stackprotector: use canary at end of stack to indicate overruns at oops time
[linux-2.6.git] / kernel / fork.c
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/mnt_namespace.h>
21 #include <linux/personality.h>
22 #include <linux/mempolicy.h>
23 #include <linux/sem.h>
24 #include <linux/file.h>
25 #include <linux/fdtable.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/cgroup.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/memcontrol.h>
45 #include <linux/profile.h>
46 #include <linux/rmap.h>
47 #include <linux/acct.h>
48 #include <linux/tsacct_kern.h>
49 #include <linux/cn_proc.h>
50 #include <linux/freezer.h>
51 #include <linux/delayacct.h>
52 #include <linux/taskstats_kern.h>
53 #include <linux/random.h>
54 #include <linux/tty.h>
55 #include <linux/proc_fs.h>
56 #include <linux/blkdev.h>
57 #include <linux/magic.h>
58
59 #include <asm/pgtable.h>
60 #include <asm/pgalloc.h>
61 #include <asm/uaccess.h>
62 #include <asm/mmu_context.h>
63 #include <asm/cacheflush.h>
64 #include <asm/tlbflush.h>
65
66 /*
67  * Protected counters by write_lock_irq(&tasklist_lock)
68  */
69 unsigned long total_forks;      /* Handle normal Linux uptimes. */
70 int nr_threads;                 /* The idle threads do not count.. */
71
72 int max_threads;                /* tunable limit on nr_threads */
73
74 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
75
76 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
77
78 int nr_processes(void)
79 {
80         int cpu;
81         int total = 0;
82
83         for_each_online_cpu(cpu)
84                 total += per_cpu(process_counts, cpu);
85
86         return total;
87 }
88
89 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
90 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
91 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
92 static struct kmem_cache *task_struct_cachep;
93 #endif
94
95 /* SLAB cache for signal_struct structures (tsk->signal) */
96 static struct kmem_cache *signal_cachep;
97
98 /* SLAB cache for sighand_struct structures (tsk->sighand) */
99 struct kmem_cache *sighand_cachep;
100
101 /* SLAB cache for files_struct structures (tsk->files) */
102 struct kmem_cache *files_cachep;
103
104 /* SLAB cache for fs_struct structures (tsk->fs) */
105 struct kmem_cache *fs_cachep;
106
107 /* SLAB cache for vm_area_struct structures */
108 struct kmem_cache *vm_area_cachep;
109
110 /* SLAB cache for mm_struct structures (tsk->mm) */
111 static struct kmem_cache *mm_cachep;
112
113 void free_task(struct task_struct *tsk)
114 {
115         prop_local_destroy_single(&tsk->dirties);
116         free_thread_info(tsk->stack);
117         rt_mutex_debug_task_free(tsk);
118         free_task_struct(tsk);
119 }
120 EXPORT_SYMBOL(free_task);
121
122 void __put_task_struct(struct task_struct *tsk)
123 {
124         WARN_ON(!tsk->exit_state);
125         WARN_ON(atomic_read(&tsk->usage));
126         WARN_ON(tsk == current);
127
128         security_task_free(tsk);
129         free_uid(tsk->user);
130         put_group_info(tsk->group_info);
131         delayacct_tsk_free(tsk);
132
133         if (!profile_handoff_task(tsk))
134                 free_task(tsk);
135 }
136
137 /*
138  * macro override instead of weak attribute alias, to workaround
139  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
140  */
141 #ifndef arch_task_cache_init
142 #define arch_task_cache_init()
143 #endif
144
145 void __init fork_init(unsigned long mempages)
146 {
147 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
148 #ifndef ARCH_MIN_TASKALIGN
149 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
150 #endif
151         /* create a slab on which task_structs can be allocated */
152         task_struct_cachep =
153                 kmem_cache_create("task_struct", sizeof(struct task_struct),
154                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
155 #endif
156
157         /* do the arch specific task caches init */
158         arch_task_cache_init();
159
160         /*
161          * The default maximum number of threads is set to a safe
162          * value: the thread structures can take up at most half
163          * of memory.
164          */
165         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
166
167         /*
168          * we need to allow at least 20 threads to boot a system
169          */
170         if(max_threads < 20)
171                 max_threads = 20;
172
173         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
174         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
175         init_task.signal->rlim[RLIMIT_SIGPENDING] =
176                 init_task.signal->rlim[RLIMIT_NPROC];
177 }
178
179 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
180                                                struct task_struct *src)
181 {
182         *dst = *src;
183         return 0;
184 }
185
186 static struct task_struct *dup_task_struct(struct task_struct *orig)
187 {
188         struct task_struct *tsk;
189         struct thread_info *ti;
190         unsigned long *stackend;
191
192         int err;
193
194         prepare_to_copy(orig);
195
196         tsk = alloc_task_struct();
197         if (!tsk)
198                 return NULL;
199
200         ti = alloc_thread_info(tsk);
201         if (!ti) {
202                 free_task_struct(tsk);
203                 return NULL;
204         }
205
206         err = arch_dup_task_struct(tsk, orig);
207         if (err)
208                 goto out;
209
210         tsk->stack = ti;
211
212         err = prop_local_init_single(&tsk->dirties);
213         if (err)
214                 goto out;
215
216         setup_thread_stack(tsk, orig);
217         stackend = end_of_stack(tsk);
218         *stackend = STACK_END_MAGIC;    /* for overflow detection */
219
220 #ifdef CONFIG_CC_STACKPROTECTOR
221         tsk->stack_canary = get_random_int();
222 #endif
223
224         /* One for us, one for whoever does the "release_task()" (usually parent) */
225         atomic_set(&tsk->usage,2);
226         atomic_set(&tsk->fs_excl, 0);
227 #ifdef CONFIG_BLK_DEV_IO_TRACE
228         tsk->btrace_seq = 0;
229 #endif
230         tsk->splice_pipe = NULL;
231         return tsk;
232
233 out:
234         free_thread_info(ti);
235         free_task_struct(tsk);
236         return NULL;
237 }
238
239 #ifdef CONFIG_MMU
240 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
241 {
242         struct vm_area_struct *mpnt, *tmp, **pprev;
243         struct rb_node **rb_link, *rb_parent;
244         int retval;
245         unsigned long charge;
246         struct mempolicy *pol;
247
248         down_write(&oldmm->mmap_sem);
249         flush_cache_dup_mm(oldmm);
250         /*
251          * Not linked in yet - no deadlock potential:
252          */
253         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
254
255         mm->locked_vm = 0;
256         mm->mmap = NULL;
257         mm->mmap_cache = NULL;
258         mm->free_area_cache = oldmm->mmap_base;
259         mm->cached_hole_size = ~0UL;
260         mm->map_count = 0;
261         cpus_clear(mm->cpu_vm_mask);
262         mm->mm_rb = RB_ROOT;
263         rb_link = &mm->mm_rb.rb_node;
264         rb_parent = NULL;
265         pprev = &mm->mmap;
266
267         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
268                 struct file *file;
269
270                 if (mpnt->vm_flags & VM_DONTCOPY) {
271                         long pages = vma_pages(mpnt);
272                         mm->total_vm -= pages;
273                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
274                                                                 -pages);
275                         continue;
276                 }
277                 charge = 0;
278                 if (mpnt->vm_flags & VM_ACCOUNT) {
279                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
280                         if (security_vm_enough_memory(len))
281                                 goto fail_nomem;
282                         charge = len;
283                 }
284                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
285                 if (!tmp)
286                         goto fail_nomem;
287                 *tmp = *mpnt;
288                 pol = mpol_dup(vma_policy(mpnt));
289                 retval = PTR_ERR(pol);
290                 if (IS_ERR(pol))
291                         goto fail_nomem_policy;
292                 vma_set_policy(tmp, pol);
293                 tmp->vm_flags &= ~VM_LOCKED;
294                 tmp->vm_mm = mm;
295                 tmp->vm_next = NULL;
296                 anon_vma_link(tmp);
297                 file = tmp->vm_file;
298                 if (file) {
299                         struct inode *inode = file->f_path.dentry->d_inode;
300                         get_file(file);
301                         if (tmp->vm_flags & VM_DENYWRITE)
302                                 atomic_dec(&inode->i_writecount);
303
304                         /* insert tmp into the share list, just after mpnt */
305                         spin_lock(&file->f_mapping->i_mmap_lock);
306                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
307                         flush_dcache_mmap_lock(file->f_mapping);
308                         vma_prio_tree_add(tmp, mpnt);
309                         flush_dcache_mmap_unlock(file->f_mapping);
310                         spin_unlock(&file->f_mapping->i_mmap_lock);
311                 }
312
313                 /*
314                  * Link in the new vma and copy the page table entries.
315                  */
316                 *pprev = tmp;
317                 pprev = &tmp->vm_next;
318
319                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
320                 rb_link = &tmp->vm_rb.rb_right;
321                 rb_parent = &tmp->vm_rb;
322
323                 mm->map_count++;
324                 retval = copy_page_range(mm, oldmm, mpnt);
325
326                 if (tmp->vm_ops && tmp->vm_ops->open)
327                         tmp->vm_ops->open(tmp);
328
329                 if (retval)
330                         goto out;
331         }
332         /* a new mm has just been created */
333         arch_dup_mmap(oldmm, mm);
334         retval = 0;
335 out:
336         up_write(&mm->mmap_sem);
337         flush_tlb_mm(oldmm);
338         up_write(&oldmm->mmap_sem);
339         return retval;
340 fail_nomem_policy:
341         kmem_cache_free(vm_area_cachep, tmp);
342 fail_nomem:
343         retval = -ENOMEM;
344         vm_unacct_memory(charge);
345         goto out;
346 }
347
348 static inline int mm_alloc_pgd(struct mm_struct * mm)
349 {
350         mm->pgd = pgd_alloc(mm);
351         if (unlikely(!mm->pgd))
352                 return -ENOMEM;
353         return 0;
354 }
355
356 static inline void mm_free_pgd(struct mm_struct * mm)
357 {
358         pgd_free(mm, mm->pgd);
359 }
360 #else
361 #define dup_mmap(mm, oldmm)     (0)
362 #define mm_alloc_pgd(mm)        (0)
363 #define mm_free_pgd(mm)
364 #endif /* CONFIG_MMU */
365
366 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
367
368 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
369 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
370
371 #include <linux/init_task.h>
372
373 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
374 {
375         atomic_set(&mm->mm_users, 1);
376         atomic_set(&mm->mm_count, 1);
377         init_rwsem(&mm->mmap_sem);
378         INIT_LIST_HEAD(&mm->mmlist);
379         mm->flags = (current->mm) ? current->mm->flags
380                                   : MMF_DUMP_FILTER_DEFAULT;
381         mm->core_waiters = 0;
382         mm->nr_ptes = 0;
383         set_mm_counter(mm, file_rss, 0);
384         set_mm_counter(mm, anon_rss, 0);
385         spin_lock_init(&mm->page_table_lock);
386         rwlock_init(&mm->ioctx_list_lock);
387         mm->ioctx_list = NULL;
388         mm->free_area_cache = TASK_UNMAPPED_BASE;
389         mm->cached_hole_size = ~0UL;
390         mm_init_owner(mm, p);
391
392         if (likely(!mm_alloc_pgd(mm))) {
393                 mm->def_flags = 0;
394                 return mm;
395         }
396
397         free_mm(mm);
398         return NULL;
399 }
400
401 /*
402  * Allocate and initialize an mm_struct.
403  */
404 struct mm_struct * mm_alloc(void)
405 {
406         struct mm_struct * mm;
407
408         mm = allocate_mm();
409         if (mm) {
410                 memset(mm, 0, sizeof(*mm));
411                 mm = mm_init(mm, current);
412         }
413         return mm;
414 }
415
416 /*
417  * Called when the last reference to the mm
418  * is dropped: either by a lazy thread or by
419  * mmput. Free the page directory and the mm.
420  */
421 void __mmdrop(struct mm_struct *mm)
422 {
423         BUG_ON(mm == &init_mm);
424         mm_free_pgd(mm);
425         destroy_context(mm);
426         free_mm(mm);
427 }
428 EXPORT_SYMBOL_GPL(__mmdrop);
429
430 /*
431  * Decrement the use count and release all resources for an mm.
432  */
433 void mmput(struct mm_struct *mm)
434 {
435         might_sleep();
436
437         if (atomic_dec_and_test(&mm->mm_users)) {
438                 exit_aio(mm);
439                 exit_mmap(mm);
440                 set_mm_exe_file(mm, NULL);
441                 if (!list_empty(&mm->mmlist)) {
442                         spin_lock(&mmlist_lock);
443                         list_del(&mm->mmlist);
444                         spin_unlock(&mmlist_lock);
445                 }
446                 put_swap_token(mm);
447                 mmdrop(mm);
448         }
449 }
450 EXPORT_SYMBOL_GPL(mmput);
451
452 /**
453  * get_task_mm - acquire a reference to the task's mm
454  *
455  * Returns %NULL if the task has no mm.  Checks PF_BORROWED_MM (meaning
456  * this kernel workthread has transiently adopted a user mm with use_mm,
457  * to do its AIO) is not set and if so returns a reference to it, after
458  * bumping up the use count.  User must release the mm via mmput()
459  * after use.  Typically used by /proc and ptrace.
460  */
461 struct mm_struct *get_task_mm(struct task_struct *task)
462 {
463         struct mm_struct *mm;
464
465         task_lock(task);
466         mm = task->mm;
467         if (mm) {
468                 if (task->flags & PF_BORROWED_MM)
469                         mm = NULL;
470                 else
471                         atomic_inc(&mm->mm_users);
472         }
473         task_unlock(task);
474         return mm;
475 }
476 EXPORT_SYMBOL_GPL(get_task_mm);
477
478 /* Please note the differences between mmput and mm_release.
479  * mmput is called whenever we stop holding onto a mm_struct,
480  * error success whatever.
481  *
482  * mm_release is called after a mm_struct has been removed
483  * from the current process.
484  *
485  * This difference is important for error handling, when we
486  * only half set up a mm_struct for a new process and need to restore
487  * the old one.  Because we mmput the new mm_struct before
488  * restoring the old one. . .
489  * Eric Biederman 10 January 1998
490  */
491 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
492 {
493         struct completion *vfork_done = tsk->vfork_done;
494
495         /* Get rid of any cached register state */
496         deactivate_mm(tsk, mm);
497
498         /* notify parent sleeping on vfork() */
499         if (vfork_done) {
500                 tsk->vfork_done = NULL;
501                 complete(vfork_done);
502         }
503
504         /*
505          * If we're exiting normally, clear a user-space tid field if
506          * requested.  We leave this alone when dying by signal, to leave
507          * the value intact in a core dump, and to save the unnecessary
508          * trouble otherwise.  Userland only wants this done for a sys_exit.
509          */
510         if (tsk->clear_child_tid
511             && !(tsk->flags & PF_SIGNALED)
512             && atomic_read(&mm->mm_users) > 1) {
513                 u32 __user * tidptr = tsk->clear_child_tid;
514                 tsk->clear_child_tid = NULL;
515
516                 /*
517                  * We don't check the error code - if userspace has
518                  * not set up a proper pointer then tough luck.
519                  */
520                 put_user(0, tidptr);
521                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
522         }
523 }
524
525 /*
526  * Allocate a new mm structure and copy contents from the
527  * mm structure of the passed in task structure.
528  */
529 struct mm_struct *dup_mm(struct task_struct *tsk)
530 {
531         struct mm_struct *mm, *oldmm = current->mm;
532         int err;
533
534         if (!oldmm)
535                 return NULL;
536
537         mm = allocate_mm();
538         if (!mm)
539                 goto fail_nomem;
540
541         memcpy(mm, oldmm, sizeof(*mm));
542
543         /* Initializing for Swap token stuff */
544         mm->token_priority = 0;
545         mm->last_interval = 0;
546
547         if (!mm_init(mm, tsk))
548                 goto fail_nomem;
549
550         if (init_new_context(tsk, mm))
551                 goto fail_nocontext;
552
553         dup_mm_exe_file(oldmm, mm);
554
555         err = dup_mmap(mm, oldmm);
556         if (err)
557                 goto free_pt;
558
559         mm->hiwater_rss = get_mm_rss(mm);
560         mm->hiwater_vm = mm->total_vm;
561
562         return mm;
563
564 free_pt:
565         mmput(mm);
566
567 fail_nomem:
568         return NULL;
569
570 fail_nocontext:
571         /*
572          * If init_new_context() failed, we cannot use mmput() to free the mm
573          * because it calls destroy_context()
574          */
575         mm_free_pgd(mm);
576         free_mm(mm);
577         return NULL;
578 }
579
580 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
581 {
582         struct mm_struct * mm, *oldmm;
583         int retval;
584
585         tsk->min_flt = tsk->maj_flt = 0;
586         tsk->nvcsw = tsk->nivcsw = 0;
587
588         tsk->mm = NULL;
589         tsk->active_mm = NULL;
590
591         /*
592          * Are we cloning a kernel thread?
593          *
594          * We need to steal a active VM for that..
595          */
596         oldmm = current->mm;
597         if (!oldmm)
598                 return 0;
599
600         if (clone_flags & CLONE_VM) {
601                 atomic_inc(&oldmm->mm_users);
602                 mm = oldmm;
603                 goto good_mm;
604         }
605
606         retval = -ENOMEM;
607         mm = dup_mm(tsk);
608         if (!mm)
609                 goto fail_nomem;
610
611 good_mm:
612         /* Initializing for Swap token stuff */
613         mm->token_priority = 0;
614         mm->last_interval = 0;
615
616         tsk->mm = mm;
617         tsk->active_mm = mm;
618         return 0;
619
620 fail_nomem:
621         return retval;
622 }
623
624 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
625 {
626         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
627         /* We don't need to lock fs - think why ;-) */
628         if (fs) {
629                 atomic_set(&fs->count, 1);
630                 rwlock_init(&fs->lock);
631                 fs->umask = old->umask;
632                 read_lock(&old->lock);
633                 fs->root = old->root;
634                 path_get(&old->root);
635                 fs->pwd = old->pwd;
636                 path_get(&old->pwd);
637                 if (old->altroot.dentry) {
638                         fs->altroot = old->altroot;
639                         path_get(&old->altroot);
640                 } else {
641                         fs->altroot.mnt = NULL;
642                         fs->altroot.dentry = NULL;
643                 }
644                 read_unlock(&old->lock);
645         }
646         return fs;
647 }
648
649 struct fs_struct *copy_fs_struct(struct fs_struct *old)
650 {
651         return __copy_fs_struct(old);
652 }
653
654 EXPORT_SYMBOL_GPL(copy_fs_struct);
655
656 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
657 {
658         if (clone_flags & CLONE_FS) {
659                 atomic_inc(&current->fs->count);
660                 return 0;
661         }
662         tsk->fs = __copy_fs_struct(current->fs);
663         if (!tsk->fs)
664                 return -ENOMEM;
665         return 0;
666 }
667
668 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
669 {
670         struct files_struct *oldf, *newf;
671         int error = 0;
672
673         /*
674          * A background process may not have any files ...
675          */
676         oldf = current->files;
677         if (!oldf)
678                 goto out;
679
680         if (clone_flags & CLONE_FILES) {
681                 atomic_inc(&oldf->count);
682                 goto out;
683         }
684
685         newf = dup_fd(oldf, &error);
686         if (!newf)
687                 goto out;
688
689         tsk->files = newf;
690         error = 0;
691 out:
692         return error;
693 }
694
695 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
696 {
697 #ifdef CONFIG_BLOCK
698         struct io_context *ioc = current->io_context;
699
700         if (!ioc)
701                 return 0;
702         /*
703          * Share io context with parent, if CLONE_IO is set
704          */
705         if (clone_flags & CLONE_IO) {
706                 tsk->io_context = ioc_task_link(ioc);
707                 if (unlikely(!tsk->io_context))
708                         return -ENOMEM;
709         } else if (ioprio_valid(ioc->ioprio)) {
710                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
711                 if (unlikely(!tsk->io_context))
712                         return -ENOMEM;
713
714                 tsk->io_context->ioprio = ioc->ioprio;
715         }
716 #endif
717         return 0;
718 }
719
720 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
721 {
722         struct sighand_struct *sig;
723
724         if (clone_flags & (CLONE_SIGHAND | CLONE_THREAD)) {
725                 atomic_inc(&current->sighand->count);
726                 return 0;
727         }
728         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
729         rcu_assign_pointer(tsk->sighand, sig);
730         if (!sig)
731                 return -ENOMEM;
732         atomic_set(&sig->count, 1);
733         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
734         return 0;
735 }
736
737 void __cleanup_sighand(struct sighand_struct *sighand)
738 {
739         if (atomic_dec_and_test(&sighand->count))
740                 kmem_cache_free(sighand_cachep, sighand);
741 }
742
743 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
744 {
745         struct signal_struct *sig;
746         int ret;
747
748         if (clone_flags & CLONE_THREAD) {
749                 atomic_inc(&current->signal->count);
750                 atomic_inc(&current->signal->live);
751                 return 0;
752         }
753         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
754         tsk->signal = sig;
755         if (!sig)
756                 return -ENOMEM;
757
758         ret = copy_thread_group_keys(tsk);
759         if (ret < 0) {
760                 kmem_cache_free(signal_cachep, sig);
761                 return ret;
762         }
763
764         atomic_set(&sig->count, 1);
765         atomic_set(&sig->live, 1);
766         init_waitqueue_head(&sig->wait_chldexit);
767         sig->flags = 0;
768         sig->group_exit_code = 0;
769         sig->group_exit_task = NULL;
770         sig->group_stop_count = 0;
771         sig->curr_target = tsk;
772         init_sigpending(&sig->shared_pending);
773         INIT_LIST_HEAD(&sig->posix_timers);
774
775         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
776         sig->it_real_incr.tv64 = 0;
777         sig->real_timer.function = it_real_fn;
778
779         sig->it_virt_expires = cputime_zero;
780         sig->it_virt_incr = cputime_zero;
781         sig->it_prof_expires = cputime_zero;
782         sig->it_prof_incr = cputime_zero;
783
784         sig->leader = 0;        /* session leadership doesn't inherit */
785         sig->tty_old_pgrp = NULL;
786
787         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
788         sig->gtime = cputime_zero;
789         sig->cgtime = cputime_zero;
790         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
791         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
792         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
793         sig->sum_sched_runtime = 0;
794         INIT_LIST_HEAD(&sig->cpu_timers[0]);
795         INIT_LIST_HEAD(&sig->cpu_timers[1]);
796         INIT_LIST_HEAD(&sig->cpu_timers[2]);
797         taskstats_tgid_init(sig);
798
799         task_lock(current->group_leader);
800         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
801         task_unlock(current->group_leader);
802
803         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
804                 /*
805                  * New sole thread in the process gets an expiry time
806                  * of the whole CPU time limit.
807                  */
808                 tsk->it_prof_expires =
809                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
810         }
811         acct_init_pacct(&sig->pacct);
812
813         tty_audit_fork(sig);
814
815         return 0;
816 }
817
818 void __cleanup_signal(struct signal_struct *sig)
819 {
820         exit_thread_group_keys(sig);
821         kmem_cache_free(signal_cachep, sig);
822 }
823
824 static void cleanup_signal(struct task_struct *tsk)
825 {
826         struct signal_struct *sig = tsk->signal;
827
828         atomic_dec(&sig->live);
829
830         if (atomic_dec_and_test(&sig->count))
831                 __cleanup_signal(sig);
832 }
833
834 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
835 {
836         unsigned long new_flags = p->flags;
837
838         new_flags &= ~PF_SUPERPRIV;
839         new_flags |= PF_FORKNOEXEC;
840         if (!(clone_flags & CLONE_PTRACE))
841                 p->ptrace = 0;
842         p->flags = new_flags;
843         clear_freeze_flag(p);
844 }
845
846 asmlinkage long sys_set_tid_address(int __user *tidptr)
847 {
848         current->clear_child_tid = tidptr;
849
850         return task_pid_vnr(current);
851 }
852
853 static void rt_mutex_init_task(struct task_struct *p)
854 {
855         spin_lock_init(&p->pi_lock);
856 #ifdef CONFIG_RT_MUTEXES
857         plist_head_init(&p->pi_waiters, &p->pi_lock);
858         p->pi_blocked_on = NULL;
859 #endif
860 }
861
862 #ifdef CONFIG_MM_OWNER
863 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
864 {
865         mm->owner = p;
866 }
867 #endif /* CONFIG_MM_OWNER */
868
869 /*
870  * This creates a new process as a copy of the old one,
871  * but does not actually start it yet.
872  *
873  * It copies the registers, and all the appropriate
874  * parts of the process environment (as per the clone
875  * flags). The actual kick-off is left to the caller.
876  */
877 static struct task_struct *copy_process(unsigned long clone_flags,
878                                         unsigned long stack_start,
879                                         struct pt_regs *regs,
880                                         unsigned long stack_size,
881                                         int __user *child_tidptr,
882                                         struct pid *pid)
883 {
884         int retval;
885         struct task_struct *p;
886         int cgroup_callbacks_done = 0;
887
888         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
889                 return ERR_PTR(-EINVAL);
890
891         /*
892          * Thread groups must share signals as well, and detached threads
893          * can only be started up within the thread group.
894          */
895         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
896                 return ERR_PTR(-EINVAL);
897
898         /*
899          * Shared signal handlers imply shared VM. By way of the above,
900          * thread groups also imply shared VM. Blocking this case allows
901          * for various simplifications in other code.
902          */
903         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
904                 return ERR_PTR(-EINVAL);
905
906         retval = security_task_create(clone_flags);
907         if (retval)
908                 goto fork_out;
909
910         retval = -ENOMEM;
911         p = dup_task_struct(current);
912         if (!p)
913                 goto fork_out;
914
915         rt_mutex_init_task(p);
916
917 #ifdef CONFIG_TRACE_IRQFLAGS
918         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
919         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
920 #endif
921         retval = -EAGAIN;
922         if (atomic_read(&p->user->processes) >=
923                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
924                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
925                     p->user != current->nsproxy->user_ns->root_user)
926                         goto bad_fork_free;
927         }
928
929         atomic_inc(&p->user->__count);
930         atomic_inc(&p->user->processes);
931         get_group_info(p->group_info);
932
933         /*
934          * If multiple threads are within copy_process(), then this check
935          * triggers too late. This doesn't hurt, the check is only there
936          * to stop root fork bombs.
937          */
938         if (nr_threads >= max_threads)
939                 goto bad_fork_cleanup_count;
940
941         if (!try_module_get(task_thread_info(p)->exec_domain->module))
942                 goto bad_fork_cleanup_count;
943
944         if (p->binfmt && !try_module_get(p->binfmt->module))
945                 goto bad_fork_cleanup_put_domain;
946
947         p->did_exec = 0;
948         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
949         copy_flags(clone_flags, p);
950         INIT_LIST_HEAD(&p->children);
951         INIT_LIST_HEAD(&p->sibling);
952 #ifdef CONFIG_PREEMPT_RCU
953         p->rcu_read_lock_nesting = 0;
954         p->rcu_flipctr_idx = 0;
955 #endif /* #ifdef CONFIG_PREEMPT_RCU */
956         p->vfork_done = NULL;
957         spin_lock_init(&p->alloc_lock);
958
959         clear_tsk_thread_flag(p, TIF_SIGPENDING);
960         init_sigpending(&p->pending);
961
962         p->utime = cputime_zero;
963         p->stime = cputime_zero;
964         p->gtime = cputime_zero;
965         p->utimescaled = cputime_zero;
966         p->stimescaled = cputime_zero;
967         p->prev_utime = cputime_zero;
968         p->prev_stime = cputime_zero;
969
970 #ifdef CONFIG_DETECT_SOFTLOCKUP
971         p->last_switch_count = 0;
972         p->last_switch_timestamp = 0;
973 #endif
974
975 #ifdef CONFIG_TASK_XACCT
976         p->rchar = 0;           /* I/O counter: bytes read */
977         p->wchar = 0;           /* I/O counter: bytes written */
978         p->syscr = 0;           /* I/O counter: read syscalls */
979         p->syscw = 0;           /* I/O counter: write syscalls */
980 #endif
981         task_io_accounting_init(p);
982         acct_clear_integrals(p);
983
984         p->it_virt_expires = cputime_zero;
985         p->it_prof_expires = cputime_zero;
986         p->it_sched_expires = 0;
987         INIT_LIST_HEAD(&p->cpu_timers[0]);
988         INIT_LIST_HEAD(&p->cpu_timers[1]);
989         INIT_LIST_HEAD(&p->cpu_timers[2]);
990
991         p->lock_depth = -1;             /* -1 = no lock */
992         do_posix_clock_monotonic_gettime(&p->start_time);
993         p->real_start_time = p->start_time;
994         monotonic_to_bootbased(&p->real_start_time);
995 #ifdef CONFIG_SECURITY
996         p->security = NULL;
997 #endif
998         p->cap_bset = current->cap_bset;
999         p->io_context = NULL;
1000         p->audit_context = NULL;
1001         cgroup_fork(p);
1002 #ifdef CONFIG_NUMA
1003         p->mempolicy = mpol_dup(p->mempolicy);
1004         if (IS_ERR(p->mempolicy)) {
1005                 retval = PTR_ERR(p->mempolicy);
1006                 p->mempolicy = NULL;
1007                 goto bad_fork_cleanup_cgroup;
1008         }
1009         mpol_fix_fork_child_flag(p);
1010 #endif
1011 #ifdef CONFIG_TRACE_IRQFLAGS
1012         p->irq_events = 0;
1013 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1014         p->hardirqs_enabled = 1;
1015 #else
1016         p->hardirqs_enabled = 0;
1017 #endif
1018         p->hardirq_enable_ip = 0;
1019         p->hardirq_enable_event = 0;
1020         p->hardirq_disable_ip = _THIS_IP_;
1021         p->hardirq_disable_event = 0;
1022         p->softirqs_enabled = 1;
1023         p->softirq_enable_ip = _THIS_IP_;
1024         p->softirq_enable_event = 0;
1025         p->softirq_disable_ip = 0;
1026         p->softirq_disable_event = 0;
1027         p->hardirq_context = 0;
1028         p->softirq_context = 0;
1029 #endif
1030 #ifdef CONFIG_LOCKDEP
1031         p->lockdep_depth = 0; /* no locks held yet */
1032         p->curr_chain_key = 0;
1033         p->lockdep_recursion = 0;
1034 #endif
1035
1036 #ifdef CONFIG_DEBUG_MUTEXES
1037         p->blocked_on = NULL; /* not blocked yet */
1038 #endif
1039
1040         /* Perform scheduler related setup. Assign this task to a CPU. */
1041         sched_fork(p, clone_flags);
1042
1043         if ((retval = security_task_alloc(p)))
1044                 goto bad_fork_cleanup_policy;
1045         if ((retval = audit_alloc(p)))
1046                 goto bad_fork_cleanup_security;
1047         /* copy all the process information */
1048         if ((retval = copy_semundo(clone_flags, p)))
1049                 goto bad_fork_cleanup_audit;
1050         if ((retval = copy_files(clone_flags, p)))
1051                 goto bad_fork_cleanup_semundo;
1052         if ((retval = copy_fs(clone_flags, p)))
1053                 goto bad_fork_cleanup_files;
1054         if ((retval = copy_sighand(clone_flags, p)))
1055                 goto bad_fork_cleanup_fs;
1056         if ((retval = copy_signal(clone_flags, p)))
1057                 goto bad_fork_cleanup_sighand;
1058         if ((retval = copy_mm(clone_flags, p)))
1059                 goto bad_fork_cleanup_signal;
1060         if ((retval = copy_keys(clone_flags, p)))
1061                 goto bad_fork_cleanup_mm;
1062         if ((retval = copy_namespaces(clone_flags, p)))
1063                 goto bad_fork_cleanup_keys;
1064         if ((retval = copy_io(clone_flags, p)))
1065                 goto bad_fork_cleanup_namespaces;
1066         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1067         if (retval)
1068                 goto bad_fork_cleanup_io;
1069
1070         if (pid != &init_struct_pid) {
1071                 retval = -ENOMEM;
1072                 pid = alloc_pid(task_active_pid_ns(p));
1073                 if (!pid)
1074                         goto bad_fork_cleanup_io;
1075
1076                 if (clone_flags & CLONE_NEWPID) {
1077                         retval = pid_ns_prepare_proc(task_active_pid_ns(p));
1078                         if (retval < 0)
1079                                 goto bad_fork_free_pid;
1080                 }
1081         }
1082
1083         p->pid = pid_nr(pid);
1084         p->tgid = p->pid;
1085         if (clone_flags & CLONE_THREAD)
1086                 p->tgid = current->tgid;
1087
1088         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1089         /*
1090          * Clear TID on mm_release()?
1091          */
1092         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1093 #ifdef CONFIG_FUTEX
1094         p->robust_list = NULL;
1095 #ifdef CONFIG_COMPAT
1096         p->compat_robust_list = NULL;
1097 #endif
1098         INIT_LIST_HEAD(&p->pi_state_list);
1099         p->pi_state_cache = NULL;
1100 #endif
1101         /*
1102          * sigaltstack should be cleared when sharing the same VM
1103          */
1104         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1105                 p->sas_ss_sp = p->sas_ss_size = 0;
1106
1107         /*
1108          * Syscall tracing should be turned off in the child regardless
1109          * of CLONE_PTRACE.
1110          */
1111         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1112 #ifdef TIF_SYSCALL_EMU
1113         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1114 #endif
1115         clear_all_latency_tracing(p);
1116
1117         /* Our parent execution domain becomes current domain
1118            These must match for thread signalling to apply */
1119         p->parent_exec_id = p->self_exec_id;
1120
1121         /* ok, now we should be set up.. */
1122         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1123         p->pdeath_signal = 0;
1124         p->exit_state = 0;
1125
1126         /*
1127          * Ok, make it visible to the rest of the system.
1128          * We dont wake it up yet.
1129          */
1130         p->group_leader = p;
1131         INIT_LIST_HEAD(&p->thread_group);
1132         INIT_LIST_HEAD(&p->ptrace_children);
1133         INIT_LIST_HEAD(&p->ptrace_list);
1134
1135         /* Now that the task is set up, run cgroup callbacks if
1136          * necessary. We need to run them before the task is visible
1137          * on the tasklist. */
1138         cgroup_fork_callbacks(p);
1139         cgroup_callbacks_done = 1;
1140
1141         /* Need tasklist lock for parent etc handling! */
1142         write_lock_irq(&tasklist_lock);
1143
1144         /*
1145          * The task hasn't been attached yet, so its cpus_allowed mask will
1146          * not be changed, nor will its assigned CPU.
1147          *
1148          * The cpus_allowed mask of the parent may have changed after it was
1149          * copied first time - so re-copy it here, then check the child's CPU
1150          * to ensure it is on a valid CPU (and if not, just force it back to
1151          * parent's CPU). This avoids alot of nasty races.
1152          */
1153         p->cpus_allowed = current->cpus_allowed;
1154         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1155         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1156                         !cpu_online(task_cpu(p))))
1157                 set_task_cpu(p, smp_processor_id());
1158
1159         /* CLONE_PARENT re-uses the old parent */
1160         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1161                 p->real_parent = current->real_parent;
1162         else
1163                 p->real_parent = current;
1164         p->parent = p->real_parent;
1165
1166         spin_lock(&current->sighand->siglock);
1167
1168         /*
1169          * Process group and session signals need to be delivered to just the
1170          * parent before the fork or both the parent and the child after the
1171          * fork. Restart if a signal comes in before we add the new process to
1172          * it's process group.
1173          * A fatal signal pending means that current will exit, so the new
1174          * thread can't slip out of an OOM kill (or normal SIGKILL).
1175          */
1176         recalc_sigpending();
1177         if (signal_pending(current)) {
1178                 spin_unlock(&current->sighand->siglock);
1179                 write_unlock_irq(&tasklist_lock);
1180                 retval = -ERESTARTNOINTR;
1181                 goto bad_fork_free_pid;
1182         }
1183
1184         if (clone_flags & CLONE_THREAD) {
1185                 p->group_leader = current->group_leader;
1186                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1187
1188                 if (!cputime_eq(current->signal->it_virt_expires,
1189                                 cputime_zero) ||
1190                     !cputime_eq(current->signal->it_prof_expires,
1191                                 cputime_zero) ||
1192                     current->signal->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY ||
1193                     !list_empty(&current->signal->cpu_timers[0]) ||
1194                     !list_empty(&current->signal->cpu_timers[1]) ||
1195                     !list_empty(&current->signal->cpu_timers[2])) {
1196                         /*
1197                          * Have child wake up on its first tick to check
1198                          * for process CPU timers.
1199                          */
1200                         p->it_prof_expires = jiffies_to_cputime(1);
1201                 }
1202         }
1203
1204         if (likely(p->pid)) {
1205                 add_parent(p);
1206                 if (unlikely(p->ptrace & PT_PTRACED))
1207                         __ptrace_link(p, current->parent);
1208
1209                 if (thread_group_leader(p)) {
1210                         if (clone_flags & CLONE_NEWPID)
1211                                 p->nsproxy->pid_ns->child_reaper = p;
1212
1213                         p->signal->leader_pid = pid;
1214                         p->signal->tty = current->signal->tty;
1215                         set_task_pgrp(p, task_pgrp_nr(current));
1216                         set_task_session(p, task_session_nr(current));
1217                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1218                         attach_pid(p, PIDTYPE_SID, task_session(current));
1219                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1220                         __get_cpu_var(process_counts)++;
1221                 }
1222                 attach_pid(p, PIDTYPE_PID, pid);
1223                 nr_threads++;
1224         }
1225
1226         total_forks++;
1227         spin_unlock(&current->sighand->siglock);
1228         write_unlock_irq(&tasklist_lock);
1229         proc_fork_connector(p);
1230         cgroup_post_fork(p);
1231         return p;
1232
1233 bad_fork_free_pid:
1234         if (pid != &init_struct_pid)
1235                 free_pid(pid);
1236 bad_fork_cleanup_io:
1237         put_io_context(p->io_context);
1238 bad_fork_cleanup_namespaces:
1239         exit_task_namespaces(p);
1240 bad_fork_cleanup_keys:
1241         exit_keys(p);
1242 bad_fork_cleanup_mm:
1243         if (p->mm)
1244                 mmput(p->mm);
1245 bad_fork_cleanup_signal:
1246         cleanup_signal(p);
1247 bad_fork_cleanup_sighand:
1248         __cleanup_sighand(p->sighand);
1249 bad_fork_cleanup_fs:
1250         exit_fs(p); /* blocking */
1251 bad_fork_cleanup_files:
1252         exit_files(p); /* blocking */
1253 bad_fork_cleanup_semundo:
1254         exit_sem(p);
1255 bad_fork_cleanup_audit:
1256         audit_free(p);
1257 bad_fork_cleanup_security:
1258         security_task_free(p);
1259 bad_fork_cleanup_policy:
1260 #ifdef CONFIG_NUMA
1261         mpol_put(p->mempolicy);
1262 bad_fork_cleanup_cgroup:
1263 #endif
1264         cgroup_exit(p, cgroup_callbacks_done);
1265         delayacct_tsk_free(p);
1266         if (p->binfmt)
1267                 module_put(p->binfmt->module);
1268 bad_fork_cleanup_put_domain:
1269         module_put(task_thread_info(p)->exec_domain->module);
1270 bad_fork_cleanup_count:
1271         put_group_info(p->group_info);
1272         atomic_dec(&p->user->processes);
1273         free_uid(p->user);
1274 bad_fork_free:
1275         free_task(p);
1276 fork_out:
1277         return ERR_PTR(retval);
1278 }
1279
1280 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1281 {
1282         memset(regs, 0, sizeof(struct pt_regs));
1283         return regs;
1284 }
1285
1286 struct task_struct * __cpuinit fork_idle(int cpu)
1287 {
1288         struct task_struct *task;
1289         struct pt_regs regs;
1290
1291         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1292                                 &init_struct_pid);
1293         if (!IS_ERR(task))
1294                 init_idle(task, cpu);
1295
1296         return task;
1297 }
1298
1299 static int fork_traceflag(unsigned clone_flags)
1300 {
1301         if (clone_flags & CLONE_UNTRACED)
1302                 return 0;
1303         else if (clone_flags & CLONE_VFORK) {
1304                 if (current->ptrace & PT_TRACE_VFORK)
1305                         return PTRACE_EVENT_VFORK;
1306         } else if ((clone_flags & CSIGNAL) != SIGCHLD) {
1307                 if (current->ptrace & PT_TRACE_CLONE)
1308                         return PTRACE_EVENT_CLONE;
1309         } else if (current->ptrace & PT_TRACE_FORK)
1310                 return PTRACE_EVENT_FORK;
1311
1312         return 0;
1313 }
1314
1315 /*
1316  *  Ok, this is the main fork-routine.
1317  *
1318  * It copies the process, and if successful kick-starts
1319  * it and waits for it to finish using the VM if required.
1320  */
1321 long do_fork(unsigned long clone_flags,
1322               unsigned long stack_start,
1323               struct pt_regs *regs,
1324               unsigned long stack_size,
1325               int __user *parent_tidptr,
1326               int __user *child_tidptr)
1327 {
1328         struct task_struct *p;
1329         int trace = 0;
1330         long nr;
1331
1332         /*
1333          * We hope to recycle these flags after 2.6.26
1334          */
1335         if (unlikely(clone_flags & CLONE_STOPPED)) {
1336                 static int __read_mostly count = 100;
1337
1338                 if (count > 0 && printk_ratelimit()) {
1339                         char comm[TASK_COMM_LEN];
1340
1341                         count--;
1342                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1343                                         "clone flags 0x%lx\n",
1344                                 get_task_comm(comm, current),
1345                                 clone_flags & CLONE_STOPPED);
1346                 }
1347         }
1348
1349         if (unlikely(current->ptrace)) {
1350                 trace = fork_traceflag (clone_flags);
1351                 if (trace)
1352                         clone_flags |= CLONE_PTRACE;
1353         }
1354
1355         p = copy_process(clone_flags, stack_start, regs, stack_size,
1356                         child_tidptr, NULL);
1357         /*
1358          * Do this prior waking up the new thread - the thread pointer
1359          * might get invalid after that point, if the thread exits quickly.
1360          */
1361         if (!IS_ERR(p)) {
1362                 struct completion vfork;
1363
1364                 nr = task_pid_vnr(p);
1365
1366                 if (clone_flags & CLONE_PARENT_SETTID)
1367                         put_user(nr, parent_tidptr);
1368
1369                 if (clone_flags & CLONE_VFORK) {
1370                         p->vfork_done = &vfork;
1371                         init_completion(&vfork);
1372                 }
1373
1374                 if ((p->ptrace & PT_PTRACED) || (clone_flags & CLONE_STOPPED)) {
1375                         /*
1376                          * We'll start up with an immediate SIGSTOP.
1377                          */
1378                         sigaddset(&p->pending.signal, SIGSTOP);
1379                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1380                 }
1381
1382                 if (!(clone_flags & CLONE_STOPPED))
1383                         wake_up_new_task(p, clone_flags);
1384                 else
1385                         __set_task_state(p, TASK_STOPPED);
1386
1387                 if (unlikely (trace)) {
1388                         current->ptrace_message = nr;
1389                         ptrace_notify ((trace << 8) | SIGTRAP);
1390                 }
1391
1392                 if (clone_flags & CLONE_VFORK) {
1393                         freezer_do_not_count();
1394                         wait_for_completion(&vfork);
1395                         freezer_count();
1396                         if (unlikely (current->ptrace & PT_TRACE_VFORK_DONE)) {
1397                                 current->ptrace_message = nr;
1398                                 ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
1399                         }
1400                 }
1401         } else {
1402                 nr = PTR_ERR(p);
1403         }
1404         return nr;
1405 }
1406
1407 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1408 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1409 #endif
1410
1411 static void sighand_ctor(struct kmem_cache *cachep, void *data)
1412 {
1413         struct sighand_struct *sighand = data;
1414
1415         spin_lock_init(&sighand->siglock);
1416         init_waitqueue_head(&sighand->signalfd_wqh);
1417 }
1418
1419 void __init proc_caches_init(void)
1420 {
1421         sighand_cachep = kmem_cache_create("sighand_cache",
1422                         sizeof(struct sighand_struct), 0,
1423                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1424                         sighand_ctor);
1425         signal_cachep = kmem_cache_create("signal_cache",
1426                         sizeof(struct signal_struct), 0,
1427                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1428         files_cachep = kmem_cache_create("files_cache",
1429                         sizeof(struct files_struct), 0,
1430                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1431         fs_cachep = kmem_cache_create("fs_cache",
1432                         sizeof(struct fs_struct), 0,
1433                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1434         vm_area_cachep = kmem_cache_create("vm_area_struct",
1435                         sizeof(struct vm_area_struct), 0,
1436                         SLAB_PANIC, NULL);
1437         mm_cachep = kmem_cache_create("mm_struct",
1438                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1439                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1440 }
1441
1442 /*
1443  * Check constraints on flags passed to the unshare system call and
1444  * force unsharing of additional process context as appropriate.
1445  */
1446 static void check_unshare_flags(unsigned long *flags_ptr)
1447 {
1448         /*
1449          * If unsharing a thread from a thread group, must also
1450          * unshare vm.
1451          */
1452         if (*flags_ptr & CLONE_THREAD)
1453                 *flags_ptr |= CLONE_VM;
1454
1455         /*
1456          * If unsharing vm, must also unshare signal handlers.
1457          */
1458         if (*flags_ptr & CLONE_VM)
1459                 *flags_ptr |= CLONE_SIGHAND;
1460
1461         /*
1462          * If unsharing signal handlers and the task was created
1463          * using CLONE_THREAD, then must unshare the thread
1464          */
1465         if ((*flags_ptr & CLONE_SIGHAND) &&
1466             (atomic_read(&current->signal->count) > 1))
1467                 *flags_ptr |= CLONE_THREAD;
1468
1469         /*
1470          * If unsharing namespace, must also unshare filesystem information.
1471          */
1472         if (*flags_ptr & CLONE_NEWNS)
1473                 *flags_ptr |= CLONE_FS;
1474 }
1475
1476 /*
1477  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1478  */
1479 static int unshare_thread(unsigned long unshare_flags)
1480 {
1481         if (unshare_flags & CLONE_THREAD)
1482                 return -EINVAL;
1483
1484         return 0;
1485 }
1486
1487 /*
1488  * Unshare the filesystem structure if it is being shared
1489  */
1490 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1491 {
1492         struct fs_struct *fs = current->fs;
1493
1494         if ((unshare_flags & CLONE_FS) &&
1495             (fs && atomic_read(&fs->count) > 1)) {
1496                 *new_fsp = __copy_fs_struct(current->fs);
1497                 if (!*new_fsp)
1498                         return -ENOMEM;
1499         }
1500
1501         return 0;
1502 }
1503
1504 /*
1505  * Unsharing of sighand is not supported yet
1506  */
1507 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1508 {
1509         struct sighand_struct *sigh = current->sighand;
1510
1511         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1512                 return -EINVAL;
1513         else
1514                 return 0;
1515 }
1516
1517 /*
1518  * Unshare vm if it is being shared
1519  */
1520 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1521 {
1522         struct mm_struct *mm = current->mm;
1523
1524         if ((unshare_flags & CLONE_VM) &&
1525             (mm && atomic_read(&mm->mm_users) > 1)) {
1526                 return -EINVAL;
1527         }
1528
1529         return 0;
1530 }
1531
1532 /*
1533  * Unshare file descriptor table if it is being shared
1534  */
1535 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1536 {
1537         struct files_struct *fd = current->files;
1538         int error = 0;
1539
1540         if ((unshare_flags & CLONE_FILES) &&
1541             (fd && atomic_read(&fd->count) > 1)) {
1542                 *new_fdp = dup_fd(fd, &error);
1543                 if (!*new_fdp)
1544                         return error;
1545         }
1546
1547         return 0;
1548 }
1549
1550 /*
1551  * unshare allows a process to 'unshare' part of the process
1552  * context which was originally shared using clone.  copy_*
1553  * functions used by do_fork() cannot be used here directly
1554  * because they modify an inactive task_struct that is being
1555  * constructed. Here we are modifying the current, active,
1556  * task_struct.
1557  */
1558 asmlinkage long sys_unshare(unsigned long unshare_flags)
1559 {
1560         int err = 0;
1561         struct fs_struct *fs, *new_fs = NULL;
1562         struct sighand_struct *new_sigh = NULL;
1563         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1564         struct files_struct *fd, *new_fd = NULL;
1565         struct nsproxy *new_nsproxy = NULL;
1566         int do_sysvsem = 0;
1567
1568         check_unshare_flags(&unshare_flags);
1569
1570         /* Return -EINVAL for all unsupported flags */
1571         err = -EINVAL;
1572         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1573                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1574                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWUSER|
1575                                 CLONE_NEWNET))
1576                 goto bad_unshare_out;
1577
1578         /*
1579          * CLONE_NEWIPC must also detach from the undolist: after switching
1580          * to a new ipc namespace, the semaphore arrays from the old
1581          * namespace are unreachable.
1582          */
1583         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1584                 do_sysvsem = 1;
1585         if ((err = unshare_thread(unshare_flags)))
1586                 goto bad_unshare_out;
1587         if ((err = unshare_fs(unshare_flags, &new_fs)))
1588                 goto bad_unshare_cleanup_thread;
1589         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1590                 goto bad_unshare_cleanup_fs;
1591         if ((err = unshare_vm(unshare_flags, &new_mm)))
1592                 goto bad_unshare_cleanup_sigh;
1593         if ((err = unshare_fd(unshare_flags, &new_fd)))
1594                 goto bad_unshare_cleanup_vm;
1595         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1596                         new_fs)))
1597                 goto bad_unshare_cleanup_fd;
1598
1599         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1600                 if (do_sysvsem) {
1601                         /*
1602                          * CLONE_SYSVSEM is equivalent to sys_exit().
1603                          */
1604                         exit_sem(current);
1605                 }
1606
1607                 if (new_nsproxy) {
1608                         switch_task_namespaces(current, new_nsproxy);
1609                         new_nsproxy = NULL;
1610                 }
1611
1612                 task_lock(current);
1613
1614                 if (new_fs) {
1615                         fs = current->fs;
1616                         current->fs = new_fs;
1617                         new_fs = fs;
1618                 }
1619
1620                 if (new_mm) {
1621                         mm = current->mm;
1622                         active_mm = current->active_mm;
1623                         current->mm = new_mm;
1624                         current->active_mm = new_mm;
1625                         activate_mm(active_mm, new_mm);
1626                         new_mm = mm;
1627                 }
1628
1629                 if (new_fd) {
1630                         fd = current->files;
1631                         current->files = new_fd;
1632                         new_fd = fd;
1633                 }
1634
1635                 task_unlock(current);
1636         }
1637
1638         if (new_nsproxy)
1639                 put_nsproxy(new_nsproxy);
1640
1641 bad_unshare_cleanup_fd:
1642         if (new_fd)
1643                 put_files_struct(new_fd);
1644
1645 bad_unshare_cleanup_vm:
1646         if (new_mm)
1647                 mmput(new_mm);
1648
1649 bad_unshare_cleanup_sigh:
1650         if (new_sigh)
1651                 if (atomic_dec_and_test(&new_sigh->count))
1652                         kmem_cache_free(sighand_cachep, new_sigh);
1653
1654 bad_unshare_cleanup_fs:
1655         if (new_fs)
1656                 put_fs_struct(new_fs);
1657
1658 bad_unshare_cleanup_thread:
1659 bad_unshare_out:
1660         return err;
1661 }
1662
1663 /*
1664  *      Helper to unshare the files of the current task.
1665  *      We don't want to expose copy_files internals to
1666  *      the exec layer of the kernel.
1667  */
1668
1669 int unshare_files(struct files_struct **displaced)
1670 {
1671         struct task_struct *task = current;
1672         struct files_struct *copy = NULL;
1673         int error;
1674
1675         error = unshare_fd(CLONE_FILES, &copy);
1676         if (error || !copy) {
1677                 *displaced = NULL;
1678                 return error;
1679         }
1680         *displaced = task->files;
1681         task_lock(task);
1682         task->files = copy;
1683         task_unlock(task);
1684         return 0;
1685 }