Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6
[powerpc.git] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/smp_lock.h>
18 #include <linux/module.h>
19 #include <linux/vmalloc.h>
20 #include <linux/completion.h>
21 #include <linux/namespace.h>
22 #include <linux/personality.h>
23 #include <linux/mempolicy.h>
24 #include <linux/sem.h>
25 #include <linux/file.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/fs.h>
30 #include <linux/capability.h>
31 #include <linux/cpu.h>
32 #include <linux/cpuset.h>
33 #include <linux/security.h>
34 #include <linux/swap.h>
35 #include <linux/syscalls.h>
36 #include <linux/jiffies.h>
37 #include <linux/futex.h>
38 #include <linux/rcupdate.h>
39 #include <linux/ptrace.h>
40 #include <linux/mount.h>
41 #include <linux/audit.h>
42 #include <linux/profile.h>
43 #include <linux/rmap.h>
44 #include <linux/acct.h>
45 #include <linux/tsacct_kern.h>
46 #include <linux/cn_proc.h>
47 #include <linux/delayacct.h>
48 #include <linux/taskstats_kern.h>
49 #include <linux/random.h>
50
51 #include <asm/pgtable.h>
52 #include <asm/pgalloc.h>
53 #include <asm/uaccess.h>
54 #include <asm/mmu_context.h>
55 #include <asm/cacheflush.h>
56 #include <asm/tlbflush.h>
57
58 /*
59  * Protected counters by write_lock_irq(&tasklist_lock)
60  */
61 unsigned long total_forks;      /* Handle normal Linux uptimes. */
62 int nr_threads;                 /* The idle threads do not count.. */
63
64 int max_threads;                /* tunable limit on nr_threads */
65
66 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
67
68 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
69
70 int nr_processes(void)
71 {
72         int cpu;
73         int total = 0;
74
75         for_each_online_cpu(cpu)
76                 total += per_cpu(process_counts, cpu);
77
78         return total;
79 }
80
81 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
82 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
83 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
84 static kmem_cache_t *task_struct_cachep;
85 #endif
86
87 /* SLAB cache for signal_struct structures (tsk->signal) */
88 static kmem_cache_t *signal_cachep;
89
90 /* SLAB cache for sighand_struct structures (tsk->sighand) */
91 kmem_cache_t *sighand_cachep;
92
93 /* SLAB cache for files_struct structures (tsk->files) */
94 kmem_cache_t *files_cachep;
95
96 /* SLAB cache for fs_struct structures (tsk->fs) */
97 kmem_cache_t *fs_cachep;
98
99 /* SLAB cache for vm_area_struct structures */
100 kmem_cache_t *vm_area_cachep;
101
102 /* SLAB cache for mm_struct structures (tsk->mm) */
103 static kmem_cache_t *mm_cachep;
104
105 void free_task(struct task_struct *tsk)
106 {
107         free_thread_info(tsk->thread_info);
108         rt_mutex_debug_task_free(tsk);
109         free_task_struct(tsk);
110 }
111 EXPORT_SYMBOL(free_task);
112
113 void __put_task_struct(struct task_struct *tsk)
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         security_task_free(tsk);
120         free_uid(tsk->user);
121         put_group_info(tsk->group_info);
122         delayacct_tsk_free(tsk);
123
124         if (!profile_handoff_task(tsk))
125                 free_task(tsk);
126 }
127
128 void __init fork_init(unsigned long mempages)
129 {
130 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
131 #ifndef ARCH_MIN_TASKALIGN
132 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
133 #endif
134         /* create a slab on which task_structs can be allocated */
135         task_struct_cachep =
136                 kmem_cache_create("task_struct", sizeof(struct task_struct),
137                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
138 #endif
139
140         /*
141          * The default maximum number of threads is set to a safe
142          * value: the thread structures can take up at most half
143          * of memory.
144          */
145         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
146
147         /*
148          * we need to allow at least 20 threads to boot a system
149          */
150         if(max_threads < 20)
151                 max_threads = 20;
152
153         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
154         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
155         init_task.signal->rlim[RLIMIT_SIGPENDING] =
156                 init_task.signal->rlim[RLIMIT_NPROC];
157 }
158
159 static struct task_struct *dup_task_struct(struct task_struct *orig)
160 {
161         struct task_struct *tsk;
162         struct thread_info *ti;
163
164         prepare_to_copy(orig);
165
166         tsk = alloc_task_struct();
167         if (!tsk)
168                 return NULL;
169
170         ti = alloc_thread_info(tsk);
171         if (!ti) {
172                 free_task_struct(tsk);
173                 return NULL;
174         }
175
176         *tsk = *orig;
177         tsk->thread_info = ti;
178         setup_thread_stack(tsk, orig);
179
180 #ifdef CONFIG_CC_STACKPROTECTOR
181         tsk->stack_canary = get_random_int();
182 #endif
183
184         /* One for us, one for whoever does the "release_task()" (usually parent) */
185         atomic_set(&tsk->usage,2);
186         atomic_set(&tsk->fs_excl, 0);
187 #ifdef CONFIG_BLK_DEV_IO_TRACE
188         tsk->btrace_seq = 0;
189 #endif
190         tsk->splice_pipe = NULL;
191         return tsk;
192 }
193
194 #ifdef CONFIG_MMU
195 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
196 {
197         struct vm_area_struct *mpnt, *tmp, **pprev;
198         struct rb_node **rb_link, *rb_parent;
199         int retval;
200         unsigned long charge;
201         struct mempolicy *pol;
202
203         down_write(&oldmm->mmap_sem);
204         flush_cache_mm(oldmm);
205         /*
206          * Not linked in yet - no deadlock potential:
207          */
208         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
209
210         mm->locked_vm = 0;
211         mm->mmap = NULL;
212         mm->mmap_cache = NULL;
213         mm->free_area_cache = oldmm->mmap_base;
214         mm->cached_hole_size = ~0UL;
215         mm->map_count = 0;
216         cpus_clear(mm->cpu_vm_mask);
217         mm->mm_rb = RB_ROOT;
218         rb_link = &mm->mm_rb.rb_node;
219         rb_parent = NULL;
220         pprev = &mm->mmap;
221
222         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
223                 struct file *file;
224
225                 if (mpnt->vm_flags & VM_DONTCOPY) {
226                         long pages = vma_pages(mpnt);
227                         mm->total_vm -= pages;
228                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
229                                                                 -pages);
230                         continue;
231                 }
232                 charge = 0;
233                 if (mpnt->vm_flags & VM_ACCOUNT) {
234                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
235                         if (security_vm_enough_memory(len))
236                                 goto fail_nomem;
237                         charge = len;
238                 }
239                 tmp = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
240                 if (!tmp)
241                         goto fail_nomem;
242                 *tmp = *mpnt;
243                 pol = mpol_copy(vma_policy(mpnt));
244                 retval = PTR_ERR(pol);
245                 if (IS_ERR(pol))
246                         goto fail_nomem_policy;
247                 vma_set_policy(tmp, pol);
248                 tmp->vm_flags &= ~VM_LOCKED;
249                 tmp->vm_mm = mm;
250                 tmp->vm_next = NULL;
251                 anon_vma_link(tmp);
252                 file = tmp->vm_file;
253                 if (file) {
254                         struct inode *inode = file->f_dentry->d_inode;
255                         get_file(file);
256                         if (tmp->vm_flags & VM_DENYWRITE)
257                                 atomic_dec(&inode->i_writecount);
258       
259                         /* insert tmp into the share list, just after mpnt */
260                         spin_lock(&file->f_mapping->i_mmap_lock);
261                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
262                         flush_dcache_mmap_lock(file->f_mapping);
263                         vma_prio_tree_add(tmp, mpnt);
264                         flush_dcache_mmap_unlock(file->f_mapping);
265                         spin_unlock(&file->f_mapping->i_mmap_lock);
266                 }
267
268                 /*
269                  * Link in the new vma and copy the page table entries.
270                  */
271                 *pprev = tmp;
272                 pprev = &tmp->vm_next;
273
274                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
275                 rb_link = &tmp->vm_rb.rb_right;
276                 rb_parent = &tmp->vm_rb;
277
278                 mm->map_count++;
279                 retval = copy_page_range(mm, oldmm, mpnt);
280
281                 if (tmp->vm_ops && tmp->vm_ops->open)
282                         tmp->vm_ops->open(tmp);
283
284                 if (retval)
285                         goto out;
286         }
287         retval = 0;
288 out:
289         up_write(&mm->mmap_sem);
290         flush_tlb_mm(oldmm);
291         up_write(&oldmm->mmap_sem);
292         return retval;
293 fail_nomem_policy:
294         kmem_cache_free(vm_area_cachep, tmp);
295 fail_nomem:
296         retval = -ENOMEM;
297         vm_unacct_memory(charge);
298         goto out;
299 }
300
301 static inline int mm_alloc_pgd(struct mm_struct * mm)
302 {
303         mm->pgd = pgd_alloc(mm);
304         if (unlikely(!mm->pgd))
305                 return -ENOMEM;
306         return 0;
307 }
308
309 static inline void mm_free_pgd(struct mm_struct * mm)
310 {
311         pgd_free(mm->pgd);
312 }
313 #else
314 #define dup_mmap(mm, oldmm)     (0)
315 #define mm_alloc_pgd(mm)        (0)
316 #define mm_free_pgd(mm)
317 #endif /* CONFIG_MMU */
318
319  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
320
321 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, SLAB_KERNEL))
322 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
323
324 #include <linux/init_task.h>
325
326 static struct mm_struct * mm_init(struct mm_struct * mm)
327 {
328         atomic_set(&mm->mm_users, 1);
329         atomic_set(&mm->mm_count, 1);
330         init_rwsem(&mm->mmap_sem);
331         INIT_LIST_HEAD(&mm->mmlist);
332         mm->core_waiters = 0;
333         mm->nr_ptes = 0;
334         set_mm_counter(mm, file_rss, 0);
335         set_mm_counter(mm, anon_rss, 0);
336         spin_lock_init(&mm->page_table_lock);
337         rwlock_init(&mm->ioctx_list_lock);
338         mm->ioctx_list = NULL;
339         mm->free_area_cache = TASK_UNMAPPED_BASE;
340         mm->cached_hole_size = ~0UL;
341
342         if (likely(!mm_alloc_pgd(mm))) {
343                 mm->def_flags = 0;
344                 return mm;
345         }
346         free_mm(mm);
347         return NULL;
348 }
349
350 /*
351  * Allocate and initialize an mm_struct.
352  */
353 struct mm_struct * mm_alloc(void)
354 {
355         struct mm_struct * mm;
356
357         mm = allocate_mm();
358         if (mm) {
359                 memset(mm, 0, sizeof(*mm));
360                 mm = mm_init(mm);
361         }
362         return mm;
363 }
364
365 /*
366  * Called when the last reference to the mm
367  * is dropped: either by a lazy thread or by
368  * mmput. Free the page directory and the mm.
369  */
370 void fastcall __mmdrop(struct mm_struct *mm)
371 {
372         BUG_ON(mm == &init_mm);
373         mm_free_pgd(mm);
374         destroy_context(mm);
375         free_mm(mm);
376 }
377
378 /*
379  * Decrement the use count and release all resources for an mm.
380  */
381 void mmput(struct mm_struct *mm)
382 {
383         might_sleep();
384
385         if (atomic_dec_and_test(&mm->mm_users)) {
386                 exit_aio(mm);
387                 exit_mmap(mm);
388                 if (!list_empty(&mm->mmlist)) {
389                         spin_lock(&mmlist_lock);
390                         list_del(&mm->mmlist);
391                         spin_unlock(&mmlist_lock);
392                 }
393                 put_swap_token(mm);
394                 mmdrop(mm);
395         }
396 }
397 EXPORT_SYMBOL_GPL(mmput);
398
399 /**
400  * get_task_mm - acquire a reference to the task's mm
401  *
402  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
403  * this kernel workthread has transiently adopted a user mm with use_mm,
404  * to do its AIO) is not set and if so returns a reference to it, after
405  * bumping up the use count.  User must release the mm via mmput()
406  * after use.  Typically used by /proc and ptrace.
407  */
408 struct mm_struct *get_task_mm(struct task_struct *task)
409 {
410         struct mm_struct *mm;
411
412         task_lock(task);
413         mm = task->mm;
414         if (mm) {
415                 if (task->flags & PF_BORROWED_MM)
416                         mm = NULL;
417                 else
418                         atomic_inc(&mm->mm_users);
419         }
420         task_unlock(task);
421         return mm;
422 }
423 EXPORT_SYMBOL_GPL(get_task_mm);
424
425 /* Please note the differences between mmput and mm_release.
426  * mmput is called whenever we stop holding onto a mm_struct,
427  * error success whatever.
428  *
429  * mm_release is called after a mm_struct has been removed
430  * from the current process.
431  *
432  * This difference is important for error handling, when we
433  * only half set up a mm_struct for a new process and need to restore
434  * the old one.  Because we mmput the new mm_struct before
435  * restoring the old one. . .
436  * Eric Biederman 10 January 1998
437  */
438 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
439 {
440         struct completion *vfork_done = tsk->vfork_done;
441
442         /* Get rid of any cached register state */
443         deactivate_mm(tsk, mm);
444
445         /* notify parent sleeping on vfork() */
446         if (vfork_done) {
447                 tsk->vfork_done = NULL;
448                 complete(vfork_done);
449         }
450         if (tsk->clear_child_tid && atomic_read(&mm->mm_users) > 1) {
451                 u32 __user * tidptr = tsk->clear_child_tid;
452                 tsk->clear_child_tid = NULL;
453
454                 /*
455                  * We don't check the error code - if userspace has
456                  * not set up a proper pointer then tough luck.
457                  */
458                 put_user(0, tidptr);
459                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
460         }
461 }
462
463 /*
464  * Allocate a new mm structure and copy contents from the
465  * mm structure of the passed in task structure.
466  */
467 static struct mm_struct *dup_mm(struct task_struct *tsk)
468 {
469         struct mm_struct *mm, *oldmm = current->mm;
470         int err;
471
472         if (!oldmm)
473                 return NULL;
474
475         mm = allocate_mm();
476         if (!mm)
477                 goto fail_nomem;
478
479         memcpy(mm, oldmm, sizeof(*mm));
480
481         if (!mm_init(mm))
482                 goto fail_nomem;
483
484         if (init_new_context(tsk, mm))
485                 goto fail_nocontext;
486
487         err = dup_mmap(mm, oldmm);
488         if (err)
489                 goto free_pt;
490
491         mm->hiwater_rss = get_mm_rss(mm);
492         mm->hiwater_vm = mm->total_vm;
493
494         return mm;
495
496 free_pt:
497         mmput(mm);
498
499 fail_nomem:
500         return NULL;
501
502 fail_nocontext:
503         /*
504          * If init_new_context() failed, we cannot use mmput() to free the mm
505          * because it calls destroy_context()
506          */
507         mm_free_pgd(mm);
508         free_mm(mm);
509         return NULL;
510 }
511
512 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
513 {
514         struct mm_struct * mm, *oldmm;
515         int retval;
516
517         tsk->min_flt = tsk->maj_flt = 0;
518         tsk->nvcsw = tsk->nivcsw = 0;
519
520         tsk->mm = NULL;
521         tsk->active_mm = NULL;
522
523         /*
524          * Are we cloning a kernel thread?
525          *
526          * We need to steal a active VM for that..
527          */
528         oldmm = current->mm;
529         if (!oldmm)
530                 return 0;
531
532         if (clone_flags & CLONE_VM) {
533                 atomic_inc(&oldmm->mm_users);
534                 mm = oldmm;
535                 goto good_mm;
536         }
537
538         retval = -ENOMEM;
539         mm = dup_mm(tsk);
540         if (!mm)
541                 goto fail_nomem;
542
543 good_mm:
544         tsk->mm = mm;
545         tsk->active_mm = mm;
546         return 0;
547
548 fail_nomem:
549         return retval;
550 }
551
552 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
553 {
554         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
555         /* We don't need to lock fs - think why ;-) */
556         if (fs) {
557                 atomic_set(&fs->count, 1);
558                 rwlock_init(&fs->lock);
559                 fs->umask = old->umask;
560                 read_lock(&old->lock);
561                 fs->rootmnt = mntget(old->rootmnt);
562                 fs->root = dget(old->root);
563                 fs->pwdmnt = mntget(old->pwdmnt);
564                 fs->pwd = dget(old->pwd);
565                 if (old->altroot) {
566                         fs->altrootmnt = mntget(old->altrootmnt);
567                         fs->altroot = dget(old->altroot);
568                 } else {
569                         fs->altrootmnt = NULL;
570                         fs->altroot = NULL;
571                 }
572                 read_unlock(&old->lock);
573         }
574         return fs;
575 }
576
577 struct fs_struct *copy_fs_struct(struct fs_struct *old)
578 {
579         return __copy_fs_struct(old);
580 }
581
582 EXPORT_SYMBOL_GPL(copy_fs_struct);
583
584 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
585 {
586         if (clone_flags & CLONE_FS) {
587                 atomic_inc(&current->fs->count);
588                 return 0;
589         }
590         tsk->fs = __copy_fs_struct(current->fs);
591         if (!tsk->fs)
592                 return -ENOMEM;
593         return 0;
594 }
595
596 static int count_open_files(struct fdtable *fdt)
597 {
598         int size = fdt->max_fdset;
599         int i;
600
601         /* Find the last open fd */
602         for (i = size/(8*sizeof(long)); i > 0; ) {
603                 if (fdt->open_fds->fds_bits[--i])
604                         break;
605         }
606         i = (i+1) * 8 * sizeof(long);
607         return i;
608 }
609
610 static struct files_struct *alloc_files(void)
611 {
612         struct files_struct *newf;
613         struct fdtable *fdt;
614
615         newf = kmem_cache_alloc(files_cachep, SLAB_KERNEL);
616         if (!newf)
617                 goto out;
618
619         atomic_set(&newf->count, 1);
620
621         spin_lock_init(&newf->file_lock);
622         newf->next_fd = 0;
623         fdt = &newf->fdtab;
624         fdt->max_fds = NR_OPEN_DEFAULT;
625         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
626         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
627         fdt->open_fds = (fd_set *)&newf->open_fds_init;
628         fdt->fd = &newf->fd_array[0];
629         INIT_RCU_HEAD(&fdt->rcu);
630         fdt->free_files = NULL;
631         fdt->next = NULL;
632         rcu_assign_pointer(newf->fdt, fdt);
633 out:
634         return newf;
635 }
636
637 /*
638  * Allocate a new files structure and copy contents from the
639  * passed in files structure.
640  * errorp will be valid only when the returned files_struct is NULL.
641  */
642 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
643 {
644         struct files_struct *newf;
645         struct file **old_fds, **new_fds;
646         int open_files, size, i, expand;
647         struct fdtable *old_fdt, *new_fdt;
648
649         *errorp = -ENOMEM;
650         newf = alloc_files();
651         if (!newf)
652                 goto out;
653
654         spin_lock(&oldf->file_lock);
655         old_fdt = files_fdtable(oldf);
656         new_fdt = files_fdtable(newf);
657         size = old_fdt->max_fdset;
658         open_files = count_open_files(old_fdt);
659         expand = 0;
660
661         /*
662          * Check whether we need to allocate a larger fd array or fd set.
663          * Note: we're not a clone task, so the open count won't  change.
664          */
665         if (open_files > new_fdt->max_fdset) {
666                 new_fdt->max_fdset = 0;
667                 expand = 1;
668         }
669         if (open_files > new_fdt->max_fds) {
670                 new_fdt->max_fds = 0;
671                 expand = 1;
672         }
673
674         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
675         if (expand) {
676                 spin_unlock(&oldf->file_lock);
677                 spin_lock(&newf->file_lock);
678                 *errorp = expand_files(newf, open_files-1);
679                 spin_unlock(&newf->file_lock);
680                 if (*errorp < 0)
681                         goto out_release;
682                 new_fdt = files_fdtable(newf);
683                 /*
684                  * Reacquire the oldf lock and a pointer to its fd table
685                  * who knows it may have a new bigger fd table. We need
686                  * the latest pointer.
687                  */
688                 spin_lock(&oldf->file_lock);
689                 old_fdt = files_fdtable(oldf);
690         }
691
692         old_fds = old_fdt->fd;
693         new_fds = new_fdt->fd;
694
695         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
696         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
697
698         for (i = open_files; i != 0; i--) {
699                 struct file *f = *old_fds++;
700                 if (f) {
701                         get_file(f);
702                 } else {
703                         /*
704                          * The fd may be claimed in the fd bitmap but not yet
705                          * instantiated in the files array if a sibling thread
706                          * is partway through open().  So make sure that this
707                          * fd is available to the new process.
708                          */
709                         FD_CLR(open_files - i, new_fdt->open_fds);
710                 }
711                 rcu_assign_pointer(*new_fds++, f);
712         }
713         spin_unlock(&oldf->file_lock);
714
715         /* compute the remainder to be cleared */
716         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
717
718         /* This is long word aligned thus could use a optimized version */ 
719         memset(new_fds, 0, size); 
720
721         if (new_fdt->max_fdset > open_files) {
722                 int left = (new_fdt->max_fdset-open_files)/8;
723                 int start = open_files / (8 * sizeof(unsigned long));
724
725                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
726                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
727         }
728
729 out:
730         return newf;
731
732 out_release:
733         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
734         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
735         free_fd_array(new_fdt->fd, new_fdt->max_fds);
736         kmem_cache_free(files_cachep, newf);
737         return NULL;
738 }
739
740 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
741 {
742         struct files_struct *oldf, *newf;
743         int error = 0;
744
745         /*
746          * A background process may not have any files ...
747          */
748         oldf = current->files;
749         if (!oldf)
750                 goto out;
751
752         if (clone_flags & CLONE_FILES) {
753                 atomic_inc(&oldf->count);
754                 goto out;
755         }
756
757         /*
758          * Note: we may be using current for both targets (See exec.c)
759          * This works because we cache current->files (old) as oldf. Don't
760          * break this.
761          */
762         tsk->files = NULL;
763         newf = dup_fd(oldf, &error);
764         if (!newf)
765                 goto out;
766
767         tsk->files = newf;
768         error = 0;
769 out:
770         return error;
771 }
772
773 /*
774  *      Helper to unshare the files of the current task.
775  *      We don't want to expose copy_files internals to
776  *      the exec layer of the kernel.
777  */
778
779 int unshare_files(void)
780 {
781         struct files_struct *files  = current->files;
782         int rc;
783
784         BUG_ON(!files);
785
786         /* This can race but the race causes us to copy when we don't
787            need to and drop the copy */
788         if(atomic_read(&files->count) == 1)
789         {
790                 atomic_inc(&files->count);
791                 return 0;
792         }
793         rc = copy_files(0, current);
794         if(rc)
795                 current->files = files;
796         return rc;
797 }
798
799 EXPORT_SYMBOL(unshare_files);
800
801 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
802 {
803         struct sighand_struct *sig;
804
805         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
806                 atomic_inc(&current->sighand->count);
807                 return 0;
808         }
809         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
810         rcu_assign_pointer(tsk->sighand, sig);
811         if (!sig)
812                 return -ENOMEM;
813         atomic_set(&sig->count, 1);
814         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
815         return 0;
816 }
817
818 void __cleanup_sighand(struct sighand_struct *sighand)
819 {
820         if (atomic_dec_and_test(&sighand->count))
821                 kmem_cache_free(sighand_cachep, sighand);
822 }
823
824 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
825 {
826         struct signal_struct *sig;
827         int ret;
828
829         if (clone_flags & CLONE_THREAD) {
830                 atomic_inc(&current->signal->count);
831                 atomic_inc(&current->signal->live);
832                 taskstats_tgid_alloc(current->signal);
833                 return 0;
834         }
835         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
836         tsk->signal = sig;
837         if (!sig)
838                 return -ENOMEM;
839
840         ret = copy_thread_group_keys(tsk);
841         if (ret < 0) {
842                 kmem_cache_free(signal_cachep, sig);
843                 return ret;
844         }
845
846         atomic_set(&sig->count, 1);
847         atomic_set(&sig->live, 1);
848         init_waitqueue_head(&sig->wait_chldexit);
849         sig->flags = 0;
850         sig->group_exit_code = 0;
851         sig->group_exit_task = NULL;
852         sig->group_stop_count = 0;
853         sig->curr_target = NULL;
854         init_sigpending(&sig->shared_pending);
855         INIT_LIST_HEAD(&sig->posix_timers);
856
857         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
858         sig->it_real_incr.tv64 = 0;
859         sig->real_timer.function = it_real_fn;
860         sig->tsk = tsk;
861
862         sig->it_virt_expires = cputime_zero;
863         sig->it_virt_incr = cputime_zero;
864         sig->it_prof_expires = cputime_zero;
865         sig->it_prof_incr = cputime_zero;
866
867         sig->leader = 0;        /* session leadership doesn't inherit */
868         sig->tty_old_pgrp = 0;
869
870         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
871         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
872         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
873         sig->sched_time = 0;
874         INIT_LIST_HEAD(&sig->cpu_timers[0]);
875         INIT_LIST_HEAD(&sig->cpu_timers[1]);
876         INIT_LIST_HEAD(&sig->cpu_timers[2]);
877         taskstats_tgid_init(sig);
878
879         task_lock(current->group_leader);
880         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
881         task_unlock(current->group_leader);
882
883         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
884                 /*
885                  * New sole thread in the process gets an expiry time
886                  * of the whole CPU time limit.
887                  */
888                 tsk->it_prof_expires =
889                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
890         }
891         acct_init_pacct(&sig->pacct);
892
893         return 0;
894 }
895
896 void __cleanup_signal(struct signal_struct *sig)
897 {
898         exit_thread_group_keys(sig);
899         taskstats_tgid_free(sig);
900         kmem_cache_free(signal_cachep, sig);
901 }
902
903 static inline void cleanup_signal(struct task_struct *tsk)
904 {
905         struct signal_struct *sig = tsk->signal;
906
907         atomic_dec(&sig->live);
908
909         if (atomic_dec_and_test(&sig->count))
910                 __cleanup_signal(sig);
911 }
912
913 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
914 {
915         unsigned long new_flags = p->flags;
916
917         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
918         new_flags |= PF_FORKNOEXEC;
919         if (!(clone_flags & CLONE_PTRACE))
920                 p->ptrace = 0;
921         p->flags = new_flags;
922 }
923
924 asmlinkage long sys_set_tid_address(int __user *tidptr)
925 {
926         current->clear_child_tid = tidptr;
927
928         return current->pid;
929 }
930
931 static inline void rt_mutex_init_task(struct task_struct *p)
932 {
933 #ifdef CONFIG_RT_MUTEXES
934         spin_lock_init(&p->pi_lock);
935         plist_head_init(&p->pi_waiters, &p->pi_lock);
936         p->pi_blocked_on = NULL;
937 #endif
938 }
939
940 /*
941  * This creates a new process as a copy of the old one,
942  * but does not actually start it yet.
943  *
944  * It copies the registers, and all the appropriate
945  * parts of the process environment (as per the clone
946  * flags). The actual kick-off is left to the caller.
947  */
948 static struct task_struct *copy_process(unsigned long clone_flags,
949                                         unsigned long stack_start,
950                                         struct pt_regs *regs,
951                                         unsigned long stack_size,
952                                         int __user *parent_tidptr,
953                                         int __user *child_tidptr,
954                                         int pid)
955 {
956         int retval;
957         struct task_struct *p = NULL;
958
959         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
960                 return ERR_PTR(-EINVAL);
961
962         /*
963          * Thread groups must share signals as well, and detached threads
964          * can only be started up within the thread group.
965          */
966         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
967                 return ERR_PTR(-EINVAL);
968
969         /*
970          * Shared signal handlers imply shared VM. By way of the above,
971          * thread groups also imply shared VM. Blocking this case allows
972          * for various simplifications in other code.
973          */
974         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
975                 return ERR_PTR(-EINVAL);
976
977         retval = security_task_create(clone_flags);
978         if (retval)
979                 goto fork_out;
980
981         retval = -ENOMEM;
982         p = dup_task_struct(current);
983         if (!p)
984                 goto fork_out;
985
986 #ifdef CONFIG_TRACE_IRQFLAGS
987         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
988         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
989 #endif
990         retval = -EAGAIN;
991         if (atomic_read(&p->user->processes) >=
992                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
993                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
994                                 p->user != &root_user)
995                         goto bad_fork_free;
996         }
997
998         atomic_inc(&p->user->__count);
999         atomic_inc(&p->user->processes);
1000         get_group_info(p->group_info);
1001
1002         /*
1003          * If multiple threads are within copy_process(), then this check
1004          * triggers too late. This doesn't hurt, the check is only there
1005          * to stop root fork bombs.
1006          */
1007         if (nr_threads >= max_threads)
1008                 goto bad_fork_cleanup_count;
1009
1010         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1011                 goto bad_fork_cleanup_count;
1012
1013         if (p->binfmt && !try_module_get(p->binfmt->module))
1014                 goto bad_fork_cleanup_put_domain;
1015
1016         p->did_exec = 0;
1017         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1018         copy_flags(clone_flags, p);
1019         p->pid = pid;
1020         retval = -EFAULT;
1021         if (clone_flags & CLONE_PARENT_SETTID)
1022                 if (put_user(p->pid, parent_tidptr))
1023                         goto bad_fork_cleanup_delays_binfmt;
1024
1025         INIT_LIST_HEAD(&p->children);
1026         INIT_LIST_HEAD(&p->sibling);
1027         p->vfork_done = NULL;
1028         spin_lock_init(&p->alloc_lock);
1029
1030         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1031         init_sigpending(&p->pending);
1032
1033         p->utime = cputime_zero;
1034         p->stime = cputime_zero;
1035         p->sched_time = 0;
1036         p->rchar = 0;           /* I/O counter: bytes read */
1037         p->wchar = 0;           /* I/O counter: bytes written */
1038         p->syscr = 0;           /* I/O counter: read syscalls */
1039         p->syscw = 0;           /* I/O counter: write syscalls */
1040         acct_clear_integrals(p);
1041
1042         p->it_virt_expires = cputime_zero;
1043         p->it_prof_expires = cputime_zero;
1044         p->it_sched_expires = 0;
1045         INIT_LIST_HEAD(&p->cpu_timers[0]);
1046         INIT_LIST_HEAD(&p->cpu_timers[1]);
1047         INIT_LIST_HEAD(&p->cpu_timers[2]);
1048
1049         p->lock_depth = -1;             /* -1 = no lock */
1050         do_posix_clock_monotonic_gettime(&p->start_time);
1051         p->security = NULL;
1052         p->io_context = NULL;
1053         p->io_wait = NULL;
1054         p->audit_context = NULL;
1055         cpuset_fork(p);
1056 #ifdef CONFIG_NUMA
1057         p->mempolicy = mpol_copy(p->mempolicy);
1058         if (IS_ERR(p->mempolicy)) {
1059                 retval = PTR_ERR(p->mempolicy);
1060                 p->mempolicy = NULL;
1061                 goto bad_fork_cleanup_cpuset;
1062         }
1063         mpol_fix_fork_child_flag(p);
1064 #endif
1065 #ifdef CONFIG_TRACE_IRQFLAGS
1066         p->irq_events = 0;
1067 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1068         p->hardirqs_enabled = 1;
1069 #else
1070         p->hardirqs_enabled = 0;
1071 #endif
1072         p->hardirq_enable_ip = 0;
1073         p->hardirq_enable_event = 0;
1074         p->hardirq_disable_ip = _THIS_IP_;
1075         p->hardirq_disable_event = 0;
1076         p->softirqs_enabled = 1;
1077         p->softirq_enable_ip = _THIS_IP_;
1078         p->softirq_enable_event = 0;
1079         p->softirq_disable_ip = 0;
1080         p->softirq_disable_event = 0;
1081         p->hardirq_context = 0;
1082         p->softirq_context = 0;
1083 #endif
1084 #ifdef CONFIG_LOCKDEP
1085         p->lockdep_depth = 0; /* no locks held yet */
1086         p->curr_chain_key = 0;
1087         p->lockdep_recursion = 0;
1088 #endif
1089
1090         rt_mutex_init_task(p);
1091
1092 #ifdef CONFIG_DEBUG_MUTEXES
1093         p->blocked_on = NULL; /* not blocked yet */
1094 #endif
1095
1096         p->tgid = p->pid;
1097         if (clone_flags & CLONE_THREAD)
1098                 p->tgid = current->tgid;
1099
1100         if ((retval = security_task_alloc(p)))
1101                 goto bad_fork_cleanup_policy;
1102         if ((retval = audit_alloc(p)))
1103                 goto bad_fork_cleanup_security;
1104         /* copy all the process information */
1105         if ((retval = copy_semundo(clone_flags, p)))
1106                 goto bad_fork_cleanup_audit;
1107         if ((retval = copy_files(clone_flags, p)))
1108                 goto bad_fork_cleanup_semundo;
1109         if ((retval = copy_fs(clone_flags, p)))
1110                 goto bad_fork_cleanup_files;
1111         if ((retval = copy_sighand(clone_flags, p)))
1112                 goto bad_fork_cleanup_fs;
1113         if ((retval = copy_signal(clone_flags, p)))
1114                 goto bad_fork_cleanup_sighand;
1115         if ((retval = copy_mm(clone_flags, p)))
1116                 goto bad_fork_cleanup_signal;
1117         if ((retval = copy_keys(clone_flags, p)))
1118                 goto bad_fork_cleanup_mm;
1119         if ((retval = copy_namespace(clone_flags, p)))
1120                 goto bad_fork_cleanup_keys;
1121         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1122         if (retval)
1123                 goto bad_fork_cleanup_namespace;
1124
1125         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1126         /*
1127          * Clear TID on mm_release()?
1128          */
1129         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1130         p->robust_list = NULL;
1131 #ifdef CONFIG_COMPAT
1132         p->compat_robust_list = NULL;
1133 #endif
1134         INIT_LIST_HEAD(&p->pi_state_list);
1135         p->pi_state_cache = NULL;
1136
1137         /*
1138          * sigaltstack should be cleared when sharing the same VM
1139          */
1140         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1141                 p->sas_ss_sp = p->sas_ss_size = 0;
1142
1143         /*
1144          * Syscall tracing should be turned off in the child regardless
1145          * of CLONE_PTRACE.
1146          */
1147         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1148 #ifdef TIF_SYSCALL_EMU
1149         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1150 #endif
1151
1152         /* Our parent execution domain becomes current domain
1153            These must match for thread signalling to apply */
1154         p->parent_exec_id = p->self_exec_id;
1155
1156         /* ok, now we should be set up.. */
1157         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1158         p->pdeath_signal = 0;
1159         p->exit_state = 0;
1160
1161         /*
1162          * Ok, make it visible to the rest of the system.
1163          * We dont wake it up yet.
1164          */
1165         p->group_leader = p;
1166         INIT_LIST_HEAD(&p->thread_group);
1167         INIT_LIST_HEAD(&p->ptrace_children);
1168         INIT_LIST_HEAD(&p->ptrace_list);
1169
1170         /* Perform scheduler related setup. Assign this task to a CPU. */
1171         sched_fork(p, clone_flags);
1172
1173         /* Need tasklist lock for parent etc handling! */
1174         write_lock_irq(&tasklist_lock);
1175
1176         /* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
1177         p->ioprio = current->ioprio;
1178
1179         /*
1180          * The task hasn't been attached yet, so its cpus_allowed mask will
1181          * not be changed, nor will its assigned CPU.
1182          *
1183          * The cpus_allowed mask of the parent may have changed after it was
1184          * copied first time - so re-copy it here, then check the child's CPU
1185          * to ensure it is on a valid CPU (and if not, just force it back to
1186          * parent's CPU). This avoids alot of nasty races.
1187          */
1188         p->cpus_allowed = current->cpus_allowed;
1189         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1190                         !cpu_online(task_cpu(p))))
1191                 set_task_cpu(p, smp_processor_id());
1192
1193         /* CLONE_PARENT re-uses the old parent */
1194         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1195                 p->real_parent = current->real_parent;
1196         else
1197                 p->real_parent = current;
1198         p->parent = p->real_parent;
1199
1200         spin_lock(&current->sighand->siglock);
1201
1202         /*
1203          * Process group and session signals need to be delivered to just the
1204          * parent before the fork or both the parent and the child after the
1205          * fork. Restart if a signal comes in before we add the new process to
1206          * it's process group.
1207          * A fatal signal pending means that current will exit, so the new
1208          * thread can't slip out of an OOM kill (or normal SIGKILL).
1209          */
1210         recalc_sigpending();
1211         if (signal_pending(current)) {
1212                 spin_unlock(&current->sighand->siglock);
1213                 write_unlock_irq(&tasklist_lock);
1214                 retval = -ERESTARTNOINTR;
1215                 goto bad_fork_cleanup_namespace;
1216         }
1217
1218         if (clone_flags & CLONE_THREAD) {
1219                 p->group_leader = current->group_leader;
1220                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1221
1222                 if (!cputime_eq(current->signal->it_virt_expires,
1223                                 cputime_zero) ||
1224                     !cputime_eq(current->signal->it_prof_expires,
1225                                 cputime_zero) ||
1226                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1227                     !list_empty(&current->signal->cpu_timers[0]) ||
1228                     !list_empty(&current->signal->cpu_timers[1]) ||
1229                     !list_empty(&current->signal->cpu_timers[2])) {
1230                         /*
1231                          * Have child wake up on its first tick to check
1232                          * for process CPU timers.
1233                          */
1234                         p->it_prof_expires = jiffies_to_cputime(1);
1235                 }
1236         }
1237
1238         if (likely(p->pid)) {
1239                 add_parent(p);
1240                 if (unlikely(p->ptrace & PT_PTRACED))
1241                         __ptrace_link(p, current->parent);
1242
1243                 if (thread_group_leader(p)) {
1244                         p->signal->tty = current->signal->tty;
1245                         p->signal->pgrp = process_group(current);
1246                         p->signal->session = current->signal->session;
1247                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1248                         attach_pid(p, PIDTYPE_SID, p->signal->session);
1249
1250                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1251                         __get_cpu_var(process_counts)++;
1252                 }
1253                 attach_pid(p, PIDTYPE_PID, p->pid);
1254                 nr_threads++;
1255         }
1256
1257         total_forks++;
1258         spin_unlock(&current->sighand->siglock);
1259         write_unlock_irq(&tasklist_lock);
1260         proc_fork_connector(p);
1261         return p;
1262
1263 bad_fork_cleanup_namespace:
1264         exit_namespace(p);
1265 bad_fork_cleanup_keys:
1266         exit_keys(p);
1267 bad_fork_cleanup_mm:
1268         if (p->mm)
1269                 mmput(p->mm);
1270 bad_fork_cleanup_signal:
1271         cleanup_signal(p);
1272 bad_fork_cleanup_sighand:
1273         __cleanup_sighand(p->sighand);
1274 bad_fork_cleanup_fs:
1275         exit_fs(p); /* blocking */
1276 bad_fork_cleanup_files:
1277         exit_files(p); /* blocking */
1278 bad_fork_cleanup_semundo:
1279         exit_sem(p);
1280 bad_fork_cleanup_audit:
1281         audit_free(p);
1282 bad_fork_cleanup_security:
1283         security_task_free(p);
1284 bad_fork_cleanup_policy:
1285 #ifdef CONFIG_NUMA
1286         mpol_free(p->mempolicy);
1287 bad_fork_cleanup_cpuset:
1288 #endif
1289         cpuset_exit(p);
1290 bad_fork_cleanup_delays_binfmt:
1291         delayacct_tsk_free(p);
1292         if (p->binfmt)
1293                 module_put(p->binfmt->module);
1294 bad_fork_cleanup_put_domain:
1295         module_put(task_thread_info(p)->exec_domain->module);
1296 bad_fork_cleanup_count:
1297         put_group_info(p->group_info);
1298         atomic_dec(&p->user->processes);
1299         free_uid(p->user);
1300 bad_fork_free:
1301         free_task(p);
1302 fork_out:
1303         return ERR_PTR(retval);
1304 }
1305
1306 struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1307 {
1308         memset(regs, 0, sizeof(struct pt_regs));
1309         return regs;
1310 }
1311
1312 struct task_struct * __devinit fork_idle(int cpu)
1313 {
1314         struct task_struct *task;
1315         struct pt_regs regs;
1316
1317         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1318         if (!task)
1319                 return ERR_PTR(-ENOMEM);
1320         init_idle(task, cpu);
1321
1322         return task;
1323 }
1324
1325 static inline int fork_traceflag (unsigned clone_flags)
1326 {
1327         if (clone_flags & CLONE_UNTRACED)
1328                 return 0;
1329         else if (clone_flags & CLONE_VFORK) {
1330                 if (current->ptrace & PT_TRACE_VFORK)
1331                         return PTRACE_EVENT_VFORK;
1332         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1333                 if (current->ptrace & PT_TRACE_CLONE)
1334                         return PTRACE_EVENT_CLONE;
1335         } else if (current->ptrace & PT_TRACE_FORK)
1336                 return PTRACE_EVENT_FORK;
1337
1338         return 0;
1339 }
1340
1341 /*
1342  *  Ok, this is the main fork-routine.
1343  *
1344  * It copies the process, and if successful kick-starts
1345  * it and waits for it to finish using the VM if required.
1346  */
1347 long do_fork(unsigned long clone_flags,
1348               unsigned long stack_start,
1349               struct pt_regs *regs,
1350               unsigned long stack_size,
1351               int __user *parent_tidptr,
1352               int __user *child_tidptr)
1353 {
1354         struct task_struct *p;
1355         int trace = 0;
1356         struct pid *pid = alloc_pid();
1357         long nr;
1358
1359         if (!pid)
1360                 return -EAGAIN;
1361         nr = pid->nr;
1362         if (unlikely(current->ptrace)) {
1363                 trace = fork_traceflag (clone_flags);
1364                 if (trace)
1365                         clone_flags |= CLONE_PTRACE;
1366         }
1367
1368         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1369         /*
1370          * Do this prior waking up the new thread - the thread pointer
1371          * might get invalid after that point, if the thread exits quickly.
1372          */
1373         if (!IS_ERR(p)) {
1374                 struct completion vfork;
1375
1376                 if (clone_flags & CLONE_VFORK) {
1377                         p->vfork_done = &vfork;
1378                         init_completion(&vfork);
1379                 }
1380
1381                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1382                         /*
1383                          * We'll start up with an immediate SIGSTOP.
1384                          */
1385                         sigaddset(&p->pending.signal, SIGSTOP);
1386                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1387                 }
1388
1389                 if (!(clone_flags & CLONE_STOPPED))
1390                         wake_up_new_task(p, clone_flags);
1391                 else
1392                         p->state = TASK_STOPPED;
1393
1394                 if (unlikely (trace)) {
1395                         current->ptrace_message = nr;
1396                         ptrace_notify ((trace << 8) | SIGTRAP);
1397                 }
1398
1399                 if (clone_flags & CLONE_VFORK) {
1400                         wait_for_completion(&vfork);
1401                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1402                                 current->ptrace_message = nr;
1403                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1404                         }
1405                 }
1406         } else {
1407                 free_pid(pid);
1408                 nr = PTR_ERR(p);
1409         }
1410         return nr;
1411 }
1412
1413 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1414 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1415 #endif
1416
1417 static void sighand_ctor(void *data, kmem_cache_t *cachep, unsigned long flags)
1418 {
1419         struct sighand_struct *sighand = data;
1420
1421         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1422                                         SLAB_CTOR_CONSTRUCTOR)
1423                 spin_lock_init(&sighand->siglock);
1424 }
1425
1426 void __init proc_caches_init(void)
1427 {
1428         sighand_cachep = kmem_cache_create("sighand_cache",
1429                         sizeof(struct sighand_struct), 0,
1430                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1431                         sighand_ctor, NULL);
1432         signal_cachep = kmem_cache_create("signal_cache",
1433                         sizeof(struct signal_struct), 0,
1434                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1435         files_cachep = kmem_cache_create("files_cache", 
1436                         sizeof(struct files_struct), 0,
1437                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1438         fs_cachep = kmem_cache_create("fs_cache", 
1439                         sizeof(struct fs_struct), 0,
1440                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1441         vm_area_cachep = kmem_cache_create("vm_area_struct",
1442                         sizeof(struct vm_area_struct), 0,
1443                         SLAB_PANIC, NULL, NULL);
1444         mm_cachep = kmem_cache_create("mm_struct",
1445                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1446                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1447 }
1448
1449
1450 /*
1451  * Check constraints on flags passed to the unshare system call and
1452  * force unsharing of additional process context as appropriate.
1453  */
1454 static inline void check_unshare_flags(unsigned long *flags_ptr)
1455 {
1456         /*
1457          * If unsharing a thread from a thread group, must also
1458          * unshare vm.
1459          */
1460         if (*flags_ptr & CLONE_THREAD)
1461                 *flags_ptr |= CLONE_VM;
1462
1463         /*
1464          * If unsharing vm, must also unshare signal handlers.
1465          */
1466         if (*flags_ptr & CLONE_VM)
1467                 *flags_ptr |= CLONE_SIGHAND;
1468
1469         /*
1470          * If unsharing signal handlers and the task was created
1471          * using CLONE_THREAD, then must unshare the thread
1472          */
1473         if ((*flags_ptr & CLONE_SIGHAND) &&
1474             (atomic_read(&current->signal->count) > 1))
1475                 *flags_ptr |= CLONE_THREAD;
1476
1477         /*
1478          * If unsharing namespace, must also unshare filesystem information.
1479          */
1480         if (*flags_ptr & CLONE_NEWNS)
1481                 *flags_ptr |= CLONE_FS;
1482 }
1483
1484 /*
1485  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1486  */
1487 static int unshare_thread(unsigned long unshare_flags)
1488 {
1489         if (unshare_flags & CLONE_THREAD)
1490                 return -EINVAL;
1491
1492         return 0;
1493 }
1494
1495 /*
1496  * Unshare the filesystem structure if it is being shared
1497  */
1498 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1499 {
1500         struct fs_struct *fs = current->fs;
1501
1502         if ((unshare_flags & CLONE_FS) &&
1503             (fs && atomic_read(&fs->count) > 1)) {
1504                 *new_fsp = __copy_fs_struct(current->fs);
1505                 if (!*new_fsp)
1506                         return -ENOMEM;
1507         }
1508
1509         return 0;
1510 }
1511
1512 /*
1513  * Unshare the namespace structure if it is being shared
1514  */
1515 static int unshare_namespace(unsigned long unshare_flags, struct namespace **new_nsp, struct fs_struct *new_fs)
1516 {
1517         struct namespace *ns = current->namespace;
1518
1519         if ((unshare_flags & CLONE_NEWNS) &&
1520             (ns && atomic_read(&ns->count) > 1)) {
1521                 if (!capable(CAP_SYS_ADMIN))
1522                         return -EPERM;
1523
1524                 *new_nsp = dup_namespace(current, new_fs ? new_fs : current->fs);
1525                 if (!*new_nsp)
1526                         return -ENOMEM;
1527         }
1528
1529         return 0;
1530 }
1531
1532 /*
1533  * Unsharing of sighand for tasks created with CLONE_SIGHAND is not
1534  * supported yet
1535  */
1536 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1537 {
1538         struct sighand_struct *sigh = current->sighand;
1539
1540         if ((unshare_flags & CLONE_SIGHAND) &&
1541             (sigh && atomic_read(&sigh->count) > 1))
1542                 return -EINVAL;
1543         else
1544                 return 0;
1545 }
1546
1547 /*
1548  * Unshare vm if it is being shared
1549  */
1550 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1551 {
1552         struct mm_struct *mm = current->mm;
1553
1554         if ((unshare_flags & CLONE_VM) &&
1555             (mm && atomic_read(&mm->mm_users) > 1)) {
1556                 return -EINVAL;
1557         }
1558
1559         return 0;
1560 }
1561
1562 /*
1563  * Unshare file descriptor table if it is being shared
1564  */
1565 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1566 {
1567         struct files_struct *fd = current->files;
1568         int error = 0;
1569
1570         if ((unshare_flags & CLONE_FILES) &&
1571             (fd && atomic_read(&fd->count) > 1)) {
1572                 *new_fdp = dup_fd(fd, &error);
1573                 if (!*new_fdp)
1574                         return error;
1575         }
1576
1577         return 0;
1578 }
1579
1580 /*
1581  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1582  * supported yet
1583  */
1584 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1585 {
1586         if (unshare_flags & CLONE_SYSVSEM)
1587                 return -EINVAL;
1588
1589         return 0;
1590 }
1591
1592 /*
1593  * unshare allows a process to 'unshare' part of the process
1594  * context which was originally shared using clone.  copy_*
1595  * functions used by do_fork() cannot be used here directly
1596  * because they modify an inactive task_struct that is being
1597  * constructed. Here we are modifying the current, active,
1598  * task_struct.
1599  */
1600 asmlinkage long sys_unshare(unsigned long unshare_flags)
1601 {
1602         int err = 0;
1603         struct fs_struct *fs, *new_fs = NULL;
1604         struct namespace *ns, *new_ns = NULL;
1605         struct sighand_struct *sigh, *new_sigh = NULL;
1606         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1607         struct files_struct *fd, *new_fd = NULL;
1608         struct sem_undo_list *new_ulist = NULL;
1609
1610         check_unshare_flags(&unshare_flags);
1611
1612         /* Return -EINVAL for all unsupported flags */
1613         err = -EINVAL;
1614         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1615                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM))
1616                 goto bad_unshare_out;
1617
1618         if ((err = unshare_thread(unshare_flags)))
1619                 goto bad_unshare_out;
1620         if ((err = unshare_fs(unshare_flags, &new_fs)))
1621                 goto bad_unshare_cleanup_thread;
1622         if ((err = unshare_namespace(unshare_flags, &new_ns, new_fs)))
1623                 goto bad_unshare_cleanup_fs;
1624         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1625                 goto bad_unshare_cleanup_ns;
1626         if ((err = unshare_vm(unshare_flags, &new_mm)))
1627                 goto bad_unshare_cleanup_sigh;
1628         if ((err = unshare_fd(unshare_flags, &new_fd)))
1629                 goto bad_unshare_cleanup_vm;
1630         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1631                 goto bad_unshare_cleanup_fd;
1632
1633         if (new_fs || new_ns || new_sigh || new_mm || new_fd || new_ulist) {
1634
1635                 task_lock(current);
1636
1637                 if (new_fs) {
1638                         fs = current->fs;
1639                         current->fs = new_fs;
1640                         new_fs = fs;
1641                 }
1642
1643                 if (new_ns) {
1644                         ns = current->namespace;
1645                         current->namespace = new_ns;
1646                         new_ns = ns;
1647                 }
1648
1649                 if (new_sigh) {
1650                         sigh = current->sighand;
1651                         rcu_assign_pointer(current->sighand, new_sigh);
1652                         new_sigh = sigh;
1653                 }
1654
1655                 if (new_mm) {
1656                         mm = current->mm;
1657                         active_mm = current->active_mm;
1658                         current->mm = new_mm;
1659                         current->active_mm = new_mm;
1660                         activate_mm(active_mm, new_mm);
1661                         new_mm = mm;
1662                 }
1663
1664                 if (new_fd) {
1665                         fd = current->files;
1666                         current->files = new_fd;
1667                         new_fd = fd;
1668                 }
1669
1670                 task_unlock(current);
1671         }
1672
1673 bad_unshare_cleanup_fd:
1674         if (new_fd)
1675                 put_files_struct(new_fd);
1676
1677 bad_unshare_cleanup_vm:
1678         if (new_mm)
1679                 mmput(new_mm);
1680
1681 bad_unshare_cleanup_sigh:
1682         if (new_sigh)
1683                 if (atomic_dec_and_test(&new_sigh->count))
1684                         kmem_cache_free(sighand_cachep, new_sigh);
1685
1686 bad_unshare_cleanup_ns:
1687         if (new_ns)
1688                 put_namespace(new_ns);
1689
1690 bad_unshare_cleanup_fs:
1691         if (new_fs)
1692                 put_fs_struct(new_fs);
1693
1694 bad_unshare_cleanup_thread:
1695 bad_unshare_out:
1696         return err;
1697 }