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