signal: re-add dead task accumulation stats.
[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/iocontext.h>
27 #include <linux/key.h>
28 #include <linux/binfmts.h>
29 #include <linux/mman.h>
30 #include <linux/mmu_notifier.h>
31 #include <linux/fs.h>
32 #include <linux/nsproxy.h>
33 #include <linux/capability.h>
34 #include <linux/cpu.h>
35 #include <linux/cgroup.h>
36 #include <linux/security.h>
37 #include <linux/hugetlb.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/tracehook.h>
42 #include <linux/futex.h>
43 #include <linux/compat.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/profile.h>
52 #include <linux/rmap.h>
53 #include <linux/acct.h>
54 #include <linux/tsacct_kern.h>
55 #include <linux/cn_proc.h>
56 #include <linux/freezer.h>
57 #include <linux/delayacct.h>
58 #include <linux/taskstats_kern.h>
59 #include <linux/random.h>
60 #include <linux/tty.h>
61 #include <linux/proc_fs.h>
62 #include <linux/blkdev.h>
63 #include <trace/sched.h>
64
65 #include <asm/pgtable.h>
66 #include <asm/pgalloc.h>
67 #include <asm/uaccess.h>
68 #include <asm/mmu_context.h>
69 #include <asm/cacheflush.h>
70 #include <asm/tlbflush.h>
71
72 /*
73  * Protected counters by write_lock_irq(&tasklist_lock)
74  */
75 unsigned long total_forks;      /* Handle normal Linux uptimes. */
76 int nr_threads;                 /* The idle threads do not count.. */
77
78 int max_threads;                /* tunable limit on nr_threads */
79
80 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
81
82 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
83
84 DEFINE_TRACE(sched_process_fork);
85
86 int nr_processes(void)
87 {
88         int cpu;
89         int total = 0;
90
91         for_each_online_cpu(cpu)
92                 total += per_cpu(process_counts, cpu);
93
94         return total;
95 }
96
97 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
98 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
99 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
100 static struct kmem_cache *task_struct_cachep;
101 #endif
102
103 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
104 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
105 {
106 #ifdef CONFIG_DEBUG_STACK_USAGE
107         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
108 #else
109         gfp_t mask = GFP_KERNEL;
110 #endif
111         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
112 }
113
114 static inline void free_thread_info(struct thread_info *ti)
115 {
116         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
117 }
118 #endif
119
120 /* SLAB cache for signal_struct structures (tsk->signal) */
121 static struct kmem_cache *signal_cachep;
122
123 /* SLAB cache for sighand_struct structures (tsk->sighand) */
124 struct kmem_cache *sighand_cachep;
125
126 /* SLAB cache for files_struct structures (tsk->files) */
127 struct kmem_cache *files_cachep;
128
129 /* SLAB cache for fs_struct structures (tsk->fs) */
130 struct kmem_cache *fs_cachep;
131
132 /* SLAB cache for vm_area_struct structures */
133 struct kmem_cache *vm_area_cachep;
134
135 /* SLAB cache for mm_struct structures (tsk->mm) */
136 static struct kmem_cache *mm_cachep;
137
138 void free_task(struct task_struct *tsk)
139 {
140         prop_local_destroy_single(&tsk->dirties);
141         free_thread_info(tsk->stack);
142         rt_mutex_debug_task_free(tsk);
143         ftrace_graph_exit_task(tsk);
144         free_task_struct(tsk);
145 }
146 EXPORT_SYMBOL(free_task);
147
148 void __put_task_struct(struct task_struct *tsk)
149 {
150         WARN_ON(!tsk->exit_state);
151         WARN_ON(atomic_read(&tsk->usage));
152         WARN_ON(tsk == current);
153
154         put_cred(tsk->real_cred);
155         put_cred(tsk->cred);
156         delayacct_tsk_free(tsk);
157
158         if (!profile_handoff_task(tsk))
159                 free_task(tsk);
160 }
161
162 /*
163  * macro override instead of weak attribute alias, to workaround
164  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
165  */
166 #ifndef arch_task_cache_init
167 #define arch_task_cache_init()
168 #endif
169
170 void __init fork_init(unsigned long mempages)
171 {
172 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
173 #ifndef ARCH_MIN_TASKALIGN
174 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
175 #endif
176         /* create a slab on which task_structs can be allocated */
177         task_struct_cachep =
178                 kmem_cache_create("task_struct", sizeof(struct task_struct),
179                         ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
180 #endif
181
182         /* do the arch specific task caches init */
183         arch_task_cache_init();
184
185         /*
186          * The default maximum number of threads is set to a safe
187          * value: the thread structures can take up at most half
188          * of memory.
189          */
190         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
191
192         /*
193          * we need to allow at least 20 threads to boot a system
194          */
195         if(max_threads < 20)
196                 max_threads = 20;
197
198         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
199         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
200         init_task.signal->rlim[RLIMIT_SIGPENDING] =
201                 init_task.signal->rlim[RLIMIT_NPROC];
202 }
203
204 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
205                                                struct task_struct *src)
206 {
207         *dst = *src;
208         return 0;
209 }
210
211 static struct task_struct *dup_task_struct(struct task_struct *orig)
212 {
213         struct task_struct *tsk;
214         struct thread_info *ti;
215         int err;
216
217         prepare_to_copy(orig);
218
219         tsk = alloc_task_struct();
220         if (!tsk)
221                 return NULL;
222
223         ti = alloc_thread_info(tsk);
224         if (!ti) {
225                 free_task_struct(tsk);
226                 return NULL;
227         }
228
229         err = arch_dup_task_struct(tsk, orig);
230         if (err)
231                 goto out;
232
233         tsk->stack = ti;
234
235         err = prop_local_init_single(&tsk->dirties);
236         if (err)
237                 goto out;
238
239         setup_thread_stack(tsk, orig);
240
241 #ifdef CONFIG_CC_STACKPROTECTOR
242         tsk->stack_canary = get_random_int();
243 #endif
244
245         /* One for us, one for whoever does the "release_task()" (usually parent) */
246         atomic_set(&tsk->usage,2);
247         atomic_set(&tsk->fs_excl, 0);
248 #ifdef CONFIG_BLK_DEV_IO_TRACE
249         tsk->btrace_seq = 0;
250 #endif
251         tsk->splice_pipe = NULL;
252         return tsk;
253
254 out:
255         free_thread_info(ti);
256         free_task_struct(tsk);
257         return NULL;
258 }
259
260 #ifdef CONFIG_MMU
261 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
262 {
263         struct vm_area_struct *mpnt, *tmp, **pprev;
264         struct rb_node **rb_link, *rb_parent;
265         int retval;
266         unsigned long charge;
267         struct mempolicy *pol;
268
269         down_write(&oldmm->mmap_sem);
270         flush_cache_dup_mm(oldmm);
271         /*
272          * Not linked in yet - no deadlock potential:
273          */
274         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
275
276         mm->locked_vm = 0;
277         mm->mmap = NULL;
278         mm->mmap_cache = NULL;
279         mm->free_area_cache = oldmm->mmap_base;
280         mm->cached_hole_size = ~0UL;
281         mm->map_count = 0;
282         cpus_clear(mm->cpu_vm_mask);
283         mm->mm_rb = RB_ROOT;
284         rb_link = &mm->mm_rb.rb_node;
285         rb_parent = NULL;
286         pprev = &mm->mmap;
287
288         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
289                 struct file *file;
290
291                 if (mpnt->vm_flags & VM_DONTCOPY) {
292                         long pages = vma_pages(mpnt);
293                         mm->total_vm -= pages;
294                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
295                                                                 -pages);
296                         continue;
297                 }
298                 charge = 0;
299                 if (mpnt->vm_flags & VM_ACCOUNT) {
300                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
301                         if (security_vm_enough_memory(len))
302                                 goto fail_nomem;
303                         charge = len;
304                 }
305                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
306                 if (!tmp)
307                         goto fail_nomem;
308                 *tmp = *mpnt;
309                 pol = mpol_dup(vma_policy(mpnt));
310                 retval = PTR_ERR(pol);
311                 if (IS_ERR(pol))
312                         goto fail_nomem_policy;
313                 vma_set_policy(tmp, pol);
314                 tmp->vm_flags &= ~VM_LOCKED;
315                 tmp->vm_mm = mm;
316                 tmp->vm_next = NULL;
317                 anon_vma_link(tmp);
318                 file = tmp->vm_file;
319                 if (file) {
320                         struct inode *inode = file->f_path.dentry->d_inode;
321                         struct address_space *mapping = file->f_mapping;
322
323                         get_file(file);
324                         if (tmp->vm_flags & VM_DENYWRITE)
325                                 atomic_dec(&inode->i_writecount);
326                         spin_lock(&mapping->i_mmap_lock);
327                         if (tmp->vm_flags & VM_SHARED)
328                                 mapping->i_mmap_writable++;
329                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
330                         flush_dcache_mmap_lock(mapping);
331                         /* insert tmp into the share list, just after mpnt */
332                         vma_prio_tree_add(tmp, mpnt);
333                         flush_dcache_mmap_unlock(mapping);
334                         spin_unlock(&mapping->i_mmap_lock);
335                 }
336
337                 /*
338                  * Clear hugetlb-related page reserves for children. This only
339                  * affects MAP_PRIVATE mappings. Faults generated by the child
340                  * are not guaranteed to succeed, even if read-only
341                  */
342                 if (is_vm_hugetlb_page(tmp))
343                         reset_vma_resv_huge_pages(tmp);
344
345                 /*
346                  * Link in the new vma and copy the page table entries.
347                  */
348                 *pprev = tmp;
349                 pprev = &tmp->vm_next;
350
351                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
352                 rb_link = &tmp->vm_rb.rb_right;
353                 rb_parent = &tmp->vm_rb;
354
355                 mm->map_count++;
356                 retval = copy_page_range(mm, oldmm, mpnt);
357
358                 if (tmp->vm_ops && tmp->vm_ops->open)
359                         tmp->vm_ops->open(tmp);
360
361                 if (retval)
362                         goto out;
363         }
364         /* a new mm has just been created */
365         arch_dup_mmap(oldmm, mm);
366         retval = 0;
367 out:
368         up_write(&mm->mmap_sem);
369         flush_tlb_mm(oldmm);
370         up_write(&oldmm->mmap_sem);
371         return retval;
372 fail_nomem_policy:
373         kmem_cache_free(vm_area_cachep, tmp);
374 fail_nomem:
375         retval = -ENOMEM;
376         vm_unacct_memory(charge);
377         goto out;
378 }
379
380 static inline int mm_alloc_pgd(struct mm_struct * mm)
381 {
382         mm->pgd = pgd_alloc(mm);
383         if (unlikely(!mm->pgd))
384                 return -ENOMEM;
385         return 0;
386 }
387
388 static inline void mm_free_pgd(struct mm_struct * mm)
389 {
390         pgd_free(mm, mm->pgd);
391 }
392 #else
393 #define dup_mmap(mm, oldmm)     (0)
394 #define mm_alloc_pgd(mm)        (0)
395 #define mm_free_pgd(mm)
396 #endif /* CONFIG_MMU */
397
398 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
399
400 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
401 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
402
403 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
404
405 static int __init coredump_filter_setup(char *s)
406 {
407         default_dump_filter =
408                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
409                 MMF_DUMP_FILTER_MASK;
410         return 1;
411 }
412
413 __setup("coredump_filter=", coredump_filter_setup);
414
415 #include <linux/init_task.h>
416
417 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
418 {
419         atomic_set(&mm->mm_users, 1);
420         atomic_set(&mm->mm_count, 1);
421         init_rwsem(&mm->mmap_sem);
422         INIT_LIST_HEAD(&mm->mmlist);
423         mm->flags = (current->mm) ? current->mm->flags : default_dump_filter;
424         mm->core_state = NULL;
425         mm->nr_ptes = 0;
426         set_mm_counter(mm, file_rss, 0);
427         set_mm_counter(mm, anon_rss, 0);
428         spin_lock_init(&mm->page_table_lock);
429         spin_lock_init(&mm->ioctx_lock);
430         INIT_HLIST_HEAD(&mm->ioctx_list);
431         mm->free_area_cache = TASK_UNMAPPED_BASE;
432         mm->cached_hole_size = ~0UL;
433         mm_init_owner(mm, p);
434
435         if (likely(!mm_alloc_pgd(mm))) {
436                 mm->def_flags = 0;
437                 mmu_notifier_mm_init(mm);
438                 return mm;
439         }
440
441         free_mm(mm);
442         return NULL;
443 }
444
445 /*
446  * Allocate and initialize an mm_struct.
447  */
448 struct mm_struct * mm_alloc(void)
449 {
450         struct mm_struct * mm;
451
452         mm = allocate_mm();
453         if (mm) {
454                 memset(mm, 0, sizeof(*mm));
455                 mm = mm_init(mm, current);
456         }
457         return mm;
458 }
459
460 /*
461  * Called when the last reference to the mm
462  * is dropped: either by a lazy thread or by
463  * mmput. Free the page directory and the mm.
464  */
465 void __mmdrop(struct mm_struct *mm)
466 {
467         BUG_ON(mm == &init_mm);
468         mm_free_pgd(mm);
469         destroy_context(mm);
470         mmu_notifier_mm_destroy(mm);
471         free_mm(mm);
472 }
473 EXPORT_SYMBOL_GPL(__mmdrop);
474
475 /*
476  * Decrement the use count and release all resources for an mm.
477  */
478 void mmput(struct mm_struct *mm)
479 {
480         might_sleep();
481
482         if (atomic_dec_and_test(&mm->mm_users)) {
483                 exit_aio(mm);
484                 exit_mmap(mm);
485                 set_mm_exe_file(mm, NULL);
486                 if (!list_empty(&mm->mmlist)) {
487                         spin_lock(&mmlist_lock);
488                         list_del(&mm->mmlist);
489                         spin_unlock(&mmlist_lock);
490                 }
491                 put_swap_token(mm);
492                 mmdrop(mm);
493         }
494 }
495 EXPORT_SYMBOL_GPL(mmput);
496
497 /**
498  * get_task_mm - acquire a reference to the task's mm
499  *
500  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
501  * this kernel workthread has transiently adopted a user mm with use_mm,
502  * to do its AIO) is not set and if so returns a reference to it, after
503  * bumping up the use count.  User must release the mm via mmput()
504  * after use.  Typically used by /proc and ptrace.
505  */
506 struct mm_struct *get_task_mm(struct task_struct *task)
507 {
508         struct mm_struct *mm;
509
510         task_lock(task);
511         mm = task->mm;
512         if (mm) {
513                 if (task->flags & PF_KTHREAD)
514                         mm = NULL;
515                 else
516                         atomic_inc(&mm->mm_users);
517         }
518         task_unlock(task);
519         return mm;
520 }
521 EXPORT_SYMBOL_GPL(get_task_mm);
522
523 /* Please note the differences between mmput and mm_release.
524  * mmput is called whenever we stop holding onto a mm_struct,
525  * error success whatever.
526  *
527  * mm_release is called after a mm_struct has been removed
528  * from the current process.
529  *
530  * This difference is important for error handling, when we
531  * only half set up a mm_struct for a new process and need to restore
532  * the old one.  Because we mmput the new mm_struct before
533  * restoring the old one. . .
534  * Eric Biederman 10 January 1998
535  */
536 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
537 {
538         struct completion *vfork_done = tsk->vfork_done;
539
540         /* Get rid of any futexes when releasing the mm */
541 #ifdef CONFIG_FUTEX
542         if (unlikely(tsk->robust_list))
543                 exit_robust_list(tsk);
544 #ifdef CONFIG_COMPAT
545         if (unlikely(tsk->compat_robust_list))
546                 compat_exit_robust_list(tsk);
547 #endif
548 #endif
549
550         /* Get rid of any cached register state */
551         deactivate_mm(tsk, mm);
552
553         /* notify parent sleeping on vfork() */
554         if (vfork_done) {
555                 tsk->vfork_done = NULL;
556                 complete(vfork_done);
557         }
558
559         /*
560          * If we're exiting normally, clear a user-space tid field if
561          * requested.  We leave this alone when dying by signal, to leave
562          * the value intact in a core dump, and to save the unnecessary
563          * trouble otherwise.  Userland only wants this done for a sys_exit.
564          */
565         if (tsk->clear_child_tid
566             && !(tsk->flags & PF_SIGNALED)
567             && atomic_read(&mm->mm_users) > 1) {
568                 u32 __user * tidptr = tsk->clear_child_tid;
569                 tsk->clear_child_tid = NULL;
570
571                 /*
572                  * We don't check the error code - if userspace has
573                  * not set up a proper pointer then tough luck.
574                  */
575                 put_user(0, tidptr);
576                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
577         }
578 }
579
580 /*
581  * Allocate a new mm structure and copy contents from the
582  * mm structure of the passed in task structure.
583  */
584 struct mm_struct *dup_mm(struct task_struct *tsk)
585 {
586         struct mm_struct *mm, *oldmm = current->mm;
587         int err;
588
589         if (!oldmm)
590                 return NULL;
591
592         mm = allocate_mm();
593         if (!mm)
594                 goto fail_nomem;
595
596         memcpy(mm, oldmm, sizeof(*mm));
597
598         /* Initializing for Swap token stuff */
599         mm->token_priority = 0;
600         mm->last_interval = 0;
601
602         if (!mm_init(mm, tsk))
603                 goto fail_nomem;
604
605         if (init_new_context(tsk, mm))
606                 goto fail_nocontext;
607
608         dup_mm_exe_file(oldmm, mm);
609
610         err = dup_mmap(mm, oldmm);
611         if (err)
612                 goto free_pt;
613
614         mm->hiwater_rss = get_mm_rss(mm);
615         mm->hiwater_vm = mm->total_vm;
616
617         return mm;
618
619 free_pt:
620         mmput(mm);
621
622 fail_nomem:
623         return NULL;
624
625 fail_nocontext:
626         /*
627          * If init_new_context() failed, we cannot use mmput() to free the mm
628          * because it calls destroy_context()
629          */
630         mm_free_pgd(mm);
631         free_mm(mm);
632         return NULL;
633 }
634
635 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
636 {
637         struct mm_struct * mm, *oldmm;
638         int retval;
639
640         tsk->min_flt = tsk->maj_flt = 0;
641         tsk->nvcsw = tsk->nivcsw = 0;
642
643         tsk->mm = NULL;
644         tsk->active_mm = NULL;
645
646         /*
647          * Are we cloning a kernel thread?
648          *
649          * We need to steal a active VM for that..
650          */
651         oldmm = current->mm;
652         if (!oldmm)
653                 return 0;
654
655         if (clone_flags & CLONE_VM) {
656                 atomic_inc(&oldmm->mm_users);
657                 mm = oldmm;
658                 goto good_mm;
659         }
660
661         retval = -ENOMEM;
662         mm = dup_mm(tsk);
663         if (!mm)
664                 goto fail_nomem;
665
666 good_mm:
667         /* Initializing for Swap token stuff */
668         mm->token_priority = 0;
669         mm->last_interval = 0;
670
671         tsk->mm = mm;
672         tsk->active_mm = mm;
673         return 0;
674
675 fail_nomem:
676         return retval;
677 }
678
679 static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
680 {
681         struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
682         /* We don't need to lock fs - think why ;-) */
683         if (fs) {
684                 atomic_set(&fs->count, 1);
685                 rwlock_init(&fs->lock);
686                 fs->umask = old->umask;
687                 read_lock(&old->lock);
688                 fs->root = old->root;
689                 path_get(&old->root);
690                 fs->pwd = old->pwd;
691                 path_get(&old->pwd);
692                 read_unlock(&old->lock);
693         }
694         return fs;
695 }
696
697 struct fs_struct *copy_fs_struct(struct fs_struct *old)
698 {
699         return __copy_fs_struct(old);
700 }
701
702 EXPORT_SYMBOL_GPL(copy_fs_struct);
703
704 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
705 {
706         if (clone_flags & CLONE_FS) {
707                 atomic_inc(&current->fs->count);
708                 return 0;
709         }
710         tsk->fs = __copy_fs_struct(current->fs);
711         if (!tsk->fs)
712                 return -ENOMEM;
713         return 0;
714 }
715
716 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
717 {
718         struct files_struct *oldf, *newf;
719         int error = 0;
720
721         /*
722          * A background process may not have any files ...
723          */
724         oldf = current->files;
725         if (!oldf)
726                 goto out;
727
728         if (clone_flags & CLONE_FILES) {
729                 atomic_inc(&oldf->count);
730                 goto out;
731         }
732
733         newf = dup_fd(oldf, &error);
734         if (!newf)
735                 goto out;
736
737         tsk->files = newf;
738         error = 0;
739 out:
740         return error;
741 }
742
743 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
744 {
745 #ifdef CONFIG_BLOCK
746         struct io_context *ioc = current->io_context;
747
748         if (!ioc)
749                 return 0;
750         /*
751          * Share io context with parent, if CLONE_IO is set
752          */
753         if (clone_flags & CLONE_IO) {
754                 tsk->io_context = ioc_task_link(ioc);
755                 if (unlikely(!tsk->io_context))
756                         return -ENOMEM;
757         } else if (ioprio_valid(ioc->ioprio)) {
758                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
759                 if (unlikely(!tsk->io_context))
760                         return -ENOMEM;
761
762                 tsk->io_context->ioprio = ioc->ioprio;
763         }
764 #endif
765         return 0;
766 }
767
768 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
769 {
770         struct sighand_struct *sig;
771
772         if (clone_flags & CLONE_SIGHAND) {
773                 atomic_inc(&current->sighand->count);
774                 return 0;
775         }
776         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
777         rcu_assign_pointer(tsk->sighand, sig);
778         if (!sig)
779                 return -ENOMEM;
780         atomic_set(&sig->count, 1);
781         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
782         return 0;
783 }
784
785 void __cleanup_sighand(struct sighand_struct *sighand)
786 {
787         if (atomic_dec_and_test(&sighand->count))
788                 kmem_cache_free(sighand_cachep, sighand);
789 }
790
791
792 /*
793  * Initialize POSIX timer handling for a thread group.
794  */
795 static void posix_cpu_timers_init_group(struct signal_struct *sig)
796 {
797         /* Thread group counters. */
798         thread_group_cputime_init(sig);
799
800         /* Expiration times and increments. */
801         sig->it_virt_expires = cputime_zero;
802         sig->it_virt_incr = cputime_zero;
803         sig->it_prof_expires = cputime_zero;
804         sig->it_prof_incr = cputime_zero;
805
806         /* Cached expiration times. */
807         sig->cputime_expires.prof_exp = cputime_zero;
808         sig->cputime_expires.virt_exp = cputime_zero;
809         sig->cputime_expires.sched_exp = 0;
810
811         /* The timer lists. */
812         INIT_LIST_HEAD(&sig->cpu_timers[0]);
813         INIT_LIST_HEAD(&sig->cpu_timers[1]);
814         INIT_LIST_HEAD(&sig->cpu_timers[2]);
815 }
816
817 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
818 {
819         struct signal_struct *sig;
820
821         if (clone_flags & CLONE_THREAD) {
822                 atomic_inc(&current->signal->count);
823                 atomic_inc(&current->signal->live);
824                 return 0;
825         }
826         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
827
828         if (sig)
829                 posix_cpu_timers_init_group(sig);
830
831         tsk->signal = sig;
832         if (!sig)
833                 return -ENOMEM;
834
835         atomic_set(&sig->count, 1);
836         atomic_set(&sig->live, 1);
837         init_waitqueue_head(&sig->wait_chldexit);
838         sig->flags = 0;
839         sig->group_exit_code = 0;
840         sig->group_exit_task = NULL;
841         sig->group_stop_count = 0;
842         sig->curr_target = tsk;
843         init_sigpending(&sig->shared_pending);
844         INIT_LIST_HEAD(&sig->posix_timers);
845
846         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
847         sig->it_real_incr.tv64 = 0;
848         sig->real_timer.function = it_real_fn;
849
850         sig->leader = 0;        /* session leadership doesn't inherit */
851         sig->tty_old_pgrp = NULL;
852         sig->tty = NULL;
853
854         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
855         sig->gtime = cputime_zero;
856         sig->cgtime = cputime_zero;
857         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
858         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
859         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
860         task_io_accounting_init(&sig->ioac);
861         sig->sum_sched_runtime = 0;
862         taskstats_tgid_init(sig);
863
864         task_lock(current->group_leader);
865         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
866         task_unlock(current->group_leader);
867
868         acct_init_pacct(&sig->pacct);
869
870         tty_audit_fork(sig);
871
872         return 0;
873 }
874
875 void __cleanup_signal(struct signal_struct *sig)
876 {
877         thread_group_cputime_free(sig);
878         tty_kref_put(sig->tty);
879         kmem_cache_free(signal_cachep, sig);
880 }
881
882 static void cleanup_signal(struct task_struct *tsk)
883 {
884         struct signal_struct *sig = tsk->signal;
885
886         atomic_dec(&sig->live);
887
888         if (atomic_dec_and_test(&sig->count))
889                 __cleanup_signal(sig);
890 }
891
892 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
893 {
894         unsigned long new_flags = p->flags;
895
896         new_flags &= ~PF_SUPERPRIV;
897         new_flags |= PF_FORKNOEXEC;
898         new_flags |= PF_STARTING;
899         p->flags = new_flags;
900         clear_freeze_flag(p);
901 }
902
903 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
904 {
905         current->clear_child_tid = tidptr;
906
907         return task_pid_vnr(current);
908 }
909
910 static void rt_mutex_init_task(struct task_struct *p)
911 {
912         spin_lock_init(&p->pi_lock);
913 #ifdef CONFIG_RT_MUTEXES
914         plist_head_init(&p->pi_waiters, &p->pi_lock);
915         p->pi_blocked_on = NULL;
916 #endif
917 }
918
919 #ifdef CONFIG_MM_OWNER
920 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
921 {
922         mm->owner = p;
923 }
924 #endif /* CONFIG_MM_OWNER */
925
926 /*
927  * Initialize POSIX timer handling for a single task.
928  */
929 static void posix_cpu_timers_init(struct task_struct *tsk)
930 {
931         tsk->cputime_expires.prof_exp = cputime_zero;
932         tsk->cputime_expires.virt_exp = cputime_zero;
933         tsk->cputime_expires.sched_exp = 0;
934         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
935         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
936         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
937 }
938
939 /*
940  * This creates a new process as a copy of the old one,
941  * but does not actually start it yet.
942  *
943  * It copies the registers, and all the appropriate
944  * parts of the process environment (as per the clone
945  * flags). The actual kick-off is left to the caller.
946  */
947 static struct task_struct *copy_process(unsigned long clone_flags,
948                                         unsigned long stack_start,
949                                         struct pt_regs *regs,
950                                         unsigned long stack_size,
951                                         int __user *child_tidptr,
952                                         struct pid *pid,
953                                         int trace)
954 {
955         int retval;
956         struct task_struct *p;
957         int cgroup_callbacks_done = 0;
958
959         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
960                 return ERR_PTR(-EINVAL);
961
962         /*
963          * Thread groups must share signals as well, and detached threads
964          * can only be started up within the thread group.
965          */
966         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
967                 return ERR_PTR(-EINVAL);
968
969         /*
970          * Shared signal handlers imply shared VM. By way of the above,
971          * thread groups also imply shared VM. Blocking this case allows
972          * for various simplifications in other code.
973          */
974         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
975                 return ERR_PTR(-EINVAL);
976
977         retval = security_task_create(clone_flags);
978         if (retval)
979                 goto fork_out;
980
981         retval = -ENOMEM;
982         p = dup_task_struct(current);
983         if (!p)
984                 goto fork_out;
985
986         rt_mutex_init_task(p);
987
988 #ifdef CONFIG_PROVE_LOCKING
989         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
990         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
991 #endif
992         retval = -EAGAIN;
993         if (atomic_read(&p->real_cred->user->processes) >=
994                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
995                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
996                     p->real_cred->user != INIT_USER)
997                         goto bad_fork_free;
998         }
999
1000         retval = copy_creds(p, clone_flags);
1001         if (retval < 0)
1002                 goto bad_fork_free;
1003
1004         /*
1005          * If multiple threads are within copy_process(), then this check
1006          * triggers too late. This doesn't hurt, the check is only there
1007          * to stop root fork bombs.
1008          */
1009         if (nr_threads >= max_threads)
1010                 goto bad_fork_cleanup_count;
1011
1012         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1013                 goto bad_fork_cleanup_count;
1014
1015         if (p->binfmt && !try_module_get(p->binfmt->module))
1016                 goto bad_fork_cleanup_put_domain;
1017
1018         p->did_exec = 0;
1019         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1020         copy_flags(clone_flags, p);
1021         INIT_LIST_HEAD(&p->children);
1022         INIT_LIST_HEAD(&p->sibling);
1023 #ifdef CONFIG_PREEMPT_RCU
1024         p->rcu_read_lock_nesting = 0;
1025         p->rcu_flipctr_idx = 0;
1026 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1027         p->vfork_done = NULL;
1028         spin_lock_init(&p->alloc_lock);
1029
1030         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1031         init_sigpending(&p->pending);
1032
1033         p->utime = cputime_zero;
1034         p->stime = cputime_zero;
1035         p->gtime = cputime_zero;
1036         p->utimescaled = cputime_zero;
1037         p->stimescaled = cputime_zero;
1038         p->prev_utime = cputime_zero;
1039         p->prev_stime = cputime_zero;
1040
1041         p->default_timer_slack_ns = current->timer_slack_ns;
1042
1043 #ifdef CONFIG_DETECT_SOFTLOCKUP
1044         p->last_switch_count = 0;
1045         p->last_switch_timestamp = 0;
1046 #endif
1047
1048         task_io_accounting_init(&p->ioac);
1049         acct_clear_integrals(p);
1050
1051         posix_cpu_timers_init(p);
1052
1053         p->lock_depth = -1;             /* -1 = no lock */
1054         do_posix_clock_monotonic_gettime(&p->start_time);
1055         p->real_start_time = p->start_time;
1056         monotonic_to_bootbased(&p->real_start_time);
1057         p->io_context = NULL;
1058         p->audit_context = NULL;
1059         cgroup_fork(p);
1060 #ifdef CONFIG_NUMA
1061         p->mempolicy = mpol_dup(p->mempolicy);
1062         if (IS_ERR(p->mempolicy)) {
1063                 retval = PTR_ERR(p->mempolicy);
1064                 p->mempolicy = NULL;
1065                 goto bad_fork_cleanup_cgroup;
1066         }
1067         mpol_fix_fork_child_flag(p);
1068 #endif
1069 #ifdef CONFIG_TRACE_IRQFLAGS
1070         p->irq_events = 0;
1071 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1072         p->hardirqs_enabled = 1;
1073 #else
1074         p->hardirqs_enabled = 0;
1075 #endif
1076         p->hardirq_enable_ip = 0;
1077         p->hardirq_enable_event = 0;
1078         p->hardirq_disable_ip = _THIS_IP_;
1079         p->hardirq_disable_event = 0;
1080         p->softirqs_enabled = 1;
1081         p->softirq_enable_ip = _THIS_IP_;
1082         p->softirq_enable_event = 0;
1083         p->softirq_disable_ip = 0;
1084         p->softirq_disable_event = 0;
1085         p->hardirq_context = 0;
1086         p->softirq_context = 0;
1087 #endif
1088 #ifdef CONFIG_LOCKDEP
1089         p->lockdep_depth = 0; /* no locks held yet */
1090         p->curr_chain_key = 0;
1091         p->lockdep_recursion = 0;
1092 #endif
1093
1094 #ifdef CONFIG_DEBUG_MUTEXES
1095         p->blocked_on = NULL; /* not blocked yet */
1096 #endif
1097         if (unlikely(ptrace_reparented(current)))
1098                 ptrace_fork(p, clone_flags);
1099
1100         /* Perform scheduler related setup. Assign this task to a CPU. */
1101         sched_fork(p, clone_flags);
1102
1103         if ((retval = audit_alloc(p)))
1104                 goto bad_fork_cleanup_policy;
1105         /* copy all the process information */
1106         if ((retval = copy_semundo(clone_flags, p)))
1107                 goto bad_fork_cleanup_audit;
1108         if ((retval = copy_files(clone_flags, p)))
1109                 goto bad_fork_cleanup_semundo;
1110         if ((retval = copy_fs(clone_flags, p)))
1111                 goto bad_fork_cleanup_files;
1112         if ((retval = copy_sighand(clone_flags, p)))
1113                 goto bad_fork_cleanup_fs;
1114         if ((retval = copy_signal(clone_flags, p)))
1115                 goto bad_fork_cleanup_sighand;
1116         if ((retval = copy_mm(clone_flags, p)))
1117                 goto bad_fork_cleanup_signal;
1118         if ((retval = copy_namespaces(clone_flags, p)))
1119                 goto bad_fork_cleanup_mm;
1120         if ((retval = copy_io(clone_flags, p)))
1121                 goto bad_fork_cleanup_namespaces;
1122         retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
1123         if (retval)
1124                 goto bad_fork_cleanup_io;
1125
1126         if (pid != &init_struct_pid) {
1127                 retval = -ENOMEM;
1128                 pid = alloc_pid(p->nsproxy->pid_ns);
1129                 if (!pid)
1130                         goto bad_fork_cleanup_io;
1131
1132                 if (clone_flags & CLONE_NEWPID) {
1133                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1134                         if (retval < 0)
1135                                 goto bad_fork_free_pid;
1136                 }
1137         }
1138
1139         ftrace_graph_init_task(p);
1140
1141         p->pid = pid_nr(pid);
1142         p->tgid = p->pid;
1143         if (clone_flags & CLONE_THREAD)
1144                 p->tgid = current->tgid;
1145
1146         if (current->nsproxy != p->nsproxy) {
1147                 retval = ns_cgroup_clone(p, pid);
1148                 if (retval)
1149                         goto bad_fork_free_graph;
1150         }
1151
1152         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1153         /*
1154          * Clear TID on mm_release()?
1155          */
1156         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1157 #ifdef CONFIG_FUTEX
1158         p->robust_list = NULL;
1159 #ifdef CONFIG_COMPAT
1160         p->compat_robust_list = NULL;
1161 #endif
1162         INIT_LIST_HEAD(&p->pi_state_list);
1163         p->pi_state_cache = NULL;
1164 #endif
1165         /*
1166          * sigaltstack should be cleared when sharing the same VM
1167          */
1168         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1169                 p->sas_ss_sp = p->sas_ss_size = 0;
1170
1171         /*
1172          * Syscall tracing should be turned off in the child regardless
1173          * of CLONE_PTRACE.
1174          */
1175         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1176 #ifdef TIF_SYSCALL_EMU
1177         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1178 #endif
1179         clear_all_latency_tracing(p);
1180
1181         /* Our parent execution domain becomes current domain
1182            These must match for thread signalling to apply */
1183         p->parent_exec_id = p->self_exec_id;
1184
1185         /* ok, now we should be set up.. */
1186         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1187         p->pdeath_signal = 0;
1188         p->exit_state = 0;
1189
1190         /*
1191          * Ok, make it visible to the rest of the system.
1192          * We dont wake it up yet.
1193          */
1194         p->group_leader = p;
1195         INIT_LIST_HEAD(&p->thread_group);
1196
1197         /* Now that the task is set up, run cgroup callbacks if
1198          * necessary. We need to run them before the task is visible
1199          * on the tasklist. */
1200         cgroup_fork_callbacks(p);
1201         cgroup_callbacks_done = 1;
1202
1203         /* Need tasklist lock for parent etc handling! */
1204         write_lock_irq(&tasklist_lock);
1205
1206         /*
1207          * The task hasn't been attached yet, so its cpus_allowed mask will
1208          * not be changed, nor will its assigned CPU.
1209          *
1210          * The cpus_allowed mask of the parent may have changed after it was
1211          * copied first time - so re-copy it here, then check the child's CPU
1212          * to ensure it is on a valid CPU (and if not, just force it back to
1213          * parent's CPU). This avoids alot of nasty races.
1214          */
1215         p->cpus_allowed = current->cpus_allowed;
1216         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1217         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1218                         !cpu_online(task_cpu(p))))
1219                 set_task_cpu(p, smp_processor_id());
1220
1221         /* CLONE_PARENT re-uses the old parent */
1222         if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
1223                 p->real_parent = current->real_parent;
1224         else
1225                 p->real_parent = current;
1226
1227         spin_lock(&current->sighand->siglock);
1228
1229         /*
1230          * Process group and session signals need to be delivered to just the
1231          * parent before the fork or both the parent and the child after the
1232          * fork. Restart if a signal comes in before we add the new process to
1233          * it's process group.
1234          * A fatal signal pending means that current will exit, so the new
1235          * thread can't slip out of an OOM kill (or normal SIGKILL).
1236          */
1237         recalc_sigpending();
1238         if (signal_pending(current)) {
1239                 spin_unlock(&current->sighand->siglock);
1240                 write_unlock_irq(&tasklist_lock);
1241                 retval = -ERESTARTNOINTR;
1242                 goto bad_fork_free_graph;
1243         }
1244
1245         if (clone_flags & CLONE_THREAD) {
1246                 p->group_leader = current->group_leader;
1247                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1248         }
1249
1250         if (likely(p->pid)) {
1251                 list_add_tail(&p->sibling, &p->real_parent->children);
1252                 tracehook_finish_clone(p, clone_flags, trace);
1253
1254                 if (thread_group_leader(p)) {
1255                         if (clone_flags & CLONE_NEWPID)
1256                                 p->nsproxy->pid_ns->child_reaper = p;
1257
1258                         p->signal->leader_pid = pid;
1259                         tty_kref_put(p->signal->tty);
1260                         p->signal->tty = tty_kref_get(current->signal->tty);
1261                         set_task_pgrp(p, task_pgrp_nr(current));
1262                         set_task_session(p, task_session_nr(current));
1263                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1264                         attach_pid(p, PIDTYPE_SID, task_session(current));
1265                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1266                         __get_cpu_var(process_counts)++;
1267                 }
1268                 attach_pid(p, PIDTYPE_PID, pid);
1269                 nr_threads++;
1270         }
1271
1272         total_forks++;
1273         spin_unlock(&current->sighand->siglock);
1274         write_unlock_irq(&tasklist_lock);
1275         proc_fork_connector(p);
1276         cgroup_post_fork(p);
1277         return p;
1278
1279 bad_fork_free_graph:
1280         ftrace_graph_exit_task(p);
1281 bad_fork_free_pid:
1282         if (pid != &init_struct_pid)
1283                 free_pid(pid);
1284 bad_fork_cleanup_io:
1285         put_io_context(p->io_context);
1286 bad_fork_cleanup_namespaces:
1287         exit_task_namespaces(p);
1288 bad_fork_cleanup_mm:
1289         if (p->mm)
1290                 mmput(p->mm);
1291 bad_fork_cleanup_signal:
1292         cleanup_signal(p);
1293 bad_fork_cleanup_sighand:
1294         __cleanup_sighand(p->sighand);
1295 bad_fork_cleanup_fs:
1296         exit_fs(p); /* blocking */
1297 bad_fork_cleanup_files:
1298         exit_files(p); /* blocking */
1299 bad_fork_cleanup_semundo:
1300         exit_sem(p);
1301 bad_fork_cleanup_audit:
1302         audit_free(p);
1303 bad_fork_cleanup_policy:
1304 #ifdef CONFIG_NUMA
1305         mpol_put(p->mempolicy);
1306 bad_fork_cleanup_cgroup:
1307 #endif
1308         cgroup_exit(p, cgroup_callbacks_done);
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         atomic_dec(&p->cred->user->processes);
1316         put_cred(p->real_cred);
1317         put_cred(p->cred);
1318 bad_fork_free:
1319         free_task(p);
1320 fork_out:
1321         return ERR_PTR(retval);
1322 }
1323
1324 noinline struct pt_regs * __cpuinit __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 * __cpuinit 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,
1336                             &init_struct_pid, 0);
1337         if (!IS_ERR(task))
1338                 init_idle(task, cpu);
1339
1340         return task;
1341 }
1342
1343 /*
1344  *  Ok, this is the main fork-routine.
1345  *
1346  * It copies the process, and if successful kick-starts
1347  * it and waits for it to finish using the VM if required.
1348  */
1349 long do_fork(unsigned long clone_flags,
1350               unsigned long stack_start,
1351               struct pt_regs *regs,
1352               unsigned long stack_size,
1353               int __user *parent_tidptr,
1354               int __user *child_tidptr)
1355 {
1356         struct task_struct *p;
1357         int trace = 0;
1358         long nr;
1359
1360         /*
1361          * Do some preliminary argument and permissions checking before we
1362          * actually start allocating stuff
1363          */
1364         if (clone_flags & CLONE_NEWUSER) {
1365                 if (clone_flags & CLONE_THREAD)
1366                         return -EINVAL;
1367                 /* hopefully this check will go away when userns support is
1368                  * complete
1369                  */
1370                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1371                                 !capable(CAP_SETGID))
1372                         return -EPERM;
1373         }
1374
1375         /*
1376          * We hope to recycle these flags after 2.6.26
1377          */
1378         if (unlikely(clone_flags & CLONE_STOPPED)) {
1379                 static int __read_mostly count = 100;
1380
1381                 if (count > 0 && printk_ratelimit()) {
1382                         char comm[TASK_COMM_LEN];
1383
1384                         count--;
1385                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1386                                         "clone flags 0x%lx\n",
1387                                 get_task_comm(comm, current),
1388                                 clone_flags & CLONE_STOPPED);
1389                 }
1390         }
1391
1392         /*
1393          * When called from kernel_thread, don't do user tracing stuff.
1394          */
1395         if (likely(user_mode(regs)))
1396                 trace = tracehook_prepare_clone(clone_flags);
1397
1398         p = copy_process(clone_flags, stack_start, regs, stack_size,
1399                          child_tidptr, NULL, trace);
1400         /*
1401          * Do this prior waking up the new thread - the thread pointer
1402          * might get invalid after that point, if the thread exits quickly.
1403          */
1404         if (!IS_ERR(p)) {
1405                 struct completion vfork;
1406
1407                 trace_sched_process_fork(current, p);
1408
1409                 nr = task_pid_vnr(p);
1410
1411                 if (clone_flags & CLONE_PARENT_SETTID)
1412                         put_user(nr, parent_tidptr);
1413
1414                 if (clone_flags & CLONE_VFORK) {
1415                         p->vfork_done = &vfork;
1416                         init_completion(&vfork);
1417                 }
1418
1419                 audit_finish_fork(p);
1420                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1421
1422                 /*
1423                  * We set PF_STARTING at creation in case tracing wants to
1424                  * use this to distinguish a fully live task from one that
1425                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1426                  * clear it and set the child going.
1427                  */
1428                 p->flags &= ~PF_STARTING;
1429
1430                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1431                         /*
1432                          * We'll start up with an immediate SIGSTOP.
1433                          */
1434                         sigaddset(&p->pending.signal, SIGSTOP);
1435                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1436                         __set_task_state(p, TASK_STOPPED);
1437                 } else {
1438                         wake_up_new_task(p, clone_flags);
1439                 }
1440
1441                 tracehook_report_clone_complete(trace, regs,
1442                                                 clone_flags, nr, p);
1443
1444                 if (clone_flags & CLONE_VFORK) {
1445                         freezer_do_not_count();
1446                         wait_for_completion(&vfork);
1447                         freezer_count();
1448                         tracehook_report_vfork_done(p, nr);
1449                 }
1450         } else {
1451                 nr = PTR_ERR(p);
1452         }
1453         return nr;
1454 }
1455
1456 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1457 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1458 #endif
1459
1460 static void sighand_ctor(void *data)
1461 {
1462         struct sighand_struct *sighand = data;
1463
1464         spin_lock_init(&sighand->siglock);
1465         init_waitqueue_head(&sighand->signalfd_wqh);
1466 }
1467
1468 void __init proc_caches_init(void)
1469 {
1470         sighand_cachep = kmem_cache_create("sighand_cache",
1471                         sizeof(struct sighand_struct), 0,
1472                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1473                         sighand_ctor);
1474         signal_cachep = kmem_cache_create("signal_cache",
1475                         sizeof(struct signal_struct), 0,
1476                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1477         files_cachep = kmem_cache_create("files_cache",
1478                         sizeof(struct files_struct), 0,
1479                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1480         fs_cachep = kmem_cache_create("fs_cache",
1481                         sizeof(struct fs_struct), 0,
1482                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1483         mm_cachep = kmem_cache_create("mm_struct",
1484                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1485                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1486         mmap_init();
1487 }
1488
1489 /*
1490  * Check constraints on flags passed to the unshare system call and
1491  * force unsharing of additional process context as appropriate.
1492  */
1493 static void check_unshare_flags(unsigned long *flags_ptr)
1494 {
1495         /*
1496          * If unsharing a thread from a thread group, must also
1497          * unshare vm.
1498          */
1499         if (*flags_ptr & CLONE_THREAD)
1500                 *flags_ptr |= CLONE_VM;
1501
1502         /*
1503          * If unsharing vm, must also unshare signal handlers.
1504          */
1505         if (*flags_ptr & CLONE_VM)
1506                 *flags_ptr |= CLONE_SIGHAND;
1507
1508         /*
1509          * If unsharing signal handlers and the task was created
1510          * using CLONE_THREAD, then must unshare the thread
1511          */
1512         if ((*flags_ptr & CLONE_SIGHAND) &&
1513             (atomic_read(&current->signal->count) > 1))
1514                 *flags_ptr |= CLONE_THREAD;
1515
1516         /*
1517          * If unsharing namespace, must also unshare filesystem information.
1518          */
1519         if (*flags_ptr & CLONE_NEWNS)
1520                 *flags_ptr |= CLONE_FS;
1521 }
1522
1523 /*
1524  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1525  */
1526 static int unshare_thread(unsigned long unshare_flags)
1527 {
1528         if (unshare_flags & CLONE_THREAD)
1529                 return -EINVAL;
1530
1531         return 0;
1532 }
1533
1534 /*
1535  * Unshare the filesystem structure if it is being shared
1536  */
1537 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1538 {
1539         struct fs_struct *fs = current->fs;
1540
1541         if ((unshare_flags & CLONE_FS) &&
1542             (fs && atomic_read(&fs->count) > 1)) {
1543                 *new_fsp = __copy_fs_struct(current->fs);
1544                 if (!*new_fsp)
1545                         return -ENOMEM;
1546         }
1547
1548         return 0;
1549 }
1550
1551 /*
1552  * Unsharing of sighand is not supported yet
1553  */
1554 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1555 {
1556         struct sighand_struct *sigh = current->sighand;
1557
1558         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1559                 return -EINVAL;
1560         else
1561                 return 0;
1562 }
1563
1564 /*
1565  * Unshare vm if it is being shared
1566  */
1567 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1568 {
1569         struct mm_struct *mm = current->mm;
1570
1571         if ((unshare_flags & CLONE_VM) &&
1572             (mm && atomic_read(&mm->mm_users) > 1)) {
1573                 return -EINVAL;
1574         }
1575
1576         return 0;
1577 }
1578
1579 /*
1580  * Unshare file descriptor table if it is being shared
1581  */
1582 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1583 {
1584         struct files_struct *fd = current->files;
1585         int error = 0;
1586
1587         if ((unshare_flags & CLONE_FILES) &&
1588             (fd && atomic_read(&fd->count) > 1)) {
1589                 *new_fdp = dup_fd(fd, &error);
1590                 if (!*new_fdp)
1591                         return error;
1592         }
1593
1594         return 0;
1595 }
1596
1597 /*
1598  * unshare allows a process to 'unshare' part of the process
1599  * context which was originally shared using clone.  copy_*
1600  * functions used by do_fork() cannot be used here directly
1601  * because they modify an inactive task_struct that is being
1602  * constructed. Here we are modifying the current, active,
1603  * task_struct.
1604  */
1605 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1606 {
1607         int err = 0;
1608         struct fs_struct *fs, *new_fs = NULL;
1609         struct sighand_struct *new_sigh = NULL;
1610         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1611         struct files_struct *fd, *new_fd = NULL;
1612         struct nsproxy *new_nsproxy = NULL;
1613         int do_sysvsem = 0;
1614
1615         check_unshare_flags(&unshare_flags);
1616
1617         /* Return -EINVAL for all unsupported flags */
1618         err = -EINVAL;
1619         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1620                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1621                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1622                 goto bad_unshare_out;
1623
1624         /*
1625          * CLONE_NEWIPC must also detach from the undolist: after switching
1626          * to a new ipc namespace, the semaphore arrays from the old
1627          * namespace are unreachable.
1628          */
1629         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1630                 do_sysvsem = 1;
1631         if ((err = unshare_thread(unshare_flags)))
1632                 goto bad_unshare_out;
1633         if ((err = unshare_fs(unshare_flags, &new_fs)))
1634                 goto bad_unshare_cleanup_thread;
1635         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1636                 goto bad_unshare_cleanup_fs;
1637         if ((err = unshare_vm(unshare_flags, &new_mm)))
1638                 goto bad_unshare_cleanup_sigh;
1639         if ((err = unshare_fd(unshare_flags, &new_fd)))
1640                 goto bad_unshare_cleanup_vm;
1641         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1642                         new_fs)))
1643                 goto bad_unshare_cleanup_fd;
1644
1645         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1646                 if (do_sysvsem) {
1647                         /*
1648                          * CLONE_SYSVSEM is equivalent to sys_exit().
1649                          */
1650                         exit_sem(current);
1651                 }
1652
1653                 if (new_nsproxy) {
1654                         switch_task_namespaces(current, new_nsproxy);
1655                         new_nsproxy = NULL;
1656                 }
1657
1658                 task_lock(current);
1659
1660                 if (new_fs) {
1661                         fs = current->fs;
1662                         current->fs = new_fs;
1663                         new_fs = fs;
1664                 }
1665
1666                 if (new_mm) {
1667                         mm = current->mm;
1668                         active_mm = current->active_mm;
1669                         current->mm = new_mm;
1670                         current->active_mm = new_mm;
1671                         activate_mm(active_mm, new_mm);
1672                         new_mm = mm;
1673                 }
1674
1675                 if (new_fd) {
1676                         fd = current->files;
1677                         current->files = new_fd;
1678                         new_fd = fd;
1679                 }
1680
1681                 task_unlock(current);
1682         }
1683
1684         if (new_nsproxy)
1685                 put_nsproxy(new_nsproxy);
1686
1687 bad_unshare_cleanup_fd:
1688         if (new_fd)
1689                 put_files_struct(new_fd);
1690
1691 bad_unshare_cleanup_vm:
1692         if (new_mm)
1693                 mmput(new_mm);
1694
1695 bad_unshare_cleanup_sigh:
1696         if (new_sigh)
1697                 if (atomic_dec_and_test(&new_sigh->count))
1698                         kmem_cache_free(sighand_cachep, new_sigh);
1699
1700 bad_unshare_cleanup_fs:
1701         if (new_fs)
1702                 put_fs_struct(new_fs);
1703
1704 bad_unshare_cleanup_thread:
1705 bad_unshare_out:
1706         return err;
1707 }
1708
1709 /*
1710  *      Helper to unshare the files of the current task.
1711  *      We don't want to expose copy_files internals to
1712  *      the exec layer of the kernel.
1713  */
1714
1715 int unshare_files(struct files_struct **displaced)
1716 {
1717         struct task_struct *task = current;
1718         struct files_struct *copy = NULL;
1719         int error;
1720
1721         error = unshare_fd(CLONE_FILES, &copy);
1722         if (error || !copy) {
1723                 *displaced = NULL;
1724                 return error;
1725         }
1726         *displaced = task->files;
1727         task_lock(task);
1728         task->files = copy;
1729         task_unlock(task);
1730         return 0;
1731 }