e0a2b449dea64ea1b0d79bdc7124a74cb5c4cbdf
[linux-2.6.git] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/config.h>
15 #include <linux/slab.h>
16 #include <linux/init.h>
17 #include <linux/unistd.h>
18 #include <linux/smp_lock.h>
19 #include <linux/module.h>
20 #include <linux/vmalloc.h>
21 #include <linux/completion.h>
22 #include <linux/namespace.h>
23 #include <linux/personality.h>
24 #include <linux/mempolicy.h>
25 #include <linux/sem.h>
26 #include <linux/file.h>
27 #include <linux/key.h>
28 #include <linux/binfmts.h>
29 #include <linux/mman.h>
30 #include <linux/fs.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/cn_proc.h>
47
48 #include <asm/pgtable.h>
49 #include <asm/pgalloc.h>
50 #include <asm/uaccess.h>
51 #include <asm/mmu_context.h>
52 #include <asm/cacheflush.h>
53 #include <asm/tlbflush.h>
54
55 /*
56  * Protected counters by write_lock_irq(&tasklist_lock)
57  */
58 unsigned long total_forks;      /* Handle normal Linux uptimes. */
59 int nr_threads;                 /* The idle threads do not count.. */
60
61 int max_threads;                /* tunable limit on nr_threads */
62
63 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
64
65  __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
66
67 EXPORT_SYMBOL(tasklist_lock);
68
69 int nr_processes(void)
70 {
71         int cpu;
72         int total = 0;
73
74         for_each_online_cpu(cpu)
75                 total += per_cpu(process_counts, cpu);
76
77         return total;
78 }
79
80 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
81 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
82 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
83 static kmem_cache_t *task_struct_cachep;
84 #endif
85
86 /* SLAB cache for signal_struct structures (tsk->signal) */
87 kmem_cache_t *signal_cachep;
88
89 /* SLAB cache for sighand_struct structures (tsk->sighand) */
90 kmem_cache_t *sighand_cachep;
91
92 /* SLAB cache for files_struct structures (tsk->files) */
93 kmem_cache_t *files_cachep;
94
95 /* SLAB cache for fs_struct structures (tsk->fs) */
96 kmem_cache_t *fs_cachep;
97
98 /* SLAB cache for vm_area_struct structures */
99 kmem_cache_t *vm_area_cachep;
100
101 /* SLAB cache for mm_struct structures (tsk->mm) */
102 static kmem_cache_t *mm_cachep;
103
104 void free_task(struct task_struct *tsk)
105 {
106         free_thread_info(tsk->thread_info);
107         free_task_struct(tsk);
108 }
109 EXPORT_SYMBOL(free_task);
110
111 void __put_task_struct_cb(struct rcu_head *rhp)
112 {
113         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
114
115         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
116         WARN_ON(atomic_read(&tsk->usage));
117         WARN_ON(tsk == current);
118
119         if (unlikely(tsk->audit_context))
120                 audit_free(tsk);
121         security_task_free(tsk);
122         free_uid(tsk->user);
123         put_group_info(tsk->group_info);
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         /* One for us, one for whoever does the "release_task()" (usually parent) */
182         atomic_set(&tsk->usage,2);
183         atomic_set(&tsk->fs_excl, 0);
184         tsk->btrace_seq = 0;
185         return tsk;
186 }
187
188 #ifdef CONFIG_MMU
189 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
190 {
191         struct vm_area_struct *mpnt, *tmp, **pprev;
192         struct rb_node **rb_link, *rb_parent;
193         int retval;
194         unsigned long charge;
195         struct mempolicy *pol;
196
197         down_write(&oldmm->mmap_sem);
198         flush_cache_mm(oldmm);
199         down_write(&mm->mmap_sem);
200
201         mm->locked_vm = 0;
202         mm->mmap = NULL;
203         mm->mmap_cache = NULL;
204         mm->free_area_cache = oldmm->mmap_base;
205         mm->cached_hole_size = ~0UL;
206         mm->map_count = 0;
207         cpus_clear(mm->cpu_vm_mask);
208         mm->mm_rb = RB_ROOT;
209         rb_link = &mm->mm_rb.rb_node;
210         rb_parent = NULL;
211         pprev = &mm->mmap;
212
213         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
214                 struct file *file;
215
216                 if (mpnt->vm_flags & VM_DONTCOPY) {
217                         long pages = vma_pages(mpnt);
218                         mm->total_vm -= pages;
219                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
220                                                                 -pages);
221                         continue;
222                 }
223                 charge = 0;
224                 if (mpnt->vm_flags & VM_ACCOUNT) {
225                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
226                         if (security_vm_enough_memory(len))
227                                 goto fail_nomem;
228                         charge = len;
229                 }
230                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
231                 if (!tmp)
232                         goto fail_nomem;
233                 *tmp = *mpnt;
234                 pol = mpol_copy(vma_policy(mpnt));
235                 retval = PTR_ERR(pol);
236                 if (IS_ERR(pol))
237                         goto fail_nomem_policy;
238                 vma_set_policy(tmp, pol);
239                 tmp->vm_flags &= ~VM_LOCKED;
240                 tmp->vm_mm = mm;
241                 tmp->vm_next = NULL;
242                 anon_vma_link(tmp);
243                 file = tmp->vm_file;
244                 if (file) {
245                         struct inode *inode = file->f_dentry->d_inode;
246                         get_file(file);
247                         if (tmp->vm_flags & VM_DENYWRITE)
248                                 atomic_dec(&inode->i_writecount);
249       
250                         /* insert tmp into the share list, just after mpnt */
251                         spin_lock(&file->f_mapping->i_mmap_lock);
252                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
253                         flush_dcache_mmap_lock(file->f_mapping);
254                         vma_prio_tree_add(tmp, mpnt);
255                         flush_dcache_mmap_unlock(file->f_mapping);
256                         spin_unlock(&file->f_mapping->i_mmap_lock);
257                 }
258
259                 /*
260                  * Link in the new vma and copy the page table entries.
261                  */
262                 *pprev = tmp;
263                 pprev = &tmp->vm_next;
264
265                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
266                 rb_link = &tmp->vm_rb.rb_right;
267                 rb_parent = &tmp->vm_rb;
268
269                 mm->map_count++;
270                 retval = copy_page_range(mm, oldmm, mpnt);
271
272                 if (tmp->vm_ops && tmp->vm_ops->open)
273                         tmp->vm_ops->open(tmp);
274
275                 if (retval)
276                         goto out;
277         }
278         retval = 0;
279 out:
280         up_write(&mm->mmap_sem);
281         flush_tlb_mm(oldmm);
282         up_write(&oldmm->mmap_sem);
283         return retval;
284 fail_nomem_policy:
285         kmem_cache_free(vm_area_cachep, tmp);
286 fail_nomem:
287         retval = -ENOMEM;
288         vm_unacct_memory(charge);
289         goto out;
290 }
291
292 static inline int mm_alloc_pgd(struct mm_struct * mm)
293 {
294         mm->pgd = pgd_alloc(mm);
295         if (unlikely(!mm->pgd))
296                 return -ENOMEM;
297         return 0;
298 }
299
300 static inline void mm_free_pgd(struct mm_struct * mm)
301 {
302         pgd_free(mm->pgd);
303 }
304 #else
305 #define dup_mmap(mm, oldmm)     (0)
306 #define mm_alloc_pgd(mm)        (0)
307 #define mm_free_pgd(mm)
308 #endif /* CONFIG_MMU */
309
310  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
311
312 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
313 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
314
315 #include <linux/init_task.h>
316
317 static struct mm_struct * mm_init(struct mm_struct * mm)
318 {
319         atomic_set(&mm->mm_users, 1);
320         atomic_set(&mm->mm_count, 1);
321         init_rwsem(&mm->mmap_sem);
322         INIT_LIST_HEAD(&mm->mmlist);
323         mm->core_waiters = 0;
324         mm->nr_ptes = 0;
325         set_mm_counter(mm, file_rss, 0);
326         set_mm_counter(mm, anon_rss, 0);
327         spin_lock_init(&mm->page_table_lock);
328         rwlock_init(&mm->ioctx_list_lock);
329         mm->ioctx_list = NULL;
330         mm->free_area_cache = TASK_UNMAPPED_BASE;
331         mm->cached_hole_size = ~0UL;
332
333         if (likely(!mm_alloc_pgd(mm))) {
334                 mm->def_flags = 0;
335                 return mm;
336         }
337         free_mm(mm);
338         return NULL;
339 }
340
341 /*
342  * Allocate and initialize an mm_struct.
343  */
344 struct mm_struct * mm_alloc(void)
345 {
346         struct mm_struct * mm;
347
348         mm = allocate_mm();
349         if (mm) {
350                 memset(mm, 0, sizeof(*mm));
351                 mm = mm_init(mm);
352         }
353         return mm;
354 }
355
356 /*
357  * Called when the last reference to the mm
358  * is dropped: either by a lazy thread or by
359  * mmput. Free the page directory and the mm.
360  */
361 void fastcall __mmdrop(struct mm_struct *mm)
362 {
363         BUG_ON(mm == &init_mm);
364         mm_free_pgd(mm);
365         destroy_context(mm);
366         free_mm(mm);
367 }
368
369 /*
370  * Decrement the use count and release all resources for an mm.
371  */
372 void mmput(struct mm_struct *mm)
373 {
374         if (atomic_dec_and_test(&mm->mm_users)) {
375                 exit_aio(mm);
376                 exit_mmap(mm);
377                 if (!list_empty(&mm->mmlist)) {
378                         spin_lock(&mmlist_lock);
379                         list_del(&mm->mmlist);
380                         spin_unlock(&mmlist_lock);
381                 }
382                 put_swap_token(mm);
383                 mmdrop(mm);
384         }
385 }
386 EXPORT_SYMBOL_GPL(mmput);
387
388 /**
389  * get_task_mm - acquire a reference to the task's mm
390  *
391  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
392  * this kernel workthread has transiently adopted a user mm with use_mm,
393  * to do its AIO) is not set and if so returns a reference to it, after
394  * bumping up the use count.  User must release the mm via mmput()
395  * after use.  Typically used by /proc and ptrace.
396  */
397 struct mm_struct *get_task_mm(struct task_struct *task)
398 {
399         struct mm_struct *mm;
400
401         task_lock(task);
402         mm = task->mm;
403         if (mm) {
404                 if (task->flags & PF_BORROWED_MM)
405                         mm = NULL;
406                 else
407                         atomic_inc(&mm->mm_users);
408         }
409         task_unlock(task);
410         return mm;
411 }
412 EXPORT_SYMBOL_GPL(get_task_mm);
413
414 /* Please note the differences between mmput and mm_release.
415  * mmput is called whenever we stop holding onto a mm_struct,
416  * error success whatever.
417  *
418  * mm_release is called after a mm_struct has been removed
419  * from the current process.
420  *
421  * This difference is important for error handling, when we
422  * only half set up a mm_struct for a new process and need to restore
423  * the old one.  Because we mmput the new mm_struct before
424  * restoring the old one. . .
425  * Eric Biederman 10 January 1998
426  */
427 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
428 {
429         struct completion *vfork_done = tsk->vfork_done;
430
431         /* Get rid of any cached register state */
432         deactivate_mm(tsk, mm);
433
434         /* notify parent sleeping on vfork() */
435         if (vfork_done) {
436                 tsk->vfork_done = NULL;
437                 complete(vfork_done);
438         }
439         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
440                 u32 __user * tidptr = tsk->clear_child_tid;
441                 tsk->clear_child_tid = NULL;
442
443                 /*
444                  * We don't check the error code - if userspace has
445                  * not set up a proper pointer then tough luck.
446                  */
447                 put_user(0, tidptr);
448                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
449         }
450 }
451
452 /*
453  * Allocate a new mm structure and copy contents from the
454  * mm structure of the passed in task structure.
455  */
456 static struct mm_struct *dup_mm(struct task_struct *tsk)
457 {
458         struct mm_struct *mm, *oldmm = current->mm;
459         int err;
460
461         if (!oldmm)
462                 return NULL;
463
464         mm = allocate_mm();
465         if (!mm)
466                 goto fail_nomem;
467
468         memcpy(mm, oldmm, sizeof(*mm));
469
470         if (!mm_init(mm))
471                 goto fail_nomem;
472
473         if (init_new_context(tsk, mm))
474                 goto fail_nocontext;
475
476         err = dup_mmap(mm, oldmm);
477         if (err)
478                 goto free_pt;
479
480         mm->hiwater_rss = get_mm_rss(mm);
481         mm->hiwater_vm = mm->total_vm;
482
483         return mm;
484
485 free_pt:
486         mmput(mm);
487
488 fail_nomem:
489         return NULL;
490
491 fail_nocontext:
492         /*
493          * If init_new_context() failed, we cannot use mmput() to free the mm
494          * because it calls destroy_context()
495          */
496         mm_free_pgd(mm);
497         free_mm(mm);
498         return NULL;
499 }
500
501 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
502 {
503         struct mm_struct * mm, *oldmm;
504         int retval;
505
506         tsk->min_flt = tsk->maj_flt = 0;
507         tsk->nvcsw = tsk->nivcsw = 0;
508
509         tsk->mm = NULL;
510         tsk->active_mm = NULL;
511
512         /*
513          * Are we cloning a kernel thread?
514          *
515          * We need to steal a active VM for that..
516          */
517         oldmm = current->mm;
518         if (!oldmm)
519                 return 0;
520
521         if (clone_flags & CLONE_VM) {
522                 atomic_inc(&oldmm->mm_users);
523                 mm = oldmm;
524                 goto good_mm;
525         }
526
527         retval = -ENOMEM;
528         mm = dup_mm(tsk);
529         if (!mm)
530                 goto fail_nomem;
531
532 good_mm:
533         tsk->mm = mm;
534         tsk->active_mm = mm;
535         return 0;
536
537 fail_nomem:
538         return retval;
539 }
540
541 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
542 {
543         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
544         /* We don't need to lock fs - think why ;-) */
545         if (fs) {
546                 atomic_set(&fs->count, 1);
547                 rwlock_init(&fs->lock);
548                 fs->umask = old->umask;
549                 read_lock(&old->lock);
550                 fs->rootmnt = mntget(old->rootmnt);
551                 fs->root = dget(old->root);
552                 fs->pwdmnt = mntget(old->pwdmnt);
553                 fs->pwd = dget(old->pwd);
554                 if (old->altroot) {
555                         fs->altrootmnt = mntget(old->altrootmnt);
556                         fs->altroot = dget(old->altroot);
557                 } else {
558                         fs->altrootmnt = NULL;
559                         fs->altroot = NULL;
560                 }
561                 read_unlock(&old->lock);
562         }
563         return fs;
564 }
565
566 struct fs_struct *copy_fs_struct(struct fs_struct *old)
567 {
568         return __copy_fs_struct(old);
569 }
570
571 EXPORT_SYMBOL_GPL(copy_fs_struct);
572
573 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
574 {
575         if (clone_flags & CLONE_FS) {
576                 atomic_inc(&current->fs->count);
577                 return 0;
578         }
579         tsk->fs = __copy_fs_struct(current->fs);
580         if (!tsk->fs)
581                 return -ENOMEM;
582         return 0;
583 }
584
585 static int count_open_files(struct fdtable *fdt)
586 {
587         int size = fdt->max_fdset;
588         int i;
589
590         /* Find the last open fd */
591         for (i = size/(8*sizeof(long)); i > 0; ) {
592                 if (fdt->open_fds->fds_bits[--i])
593                         break;
594         }
595         i = (i+1) * 8 * sizeof(long);
596         return i;
597 }
598
599 static struct files_struct *alloc_files(void)
600 {
601         struct files_struct *newf;
602         struct fdtable *fdt;
603
604         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
605         if (!newf)
606                 goto out;
607
608         atomic_set(&newf->count, 1);
609
610         spin_lock_init(&newf->file_lock);
611         newf->next_fd = 0;
612         fdt = &newf->fdtab;
613         fdt->max_fds = NR_OPEN_DEFAULT;
614         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
615         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
616         fdt->open_fds = (fd_set *)&newf->open_fds_init;
617         fdt->fd = &newf->fd_array[0];
618         INIT_RCU_HEAD(&fdt->rcu);
619         fdt->free_files = NULL;
620         fdt->next = NULL;
621         rcu_assign_pointer(newf->fdt, fdt);
622 out:
623         return newf;
624 }
625
626 /*
627  * Allocate a new files structure and copy contents from the
628  * passed in files structure.
629  */
630 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
631 {
632         struct files_struct *newf;
633         struct file **old_fds, **new_fds;
634         int open_files, size, i, expand;
635         struct fdtable *old_fdt, *new_fdt;
636
637         newf = alloc_files();
638         if (!newf)
639                 goto out;
640
641         spin_lock(&oldf->file_lock);
642         old_fdt = files_fdtable(oldf);
643         new_fdt = files_fdtable(newf);
644         size = old_fdt->max_fdset;
645         open_files = count_open_files(old_fdt);
646         expand = 0;
647
648         /*
649          * Check whether we need to allocate a larger fd array or fd set.
650          * Note: we're not a clone task, so the open count won't  change.
651          */
652         if (open_files > new_fdt->max_fdset) {
653                 new_fdt->max_fdset = 0;
654                 expand = 1;
655         }
656         if (open_files > new_fdt->max_fds) {
657                 new_fdt->max_fds = 0;
658                 expand = 1;
659         }
660
661         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
662         if (expand) {
663                 spin_unlock(&oldf->file_lock);
664                 spin_lock(&newf->file_lock);
665                 *errorp = expand_files(newf, open_files-1);
666                 spin_unlock(&newf->file_lock);
667                 if (*errorp < 0)
668                         goto out_release;
669                 new_fdt = files_fdtable(newf);
670                 /*
671                  * Reacquire the oldf lock and a pointer to its fd table
672                  * who knows it may have a new bigger fd table. We need
673                  * the latest pointer.
674                  */
675                 spin_lock(&oldf->file_lock);
676                 old_fdt = files_fdtable(oldf);
677         }
678
679         old_fds = old_fdt->fd;
680         new_fds = new_fdt->fd;
681
682         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
683         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
684
685         for (i = open_files; i != 0; i--) {
686                 struct file *f = *old_fds++;
687                 if (f) {
688                         get_file(f);
689                 } else {
690                         /*
691                          * The fd may be claimed in the fd bitmap but not yet
692                          * instantiated in the files array if a sibling thread
693                          * is partway through open().  So make sure that this
694                          * fd is available to the new process.
695                          */
696                         FD_CLR(open_files - i, new_fdt->open_fds);
697                 }
698                 rcu_assign_pointer(*new_fds++, f);
699         }
700         spin_unlock(&oldf->file_lock);
701
702         /* compute the remainder to be cleared */
703         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
704
705         /* This is long word aligned thus could use a optimized version */ 
706         memset(new_fds, 0, size); 
707
708         if (new_fdt->max_fdset > open_files) {
709                 int left = (new_fdt->max_fdset-open_files)/8;
710                 int start = open_files / (8 * sizeof(unsigned long));
711
712                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
713                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
714         }
715
716 out:
717         return newf;
718
719 out_release:
720         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
721         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
722         free_fd_array(new_fdt->fd, new_fdt->max_fds);
723         kmem_cache_free(files_cachep, newf);
724         goto out;
725 }
726
727 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
728 {
729         struct files_struct *oldf, *newf;
730         int error = 0;
731
732         /*
733          * A background process may not have any files ...
734          */
735         oldf = current->files;
736         if (!oldf)
737                 goto out;
738
739         if (clone_flags & CLONE_FILES) {
740                 atomic_inc(&oldf->count);
741                 goto out;
742         }
743
744         /*
745          * Note: we may be using current for both targets (See exec.c)
746          * This works because we cache current->files (old) as oldf. Don't
747          * break this.
748          */
749         tsk->files = NULL;
750         error = -ENOMEM;
751         newf = dup_fd(oldf, &error);
752         if (!newf)
753                 goto out;
754
755         tsk->files = newf;
756         error = 0;
757 out:
758         return error;
759 }
760
761 /*
762  *      Helper to unshare the files of the current task.
763  *      We don't want to expose copy_files internals to
764  *      the exec layer of the kernel.
765  */
766
767 int unshare_files(void)
768 {
769         struct files_struct *files  = current->files;
770         int rc;
771
772         BUG_ON(!files);
773
774         /* This can race but the race causes us to copy when we don't
775            need to and drop the copy */
776         if(atomic_read(&files->count) == 1)
777         {
778                 atomic_inc(&files->count);
779                 return 0;
780         }
781         rc = copy_files(0, current);
782         if(rc)
783                 current->files = files;
784         return rc;
785 }
786
787 EXPORT_SYMBOL(unshare_files);
788
789 void sighand_free_cb(struct rcu_head *rhp)
790 {
791         struct sighand_struct *sp;
792
793         sp = container_of(rhp, struct sighand_struct, rcu);
794         kmem_cache_free(sighand_cachep, sp);
795 }
796
797 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
798 {
799         struct sighand_struct *sig;
800
801         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
802                 atomic_inc(&current->sighand->count);
803                 return 0;
804         }
805         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
806         rcu_assign_pointer(tsk->sighand, sig);
807         if (!sig)
808                 return -ENOMEM;
809         spin_lock_init(&sig->siglock);
810         atomic_set(&sig->count, 1);
811         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
812         return 0;
813 }
814
815 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
816 {
817         struct signal_struct *sig;
818         int ret;
819
820         if (clone_flags & CLONE_THREAD) {
821                 atomic_inc(&current->signal->count);
822                 atomic_inc(&current->signal->live);
823                 return 0;
824         }
825         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
826         tsk->signal = sig;
827         if (!sig)
828                 return -ENOMEM;
829
830         ret = copy_thread_group_keys(tsk);
831         if (ret < 0) {
832                 kmem_cache_free(signal_cachep, sig);
833                 return ret;
834         }
835
836         atomic_set(&sig->count, 1);
837         atomic_set(&sig->live, 1);
838         init_waitqueue_head(&sig->wait_chldexit);
839         sig->flags = 0;
840         sig->group_exit_code = 0;
841         sig->group_exit_task = NULL;
842         sig->group_stop_count = 0;
843         sig->curr_target = NULL;
844         init_sigpending(&sig->shared_pending);
845         INIT_LIST_HEAD(&sig->posix_timers);
846
847         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
848         sig->it_real_incr.tv64 = 0;
849         sig->real_timer.function = it_real_fn;
850         sig->tsk = tsk;
851
852         sig->it_virt_expires = cputime_zero;
853         sig->it_virt_incr = cputime_zero;
854         sig->it_prof_expires = cputime_zero;
855         sig->it_prof_incr = cputime_zero;
856
857         sig->leader = 0;        /* session leadership doesn't inherit */
858         sig->tty_old_pgrp = 0;
859
860         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
861         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
862         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
863         sig->sched_time = 0;
864         INIT_LIST_HEAD(&sig->cpu_timers[0]);
865         INIT_LIST_HEAD(&sig->cpu_timers[1]);
866         INIT_LIST_HEAD(&sig->cpu_timers[2]);
867
868         task_lock(current->group_leader);
869         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
870         task_unlock(current->group_leader);
871
872         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
873                 /*
874                  * New sole thread in the process gets an expiry time
875                  * of the whole CPU time limit.
876                  */
877                 tsk->it_prof_expires =
878                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
879         }
880
881         return 0;
882 }
883
884 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
885 {
886         unsigned long new_flags = p->flags;
887
888         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
889         new_flags |= PF_FORKNOEXEC;
890         if (!(clone_flags & CLONE_PTRACE))
891                 p->ptrace = 0;
892         p->flags = new_flags;
893 }
894
895 asmlinkage long sys_set_tid_address(int __user *tidptr)
896 {
897         current->clear_child_tid = tidptr;
898
899         return current->pid;
900 }
901
902 /*
903  * This creates a new process as a copy of the old one,
904  * but does not actually start it yet.
905  *
906  * It copies the registers, and all the appropriate
907  * parts of the process environment (as per the clone
908  * flags). The actual kick-off is left to the caller.
909  */
910 static task_t *copy_process(unsigned long clone_flags,
911                                  unsigned long stack_start,
912                                  struct pt_regs *regs,
913                                  unsigned long stack_size,
914                                  int __user *parent_tidptr,
915                                  int __user *child_tidptr,
916                                  int pid)
917 {
918         int retval;
919         struct task_struct *p = NULL;
920
921         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
922                 return ERR_PTR(-EINVAL);
923
924         /*
925          * Thread groups must share signals as well, and detached threads
926          * can only be started up within the thread group.
927          */
928         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
929                 return ERR_PTR(-EINVAL);
930
931         /*
932          * Shared signal handlers imply shared VM. By way of the above,
933          * thread groups also imply shared VM. Blocking this case allows
934          * for various simplifications in other code.
935          */
936         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
937                 return ERR_PTR(-EINVAL);
938
939         retval = security_task_create(clone_flags);
940         if (retval)
941                 goto fork_out;
942
943         retval = -ENOMEM;
944         p = dup_task_struct(current);
945         if (!p)
946                 goto fork_out;
947
948         retval = -EAGAIN;
949         if (atomic_read(&p->user->processes) >=
950                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
951                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
952                                 p->user != &root_user)
953                         goto bad_fork_free;
954         }
955
956         atomic_inc(&p->user->__count);
957         atomic_inc(&p->user->processes);
958         get_group_info(p->group_info);
959
960         /*
961          * If multiple threads are within copy_process(), then this check
962          * triggers too late. This doesn't hurt, the check is only there
963          * to stop root fork bombs.
964          */
965         if (nr_threads >= max_threads)
966                 goto bad_fork_cleanup_count;
967
968         if (!try_module_get(task_thread_info(p)->exec_domain->module))
969                 goto bad_fork_cleanup_count;
970
971         if (p->binfmt && !try_module_get(p->binfmt->module))
972                 goto bad_fork_cleanup_put_domain;
973
974         p->did_exec = 0;
975         copy_flags(clone_flags, p);
976         p->pid = pid;
977         retval = -EFAULT;
978         if (clone_flags & CLONE_PARENT_SETTID)
979                 if (put_user(p->pid, parent_tidptr))
980                         goto bad_fork_cleanup;
981
982         p->proc_dentry = NULL;
983
984         INIT_LIST_HEAD(&p->children);
985         INIT_LIST_HEAD(&p->sibling);
986         p->vfork_done = NULL;
987         spin_lock_init(&p->alloc_lock);
988         spin_lock_init(&p->proc_lock);
989
990         clear_tsk_thread_flag(p, TIF_SIGPENDING);
991         init_sigpending(&p->pending);
992
993         p->utime = cputime_zero;
994         p->stime = cputime_zero;
995         p->sched_time = 0;
996         p->rchar = 0;           /* I/O counter: bytes read */
997         p->wchar = 0;           /* I/O counter: bytes written */
998         p->syscr = 0;           /* I/O counter: read syscalls */
999         p->syscw = 0;           /* I/O counter: write syscalls */
1000         acct_clear_integrals(p);
1001
1002         p->it_virt_expires = cputime_zero;
1003         p->it_prof_expires = cputime_zero;
1004         p->it_sched_expires = 0;
1005         INIT_LIST_HEAD(&p->cpu_timers[0]);
1006         INIT_LIST_HEAD(&p->cpu_timers[1]);
1007         INIT_LIST_HEAD(&p->cpu_timers[2]);
1008
1009         p->lock_depth = -1;             /* -1 = no lock */
1010         do_posix_clock_monotonic_gettime(&p->start_time);
1011         p->security = NULL;
1012         p->io_context = NULL;
1013         p->io_wait = NULL;
1014         p->audit_context = NULL;
1015         cpuset_fork(p);
1016 #ifdef CONFIG_NUMA
1017         p->mempolicy = mpol_copy(p->mempolicy);
1018         if (IS_ERR(p->mempolicy)) {
1019                 retval = PTR_ERR(p->mempolicy);
1020                 p->mempolicy = NULL;
1021                 goto bad_fork_cleanup_cpuset;
1022         }
1023         mpol_fix_fork_child_flag(p);
1024 #endif
1025
1026 #ifdef CONFIG_DEBUG_MUTEXES
1027         p->blocked_on = NULL; /* not blocked yet */
1028 #endif
1029
1030         p->tgid = p->pid;
1031         if (clone_flags & CLONE_THREAD)
1032                 p->tgid = current->tgid;
1033
1034         if ((retval = security_task_alloc(p)))
1035                 goto bad_fork_cleanup_policy;
1036         if ((retval = audit_alloc(p)))
1037                 goto bad_fork_cleanup_security;
1038         /* copy all the process information */
1039         if ((retval = copy_semundo(clone_flags, p)))
1040                 goto bad_fork_cleanup_audit;
1041         if ((retval = copy_files(clone_flags, p)))
1042                 goto bad_fork_cleanup_semundo;
1043         if ((retval = copy_fs(clone_flags, p)))
1044                 goto bad_fork_cleanup_files;
1045         if ((retval = copy_sighand(clone_flags, p)))
1046                 goto bad_fork_cleanup_fs;
1047         if ((retval = copy_signal(clone_flags, p)))
1048                 goto bad_fork_cleanup_sighand;
1049         if ((retval = copy_mm(clone_flags, p)))
1050                 goto bad_fork_cleanup_signal;
1051         if ((retval = copy_keys(clone_flags, p)))
1052                 goto bad_fork_cleanup_mm;
1053         if ((retval = copy_namespace(clone_flags, p)))
1054                 goto bad_fork_cleanup_keys;
1055         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1056         if (retval)
1057                 goto bad_fork_cleanup_namespace;
1058
1059         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1060         /*
1061          * Clear TID on mm_release()?
1062          */
1063         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1064
1065         /*
1066          * sigaltstack should be cleared when sharing the same VM
1067          */
1068         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1069                 p->sas_ss_sp = p->sas_ss_size = 0;
1070
1071         /*
1072          * Syscall tracing should be turned off in the child regardless
1073          * of CLONE_PTRACE.
1074          */
1075         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1076 #ifdef TIF_SYSCALL_EMU
1077         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1078 #endif
1079
1080         /* Our parent execution domain becomes current domain
1081            These must match for thread signalling to apply */
1082            
1083         p->parent_exec_id = p->self_exec_id;
1084
1085         /* ok, now we should be set up.. */
1086         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1087         p->pdeath_signal = 0;
1088         p->exit_state = 0;
1089
1090         /*
1091          * Ok, make it visible to the rest of the system.
1092          * We dont wake it up yet.
1093          */
1094         p->group_leader = p;
1095         INIT_LIST_HEAD(&p->ptrace_children);
1096         INIT_LIST_HEAD(&p->ptrace_list);
1097
1098         /* Perform scheduler related setup. Assign this task to a CPU. */
1099         sched_fork(p, clone_flags);
1100
1101         /* Need tasklist lock for parent etc handling! */
1102         write_lock_irq(&tasklist_lock);
1103
1104         /*
1105          * The task hasn't been attached yet, so its cpus_allowed mask will
1106          * not be changed, nor will its assigned CPU.
1107          *
1108          * The cpus_allowed mask of the parent may have changed after it was
1109          * copied first time - so re-copy it here, then check the child's CPU
1110          * to ensure it is on a valid CPU (and if not, just force it back to
1111          * parent's CPU). This avoids alot of nasty races.
1112          */
1113         p->cpus_allowed = current->cpus_allowed;
1114         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1115                         !cpu_online(task_cpu(p))))
1116                 set_task_cpu(p, smp_processor_id());
1117
1118         /*
1119          * Check for pending SIGKILL! The new thread should not be allowed
1120          * to slip out of an OOM kill. (or normal SIGKILL.)
1121          */
1122         if (sigismember(&current->pending.signal, SIGKILL)) {
1123                 write_unlock_irq(&tasklist_lock);
1124                 retval = -EINTR;
1125                 goto bad_fork_cleanup_namespace;
1126         }
1127
1128         /* CLONE_PARENT re-uses the old parent */
1129         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1130                 p->real_parent = current->real_parent;
1131         else
1132                 p->real_parent = current;
1133         p->parent = p->real_parent;
1134
1135         spin_lock(&current->sighand->siglock);
1136         if (clone_flags & CLONE_THREAD) {
1137                 /*
1138                  * Important: if an exit-all has been started then
1139                  * do not create this new thread - the whole thread
1140                  * group is supposed to exit anyway.
1141                  */
1142                 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1143                         spin_unlock(&current->sighand->siglock);
1144                         write_unlock_irq(&tasklist_lock);
1145                         retval = -EAGAIN;
1146                         goto bad_fork_cleanup_namespace;
1147                 }
1148                 p->group_leader = current->group_leader;
1149
1150                 if (current->signal->group_stop_count > 0) {
1151                         /*
1152                          * There is an all-stop in progress for the group.
1153                          * We ourselves will stop as soon as we check signals.
1154                          * Make the new thread part of that group stop too.
1155                          */
1156                         current->signal->group_stop_count++;
1157                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1158                 }
1159
1160                 if (!cputime_eq(current->signal->it_virt_expires,
1161                                 cputime_zero) ||
1162                     !cputime_eq(current->signal->it_prof_expires,
1163                                 cputime_zero) ||
1164                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1165                     !list_empty(&current->signal->cpu_timers[0]) ||
1166                     !list_empty(&current->signal->cpu_timers[1]) ||
1167                     !list_empty(&current->signal->cpu_timers[2])) {
1168                         /*
1169                          * Have child wake up on its first tick to check
1170                          * for process CPU timers.
1171                          */
1172                         p->it_prof_expires = jiffies_to_cputime(1);
1173                 }
1174         }
1175
1176         /*
1177          * inherit ioprio
1178          */
1179         p->ioprio = current->ioprio;
1180
1181         SET_LINKS(p);
1182         if (unlikely(p->ptrace & PT_PTRACED))
1183                 __ptrace_link(p, current->parent);
1184
1185         if (thread_group_leader(p)) {
1186                 p->signal->tty = current->signal->tty;
1187                 p->signal->pgrp = process_group(current);
1188                 p->signal->session = current->signal->session;
1189                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1190                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1191                 if (p->pid)
1192                         __get_cpu_var(process_counts)++;
1193         }
1194         attach_pid(p, PIDTYPE_TGID, p->tgid);
1195         attach_pid(p, PIDTYPE_PID, p->pid);
1196
1197         nr_threads++;
1198         total_forks++;
1199         spin_unlock(&current->sighand->siglock);
1200         write_unlock_irq(&tasklist_lock);
1201         proc_fork_connector(p);
1202         return p;
1203
1204 bad_fork_cleanup_namespace:
1205         exit_namespace(p);
1206 bad_fork_cleanup_keys:
1207         exit_keys(p);
1208 bad_fork_cleanup_mm:
1209         if (p->mm)
1210                 mmput(p->mm);
1211 bad_fork_cleanup_signal:
1212         exit_signal(p);
1213 bad_fork_cleanup_sighand:
1214         exit_sighand(p);
1215 bad_fork_cleanup_fs:
1216         exit_fs(p); /* blocking */
1217 bad_fork_cleanup_files:
1218         exit_files(p); /* blocking */
1219 bad_fork_cleanup_semundo:
1220         exit_sem(p);
1221 bad_fork_cleanup_audit:
1222         audit_free(p);
1223 bad_fork_cleanup_security:
1224         security_task_free(p);
1225 bad_fork_cleanup_policy:
1226 #ifdef CONFIG_NUMA
1227         mpol_free(p->mempolicy);
1228 bad_fork_cleanup_cpuset:
1229 #endif
1230         cpuset_exit(p);
1231 bad_fork_cleanup:
1232         if (p->binfmt)
1233                 module_put(p->binfmt->module);
1234 bad_fork_cleanup_put_domain:
1235         module_put(task_thread_info(p)->exec_domain->module);
1236 bad_fork_cleanup_count:
1237         put_group_info(p->group_info);
1238         atomic_dec(&p->user->processes);
1239         free_uid(p->user);
1240 bad_fork_free:
1241         free_task(p);
1242 fork_out:
1243         return ERR_PTR(retval);
1244 }
1245
1246 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1247 {
1248         memset(regs, 0, sizeof(struct pt_regs));
1249         return regs;
1250 }
1251
1252 task_t * __devinit fork_idle(int cpu)
1253 {
1254         task_t *task;
1255         struct pt_regs regs;
1256
1257         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1258         if (!task)
1259                 return ERR_PTR(-ENOMEM);
1260         init_idle(task, cpu);
1261         unhash_process(task);
1262         return task;
1263 }
1264
1265 static inline int fork_traceflag (unsigned clone_flags)
1266 {
1267         if (clone_flags & CLONE_UNTRACED)
1268                 return 0;
1269         else if (clone_flags & CLONE_VFORK) {
1270                 if (current->ptrace & PT_TRACE_VFORK)
1271                         return PTRACE_EVENT_VFORK;
1272         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1273                 if (current->ptrace & PT_TRACE_CLONE)
1274                         return PTRACE_EVENT_CLONE;
1275         } else if (current->ptrace & PT_TRACE_FORK)
1276                 return PTRACE_EVENT_FORK;
1277
1278         return 0;
1279 }
1280
1281 /*
1282  *  Ok, this is the main fork-routine.
1283  *
1284  * It copies the process, and if successful kick-starts
1285  * it and waits for it to finish using the VM if required.
1286  */
1287 long do_fork(unsigned long clone_flags,
1288               unsigned long stack_start,
1289               struct pt_regs *regs,
1290               unsigned long stack_size,
1291               int __user *parent_tidptr,
1292               int __user *child_tidptr)
1293 {
1294         struct task_struct *p;
1295         int trace = 0;
1296         long pid = alloc_pidmap();
1297
1298         if (pid < 0)
1299                 return -EAGAIN;
1300         if (unlikely(current->ptrace)) {
1301                 trace = fork_traceflag (clone_flags);
1302                 if (trace)
1303                         clone_flags |= CLONE_PTRACE;
1304         }
1305
1306         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1307         /*
1308          * Do this prior waking up the new thread - the thread pointer
1309          * might get invalid after that point, if the thread exits quickly.
1310          */
1311         if (!IS_ERR(p)) {
1312                 struct completion vfork;
1313
1314                 if (clone_flags & CLONE_VFORK) {
1315                         p->vfork_done = &vfork;
1316                         init_completion(&vfork);
1317                 }
1318
1319                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1320                         /*
1321                          * We'll start up with an immediate SIGSTOP.
1322                          */
1323                         sigaddset(&p->pending.signal, SIGSTOP);
1324                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1325                 }
1326
1327                 if (!(clone_flags & CLONE_STOPPED))
1328                         wake_up_new_task(p, clone_flags);
1329                 else
1330                         p->state = TASK_STOPPED;
1331
1332                 if (unlikely (trace)) {
1333                         current->ptrace_message = pid;
1334                         ptrace_notify ((trace << 8) | SIGTRAP);
1335                 }
1336
1337                 if (clone_flags & CLONE_VFORK) {
1338                         wait_for_completion(&vfork);
1339                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1340                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1341                 }
1342         } else {
1343                 free_pidmap(pid);
1344                 pid = PTR_ERR(p);
1345         }
1346         return pid;
1347 }
1348
1349 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1350 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1351 #endif
1352
1353 void __init proc_caches_init(void)
1354 {
1355         sighand_cachep = kmem_cache_create("sighand_cache",
1356                         sizeof(struct sighand_struct), 0,
1357                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1358         signal_cachep = kmem_cache_create("signal_cache",
1359                         sizeof(struct signal_struct), 0,
1360                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1361         files_cachep = kmem_cache_create("files_cache", 
1362                         sizeof(struct files_struct), 0,
1363                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1364         fs_cachep = kmem_cache_create("fs_cache", 
1365                         sizeof(struct fs_struct), 0,
1366                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1367         vm_area_cachep = kmem_cache_create("vm_area_struct",
1368                         sizeof(struct vm_area_struct), 0,
1369                         SLAB_PANIC, NULL, NULL);
1370         mm_cachep = kmem_cache_create("mm_struct",
1371                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1372                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1373 }
1374
1375
1376 /*
1377  * Check constraints on flags passed to the unshare system call and
1378  * force unsharing of additional process context as appropriate.
1379  */
1380 static inline void check_unshare_flags(unsigned long *flags_ptr)
1381 {
1382         /*
1383          * If unsharing a thread from a thread group, must also
1384          * unshare vm.
1385          */
1386         if (*flags_ptr & CLONE_THREAD)
1387                 *flags_ptr |= CLONE_VM;
1388
1389         /*
1390          * If unsharing vm, must also unshare signal handlers.
1391          */
1392         if (*flags_ptr & CLONE_VM)
1393                 *flags_ptr |= CLONE_SIGHAND;
1394
1395         /*
1396          * If unsharing signal handlers and the task was created
1397          * using CLONE_THREAD, then must unshare the thread
1398          */
1399         if ((*flags_ptr & CLONE_SIGHAND) &&
1400             (atomic_read(&current->signal->count) > 1))
1401                 *flags_ptr |= CLONE_THREAD;
1402
1403         /*
1404          * If unsharing namespace, must also unshare filesystem information.
1405          */
1406         if (*flags_ptr & CLONE_NEWNS)
1407                 *flags_ptr |= CLONE_FS;
1408 }
1409
1410 /*
1411  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1412  */
1413 static int unshare_thread(unsigned long unshare_flags)
1414 {
1415         if (unshare_flags & CLONE_THREAD)
1416                 return -EINVAL;
1417
1418         return 0;
1419 }
1420
1421 /*
1422  * Unshare the filesystem structure if it is being shared
1423  */
1424 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1425 {
1426         struct fs_struct *fs = current->fs;
1427
1428         if ((unshare_flags & CLONE_FS) &&
1429             (fs && atomic_read(&fs->count) > 1)) {
1430                 *new_fsp = __copy_fs_struct(current->fs);
1431                 if (!*new_fsp)
1432                         return -ENOMEM;
1433         }
1434
1435         return 0;
1436 }
1437
1438 /*
1439  * Unshare the namespace structure if it is being shared
1440  */
1441 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1442 {
1443         struct namespace *ns = current->namespace;
1444
1445         if ((unshare_flags & CLONE_NEWNS) &&
1446             (ns && atomic_read(&ns->count) > 1)) {
1447                 if (!capable(CAP_SYS_ADMIN))
1448                         return -EPERM;
1449
1450                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1451                 if (!*new_nsp)
1452                         return -ENOMEM;
1453         }
1454
1455         return 0;
1456 }
1457
1458 /*
1459  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1460  * supported yet
1461  */
1462 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1463 {
1464         struct sighand_struct *sigh = current->sighand;
1465
1466         if ((unshare_flags & CLONE_SIGHAND) &&
1467             (sigh && atomic_read(&sigh->count) > 1))
1468                 return -EINVAL;
1469         else
1470                 return 0;
1471 }
1472
1473 /*
1474  * Unshare vm if it is being shared
1475  */
1476 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1477 {
1478         struct mm_struct *mm = current->mm;
1479
1480         if ((unshare_flags & CLONE_VM) &&
1481             (mm && atomic_read(&mm->mm_users) > 1)) {
1482                 return -EINVAL;
1483         }
1484
1485         return 0;
1486 }
1487
1488 /*
1489  * Unshare file descriptor table if it is being shared
1490  */
1491 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1492 {
1493         struct files_struct *fd = current->files;
1494         int error = 0;
1495
1496         if ((unshare_flags & CLONE_FILES) &&
1497             (fd && atomic_read(&fd->count) > 1)) {
1498                 *new_fdp = dup_fd(fd, &error);
1499                 if (!*new_fdp)
1500                         return error;
1501         }
1502
1503         return 0;
1504 }
1505
1506 /*
1507  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1508  * supported yet
1509  */
1510 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1511 {
1512         if (unshare_flags & CLONE_SYSVSEM)
1513                 return -EINVAL;
1514
1515         return 0;
1516 }
1517
1518 /*
1519  * unshare allows a process to 'unshare' part of the process
1520  * context which was originally shared using clone.  copy_*
1521  * functions used by do_fork() cannot be used here directly
1522  * because they modify an inactive task_struct that is being
1523  * constructed. Here we are modifying the current, active,
1524  * task_struct.
1525  */
1526 asmlinkage long sys_unshare(unsigned long unshare_flags)
1527 {
1528         int err = 0;
1529         struct fs_struct *fs, *new_fs = NULL;
1530         struct namespace *ns, *new_ns = NULL;
1531         struct sighand_struct *sigh, *new_sigh = NULL;
1532         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1533         struct files_struct *fd, *new_fd = NULL;
1534         struct sem_undo_list *new_ulist = NULL;
1535
1536         check_unshare_flags(&unshare_flags);
1537
1538         /* Return -EINVAL for all unsupported flags */
1539         err = -EINVAL;
1540         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1541                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1542                 goto bad_unshare_out;
1543
1544         if ((err = unshare_thread(unshare_flags)))
1545                 goto bad_unshare_out;
1546         if ((err = unshare_fs(unshare_flags, &new_fs)))
1547                 goto bad_unshare_cleanup_thread;
1548         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1549                 goto bad_unshare_cleanup_fs;
1550         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1551                 goto bad_unshare_cleanup_ns;
1552         if ((err = unshare_vm(unshare_flags, &new_mm)))
1553                 goto bad_unshare_cleanup_sigh;
1554         if ((err = unshare_fd(unshare_flags, &new_fd)))
1555                 goto bad_unshare_cleanup_vm;
1556         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1557                 goto bad_unshare_cleanup_fd;
1558
1559         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1560
1561                 task_lock(current);
1562
1563                 if (new_fs) {
1564                         fs = current->fs;
1565                         current->fs = new_fs;
1566                         new_fs = fs;
1567                 }
1568
1569                 if (new_ns) {
1570                         ns = current->namespace;
1571                         current->namespace = new_ns;
1572                         new_ns = ns;
1573                 }
1574
1575                 if (new_sigh) {
1576                         sigh = current->sighand;
1577                         rcu_assign_pointer(current->sighand, new_sigh);
1578                         new_sigh = sigh;
1579                 }
1580
1581                 if (new_mm) {
1582                         mm = current->mm;
1583                         active_mm = current->active_mm;
1584                         current->mm = new_mm;
1585                         current->active_mm = new_mm;
1586                         activate_mm(active_mm, new_mm);
1587                         new_mm = mm;
1588                 }
1589
1590                 if (new_fd) {
1591                         fd = current->files;
1592                         current->files = new_fd;
1593                         new_fd = fd;
1594                 }
1595
1596                 task_unlock(current);
1597         }
1598
1599 bad_unshare_cleanup_fd:
1600         if (new_fd)
1601                 put_files_struct(new_fd);
1602
1603 bad_unshare_cleanup_vm:
1604         if (new_mm)
1605                 mmput(new_mm);
1606
1607 bad_unshare_cleanup_sigh:
1608         if (new_sigh)
1609                 if (atomic_dec_and_test(&new_sigh->count))
1610                         kmem_cache_free(sighand_cachep, new_sigh);
1611
1612 bad_unshare_cleanup_ns:
1613         if (new_ns)
1614                 put_namespace(new_ns);
1615
1616 bad_unshare_cleanup_fs:
1617         if (new_fs)
1618                 put_fs_struct(new_fs);
1619
1620 bad_unshare_cleanup_thread:
1621 bad_unshare_out:
1622         return err;
1623 }