8c859eef8e6a5b825612b750608f452459626643
[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/mnt_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 struct kmem_cache *task_struct_cachep;
86 #endif
87
88 /* SLAB cache for signal_struct structures (tsk->signal) */
89 static struct kmem_cache *signal_cachep;
90
91 /* SLAB cache for sighand_struct structures (tsk->sighand) */
92 struct kmem_cache *sighand_cachep;
93
94 /* SLAB cache for files_struct structures (tsk->files) */
95 struct kmem_cache *files_cachep;
96
97 /* SLAB cache for fs_struct structures (tsk->fs) */
98 struct kmem_cache *fs_cachep;
99
100 /* SLAB cache for vm_area_struct structures */
101 struct kmem_cache *vm_area_cachep;
102
103 /* SLAB cache for mm_struct structures (tsk->mm) */
104 static struct kmem_cache *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, GFP_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_path.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, GFP_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
452         /*
453          * If we're exiting normally, clear a user-space tid field if
454          * requested.  We leave this alone when dying by signal, to leave
455          * the value intact in a core dump, and to save the unnecessary
456          * trouble otherwise.  Userland only wants this done for a sys_exit.
457          */
458         if (tsk->clear_child_tid
459             && !(tsk->flags & PF_SIGNALED)
460             && atomic_read(&mm->mm_users) > 1) {
461                 u32 __user * tidptr = tsk->clear_child_tid;
462                 tsk->clear_child_tid = NULL;
463
464                 /*
465                  * We don't check the error code - if userspace has
466                  * not set up a proper pointer then tough luck.
467                  */
468                 put_user(0, tidptr);
469                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
470         }
471 }
472
473 /*
474  * Allocate a new mm structure and copy contents from the
475  * mm structure of the passed in task structure.
476  */
477 static struct mm_struct *dup_mm(struct task_struct *tsk)
478 {
479         struct mm_struct *mm, *oldmm = current->mm;
480         int err;
481
482         if (!oldmm)
483                 return NULL;
484
485         mm = allocate_mm();
486         if (!mm)
487                 goto fail_nomem;
488
489         memcpy(mm, oldmm, sizeof(*mm));
490
491         /* Initializing for Swap token stuff */
492         mm->token_priority = 0;
493         mm->last_interval = 0;
494
495         if (!mm_init(mm))
496                 goto fail_nomem;
497
498         if (init_new_context(tsk, mm))
499                 goto fail_nocontext;
500
501         err = dup_mmap(mm, oldmm);
502         if (err)
503                 goto free_pt;
504
505         mm->hiwater_rss = get_mm_rss(mm);
506         mm->hiwater_vm = mm->total_vm;
507
508         return mm;
509
510 free_pt:
511         mmput(mm);
512
513 fail_nomem:
514         return NULL;
515
516 fail_nocontext:
517         /*
518          * If init_new_context() failed, we cannot use mmput() to free the mm
519          * because it calls destroy_context()
520          */
521         mm_free_pgd(mm);
522         free_mm(mm);
523         return NULL;
524 }
525
526 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
527 {
528         struct mm_struct * mm, *oldmm;
529         int retval;
530
531         tsk->min_flt = tsk->maj_flt = 0;
532         tsk->nvcsw = tsk->nivcsw = 0;
533
534         tsk->mm = NULL;
535         tsk->active_mm = NULL;
536
537         /*
538          * Are we cloning a kernel thread?
539          *
540          * We need to steal a active VM for that..
541          */
542         oldmm = current->mm;
543         if (!oldmm)
544                 return 0;
545
546         if (clone_flags & CLONE_VM) {
547                 atomic_inc(&oldmm->mm_users);
548                 mm = oldmm;
549                 goto good_mm;
550         }
551
552         retval = -ENOMEM;
553         mm = dup_mm(tsk);
554         if (!mm)
555                 goto fail_nomem;
556
557 good_mm:
558         /* Initializing for Swap token stuff */
559         mm->token_priority = 0;
560         mm->last_interval = 0;
561
562         tsk->mm = mm;
563         tsk->active_mm = mm;
564         return 0;
565
566 fail_nomem:
567         return retval;
568 }
569
570 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
571 {
572         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
573         /* We don't need to lock fs - think why ;-) */
574         if (fs) {
575                 atomic_set(&fs->count, 1);
576                 rwlock_init(&fs->lock);
577                 fs->umask = old->umask;
578                 read_lock(&old->lock);
579                 fs->rootmnt = mntget(old->rootmnt);
580                 fs->root = dget(old->root);
581                 fs->pwdmnt = mntget(old->pwdmnt);
582                 fs->pwd = dget(old->pwd);
583                 if (old->altroot) {
584                         fs->altrootmnt = mntget(old->altrootmnt);
585                         fs->altroot = dget(old->altroot);
586                 } else {
587                         fs->altrootmnt = NULL;
588                         fs->altroot = NULL;
589                 }
590                 read_unlock(&old->lock);
591         }
592         return fs;
593 }
594
595 struct fs_struct *copy_fs_struct(struct fs_struct *old)
596 {
597         return __copy_fs_struct(old);
598 }
599
600 EXPORT_SYMBOL_GPL(copy_fs_struct);
601
602 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
603 {
604         if (clone_flags & CLONE_FS) {
605                 atomic_inc(&current->fs->count);
606                 return 0;
607         }
608         tsk->fs = __copy_fs_struct(current->fs);
609         if (!tsk->fs)
610                 return -ENOMEM;
611         return 0;
612 }
613
614 static int count_open_files(struct fdtable *fdt)
615 {
616         int size = fdt->max_fdset;
617         int i;
618
619         /* Find the last open fd */
620         for (i = size/(8*sizeof(long)); i > 0; ) {
621                 if (fdt->open_fds->fds_bits[--i])
622                         break;
623         }
624         i = (i+1) * 8 * sizeof(long);
625         return i;
626 }
627
628 static struct files_struct *alloc_files(void)
629 {
630         struct files_struct *newf;
631         struct fdtable *fdt;
632
633         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
634         if (!newf)
635                 goto out;
636
637         atomic_set(&newf->count, 1);
638
639         spin_lock_init(&newf->file_lock);
640         newf->next_fd = 0;
641         fdt = &newf->fdtab;
642         fdt->max_fds = NR_OPEN_DEFAULT;
643         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
644         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
645         fdt->open_fds = (fd_set *)&newf->open_fds_init;
646         fdt->fd = &newf->fd_array[0];
647         INIT_RCU_HEAD(&fdt->rcu);
648         fdt->free_files = NULL;
649         fdt->next = NULL;
650         rcu_assign_pointer(newf->fdt, fdt);
651 out:
652         return newf;
653 }
654
655 /*
656  * Allocate a new files structure and copy contents from the
657  * passed in files structure.
658  * errorp will be valid only when the returned files_struct is NULL.
659  */
660 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
661 {
662         struct files_struct *newf;
663         struct file **old_fds, **new_fds;
664         int open_files, size, i, expand;
665         struct fdtable *old_fdt, *new_fdt;
666
667         *errorp = -ENOMEM;
668         newf = alloc_files();
669         if (!newf)
670                 goto out;
671
672         spin_lock(&oldf->file_lock);
673         old_fdt = files_fdtable(oldf);
674         new_fdt = files_fdtable(newf);
675         size = old_fdt->max_fdset;
676         open_files = count_open_files(old_fdt);
677         expand = 0;
678
679         /*
680          * Check whether we need to allocate a larger fd array or fd set.
681          * Note: we're not a clone task, so the open count won't  change.
682          */
683         if (open_files > new_fdt->max_fdset) {
684                 new_fdt->max_fdset = 0;
685                 expand = 1;
686         }
687         if (open_files > new_fdt->max_fds) {
688                 new_fdt->max_fds = 0;
689                 expand = 1;
690         }
691
692         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
693         if (expand) {
694                 spin_unlock(&oldf->file_lock);
695                 spin_lock(&newf->file_lock);
696                 *errorp = expand_files(newf, open_files-1);
697                 spin_unlock(&newf->file_lock);
698                 if (*errorp < 0)
699                         goto out_release;
700                 new_fdt = files_fdtable(newf);
701                 /*
702                  * Reacquire the oldf lock and a pointer to its fd table
703                  * who knows it may have a new bigger fd table. We need
704                  * the latest pointer.
705                  */
706                 spin_lock(&oldf->file_lock);
707                 old_fdt = files_fdtable(oldf);
708         }
709
710         old_fds = old_fdt->fd;
711         new_fds = new_fdt->fd;
712
713         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
714         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
715
716         for (i = open_files; i != 0; i--) {
717                 struct file *f = *old_fds++;
718                 if (f) {
719                         get_file(f);
720                 } else {
721                         /*
722                          * The fd may be claimed in the fd bitmap but not yet
723                          * instantiated in the files array if a sibling thread
724                          * is partway through open().  So make sure that this
725                          * fd is available to the new process.
726                          */
727                         FD_CLR(open_files - i, new_fdt->open_fds);
728                 }
729                 rcu_assign_pointer(*new_fds++, f);
730         }
731         spin_unlock(&oldf->file_lock);
732
733         /* compute the remainder to be cleared */
734         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
735
736         /* This is long word aligned thus could use a optimized version */ 
737         memset(new_fds, 0, size); 
738
739         if (new_fdt->max_fdset > open_files) {
740                 int left = (new_fdt->max_fdset-open_files)/8;
741                 int start = open_files / (8 * sizeof(unsigned long));
742
743                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
744                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
745         }
746
747 out:
748         return newf;
749
750 out_release:
751         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
752         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
753         free_fd_array(new_fdt->fd, new_fdt->max_fds);
754         kmem_cache_free(files_cachep, newf);
755         return NULL;
756 }
757
758 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
759 {
760         struct files_struct *oldf, *newf;
761         int error = 0;
762
763         /*
764          * A background process may not have any files ...
765          */
766         oldf = current->files;
767         if (!oldf)
768                 goto out;
769
770         if (clone_flags & CLONE_FILES) {
771                 atomic_inc(&oldf->count);
772                 goto out;
773         }
774
775         /*
776          * Note: we may be using current for both targets (See exec.c)
777          * This works because we cache current->files (old) as oldf. Don't
778          * break this.
779          */
780         tsk->files = NULL;
781         newf = dup_fd(oldf, &error);
782         if (!newf)
783                 goto out;
784
785         tsk->files = newf;
786         error = 0;
787 out:
788         return error;
789 }
790
791 /*
792  *      Helper to unshare the files of the current task.
793  *      We don't want to expose copy_files internals to
794  *      the exec layer of the kernel.
795  */
796
797 int unshare_files(void)
798 {
799         struct files_struct *files  = current->files;
800         int rc;
801
802         BUG_ON(!files);
803
804         /* This can race but the race causes us to copy when we don't
805            need to and drop the copy */
806         if(atomic_read(&files->count) == 1)
807         {
808                 atomic_inc(&files->count);
809                 return 0;
810         }
811         rc = copy_files(0, current);
812         if(rc)
813                 current->files = files;
814         return rc;
815 }
816
817 EXPORT_SYMBOL(unshare_files);
818
819 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
820 {
821         struct sighand_struct *sig;
822
823         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
824                 atomic_inc(&current->sighand->count);
825                 return 0;
826         }
827         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
828         rcu_assign_pointer(tsk->sighand, sig);
829         if (!sig)
830                 return -ENOMEM;
831         atomic_set(&sig->count, 1);
832         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
833         return 0;
834 }
835
836 void __cleanup_sighand(struct sighand_struct *sighand)
837 {
838         if (atomic_dec_and_test(&sighand->count))
839                 kmem_cache_free(sighand_cachep, sighand);
840 }
841
842 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
843 {
844         struct signal_struct *sig;
845         int ret;
846
847         if (clone_flags & CLONE_THREAD) {
848                 atomic_inc(&current->signal->count);
849                 atomic_inc(&current->signal->live);
850                 return 0;
851         }
852         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
853         tsk->signal = sig;
854         if (!sig)
855                 return -ENOMEM;
856
857         ret = copy_thread_group_keys(tsk);
858         if (ret < 0) {
859                 kmem_cache_free(signal_cachep, sig);
860                 return ret;
861         }
862
863         atomic_set(&sig->count, 1);
864         atomic_set(&sig->live, 1);
865         init_waitqueue_head(&sig->wait_chldexit);
866         sig->flags = 0;
867         sig->group_exit_code = 0;
868         sig->group_exit_task = NULL;
869         sig->group_stop_count = 0;
870         sig->curr_target = NULL;
871         init_sigpending(&sig->shared_pending);
872         INIT_LIST_HEAD(&sig->posix_timers);
873
874         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
875         sig->it_real_incr.tv64 = 0;
876         sig->real_timer.function = it_real_fn;
877         sig->tsk = tsk;
878
879         sig->it_virt_expires = cputime_zero;
880         sig->it_virt_incr = cputime_zero;
881         sig->it_prof_expires = cputime_zero;
882         sig->it_prof_incr = cputime_zero;
883
884         sig->leader = 0;        /* session leadership doesn't inherit */
885         sig->tty_old_pgrp = 0;
886
887         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
888         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
889         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
890         sig->sched_time = 0;
891         INIT_LIST_HEAD(&sig->cpu_timers[0]);
892         INIT_LIST_HEAD(&sig->cpu_timers[1]);
893         INIT_LIST_HEAD(&sig->cpu_timers[2]);
894         taskstats_tgid_init(sig);
895
896         task_lock(current->group_leader);
897         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
898         task_unlock(current->group_leader);
899
900         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
901                 /*
902                  * New sole thread in the process gets an expiry time
903                  * of the whole CPU time limit.
904                  */
905                 tsk->it_prof_expires =
906                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
907         }
908         acct_init_pacct(&sig->pacct);
909
910         return 0;
911 }
912
913 void __cleanup_signal(struct signal_struct *sig)
914 {
915         exit_thread_group_keys(sig);
916         kmem_cache_free(signal_cachep, sig);
917 }
918
919 static inline void cleanup_signal(struct task_struct *tsk)
920 {
921         struct signal_struct *sig = tsk->signal;
922
923         atomic_dec(&sig->live);
924
925         if (atomic_dec_and_test(&sig->count))
926                 __cleanup_signal(sig);
927 }
928
929 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
930 {
931         unsigned long new_flags = p->flags;
932
933         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
934         new_flags |= PF_FORKNOEXEC;
935         if (!(clone_flags & CLONE_PTRACE))
936                 p->ptrace = 0;
937         p->flags = new_flags;
938 }
939
940 asmlinkage long sys_set_tid_address(int __user *tidptr)
941 {
942         current->clear_child_tid = tidptr;
943
944         return current->pid;
945 }
946
947 static inline void rt_mutex_init_task(struct task_struct *p)
948 {
949 #ifdef CONFIG_RT_MUTEXES
950         spin_lock_init(&p->pi_lock);
951         plist_head_init(&p->pi_waiters, &p->pi_lock);
952         p->pi_blocked_on = NULL;
953 #endif
954 }
955
956 /*
957  * This creates a new process as a copy of the old one,
958  * but does not actually start it yet.
959  *
960  * It copies the registers, and all the appropriate
961  * parts of the process environment (as per the clone
962  * flags). The actual kick-off is left to the caller.
963  */
964 static struct task_struct *copy_process(unsigned long clone_flags,
965                                         unsigned long stack_start,
966                                         struct pt_regs *regs,
967                                         unsigned long stack_size,
968                                         int __user *parent_tidptr,
969                                         int __user *child_tidptr,
970                                         int pid)
971 {
972         int retval;
973         struct task_struct *p = NULL;
974
975         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
976                 return ERR_PTR(-EINVAL);
977
978         /*
979          * Thread groups must share signals as well, and detached threads
980          * can only be started up within the thread group.
981          */
982         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
983                 return ERR_PTR(-EINVAL);
984
985         /*
986          * Shared signal handlers imply shared VM. By way of the above,
987          * thread groups also imply shared VM. Blocking this case allows
988          * for various simplifications in other code.
989          */
990         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
991                 return ERR_PTR(-EINVAL);
992
993         retval = security_task_create(clone_flags);
994         if (retval)
995                 goto fork_out;
996
997         retval = -ENOMEM;
998         p = dup_task_struct(current);
999         if (!p)
1000                 goto fork_out;
1001
1002         rt_mutex_init_task(p);
1003
1004 #ifdef CONFIG_TRACE_IRQFLAGS
1005         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1006         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1007 #endif
1008         retval = -EAGAIN;
1009         if (atomic_read(&p->user->processes) >=
1010                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1011                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1012                                 p->user != &root_user)
1013                         goto bad_fork_free;
1014         }
1015
1016         atomic_inc(&p->user->__count);
1017         atomic_inc(&p->user->processes);
1018         get_group_info(p->group_info);
1019
1020         /*
1021          * If multiple threads are within copy_process(), then this check
1022          * triggers too late. This doesn't hurt, the check is only there
1023          * to stop root fork bombs.
1024          */
1025         if (nr_threads >= max_threads)
1026                 goto bad_fork_cleanup_count;
1027
1028         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1029                 goto bad_fork_cleanup_count;
1030
1031         if (p->binfmt && !try_module_get(p->binfmt->module))
1032                 goto bad_fork_cleanup_put_domain;
1033
1034         p->did_exec = 0;
1035         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1036         copy_flags(clone_flags, p);
1037         p->pid = pid;
1038         retval = -EFAULT;
1039         if (clone_flags & CLONE_PARENT_SETTID)
1040                 if (put_user(p->pid, parent_tidptr))
1041                         goto bad_fork_cleanup_delays_binfmt;
1042
1043         INIT_LIST_HEAD(&p->children);
1044         INIT_LIST_HEAD(&p->sibling);
1045         p->vfork_done = NULL;
1046         spin_lock_init(&p->alloc_lock);
1047
1048         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1049         init_sigpending(&p->pending);
1050
1051         p->utime = cputime_zero;
1052         p->stime = cputime_zero;
1053         p->sched_time = 0;
1054         p->rchar = 0;           /* I/O counter: bytes read */
1055         p->wchar = 0;           /* I/O counter: bytes written */
1056         p->syscr = 0;           /* I/O counter: read syscalls */
1057         p->syscw = 0;           /* I/O counter: write syscalls */
1058         acct_clear_integrals(p);
1059
1060         p->it_virt_expires = cputime_zero;
1061         p->it_prof_expires = cputime_zero;
1062         p->it_sched_expires = 0;
1063         INIT_LIST_HEAD(&p->cpu_timers[0]);
1064         INIT_LIST_HEAD(&p->cpu_timers[1]);
1065         INIT_LIST_HEAD(&p->cpu_timers[2]);
1066
1067         p->lock_depth = -1;             /* -1 = no lock */
1068         do_posix_clock_monotonic_gettime(&p->start_time);
1069         p->security = NULL;
1070         p->io_context = NULL;
1071         p->io_wait = NULL;
1072         p->audit_context = NULL;
1073         cpuset_fork(p);
1074 #ifdef CONFIG_NUMA
1075         p->mempolicy = mpol_copy(p->mempolicy);
1076         if (IS_ERR(p->mempolicy)) {
1077                 retval = PTR_ERR(p->mempolicy);
1078                 p->mempolicy = NULL;
1079                 goto bad_fork_cleanup_cpuset;
1080         }
1081         mpol_fix_fork_child_flag(p);
1082 #endif
1083 #ifdef CONFIG_TRACE_IRQFLAGS
1084         p->irq_events = 0;
1085 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1086         p->hardirqs_enabled = 1;
1087 #else
1088         p->hardirqs_enabled = 0;
1089 #endif
1090         p->hardirq_enable_ip = 0;
1091         p->hardirq_enable_event = 0;
1092         p->hardirq_disable_ip = _THIS_IP_;
1093         p->hardirq_disable_event = 0;
1094         p->softirqs_enabled = 1;
1095         p->softirq_enable_ip = _THIS_IP_;
1096         p->softirq_enable_event = 0;
1097         p->softirq_disable_ip = 0;
1098         p->softirq_disable_event = 0;
1099         p->hardirq_context = 0;
1100         p->softirq_context = 0;
1101 #endif
1102 #ifdef CONFIG_LOCKDEP
1103         p->lockdep_depth = 0; /* no locks held yet */
1104         p->curr_chain_key = 0;
1105         p->lockdep_recursion = 0;
1106 #endif
1107
1108 #ifdef CONFIG_DEBUG_MUTEXES
1109         p->blocked_on = NULL; /* not blocked yet */
1110 #endif
1111
1112         p->tgid = p->pid;
1113         if (clone_flags & CLONE_THREAD)
1114                 p->tgid = current->tgid;
1115
1116         if ((retval = security_task_alloc(p)))
1117                 goto bad_fork_cleanup_policy;
1118         if ((retval = audit_alloc(p)))
1119                 goto bad_fork_cleanup_security;
1120         /* copy all the process information */
1121         if ((retval = copy_semundo(clone_flags, p)))
1122                 goto bad_fork_cleanup_audit;
1123         if ((retval = copy_files(clone_flags, p)))
1124                 goto bad_fork_cleanup_semundo;
1125         if ((retval = copy_fs(clone_flags, p)))
1126                 goto bad_fork_cleanup_files;
1127         if ((retval = copy_sighand(clone_flags, p)))
1128                 goto bad_fork_cleanup_fs;
1129         if ((retval = copy_signal(clone_flags, p)))
1130                 goto bad_fork_cleanup_sighand;
1131         if ((retval = copy_mm(clone_flags, p)))
1132                 goto bad_fork_cleanup_signal;
1133         if ((retval = copy_keys(clone_flags, p)))
1134                 goto bad_fork_cleanup_mm;
1135         if ((retval = copy_namespaces(clone_flags, p)))
1136                 goto bad_fork_cleanup_keys;
1137         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1138         if (retval)
1139                 goto bad_fork_cleanup_namespaces;
1140
1141         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1142         /*
1143          * Clear TID on mm_release()?
1144          */
1145         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1146         p->robust_list = NULL;
1147 #ifdef CONFIG_COMPAT
1148         p->compat_robust_list = NULL;
1149 #endif
1150         INIT_LIST_HEAD(&p->pi_state_list);
1151         p->pi_state_cache = NULL;
1152
1153         /*
1154          * sigaltstack should be cleared when sharing the same VM
1155          */
1156         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1157                 p->sas_ss_sp = p->sas_ss_size = 0;
1158
1159         /*
1160          * Syscall tracing should be turned off in the child regardless
1161          * of CLONE_PTRACE.
1162          */
1163         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1164 #ifdef TIF_SYSCALL_EMU
1165         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1166 #endif
1167
1168         /* Our parent execution domain becomes current domain
1169            These must match for thread signalling to apply */
1170         p->parent_exec_id = p->self_exec_id;
1171
1172         /* ok, now we should be set up.. */
1173         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1174         p->pdeath_signal = 0;
1175         p->exit_state = 0;
1176
1177         /*
1178          * Ok, make it visible to the rest of the system.
1179          * We dont wake it up yet.
1180          */
1181         p->group_leader = p;
1182         INIT_LIST_HEAD(&p->thread_group);
1183         INIT_LIST_HEAD(&p->ptrace_children);
1184         INIT_LIST_HEAD(&p->ptrace_list);
1185
1186         /* Perform scheduler related setup. Assign this task to a CPU. */
1187         sched_fork(p, clone_flags);
1188
1189         /* Need tasklist lock for parent etc handling! */
1190         write_lock_irq(&tasklist_lock);
1191
1192         /* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
1193         p->ioprio = current->ioprio;
1194
1195         /*
1196          * The task hasn't been attached yet, so its cpus_allowed mask will
1197          * not be changed, nor will its assigned CPU.
1198          *
1199          * The cpus_allowed mask of the parent may have changed after it was
1200          * copied first time - so re-copy it here, then check the child's CPU
1201          * to ensure it is on a valid CPU (and if not, just force it back to
1202          * parent's CPU). This avoids alot of nasty races.
1203          */
1204         p->cpus_allowed = current->cpus_allowed;
1205         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1206                         !cpu_online(task_cpu(p))))
1207                 set_task_cpu(p, smp_processor_id());
1208
1209         /* CLONE_PARENT re-uses the old parent */
1210         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1211                 p->real_parent = current->real_parent;
1212         else
1213                 p->real_parent = current;
1214         p->parent = p->real_parent;
1215
1216         spin_lock(&current->sighand->siglock);
1217
1218         /*
1219          * Process group and session signals need to be delivered to just the
1220          * parent before the fork or both the parent and the child after the
1221          * fork. Restart if a signal comes in before we add the new process to
1222          * it's process group.
1223          * A fatal signal pending means that current will exit, so the new
1224          * thread can't slip out of an OOM kill (or normal SIGKILL).
1225          */
1226         recalc_sigpending();
1227         if (signal_pending(current)) {
1228                 spin_unlock(&current->sighand->siglock);
1229                 write_unlock_irq(&tasklist_lock);
1230                 retval = -ERESTARTNOINTR;
1231                 goto bad_fork_cleanup_namespaces;
1232         }
1233
1234         if (clone_flags & CLONE_THREAD) {
1235                 p->group_leader = current->group_leader;
1236                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1237
1238                 if (!cputime_eq(current->signal->it_virt_expires,
1239                                 cputime_zero) ||
1240                     !cputime_eq(current->signal->it_prof_expires,
1241                                 cputime_zero) ||
1242                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1243                     !list_empty(&current->signal->cpu_timers[0]) ||
1244                     !list_empty(&current->signal->cpu_timers[1]) ||
1245                     !list_empty(&current->signal->cpu_timers[2])) {
1246                         /*
1247                          * Have child wake up on its first tick to check
1248                          * for process CPU timers.
1249                          */
1250                         p->it_prof_expires = jiffies_to_cputime(1);
1251                 }
1252         }
1253
1254         if (likely(p->pid)) {
1255                 add_parent(p);
1256                 if (unlikely(p->ptrace & PT_PTRACED))
1257                         __ptrace_link(p, current->parent);
1258
1259                 if (thread_group_leader(p)) {
1260                         p->signal->tty = current->signal->tty;
1261                         p->signal->pgrp = process_group(current);
1262                         set_signal_session(p->signal, process_session(current));
1263                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1264                         attach_pid(p, PIDTYPE_SID, process_session(p));
1265
1266                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1267                         __get_cpu_var(process_counts)++;
1268                 }
1269                 attach_pid(p, PIDTYPE_PID, p->pid);
1270                 nr_threads++;
1271         }
1272
1273         total_forks++;
1274         spin_unlock(&current->sighand->siglock);
1275         write_unlock_irq(&tasklist_lock);
1276         proc_fork_connector(p);
1277         return p;
1278
1279 bad_fork_cleanup_namespaces:
1280         exit_task_namespaces(p);
1281 bad_fork_cleanup_keys:
1282         exit_keys(p);
1283 bad_fork_cleanup_mm:
1284         if (p->mm)
1285                 mmput(p->mm);
1286 bad_fork_cleanup_signal:
1287         cleanup_signal(p);
1288 bad_fork_cleanup_sighand:
1289         __cleanup_sighand(p->sighand);
1290 bad_fork_cleanup_fs:
1291         exit_fs(p); /* blocking */
1292 bad_fork_cleanup_files:
1293         exit_files(p); /* blocking */
1294 bad_fork_cleanup_semundo:
1295         exit_sem(p);
1296 bad_fork_cleanup_audit:
1297         audit_free(p);
1298 bad_fork_cleanup_security:
1299         security_task_free(p);
1300 bad_fork_cleanup_policy:
1301 #ifdef CONFIG_NUMA
1302         mpol_free(p->mempolicy);
1303 bad_fork_cleanup_cpuset:
1304 #endif
1305         cpuset_exit(p);
1306 bad_fork_cleanup_delays_binfmt:
1307         delayacct_tsk_free(p);
1308         if (p->binfmt)
1309                 module_put(p->binfmt->module);
1310 bad_fork_cleanup_put_domain:
1311         module_put(task_thread_info(p)->exec_domain->module);
1312 bad_fork_cleanup_count:
1313         put_group_info(p->group_info);
1314         atomic_dec(&p->user->processes);
1315         free_uid(p->user);
1316 bad_fork_free:
1317         free_task(p);
1318 fork_out:
1319         return ERR_PTR(retval);
1320 }
1321
1322 noinline struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1323 {
1324         memset(regs, 0, sizeof(struct pt_regs));
1325         return regs;
1326 }
1327
1328 struct task_struct * __devinit fork_idle(int cpu)
1329 {
1330         struct task_struct *task;
1331         struct pt_regs regs;
1332
1333         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1334         if (!IS_ERR(task))
1335                 init_idle(task, cpu);
1336
1337         return task;
1338 }
1339
1340 static inline int fork_traceflag (unsigned clone_flags)
1341 {
1342         if (clone_flags & CLONE_UNTRACED)
1343                 return 0;
1344         else if (clone_flags & CLONE_VFORK) {
1345                 if (current->ptrace & PT_TRACE_VFORK)
1346                         return PTRACE_EVENT_VFORK;
1347         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1348                 if (current->ptrace & PT_TRACE_CLONE)
1349                         return PTRACE_EVENT_CLONE;
1350         } else if (current->ptrace & PT_TRACE_FORK)
1351                 return PTRACE_EVENT_FORK;
1352
1353         return 0;
1354 }
1355
1356 /*
1357  *  Ok, this is the main fork-routine.
1358  *
1359  * It copies the process, and if successful kick-starts
1360  * it and waits for it to finish using the VM if required.
1361  */
1362 long do_fork(unsigned long clone_flags,
1363               unsigned long stack_start,
1364               struct pt_regs *regs,
1365               unsigned long stack_size,
1366               int __user *parent_tidptr,
1367               int __user *child_tidptr)
1368 {
1369         struct task_struct *p;
1370         int trace = 0;
1371         struct pid *pid = alloc_pid();
1372         long nr;
1373
1374         if (!pid)
1375                 return -EAGAIN;
1376         nr = pid->nr;
1377         if (unlikely(current->ptrace)) {
1378                 trace = fork_traceflag (clone_flags);
1379                 if (trace)
1380                         clone_flags |= CLONE_PTRACE;
1381         }
1382
1383         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1384         /*
1385          * Do this prior waking up the new thread - the thread pointer
1386          * might get invalid after that point, if the thread exits quickly.
1387          */
1388         if (!IS_ERR(p)) {
1389                 struct completion vfork;
1390
1391                 if (clone_flags & CLONE_VFORK) {
1392                         p->vfork_done = &vfork;
1393                         init_completion(&vfork);
1394                 }
1395
1396                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1397                         /*
1398                          * We'll start up with an immediate SIGSTOP.
1399                          */
1400                         sigaddset(&p->pending.signal, SIGSTOP);
1401                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1402                 }
1403
1404                 if (!(clone_flags & CLONE_STOPPED))
1405                         wake_up_new_task(p, clone_flags);
1406                 else
1407                         p->state = TASK_STOPPED;
1408
1409                 if (unlikely (trace)) {
1410                         current->ptrace_message = nr;
1411                         ptrace_notify ((trace << 8) | SIGTRAP);
1412                 }
1413
1414                 if (clone_flags & CLONE_VFORK) {
1415                         wait_for_completion(&vfork);
1416                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1417                                 current->ptrace_message = nr;
1418                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1419                         }
1420                 }
1421         } else {
1422                 free_pid(pid);
1423                 nr = PTR_ERR(p);
1424         }
1425         return nr;
1426 }
1427
1428 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1429 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1430 #endif
1431
1432 static void sighand_ctor(void *data, struct kmem_cache *cachep, unsigned long flags)
1433 {
1434         struct sighand_struct *sighand = data;
1435
1436         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1437                                         SLAB_CTOR_CONSTRUCTOR)
1438                 spin_lock_init(&sighand->siglock);
1439 }
1440
1441 void __init proc_caches_init(void)
1442 {
1443         sighand_cachep = kmem_cache_create("sighand_cache",
1444                         sizeof(struct sighand_struct), 0,
1445                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1446                         sighand_ctor, NULL);
1447         signal_cachep = kmem_cache_create("signal_cache",
1448                         sizeof(struct signal_struct), 0,
1449                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1450         files_cachep = kmem_cache_create("files_cache", 
1451                         sizeof(struct files_struct), 0,
1452                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1453         fs_cachep = kmem_cache_create("fs_cache", 
1454                         sizeof(struct fs_struct), 0,
1455                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1456         vm_area_cachep = kmem_cache_create("vm_area_struct",
1457                         sizeof(struct vm_area_struct), 0,
1458                         SLAB_PANIC, NULL, NULL);
1459         mm_cachep = kmem_cache_create("mm_struct",
1460                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1461                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1462 }
1463
1464
1465 /*
1466  * Check constraints on flags passed to the unshare system call and
1467  * force unsharing of additional process context as appropriate.
1468  */
1469 static inline void check_unshare_flags(unsigned long *flags_ptr)
1470 {
1471         /*
1472          * If unsharing a thread from a thread group, must also
1473          * unshare vm.
1474          */
1475         if (*flags_ptr & CLONE_THREAD)
1476                 *flags_ptr |= CLONE_VM;
1477
1478         /*
1479          * If unsharing vm, must also unshare signal handlers.
1480          */
1481         if (*flags_ptr & CLONE_VM)
1482                 *flags_ptr |= CLONE_SIGHAND;
1483
1484         /*
1485          * If unsharing signal handlers and the task was created
1486          * using CLONE_THREAD, then must unshare the thread
1487          */
1488         if ((*flags_ptr & CLONE_SIGHAND) &&
1489             (atomic_read(&current->signal->count) > 1))
1490                 *flags_ptr |= CLONE_THREAD;
1491
1492         /*
1493          * If unsharing namespace, must also unshare filesystem information.
1494          */
1495         if (*flags_ptr & CLONE_NEWNS)
1496                 *flags_ptr |= CLONE_FS;
1497 }
1498
1499 /*
1500  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1501  */
1502 static int unshare_thread(unsigned long unshare_flags)
1503 {
1504         if (unshare_flags & CLONE_THREAD)
1505                 return -EINVAL;
1506
1507         return 0;
1508 }
1509
1510 /*
1511  * Unshare the filesystem structure if it is being shared
1512  */
1513 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1514 {
1515         struct fs_struct *fs = current->fs;
1516
1517         if ((unshare_flags & CLONE_FS) &&
1518             (fs && atomic_read(&fs->count) > 1)) {
1519                 *new_fsp = __copy_fs_struct(current->fs);
1520                 if (!*new_fsp)
1521                         return -ENOMEM;
1522         }
1523
1524         return 0;
1525 }
1526
1527 /*
1528  * Unshare the mnt_namespace structure if it is being shared
1529  */
1530 static int unshare_mnt_namespace(unsigned long unshare_flags,
1531                 struct mnt_namespace **new_nsp, struct fs_struct *new_fs)
1532 {
1533         struct mnt_namespace *ns = current->nsproxy->mnt_ns;
1534
1535         if ((unshare_flags & CLONE_NEWNS) && ns) {
1536                 if (!capable(CAP_SYS_ADMIN))
1537                         return -EPERM;
1538
1539                 *new_nsp = dup_mnt_ns(current, new_fs ? new_fs : current->fs);
1540                 if (!*new_nsp)
1541                         return -ENOMEM;
1542         }
1543
1544         return 0;
1545 }
1546
1547 /*
1548  * Unsharing of sighand is not supported yet
1549  */
1550 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1551 {
1552         struct sighand_struct *sigh = current->sighand;
1553
1554         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1555                 return -EINVAL;
1556         else
1557                 return 0;
1558 }
1559
1560 /*
1561  * Unshare vm if it is being shared
1562  */
1563 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1564 {
1565         struct mm_struct *mm = current->mm;
1566
1567         if ((unshare_flags & CLONE_VM) &&
1568             (mm && atomic_read(&mm->mm_users) > 1)) {
1569                 return -EINVAL;
1570         }
1571
1572         return 0;
1573 }
1574
1575 /*
1576  * Unshare file descriptor table if it is being shared
1577  */
1578 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1579 {
1580         struct files_struct *fd = current->files;
1581         int error = 0;
1582
1583         if ((unshare_flags & CLONE_FILES) &&
1584             (fd && atomic_read(&fd->count) > 1)) {
1585                 *new_fdp = dup_fd(fd, &error);
1586                 if (!*new_fdp)
1587                         return error;
1588         }
1589
1590         return 0;
1591 }
1592
1593 /*
1594  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1595  * supported yet
1596  */
1597 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1598 {
1599         if (unshare_flags & CLONE_SYSVSEM)
1600                 return -EINVAL;
1601
1602         return 0;
1603 }
1604
1605 #ifndef CONFIG_IPC_NS
1606 static inline int unshare_ipcs(unsigned long flags, struct ipc_namespace **ns)
1607 {
1608         if (flags & CLONE_NEWIPC)
1609                 return -EINVAL;
1610
1611         return 0;
1612 }
1613 #endif
1614
1615 /*
1616  * unshare allows a process to 'unshare' part of the process
1617  * context which was originally shared using clone.  copy_*
1618  * functions used by do_fork() cannot be used here directly
1619  * because they modify an inactive task_struct that is being
1620  * constructed. Here we are modifying the current, active,
1621  * task_struct.
1622  */
1623 asmlinkage long sys_unshare(unsigned long unshare_flags)
1624 {
1625         int err = 0;
1626         struct fs_struct *fs, *new_fs = NULL;
1627         struct mnt_namespace *ns, *new_ns = NULL;
1628         struct sighand_struct *new_sigh = NULL;
1629         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1630         struct files_struct *fd, *new_fd = NULL;
1631         struct sem_undo_list *new_ulist = NULL;
1632         struct nsproxy *new_nsproxy = NULL, *old_nsproxy = NULL;
1633         struct uts_namespace *uts, *new_uts = NULL;
1634         struct ipc_namespace *ipc, *new_ipc = NULL;
1635
1636         check_unshare_flags(&unshare_flags);
1637
1638         /* Return -EINVAL for all unsupported flags */
1639         err = -EINVAL;
1640         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1641                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1642                                 CLONE_NEWUTS|CLONE_NEWIPC))
1643                 goto bad_unshare_out;
1644
1645         if ((err = unshare_thread(unshare_flags)))
1646                 goto bad_unshare_out;
1647         if ((err = unshare_fs(unshare_flags, &new_fs)))
1648                 goto bad_unshare_cleanup_thread;
1649         if ((err = unshare_mnt_namespace(unshare_flags, &new_ns, new_fs)))
1650                 goto bad_unshare_cleanup_fs;
1651         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1652                 goto bad_unshare_cleanup_ns;
1653         if ((err = unshare_vm(unshare_flags, &new_mm)))
1654                 goto bad_unshare_cleanup_sigh;
1655         if ((err = unshare_fd(unshare_flags, &new_fd)))
1656                 goto bad_unshare_cleanup_vm;
1657         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1658                 goto bad_unshare_cleanup_fd;
1659         if ((err = unshare_utsname(unshare_flags, &new_uts)))
1660                 goto bad_unshare_cleanup_semundo;
1661         if ((err = unshare_ipcs(unshare_flags, &new_ipc)))
1662                 goto bad_unshare_cleanup_uts;
1663
1664         if (new_ns || new_uts || new_ipc) {
1665                 old_nsproxy = current->nsproxy;
1666                 new_nsproxy = dup_namespaces(old_nsproxy);
1667                 if (!new_nsproxy) {
1668                         err = -ENOMEM;
1669                         goto bad_unshare_cleanup_ipc;
1670                 }
1671         }
1672
1673         if (new_fs || new_ns || new_mm || new_fd || new_ulist ||
1674                                 new_uts || new_ipc) {
1675
1676                 task_lock(current);
1677
1678                 if (new_nsproxy) {
1679                         current->nsproxy = new_nsproxy;
1680                         new_nsproxy = old_nsproxy;
1681                 }
1682
1683                 if (new_fs) {
1684                         fs = current->fs;
1685                         current->fs = new_fs;
1686                         new_fs = fs;
1687                 }
1688
1689                 if (new_ns) {
1690                         ns = current->nsproxy->mnt_ns;
1691                         current->nsproxy->mnt_ns = new_ns;
1692                         new_ns = ns;
1693                 }
1694
1695                 if (new_mm) {
1696                         mm = current->mm;
1697                         active_mm = current->active_mm;
1698                         current->mm = new_mm;
1699                         current->active_mm = new_mm;
1700                         activate_mm(active_mm, new_mm);
1701                         new_mm = mm;
1702                 }
1703
1704                 if (new_fd) {
1705                         fd = current->files;
1706                         current->files = new_fd;
1707                         new_fd = fd;
1708                 }
1709
1710                 if (new_uts) {
1711                         uts = current->nsproxy->uts_ns;
1712                         current->nsproxy->uts_ns = new_uts;
1713                         new_uts = uts;
1714                 }
1715
1716                 if (new_ipc) {
1717                         ipc = current->nsproxy->ipc_ns;
1718                         current->nsproxy->ipc_ns = new_ipc;
1719                         new_ipc = ipc;
1720                 }
1721
1722                 task_unlock(current);
1723         }
1724
1725         if (new_nsproxy)
1726                 put_nsproxy(new_nsproxy);
1727
1728 bad_unshare_cleanup_ipc:
1729         if (new_ipc)
1730                 put_ipc_ns(new_ipc);
1731
1732 bad_unshare_cleanup_uts:
1733         if (new_uts)
1734                 put_uts_ns(new_uts);
1735
1736 bad_unshare_cleanup_semundo:
1737 bad_unshare_cleanup_fd:
1738         if (new_fd)
1739                 put_files_struct(new_fd);
1740
1741 bad_unshare_cleanup_vm:
1742         if (new_mm)
1743                 mmput(new_mm);
1744
1745 bad_unshare_cleanup_sigh:
1746         if (new_sigh)
1747                 if (atomic_dec_and_test(&new_sigh->count))
1748                         kmem_cache_free(sighand_cachep, new_sigh);
1749
1750 bad_unshare_cleanup_ns:
1751         if (new_ns)
1752                 put_mnt_ns(new_ns);
1753
1754 bad_unshare_cleanup_fs:
1755         if (new_fs)
1756                 put_fs_struct(new_fs);
1757
1758 bad_unshare_cleanup_thread:
1759 bad_unshare_out:
1760         return err;
1761 }