[PATCH] io-accounting: core statistics
[linux-3.10.git] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/smp_lock.h>
18 #include <linux/module.h>
19 #include <linux/vmalloc.h>
20 #include <linux/completion.h>
21 #include <linux/mnt_namespace.h>
22 #include <linux/personality.h>
23 #include <linux/mempolicy.h>
24 #include <linux/sem.h>
25 #include <linux/file.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/fs.h>
30 #include <linux/nsproxy.h>
31 #include <linux/capability.h>
32 #include <linux/cpu.h>
33 #include <linux/cpuset.h>
34 #include <linux/security.h>
35 #include <linux/swap.h>
36 #include <linux/syscalls.h>
37 #include <linux/jiffies.h>
38 #include <linux/futex.h>
39 #include <linux/task_io_accounting_ops.h>
40 #include <linux/rcupdate.h>
41 #include <linux/ptrace.h>
42 #include <linux/mount.h>
43 #include <linux/audit.h>
44 #include <linux/profile.h>
45 #include <linux/rmap.h>
46 #include <linux/acct.h>
47 #include <linux/tsacct_kern.h>
48 #include <linux/cn_proc.h>
49 #include <linux/delayacct.h>
50 #include <linux/taskstats_kern.h>
51 #include <linux/random.h>
52
53 #include <asm/pgtable.h>
54 #include <asm/pgalloc.h>
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
57 #include <asm/cacheflush.h>
58 #include <asm/tlbflush.h>
59
60 /*
61  * Protected counters by write_lock_irq(&tasklist_lock)
62  */
63 unsigned long total_forks;      /* Handle normal Linux uptimes. */
64 int nr_threads;                 /* The idle threads do not count.. */
65
66 int max_threads;                /* tunable limit on nr_threads */
67
68 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
69
70 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
71
72 int nr_processes(void)
73 {
74         int cpu;
75         int total = 0;
76
77         for_each_online_cpu(cpu)
78                 total += per_cpu(process_counts, cpu);
79
80         return total;
81 }
82
83 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
84 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
85 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
86 static struct kmem_cache *task_struct_cachep;
87 #endif
88
89 /* SLAB cache for signal_struct structures (tsk->signal) */
90 static struct kmem_cache *signal_cachep;
91
92 /* SLAB cache for sighand_struct structures (tsk->sighand) */
93 struct kmem_cache *sighand_cachep;
94
95 /* SLAB cache for files_struct structures (tsk->files) */
96 struct kmem_cache *files_cachep;
97
98 /* SLAB cache for fs_struct structures (tsk->fs) */
99 struct kmem_cache *fs_cachep;
100
101 /* SLAB cache for vm_area_struct structures */
102 struct kmem_cache *vm_area_cachep;
103
104 /* SLAB cache for mm_struct structures (tsk->mm) */
105 static struct kmem_cache *mm_cachep;
106
107 void free_task(struct task_struct *tsk)
108 {
109         free_thread_info(tsk->thread_info);
110         rt_mutex_debug_task_free(tsk);
111         free_task_struct(tsk);
112 }
113 EXPORT_SYMBOL(free_task);
114
115 void __put_task_struct(struct task_struct *tsk)
116 {
117         WARN_ON(!(tsk->exit_state & (EXIT_DEAD | EXIT_ZOMBIE)));
118         WARN_ON(atomic_read(&tsk->usage));
119         WARN_ON(tsk == current);
120
121         security_task_free(tsk);
122         free_uid(tsk->user);
123         put_group_info(tsk->group_info);
124         delayacct_tsk_free(tsk);
125
126         if (!profile_handoff_task(tsk))
127                 free_task(tsk);
128 }
129
130 void __init fork_init(unsigned long mempages)
131 {
132 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
133 #ifndef ARCH_MIN_TASKALIGN
134 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
135 #endif
136         /* create a slab on which task_structs can be allocated */
137         task_struct_cachep =
138                 kmem_cache_create("task_struct", sizeof(struct task_struct),
139                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL, NULL);
140 #endif
141
142         /*
143          * The default maximum number of threads is set to a safe
144          * value: the thread structures can take up at most half
145          * of memory.
146          */
147         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
148
149         /*
150          * we need to allow at least 20 threads to boot a system
151          */
152         if(max_threads < 20)
153                 max_threads = 20;
154
155         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
156         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
157         init_task.signal->rlim[RLIMIT_SIGPENDING] =
158                 init_task.signal->rlim[RLIMIT_NPROC];
159 }
160
161 static struct task_struct *dup_task_struct(struct task_struct *orig)
162 {
163         struct task_struct *tsk;
164         struct thread_info *ti;
165
166         prepare_to_copy(orig);
167
168         tsk = alloc_task_struct();
169         if (!tsk)
170                 return NULL;
171
172         ti = alloc_thread_info(tsk);
173         if (!ti) {
174                 free_task_struct(tsk);
175                 return NULL;
176         }
177
178         *tsk = *orig;
179         tsk->thread_info = ti;
180         setup_thread_stack(tsk, orig);
181
182 #ifdef CONFIG_CC_STACKPROTECTOR
183         tsk->stack_canary = get_random_int();
184 #endif
185
186         /* One for us, one for whoever does the "release_task()" (usually parent) */
187         atomic_set(&tsk->usage,2);
188         atomic_set(&tsk->fs_excl, 0);
189 #ifdef CONFIG_BLK_DEV_IO_TRACE
190         tsk->btrace_seq = 0;
191 #endif
192         tsk->splice_pipe = NULL;
193         return tsk;
194 }
195
196 #ifdef CONFIG_MMU
197 static inline int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
198 {
199         struct vm_area_struct *mpnt, *tmp, **pprev;
200         struct rb_node **rb_link, *rb_parent;
201         int retval;
202         unsigned long charge;
203         struct mempolicy *pol;
204
205         down_write(&oldmm->mmap_sem);
206         flush_cache_mm(oldmm);
207         /*
208          * Not linked in yet - no deadlock potential:
209          */
210         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
211
212         mm->locked_vm = 0;
213         mm->mmap = NULL;
214         mm->mmap_cache = NULL;
215         mm->free_area_cache = oldmm->mmap_base;
216         mm->cached_hole_size = ~0UL;
217         mm->map_count = 0;
218         cpus_clear(mm->cpu_vm_mask);
219         mm->mm_rb = RB_ROOT;
220         rb_link = &mm->mm_rb.rb_node;
221         rb_parent = NULL;
222         pprev = &mm->mmap;
223
224         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
225                 struct file *file;
226
227                 if (mpnt->vm_flags & VM_DONTCOPY) {
228                         long pages = vma_pages(mpnt);
229                         mm->total_vm -= pages;
230                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
231                                                                 -pages);
232                         continue;
233                 }
234                 charge = 0;
235                 if (mpnt->vm_flags & VM_ACCOUNT) {
236                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
237                         if (security_vm_enough_memory(len))
238                                 goto fail_nomem;
239                         charge = len;
240                 }
241                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
242                 if (!tmp)
243                         goto fail_nomem;
244                 *tmp = *mpnt;
245                 pol = mpol_copy(vma_policy(mpnt));
246                 retval = PTR_ERR(pol);
247                 if (IS_ERR(pol))
248                         goto fail_nomem_policy;
249                 vma_set_policy(tmp, pol);
250                 tmp->vm_flags &= ~VM_LOCKED;
251                 tmp->vm_mm = mm;
252                 tmp->vm_next = NULL;
253                 anon_vma_link(tmp);
254                 file = tmp->vm_file;
255                 if (file) {
256                         struct inode *inode = file->f_path.dentry->d_inode;
257                         get_file(file);
258                         if (tmp->vm_flags & VM_DENYWRITE)
259                                 atomic_dec(&inode->i_writecount);
260       
261                         /* insert tmp into the share list, just after mpnt */
262                         spin_lock(&file->f_mapping->i_mmap_lock);
263                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
264                         flush_dcache_mmap_lock(file->f_mapping);
265                         vma_prio_tree_add(tmp, mpnt);
266                         flush_dcache_mmap_unlock(file->f_mapping);
267                         spin_unlock(&file->f_mapping->i_mmap_lock);
268                 }
269
270                 /*
271                  * Link in the new vma and copy the page table entries.
272                  */
273                 *pprev = tmp;
274                 pprev = &tmp->vm_next;
275
276                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
277                 rb_link = &tmp->vm_rb.rb_right;
278                 rb_parent = &tmp->vm_rb;
279
280                 mm->map_count++;
281                 retval = copy_page_range(mm, oldmm, mpnt);
282
283                 if (tmp->vm_ops && tmp->vm_ops->open)
284                         tmp->vm_ops->open(tmp);
285
286                 if (retval)
287                         goto out;
288         }
289         retval = 0;
290 out:
291         up_write(&mm->mmap_sem);
292         flush_tlb_mm(oldmm);
293         up_write(&oldmm->mmap_sem);
294         return retval;
295 fail_nomem_policy:
296         kmem_cache_free(vm_area_cachep, tmp);
297 fail_nomem:
298         retval = -ENOMEM;
299         vm_unacct_memory(charge);
300         goto out;
301 }
302
303 static inline int mm_alloc_pgd(struct mm_struct * mm)
304 {
305         mm->pgd = pgd_alloc(mm);
306         if (unlikely(!mm->pgd))
307                 return -ENOMEM;
308         return 0;
309 }
310
311 static inline void mm_free_pgd(struct mm_struct * mm)
312 {
313         pgd_free(mm->pgd);
314 }
315 #else
316 #define dup_mmap(mm, oldmm)     (0)
317 #define mm_alloc_pgd(mm)        (0)
318 #define mm_free_pgd(mm)
319 #endif /* CONFIG_MMU */
320
321  __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
322
323 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
324 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
325
326 #include <linux/init_task.h>
327
328 static struct mm_struct * mm_init(struct mm_struct * mm)
329 {
330         atomic_set(&mm->mm_users, 1);
331         atomic_set(&mm->mm_count, 1);
332         init_rwsem(&mm->mmap_sem);
333         INIT_LIST_HEAD(&mm->mmlist);
334         mm->core_waiters = 0;
335         mm->nr_ptes = 0;
336         set_mm_counter(mm, file_rss, 0);
337         set_mm_counter(mm, anon_rss, 0);
338         spin_lock_init(&mm->page_table_lock);
339         rwlock_init(&mm->ioctx_list_lock);
340         mm->ioctx_list = NULL;
341         mm->free_area_cache = TASK_UNMAPPED_BASE;
342         mm->cached_hole_size = ~0UL;
343
344         if (likely(!mm_alloc_pgd(mm))) {
345                 mm->def_flags = 0;
346                 return mm;
347         }
348         free_mm(mm);
349         return NULL;
350 }
351
352 /*
353  * Allocate and initialize an mm_struct.
354  */
355 struct mm_struct * mm_alloc(void)
356 {
357         struct mm_struct * mm;
358
359         mm = allocate_mm();
360         if (mm) {
361                 memset(mm, 0, sizeof(*mm));
362                 mm = mm_init(mm);
363         }
364         return mm;
365 }
366
367 /*
368  * Called when the last reference to the mm
369  * is dropped: either by a lazy thread or by
370  * mmput. Free the page directory and the mm.
371  */
372 void fastcall __mmdrop(struct mm_struct *mm)
373 {
374         BUG_ON(mm == &init_mm);
375         mm_free_pgd(mm);
376         destroy_context(mm);
377         free_mm(mm);
378 }
379
380 /*
381  * Decrement the use count and release all resources for an mm.
382  */
383 void mmput(struct mm_struct *mm)
384 {
385         might_sleep();
386
387         if (atomic_dec_and_test(&mm->mm_users)) {
388                 exit_aio(mm);
389                 exit_mmap(mm);
390                 if (!list_empty(&mm->mmlist)) {
391                         spin_lock(&mmlist_lock);
392                         list_del(&mm->mmlist);
393                         spin_unlock(&mmlist_lock);
394                 }
395                 put_swap_token(mm);
396                 mmdrop(mm);
397         }
398 }
399 EXPORT_SYMBOL_GPL(mmput);
400
401 /**
402  * get_task_mm - acquire a reference to the task's mm
403  *
404  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
405  * this kernel workthread has transiently adopted a user mm with use_mm,
406  * to do its AIO) is not set and if so returns a reference to it, after
407  * bumping up the use count.  User must release the mm via mmput()
408  * after use.  Typically used by /proc and ptrace.
409  */
410 struct mm_struct *get_task_mm(struct task_struct *task)
411 {
412         struct mm_struct *mm;
413
414         task_lock(task);
415         mm = task->mm;
416         if (mm) {
417                 if (task->flags & PF_BORROWED_MM)
418                         mm = NULL;
419                 else
420                         atomic_inc(&mm->mm_users);
421         }
422         task_unlock(task);
423         return mm;
424 }
425 EXPORT_SYMBOL_GPL(get_task_mm);
426
427 /* Please note the differences between mmput and mm_release.
428  * mmput is called whenever we stop holding onto a mm_struct,
429  * error success whatever.
430  *
431  * mm_release is called after a mm_struct has been removed
432  * from the current process.
433  *
434  * This difference is important for error handling, when we
435  * only half set up a mm_struct for a new process and need to restore
436  * the old one.  Because we mmput the new mm_struct before
437  * restoring the old one. . .
438  * Eric Biederman 10 January 1998
439  */
440 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
441 {
442         struct completion *vfork_done = tsk->vfork_done;
443
444         /* Get rid of any cached register state */
445         deactivate_mm(tsk, mm);
446
447         /* notify parent sleeping on vfork() */
448         if (vfork_done) {
449                 tsk->vfork_done = NULL;
450                 complete(vfork_done);
451         }
452
453         /*
454          * If we're exiting normally, clear a user-space tid field if
455          * requested.  We leave this alone when dying by signal, to leave
456          * the value intact in a core dump, and to save the unnecessary
457          * trouble otherwise.  Userland only wants this done for a sys_exit.
458          */
459         if (tsk->clear_child_tid
460             && !(tsk->flags & PF_SIGNALED)
461             && atomic_read(&mm->mm_users) > 1) {
462                 u32 __user * tidptr = tsk->clear_child_tid;
463                 tsk->clear_child_tid = NULL;
464
465                 /*
466                  * We don't check the error code - if userspace has
467                  * not set up a proper pointer then tough luck.
468                  */
469                 put_user(0, tidptr);
470                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
471         }
472 }
473
474 /*
475  * Allocate a new mm structure and copy contents from the
476  * mm structure of the passed in task structure.
477  */
478 static struct mm_struct *dup_mm(struct task_struct *tsk)
479 {
480         struct mm_struct *mm, *oldmm = current->mm;
481         int err;
482
483         if (!oldmm)
484                 return NULL;
485
486         mm = allocate_mm();
487         if (!mm)
488                 goto fail_nomem;
489
490         memcpy(mm, oldmm, sizeof(*mm));
491
492         /* Initializing for Swap token stuff */
493         mm->token_priority = 0;
494         mm->last_interval = 0;
495
496         if (!mm_init(mm))
497                 goto fail_nomem;
498
499         if (init_new_context(tsk, mm))
500                 goto fail_nocontext;
501
502         err = dup_mmap(mm, oldmm);
503         if (err)
504                 goto free_pt;
505
506         mm->hiwater_rss = get_mm_rss(mm);
507         mm->hiwater_vm = mm->total_vm;
508
509         return mm;
510
511 free_pt:
512         mmput(mm);
513
514 fail_nomem:
515         return NULL;
516
517 fail_nocontext:
518         /*
519          * If init_new_context() failed, we cannot use mmput() to free the mm
520          * because it calls destroy_context()
521          */
522         mm_free_pgd(mm);
523         free_mm(mm);
524         return NULL;
525 }
526
527 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
528 {
529         struct mm_struct * mm, *oldmm;
530         int retval;
531
532         tsk->min_flt = tsk->maj_flt = 0;
533         tsk->nvcsw = tsk->nivcsw = 0;
534
535         tsk->mm = NULL;
536         tsk->active_mm = NULL;
537
538         /*
539          * Are we cloning a kernel thread?
540          *
541          * We need to steal a active VM for that..
542          */
543         oldmm = current->mm;
544         if (!oldmm)
545                 return 0;
546
547         if (clone_flags & CLONE_VM) {
548                 atomic_inc(&oldmm->mm_users);
549                 mm = oldmm;
550                 goto good_mm;
551         }
552
553         retval = -ENOMEM;
554         mm = dup_mm(tsk);
555         if (!mm)
556                 goto fail_nomem;
557
558 good_mm:
559         /* Initializing for Swap token stuff */
560         mm->token_priority = 0;
561         mm->last_interval = 0;
562
563         tsk->mm = mm;
564         tsk->active_mm = mm;
565         return 0;
566
567 fail_nomem:
568         return retval;
569 }
570
571 static inline struct fs_struct *__copy_fs_struct(struct fs_struct *old)
572 {
573         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
574         /* We don't need to lock fs - think why ;-) */
575         if (fs) {
576                 atomic_set(&fs->count, 1);
577                 rwlock_init(&fs->lock);
578                 fs->umask = old->umask;
579                 read_lock(&old->lock);
580                 fs->rootmnt = mntget(old->rootmnt);
581                 fs->root = dget(old->root);
582                 fs->pwdmnt = mntget(old->pwdmnt);
583                 fs->pwd = dget(old->pwd);
584                 if (old->altroot) {
585                         fs->altrootmnt = mntget(old->altrootmnt);
586                         fs->altroot = dget(old->altroot);
587                 } else {
588                         fs->altrootmnt = NULL;
589                         fs->altroot = NULL;
590                 }
591                 read_unlock(&old->lock);
592         }
593         return fs;
594 }
595
596 struct fs_struct *copy_fs_struct(struct fs_struct *old)
597 {
598         return __copy_fs_struct(old);
599 }
600
601 EXPORT_SYMBOL_GPL(copy_fs_struct);
602
603 static inline int copy_fs(unsigned long clone_flags, struct task_struct * tsk)
604 {
605         if (clone_flags & CLONE_FS) {
606                 atomic_inc(&current->fs->count);
607                 return 0;
608         }
609         tsk->fs = __copy_fs_struct(current->fs);
610         if (!tsk->fs)
611                 return -ENOMEM;
612         return 0;
613 }
614
615 static int count_open_files(struct fdtable *fdt)
616 {
617         int size = fdt->max_fdset;
618         int i;
619
620         /* Find the last open fd */
621         for (i = size/(8*sizeof(long)); i > 0; ) {
622                 if (fdt->open_fds->fds_bits[--i])
623                         break;
624         }
625         i = (i+1) * 8 * sizeof(long);
626         return i;
627 }
628
629 static struct files_struct *alloc_files(void)
630 {
631         struct files_struct *newf;
632         struct fdtable *fdt;
633
634         newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
635         if (!newf)
636                 goto out;
637
638         atomic_set(&newf->count, 1);
639
640         spin_lock_init(&newf->file_lock);
641         newf->next_fd = 0;
642         fdt = &newf->fdtab;
643         fdt->max_fds = NR_OPEN_DEFAULT;
644         fdt->max_fdset = EMBEDDED_FD_SET_SIZE;
645         fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
646         fdt->open_fds = (fd_set *)&newf->open_fds_init;
647         fdt->fd = &newf->fd_array[0];
648         INIT_RCU_HEAD(&fdt->rcu);
649         fdt->free_files = NULL;
650         fdt->next = NULL;
651         rcu_assign_pointer(newf->fdt, fdt);
652 out:
653         return newf;
654 }
655
656 /*
657  * Allocate a new files structure and copy contents from the
658  * passed in files structure.
659  * errorp will be valid only when the returned files_struct is NULL.
660  */
661 static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
662 {
663         struct files_struct *newf;
664         struct file **old_fds, **new_fds;
665         int open_files, size, i, expand;
666         struct fdtable *old_fdt, *new_fdt;
667
668         *errorp = -ENOMEM;
669         newf = alloc_files();
670         if (!newf)
671                 goto out;
672
673         spin_lock(&oldf->file_lock);
674         old_fdt = files_fdtable(oldf);
675         new_fdt = files_fdtable(newf);
676         size = old_fdt->max_fdset;
677         open_files = count_open_files(old_fdt);
678         expand = 0;
679
680         /*
681          * Check whether we need to allocate a larger fd array or fd set.
682          * Note: we're not a clone task, so the open count won't  change.
683          */
684         if (open_files > new_fdt->max_fdset) {
685                 new_fdt->max_fdset = 0;
686                 expand = 1;
687         }
688         if (open_files > new_fdt->max_fds) {
689                 new_fdt->max_fds = 0;
690                 expand = 1;
691         }
692
693         /* if the old fdset gets grown now, we'll only copy up to "size" fds */
694         if (expand) {
695                 spin_unlock(&oldf->file_lock);
696                 spin_lock(&newf->file_lock);
697                 *errorp = expand_files(newf, open_files-1);
698                 spin_unlock(&newf->file_lock);
699                 if (*errorp < 0)
700                         goto out_release;
701                 new_fdt = files_fdtable(newf);
702                 /*
703                  * Reacquire the oldf lock and a pointer to its fd table
704                  * who knows it may have a new bigger fd table. We need
705                  * the latest pointer.
706                  */
707                 spin_lock(&oldf->file_lock);
708                 old_fdt = files_fdtable(oldf);
709         }
710
711         old_fds = old_fdt->fd;
712         new_fds = new_fdt->fd;
713
714         memcpy(new_fdt->open_fds->fds_bits, old_fdt->open_fds->fds_bits, open_files/8);
715         memcpy(new_fdt->close_on_exec->fds_bits, old_fdt->close_on_exec->fds_bits, open_files/8);
716
717         for (i = open_files; i != 0; i--) {
718                 struct file *f = *old_fds++;
719                 if (f) {
720                         get_file(f);
721                 } else {
722                         /*
723                          * The fd may be claimed in the fd bitmap but not yet
724                          * instantiated in the files array if a sibling thread
725                          * is partway through open().  So make sure that this
726                          * fd is available to the new process.
727                          */
728                         FD_CLR(open_files - i, new_fdt->open_fds);
729                 }
730                 rcu_assign_pointer(*new_fds++, f);
731         }
732         spin_unlock(&oldf->file_lock);
733
734         /* compute the remainder to be cleared */
735         size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
736
737         /* This is long word aligned thus could use a optimized version */ 
738         memset(new_fds, 0, size); 
739
740         if (new_fdt->max_fdset > open_files) {
741                 int left = (new_fdt->max_fdset-open_files)/8;
742                 int start = open_files / (8 * sizeof(unsigned long));
743
744                 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
745                 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
746         }
747
748 out:
749         return newf;
750
751 out_release:
752         free_fdset (new_fdt->close_on_exec, new_fdt->max_fdset);
753         free_fdset (new_fdt->open_fds, new_fdt->max_fdset);
754         free_fd_array(new_fdt->fd, new_fdt->max_fds);
755         kmem_cache_free(files_cachep, newf);
756         return NULL;
757 }
758
759 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
760 {
761         struct files_struct *oldf, *newf;
762         int error = 0;
763
764         /*
765          * A background process may not have any files ...
766          */
767         oldf = current->files;
768         if (!oldf)
769                 goto out;
770
771         if (clone_flags & CLONE_FILES) {
772                 atomic_inc(&oldf->count);
773                 goto out;
774         }
775
776         /*
777          * Note: we may be using current for both targets (See exec.c)
778          * This works because we cache current->files (old) as oldf. Don't
779          * break this.
780          */
781         tsk->files = NULL;
782         newf = dup_fd(oldf, &error);
783         if (!newf)
784                 goto out;
785
786         tsk->files = newf;
787         error = 0;
788 out:
789         return error;
790 }
791
792 /*
793  *      Helper to unshare the files of the current task.
794  *      We don't want to expose copy_files internals to
795  *      the exec layer of the kernel.
796  */
797
798 int unshare_files(void)
799 {
800         struct files_struct *files  = current->files;
801         int rc;
802
803         BUG_ON(!files);
804
805         /* This can race but the race causes us to copy when we don't
806            need to and drop the copy */
807         if(atomic_read(&files->count) == 1)
808         {
809                 atomic_inc(&files->count);
810                 return 0;
811         }
812         rc = copy_files(0, current);
813         if(rc)
814                 current->files = files;
815         return rc;
816 }
817
818 EXPORT_SYMBOL(unshare_files);
819
820 static inline int copy_sighand(unsigned long clone_flags, struct task_struct * tsk)
821 {
822         struct sighand_struct *sig;
823
824         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
825                 atomic_inc(&current->sighand->count);
826                 return 0;
827         }
828         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
829         rcu_assign_pointer(tsk->sighand, sig);
830         if (!sig)
831                 return -ENOMEM;
832         atomic_set(&sig->count, 1);
833         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
834         return 0;
835 }
836
837 void __cleanup_sighand(struct sighand_struct *sighand)
838 {
839         if (atomic_dec_and_test(&sighand->count))
840                 kmem_cache_free(sighand_cachep, sighand);
841 }
842
843 static inline int copy_signal(unsigned long clone_flags, struct task_struct * tsk)
844 {
845         struct signal_struct *sig;
846         int ret;
847
848         if (clone_flags & CLONE_THREAD) {
849                 atomic_inc(&current->signal->count);
850                 atomic_inc(&current->signal->live);
851                 return 0;
852         }
853         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
854         tsk->signal = sig;
855         if (!sig)
856                 return -ENOMEM;
857
858         ret = copy_thread_group_keys(tsk);
859         if (ret < 0) {
860                 kmem_cache_free(signal_cachep, sig);
861                 return ret;
862         }
863
864         atomic_set(&sig->count, 1);
865         atomic_set(&sig->live, 1);
866         init_waitqueue_head(&sig->wait_chldexit);
867         sig->flags = 0;
868         sig->group_exit_code = 0;
869         sig->group_exit_task = NULL;
870         sig->group_stop_count = 0;
871         sig->curr_target = NULL;
872         init_sigpending(&sig->shared_pending);
873         INIT_LIST_HEAD(&sig->posix_timers);
874
875         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_REL);
876         sig->it_real_incr.tv64 = 0;
877         sig->real_timer.function = it_real_fn;
878         sig->tsk = tsk;
879
880         sig->it_virt_expires = cputime_zero;
881         sig->it_virt_incr = cputime_zero;
882         sig->it_prof_expires = cputime_zero;
883         sig->it_prof_incr = cputime_zero;
884
885         sig->leader = 0;        /* session leadership doesn't inherit */
886         sig->tty_old_pgrp = 0;
887
888         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
889         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
890         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
891         sig->sched_time = 0;
892         INIT_LIST_HEAD(&sig->cpu_timers[0]);
893         INIT_LIST_HEAD(&sig->cpu_timers[1]);
894         INIT_LIST_HEAD(&sig->cpu_timers[2]);
895         taskstats_tgid_init(sig);
896
897         task_lock(current->group_leader);
898         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
899         task_unlock(current->group_leader);
900
901         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
902                 /*
903                  * New sole thread in the process gets an expiry time
904                  * of the whole CPU time limit.
905                  */
906                 tsk->it_prof_expires =
907                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
908         }
909         acct_init_pacct(&sig->pacct);
910
911         return 0;
912 }
913
914 void __cleanup_signal(struct signal_struct *sig)
915 {
916         exit_thread_group_keys(sig);
917         kmem_cache_free(signal_cachep, sig);
918 }
919
920 static inline void cleanup_signal(struct task_struct *tsk)
921 {
922         struct signal_struct *sig = tsk->signal;
923
924         atomic_dec(&sig->live);
925
926         if (atomic_dec_and_test(&sig->count))
927                 __cleanup_signal(sig);
928 }
929
930 static inline void copy_flags(unsigned long clone_flags, struct task_struct *p)
931 {
932         unsigned long new_flags = p->flags;
933
934         new_flags &= ~(PF_SUPERPRIV | PF_NOFREEZE);
935         new_flags |= PF_FORKNOEXEC;
936         if (!(clone_flags & CLONE_PTRACE))
937                 p->ptrace = 0;
938         p->flags = new_flags;
939 }
940
941 asmlinkage long sys_set_tid_address(int __user *tidptr)
942 {
943         current->clear_child_tid = tidptr;
944
945         return current->pid;
946 }
947
948 static inline void rt_mutex_init_task(struct task_struct *p)
949 {
950 #ifdef CONFIG_RT_MUTEXES
951         spin_lock_init(&p->pi_lock);
952         plist_head_init(&p->pi_waiters, &p->pi_lock);
953         p->pi_blocked_on = NULL;
954 #endif
955 }
956
957 /*
958  * This creates a new process as a copy of the old one,
959  * but does not actually start it yet.
960  *
961  * It copies the registers, and all the appropriate
962  * parts of the process environment (as per the clone
963  * flags). The actual kick-off is left to the caller.
964  */
965 static struct task_struct *copy_process(unsigned long clone_flags,
966                                         unsigned long stack_start,
967                                         struct pt_regs *regs,
968                                         unsigned long stack_size,
969                                         int __user *parent_tidptr,
970                                         int __user *child_tidptr,
971                                         int pid)
972 {
973         int retval;
974         struct task_struct *p = NULL;
975
976         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
977                 return ERR_PTR(-EINVAL);
978
979         /*
980          * Thread groups must share signals as well, and detached threads
981          * can only be started up within the thread group.
982          */
983         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
984                 return ERR_PTR(-EINVAL);
985
986         /*
987          * Shared signal handlers imply shared VM. By way of the above,
988          * thread groups also imply shared VM. Blocking this case allows
989          * for various simplifications in other code.
990          */
991         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
992                 return ERR_PTR(-EINVAL);
993
994         retval = security_task_create(clone_flags);
995         if (retval)
996                 goto fork_out;
997
998         retval = -ENOMEM;
999         p = dup_task_struct(current);
1000         if (!p)
1001                 goto fork_out;
1002
1003         rt_mutex_init_task(p);
1004
1005 #ifdef CONFIG_TRACE_IRQFLAGS
1006         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1007         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1008 #endif
1009         retval = -EAGAIN;
1010         if (atomic_read(&p->user->processes) >=
1011                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1012                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1013                                 p->user != &root_user)
1014                         goto bad_fork_free;
1015         }
1016
1017         atomic_inc(&p->user->__count);
1018         atomic_inc(&p->user->processes);
1019         get_group_info(p->group_info);
1020
1021         /*
1022          * If multiple threads are within copy_process(), then this check
1023          * triggers too late. This doesn't hurt, the check is only there
1024          * to stop root fork bombs.
1025          */
1026         if (nr_threads >= max_threads)
1027                 goto bad_fork_cleanup_count;
1028
1029         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1030                 goto bad_fork_cleanup_count;
1031
1032         if (p->binfmt && !try_module_get(p->binfmt->module))
1033                 goto bad_fork_cleanup_put_domain;
1034
1035         p->did_exec = 0;
1036         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1037         copy_flags(clone_flags, p);
1038         p->pid = pid;
1039         retval = -EFAULT;
1040         if (clone_flags & CLONE_PARENT_SETTID)
1041                 if (put_user(p->pid, parent_tidptr))
1042                         goto bad_fork_cleanup_delays_binfmt;
1043
1044         INIT_LIST_HEAD(&p->children);
1045         INIT_LIST_HEAD(&p->sibling);
1046         p->vfork_done = NULL;
1047         spin_lock_init(&p->alloc_lock);
1048
1049         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1050         init_sigpending(&p->pending);
1051
1052         p->utime = cputime_zero;
1053         p->stime = cputime_zero;
1054         p->sched_time = 0;
1055         p->rchar = 0;           /* I/O counter: bytes read */
1056         p->wchar = 0;           /* I/O counter: bytes written */
1057         p->syscr = 0;           /* I/O counter: read syscalls */
1058         p->syscw = 0;           /* I/O counter: write syscalls */
1059         task_io_accounting_init(p);
1060         acct_clear_integrals(p);
1061
1062         p->it_virt_expires = cputime_zero;
1063         p->it_prof_expires = cputime_zero;
1064         p->it_sched_expires = 0;
1065         INIT_LIST_HEAD(&p->cpu_timers[0]);
1066         INIT_LIST_HEAD(&p->cpu_timers[1]);
1067         INIT_LIST_HEAD(&p->cpu_timers[2]);
1068
1069         p->lock_depth = -1;             /* -1 = no lock */
1070         do_posix_clock_monotonic_gettime(&p->start_time);
1071         p->security = NULL;
1072         p->io_context = NULL;
1073         p->io_wait = NULL;
1074         p->audit_context = NULL;
1075         cpuset_fork(p);
1076 #ifdef CONFIG_NUMA
1077         p->mempolicy = mpol_copy(p->mempolicy);
1078         if (IS_ERR(p->mempolicy)) {
1079                 retval = PTR_ERR(p->mempolicy);
1080                 p->mempolicy = NULL;
1081                 goto bad_fork_cleanup_cpuset;
1082         }
1083         mpol_fix_fork_child_flag(p);
1084 #endif
1085 #ifdef CONFIG_TRACE_IRQFLAGS
1086         p->irq_events = 0;
1087 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1088         p->hardirqs_enabled = 1;
1089 #else
1090         p->hardirqs_enabled = 0;
1091 #endif
1092         p->hardirq_enable_ip = 0;
1093         p->hardirq_enable_event = 0;
1094         p->hardirq_disable_ip = _THIS_IP_;
1095         p->hardirq_disable_event = 0;
1096         p->softirqs_enabled = 1;
1097         p->softirq_enable_ip = _THIS_IP_;
1098         p->softirq_enable_event = 0;
1099         p->softirq_disable_ip = 0;
1100         p->softirq_disable_event = 0;
1101         p->hardirq_context = 0;
1102         p->softirq_context = 0;
1103 #endif
1104 #ifdef CONFIG_LOCKDEP
1105         p->lockdep_depth = 0; /* no locks held yet */
1106         p->curr_chain_key = 0;
1107         p->lockdep_recursion = 0;
1108 #endif
1109
1110 #ifdef CONFIG_DEBUG_MUTEXES
1111         p->blocked_on = NULL; /* not blocked yet */
1112 #endif
1113
1114         p->tgid = p->pid;
1115         if (clone_flags & CLONE_THREAD)
1116                 p->tgid = current->tgid;
1117
1118         if ((retval = security_task_alloc(p)))
1119                 goto bad_fork_cleanup_policy;
1120         if ((retval = audit_alloc(p)))
1121                 goto bad_fork_cleanup_security;
1122         /* copy all the process information */
1123         if ((retval = copy_semundo(clone_flags, p)))
1124                 goto bad_fork_cleanup_audit;
1125         if ((retval = copy_files(clone_flags, p)))
1126                 goto bad_fork_cleanup_semundo;
1127         if ((retval = copy_fs(clone_flags, p)))
1128                 goto bad_fork_cleanup_files;
1129         if ((retval = copy_sighand(clone_flags, p)))
1130                 goto bad_fork_cleanup_fs;
1131         if ((retval = copy_signal(clone_flags, p)))
1132                 goto bad_fork_cleanup_sighand;
1133         if ((retval = copy_mm(clone_flags, p)))
1134                 goto bad_fork_cleanup_signal;
1135         if ((retval = copy_keys(clone_flags, p)))
1136                 goto bad_fork_cleanup_mm;
1137         if ((retval = copy_namespaces(clone_flags, p)))
1138                 goto bad_fork_cleanup_keys;
1139         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1140         if (retval)
1141                 goto bad_fork_cleanup_namespaces;
1142
1143         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1144         /*
1145          * Clear TID on mm_release()?
1146          */
1147         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1148         p->robust_list = NULL;
1149 #ifdef CONFIG_COMPAT
1150         p->compat_robust_list = NULL;
1151 #endif
1152         INIT_LIST_HEAD(&p->pi_state_list);
1153         p->pi_state_cache = NULL;
1154
1155         /*
1156          * sigaltstack should be cleared when sharing the same VM
1157          */
1158         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1159                 p->sas_ss_sp = p->sas_ss_size = 0;
1160
1161         /*
1162          * Syscall tracing should be turned off in the child regardless
1163          * of CLONE_PTRACE.
1164          */
1165         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1166 #ifdef TIF_SYSCALL_EMU
1167         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1168 #endif
1169
1170         /* Our parent execution domain becomes current domain
1171            These must match for thread signalling to apply */
1172         p->parent_exec_id = p->self_exec_id;
1173
1174         /* ok, now we should be set up.. */
1175         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1176         p->pdeath_signal = 0;
1177         p->exit_state = 0;
1178
1179         /*
1180          * Ok, make it visible to the rest of the system.
1181          * We dont wake it up yet.
1182          */
1183         p->group_leader = p;
1184         INIT_LIST_HEAD(&p->thread_group);
1185         INIT_LIST_HEAD(&p->ptrace_children);
1186         INIT_LIST_HEAD(&p->ptrace_list);
1187
1188         /* Perform scheduler related setup. Assign this task to a CPU. */
1189         sched_fork(p, clone_flags);
1190
1191         /* Need tasklist lock for parent etc handling! */
1192         write_lock_irq(&tasklist_lock);
1193
1194         /* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
1195         p->ioprio = current->ioprio;
1196
1197         /*
1198          * The task hasn't been attached yet, so its cpus_allowed mask will
1199          * not be changed, nor will its assigned CPU.
1200          *
1201          * The cpus_allowed mask of the parent may have changed after it was
1202          * copied first time - so re-copy it here, then check the child's CPU
1203          * to ensure it is on a valid CPU (and if not, just force it back to
1204          * parent's CPU). This avoids alot of nasty races.
1205          */
1206         p->cpus_allowed = current->cpus_allowed;
1207         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1208                         !cpu_online(task_cpu(p))))
1209                 set_task_cpu(p, smp_processor_id());
1210
1211         /* CLONE_PARENT re-uses the old parent */
1212         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1213                 p->real_parent = current->real_parent;
1214         else
1215                 p->real_parent = current;
1216         p->parent = p->real_parent;
1217
1218         spin_lock(&current->sighand->siglock);
1219
1220         /*
1221          * Process group and session signals need to be delivered to just the
1222          * parent before the fork or both the parent and the child after the
1223          * fork. Restart if a signal comes in before we add the new process to
1224          * it's process group.
1225          * A fatal signal pending means that current will exit, so the new
1226          * thread can't slip out of an OOM kill (or normal SIGKILL).
1227          */
1228         recalc_sigpending();
1229         if (signal_pending(current)) {
1230                 spin_unlock(&current->sighand->siglock);
1231                 write_unlock_irq(&tasklist_lock);
1232                 retval = -ERESTARTNOINTR;
1233                 goto bad_fork_cleanup_namespaces;
1234         }
1235
1236         if (clone_flags & CLONE_THREAD) {
1237                 p->group_leader = current->group_leader;
1238                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1239
1240                 if (!cputime_eq(current->signal->it_virt_expires,
1241                                 cputime_zero) ||
1242                     !cputime_eq(current->signal->it_prof_expires,
1243                                 cputime_zero) ||
1244                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1245                     !list_empty(&current->signal->cpu_timers[0]) ||
1246                     !list_empty(&current->signal->cpu_timers[1]) ||
1247                     !list_empty(&current->signal->cpu_timers[2])) {
1248                         /*
1249                          * Have child wake up on its first tick to check
1250                          * for process CPU timers.
1251                          */
1252                         p->it_prof_expires = jiffies_to_cputime(1);
1253                 }
1254         }
1255
1256         if (likely(p->pid)) {
1257                 add_parent(p);
1258                 if (unlikely(p->ptrace & PT_PTRACED))
1259                         __ptrace_link(p, current->parent);
1260
1261                 if (thread_group_leader(p)) {
1262                         p->signal->tty = current->signal->tty;
1263                         p->signal->pgrp = process_group(current);
1264                         set_signal_session(p->signal, process_session(current));
1265                         attach_pid(p, PIDTYPE_PGID, process_group(p));
1266                         attach_pid(p, PIDTYPE_SID, process_session(p));
1267
1268                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1269                         __get_cpu_var(process_counts)++;
1270                 }
1271                 attach_pid(p, PIDTYPE_PID, p->pid);
1272                 nr_threads++;
1273         }
1274
1275         total_forks++;
1276         spin_unlock(&current->sighand->siglock);
1277         write_unlock_irq(&tasklist_lock);
1278         proc_fork_connector(p);
1279         return p;
1280
1281 bad_fork_cleanup_namespaces:
1282         exit_task_namespaces(p);
1283 bad_fork_cleanup_keys:
1284         exit_keys(p);
1285 bad_fork_cleanup_mm:
1286         if (p->mm)
1287                 mmput(p->mm);
1288 bad_fork_cleanup_signal:
1289         cleanup_signal(p);
1290 bad_fork_cleanup_sighand:
1291         __cleanup_sighand(p->sighand);
1292 bad_fork_cleanup_fs:
1293         exit_fs(p); /* blocking */
1294 bad_fork_cleanup_files:
1295         exit_files(p); /* blocking */
1296 bad_fork_cleanup_semundo:
1297         exit_sem(p);
1298 bad_fork_cleanup_audit:
1299         audit_free(p);
1300 bad_fork_cleanup_security:
1301         security_task_free(p);
1302 bad_fork_cleanup_policy:
1303 #ifdef CONFIG_NUMA
1304         mpol_free(p->mempolicy);
1305 bad_fork_cleanup_cpuset:
1306 #endif
1307         cpuset_exit(p);
1308 bad_fork_cleanup_delays_binfmt:
1309         delayacct_tsk_free(p);
1310         if (p->binfmt)
1311                 module_put(p->binfmt->module);
1312 bad_fork_cleanup_put_domain:
1313         module_put(task_thread_info(p)->exec_domain->module);
1314 bad_fork_cleanup_count:
1315         put_group_info(p->group_info);
1316         atomic_dec(&p->user->processes);
1317         free_uid(p->user);
1318 bad_fork_free:
1319         free_task(p);
1320 fork_out:
1321         return ERR_PTR(retval);
1322 }
1323
1324 noinline struct pt_regs * __devinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1325 {
1326         memset(regs, 0, sizeof(struct pt_regs));
1327         return regs;
1328 }
1329
1330 struct task_struct * __devinit fork_idle(int cpu)
1331 {
1332         struct task_struct *task;
1333         struct pt_regs regs;
1334
1335         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL, NULL, 0);
1336         if (!IS_ERR(task))
1337                 init_idle(task, cpu);
1338
1339         return task;
1340 }
1341
1342 static inline int fork_traceflag (unsigned clone_flags)
1343 {
1344         if (clone_flags & CLONE_UNTRACED)
1345                 return 0;
1346         else if (clone_flags & CLONE_VFORK) {
1347                 if (current->ptrace & PT_TRACE_VFORK)
1348                         return PTRACE_EVENT_VFORK;
1349         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1350                 if (current->ptrace & PT_TRACE_CLONE)
1351                         return PTRACE_EVENT_CLONE;
1352         } else if (current->ptrace & PT_TRACE_FORK)
1353                 return PTRACE_EVENT_FORK;
1354
1355         return 0;
1356 }
1357
1358 /*
1359  *  Ok, this is the main fork-routine.
1360  *
1361  * It copies the process, and if successful kick-starts
1362  * it and waits for it to finish using the VM if required.
1363  */
1364 long do_fork(unsigned long clone_flags,
1365               unsigned long stack_start,
1366               struct pt_regs *regs,
1367               unsigned long stack_size,
1368               int __user *parent_tidptr,
1369               int __user *child_tidptr)
1370 {
1371         struct task_struct *p;
1372         int trace = 0;
1373         struct pid *pid = alloc_pid();
1374         long nr;
1375
1376         if (!pid)
1377                 return -EAGAIN;
1378         nr = pid->nr;
1379         if (unlikely(current->ptrace)) {
1380                 trace = fork_traceflag (clone_flags);
1381                 if (trace)
1382                         clone_flags |= CLONE_PTRACE;
1383         }
1384
1385         p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, nr);
1386         /*
1387          * Do this prior waking up the new thread - the thread pointer
1388          * might get invalid after that point, if the thread exits quickly.
1389          */
1390         if (!IS_ERR(p)) {
1391                 struct completion vfork;
1392
1393                 if (clone_flags & CLONE_VFORK) {
1394                         p->vfork_done = &vfork;
1395                         init_completion(&vfork);
1396                 }
1397
1398                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1399                         /*
1400                          * We'll start up with an immediate SIGSTOP.
1401                          */
1402                         sigaddset(&p->pending.signal, SIGSTOP);
1403                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1404                 }
1405
1406                 if (!(clone_flags & CLONE_STOPPED))
1407                         wake_up_new_task(p, clone_flags);
1408                 else
1409                         p->state = TASK_STOPPED;
1410
1411                 if (unlikely (trace)) {
1412                         current->ptrace_message = nr;
1413                         ptrace_notify ((trace << 8) | SIGTRAP);
1414                 }
1415
1416                 if (clone_flags & CLONE_VFORK) {
1417                         wait_for_completion(&vfork);
1418                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1419                                 current->ptrace_message = nr;
1420                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1421                         }
1422                 }
1423         } else {
1424                 free_pid(pid);
1425                 nr = PTR_ERR(p);
1426         }
1427         return nr;
1428 }
1429
1430 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1431 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1432 #endif
1433
1434 static void sighand_ctor(void *data, struct kmem_cache *cachep, unsigned long flags)
1435 {
1436         struct sighand_struct *sighand = data;
1437
1438         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
1439                                         SLAB_CTOR_CONSTRUCTOR)
1440                 spin_lock_init(&sighand->siglock);
1441 }
1442
1443 void __init proc_caches_init(void)
1444 {
1445         sighand_cachep = kmem_cache_create("sighand_cache",
1446                         sizeof(struct sighand_struct), 0,
1447                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1448                         sighand_ctor, NULL);
1449         signal_cachep = kmem_cache_create("signal_cache",
1450                         sizeof(struct signal_struct), 0,
1451                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1452         files_cachep = kmem_cache_create("files_cache", 
1453                         sizeof(struct files_struct), 0,
1454                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1455         fs_cachep = kmem_cache_create("fs_cache", 
1456                         sizeof(struct fs_struct), 0,
1457                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1458         vm_area_cachep = kmem_cache_create("vm_area_struct",
1459                         sizeof(struct vm_area_struct), 0,
1460                         SLAB_PANIC, NULL, NULL);
1461         mm_cachep = kmem_cache_create("mm_struct",
1462                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1463                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1464 }
1465
1466
1467 /*
1468  * Check constraints on flags passed to the unshare system call and
1469  * force unsharing of additional process context as appropriate.
1470  */
1471 static inline void check_unshare_flags(unsigned long *flags_ptr)
1472 {
1473         /*
1474          * If unsharing a thread from a thread group, must also
1475          * unshare vm.
1476          */
1477         if (*flags_ptr & CLONE_THREAD)
1478                 *flags_ptr |= CLONE_VM;
1479
1480         /*
1481          * If unsharing vm, must also unshare signal handlers.
1482          */
1483         if (*flags_ptr & CLONE_VM)
1484                 *flags_ptr |= CLONE_SIGHAND;
1485
1486         /*
1487          * If unsharing signal handlers and the task was created
1488          * using CLONE_THREAD, then must unshare the thread
1489          */
1490         if ((*flags_ptr & CLONE_SIGHAND) &&
1491             (atomic_read(&current->signal->count) > 1))
1492                 *flags_ptr |= CLONE_THREAD;
1493
1494         /*
1495          * If unsharing namespace, must also unshare filesystem information.
1496          */
1497         if (*flags_ptr & CLONE_NEWNS)
1498                 *flags_ptr |= CLONE_FS;
1499 }
1500
1501 /*
1502  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1503  */
1504 static int unshare_thread(unsigned long unshare_flags)
1505 {
1506         if (unshare_flags & CLONE_THREAD)
1507                 return -EINVAL;
1508
1509         return 0;
1510 }
1511
1512 /*
1513  * Unshare the filesystem structure if it is being shared
1514  */
1515 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1516 {
1517         struct fs_struct *fs = current->fs;
1518
1519         if ((unshare_flags & CLONE_FS) &&
1520             (fs && atomic_read(&fs->count) > 1)) {
1521                 *new_fsp = __copy_fs_struct(current->fs);
1522                 if (!*new_fsp)
1523                         return -ENOMEM;
1524         }
1525
1526         return 0;
1527 }
1528
1529 /*
1530  * Unshare the mnt_namespace structure if it is being shared
1531  */
1532 static int unshare_mnt_namespace(unsigned long unshare_flags,
1533                 struct mnt_namespace **new_nsp, struct fs_struct *new_fs)
1534 {
1535         struct mnt_namespace *ns = current->nsproxy->mnt_ns;
1536
1537         if ((unshare_flags & CLONE_NEWNS) && ns) {
1538                 if (!capable(CAP_SYS_ADMIN))
1539                         return -EPERM;
1540
1541                 *new_nsp = dup_mnt_ns(current, new_fs ? new_fs : current->fs);
1542                 if (!*new_nsp)
1543                         return -ENOMEM;
1544         }
1545
1546         return 0;
1547 }
1548
1549 /*
1550  * Unsharing of sighand is not supported yet
1551  */
1552 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1553 {
1554         struct sighand_struct *sigh = current->sighand;
1555
1556         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1557                 return -EINVAL;
1558         else
1559                 return 0;
1560 }
1561
1562 /*
1563  * Unshare vm if it is being shared
1564  */
1565 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1566 {
1567         struct mm_struct *mm = current->mm;
1568
1569         if ((unshare_flags & CLONE_VM) &&
1570             (mm && atomic_read(&mm->mm_users) > 1)) {
1571                 return -EINVAL;
1572         }
1573
1574         return 0;
1575 }
1576
1577 /*
1578  * Unshare file descriptor table if it is being shared
1579  */
1580 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1581 {
1582         struct files_struct *fd = current->files;
1583         int error = 0;
1584
1585         if ((unshare_flags & CLONE_FILES) &&
1586             (fd && atomic_read(&fd->count) > 1)) {
1587                 *new_fdp = dup_fd(fd, &error);
1588                 if (!*new_fdp)
1589                         return error;
1590         }
1591
1592         return 0;
1593 }
1594
1595 /*
1596  * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
1597  * supported yet
1598  */
1599 static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
1600 {
1601         if (unshare_flags & CLONE_SYSVSEM)
1602                 return -EINVAL;
1603
1604         return 0;
1605 }
1606
1607 #ifndef CONFIG_IPC_NS
1608 static inline int unshare_ipcs(unsigned long flags, struct ipc_namespace **ns)
1609 {
1610         if (flags & CLONE_NEWIPC)
1611                 return -EINVAL;
1612
1613         return 0;
1614 }
1615 #endif
1616
1617 /*
1618  * unshare allows a process to 'unshare' part of the process
1619  * context which was originally shared using clone.  copy_*
1620  * functions used by do_fork() cannot be used here directly
1621  * because they modify an inactive task_struct that is being
1622  * constructed. Here we are modifying the current, active,
1623  * task_struct.
1624  */
1625 asmlinkage long sys_unshare(unsigned long unshare_flags)
1626 {
1627         int err = 0;
1628         struct fs_struct *fs, *new_fs = NULL;
1629         struct mnt_namespace *ns, *new_ns = NULL;
1630         struct sighand_struct *new_sigh = NULL;
1631         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1632         struct files_struct *fd, *new_fd = NULL;
1633         struct sem_undo_list *new_ulist = NULL;
1634         struct nsproxy *new_nsproxy = NULL, *old_nsproxy = NULL;
1635         struct uts_namespace *uts, *new_uts = NULL;
1636         struct ipc_namespace *ipc, *new_ipc = NULL;
1637
1638         check_unshare_flags(&unshare_flags);
1639
1640         /* Return -EINVAL for all unsupported flags */
1641         err = -EINVAL;
1642         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1643                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1644                                 CLONE_NEWUTS|CLONE_NEWIPC))
1645                 goto bad_unshare_out;
1646
1647         if ((err = unshare_thread(unshare_flags)))
1648                 goto bad_unshare_out;
1649         if ((err = unshare_fs(unshare_flags, &new_fs)))
1650                 goto bad_unshare_cleanup_thread;
1651         if ((err = unshare_mnt_namespace(unshare_flags, &new_ns, new_fs)))
1652                 goto bad_unshare_cleanup_fs;
1653         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1654                 goto bad_unshare_cleanup_ns;
1655         if ((err = unshare_vm(unshare_flags, &new_mm)))
1656                 goto bad_unshare_cleanup_sigh;
1657         if ((err = unshare_fd(unshare_flags, &new_fd)))
1658                 goto bad_unshare_cleanup_vm;
1659         if ((err = unshare_semundo(unshare_flags, &new_ulist)))
1660                 goto bad_unshare_cleanup_fd;
1661         if ((err = unshare_utsname(unshare_flags, &new_uts)))
1662                 goto bad_unshare_cleanup_semundo;
1663         if ((err = unshare_ipcs(unshare_flags, &new_ipc)))
1664                 goto bad_unshare_cleanup_uts;
1665
1666         if (new_ns || new_uts || new_ipc) {
1667                 old_nsproxy = current->nsproxy;
1668                 new_nsproxy = dup_namespaces(old_nsproxy);
1669                 if (!new_nsproxy) {
1670                         err = -ENOMEM;
1671                         goto bad_unshare_cleanup_ipc;
1672                 }
1673         }
1674
1675         if (new_fs || new_ns || new_mm || new_fd || new_ulist ||
1676                                 new_uts || new_ipc) {
1677
1678                 task_lock(current);
1679
1680                 if (new_nsproxy) {
1681                         current->nsproxy = new_nsproxy;
1682                         new_nsproxy = old_nsproxy;
1683                 }
1684
1685                 if (new_fs) {
1686                         fs = current->fs;
1687                         current->fs = new_fs;
1688                         new_fs = fs;
1689                 }
1690
1691                 if (new_ns) {
1692                         ns = current->nsproxy->mnt_ns;
1693                         current->nsproxy->mnt_ns = new_ns;
1694                         new_ns = ns;
1695                 }
1696
1697                 if (new_mm) {
1698                         mm = current->mm;
1699                         active_mm = current->active_mm;
1700                         current->mm = new_mm;
1701                         current->active_mm = new_mm;
1702                         activate_mm(active_mm, new_mm);
1703                         new_mm = mm;
1704                 }
1705
1706                 if (new_fd) {
1707                         fd = current->files;
1708                         current->files = new_fd;
1709                         new_fd = fd;
1710                 }
1711
1712                 if (new_uts) {
1713                         uts = current->nsproxy->uts_ns;
1714                         current->nsproxy->uts_ns = new_uts;
1715                         new_uts = uts;
1716                 }
1717
1718                 if (new_ipc) {
1719                         ipc = current->nsproxy->ipc_ns;
1720                         current->nsproxy->ipc_ns = new_ipc;
1721                         new_ipc = ipc;
1722                 }
1723
1724                 task_unlock(current);
1725         }
1726
1727         if (new_nsproxy)
1728                 put_nsproxy(new_nsproxy);
1729
1730 bad_unshare_cleanup_ipc:
1731         if (new_ipc)
1732                 put_ipc_ns(new_ipc);
1733
1734 bad_unshare_cleanup_uts:
1735         if (new_uts)
1736                 put_uts_ns(new_uts);
1737
1738 bad_unshare_cleanup_semundo:
1739 bad_unshare_cleanup_fd:
1740         if (new_fd)
1741                 put_files_struct(new_fd);
1742
1743 bad_unshare_cleanup_vm:
1744         if (new_mm)
1745                 mmput(new_mm);
1746
1747 bad_unshare_cleanup_sigh:
1748         if (new_sigh)
1749                 if (atomic_dec_and_test(&new_sigh->count))
1750                         kmem_cache_free(sighand_cachep, new_sigh);
1751
1752 bad_unshare_cleanup_ns:
1753         if (new_ns)
1754                 put_mnt_ns(new_ns);
1755
1756 bad_unshare_cleanup_fs:
1757         if (new_fs)
1758                 put_fs_struct(new_fs);
1759
1760 bad_unshare_cleanup_thread:
1761 bad_unshare_out:
1762         return err;
1763 }