Merge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
[linux-2.6] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/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         return tsk;
185 }
186
187 #ifdef CONFIG_MMU
188 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
189 {
190         struct vm_area_struct *mpnt, *tmp, **pprev;
191         struct rb_node **rb_link, *rb_parent;
192         int retval;
193         unsigned long charge;
194         struct mempolicy *pol;
195
196         down_write(&oldmm->mmap_sem);
197         flush_cache_mm(oldmm);
198         down_write(&mm->mmap_sem);
199
200         mm->locked_vm = 0;
201         mm->mmap = NULL;
202         mm->mmap_cache = NULL;
203         mm->free_area_cache = oldmm->mmap_base;
204         mm->cached_hole_size = ~0UL;
205         mm->map_count = 0;
206         cpus_clear(mm->cpu_vm_mask);
207         mm->mm_rb = RB_ROOT;
208         rb_link = &mm->mm_rb.rb_node;
209         rb_parent = NULL;
210         pprev = &mm->mmap;
211
212         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
213                 struct file *file;
214
215                 if (mpnt->vm_flags & VM_DONTCOPY) {
216                         long pages = vma_pages(mpnt);
217                         mm->total_vm -= pages;
218                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
219                                                                 -pages);
220                         continue;
221                 }
222                 charge = 0;
223                 if (mpnt->vm_flags & VM_ACCOUNT) {
224                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
225                         if (security_vm_enough_memory(len))
226                                 goto fail_nomem;
227                         charge = len;
228                 }
229                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
230                 if (!tmp)
231                         goto fail_nomem;
232                 *tmp = *mpnt;
233                 pol = mpol_copy(vma_policy(mpnt));
234                 retval = PTR_ERR(pol);
235                 if (IS_ERR(pol))
236                         goto fail_nomem_policy;
237                 vma_set_policy(tmp, pol);
238                 tmp->vm_flags &= ~VM_LOCKED;
239                 tmp->vm_mm = mm;
240                 tmp->vm_next = NULL;
241                 anon_vma_link(tmp);
242                 file = tmp->vm_file;
243                 if (file) {
244                         struct inode *inode = file->f_dentry->d_inode;
245                         get_file(file);
246                         if (tmp->vm_flags & VM_DENYWRITE)
247                                 atomic_dec(&inode->i_writecount);
248       
249                         /* insert tmp into the share list, just after mpnt */
250                         spin_lock(&file->f_mapping->i_mmap_lock);
251                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
252                         flush_dcache_mmap_lock(file->f_mapping);
253                         vma_prio_tree_add(tmp, mpnt);
254                         flush_dcache_mmap_unlock(file->f_mapping);
255                         spin_unlock(&file->f_mapping->i_mmap_lock);
256                 }
257
258                 /*
259                  * Link in the new vma and copy the page table entries.
260                  */
261                 *pprev = tmp;
262                 pprev = &tmp->vm_next;
263
264                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
265                 rb_link = &tmp->vm_rb.rb_right;
266                 rb_parent = &tmp->vm_rb;
267
268                 mm->map_count++;
269                 retval = copy_page_range(mm, oldmm, mpnt);
270
271                 if (tmp->vm_ops && tmp->vm_ops->open)
272                         tmp->vm_ops->open(tmp);
273
274                 if (retval)
275                         goto out;
276         }
277         retval = 0;
278 out:
279         up_write(&mm->mmap_sem);
280         flush_tlb_mm(oldmm);
281         up_write(&oldmm->mmap_sem);
282         return retval;
283 fail_nomem_policy:
284         kmem_cache_free(vm_area_cachep, tmp);
285 fail_nomem:
286         retval = -ENOMEM;
287         vm_unacct_memory(charge);
288         goto out;
289 }
290
291 static inline int mm_alloc_pgd(struct mm_struct * mm)
292 {
293         mm->pgd = pgd_alloc(mm);
294         if (unlikely(!mm->pgd))
295                 return -ENOMEM;
296         return 0;
297 }
298
299 static inline void mm_free_pgd(struct mm_struct * mm)
300 {
301         pgd_free(mm->pgd);
302 }
303 #else
304 #define dup_mmap(mm, oldmm)     (0)
305 #define mm_alloc_pgd(mm)        (0)
306 #define mm_free_pgd(mm)
307 #endif /* CONFIG_MMU */
308
309  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
310
311 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
312 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
313
314 #include <linux/init_task.h>
315
316 static struct mm_struct * mm_init(struct mm_struct * mm)
317 {
318         atomic_set(&mm->mm_users, 1);
319         atomic_set(&mm->mm_count, 1);
320         init_rwsem(&mm->mmap_sem);
321         INIT_LIST_HEAD(&mm->mmlist);
322         mm->core_waiters = 0;
323         mm->nr_ptes = 0;
324         set_mm_counter(mm, file_rss, 0);
325         set_mm_counter(mm, anon_rss, 0);
326         spin_lock_init(&mm->page_table_lock);
327         rwlock_init(&mm->ioctx_list_lock);
328         mm->ioctx_list = NULL;
329         mm->free_area_cache = TASK_UNMAPPED_BASE;
330         mm->cached_hole_size = ~0UL;
331
332         if (likely(!mm_alloc_pgd(mm))) {
333                 mm->def_flags = 0;
334                 return mm;
335         }
336         free_mm(mm);
337         return NULL;
338 }
339
340 /*
341  * Allocate and initialize an mm_struct.
342  */
343 struct mm_struct * mm_alloc(void)
344 {
345         struct mm_struct * mm;
346
347         mm = allocate_mm();
348         if (mm) {
349                 memset(mm, 0, sizeof(*mm));
350                 mm = mm_init(mm);
351         }
352         return mm;
353 }
354
355 /*
356  * Called when the last reference to the mm
357  * is dropped: either by a lazy thread or by
358  * mmput. Free the page directory and the mm.
359  */
360 void fastcall __mmdrop(struct mm_struct *mm)
361 {
362         BUG_ON(mm == &init_mm);
363         mm_free_pgd(mm);
364         destroy_context(mm);
365         free_mm(mm);
366 }
367
368 /*
369  * Decrement the use count and release all resources for an mm.
370  */
371 void mmput(struct mm_struct *mm)
372 {
373         if (atomic_dec_and_test(&mm->mm_users)) {
374                 exit_aio(mm);
375                 exit_mmap(mm);
376                 if (!list_empty(&mm->mmlist)) {
377                         spin_lock(&mmlist_lock);
378                         list_del(&mm->mmlist);
379                         spin_unlock(&mmlist_lock);
380                 }
381                 put_swap_token(mm);
382                 mmdrop(mm);
383         }
384 }
385 EXPORT_SYMBOL_GPL(mmput);
386
387 /**
388  * get_task_mm - acquire a reference to the task's mm
389  *
390  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
391  * this kernel workthread has transiently adopted a user mm with use_mm,
392  * to do its AIO) is not set and if so returns a reference to it, after
393  * bumping up the use count.  User must release the mm via mmput()
394  * after use.  Typically used by /proc and ptrace.
395  */
396 struct mm_struct *get_task_mm(struct task_struct *task)
397 {
398         struct mm_struct *mm;
399
400         task_lock(task);
401         mm = task->mm;
402         if (mm) {
403                 if (task->flags & PF_BORROWED_MM)
404                         mm = NULL;
405                 else
406                         atomic_inc(&mm->mm_users);
407         }
408         task_unlock(task);
409         return mm;
410 }
411 EXPORT_SYMBOL_GPL(get_task_mm);
412
413 /* Please note the differences between mmput and mm_release.
414  * mmput is called whenever we stop holding onto a mm_struct,
415  * error success whatever.
416  *
417  * mm_release is called after a mm_struct has been removed
418  * from the current process.
419  *
420  * This difference is important for error handling, when we
421  * only half set up a mm_struct for a new process and need to restore
422  * the old one.  Because we mmput the new mm_struct before
423  * restoring the old one. . .
424  * Eric Biederman 10 January 1998
425  */
426 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
427 {
428         struct completion *vfork_done = tsk->vfork_done;
429
430         /* Get rid of any cached register state */
431         deactivate_mm(tsk, mm);
432
433         /* notify parent sleeping on vfork() */
434         if (vfork_done) {
435                 tsk->vfork_done = NULL;
436                 complete(vfork_done);
437         }
438         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
439                 u32 __user * tidptr = tsk->clear_child_tid;
440                 tsk->clear_child_tid = NULL;
441
442                 /*
443                  * We don't check the error code - if userspace has
444                  * not set up a proper pointer then tough luck.
445                  */
446                 put_user(0, tidptr);
447                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
448         }
449 }
450
451 /*
452  * Allocate a new mm structure and copy contents from the
453  * mm structure of the passed in task structure.
454  */
455 static struct mm_struct *dup_mm(struct task_struct *tsk)
456 {
457         struct mm_struct *mm, *oldmm = current->mm;
458         int err;
459
460         if (!oldmm)
461                 return NULL;
462
463         mm = allocate_mm();
464         if (!mm)
465                 goto fail_nomem;
466
467         memcpy(mm, oldmm, sizeof(*mm));
468
469         if (!mm_init(mm))
470                 goto fail_nomem;
471
472         if (init_new_context(tsk, mm))
473                 goto fail_nocontext;
474
475         err = dup_mmap(mm, oldmm);
476         if (err)
477                 goto free_pt;
478
479         mm->hiwater_rss = get_mm_rss(mm);
480         mm->hiwater_vm = mm->total_vm;
481
482         return mm;
483
484 free_pt:
485         mmput(mm);
486
487 fail_nomem:
488         return NULL;
489
490 fail_nocontext:
491         /*
492          * If init_new_context() failed, we cannot use mmput() to free the mm
493          * because it calls destroy_context()
494          */
495         mm_free_pgd(mm);
496         free_mm(mm);
497         return NULL;
498 }
499
500 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
501 {
502         struct mm_struct * mm, *oldmm;
503         int retval;
504
505         tsk->min_flt = tsk->maj_flt = 0;
506         tsk->nvcsw = tsk->nivcsw = 0;
507
508         tsk->mm = NULL;
509         tsk->active_mm = NULL;
510
511         /*
512          * Are we cloning a kernel thread?
513          *
514          * We need to steal a active VM for that..
515          */
516         oldmm = current->mm;
517         if (!oldmm)
518                 return 0;
519
520         if (clone_flags & CLONE_VM) {
521                 atomic_inc(&oldmm->mm_users);
522                 mm = oldmm;
523                 goto good_mm;
524         }
525
526         retval = -ENOMEM;
527         mm = dup_mm(tsk);
528         if (!mm)
529                 goto fail_nomem;
530
531 good_mm:
532         tsk->mm = mm;
533         tsk->active_mm = mm;
534         return 0;
535
536 fail_nomem:
537         return retval;
538 }
539
540 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
541 {
542         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
543         /* We don't need to lock fs - think why ;-) */
544         if (fs) {
545                 atomic_set(&fs->count, 1);
546                 rwlock_init(&fs->lock);
547                 fs->umask = old->umask;
548                 read_lock(&old->lock);
549                 fs->rootmnt = mntget(old->rootmnt);
550                 fs->root = dget(old->root);
551                 fs->pwdmnt = mntget(old->pwdmnt);
552                 fs->pwd = dget(old->pwd);
553                 if (old->altroot) {
554                         fs->altrootmnt = mntget(old->altrootmnt);
555                         fs->altroot = dget(old->altroot);
556                 } else {
557                         fs->altrootmnt = NULL;
558                         fs->altroot = NULL;
559                 }
560                 read_unlock(&old->lock);
561         }
562         return fs;
563 }
564
565 struct fs_struct *copy_fs_struct(struct fs_struct *old)
566 {
567         return __copy_fs_struct(old);
568 }
569
570 EXPORT_SYMBOL_GPL(copy_fs_struct);
571
572 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
573 {
574         if (clone_flags & CLONE_FS) {
575                 atomic_inc(&current->fs->count);
576                 return 0;
577         }
578         tsk->fs = __copy_fs_struct(current->fs);
579         if (!tsk->fs)
580                 return -ENOMEM;
581         return 0;
582 }
583
584 static int count_open_files(struct fdtable *fdt)
585 {
586         int size = fdt->max_fdset;
587         int i;
588
589         /* Find the last open fd */
590         for (i = size/(8*sizeof(long)); i > 0; ) {
591                 if (fdt->open_fds->fds_bits[--i])
592                         break;
593         }
594         i = (i+1) * 8 * sizeof(long);
595         return i;
596 }
597
598 static struct files_struct *alloc_files(void)
599 {
600         struct files_struct *newf;
601         struct fdtable *fdt;
602
603         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
604         if (!newf)
605                 goto out;
606
607         atomic_set(&newf->count, 1);
608
609         spin_lock_init(&newf->file_lock);
610         fdt = &newf->fdtab;
611         fdt->next_fd = 0;
612         fdt->max_fds = NR_OPEN_DEFAULT;
613         fdt->max_fdset = __FD_SETSIZE;
614         fdt->close_on_exec = &newf->close_on_exec_init;
615         fdt->open_fds = &newf->open_fds_init;
616         fdt->fd = &newf->fd_array[0];
617         INIT_RCU_HEAD(&fdt->rcu);
618         fdt->free_files = NULL;
619         fdt->next = NULL;
620         rcu_assign_pointer(newf->fdt, fdt);
621 out:
622         return newf;
623 }
624
625 /*
626  * Allocate a new files structure and copy contents from the
627  * passed in files structure.
628  */
629 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
630 {
631         struct files_struct *newf;
632         struct file **old_fds, **new_fds;
633         int open_files, size, i, expand;
634         struct fdtable *old_fdt, *new_fdt;
635
636         newf = alloc_files();
637         if (!newf)
638                 goto out;
639
640         spin_lock(&oldf->file_lock);
641         old_fdt = files_fdtable(oldf);
642         new_fdt = files_fdtable(newf);
643         size = old_fdt->max_fdset;
644         open_files = count_open_files(old_fdt);
645         expand = 0;
646
647         /*
648          * Check whether we need to allocate a larger fd array or fd set.
649          * Note: we're not a clone task, so the open count won't  change.
650          */
651         if (open_files > new_fdt->max_fdset) {
652                 new_fdt->max_fdset = 0;
653                 expand = 1;
654         }
655         if (open_files > new_fdt->max_fds) {
656                 new_fdt->max_fds = 0;
657                 expand = 1;
658         }
659
660         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
661         if (expand) {
662                 spin_unlock(&oldf->file_lock);
663                 spin_lock(&newf->file_lock);
664                 *errorp = expand_files(newf, open_files-1);
665                 spin_unlock(&newf->file_lock);
666                 if (*errorp < 0)
667                         goto out_release;
668                 new_fdt = files_fdtable(newf);
669                 /*
670                  * Reacquire the oldf lock and a pointer to its fd table
671                  * who knows it may have a new bigger fd table. We need
672                  * the latest pointer.
673                  */
674                 spin_lock(&oldf->file_lock);
675                 old_fdt = files_fdtable(oldf);
676         }
677
678         old_fds = old_fdt->fd;
679         new_fds = new_fdt->fd;
680
681         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
682         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
683
684         for (i = open_files; i != 0; i--) {
685                 struct file *f = *old_fds++;
686                 if (f) {
687                         get_file(f);
688                 } else {
689                         /*
690                          * The fd may be claimed in the fd bitmap but not yet
691                          * instantiated in the files array if a sibling thread
692                          * is partway through open().  So make sure that this
693                          * fd is available to the new process.
694                          */
695                         FD_CLR(open_files - i, new_fdt->open_fds);
696                 }
697                 rcu_assign_pointer(*new_fds++, f);
698         }
699         spin_unlock(&oldf->file_lock);
700
701         /* compute the remainder to be cleared */
702         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
703
704         /* This is long word aligned thus could use a optimized version */ 
705         memset(new_fds, 0, size); 
706
707         if (new_fdt->max_fdset > open_files) {
708                 int left = (new_fdt->max_fdset-open_files)/8;
709                 int start = open_files / (8 * sizeof(unsigned long));
710
711                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
712                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
713         }
714
715 out:
716         return newf;
717
718 out_release:
719         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
720         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
721         free_fd_array(new_fdt->fd, new_fdt->max_fds);
722         kmem_cache_free(files_cachep, newf);
723         goto out;
724 }
725
726 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
727 {
728         struct files_struct *oldf, *newf;
729         int error = 0;
730
731         /*
732          * A background process may not have any files ...
733          */
734         oldf = current->files;
735         if (!oldf)
736                 goto out;
737
738         if (clone_flags & CLONE_FILES) {
739                 atomic_inc(&oldf->count);
740                 goto out;
741         }
742
743         /*
744          * Note: we may be using current for both targets (See exec.c)
745          * This works because we cache current->files (old) as oldf. Don't
746          * break this.
747          */
748         tsk->files = NULL;
749         error = -ENOMEM;
750         newf = dup_fd(oldf, &error);
751         if (!newf)
752                 goto out;
753
754         tsk->files = newf;
755         error = 0;
756 out:
757         return error;
758 }
759
760 /*
761  *      Helper to unshare the files of the current task.
762  *      We don't want to expose copy_files internals to
763  *      the exec layer of the kernel.
764  */
765
766 int unshare_files(void)
767 {
768         struct files_struct *files  = current->files;
769         int rc;
770
771         if(!files)
772                 BUG();
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->real_timer.data = 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 #endif
1024
1025 #ifdef CONFIG_DEBUG_MUTEXES
1026         p->blocked_on = NULL; /* not blocked yet */
1027 #endif
1028
1029         p->tgid = p->pid;
1030         if (clone_flags & CLONE_THREAD)
1031                 p->tgid = current->tgid;
1032
1033         if ((retval = security_task_alloc(p)))
1034                 goto bad_fork_cleanup_policy;
1035         if ((retval = audit_alloc(p)))
1036                 goto bad_fork_cleanup_security;
1037         /* copy all the process information */
1038         if ((retval = copy_semundo(clone_flags, p)))
1039                 goto bad_fork_cleanup_audit;
1040         if ((retval = copy_files(clone_flags, p)))
1041                 goto bad_fork_cleanup_semundo;
1042         if ((retval = copy_fs(clone_flags, p)))
1043                 goto bad_fork_cleanup_files;
1044         if ((retval = copy_sighand(clone_flags, p)))
1045                 goto bad_fork_cleanup_fs;
1046         if ((retval = copy_signal(clone_flags, p)))
1047                 goto bad_fork_cleanup_sighand;
1048         if ((retval = copy_mm(clone_flags, p)))
1049                 goto bad_fork_cleanup_signal;
1050         if ((retval = copy_keys(clone_flags, p)))
1051                 goto bad_fork_cleanup_mm;
1052         if ((retval = copy_namespace(clone_flags, p)))
1053                 goto bad_fork_cleanup_keys;
1054         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1055         if (retval)
1056                 goto bad_fork_cleanup_namespace;
1057
1058         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1059         /*
1060          * Clear TID on mm_release()?
1061          */
1062         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1063
1064         /*
1065          * sigaltstack should be cleared when sharing the same VM
1066          */
1067         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1068                 p->sas_ss_sp = p->sas_ss_size = 0;
1069
1070         /*
1071          * Syscall tracing should be turned off in the child regardless
1072          * of CLONE_PTRACE.
1073          */
1074         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1075 #ifdef TIF_SYSCALL_EMU
1076         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1077 #endif
1078
1079         /* Our parent execution domain becomes current domain
1080            These must match for thread signalling to apply */
1081            
1082         p->parent_exec_id = p->self_exec_id;
1083
1084         /* ok, now we should be set up.. */
1085         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1086         p->pdeath_signal = 0;
1087         p->exit_state = 0;
1088
1089         /*
1090          * Ok, make it visible to the rest of the system.
1091          * We dont wake it up yet.
1092          */
1093         p->group_leader = p;
1094         INIT_LIST_HEAD(&p->ptrace_children);
1095         INIT_LIST_HEAD(&p->ptrace_list);
1096
1097         /* Perform scheduler related setup. Assign this task to a CPU. */
1098         sched_fork(p, clone_flags);
1099
1100         /* Need tasklist lock for parent etc handling! */
1101         write_lock_irq(&tasklist_lock);
1102
1103         /*
1104          * The task hasn't been attached yet, so its cpus_allowed mask will
1105          * not be changed, nor will its assigned CPU.
1106          *
1107          * The cpus_allowed mask of the parent may have changed after it was
1108          * copied first time - so re-copy it here, then check the child's CPU
1109          * to ensure it is on a valid CPU (and if not, just force it back to
1110          * parent's CPU). This avoids alot of nasty races.
1111          */
1112         p->cpus_allowed = current->cpus_allowed;
1113         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1114                         !cpu_online(task_cpu(p))))
1115                 set_task_cpu(p, smp_processor_id());
1116
1117         /*
1118          * Check for pending SIGKILL! The new thread should not be allowed
1119          * to slip out of an OOM kill. (or normal SIGKILL.)
1120          */
1121         if (sigismember(&current->pending.signal, SIGKILL)) {
1122                 write_unlock_irq(&tasklist_lock);
1123                 retval = -EINTR;
1124                 goto bad_fork_cleanup_namespace;
1125         }
1126
1127         /* CLONE_PARENT re-uses the old parent */
1128         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1129                 p->real_parent = current->real_parent;
1130         else
1131                 p->real_parent = current;
1132         p->parent = p->real_parent;
1133
1134         spin_lock(&current->sighand->siglock);
1135         if (clone_flags & CLONE_THREAD) {
1136                 /*
1137                  * Important: if an exit-all has been started then
1138                  * do not create this new thread - the whole thread
1139                  * group is supposed to exit anyway.
1140                  */
1141                 if (current->signal->flags & SIGNAL_GROUP_EXIT) {
1142                         spin_unlock(&current->sighand->siglock);
1143                         write_unlock_irq(&tasklist_lock);
1144                         retval = -EAGAIN;
1145                         goto bad_fork_cleanup_namespace;
1146                 }
1147                 p->group_leader = current->group_leader;
1148
1149                 if (current->signal->group_stop_count > 0) {
1150                         /*
1151                          * There is an all-stop in progress for the group.
1152                          * We ourselves will stop as soon as we check signals.
1153                          * Make the new thread part of that group stop too.
1154                          */
1155                         current->signal->group_stop_count++;
1156                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1157                 }
1158
1159                 if (!cputime_eq(current->signal->it_virt_expires,
1160                                 cputime_zero) ||
1161                     !cputime_eq(current->signal->it_prof_expires,
1162                                 cputime_zero) ||
1163                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1164                     !list_empty(&current->signal->cpu_timers[0]) ||
1165                     !list_empty(&current->signal->cpu_timers[1]) ||
1166                     !list_empty(&current->signal->cpu_timers[2])) {
1167                         /*
1168                          * Have child wake up on its first tick to check
1169                          * for process CPU timers.
1170                          */
1171                         p->it_prof_expires = jiffies_to_cputime(1);
1172                 }
1173         }
1174
1175         /*
1176          * inherit ioprio
1177          */
1178         p->ioprio = current->ioprio;
1179
1180         SET_LINKS(p);
1181         if (unlikely(p->ptrace & PT_PTRACED))
1182                 __ptrace_link(p, current->parent);
1183
1184         if (thread_group_leader(p)) {
1185                 p->signal->tty = current->signal->tty;
1186                 p->signal->pgrp = process_group(current);
1187                 p->signal->session = current->signal->session;
1188                 attach_pid(p, PIDTYPE_PGID, process_group(p));
1189                 attach_pid(p, PIDTYPE_SID, p->signal->session);
1190                 if (p->pid)
1191                         __get_cpu_var(process_counts)++;
1192         }
1193         attach_pid(p, PIDTYPE_TGID, p->tgid);
1194         attach_pid(p, PIDTYPE_PID, p->pid);
1195
1196         nr_threads++;
1197         total_forks++;
1198         spin_unlock(&current->sighand->siglock);
1199         write_unlock_irq(&tasklist_lock);
1200         proc_fork_connector(p);
1201         return p;
1202
1203 bad_fork_cleanup_namespace:
1204         exit_namespace(p);
1205 bad_fork_cleanup_keys:
1206         exit_keys(p);
1207 bad_fork_cleanup_mm:
1208         if (p->mm)
1209                 mmput(p->mm);
1210 bad_fork_cleanup_signal:
1211         exit_signal(p);
1212 bad_fork_cleanup_sighand:
1213         exit_sighand(p);
1214 bad_fork_cleanup_fs:
1215         exit_fs(p); /* blocking */
1216 bad_fork_cleanup_files:
1217         exit_files(p); /* blocking */
1218 bad_fork_cleanup_semundo:
1219         exit_sem(p);
1220 bad_fork_cleanup_audit:
1221         audit_free(p);
1222 bad_fork_cleanup_security:
1223         security_task_free(p);
1224 bad_fork_cleanup_policy:
1225 #ifdef CONFIG_NUMA
1226         mpol_free(p->mempolicy);
1227 bad_fork_cleanup_cpuset:
1228 #endif
1229         cpuset_exit(p);
1230 bad_fork_cleanup:
1231         if (p->binfmt)
1232                 module_put(p->binfmt->module);
1233 bad_fork_cleanup_put_domain:
1234         module_put(task_thread_info(p)->exec_domain->module);
1235 bad_fork_cleanup_count:
1236         put_group_info(p->group_info);
1237         atomic_dec(&p->user->processes);
1238         free_uid(p->user);
1239 bad_fork_free:
1240         free_task(p);
1241 fork_out:
1242         return ERR_PTR(retval);
1243 }
1244
1245 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1246 {
1247         memset(regs, 0, sizeof(struct pt_regs));
1248         return regs;
1249 }
1250
1251 task_t * __devinit fork_idle(int cpu)
1252 {
1253         task_t *task;
1254         struct pt_regs regs;
1255
1256         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1257         if (!task)
1258                 return ERR_PTR(-ENOMEM);
1259         init_idle(task, cpu);
1260         unhash_process(task);
1261         return task;
1262 }
1263
1264 static inline int fork_traceflag (unsigned clone_flags)
1265 {
1266         if (clone_flags & CLONE_UNTRACED)
1267                 return 0;
1268         else if (clone_flags & CLONE_VFORK) {
1269                 if (current->ptrace & PT_TRACE_VFORK)
1270                         return PTRACE_EVENT_VFORK;
1271         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1272                 if (current->ptrace & PT_TRACE_CLONE)
1273                         return PTRACE_EVENT_CLONE;
1274         } else if (current->ptrace & PT_TRACE_FORK)
1275                 return PTRACE_EVENT_FORK;
1276
1277         return 0;
1278 }
1279
1280 /*
1281  *  Ok, this is the main fork-routine.
1282  *
1283  * It copies the process, and if successful kick-starts
1284  * it and waits for it to finish using the VM if required.
1285  */
1286 long do_fork(unsigned long clone_flags,
1287               unsigned long stack_start,
1288               struct pt_regs *regs,
1289               unsigned long stack_size,
1290               int __user *parent_tidptr,
1291               int __user *child_tidptr)
1292 {
1293         struct task_struct *p;
1294         int trace = 0;
1295         long pid = alloc_pidmap();
1296
1297         if (pid < 0)
1298                 return -EAGAIN;
1299         if (unlikely(current->ptrace)) {
1300                 trace = fork_traceflag (clone_flags);
1301                 if (trace)
1302                         clone_flags |= CLONE_PTRACE;
1303         }
1304
1305         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
1306         /*
1307          * Do this prior waking up the new thread - the thread pointer
1308          * might get invalid after that point, if the thread exits quickly.
1309          */
1310         if (!IS_ERR(p)) {
1311                 struct completion vfork;
1312
1313                 if (clone_flags & CLONE_VFORK) {
1314                         p->vfork_done = &vfork;
1315                         init_completion(&vfork);
1316                 }
1317
1318                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1319                         /*
1320                          * We'll start up with an immediate SIGSTOP.
1321                          */
1322                         sigaddset(&p->pending.signal, SIGSTOP);
1323                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1324                 }
1325
1326                 if (!(clone_flags & CLONE_STOPPED))
1327                         wake_up_new_task(p, clone_flags);
1328                 else
1329                         p->state = TASK_STOPPED;
1330
1331                 if (unlikely (trace)) {
1332                         current->ptrace_message = pid;
1333                         ptrace_notify ((trace << 8) | SIGTRAP);
1334                 }
1335
1336                 if (clone_flags & CLONE_VFORK) {
1337                         wait_for_completion(&vfork);
1338                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE))
1339                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1340                 }
1341         } else {
1342                 free_pidmap(pid);
1343                 pid = PTR_ERR(p);
1344         }
1345         return pid;
1346 }
1347
1348 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1349 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1350 #endif
1351
1352 void __init proc_caches_init(void)
1353 {
1354         sighand_cachep = kmem_cache_create("sighand_cache",
1355                         sizeof(struct sighand_struct), 0,
1356                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1357         signal_cachep = kmem_cache_create("signal_cache",
1358                         sizeof(struct signal_struct), 0,
1359                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1360         files_cachep = kmem_cache_create("files_cache", 
1361                         sizeof(struct files_struct), 0,
1362                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1363         fs_cachep = kmem_cache_create("fs_cache", 
1364                         sizeof(struct fs_struct), 0,
1365                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1366         vm_area_cachep = kmem_cache_create("vm_area_struct",
1367                         sizeof(struct vm_area_struct), 0,
1368                         SLAB_PANIC, NULL, NULL);
1369         mm_cachep = kmem_cache_create("mm_struct",
1370                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1371                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1372 }
1373
1374
1375 /*
1376  * Check constraints on flags passed to the unshare system call and
1377  * force unsharing of additional process context as appropriate.
1378  */
1379 static inline void check_unshare_flags(unsigned long *flags_ptr)
1380 {
1381         /*
1382          * If unsharing a thread from a thread group, must also
1383          * unshare vm.
1384          */
1385         if (*flags_ptr & CLONE_THREAD)
1386                 *flags_ptr |= CLONE_VM;
1387
1388         /*
1389          * If unsharing vm, must also unshare signal handlers.
1390          */
1391         if (*flags_ptr & CLONE_VM)
1392                 *flags_ptr |= CLONE_SIGHAND;
1393
1394         /*
1395          * If unsharing signal handlers and the task was created
1396          * using CLONE_THREAD, then must unshare the thread
1397          */
1398         if ((*flags_ptr & CLONE_SIGHAND) &&
1399             (atomic_read(&current->signal->count) > 1))
1400                 *flags_ptr |= CLONE_THREAD;
1401
1402         /*
1403          * If unsharing namespace, must also unshare filesystem information.
1404          */
1405         if (*flags_ptr & CLONE_NEWNS)
1406                 *flags_ptr |= CLONE_FS;
1407 }
1408
1409 /*
1410  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1411  */
1412 static int unshare_thread(unsigned long unshare_flags)
1413 {
1414         if (unshare_flags & CLONE_THREAD)
1415                 return -EINVAL;
1416
1417         return 0;
1418 }
1419
1420 /*
1421  * Unshare the filesystem structure if it is being shared
1422  */
1423 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1424 {
1425         struct fs_struct *fs = current->fs;
1426
1427         if ((unshare_flags & CLONE_FS) &&
1428             (fs && atomic_read(&fs->count) > 1)) {
1429                 *new_fsp = __copy_fs_struct(current->fs);
1430                 if (!*new_fsp)
1431                         return -ENOMEM;
1432         }
1433
1434         return 0;
1435 }
1436
1437 /*
1438  * Unshare the namespace structure if it is being shared
1439  */
1440 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1441 {
1442         struct namespace *ns = current->namespace;
1443
1444         if ((unshare_flags & CLONE_NEWNS) &&
1445             (ns && atomic_read(&ns->count) > 1)) {
1446                 if (!capable(CAP_SYS_ADMIN))
1447                         return -EPERM;
1448
1449                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1450                 if (!*new_nsp)
1451                         return -ENOMEM;
1452         }
1453
1454         return 0;
1455 }
1456
1457 /*
1458  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1459  * supported yet
1460  */
1461 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1462 {
1463         struct sighand_struct *sigh = current->sighand;
1464
1465         if ((unshare_flags & CLONE_SIGHAND) &&
1466             (sigh && atomic_read(&sigh->count) > 1))
1467                 return -EINVAL;
1468         else
1469                 return 0;
1470 }
1471
1472 /*
1473  * Unshare vm if it is being shared
1474  */
1475 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1476 {
1477         struct mm_struct *mm = current->mm;
1478
1479         if ((unshare_flags & CLONE_VM) &&
1480             (mm && atomic_read(&mm->mm_users) > 1)) {
1481                 return -EINVAL;
1482         }
1483
1484         return 0;
1485 }
1486
1487 /*
1488  * Unshare file descriptor table if it is being shared
1489  */
1490 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1491 {
1492         struct files_struct *fd = current->files;
1493         int error = 0;
1494
1495         if ((unshare_flags & CLONE_FILES) &&
1496             (fd && atomic_read(&fd->count) > 1)) {
1497                 *new_fdp = dup_fd(fd, &error);
1498                 if (!*new_fdp)
1499                         return error;
1500         }
1501
1502         return 0;
1503 }
1504
1505 /*
1506  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1507  * supported yet
1508  */
1509 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1510 {
1511         if (unshare_flags & CLONE_SYSVSEM)
1512                 return -EINVAL;
1513
1514         return 0;
1515 }
1516
1517 /*
1518  * unshare allows a process to 'unshare' part of the process
1519  * context which was originally shared using clone.  copy_*
1520  * functions used by do_fork() cannot be used here directly
1521  * because they modify an inactive task_struct that is being
1522  * constructed. Here we are modifying the current, active,
1523  * task_struct.
1524  */
1525 asmlinkage long sys_unshare(unsigned long unshare_flags)
1526 {
1527         int err = 0;
1528         struct fs_struct *fs, *new_fs = NULL;
1529         struct namespace *ns, *new_ns = NULL;
1530         struct sighand_struct *sigh, *new_sigh = NULL;
1531         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1532         struct files_struct *fd, *new_fd = NULL;
1533         struct sem_undo_list *new_ulist = NULL;
1534
1535         check_unshare_flags(&unshare_flags);
1536
1537         /* Return -EINVAL for all unsupported flags */
1538         err = -EINVAL;
1539         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1540                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1541                 goto bad_unshare_out;
1542
1543         if ((err = unshare_thread(unshare_flags)))
1544                 goto bad_unshare_out;
1545         if ((err = unshare_fs(unshare_flags, &new_fs)))
1546                 goto bad_unshare_cleanup_thread;
1547         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1548                 goto bad_unshare_cleanup_fs;
1549         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1550                 goto bad_unshare_cleanup_ns;
1551         if ((err = unshare_vm(unshare_flags, &new_mm)))
1552                 goto bad_unshare_cleanup_sigh;
1553         if ((err = unshare_fd(unshare_flags, &new_fd)))
1554                 goto bad_unshare_cleanup_vm;
1555         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1556                 goto bad_unshare_cleanup_fd;
1557
1558         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1559
1560                 task_lock(current);
1561
1562                 if (new_fs) {
1563                         fs = current->fs;
1564                         current->fs = new_fs;
1565                         new_fs = fs;
1566                 }
1567
1568                 if (new_ns) {
1569                         ns = current->namespace;
1570                         current->namespace = new_ns;
1571                         new_ns = ns;
1572                 }
1573
1574                 if (new_sigh) {
1575                         sigh = current->sighand;
1576                         rcu_assign_pointer(current->sighand, new_sigh);
1577                         new_sigh = sigh;
1578                 }
1579
1580                 if (new_mm) {
1581                         mm = current->mm;
1582                         active_mm = current->active_mm;
1583                         current->mm = new_mm;
1584                         current->active_mm = new_mm;
1585                         activate_mm(active_mm, new_mm);
1586                         new_mm = mm;
1587                 }
1588
1589                 if (new_fd) {
1590                         fd = current->files;
1591                         current->files = new_fd;
1592                         new_fd = fd;
1593                 }
1594
1595                 task_unlock(current);
1596         }
1597
1598 bad_unshare_cleanup_fd:
1599         if (new_fd)
1600                 put_files_struct(new_fd);
1601
1602 bad_unshare_cleanup_vm:
1603         if (new_mm)
1604                 mmput(new_mm);
1605
1606 bad_unshare_cleanup_sigh:
1607         if (new_sigh)
1608                 if (atomic_dec_and_test(&new_sigh->count))
1609                         kmem_cache_free(sighand_cachep, new_sigh);
1610
1611 bad_unshare_cleanup_ns:
1612         if (new_ns)
1613                 put_namespace(new_ns);
1614
1615 bad_unshare_cleanup_fs:
1616         if (new_fs)
1617                 put_fs_struct(new_fs);
1618
1619 bad_unshare_cleanup_thread:
1620 bad_unshare_out:
1621         return err;
1622 }