capabilities: introduce per-process capability bounding set
[linux-3.10.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 #include <linux/blkdev.h>
55
56 #include <asm/pgtable.h>
57 #include <asm/pgalloc.h>
58 #include <asm/uaccess.h>
59 #include <asm/mmu_context.h>
60 #include <asm/cacheflush.h>
61 #include <asm/tlbflush.h>
62
63 /*
64  * Protected counters by write_lock_irq(&tasklist_lock)
65  */
66 unsigned long total_forks;      /* Handle normal Linux uptimes. */
67 int nr_threads;                 /* The idle threads do not count.. */
68
69 int max_threads;                /* tunable limit on nr_threads */
70
71 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
72
73 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
74
75 int nr_processes(void)
76 {
77         int cpu;
78         int total = 0;
79
80         for_each_online_cpu(cpu)
81                 total += per_cpu(process_counts, cpu);
82
83         return total;
84 }
85
86 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
87 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
88 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
89 static struct kmem_cache *task_struct_cachep;
90 #endif
91
92 /* SLAB cache for signal_struct structures (tsk->signal) */
93 static struct kmem_cache *signal_cachep;
94
95 /* SLAB cache for sighand_struct structures (tsk->sighand) */
96 struct kmem_cache *sighand_cachep;
97
98 /* SLAB cache for files_struct structures (tsk->files) */
99 struct kmem_cache *files_cachep;
100
101 /* SLAB cache for fs_struct structures (tsk->fs) */
102 struct kmem_cache *fs_cachep;
103
104 /* SLAB cache for vm_area_struct structures */
105 struct kmem_cache *vm_area_cachep;
106
107 /* SLAB cache for mm_struct structures (tsk->mm) */
108 static struct kmem_cache *mm_cachep;
109
110 void free_task(struct task_struct *tsk)
111 {
112         prop_local_destroy_single(&tsk->dirties);
113         free_thread_info(tsk->stack);
114         rt_mutex_debug_task_free(tsk);
115         free_task_struct(tsk);
116 }
117 EXPORT_SYMBOL(free_task);
118
119 void __put_task_struct(struct task_struct *tsk)
120 {
121         WARN_ON(!tsk->exit_state);
122         WARN_ON(atomic_read(&tsk->usage));
123         WARN_ON(tsk == current);
124
125         security_task_free(tsk);
126         free_uid(tsk->user);
127         put_group_info(tsk->group_info);
128         delayacct_tsk_free(tsk);
129
130         if (!profile_handoff_task(tsk))
131                 free_task(tsk);
132 }
133
134 void __init fork_init(unsigned long mempages)
135 {
136 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
137 #ifndef ARCH_MIN_TASKALIGN
138 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
139 #endif
140         /* create a slab on which task_structs can be allocated */
141         task_struct_cachep =
142                 kmem_cache_create("task_struct", sizeof(struct task_struct),
143                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
144 #endif
145
146         /*
147          * The default maximum number of threads is set to a safe
148          * value: the thread structures can take up at most half
149          * of memory.
150          */
151         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
152
153         /*
154          * we need to allow at least 20 threads to boot a system
155          */
156         if(max_threads < 20)
157                 max_threads = 20;
158
159         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
160         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
161         init_task.signal->rlim[RLIMIT_SIGPENDING] =
162                 init_task.signal->rlim[RLIMIT_NPROC];
163 }
164
165 static struct task_struct *dup_task_struct(struct task_struct *orig)
166 {
167         struct task_struct *tsk;
168         struct thread_info *ti;
169         int err;
170
171         prepare_to_copy(orig);
172
173         tsk = alloc_task_struct();
174         if (!tsk)
175                 return NULL;
176
177         ti = alloc_thread_info(tsk);
178         if (!ti) {
179                 free_task_struct(tsk);
180                 return NULL;
181         }
182
183         *tsk = *orig;
184         tsk->stack = ti;
185
186         err = prop_local_init_single(&tsk->dirties);
187         if (err) {
188                 free_thread_info(ti);
189                 free_task_struct(tsk);
190                 return NULL;
191         }
192
193         setup_thread_stack(tsk, orig);
194
195 #ifdef CONFIG_CC_STACKPROTECTOR
196         tsk->stack_canary = get_random_int();
197 #endif
198
199         /* One for us, one for whoever does the "release_task()" (usually parent) */
200         atomic_set(&tsk->usage,2);
201         atomic_set(&tsk->fs_excl, 0);
202 #ifdef CONFIG_BLK_DEV_IO_TRACE
203         tsk->btrace_seq = 0;
204 #endif
205         tsk->splice_pipe = NULL;
206         return tsk;
207 }
208
209 #ifdef CONFIG_MMU
210 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
211 {
212         struct vm_area_struct *mpnt, *tmp, **pprev;
213         struct rb_node **rb_link, *rb_parent;
214         int retval;
215         unsigned long charge;
216         struct mempolicy *pol;
217
218         down_write(&oldmm->mmap_sem);
219         flush_cache_dup_mm(oldmm);
220         /*
221          * Not linked in yet - no deadlock potential:
222          */
223         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
224
225         mm->locked_vm = 0;
226         mm->mmap = NULL;
227         mm->mmap_cache = NULL;
228         mm->free_area_cache = oldmm->mmap_base;
229         mm->cached_hole_size = ~0UL;
230         mm->map_count = 0;
231         cpus_clear(mm->cpu_vm_mask);
232         mm->mm_rb = RB_ROOT;
233         rb_link = &mm->mm_rb.rb_node;
234         rb_parent = NULL;
235         pprev = &mm->mmap;
236
237         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
238                 struct file *file;
239
240                 if (mpnt->vm_flags & VM_DONTCOPY) {
241                         long pages = vma_pages(mpnt);
242                         mm->total_vm -= pages;
243                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
244                                                                 -pages);
245                         continue;
246                 }
247                 charge = 0;
248                 if (mpnt->vm_flags & VM_ACCOUNT) {
249                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
250                         if (security_vm_enough_memory(len))
251                                 goto fail_nomem;
252                         charge = len;
253                 }
254                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
255                 if (!tmp)
256                         goto fail_nomem;
257                 *tmp = *mpnt;
258                 pol = mpol_copy(vma_policy(mpnt));
259                 retval = PTR_ERR(pol);
260                 if (IS_ERR(pol))
261                         goto fail_nomem_policy;
262                 vma_set_policy(tmp, pol);
263                 tmp->vm_flags &= ~VM_LOCKED;
264                 tmp->vm_mm = mm;
265                 tmp->vm_next = NULL;
266                 anon_vma_link(tmp);
267                 file = tmp->vm_file;
268                 if (file) {
269                         struct inode *inode = file->f_path.dentry->d_inode;
270                         get_file(file);
271                         if (tmp->vm_flags & VM_DENYWRITE)
272                                 atomic_dec(&inode->i_writecount);
273
274                         /* insert tmp into the share list, just after mpnt */
275                         spin_lock(&file->f_mapping->i_mmap_lock);
276                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
277                         flush_dcache_mmap_lock(file->f_mapping);
278                         vma_prio_tree_add(tmp, mpnt);
279                         flush_dcache_mmap_unlock(file->f_mapping);
280                         spin_unlock(&file->f_mapping->i_mmap_lock);
281                 }
282
283                 /*
284                  * Link in the new vma and copy the page table entries.
285                  */
286                 *pprev = tmp;
287                 pprev = &tmp->vm_next;
288
289                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
290                 rb_link = &tmp->vm_rb.rb_right;
291                 rb_parent = &tmp->vm_rb;
292
293                 mm->map_count++;
294                 retval = copy_page_range(mm, oldmm, mpnt);
295
296                 if (tmp->vm_ops && tmp->vm_ops->open)
297                         tmp->vm_ops->open(tmp);
298
299                 if (retval)
300                         goto out;
301         }
302         /* a new mm has just been created */
303         arch_dup_mmap(oldmm, mm);
304         retval = 0;
305 out:
306         up_write(&mm->mmap_sem);
307         flush_tlb_mm(oldmm);
308         up_write(&oldmm->mmap_sem);
309         return retval;
310 fail_nomem_policy:
311         kmem_cache_free(vm_area_cachep, tmp);
312 fail_nomem:
313         retval = -ENOMEM;
314         vm_unacct_memory(charge);
315         goto out;
316 }
317
318 static inline int mm_alloc_pgd(struct mm_struct * mm)
319 {
320         mm->pgd = pgd_alloc(mm);
321         if (unlikely(!mm->pgd))
322                 return -ENOMEM;
323         return 0;
324 }
325
326 static inline void mm_free_pgd(struct mm_struct * mm)
327 {
328         pgd_free(mm, mm->pgd);
329 }
330 #else
331 #define dup_mmap(mm, oldmm)     (0)
332 #define mm_alloc_pgd(mm)        (0)
333 #define mm_free_pgd(mm)
334 #endif /* CONFIG_MMU */
335
336 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
337
338 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
339 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
340
341 #include <linux/init_task.h>
342
343 static struct mm_struct * mm_init(struct mm_struct * mm)
344 {
345         atomic_set(&mm->mm_users, 1);
346         atomic_set(&mm->mm_count, 1);
347         init_rwsem(&mm->mmap_sem);
348         INIT_LIST_HEAD(&mm->mmlist);
349         mm->flags = (current->mm) ? current->mm->flags
350                                   : MMF_DUMP_FILTER_DEFAULT;
351         mm->core_waiters = 0;
352         mm->nr_ptes = 0;
353         set_mm_counter(mm, file_rss, 0);
354         set_mm_counter(mm, anon_rss, 0);
355         spin_lock_init(&mm->page_table_lock);
356         rwlock_init(&mm->ioctx_list_lock);
357         mm->ioctx_list = NULL;
358         mm->free_area_cache = TASK_UNMAPPED_BASE;
359         mm->cached_hole_size = ~0UL;
360
361         if (likely(!mm_alloc_pgd(mm))) {
362                 mm->def_flags = 0;
363                 return mm;
364         }
365         free_mm(mm);
366         return NULL;
367 }
368
369 /*
370  * Allocate and initialize an mm_struct.
371  */
372 struct mm_struct * mm_alloc(void)
373 {
374         struct mm_struct * mm;
375
376         mm = allocate_mm();
377         if (mm) {
378                 memset(mm, 0, sizeof(*mm));
379                 mm = mm_init(mm);
380         }
381         return mm;
382 }
383
384 /*
385  * Called when the last reference to the mm
386  * is dropped: either by a lazy thread or by
387  * mmput. Free the page directory and the mm.
388  */
389 void fastcall __mmdrop(struct mm_struct *mm)
390 {
391         BUG_ON(mm == &init_mm);
392         mm_free_pgd(mm);
393         destroy_context(mm);
394         free_mm(mm);
395 }
396 EXPORT_SYMBOL_GPL(__mmdrop);
397
398 /*
399  * Decrement the use count and release all resources for an mm.
400  */
401 void mmput(struct mm_struct *mm)
402 {
403         might_sleep();
404
405         if (atomic_dec_and_test(&mm->mm_users)) {
406                 exit_aio(mm);
407                 exit_mmap(mm);
408                 if (!list_empty(&mm->mmlist)) {
409                         spin_lock(&mmlist_lock);
410                         list_del(&mm->mmlist);
411                         spin_unlock(&mmlist_lock);
412                 }
413                 put_swap_token(mm);
414                 mmdrop(mm);
415         }
416 }
417 EXPORT_SYMBOL_GPL(mmput);
418
419 /**
420  * get_task_mm - acquire a reference to the task's mm
421  *
422  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
423  * this kernel workthread has transiently adopted a user mm with use_mm,
424  * to do its AIO) is not set and if so returns a reference to it, after
425  * bumping up the use count.  User must release the mm via mmput()
426  * after use.  Typically used by /proc and ptrace.
427  */
428 struct mm_struct *get_task_mm(struct task_struct *task)
429 {
430         struct mm_struct *mm;
431
432         task_lock(task);
433         mm = task->mm;
434         if (mm) {
435                 if (task->flags & PF_BORROWED_MM)
436                         mm = NULL;
437                 else
438                         atomic_inc(&mm->mm_users);
439         }
440         task_unlock(task);
441         return mm;
442 }
443 EXPORT_SYMBOL_GPL(get_task_mm);
444
445 /* Please note the differences between mmput and mm_release.
446  * mmput is called whenever we stop holding onto a mm_struct,
447  * error success whatever.
448  *
449  * mm_release is called after a mm_struct has been removed
450  * from the current process.
451  *
452  * This difference is important for error handling, when we
453  * only half set up a mm_struct for a new process and need to restore
454  * the old one.  Because we mmput the new mm_struct before
455  * restoring the old one. . .
456  * Eric Biederman 10 January 1998
457  */
458 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
459 {
460         struct completion *vfork_done = tsk->vfork_done;
461
462         /* Get rid of any cached register state */
463         deactivate_mm(tsk, mm);
464
465         /* notify parent sleeping on vfork() */
466         if (vfork_done) {
467                 tsk->vfork_done = NULL;
468                 complete(vfork_done);
469         }
470
471         /*
472          * If we're exiting normally, clear a user-space tid field if
473          * requested.  We leave this alone when dying by signal, to leave
474          * the value intact in a core dump, and to save the unnecessary
475          * trouble otherwise.  Userland only wants this done for a sys_exit.
476          */
477         if (tsk->clear_child_tid
478             && !(tsk->flags & PF_SIGNALED)
479             && atomic_read(&mm->mm_users) > 1) {
480                 u32 __user * tidptr = tsk->clear_child_tid;
481                 tsk->clear_child_tid = NULL;
482
483                 /*
484                  * We don't check the error code - if userspace has
485                  * not set up a proper pointer then tough luck.
486                  */
487                 put_user(0, tidptr);
488                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
489         }
490 }
491
492 /*
493  * Allocate a new mm structure and copy contents from the
494  * mm structure of the passed in task structure.
495  */
496 static struct mm_struct *dup_mm(struct task_struct *tsk)
497 {
498         struct mm_struct *mm, *oldmm = current->mm;
499         int err;
500
501         if (!oldmm)
502                 return NULL;
503
504         mm = allocate_mm();
505         if (!mm)
506                 goto fail_nomem;
507
508         memcpy(mm, oldmm, sizeof(*mm));
509
510         /* Initializing for Swap token stuff */
511         mm->token_priority = 0;
512         mm->last_interval = 0;
513
514         if (!mm_init(mm))
515                 goto fail_nomem;
516
517         if (init_new_context(tsk, mm))
518                 goto fail_nocontext;
519
520         err = dup_mmap(mm, oldmm);
521         if (err)
522                 goto free_pt;
523
524         mm->hiwater_rss = get_mm_rss(mm);
525         mm->hiwater_vm = mm->total_vm;
526
527         return mm;
528
529 free_pt:
530         mmput(mm);
531
532 fail_nomem:
533         return NULL;
534
535 fail_nocontext:
536         /*
537          * If init_new_context() failed, we cannot use mmput() to free the mm
538          * because it calls destroy_context()
539          */
540         mm_free_pgd(mm);
541         free_mm(mm);
542         return NULL;
543 }
544
545 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
546 {
547         struct mm_struct * mm, *oldmm;
548         int retval;
549
550         tsk->min_flt = tsk->maj_flt = 0;
551         tsk->nvcsw = tsk->nivcsw = 0;
552
553         tsk->mm = NULL;
554         tsk->active_mm = NULL;
555
556         /*
557          * Are we cloning a kernel thread?
558          *
559          * We need to steal a active VM for that..
560          */
561         oldmm = current->mm;
562         if (!oldmm)
563                 return 0;
564
565         if (clone_flags & CLONE_VM) {
566                 atomic_inc(&oldmm->mm_users);
567                 mm = oldmm;
568                 goto good_mm;
569         }
570
571         retval = -ENOMEM;
572         mm = dup_mm(tsk);
573         if (!mm)
574                 goto fail_nomem;
575
576 good_mm:
577         /* Initializing for Swap token stuff */
578         mm->token_priority = 0;
579         mm->last_interval = 0;
580
581         tsk->mm = mm;
582         tsk->active_mm = mm;
583         return 0;
584
585 fail_nomem:
586         return retval;
587 }
588
589 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
590 {
591         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
592         /* We don't need to lock fs - think why ;-) */
593         if (fs) {
594                 atomic_set(&fs->count, 1);
595                 rwlock_init(&fs->lock);
596                 fs->umask = old->umask;
597                 read_lock(&old->lock);
598                 fs->rootmnt = mntget(old->rootmnt);
599                 fs->root = dget(old->root);
600                 fs->pwdmnt = mntget(old->pwdmnt);
601                 fs->pwd = dget(old->pwd);
602                 if (old->altroot) {
603                         fs->altrootmnt = mntget(old->altrootmnt);
604                         fs->altroot = dget(old->altroot);
605                 } else {
606                         fs->altrootmnt = NULL;
607                         fs->altroot = NULL;
608                 }
609                 read_unlock(&old->lock);
610         }
611         return fs;
612 }
613
614 struct fs_struct *copy_fs_struct(struct fs_struct *old)
615 {
616         return __copy_fs_struct(old);
617 }
618
619 EXPORT_SYMBOL_GPL(copy_fs_struct);
620
621 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
622 {
623         if (clone_flags & CLONE_FS) {
624                 atomic_inc(&current->fs->count);
625                 return 0;
626         }
627         tsk->fs = __copy_fs_struct(current->fs);
628         if (!tsk->fs)
629                 return -ENOMEM;
630         return 0;
631 }
632
633 static int count_open_files(struct fdtable *fdt)
634 {
635         int size = fdt->max_fds;
636         int i;
637
638         /* Find the last open fd */
639         for (i = size/(8*sizeof(long)); i > 0; ) {
640                 if (fdt->open_fds->fds_bits[--i])
641                         break;
642         }
643         i = (i+1) * 8 * sizeof(long);
644         return i;
645 }
646
647 static struct files_struct *alloc_files(void)
648 {
649         struct files_struct *newf;
650         struct fdtable *fdt;
651
652         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
653         if (!newf)
654                 goto out;
655
656         atomic_set(&newf->count, 1);
657
658         spin_lock_init(&newf->file_lock);
659         newf->next_fd = 0;
660         fdt = &newf->fdtab;
661         fdt->max_fds = NR_OPEN_DEFAULT;
662         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
663         fdt->open_fds = (fd_set *)&newf->open_fds_init;
664         fdt->fd = &newf->fd_array[0];
665         INIT_RCU_HEAD(&fdt->rcu);
666         fdt->next = NULL;
667         rcu_assign_pointer(newf->fdt, fdt);
668 out:
669         return newf;
670 }
671
672 /*
673  * Allocate a new files structure and copy contents from the
674  * passed in files structure.
675  * errorp will be valid only when the returned files_struct is NULL.
676  */
677 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
678 {
679         struct files_struct *newf;
680         struct file **old_fds, **new_fds;
681         int open_files, size, i;
682         struct fdtable *old_fdt, *new_fdt;
683
684         *errorp = -ENOMEM;
685         newf = alloc_files();
686         if (!newf)
687                 goto out;
688
689         spin_lock(&oldf->file_lock);
690         old_fdt = files_fdtable(oldf);
691         new_fdt = files_fdtable(newf);
692         open_files = count_open_files(old_fdt);
693
694         /*
695          * Check whether we need to allocate a larger fd array and fd set.
696          * Note: we're not a clone task, so the open count won't change.
697          */
698         if (open_files > new_fdt->max_fds) {
699                 new_fdt->max_fds = 0;
700                 spin_unlock(&oldf->file_lock);
701                 spin_lock(&newf->file_lock);
702                 *errorp = expand_files(newf, open_files-1);
703                 spin_unlock(&newf->file_lock);
704                 if (*errorp < 0)
705                         goto out_release;
706                 new_fdt = files_fdtable(newf);
707                 /*
708                  * Reacquire the oldf lock and a pointer to its fd table
709                  * who knows it may have a new bigger fd table. We need
710                  * the latest pointer.
711                  */
712                 spin_lock(&oldf->file_lock);
713                 old_fdt = files_fdtable(oldf);
714         }
715
716         old_fds = old_fdt->fd;
717         new_fds = new_fdt->fd;
718
719         memcpy(new_fdt->open_fds->fds_bits,
720                 old_fdt->open_fds->fds_bits, open_files/8);
721         memcpy(new_fdt->close_on_exec->fds_bits,
722                 old_fdt->close_on_exec->fds_bits, open_files/8);
723
724         for (i = open_files; i != 0; i--) {
725                 struct file *f = *old_fds++;
726                 if (f) {
727                         get_file(f);
728                 } else {
729                         /*
730                          * The fd may be claimed in the fd bitmap but not yet
731                          * instantiated in the files array if a sibling thread
732                          * is partway through open().  So make sure that this
733                          * fd is available to the new process.
734                          */
735                         FD_CLR(open_files - i, new_fdt->open_fds);
736                 }
737                 rcu_assign_pointer(*new_fds++, f);
738         }
739         spin_unlock(&oldf->file_lock);
740
741         /* compute the remainder to be cleared */
742         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
743
744         /* This is long word aligned thus could use a optimized version */
745         memset(new_fds, 0, size);
746
747         if (new_fdt->max_fds > open_files) {
748                 int left = (new_fdt->max_fds-open_files)/8;
749                 int start = open_files / (8 * sizeof(unsigned long));
750
751                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
752                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
753         }
754
755         return newf;
756
757 out_release:
758         kmem_cache_free(files_cachep, newf);
759 out:
760         return NULL;
761 }
762
763 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
764 {
765         struct files_struct *oldf, *newf;
766         int error = 0;
767
768         /*
769          * A background process may not have any files ...
770          */
771         oldf = current->files;
772         if (!oldf)
773                 goto out;
774
775         if (clone_flags & CLONE_FILES) {
776                 atomic_inc(&oldf->count);
777                 goto out;
778         }
779
780         /*
781          * Note: we may be using current for both targets (See exec.c)
782          * This works because we cache current->files (old) as oldf. Don't
783          * break this.
784          */
785         tsk->files = NULL;
786         newf = dup_fd(oldf, &error);
787         if (!newf)
788                 goto out;
789
790         tsk->files = newf;
791         error = 0;
792 out:
793         return error;
794 }
795
796 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
797 {
798 #ifdef CONFIG_BLOCK
799         struct io_context *ioc = current->io_context;
800
801         if (!ioc)
802                 return 0;
803         /*
804          * Share io context with parent, if CLONE_IO is set
805          */
806         if (clone_flags & CLONE_IO) {
807                 tsk->io_context = ioc_task_link(ioc);
808                 if (unlikely(!tsk->io_context))
809                         return -ENOMEM;
810         } else if (ioprio_valid(ioc->ioprio)) {
811                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
812                 if (unlikely(!tsk->io_context))
813                         return -ENOMEM;
814
815                 tsk->io_context->ioprio = ioc->ioprio;
816         }
817 #endif
818         return 0;
819 }
820
821 /*
822  *      Helper to unshare the files of the current task.
823  *      We don't want to expose copy_files internals to
824  *      the exec layer of the kernel.
825  */
826
827 int unshare_files(void)
828 {
829         struct files_struct *files  = current->files;
830         int rc;
831
832         BUG_ON(!files);
833
834         /* This can race but the race causes us to copy when we don't
835            need to and drop the copy */
836         if(atomic_read(&files->count) == 1)
837         {
838                 atomic_inc(&files->count);
839                 return 0;
840         }
841         rc = copy_files(0, current);
842         if(rc)
843                 current->files = files;
844         return rc;
845 }
846
847 EXPORT_SYMBOL(unshare_files);
848
849 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
850 {
851         struct sighand_struct *sig;
852
853         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
854                 atomic_inc(&current->sighand->count);
855                 return 0;
856         }
857         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
858         rcu_assign_pointer(tsk->sighand, sig);
859         if (!sig)
860                 return -ENOMEM;
861         atomic_set(&sig->count, 1);
862         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
863         return 0;
864 }
865
866 void __cleanup_sighand(struct sighand_struct *sighand)
867 {
868         if (atomic_dec_and_test(&sighand->count))
869                 kmem_cache_free(sighand_cachep, sighand);
870 }
871
872 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
873 {
874         struct signal_struct *sig;
875         int ret;
876
877         if (clone_flags & CLONE_THREAD) {
878                 atomic_inc(&current->signal->count);
879                 atomic_inc(&current->signal->live);
880                 return 0;
881         }
882         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
883         tsk->signal = sig;
884         if (!sig)
885                 return -ENOMEM;
886
887         ret = copy_thread_group_keys(tsk);
888         if (ret < 0) {
889                 kmem_cache_free(signal_cachep, sig);
890                 return ret;
891         }
892
893         atomic_set(&sig->count, 1);
894         atomic_set(&sig->live, 1);
895         init_waitqueue_head(&sig->wait_chldexit);
896         sig->flags = 0;
897         sig->group_exit_code = 0;
898         sig->group_exit_task = NULL;
899         sig->group_stop_count = 0;
900         sig->curr_target = NULL;
901         init_sigpending(&sig->shared_pending);
902         INIT_LIST_HEAD(&sig->posix_timers);
903
904         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
905         sig->it_real_incr.tv64 = 0;
906         sig->real_timer.function = it_real_fn;
907         sig->tsk = tsk;
908
909         sig->it_virt_expires = cputime_zero;
910         sig->it_virt_incr = cputime_zero;
911         sig->it_prof_expires = cputime_zero;
912         sig->it_prof_incr = cputime_zero;
913
914         sig->leader = 0;        /* session leadership doesn't inherit */
915         sig->tty_old_pgrp = NULL;
916
917         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
918         sig->gtime = cputime_zero;
919         sig->cgtime = cputime_zero;
920         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
921         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
922         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
923         sig->sum_sched_runtime = 0;
924         INIT_LIST_HEAD(&sig->cpu_timers[0]);
925         INIT_LIST_HEAD(&sig->cpu_timers[1]);
926         INIT_LIST_HEAD(&sig->cpu_timers[2]);
927         taskstats_tgid_init(sig);
928
929         task_lock(current->group_leader);
930         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
931         task_unlock(current->group_leader);
932
933         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
934                 /*
935                  * New sole thread in the process gets an expiry time
936                  * of the whole CPU time limit.
937                  */
938                 tsk->it_prof_expires =
939                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
940         }
941         acct_init_pacct(&sig->pacct);
942
943         tty_audit_fork(sig);
944
945         return 0;
946 }
947
948 void __cleanup_signal(struct signal_struct *sig)
949 {
950         exit_thread_group_keys(sig);
951         kmem_cache_free(signal_cachep, sig);
952 }
953
954 static void cleanup_signal(struct task_struct *tsk)
955 {
956         struct signal_struct *sig = tsk->signal;
957
958         atomic_dec(&sig->live);
959
960         if (atomic_dec_and_test(&sig->count))
961                 __cleanup_signal(sig);
962 }
963
964 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
965 {
966         unsigned long new_flags = p->flags;
967
968         new_flags &= ~PF_SUPERPRIV;
969         new_flags |= PF_FORKNOEXEC;
970         if (!(clone_flags & CLONE_PTRACE))
971                 p->ptrace = 0;
972         p->flags = new_flags;
973         clear_freeze_flag(p);
974 }
975
976 asmlinkage long sys_set_tid_address(int __user *tidptr)
977 {
978         current->clear_child_tid = tidptr;
979
980         return task_pid_vnr(current);
981 }
982
983 static void rt_mutex_init_task(struct task_struct *p)
984 {
985         spin_lock_init(&p->pi_lock);
986 #ifdef CONFIG_RT_MUTEXES
987         plist_head_init(&p->pi_waiters, &p->pi_lock);
988         p->pi_blocked_on = NULL;
989 #endif
990 }
991
992 /*
993  * This creates a new process as a copy of the old one,
994  * but does not actually start it yet.
995  *
996  * It copies the registers, and all the appropriate
997  * parts of the process environment (as per the clone
998  * flags). The actual kick-off is left to the caller.
999  */
1000 static struct task_struct *copy_process(unsigned long clone_flags,
1001                                         unsigned long stack_start,
1002                                         struct pt_regs *regs,
1003                                         unsigned long stack_size,
1004                                         int __user *child_tidptr,
1005                                         struct pid *pid)
1006 {
1007         int retval;
1008         struct task_struct *p;
1009         int cgroup_callbacks_done = 0;
1010
1011         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1012                 return ERR_PTR(-EINVAL);
1013
1014         /*
1015          * Thread groups must share signals as well, and detached threads
1016          * can only be started up within the thread group.
1017          */
1018         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1019                 return ERR_PTR(-EINVAL);
1020
1021         /*
1022          * Shared signal handlers imply shared VM. By way of the above,
1023          * thread groups also imply shared VM. Blocking this case allows
1024          * for various simplifications in other code.
1025          */
1026         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1027                 return ERR_PTR(-EINVAL);
1028
1029         retval = security_task_create(clone_flags);
1030         if (retval)
1031                 goto fork_out;
1032
1033         retval = -ENOMEM;
1034         p = dup_task_struct(current);
1035         if (!p)
1036                 goto fork_out;
1037
1038         rt_mutex_init_task(p);
1039
1040 #ifdef CONFIG_TRACE_IRQFLAGS
1041         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1042         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1043 #endif
1044         retval = -EAGAIN;
1045         if (atomic_read(&p->user->processes) >=
1046                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1047                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1048                     p->user != current->nsproxy->user_ns->root_user)
1049                         goto bad_fork_free;
1050         }
1051
1052         atomic_inc(&p->user->__count);
1053         atomic_inc(&p->user->processes);
1054         get_group_info(p->group_info);
1055
1056         /*
1057          * If multiple threads are within copy_process(), then this check
1058          * triggers too late. This doesn't hurt, the check is only there
1059          * to stop root fork bombs.
1060          */
1061         if (nr_threads >= max_threads)
1062                 goto bad_fork_cleanup_count;
1063
1064         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1065                 goto bad_fork_cleanup_count;
1066
1067         if (p->binfmt && !try_module_get(p->binfmt->module))
1068                 goto bad_fork_cleanup_put_domain;
1069
1070         p->did_exec = 0;
1071         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1072         copy_flags(clone_flags, p);
1073         INIT_LIST_HEAD(&p->children);
1074         INIT_LIST_HEAD(&p->sibling);
1075 #ifdef CONFIG_PREEMPT_RCU
1076         p->rcu_read_lock_nesting = 0;
1077         p->rcu_flipctr_idx = 0;
1078 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1079         p->vfork_done = NULL;
1080         spin_lock_init(&p->alloc_lock);
1081
1082         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1083         init_sigpending(&p->pending);
1084
1085         p->utime = cputime_zero;
1086         p->stime = cputime_zero;
1087         p->gtime = cputime_zero;
1088         p->utimescaled = cputime_zero;
1089         p->stimescaled = cputime_zero;
1090         p->prev_utime = cputime_zero;
1091         p->prev_stime = cputime_zero;
1092
1093 #ifdef CONFIG_DETECT_SOFTLOCKUP
1094         p->last_switch_count = 0;
1095         p->last_switch_timestamp = 0;
1096 #endif
1097
1098 #ifdef CONFIG_TASK_XACCT
1099         p->rchar = 0;           /* I/O counter: bytes read */
1100         p->wchar = 0;           /* I/O counter: bytes written */
1101         p->syscr = 0;           /* I/O counter: read syscalls */
1102         p->syscw = 0;           /* I/O counter: write syscalls */
1103 #endif
1104         task_io_accounting_init(p);
1105         acct_clear_integrals(p);
1106
1107         p->it_virt_expires = cputime_zero;
1108         p->it_prof_expires = cputime_zero;
1109         p->it_sched_expires = 0;
1110         INIT_LIST_HEAD(&p->cpu_timers[0]);
1111         INIT_LIST_HEAD(&p->cpu_timers[1]);
1112         INIT_LIST_HEAD(&p->cpu_timers[2]);
1113
1114         p->lock_depth = -1;             /* -1 = no lock */
1115         do_posix_clock_monotonic_gettime(&p->start_time);
1116         p->real_start_time = p->start_time;
1117         monotonic_to_bootbased(&p->real_start_time);
1118 #ifdef CONFIG_SECURITY
1119         p->security = NULL;
1120 #endif
1121         p->cap_bset = current->cap_bset;
1122         p->io_context = NULL;
1123         p->audit_context = NULL;
1124         cgroup_fork(p);
1125 #ifdef CONFIG_NUMA
1126         p->mempolicy = mpol_copy(p->mempolicy);
1127         if (IS_ERR(p->mempolicy)) {
1128                 retval = PTR_ERR(p->mempolicy);
1129                 p->mempolicy = NULL;
1130                 goto bad_fork_cleanup_cgroup;
1131         }
1132         mpol_fix_fork_child_flag(p);
1133 #endif
1134 #ifdef CONFIG_TRACE_IRQFLAGS
1135         p->irq_events = 0;
1136 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1137         p->hardirqs_enabled = 1;
1138 #else
1139         p->hardirqs_enabled = 0;
1140 #endif
1141         p->hardirq_enable_ip = 0;
1142         p->hardirq_enable_event = 0;
1143         p->hardirq_disable_ip = _THIS_IP_;
1144         p->hardirq_disable_event = 0;
1145         p->softirqs_enabled = 1;
1146         p->softirq_enable_ip = _THIS_IP_;
1147         p->softirq_enable_event = 0;
1148         p->softirq_disable_ip = 0;
1149         p->softirq_disable_event = 0;
1150         p->hardirq_context = 0;
1151         p->softirq_context = 0;
1152 #endif
1153 #ifdef CONFIG_LOCKDEP
1154         p->lockdep_depth = 0; /* no locks held yet */
1155         p->curr_chain_key = 0;
1156         p->lockdep_recursion = 0;
1157 #endif
1158
1159 #ifdef CONFIG_DEBUG_MUTEXES
1160         p->blocked_on = NULL; /* not blocked yet */
1161 #endif
1162
1163         /* Perform scheduler related setup. Assign this task to a CPU. */
1164         sched_fork(p, clone_flags);
1165
1166         if ((retval = security_task_alloc(p)))
1167                 goto bad_fork_cleanup_policy;
1168         if ((retval = audit_alloc(p)))
1169                 goto bad_fork_cleanup_security;
1170         /* copy all the process information */
1171         if ((retval = copy_semundo(clone_flags, p)))
1172                 goto bad_fork_cleanup_audit;
1173         if ((retval = copy_files(clone_flags, p)))
1174                 goto bad_fork_cleanup_semundo;
1175         if ((retval = copy_fs(clone_flags, p)))
1176                 goto bad_fork_cleanup_files;
1177         if ((retval = copy_sighand(clone_flags, p)))
1178                 goto bad_fork_cleanup_fs;
1179         if ((retval = copy_signal(clone_flags, p)))
1180                 goto bad_fork_cleanup_sighand;
1181         if ((retval = copy_mm(clone_flags, p)))
1182                 goto bad_fork_cleanup_signal;
1183         if ((retval = copy_keys(clone_flags, p)))
1184                 goto bad_fork_cleanup_mm;
1185         if ((retval = copy_namespaces(clone_flags, p)))
1186                 goto bad_fork_cleanup_keys;
1187         if ((retval = copy_io(clone_flags, p)))
1188                 goto bad_fork_cleanup_namespaces;
1189         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1190         if (retval)
1191                 goto bad_fork_cleanup_io;
1192
1193         if (pid != &init_struct_pid) {
1194                 retval = -ENOMEM;
1195                 pid = alloc_pid(task_active_pid_ns(p));
1196                 if (!pid)
1197                         goto bad_fork_cleanup_io;
1198
1199                 if (clone_flags & CLONE_NEWPID) {
1200                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1201                         if (retval < 0)
1202                                 goto bad_fork_free_pid;
1203                 }
1204         }
1205
1206         p->pid = pid_nr(pid);
1207         p->tgid = p->pid;
1208         if (clone_flags & CLONE_THREAD)
1209                 p->tgid = current->tgid;
1210
1211         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1212         /*
1213          * Clear TID on mm_release()?
1214          */
1215         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1216 #ifdef CONFIG_FUTEX
1217         p->robust_list = NULL;
1218 #ifdef CONFIG_COMPAT
1219         p->compat_robust_list = NULL;
1220 #endif
1221         INIT_LIST_HEAD(&p->pi_state_list);
1222         p->pi_state_cache = NULL;
1223 #endif
1224         /*
1225          * sigaltstack should be cleared when sharing the same VM
1226          */
1227         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1228                 p->sas_ss_sp = p->sas_ss_size = 0;
1229
1230         /*
1231          * Syscall tracing should be turned off in the child regardless
1232          * of CLONE_PTRACE.
1233          */
1234         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1235 #ifdef TIF_SYSCALL_EMU
1236         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1237 #endif
1238         clear_all_latency_tracing(p);
1239
1240         /* Our parent execution domain becomes current domain
1241            These must match for thread signalling to apply */
1242         p->parent_exec_id = p->self_exec_id;
1243
1244         /* ok, now we should be set up.. */
1245         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1246         p->pdeath_signal = 0;
1247         p->exit_state = 0;
1248
1249         /*
1250          * Ok, make it visible to the rest of the system.
1251          * We dont wake it up yet.
1252          */
1253         p->group_leader = p;
1254         INIT_LIST_HEAD(&p->thread_group);
1255         INIT_LIST_HEAD(&p->ptrace_children);
1256         INIT_LIST_HEAD(&p->ptrace_list);
1257
1258         /* Now that the task is set up, run cgroup callbacks if
1259          * necessary. We need to run them before the task is visible
1260          * on the tasklist. */
1261         cgroup_fork_callbacks(p);
1262         cgroup_callbacks_done = 1;
1263
1264         /* Need tasklist lock for parent etc handling! */
1265         write_lock_irq(&tasklist_lock);
1266
1267         /*
1268          * The task hasn't been attached yet, so its cpus_allowed mask will
1269          * not be changed, nor will its assigned CPU.
1270          *
1271          * The cpus_allowed mask of the parent may have changed after it was
1272          * copied first time - so re-copy it here, then check the child's CPU
1273          * to ensure it is on a valid CPU (and if not, just force it back to
1274          * parent's CPU). This avoids alot of nasty races.
1275          */
1276         p->cpus_allowed = current->cpus_allowed;
1277         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1278         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1279                         !cpu_online(task_cpu(p))))
1280                 set_task_cpu(p, smp_processor_id());
1281
1282         /* CLONE_PARENT re-uses the old parent */
1283         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1284                 p->real_parent = current->real_parent;
1285         else
1286                 p->real_parent = current;
1287         p->parent = p->real_parent;
1288
1289         spin_lock(&current->sighand->siglock);
1290
1291         /*
1292          * Process group and session signals need to be delivered to just the
1293          * parent before the fork or both the parent and the child after the
1294          * fork. Restart if a signal comes in before we add the new process to
1295          * it's process group.
1296          * A fatal signal pending means that current will exit, so the new
1297          * thread can't slip out of an OOM kill (or normal SIGKILL).
1298          */
1299         recalc_sigpending();
1300         if (signal_pending(current)) {
1301                 spin_unlock(&current->sighand->siglock);
1302                 write_unlock_irq(&tasklist_lock);
1303                 retval = -ERESTARTNOINTR;
1304                 goto bad_fork_free_pid;
1305         }
1306
1307         if (clone_flags & CLONE_THREAD) {
1308                 p->group_leader = current->group_leader;
1309                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1310
1311                 if (!cputime_eq(current->signal->it_virt_expires,
1312                                 cputime_zero) ||
1313                     !cputime_eq(current->signal->it_prof_expires,
1314                                 cputime_zero) ||
1315                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1316                     !list_empty(&current->signal->cpu_timers[0]) ||
1317                     !list_empty(&current->signal->cpu_timers[1]) ||
1318                     !list_empty(&current->signal->cpu_timers[2])) {
1319                         /*
1320                          * Have child wake up on its first tick to check
1321                          * for process CPU timers.
1322                          */
1323                         p->it_prof_expires = jiffies_to_cputime(1);
1324                 }
1325         }
1326
1327         if (likely(p->pid)) {
1328                 add_parent(p);
1329                 if (unlikely(p->ptrace & PT_PTRACED))
1330                         __ptrace_link(p, current->parent);
1331
1332                 if (thread_group_leader(p)) {
1333                         if (clone_flags & CLONE_NEWPID)
1334                                 p->nsproxy->pid_ns->child_reaper = p;
1335
1336                         p->signal->tty = current->signal->tty;
1337                         set_task_pgrp(p, task_pgrp_nr(current));
1338                         set_task_session(p, task_session_nr(current));
1339                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1340                         attach_pid(p, PIDTYPE_SID, task_session(current));
1341                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1342                         __get_cpu_var(process_counts)++;
1343                 }
1344                 attach_pid(p, PIDTYPE_PID, pid);
1345                 nr_threads++;
1346         }
1347
1348         total_forks++;
1349         spin_unlock(&current->sighand->siglock);
1350         write_unlock_irq(&tasklist_lock);
1351         proc_fork_connector(p);
1352         cgroup_post_fork(p);
1353         return p;
1354
1355 bad_fork_free_pid:
1356         if (pid != &init_struct_pid)
1357                 free_pid(pid);
1358 bad_fork_cleanup_io:
1359         put_io_context(p->io_context);
1360 bad_fork_cleanup_namespaces:
1361         exit_task_namespaces(p);
1362 bad_fork_cleanup_keys:
1363         exit_keys(p);
1364 bad_fork_cleanup_mm:
1365         if (p->mm)
1366                 mmput(p->mm);
1367 bad_fork_cleanup_signal:
1368         cleanup_signal(p);
1369 bad_fork_cleanup_sighand:
1370         __cleanup_sighand(p->sighand);
1371 bad_fork_cleanup_fs:
1372         exit_fs(p); /* blocking */
1373 bad_fork_cleanup_files:
1374         exit_files(p); /* blocking */
1375 bad_fork_cleanup_semundo:
1376         exit_sem(p);
1377 bad_fork_cleanup_audit:
1378         audit_free(p);
1379 bad_fork_cleanup_security:
1380         security_task_free(p);
1381 bad_fork_cleanup_policy:
1382 #ifdef CONFIG_NUMA
1383         mpol_free(p->mempolicy);
1384 bad_fork_cleanup_cgroup:
1385 #endif
1386         cgroup_exit(p, cgroup_callbacks_done);
1387         delayacct_tsk_free(p);
1388         if (p->binfmt)
1389                 module_put(p->binfmt->module);
1390 bad_fork_cleanup_put_domain:
1391         module_put(task_thread_info(p)->exec_domain->module);
1392 bad_fork_cleanup_count:
1393         put_group_info(p->group_info);
1394         atomic_dec(&p->user->processes);
1395         free_uid(p->user);
1396 bad_fork_free:
1397         free_task(p);
1398 fork_out:
1399         return ERR_PTR(retval);
1400 }
1401
1402 noinline struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1403 {
1404         memset(regs, 0, sizeof(struct pt_regs));
1405         return regs;
1406 }
1407
1408 struct task_struct * __cpuinit fork_idle(int cpu)
1409 {
1410         struct task_struct *task;
1411         struct pt_regs regs;
1412
1413         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1414                                 &init_struct_pid);
1415         if (!IS_ERR(task))
1416                 init_idle(task, cpu);
1417
1418         return task;
1419 }
1420
1421 static int fork_traceflag(unsigned clone_flags)
1422 {
1423         if (clone_flags & CLONE_UNTRACED)
1424                 return 0;
1425         else if (clone_flags & CLONE_VFORK) {
1426                 if (current->ptrace & PT_TRACE_VFORK)
1427                         return PTRACE_EVENT_VFORK;
1428         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1429                 if (current->ptrace & PT_TRACE_CLONE)
1430                         return PTRACE_EVENT_CLONE;
1431         } else if (current->ptrace & PT_TRACE_FORK)
1432                 return PTRACE_EVENT_FORK;
1433
1434         return 0;
1435 }
1436
1437 /*
1438  *  Ok, this is the main fork-routine.
1439  *
1440  * It copies the process, and if successful kick-starts
1441  * it and waits for it to finish using the VM if required.
1442  */
1443 long do_fork(unsigned long clone_flags,
1444               unsigned long stack_start,
1445               struct pt_regs *regs,
1446               unsigned long stack_size,
1447               int __user *parent_tidptr,
1448               int __user *child_tidptr)
1449 {
1450         struct task_struct *p;
1451         int trace = 0;
1452         long nr;
1453
1454         /*
1455          * We hope to recycle these flags after 2.6.26
1456          */
1457         if (unlikely(clone_flags & CLONE_STOPPED)) {
1458                 static int __read_mostly count = 100;
1459
1460                 if (count > 0 && printk_ratelimit()) {
1461                         char comm[TASK_COMM_LEN];
1462
1463                         count--;
1464                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1465                                         "clone flags 0x%lx\n",
1466                                 get_task_comm(comm, current),
1467                                 clone_flags & CLONE_STOPPED);
1468                 }
1469         }
1470
1471         if (unlikely(current->ptrace)) {
1472                 trace = fork_traceflag (clone_flags);
1473                 if (trace)
1474                         clone_flags |= CLONE_PTRACE;
1475         }
1476
1477         p = copy_process(clone_flags, stack_start, regs, stack_size,
1478                         child_tidptr, NULL);
1479         /*
1480          * Do this prior waking up the new thread - the thread pointer
1481          * might get invalid after that point, if the thread exits quickly.
1482          */
1483         if (!IS_ERR(p)) {
1484                 struct completion vfork;
1485
1486                 /*
1487                  * this is enough to call pid_nr_ns here, but this if
1488                  * improves optimisation of regular fork()
1489                  */
1490                 nr = (clone_flags & CLONE_NEWPID) ?
1491                         task_pid_nr_ns(p, current->nsproxy->pid_ns) :
1492                                 task_pid_vnr(p);
1493
1494                 if (clone_flags & CLONE_PARENT_SETTID)
1495                         put_user(nr, parent_tidptr);
1496
1497                 if (clone_flags & CLONE_VFORK) {
1498                         p->vfork_done = &vfork;
1499                         init_completion(&vfork);
1500                 }
1501
1502                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1503                         /*
1504                          * We'll start up with an immediate SIGSTOP.
1505                          */
1506                         sigaddset(&p->pending.signal, SIGSTOP);
1507                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1508                 }
1509
1510                 if (!(clone_flags & CLONE_STOPPED))
1511                         wake_up_new_task(p, clone_flags);
1512                 else
1513                         p->state = TASK_STOPPED;
1514
1515                 if (unlikely (trace)) {
1516                         current->ptrace_message = nr;
1517                         ptrace_notify ((trace << 8) | SIGTRAP);
1518                 }
1519
1520                 if (clone_flags & CLONE_VFORK) {
1521                         freezer_do_not_count();
1522                         wait_for_completion(&vfork);
1523                         freezer_count();
1524                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1525                                 current->ptrace_message = nr;
1526                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1527                         }
1528                 }
1529         } else {
1530                 nr = PTR_ERR(p);
1531         }
1532         return nr;
1533 }
1534
1535 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1536 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1537 #endif
1538
1539 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1540 {
1541         struct sighand_struct *sighand = data;
1542
1543         spin_lock_init(&sighand->siglock);
1544         init_waitqueue_head(&sighand->signalfd_wqh);
1545 }
1546
1547 void __init proc_caches_init(void)
1548 {
1549         sighand_cachep = kmem_cache_create("sighand_cache",
1550                         sizeof(struct sighand_struct), 0,
1551                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1552                         sighand_ctor);
1553         signal_cachep = kmem_cache_create("signal_cache",
1554                         sizeof(struct signal_struct), 0,
1555                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1556         files_cachep = kmem_cache_create("files_cache",
1557                         sizeof(struct files_struct), 0,
1558                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1559         fs_cachep = kmem_cache_create("fs_cache",
1560                         sizeof(struct fs_struct), 0,
1561                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1562         vm_area_cachep = kmem_cache_create("vm_area_struct",
1563                         sizeof(struct vm_area_struct), 0,
1564                         SLAB_PANIC, NULL);
1565         mm_cachep = kmem_cache_create("mm_struct",
1566                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1567                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1568 }
1569
1570 /*
1571  * Check constraints on flags passed to the unshare system call and
1572  * force unsharing of additional process context as appropriate.
1573  */
1574 static void check_unshare_flags(unsigned long *flags_ptr)
1575 {
1576         /*
1577          * If unsharing a thread from a thread group, must also
1578          * unshare vm.
1579          */
1580         if (*flags_ptr & CLONE_THREAD)
1581                 *flags_ptr |= CLONE_VM;
1582
1583         /*
1584          * If unsharing vm, must also unshare signal handlers.
1585          */
1586         if (*flags_ptr & CLONE_VM)
1587                 *flags_ptr |= CLONE_SIGHAND;
1588
1589         /*
1590          * If unsharing signal handlers and the task was created
1591          * using CLONE_THREAD, then must unshare the thread
1592          */
1593         if ((*flags_ptr & CLONE_SIGHAND) &&
1594             (atomic_read(&current->signal->count) > 1))
1595                 *flags_ptr |= CLONE_THREAD;
1596
1597         /*
1598          * If unsharing namespace, must also unshare filesystem information.
1599          */
1600         if (*flags_ptr & CLONE_NEWNS)
1601                 *flags_ptr |= CLONE_FS;
1602 }
1603
1604 /*
1605  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1606  */
1607 static int unshare_thread(unsigned long unshare_flags)
1608 {
1609         if (unshare_flags & CLONE_THREAD)
1610                 return -EINVAL;
1611
1612         return 0;
1613 }
1614
1615 /*
1616  * Unshare the filesystem structure if it is being shared
1617  */
1618 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1619 {
1620         struct fs_struct *fs = current->fs;
1621
1622         if ((unshare_flags & CLONE_FS) &&
1623             (fs && atomic_read(&fs->count) > 1)) {
1624                 *new_fsp = __copy_fs_struct(current->fs);
1625                 if (!*new_fsp)
1626                         return -ENOMEM;
1627         }
1628
1629         return 0;
1630 }
1631
1632 /*
1633  * Unsharing of sighand is not supported yet
1634  */
1635 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1636 {
1637         struct sighand_struct *sigh = current->sighand;
1638
1639         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1640                 return -EINVAL;
1641         else
1642                 return 0;
1643 }
1644
1645 /*
1646  * Unshare vm if it is being shared
1647  */
1648 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1649 {
1650         struct mm_struct *mm = current->mm;
1651
1652         if ((unshare_flags & CLONE_VM) &&
1653             (mm && atomic_read(&mm->mm_users) > 1)) {
1654                 return -EINVAL;
1655         }
1656
1657         return 0;
1658 }
1659
1660 /*
1661  * Unshare file descriptor table if it is being shared
1662  */
1663 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1664 {
1665         struct files_struct *fd = current->files;
1666         int error = 0;
1667
1668         if ((unshare_flags & CLONE_FILES) &&
1669             (fd && atomic_read(&fd->count) > 1)) {
1670                 *new_fdp = dup_fd(fd, &error);
1671                 if (!*new_fdp)
1672                         return error;
1673         }
1674
1675         return 0;
1676 }
1677
1678 /*
1679  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1680  * supported yet
1681  */
1682 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1683 {
1684         if (unshare_flags & CLONE_SYSVSEM)
1685                 return -EINVAL;
1686
1687         return 0;
1688 }
1689
1690 /*
1691  * unshare allows a process to 'unshare' part of the process
1692  * context which was originally shared using clone.  copy_*
1693  * functions used by do_fork() cannot be used here directly
1694  * because they modify an inactive task_struct that is being
1695  * constructed. Here we are modifying the current, active,
1696  * task_struct.
1697  */
1698 asmlinkage long sys_unshare(unsigned long unshare_flags)
1699 {
1700         int err = 0;
1701         struct fs_struct *fs, *new_fs = NULL;
1702         struct sighand_struct *new_sigh = NULL;
1703         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1704         struct files_struct *fd, *new_fd = NULL;
1705         struct sem_undo_list *new_ulist = NULL;
1706         struct nsproxy *new_nsproxy = NULL;
1707
1708         check_unshare_flags(&unshare_flags);
1709
1710         /* Return -EINVAL for all unsupported flags */
1711         err = -EINVAL;
1712         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1713                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1714                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1715                                 CLONE_NEWNET))
1716                 goto bad_unshare_out;
1717
1718         if ((err = unshare_thread(unshare_flags)))
1719                 goto bad_unshare_out;
1720         if ((err = unshare_fs(unshare_flags, &new_fs)))
1721                 goto bad_unshare_cleanup_thread;
1722         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1723                 goto bad_unshare_cleanup_fs;
1724         if ((err = unshare_vm(unshare_flags, &new_mm)))
1725                 goto bad_unshare_cleanup_sigh;
1726         if ((err = unshare_fd(unshare_flags, &new_fd)))
1727                 goto bad_unshare_cleanup_vm;
1728         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1729                 goto bad_unshare_cleanup_fd;
1730         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1731                         new_fs)))
1732                 goto bad_unshare_cleanup_semundo;
1733
1734         if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
1735
1736                 if (new_nsproxy) {
1737                         switch_task_namespaces(current, new_nsproxy);
1738                         new_nsproxy = NULL;
1739                 }
1740
1741                 task_lock(current);
1742
1743                 if (new_fs) {
1744                         fs = current->fs;
1745                         current->fs = new_fs;
1746                         new_fs = fs;
1747                 }
1748
1749                 if (new_mm) {
1750                         mm = current->mm;
1751                         active_mm = current->active_mm;
1752                         current->mm = new_mm;
1753                         current->active_mm = new_mm;
1754                         activate_mm(active_mm, new_mm);
1755                         new_mm = mm;
1756                 }
1757
1758                 if (new_fd) {
1759                         fd = current->files;
1760                         current->files = new_fd;
1761                         new_fd = fd;
1762                 }
1763
1764                 task_unlock(current);
1765         }
1766
1767         if (new_nsproxy)
1768                 put_nsproxy(new_nsproxy);
1769
1770 bad_unshare_cleanup_semundo:
1771 bad_unshare_cleanup_fd:
1772         if (new_fd)
1773                 put_files_struct(new_fd);
1774
1775 bad_unshare_cleanup_vm:
1776         if (new_mm)
1777                 mmput(new_mm);
1778
1779 bad_unshare_cleanup_sigh:
1780         if (new_sigh)
1781                 if (atomic_dec_and_test(&new_sigh->count))
1782                         kmem_cache_free(sighand_cachep, new_sigh);
1783
1784 bad_unshare_cleanup_fs:
1785         if (new_fs)
1786                 put_fs_struct(new_fs);
1787
1788 bad_unshare_cleanup_thread:
1789 bad_unshare_out:
1790         return err;
1791 }