TASK_STOPPED;
1463
1464                 if (unlikely (trace)) {
1465                         current->ptrace_message = nr;
1466                         ptrace_notify ((trace << 8) | SIGTRAP);
1467                 }
1468
1469                 if (clone_flags & CLONE_VFORK) {
1470                         freezer_do_not_count();
1471                         wait_for_completion(&vfork);
1472                         freezer_count();
1473                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1474                                 current->ptrace_message = nr;
1475                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1476                         }
1477                 }
1478         } else {
1479                 nr = PTR_ERR(p);
1480         }
1481         return nr;
1482 }
1483
1484 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1485 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1486 #endif
1487
1488 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1489 {
1490         struct sighand_struct *sighand = data;
1491
1492         spin_lock_init(&sighand->siglock);
1493         init_waitqueue_head(&sighand->signalfd_wqh);
1494 }
1495
1496 void __init proc_caches_init(void)
1497 {
1498         sighand_cachep = kmem_cache_create("sighand_cache",
1499                         sizeof(struct sighand_struct), 0,
1500                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1501                         sighand_ctor);
1502         signal_cachep = kmem_cache_create("signal_cache",
1503                         sizeof(struct signal_struct), 0,
1504                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1505         files_cachep = kmem_cache_create("files_cache",
1506                         sizeof(struct files_struct), 0,
1507                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1508         fs_cachep = kmem_cache_create("fs_cache",
1509                         sizeof(struct fs_struct), 0,
1510                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1511         vm_area_cachep = kmem_cache_create("vm_area_struct",
1512                         sizeof(struct vm_area_struct), 0,
1513                         SLAB_PANIC, NULL);
1514         mm_cachep = kmem_cache_create("mm_struct",
1515                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1516                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1517 }
1518
1519 /*
1520  * Check constraints on flags passed to the unshare system call and
1521  * force unsharing of additional process context as appropriate.
1522  */
1523 static void check_unshare_flags(unsigned long *flags_ptr)
1524 {
1525         /*
1526          * If unsharing a thread from a thread group, must also
1527          * unshare vm.
1528          */
1529         if (*flags_ptr & CLONE_THREAD)
1530                 *flags_ptr |= CLONE_VM;
1531
1532         /*
1533          * If unsharing vm, must also unshare signal handlers.
1534          */
1535         if (*flags_ptr & CLONE_VM)
1536                 *flags_ptr |= CLONE_SIGHAND;
1537
1538         /*
1539          * If unsharing signal handlers and the task was created
1540          * using CLONE_THREAD, then must unshare the thread
1541          */
1542         if ((*flags_ptr & CLONE_SIGHAND) &&
1543             (atomic_read(&current->signal->count) > 1))
1544                 *flags_ptr |= CLONE_THREAD;
1545
1546         /*
1547          * If unsharing namespace, must also unshare filesystem information.
1548          */
1549         if (*flags_ptr & CLONE_NEWNS)
1550                 *flags_ptr |= CLONE_FS;
1551 }
1552
1553 /*
1554  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1555  */
1556 static int unshare_thread(unsigned long unshare_flags)
1557 {
1558         if (unshare_flags & CLONE_THREAD)
1559                 return -EINVAL;
1560
1561         return 0;
1562 }
1563
1564 /*
1565  * Unshare the filesystem structure if it is being shared
1566  */
1567 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1568 {
1569         struct fs_struct *fs = current->fs;
1570
1571         if ((unshare_flags & CLONE_FS) &&
1572             (fs && atomic_read(&fs->count) > 1)) {
1573                 *new_fsp = __copy_fs_struct(current->fs);
1574                 if (!*new_fsp)
1575                         return -ENOMEM;
1576         }
1577
1578         return 0;
1579 }
1580
1581 /*
1582  * Unsharing of sighand is not supported yet
1583  */
1584 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1585 {
1586         struct sighand_struct *sigh = current->sighand;
1587
1588         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1589                 return -EINVAL;
1590         else
1591                 return 0;
1592 }
1593
1594 /*
1595  * Unshare vm if it is being shared
1596  */
1597 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1598 {
1599         struct mm_struct *mm = current->mm;
1600
1601         if ((unshare_flags & CLONE_VM) &&
1602             (mm && atomic_read(&mm->mm_users) > 1)) {
1603                 return -EINVAL;
1604         }
1605
1606         return 0;
1607 }
1608
1609 /*
1610  * Unshare file descriptor table if it is being shared
1611  */
1612 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1613 {
1614         struct files_struct *fd = current->files;
1615         int error = 0;
1616
1617         if ((unshare_flags & CLONE_FILES) &&
1618             (fd && atomic_read(&fd->count) > 1)) {
1619                 *new_fdp = dup_fd(fd, &error);
1620                 if (!*new_fdp)
1621                         return error;
1622         }
1623
1624         return 0;
1625 }
1626
1627 /*
1628  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1629  * supported yet
1630  */
1631 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1632 {
1633         if (unshare_flags & CLONE_SYSVSEM)
1634                 return -EINVAL;
1635
1636         return 0;
1637 }
1638
1639 /*
1640  * unshare allows a process to 'unshare' part of the process
1641  * context which was originally shared using clone.  copy_*
1642  * functions used by do_fork() cannot be used here directly
1643  * because they modify an inactive task_struct that is being
1644  * constructed. Here we are modifying the current, active,
1645  * task_struct.
1646  */
1647 asmlinkage long sys_unshare(unsigned long unshare_flags)
1648 {
1649         int err = 0;
1650         struct fs_struct *fs, *new_fs = NULL;
1651         struct sighand_struct *new_sigh = NULL;
1652         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1653         struct files_struct *fd, *new_fd = NULL;
1654         struct sem_undo_list *new_ulist = NULL;
1655         struct nsproxy *new_nsproxy = NULL;
1656
1657         check_unshare_flags(&unshare_flags);
1658
1659         /* Return -EINVAL for all unsupported flags */
1660         err = -EINVAL;
1661         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1662                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1663                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1664                                 CLONE_NEWNET))
1665                 goto bad_unshare_out;
1666
1667         if ((err = unshare_thread(unshare_flags)))
1668                 goto bad_unshare_out;
1669         if ((err = unshare_fs(unshare_flags, &new_fs)))
1670                 goto bad_unshare_cleanup_thread;
1671         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1672                 goto bad_unshare_cleanup_fs;
1673         if ((err = unshare_vm(unshare_flags, &new_mm)))
1674                 goto bad_unshare_cleanup_sigh;
1675         if ((err = unshare_fd(unshare_flags, &new_fd)))
1676                 goto bad_unshare_cleanup_vm;
1677         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1678                 goto bad_unshare_cleanup_fd;
1679         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1680                         new_fs)))
1681                 goto bad_unshare_cleanup_semundo;
1682
1683         if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
1684
1685                 if (new_nsproxy) {
1686                         switch_task_namespaces(current, new_nsproxy);
1687                         new_nsproxy = NULL;
1688                 }
1689
1690                 task_lock(current);
1691
1692                 if (new_fs) {
1693                         fs = current->fs;
1694                         current->fs = new_fs;
1695                         new_fs = fs;
1696                 }
1697
1698                 if (new_mm) {
1699                         mm = current->mm;
1700                         active_mm = current->active_mm;
1701                         current->mm = new_mm;
1702                         current->active_mm = new_mm;
1703                         activate_mm(active_mm, new_mm);
1704                         new_mm = mm;
1705                 }
1706
1707                 if (new_fd) {
1708                         fd = current->files;
1709                         current->files = new_fd;
1710                         new_fd = fd;
1711                 }
1712
1713                 task_unlock(current);
1714         }
1715
1716         if (new_nsproxy)
1717                 put_nsproxy(new_nsproxy);
1718
1719 bad_unshare_cleanup_semundo:
1720 bad_unshare_cleanup_fd:
1721         if (new_fd)
1722                 put_files_struct(new_fd);
1723
1724 bad_unshare_cleanup_vm:
1725         if (new_mm)
1726                 mmput(new_mm);
1727
1728 bad_unshare_cleanup_sigh:
1729         if (new_sigh)
1730                 if (atomic_dec_and_test(&new_sigh->count))
1731                         kmem_cache_free(sighand_cachep, new_sigh);
1732
1733 bad_unshare_cleanup_fs:
1734         if (new_fs)
1735                 put_fs_struct(new_fs);
1736
1737 bad_unshare_cleanup_thread:
1738 bad_unshare_out:
1739         return err;
1740 }