Merge rsync://rsync.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/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/capability.h>
31 #include <linux/cpu.h>
32 #include <linux/cpuset.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/rcupdate.h>
39 #include <linux/ptrace.h>
40 #include <linux/mount.h>
41 #include <linux/audit.h>
42 #include <linux/profile.h>
43 #include <linux/rmap.h>
44 #include <linux/acct.h>
45 #include <linux/cn_proc.h>
46 #include <linux/delayacct.h>
47 #include <linux/taskstats_kern.h>
48
49 #include <asm/pgtable.h>
50 #include <asm/pgalloc.h>
51 #include <asm/uaccess.h>
52 #include <asm/mmu_context.h>
53 #include <asm/cacheflush.h>
54 #include <asm/tlbflush.h>
55
56 /*
57  * Protected counters by write_lock_irq(&tasklist_lock)
58  */
59 unsigned long total_forks;      /* Handle normal Linux uptimes. */
60 int nr_threads;                 /* The idle threads do not count.. */
61
62 int max_threads;                /* tunable limit on nr_threads */
63
64 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
65
66 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
67
68 int nr_processes(void)
69 {
70         int cpu;
71         int total = 0;
72
73         for_each_online_cpu(cpu)
74                 total += per_cpu(process_counts, cpu);
75
76         return total;
77 }
78
79 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
80 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
81 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
82 static kmem_cache_t *task_struct_cachep;
83 #endif
84
85 /* SLAB cache for signal_struct structures (tsk->signal) */
86 static kmem_cache_t *signal_cachep;
87
88 /* SLAB cache for sighand_struct structures (tsk->sighand) */
89 kmem_cache_t *sighand_cachep;
90
91 /* SLAB cache for files_struct structures (tsk->files) */
92 kmem_cache_t *files_cachep;
93
94 /* SLAB cache for fs_struct structures (tsk->fs) */
95 kmem_cache_t *fs_cachep;
96
97 /* SLAB cache for vm_area_struct structures */
98 kmem_cache_t *vm_area_cachep;
99
100 /* SLAB cache for mm_struct structures (tsk->mm) */
101 static kmem_cache_t *mm_cachep;
102
103 void free_task(struct task_struct *tsk)
104 {
105         free_thread_info(tsk->thread_info);
106         rt_mutex_debug_task_free(tsk);
107         free_task_struct(tsk);
108 }
109 EXPORT_SYMBOL(free_task);
110
111 void __put_task_struct(struct task_struct *tsk)
112 {
113         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
114         WARN_ON(atomic_read(&tsk->usage));
115         WARN_ON(tsk == current);
116
117         security_task_free(tsk);
118         free_uid(tsk->user);
119         put_group_info(tsk->group_info);
120         delayacct_tsk_free(tsk);
121
122         if (!profile_handoff_task(tsk))
123                 free_task(tsk);
124 }
125
126 void __init fork_init(unsigned long mempages)
127 {
128 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
129 #ifndef ARCH_MIN_TASKALIGN
130 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
131 #endif
132         /* create a slab on which task_structs can be allocated */
133         task_struct_cachep =
134                 kmem_cache_create("task_struct", sizeof(struct task_struct),
135                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
136 #endif
137
138         /*
139          * The default maximum number of threads is set to a safe
140          * value: the thread structures can take up at most half
141          * of memory.
142          */
143         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
144
145         /*
146          * we need to allow at least 20 threads to boot a system
147          */
148         if(max_threads < 20)
149                 max_threads = 20;
150
151         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
152         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
153         init_task.signal->rlim[RLIMIT_SIGPENDING] =
154                 init_task.signal->rlim[RLIMIT_NPROC];
155 }
156
157 static struct task_struct *dup_task_struct(struct task_struct *orig)
158 {
159         struct task_struct *tsk;
160         struct thread_info *ti;
161
162         prepare_to_copy(orig);
163
164         tsk = alloc_task_struct();
165         if (!tsk)
166                 return NULL;
167
168         ti = alloc_thread_info(tsk);
169         if (!ti) {
170                 free_task_struct(tsk);
171                 return NULL;
172         }
173
174         *tsk = *orig;
175         tsk->thread_info = ti;
176         setup_thread_stack(tsk, orig);
177
178         /* One for us, one for whoever does the "release_task()" (usually parent) */
179         atomic_set(&tsk->usage,2);
180         atomic_set(&tsk->fs_excl, 0);
181         tsk->btrace_seq = 0;
182         tsk->splice_pipe = NULL;
183         return tsk;
184 }
185
186 #ifdef CONFIG_MMU
187 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
188 {
189         struct vm_area_struct *mpnt, *tmp, **pprev;
190         struct rb_node **rb_link, *rb_parent;
191         int retval;
192         unsigned long charge;
193         struct mempolicy *pol;
194
195         down_write(&oldmm->mmap_sem);
196         flush_cache_mm(oldmm);
197         /*
198          * Not linked in yet - no deadlock potential:
199          */
200         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
201
202         mm->locked_vm = 0;
203         mm->mmap = NULL;
204         mm->mmap_cache = NULL;
205         mm->free_area_cache = oldmm->mmap_base;
206         mm->cached_hole_size = ~0UL;
207         mm->map_count = 0;
208         cpus_clear(mm->cpu_vm_mask);
209         mm->mm_rb = RB_ROOT;
210         rb_link = &mm->mm_rb.rb_node;
211         rb_parent = NULL;
212         pprev = &mm->mmap;
213
214         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
215                 struct file *file;
216
217                 if (mpnt->vm_flags & VM_DONTCOPY) {
218                         long pages = vma_pages(mpnt);
219                         mm->total_vm -= pages;
220                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
221                                                                 -pages);
222                         continue;
223                 }
224                 charge = 0;
225                 if (mpnt->vm_flags & VM_ACCOUNT) {
226                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
227                         if (security_vm_enough_memory(len))
228                                 goto fail_nomem;
229                         charge = len;
230                 }
231                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
232                 if (!tmp)
233                         goto fail_nomem;
234                 *tmp = *mpnt;
235                 pol = mpol_copy(vma_policy(mpnt));
236                 retval = PTR_ERR(pol);
237                 if (IS_ERR(pol))
238                         goto fail_nomem_policy;
239                 vma_set_policy(tmp, pol);
240                 tmp->vm_flags &= ~VM_LOCKED;
241                 tmp->vm_mm = mm;
242                 tmp->vm_next = NULL;
243                 anon_vma_link(tmp);
244                 file = tmp->vm_file;
245                 if (file) {
246                         struct inode *inode = file->f_dentry->d_inode;
247                         get_file(file);
248                         if (tmp->vm_flags & VM_DENYWRITE)
249                                 atomic_dec(&inode->i_writecount);
250       
251                         /* insert tmp into the share list, just after mpnt */
252                         spin_lock(&file->f_mapping->i_mmap_lock);
253                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
254                         flush_dcache_mmap_lock(file->f_mapping);
255                         vma_prio_tree_add(tmp, mpnt);
256                         flush_dcache_mmap_unlock(file->f_mapping);
257                         spin_unlock(&file->f_mapping->i_mmap_lock);
258                 }
259
260                 /*
261                  * Link in the new vma and copy the page table entries.
262                  */
263                 *pprev = tmp;
264                 pprev = &tmp->vm_next;
265
266                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
267                 rb_link = &tmp->vm_rb.rb_right;
268                 rb_parent = &tmp->vm_rb;
269
270                 mm->map_count++;
271                 retval = copy_page_range(mm, oldmm, mpnt);
272
273                 if (tmp->vm_ops && tmp->vm_ops->open)
274                         tmp->vm_ops->open(tmp);
275
276                 if (retval)
277                         goto out;
278         }
279         retval = 0;
280 out:
281         up_write(&mm->mmap_sem);
282         flush_tlb_mm(oldmm);
283         up_write(&oldmm->mmap_sem);
284         return retval;
285 fail_nomem_policy:
286         kmem_cache_free(vm_area_cachep, tmp);
287 fail_nomem:
288         retval = -ENOMEM;
289         vm_unacct_memory(charge);
290         goto out;
291 }
292
293 static inline int mm_alloc_pgd(struct mm_struct * mm)
294 {
295         mm->pgd = pgd_alloc(mm);
296         if (unlikely(!mm->pgd))
297                 return -ENOMEM;
298         return 0;
299 }
300
301 static inline void mm_free_pgd(struct mm_struct * mm)
302 {
303         pgd_free(mm->pgd);
304 }
305 #else
306 #define dup_mmap(mm, oldmm)     (0)
307 #define mm_alloc_pgd(mm)        (0)
308 #define mm_free_pgd(mm)
309 #endif /* CONFIG_MMU */
310
311  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
312
313 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
314 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
315
316 #include <linux/init_task.h>
317
318 static struct mm_struct * mm_init(struct mm_struct * mm)
319 {
320         atomic_set(&mm->mm_users, 1);
321         atomic_set(&mm->mm_count, 1);
322         init_rwsem(&mm->mmap_sem);
323         INIT_LIST_HEAD(&mm->mmlist);
324         mm->core_waiters = 0;
325         mm->nr_ptes = 0;
326         set_mm_counter(mm, file_rss, 0);
327         set_mm_counter(mm, anon_rss, 0);
328         spin_lock_init(&mm->page_table_lock);
329         rwlock_init(&mm->ioctx_list_lock);
330         mm->ioctx_list = NULL;
331         mm->free_area_cache = TASK_UNMAPPED_BASE;
332         mm->cached_hole_size = ~0UL;
333
334         if (likely(!mm_alloc_pgd(mm))) {
335                 mm->def_flags = 0;
336                 return mm;
337         }
338         free_mm(mm);
339         return NULL;
340 }
341
342 /*
343  * Allocate and initialize an mm_struct.
344  */
345 struct mm_struct * mm_alloc(void)
346 {
347         struct mm_struct * mm;
348
349         mm = allocate_mm();
350         if (mm) {
351                 memset(mm, 0, sizeof(*mm));
352                 mm = mm_init(mm);
353         }
354         return mm;
355 }
356
357 /*
358  * Called when the last reference to the mm
359  * is dropped: either by a lazy thread or by
360  * mmput. Free the page directory and the mm.
361  */
362 void fastcall __mmdrop(struct mm_struct *mm)
363 {
364         BUG_ON(mm == &init_mm);
365         mm_free_pgd(mm);
366         destroy_context(mm);
367         free_mm(mm);
368 }
369
370 /*
371  * Decrement the use count and release all resources for an mm.
372  */
373 void mmput(struct mm_struct *mm)
374 {
375         might_sleep();
376
377         if (atomic_dec_and_test(&mm->mm_users)) {
378                 exit_aio(mm);
379                 exit_mmap(mm);
380                 if (!list_empty(&mm->mmlist)) {
381                         spin_lock(&mmlist_lock);
382                         list_del(&mm->mmlist);
383                         spin_unlock(&mmlist_lock);
384                 }
385                 put_swap_token(mm);
386                 mmdrop(mm);
387         }
388 }
389 EXPORT_SYMBOL_GPL(mmput);
390
391 /**
392  * get_task_mm - acquire a reference to the task's mm
393  *
394  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
395  * this kernel workthread has transiently adopted a user mm with use_mm,
396  * to do its AIO) is not set and if so returns a reference to it, after
397  * bumping up the use count.  User must release the mm via mmput()
398  * after use.  Typically used by /proc and ptrace.
399  */
400 struct mm_struct *get_task_mm(struct task_struct *task)
401 {
402         struct mm_struct *mm;
403
404         task_lock(task);
405         mm = task->mm;
406         if (mm) {
407                 if (task->flags & PF_BORROWED_MM)
408                         mm = NULL;
409                 else
410                         atomic_inc(&mm->mm_users);
411         }
412         task_unlock(task);
413         return mm;
414 }
415 EXPORT_SYMBOL_GPL(get_task_mm);
416
417 /* Please note the differences between mmput and mm_release.
418  * mmput is called whenever we stop holding onto a mm_struct,
419  * error success whatever.
420  *
421  * mm_release is called after a mm_struct has been removed
422  * from the current process.
423  *
424  * This difference is important for error handling, when we
425  * only half set up a mm_struct for a new process and need to restore
426  * the old one.  Because we mmput the new mm_struct before
427  * restoring the old one. . .
428  * Eric Biederman 10 January 1998
429  */
430 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
431 {
432         struct completion *vfork_done = tsk->vfork_done;
433
434         /* Get rid of any cached register state */
435         deactivate_mm(tsk, mm);
436
437         /* notify parent sleeping on vfork() */
438         if (vfork_done) {
439                 tsk->vfork_done = NULL;
440                 complete(vfork_done);
441         }
442         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
443                 u32 __user * tidptr = tsk->clear_child_tid;
444                 tsk->clear_child_tid = NULL;
445
446                 /*
447                  * We don't check the error code - if userspace has
448                  * not set up a proper pointer then tough luck.
449                  */
450                 put_user(0, tidptr);
451                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
452         }
453 }
454
455 /*
456  * Allocate a new mm structure and copy contents from the
457  * mm structure of the passed in task structure.
458  */
459 static struct mm_struct *dup_mm(struct task_struct *tsk)
460 {
461         struct mm_struct *mm, *oldmm = current->mm;
462         int err;
463
464         if (!oldmm)
465                 return NULL;
466
467         mm = allocate_mm();
468         if (!mm)
469                 goto fail_nomem;
470
471         memcpy(mm, oldmm, sizeof(*mm));
472
473         if (!mm_init(mm))
474                 goto fail_nomem;
475
476         if (init_new_context(tsk, mm))
477                 goto fail_nocontext;
478
479         err = dup_mmap(mm, oldmm);
480         if (err)
481                 goto free_pt;
482
483         mm->hiwater_rss = get_mm_rss(mm);
484         mm->hiwater_vm = mm->total_vm;
485
486         return mm;
487
488 free_pt:
489         mmput(mm);
490
491 fail_nomem:
492         return NULL;
493
494 fail_nocontext:
495         /*
496          * If init_new_context() failed, we cannot use mmput() to free the mm
497          * because it calls destroy_context()
498          */
499         mm_free_pgd(mm);
500         free_mm(mm);
501         return NULL;
502 }
503
504 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
505 {
506         struct mm_struct * mm, *oldmm;
507         int retval;
508
509         tsk->min_flt = tsk->maj_flt = 0;
510         tsk->nvcsw = tsk->nivcsw = 0;
511
512         tsk->mm = NULL;
513         tsk->active_mm = NULL;
514
515         /*
516          * Are we cloning a kernel thread?
517          *
518          * We need to steal a active VM for that..
519          */
520         oldmm = current->mm;
521         if (!oldmm)
522                 return 0;
523
524         if (clone_flags & CLONE_VM) {
525                 atomic_inc(&oldmm->mm_users);
526                 mm = oldmm;
527                 goto good_mm;
528         }
529
530         retval = -ENOMEM;
531         mm = dup_mm(tsk);
532         if (!mm)
533                 goto fail_nomem;
534
535 good_mm:
536         tsk->mm = mm;
537         tsk->active_mm = mm;
538         return 0;
539
540 fail_nomem:
541         return retval;
542 }
543
544 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
545 {
546         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
547         /* We don't need to lock fs - think why ;-) */
548         if (fs) {
549                 atomic_set(&fs->count, 1);
550                 rwlock_init(&fs->lock);
551                 fs->umask = old->umask;
552                 read_lock(&old->lock);
553                 fs->rootmnt = mntget(old->rootmnt);
554                 fs->root = dget(old->root);
555                 fs->pwdmnt = mntget(old->pwdmnt);
556                 fs->pwd = dget(old->pwd);
557                 if (old->altroot) {
558                         fs->altrootmnt = mntget(old->altrootmnt);
559                         fs->altroot = dget(old->altroot);
560                 } else {
561                         fs->altrootmnt = NULL;
562                         fs->altroot = NULL;
563                 }
564                 read_unlock(&old->lock);
565         }
566         return fs;
567 }
568
569 struct fs_struct *copy_fs_struct(struct fs_struct *old)
570 {
571         return __copy_fs_struct(old);
572 }
573
574 EXPORT_SYMBOL_GPL(copy_fs_struct);
575
576 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
577 {
578         if (clone_flags & CLONE_FS) {
579                 atomic_inc(&current->fs->count);
580                 return 0;
581         }
582         tsk->fs = __copy_fs_struct(current->fs);
583         if (!tsk->fs)
584                 return -ENOMEM;
585         return 0;
586 }
587
588 static int count_open_files(struct fdtable *fdt)
589 {
590         int size = fdt->max_fdset;
591         int i;
592
593         /* Find the last open fd */
594         for (i = size/(8*sizeof(long)); i > 0; ) {
595                 if (fdt->open_fds->fds_bits[--i])
596                         break;
597         }
598         i = (i+1) * 8 * sizeof(long);
599         return i;
600 }
601
602 static struct files_struct *alloc_files(void)
603 {
604         struct files_struct *newf;
605         struct fdtable *fdt;
606
607         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
608         if (!newf)
609                 goto out;
610
611         atomic_set(&newf->count, 1);
612
613         spin_lock_init(&newf->file_lock);
614         newf->next_fd = 0;
615         fdt = &newf->fdtab;
616         fdt->max_fds = NR_OPEN_DEFAULT;
617         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
618         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
619         fdt->open_fds = (fd_set *)&newf->open_fds_init;
620         fdt->fd = &newf->fd_array[0];
621         INIT_RCU_HEAD(&fdt->rcu);
622         fdt->free_files = NULL;
623         fdt->next = NULL;
624         rcu_assign_pointer(newf->fdt, fdt);
625 out:
626         return newf;
627 }
628
629 /*
630  * Allocate a new files structure and copy contents from the
631  * passed in files structure.
632  * errorp will be valid only when the returned files_struct is NULL.
633  */
634 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
635 {
636         struct files_struct *newf;
637         struct file **old_fds, **new_fds;
638         int open_files, size, i, expand;
639         struct fdtable *old_fdt, *new_fdt;
640
641         *errorp = -ENOMEM;
642         newf = alloc_files();
643         if (!newf)
644                 goto out;
645
646         spin_lock(&oldf->file_lock);
647         old_fdt = files_fdtable(oldf);
648         new_fdt = files_fdtable(newf);
649         size = old_fdt->max_fdset;
650         open_files = count_open_files(old_fdt);
651         expand = 0;
652
653         /*
654          * Check whether we need to allocate a larger fd array or fd set.
655          * Note: we're not a clone task, so the open count won't  change.
656          */
657         if (open_files > new_fdt->max_fdset) {
658                 new_fdt->max_fdset = 0;
659                 expand = 1;
660         }
661         if (open_files > new_fdt->max_fds) {
662                 new_fdt->max_fds = 0;
663                 expand = 1;
664         }
665
666         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
667         if (expand) {
668                 spin_unlock(&oldf->file_lock);
669                 spin_lock(&newf->file_lock);
670                 *errorp = expand_files(newf, open_files-1);
671                 spin_unlock(&newf->file_lock);
672                 if (*errorp < 0)
673                         goto out_release;
674                 new_fdt = files_fdtable(newf);
675                 /*
676                  * Reacquire the oldf lock and a pointer to its fd table
677                  * who knows it may have a new bigger fd table. We need
678                  * the latest pointer.
679                  */
680                 spin_lock(&oldf->file_lock);
681                 old_fdt = files_fdtable(oldf);
682         }
683
684         old_fds = old_fdt->fd;
685         new_fds = new_fdt->fd;
686
687         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
688         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
689
690         for (i = open_files; i != 0; i--) {
691                 struct file *f = *old_fds++;
692                 if (f) {
693                         get_file(f);
694                 } else {
695                         /*
696                          * The fd may be claimed in the fd bitmap but not yet
697                          * instantiated in the files array if a sibling thread
698                          * is partway through open().  So make sure that this
699                          * fd is available to the new process.
700                          */
701                         FD_CLR(open_files - i, new_fdt->open_fds);
702                 }
703                 rcu_assign_pointer(*new_fds++, f);
704         }
705         spin_unlock(&oldf->file_lock);
706
707         /* compute the remainder to be cleared */
708         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
709
710         /* This is long word aligned thus could use a optimized version */ 
711         memset(new_fds, 0, size); 
712
713         if (new_fdt->max_fdset > open_files) {
714                 int left = (new_fdt->max_fdset-open_files)/8;
715                 int start = open_files / (8 * sizeof(unsigned long));
716
717                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
718                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
719         }
720
721 out:
722         return newf;
723
724 out_release:
725         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
726         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
727         free_fd_array(new_fdt->fd, new_fdt->max_fds);
728         kmem_cache_free(files_cachep, newf);
729         return NULL;
730 }
731
732 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
733 {
734         struct files_struct *oldf, *newf;
735         int error = 0;
736
737         /*
738          * A background process may not have any files ...
739          */
740         oldf = current->files;
741         if (!oldf)
742                 goto out;
743
744         if (clone_flags & CLONE_FILES) {
745                 atomic_inc(&oldf->count);
746                 goto out;
747         }
748
749         /*
750          * Note: we may be using current for both targets (See exec.c)
751          * This works because we cache current->files (old) as oldf. Don't
752          * break this.
753          */
754         tsk->files = NULL;
755         newf = dup_fd(oldf, &error);
756         if (!newf)
757                 goto out;
758
759         tsk->files = newf;
760         error = 0;
761 out:
762         return error;
763 }
764
765 /*
766  *      Helper to unshare the files of the current task.
767  *      We don't want to expose copy_files internals to
768  *      the exec layer of the kernel.
769  */
770
771 int unshare_files(void)
772 {
773         struct files_struct *files  = current->files;
774         int rc;
775
776         BUG_ON(!files);
777
778         /* This can race but the race causes us to copy when we don't
779            need to and drop the copy */
780         if(atomic_read(&files->count) == 1)
781         {
782                 atomic_inc(&files->count);
783                 return 0;
784         }
785         rc = copy_files(0, current);
786         if(rc)
787                 current->files = files;
788         return rc;
789 }
790
791 EXPORT_SYMBOL(unshare_files);
792
793 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
794 {
795         struct sighand_struct *sig;
796
797         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
798                 atomic_inc(&current->sighand->count);
799                 return 0;
800         }
801         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
802         rcu_assign_pointer(tsk->sighand, sig);
803         if (!sig)
804                 return -ENOMEM;
805         atomic_set(&sig->count, 1);
806         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
807         return 0;
808 }
809
810 void __cleanup_sighand(struct sighand_struct *sighand)
811 {
812         if (atomic_dec_and_test(&sighand->count))
813                 kmem_cache_free(sighand_cachep, sighand);
814 }
815
816 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
817 {
818         struct signal_struct *sig;
819         int ret;
820
821         if (clone_flags & CLONE_THREAD) {
822                 atomic_inc(&current->signal->count);
823                 atomic_inc(&current->signal->live);
824                 taskstats_tgid_alloc(current->signal);
825                 return 0;
826         }
827         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
828         tsk->signal = sig;
829         if (!sig)
830                 return -ENOMEM;
831
832         ret = copy_thread_group_keys(tsk);
833         if (ret < 0) {
834                 kmem_cache_free(signal_cachep, sig);
835                 return ret;
836         }
837
838         atomic_set(&sig->count, 1);
839         atomic_set(&sig->live, 1);
840         init_waitqueue_head(&sig->wait_chldexit);
841         sig->flags = 0;
842         sig->group_exit_code = 0;
843         sig->group_exit_task = NULL;
844         sig->group_stop_count = 0;
845         sig->curr_target = NULL;
846         init_sigpending(&sig->shared_pending);
847         INIT_LIST_HEAD(&sig->posix_timers);
848
849         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
850         sig->it_real_incr.tv64 = 0;
851         sig->real_timer.function = it_real_fn;
852         sig->tsk = tsk;
853
854         sig->it_virt_expires = cputime_zero;
855         sig->it_virt_incr = cputime_zero;
856         sig->it_prof_expires = cputime_zero;
857         sig->it_prof_incr = cputime_zero;
858
859         sig->leader = 0;        /* session leadership doesn't inherit */
860         sig->tty_old_pgrp = 0;
861
862         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
863         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
864         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
865         sig->sched_time = 0;
866         INIT_LIST_HEAD(&sig->cpu_timers[0]);
867         INIT_LIST_HEAD(&sig->cpu_timers[1]);
868         INIT_LIST_HEAD(&sig->cpu_timers[2]);
869         taskstats_tgid_init(sig);
870
871         task_lock(current->group_leader);
872         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
873         task_unlock(current->group_leader);
874
875         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
876                 /*
877                  * New sole thread in the process gets an expiry time
878                  * of the whole CPU time limit.
879                  */
880                 tsk->it_prof_expires =
881                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
882         }
883         acct_init_pacct(&sig->pacct);
884
885         return 0;
886 }
887
888 void __cleanup_signal(struct signal_struct *sig)
889 {
890         exit_thread_group_keys(sig);
891         taskstats_tgid_free(sig);
892         kmem_cache_free(signal_cachep, sig);
893 }
894
895 static inline void cleanup_signal(struct task_struct *tsk)
896 {
897         struct signal_struct *sig = tsk->signal;
898
899         atomic_dec(&sig->live);
900
901         if (atomic_dec_and_test(&sig->count))
902                 __cleanup_signal(sig);
903 }
904
905 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
906 {
907         unsigned long new_flags = p->flags;
908
909         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
910         new_flags |= PF_FORKNOEXEC;
911         if (!(clone_flags & CLONE_PTRACE))
912                 p->ptrace = 0;
913         p->flags = new_flags;
914 }
915
916 asmlinkage long sys_set_tid_address(int __user *tidptr)
917 {
918         current->clear_child_tid = tidptr;
919
920         return current->pid;
921 }
922
923 static inline void rt_mutex_init_task(struct task_struct *p)
924 {
925 #ifdef CONFIG_RT_MUTEXES
926         spin_lock_init(&p->pi_lock);
927         plist_head_init(&p->pi_waiters, &p->pi_lock);
928         p->pi_blocked_on = NULL;
929 #endif
930 }
931
932 /*
933  * This creates a new process as a copy of the old one,
934  * but does not actually start it yet.
935  *
936  * It copies the registers, and all the appropriate
937  * parts of the process environment (as per the clone
938  * flags). The actual kick-off is left to the caller.
939  */
940 static struct task_struct *copy_process(unsigned long clone_flags,
941                                         unsigned long stack_start,
942                                         struct pt_regs *regs,
943                                         unsigned long stack_size,
944                                         int __user *parent_tidptr,
945                                         int __user *child_tidptr,
946                                         int pid)
947 {
948         int retval;
949         struct task_struct *p = NULL;
950
951         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
952                 return ERR_PTR(-EINVAL);
953
954         /*
955          * Thread groups must share signals as well, and detached threads
956          * can only be started up within the thread group.
957          */
958         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
959                 return ERR_PTR(-EINVAL);
960
961         /*
962          * Shared signal handlers imply shared VM. By way of the above,
963          * thread groups also imply shared VM. Blocking this case allows
964          * for various simplifications in other code.
965          */
966         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
967                 return ERR_PTR(-EINVAL);
968
969         retval = security_task_create(clone_flags);
970         if (retval)
971                 goto fork_out;
972
973         retval = -ENOMEM;
974         p = dup_task_struct(current);
975         if (!p)
976                 goto fork_out;
977
978 #ifdef CONFIG_TRACE_IRQFLAGS
979         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
980         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
981 #endif
982         retval = -EAGAIN;
983         if (atomic_read(&p->user->processes) >=
984                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
985                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
986                                 p->user != &root_user)
987                         goto bad_fork_free;
988         }
989
990         atomic_inc(&p->user->__count);
991         atomic_inc(&p->user->processes);
992         get_group_info(p->group_info);
993
994         /*
995          * If multiple threads are within copy_process(), then this check
996          * triggers too late. This doesn't hurt, the check is only there
997          * to stop root fork bombs.
998          */
999         if (nr_threads >= max_threads)
1000                 goto bad_fork_cleanup_count;
1001
1002         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1003                 goto bad_fork_cleanup_count;
1004
1005         if (p->binfmt && !try_module_get(p->binfmt->module))
1006                 goto bad_fork_cleanup_put_domain;
1007
1008         p->did_exec = 0;
1009         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1010         copy_flags(clone_flags, p);
1011         p->pid = pid;
1012         retval = -EFAULT;
1013         if (clone_flags & CLONE_PARENT_SETTID)
1014                 if (put_user(p->pid, parent_tidptr))
1015                         goto bad_fork_cleanup_delays_binfmt;
1016
1017         INIT_LIST_HEAD(&p->children);
1018         INIT_LIST_HEAD(&p->sibling);
1019         p->vfork_done = NULL;
1020         spin_lock_init(&p->alloc_lock);
1021
1022         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1023         init_sigpending(&p->pending);
1024
1025         p->utime = cputime_zero;
1026         p->stime = cputime_zero;
1027         p->sched_time = 0;
1028         p->rchar = 0;           /* I/O counter: bytes read */
1029         p->wchar = 0;           /* I/O counter: bytes written */
1030         p->syscr = 0;           /* I/O counter: read syscalls */
1031         p->syscw = 0;           /* I/O counter: write syscalls */
1032         acct_clear_integrals(p);
1033
1034         p->it_virt_expires = cputime_zero;
1035         p->it_prof_expires = cputime_zero;
1036         p->it_sched_expires = 0;
1037         INIT_LIST_HEAD(&p->cpu_timers[0]);
1038         INIT_LIST_HEAD(&p->cpu_timers[1]);
1039         INIT_LIST_HEAD(&p->cpu_timers[2]);
1040
1041         p->lock_depth = -1;             /* -1 = no lock */
1042         do_posix_clock_monotonic_gettime(&p->start_time);
1043         p->security = NULL;
1044         p->io_context = NULL;
1045         p->io_wait = NULL;
1046         p->audit_context = NULL;
1047         cpuset_fork(p);
1048 #ifdef CONFIG_NUMA
1049         p->mempolicy = mpol_copy(p->mempolicy);
1050         if (IS_ERR(p->mempolicy)) {
1051                 retval = PTR_ERR(p->mempolicy);
1052                 p->mempolicy = NULL;
1053                 goto bad_fork_cleanup_cpuset;
1054         }
1055         mpol_fix_fork_child_flag(p);
1056 #endif
1057 #ifdef CONFIG_TRACE_IRQFLAGS
1058         p->irq_events = 0;
1059         p->hardirqs_enabled = 0;
1060         p->hardirq_enable_ip = 0;
1061         p->hardirq_enable_event = 0;
1062         p->hardirq_disable_ip = _THIS_IP_;
1063         p->hardirq_disable_event = 0;
1064         p->softirqs_enabled = 1;
1065         p->softirq_enable_ip = _THIS_IP_;
1066         p->softirq_enable_event = 0;
1067         p->softirq_disable_ip = 0;
1068         p->softirq_disable_event = 0;
1069         p->hardirq_context = 0;
1070         p->softirq_context = 0;
1071 #endif
1072 #ifdef CONFIG_LOCKDEP
1073         p->lockdep_depth = 0; /* no locks held yet */
1074         p->curr_chain_key = 0;
1075         p->lockdep_recursion = 0;
1076 #endif
1077
1078         rt_mutex_init_task(p);
1079
1080 #ifdef CONFIG_DEBUG_MUTEXES
1081         p->blocked_on = NULL; /* not blocked yet */
1082 #endif
1083
1084         p->tgid = p->pid;
1085         if (clone_flags & CLONE_THREAD)
1086                 p->tgid = current->tgid;
1087
1088         if ((retval = security_task_alloc(p)))
1089                 goto bad_fork_cleanup_policy;
1090         if ((retval = audit_alloc(p)))
1091                 goto bad_fork_cleanup_security;
1092         /* copy all the process information */
1093         if ((retval = copy_semundo(clone_flags, p)))
1094                 goto bad_fork_cleanup_audit;
1095         if ((retval = copy_files(clone_flags, p)))
1096                 goto bad_fork_cleanup_semundo;
1097         if ((retval = copy_fs(clone_flags, p)))
1098                 goto bad_fork_cleanup_files;
1099         if ((retval = copy_sighand(clone_flags, p)))
1100                 goto bad_fork_cleanup_fs;
1101         if ((retval = copy_signal(clone_flags, p)))
1102                 goto bad_fork_cleanup_sighand;
1103         if ((retval = copy_mm(clone_flags, p)))
1104                 goto bad_fork_cleanup_signal;
1105         if ((retval = copy_keys(clone_flags, p)))
1106                 goto bad_fork_cleanup_mm;
1107         if ((retval = copy_namespace(clone_flags, p)))
1108                 goto bad_fork_cleanup_keys;
1109         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1110         if (retval)
1111                 goto bad_fork_cleanup_namespace;
1112
1113         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1114         /*
1115          * Clear TID on mm_release()?
1116          */
1117         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1118         p->robust_list = NULL;
1119 #ifdef CONFIG_COMPAT
1120         p->compat_robust_list = NULL;
1121 #endif
1122         INIT_LIST_HEAD(&p->pi_state_list);
1123         p->pi_state_cache = NULL;
1124
1125         /*
1126          * sigaltstack should be cleared when sharing the same VM
1127          */
1128         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1129                 p->sas_ss_sp = p->sas_ss_size = 0;
1130
1131         /*
1132          * Syscall tracing should be turned off in the child regardless
1133          * of CLONE_PTRACE.
1134          */
1135         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1136 #ifdef TIF_SYSCALL_EMU
1137         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1138 #endif
1139
1140         /* Our parent execution domain becomes current domain
1141            These must match for thread signalling to apply */
1142            
1143         p->parent_exec_id = p->self_exec_id;
1144
1145         /* ok, now we should be set up.. */
1146         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1147         p->pdeath_signal = 0;
1148         p->exit_state = 0;
1149
1150         /*
1151          * Ok, make it visible to the rest of the system.
1152          * We dont wake it up yet.
1153          */
1154         p->group_leader = p;
1155         INIT_LIST_HEAD(&p->thread_group);
1156         INIT_LIST_HEAD(&p->ptrace_children);
1157         INIT_LIST_HEAD(&p->ptrace_list);
1158
1159         /* Perform scheduler related setup. Assign this task to a CPU. */
1160         sched_fork(p, clone_flags);
1161
1162         /* Need tasklist lock for parent etc handling! */
1163         write_lock_irq(&tasklist_lock);
1164
1165         /*
1166          * The task hasn't been attached yet, so its cpus_allowed mask will
1167          * not be changed, nor will its assigned CPU.
1168          *
1169          * The cpus_allowed mask of the parent may have changed after it was
1170          * copied first time - so re-copy it here, then check the child's CPU
1171          * to ensure it is on a valid CPU (and if not, just force it back to
1172          * parent's CPU). This avoids alot of nasty races.
1173          */
1174         p->cpus_allowed = current->cpus_allowed;
1175         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1176                         !cpu_online(task_cpu(p))))
1177                 set_task_cpu(p, smp_processor_id());
1178
1179         /* CLONE_PARENT re-uses the old parent */
1180         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1181                 p->real_parent = current->real_parent;
1182         else
1183                 p->real_parent = current;
1184         p->parent = p->real_parent;
1185
1186         spin_lock(&current->sighand->siglock);
1187
1188         /*
1189          * Process group and session signals need to be delivered to just the
1190          * parent before the fork or both the parent and the child after the
1191          * fork. Restart if a signal comes in before we add the new process to
1192          * it's process group.
1193          * A fatal signal pending means that current will exit, so the new
1194          * thread can't slip out of an OOM kill (or normal SIGKILL).
1195          */
1196         recalc_sigpending();
1197         if (signal_pending(current)) {
1198                 spin_unlock(&current->sighand->siglock);
1199                 write_unlock_irq(&tasklist_lock);
1200                 retval = -ERESTARTNOINTR;
1201                 goto bad_fork_cleanup_namespace;
1202         }
1203
1204         if (clone_flags & CLONE_THREAD) {
1205                 p->group_leader = current->group_leader;
1206                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1207
1208                 if (!cputime_eq(current->signal->it_virt_expires,
1209                                 cputime_zero) ||
1210                     !cputime_eq(current->signal->it_prof_expires,
1211                                 cputime_zero) ||
1212                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1213                     !list_empty(&current->signal->cpu_timers[0]) ||
1214                     !list_empty(&current->signal->cpu_timers[1]) ||
1215                     !list_empty(&current->signal->cpu_timers[2])) {
1216                         /*
1217                          * Have child wake up on its first tick to check
1218                          * for process CPU timers.
1219                          */
1220                         p->it_prof_expires = jiffies_to_cputime(1);
1221                 }
1222         }
1223
1224         /*
1225          * inherit ioprio
1226          */
1227         p->ioprio = current->ioprio;
1228
1229         if (likely(p->pid)) {
1230                 add_parent(p);
1231                 if (unlikely(p->ptrace & PT_PTRACED))
1232                         __ptrace_link(p, current->parent);
1233
1234                 if (thread_group_leader(p)) {
1235                         p->signal->tty = current->signal->tty;
1236                         p->signal->pgrp = process_group(current);
1237                         p->signal->session = current->signal->session;
1238                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1239                         attach_pid(p, PIDTYPE_SID, p->signal->session);
1240
1241                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1242                         __get_cpu_var(process_counts)++;
1243                 }
1244                 attach_pid(p, PIDTYPE_PID, p->pid);
1245                 nr_threads++;
1246         }
1247
1248         total_forks++;
1249         spin_unlock(&current->sighand->siglock);
1250         write_unlock_irq(&tasklist_lock);
1251         proc_fork_connector(p);
1252         return p;
1253
1254 bad_fork_cleanup_namespace:
1255         exit_namespace(p);
1256 bad_fork_cleanup_keys:
1257         exit_keys(p);
1258 bad_fork_cleanup_mm:
1259         if (p->mm)
1260                 mmput(p->mm);
1261 bad_fork_cleanup_signal:
1262         cleanup_signal(p);
1263 bad_fork_cleanup_sighand:
1264         __cleanup_sighand(p->sighand);
1265 bad_fork_cleanup_fs:
1266         exit_fs(p); /* blocking */
1267 bad_fork_cleanup_files:
1268         exit_files(p); /* blocking */
1269 bad_fork_cleanup_semundo:
1270         exit_sem(p);
1271 bad_fork_cleanup_audit:
1272         audit_free(p);
1273 bad_fork_cleanup_security:
1274         security_task_free(p);
1275 bad_fork_cleanup_policy:
1276 #ifdef CONFIG_NUMA
1277         mpol_free(p->mempolicy);
1278 bad_fork_cleanup_cpuset:
1279 #endif
1280         cpuset_exit(p);
1281 bad_fork_cleanup_delays_binfmt:
1282         delayacct_tsk_free(p);
1283         if (p->binfmt)
1284                 module_put(p->binfmt->module);
1285 bad_fork_cleanup_put_domain:
1286         module_put(task_thread_info(p)->exec_domain->module);
1287 bad_fork_cleanup_count:
1288         put_group_info(p->group_info);
1289         atomic_dec(&p->user->processes);
1290         free_uid(p->user);
1291 bad_fork_free:
1292         free_task(p);
1293 fork_out:
1294         return ERR_PTR(retval);
1295 }
1296
1297 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1298 {
1299         memset(regs, 0, sizeof(struct pt_regs));
1300         return regs;
1301 }
1302
1303 struct task_struct * __devinit fork_idle(int cpu)
1304 {
1305         struct task_struct *task;
1306         struct pt_regs regs;
1307
1308         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1309         if (!task)
1310                 return ERR_PTR(-ENOMEM);
1311         init_idle(task, cpu);
1312
1313         return task;
1314 }
1315
1316 static inline int fork_traceflag (unsigned clone_flags)
1317 {
1318         if (clone_flags & CLONE_UNTRACED)
1319                 return 0;
1320         else if (clone_flags & CLONE_VFORK) {
1321                 if (current->ptrace & PT_TRACE_VFORK)
1322                         return PTRACE_EVENT_VFORK;
1323         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1324                 if (current->ptrace & PT_TRACE_CLONE)
1325                         return PTRACE_EVENT_CLONE;
1326         } else if (current->ptrace & PT_TRACE_FORK)
1327                 return PTRACE_EVENT_FORK;
1328
1329         return 0;
1330 }
1331
1332 /*
1333  *  Ok, this is the main fork-routine.
1334  *
1335  * It copies the process, and if successful kick-starts
1336  * it and waits for it to finish using the VM if required.
1337  */
1338 long do_fork(unsigned long clone_flags,
1339               unsigned long stack_start,
1340               struct pt_regs *regs,
1341               unsigned long stack_size,
1342               int __user *parent_tidptr,
1343               int __user *child_tidptr)
1344 {
1345         struct task_struct *p;
1346         int trace = 0;
1347         struct pid *pid = alloc_pid();
1348         long nr;
1349
1350         if (!pid)
1351                 return -EAGAIN;
1352         nr = pid->nr;
1353         if (unlikely(current->ptrace)) {
1354                 trace = fork_traceflag (clone_flags);
1355                 if (trace)
1356                         clone_flags |= CLONE_PTRACE;
1357         }
1358
1359         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1360         /*
1361          * Do this prior waking up the new thread - the thread pointer
1362          * might get invalid after that point, if the thread exits quickly.
1363          */
1364         if (!IS_ERR(p)) {
1365                 struct completion vfork;
1366
1367                 if (clone_flags & CLONE_VFORK) {
1368                         p->vfork_done = &vfork;
1369                         init_completion(&vfork);
1370                 }
1371
1372                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1373                         /*
1374                          * We'll start up with an immediate SIGSTOP.
1375                          */
1376                         sigaddset(&p->pending.signal, SIGSTOP);
1377                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1378                 }
1379
1380                 if (!(clone_flags & CLONE_STOPPED))
1381                         wake_up_new_task(p, clone_flags);
1382                 else
1383                         p->state = TASK_STOPPED;
1384
1385                 if (unlikely (trace)) {
1386                         current->ptrace_message = nr;
1387                         ptrace_notify ((trace << 8) | SIGTRAP);
1388                 }
1389
1390                 if (clone_flags & CLONE_VFORK) {
1391                         wait_for_completion(&vfork);
1392                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1393                                 current->ptrace_message = nr;
1394                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1395                         }
1396                 }
1397         } else {
1398                 free_pid(pid);
1399                 nr = PTR_ERR(p);
1400         }
1401         return nr;
1402 }
1403
1404 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1405 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1406 #endif
1407
1408 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1409 {
1410         struct sighand_struct *sighand = data;
1411
1412         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1413                                         SLAB_CTOR_CONSTRUCTOR)
1414                 spin_lock_init(&sighand->siglock);
1415 }
1416
1417 void __init proc_caches_init(void)
1418 {
1419         sighand_cachep = kmem_cache_create("sighand_cache",
1420                         sizeof(struct sighand_struct), 0,
1421                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1422                         sighand_ctor, NULL);
1423         signal_cachep = kmem_cache_create("signal_cache",
1424                         sizeof(struct signal_struct), 0,
1425                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1426         files_cachep = kmem_cache_create("files_cache", 
1427                         sizeof(struct files_struct), 0,
1428                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1429         fs_cachep = kmem_cache_create("fs_cache", 
1430                         sizeof(struct fs_struct), 0,
1431                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1432         vm_area_cachep = kmem_cache_create("vm_area_struct",
1433                         sizeof(struct vm_area_struct), 0,
1434                         SLAB_PANIC, NULL, NULL);
1435         mm_cachep = kmem_cache_create("mm_struct",
1436                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1437                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1438 }
1439
1440
1441 /*
1442  * Check constraints on flags passed to the unshare system call and
1443  * force unsharing of additional process context as appropriate.
1444  */
1445 static inline void check_unshare_flags(unsigned long *flags_ptr)
1446 {
1447         /*
1448          * If unsharing a thread from a thread group, must also
1449          * unshare vm.
1450          */
1451         if (*flags_ptr & CLONE_THREAD)
1452                 *flags_ptr |= CLONE_VM;
1453
1454         /*
1455          * If unsharing vm, must also unshare signal handlers.
1456          */
1457         if (*flags_ptr & CLONE_VM)
1458                 *flags_ptr |= CLONE_SIGHAND;
1459
1460         /*
1461          * If unsharing signal handlers and the task was created
1462          * using CLONE_THREAD, then must unshare the thread
1463          */
1464         if ((*flags_ptr & CLONE_SIGHAND) &&
1465             (atomic_read(&current->signal->count) > 1))
1466                 *flags_ptr |= CLONE_THREAD;
1467
1468         /*
1469          * If unsharing namespace, must also unshare filesystem information.
1470          */
1471         if (*flags_ptr & CLONE_NEWNS)
1472                 *flags_ptr |= CLONE_FS;
1473 }
1474
1475 /*
1476  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1477  */
1478 static int unshare_thread(unsigned long unshare_flags)
1479 {
1480         if (unshare_flags & CLONE_THREAD)
1481                 return -EINVAL;
1482
1483         return 0;
1484 }
1485
1486 /*
1487  * Unshare the filesystem structure if it is being shared
1488  */
1489 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1490 {
1491         struct fs_struct *fs = current->fs;
1492
1493         if ((unshare_flags & CLONE_FS) &&
1494             (fs && atomic_read(&fs->count) > 1)) {
1495                 *new_fsp = __copy_fs_struct(current->fs);
1496                 if (!*new_fsp)
1497                         return -ENOMEM;
1498         }
1499
1500         return 0;
1501 }
1502
1503 /*
1504  * Unshare the namespace structure if it is being shared
1505  */
1506 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1507 {
1508         struct namespace *ns = current->namespace;
1509
1510         if ((unshare_flags & CLONE_NEWNS) &&
1511             (ns && atomic_read(&ns->count) > 1)) {
1512                 if (!capable(CAP_SYS_ADMIN))
1513                         return -EPERM;
1514
1515                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1516                 if (!*new_nsp)
1517                         return -ENOMEM;
1518         }
1519
1520         return 0;
1521 }
1522
1523 /*
1524  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1525  * supported yet
1526  */
1527 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1528 {
1529         struct sighand_struct *sigh = current->sighand;
1530
1531         if ((unshare_flags & CLONE_SIGHAND) &&
1532             (sigh && atomic_read(&sigh->count) > 1))
1533                 return -EINVAL;
1534         else
1535                 return 0;
1536 }
1537
1538 /*
1539  * Unshare vm if it is being shared
1540  */
1541 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1542 {
1543         struct mm_struct *mm = current->mm;
1544
1545         if ((unshare_flags & CLONE_VM) &&
1546             (mm && atomic_read(&mm->mm_users) > 1)) {
1547                 return -EINVAL;
1548         }
1549
1550         return 0;
1551 }
1552
1553 /*
1554  * Unshare file descriptor table if it is being shared
1555  */
1556 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1557 {
1558         struct files_struct *fd = current->files;
1559         int error = 0;
1560
1561         if ((unshare_flags & CLONE_FILES) &&
1562             (fd && atomic_read(&fd->count) > 1)) {
1563                 *new_fdp = dup_fd(fd, &error);
1564                 if (!*new_fdp)
1565                         return error;
1566         }
1567
1568         return 0;
1569 }
1570
1571 /*
1572  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1573  * supported yet
1574  */
1575 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1576 {
1577         if (unshare_flags & CLONE_SYSVSEM)
1578                 return -EINVAL;
1579
1580         return 0;
1581 }
1582
1583 /*
1584  * unshare allows a process to 'unshare' part of the process
1585  * context which was originally shared using clone.  copy_*
1586  * functions used by do_fork() cannot be used here directly
1587  * because they modify an inactive task_struct that is being
1588  * constructed. Here we are modifying the current, active,
1589  * task_struct.
1590  */
1591 asmlinkage long sys_unshare(unsigned long unshare_flags)
1592 {
1593         int err = 0;
1594         struct fs_struct *fs, *new_fs = NULL;
1595         struct namespace *ns, *new_ns = NULL;
1596         struct sighand_struct *sigh, *new_sigh = NULL;
1597         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1598         struct files_struct *fd, *new_fd = NULL;
1599         struct sem_undo_list *new_ulist = NULL;
1600
1601         check_unshare_flags(&unshare_flags);
1602
1603         /* Return -EINVAL for all unsupported flags */
1604         err = -EINVAL;
1605         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1606                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1607                 goto bad_unshare_out;
1608
1609         if ((err = unshare_thread(unshare_flags)))
1610                 goto bad_unshare_out;
1611         if ((err = unshare_fs(unshare_flags, &new_fs)))
1612                 goto bad_unshare_cleanup_thread;
1613         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1614                 goto bad_unshare_cleanup_fs;
1615         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1616                 goto bad_unshare_cleanup_ns;
1617         if ((err = unshare_vm(unshare_flags, &new_mm)))
1618                 goto bad_unshare_cleanup_sigh;
1619         if ((err = unshare_fd(unshare_flags, &new_fd)))
1620                 goto bad_unshare_cleanup_vm;
1621         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1622                 goto bad_unshare_cleanup_fd;
1623
1624         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1625
1626                 task_lock(current);
1627
1628                 if (new_fs) {
1629                         fs = current->fs;
1630                         current->fs = new_fs;
1631                         new_fs = fs;
1632                 }
1633
1634                 if (new_ns) {
1635                         ns = current->namespace;
1636                         current->namespace = new_ns;
1637                         new_ns = ns;
1638                 }
1639
1640                 if (new_sigh) {
1641                         sigh = current->sighand;
1642                         rcu_assign_pointer(current->sighand, new_sigh);
1643                         new_sigh = sigh;
1644                 }
1645
1646                 if (new_mm) {
1647                         mm = current->mm;
1648                         active_mm = current->active_mm;
1649                         current->mm = new_mm;
1650                         current->active_mm = new_mm;
1651                         activate_mm(active_mm, new_mm);
1652                         new_mm = mm;
1653                 }
1654
1655                 if (new_fd) {
1656                         fd = current->files;
1657                         current->files = new_fd;
1658                         new_fd = fd;
1659                 }
1660
1661                 task_unlock(current);
1662         }
1663
1664 bad_unshare_cleanup_fd:
1665         if (new_fd)
1666                 put_files_struct(new_fd);
1667
1668 bad_unshare_cleanup_vm:
1669         if (new_mm)
1670                 mmput(new_mm);
1671
1672 bad_unshare_cleanup_sigh:
1673         if (new_sigh)
1674                 if (atomic_dec_and_test(&new_sigh->count))
1675                         kmem_cache_free(sighand_cachep, new_sigh);
1676
1677 bad_unshare_cleanup_ns:
1678         if (new_ns)
1679                 put_namespace(new_ns);
1680
1681 bad_unshare_cleanup_fs:
1682         if (new_fs)
1683                 put_fs_struct(new_fs);
1684
1685 bad_unshare_cleanup_thread:
1686 bad_unshare_out:
1687         return err;
1688 }