execve: must clear current->clear_child_tid
[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/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/nsproxy.h>
32 #include <linux/capability.h>
33 #include <linux/cpu.h>
34 #include <linux/cgroup.h>
35 #include <linux/security.h>
36 #include <linux/hugetlb.h>
37 #include <linux/swap.h>
38 #include <linux/syscalls.h>
39 #include <linux/jiffies.h>
40 #include <linux/tracehook.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/task_io_accounting_ops.h>
44 #include <linux/rcupdate.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/audit.h>
48 #include <linux/memcontrol.h>
49 #include <linux/ftrace.h>
50 #include <linux/profile.h>
51 #include <linux/rmap.h>
52 #include <linux/acct.h>
53 #include <linux/tsacct_kern.h>
54 #include <linux/cn_proc.h>
55 #include <linux/freezer.h>
56 #include <linux/delayacct.h>
57 #include <linux/taskstats_kern.h>
58 #include <linux/random.h>
59 #include <linux/tty.h>
60 #include <linux/proc_fs.h>
61 #include <linux/blkdev.h>
62 #include <linux/fs_struct.h>
63 #include <linux/magic.h>
64 #include <linux/perf_counter.h>
65
66 #include <asm/pgtable.h>
67 #include <asm/pgalloc.h>
68 #include <asm/uaccess.h>
69 #include <asm/mmu_context.h>
70 #include <asm/cacheflush.h>
71 #include <asm/tlbflush.h>
72
73 #include <trace/events/sched.h>
74
75 /*
76  * Protected counters by write_lock_irq(&tasklist_lock)
77  */
78 unsigned long total_forks;      /* Handle normal Linux uptimes. */
79 int nr_threads;                 /* The idle threads do not count.. */
80
81 int max_threads;                /* tunable limit on nr_threads */
82
83 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
84
85 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
86
87 int nr_processes(void)
88 {
89         int cpu;
90         int total = 0;
91
92         for_each_online_cpu(cpu)
93                 total += per_cpu(process_counts, cpu);
94
95         return total;
96 }
97
98 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
99 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
100 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
101 static struct kmem_cache *task_struct_cachep;
102 #endif
103
104 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
105 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
106 {
107 #ifdef CONFIG_DEBUG_STACK_USAGE
108         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
109 #else
110         gfp_t mask = GFP_KERNEL;
111 #endif
112         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
113 }
114
115 static inline void free_thread_info(struct thread_info *ti)
116 {
117         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
118 }
119 #endif
120
121 /* SLAB cache for signal_struct structures (tsk->signal) */
122 static struct kmem_cache *signal_cachep;
123
124 /* SLAB cache for sighand_struct structures (tsk->sighand) */
125 struct kmem_cache *sighand_cachep;
126
127 /* SLAB cache for files_struct structures (tsk->files) */
128 struct kmem_cache *files_cachep;
129
130 /* SLAB cache for fs_struct structures (tsk->fs) */
131 struct kmem_cache *fs_cachep;
132
133 /* SLAB cache for vm_area_struct structures */
134 struct kmem_cache *vm_area_cachep;
135
136 /* SLAB cache for mm_struct structures (tsk->mm) */
137 static struct kmem_cache *mm_cachep;
138
139 void free_task(struct task_struct *tsk)
140 {
141         prop_local_destroy_single(&tsk->dirties);
142         free_thread_info(tsk->stack);
143         rt_mutex_debug_task_free(tsk);
144         ftrace_graph_exit_task(tsk);
145         free_task_struct(tsk);
146 }
147 EXPORT_SYMBOL(free_task);
148
149 void __put_task_struct(struct task_struct *tsk)
150 {
151         WARN_ON(!tsk->exit_state);
152         WARN_ON(atomic_read(&tsk->usage));
153         WARN_ON(tsk == current);
154
155         put_cred(tsk->real_cred);
156         put_cred(tsk->cred);
157         delayacct_tsk_free(tsk);
158
159         if (!profile_handoff_task(tsk))
160                 free_task(tsk);
161 }
162
163 /*
164  * macro override instead of weak attribute alias, to workaround
165  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
166  */
167 #ifndef arch_task_cache_init
168 #define arch_task_cache_init()
169 #endif
170
171 void __init fork_init(unsigned long mempages)
172 {
173 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
174 #ifndef ARCH_MIN_TASKALIGN
175 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
176 #endif
177         /* create a slab on which task_structs can be allocated */
178         task_struct_cachep =
179                 kmem_cache_create("task_struct", sizeof(struct task_struct),
180                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
181 #endif
182
183         /* do the arch specific task caches init */
184         arch_task_cache_init();
185
186         /*
187          * The default maximum number of threads is set to a safe
188          * value: the thread structures can take up at most half
189          * of memory.
190          */
191         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
192
193         /*
194          * we need to allow at least 20 threads to boot a system
195          */
196         if(max_threads < 20)
197                 max_threads = 20;
198
199         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
200         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
201         init_task.signal->rlim[RLIMIT_SIGPENDING] =
202                 init_task.signal->rlim[RLIMIT_NPROC];
203 }
204
205 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
206                                                struct task_struct *src)
207 {
208         *dst = *src;
209         return 0;
210 }
211
212 static struct task_struct *dup_task_struct(struct task_struct *orig)
213 {
214         struct task_struct *tsk;
215         struct thread_info *ti;
216         unsigned long *stackend;
217
218         int err;
219
220         prepare_to_copy(orig);
221
222         tsk = alloc_task_struct();
223         if (!tsk)
224                 return NULL;
225
226         ti = alloc_thread_info(tsk);
227         if (!ti) {
228                 free_task_struct(tsk);
229                 return NULL;
230         }
231
232         err = arch_dup_task_struct(tsk, orig);
233         if (err)
234                 goto out;
235
236         tsk->stack = ti;
237
238         err = prop_local_init_single(&tsk->dirties);
239         if (err)
240                 goto out;
241
242         setup_thread_stack(tsk, orig);
243         stackend = end_of_stack(tsk);
244         *stackend = STACK_END_MAGIC;    /* for overflow detection */
245
246 #ifdef CONFIG_CC_STACKPROTECTOR
247         tsk->stack_canary = get_random_int();
248 #endif
249
250         /* One for us, one for whoever does the "release_task()" (usually parent) */
251         atomic_set(&tsk->usage,2);
252         atomic_set(&tsk->fs_excl, 0);
253 #ifdef CONFIG_BLK_DEV_IO_TRACE
254         tsk->btrace_seq = 0;
255 #endif
256         tsk->splice_pipe = NULL;
257         return tsk;
258
259 out:
260         free_thread_info(ti);
261         free_task_struct(tsk);
262         return NULL;
263 }
264
265 #ifdef CONFIG_MMU
266 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
267 {
268         struct vm_area_struct *mpnt, *tmp, **pprev;
269         struct rb_node **rb_link, *rb_parent;
270         int retval;
271         unsigned long charge;
272         struct mempolicy *pol;
273
274         down_write(&oldmm->mmap_sem);
275         flush_cache_dup_mm(oldmm);
276         /*
277          * Not linked in yet - no deadlock potential:
278          */
279         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
280
281         mm->locked_vm = 0;
282         mm->mmap = NULL;
283         mm->mmap_cache = NULL;
284         mm->free_area_cache = oldmm->mmap_base;
285         mm->cached_hole_size = ~0UL;
286         mm->map_count = 0;
287         cpumask_clear(mm_cpumask(mm));
288         mm->mm_rb = RB_ROOT;
289         rb_link = &mm->mm_rb.rb_node;
290         rb_parent = NULL;
291         pprev = &mm->mmap;
292
293         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
294                 struct file *file;
295
296                 if (mpnt->vm_flags & VM_DONTCOPY) {
297                         long pages = vma_pages(mpnt);
298                         mm->total_vm -= pages;
299                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
300                                                                 -pages);
301                         continue;
302                 }
303                 charge = 0;
304                 if (mpnt->vm_flags & VM_ACCOUNT) {
305                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
306                         if (security_vm_enough_memory(len))
307                                 goto fail_nomem;
308                         charge = len;
309                 }
310                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
311                 if (!tmp)
312                         goto fail_nomem;
313                 *tmp = *mpnt;
314                 pol = mpol_dup(vma_policy(mpnt));
315                 retval = PTR_ERR(pol);
316                 if (IS_ERR(pol))
317                         goto fail_nomem_policy;
318                 vma_set_policy(tmp, pol);
319                 tmp->vm_flags &= ~VM_LOCKED;
320                 tmp->vm_mm = mm;
321                 tmp->vm_next = NULL;
322                 anon_vma_link(tmp);
323                 file = tmp->vm_file;
324                 if (file) {
325                         struct inode *inode = file->f_path.dentry->d_inode;
326                         struct address_space *mapping = file->f_mapping;
327
328                         get_file(file);
329                         if (tmp->vm_flags & VM_DENYWRITE)
330                                 atomic_dec(&inode->i_writecount);
331                         spin_lock(&mapping->i_mmap_lock);
332                         if (tmp->vm_flags & VM_SHARED)
333                                 mapping->i_mmap_writable++;
334                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
335                         flush_dcache_mmap_lock(mapping);
336                         /* insert tmp into the share list, just after mpnt */
337                         vma_prio_tree_add(tmp, mpnt);
338                         flush_dcache_mmap_unlock(mapping);
339                         spin_unlock(&mapping->i_mmap_lock);
340                 }
341
342                 /*
343                  * Clear hugetlb-related page reserves for children. This only
344                  * affects MAP_PRIVATE mappings. Faults generated by the child
345                  * are not guaranteed to succeed, even if read-only
346                  */
347                 if (is_vm_hugetlb_page(tmp))
348                         reset_vma_resv_huge_pages(tmp);
349
350                 /*
351                  * Link in the new vma and copy the page table entries.
352                  */
353                 *pprev = tmp;
354                 pprev = &tmp->vm_next;
355
356                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
357                 rb_link = &tmp->vm_rb.rb_right;
358                 rb_parent = &tmp->vm_rb;
359
360                 mm->map_count++;
361                 retval = copy_page_range(mm, oldmm, mpnt);
362
363                 if (tmp->vm_ops && tmp->vm_ops->open)
364                         tmp->vm_ops->open(tmp);
365
366                 if (retval)
367                         goto out;
368         }
369         /* a new mm has just been created */
370         arch_dup_mmap(oldmm, mm);
371         retval = 0;
372 out:
373         up_write(&mm->mmap_sem);
374         flush_tlb_mm(oldmm);
375         up_write(&oldmm->mmap_sem);
376         return retval;
377 fail_nomem_policy:
378         kmem_cache_free(vm_area_cachep, tmp);
379 fail_nomem:
380         retval = -ENOMEM;
381         vm_unacct_memory(charge);
382         goto out;
383 }
384
385 static inline int mm_alloc_pgd(struct mm_struct * mm)
386 {
387         mm->pgd = pgd_alloc(mm);
388         if (unlikely(!mm->pgd))
389                 return -ENOMEM;
390         return 0;
391 }
392
393 static inline void mm_free_pgd(struct mm_struct * mm)
394 {
395         pgd_free(mm, mm->pgd);
396 }
397 #else
398 #define dup_mmap(mm, oldmm)     (0)
399 #define mm_alloc_pgd(mm)        (0)
400 #define mm_free_pgd(mm)
401 #endif /* CONFIG_MMU */
402
403 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
404
405 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
406 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
407
408 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
409
410 static int __init coredump_filter_setup(char *s)
411 {
412         default_dump_filter =
413                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
414                 MMF_DUMP_FILTER_MASK;
415         return 1;
416 }
417
418 __setup("coredump_filter=", coredump_filter_setup);
419
420 #include <linux/init_task.h>
421
422 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
423 {
424         atomic_set(&mm->mm_users, 1);
425         atomic_set(&mm->mm_count, 1);
426         init_rwsem(&mm->mmap_sem);
427         INIT_LIST_HEAD(&mm->mmlist);
428         mm->flags = (current->mm) ? current->mm->flags : default_dump_filter;
429         mm->oom_adj = (current->mm) ? current->mm->oom_adj : 0;
430         mm->core_state = NULL;
431         mm->nr_ptes = 0;
432         set_mm_counter(mm, file_rss, 0);
433         set_mm_counter(mm, anon_rss, 0);
434         spin_lock_init(&mm->page_table_lock);
435         spin_lock_init(&mm->ioctx_lock);
436         INIT_HLIST_HEAD(&mm->ioctx_list);
437         mm->free_area_cache = TASK_UNMAPPED_BASE;
438         mm->cached_hole_size = ~0UL;
439         mm_init_owner(mm, p);
440
441         if (likely(!mm_alloc_pgd(mm))) {
442                 mm->def_flags = 0;
443                 mmu_notifier_mm_init(mm);
444                 return mm;
445         }
446
447         free_mm(mm);
448         return NULL;
449 }
450
451 /*
452  * Allocate and initialize an mm_struct.
453  */
454 struct mm_struct * mm_alloc(void)
455 {
456         struct mm_struct * mm;
457
458         mm = allocate_mm();
459         if (mm) {
460                 memset(mm, 0, sizeof(*mm));
461                 mm = mm_init(mm, current);
462         }
463         return mm;
464 }
465
466 /*
467  * Called when the last reference to the mm
468  * is dropped: either by a lazy thread or by
469  * mmput. Free the page directory and the mm.
470  */
471 void __mmdrop(struct mm_struct *mm)
472 {
473         BUG_ON(mm == &init_mm);
474         mm_free_pgd(mm);
475         destroy_context(mm);
476         mmu_notifier_mm_destroy(mm);
477         free_mm(mm);
478 }
479 EXPORT_SYMBOL_GPL(__mmdrop);
480
481 /*
482  * Decrement the use count and release all resources for an mm.
483  */
484 void mmput(struct mm_struct *mm)
485 {
486         might_sleep();
487
488         if (atomic_dec_and_test(&mm->mm_users)) {
489                 exit_aio(mm);
490                 exit_mmap(mm);
491                 set_mm_exe_file(mm, NULL);
492                 if (!list_empty(&mm->mmlist)) {
493                         spin_lock(&mmlist_lock);
494                         list_del(&mm->mmlist);
495                         spin_unlock(&mmlist_lock);
496                 }
497                 put_swap_token(mm);
498                 mmdrop(mm);
499         }
500 }
501 EXPORT_SYMBOL_GPL(mmput);
502
503 /**
504  * get_task_mm - acquire a reference to the task's mm
505  *
506  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
507  * this kernel workthread has transiently adopted a user mm with use_mm,
508  * to do its AIO) is not set and if so returns a reference to it, after
509  * bumping up the use count.  User must release the mm via mmput()
510  * after use.  Typically used by /proc and ptrace.
511  */
512 struct mm_struct *get_task_mm(struct task_struct *task)
513 {
514         struct mm_struct *mm;
515
516         task_lock(task);
517         mm = task->mm;
518         if (mm) {
519                 if (task->flags & PF_KTHREAD)
520                         mm = NULL;
521                 else
522                         atomic_inc(&mm->mm_users);
523         }
524         task_unlock(task);
525         return mm;
526 }
527 EXPORT_SYMBOL_GPL(get_task_mm);
528
529 /* Please note the differences between mmput and mm_release.
530  * mmput is called whenever we stop holding onto a mm_struct,
531  * error success whatever.
532  *
533  * mm_release is called after a mm_struct has been removed
534  * from the current process.
535  *
536  * This difference is important for error handling, when we
537  * only half set up a mm_struct for a new process and need to restore
538  * the old one.  Because we mmput the new mm_struct before
539  * restoring the old one. . .
540  * Eric Biederman 10 January 1998
541  */
542 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
543 {
544         struct completion *vfork_done = tsk->vfork_done;
545
546         /* Get rid of any futexes when releasing the mm */
547 #ifdef CONFIG_FUTEX
548         if (unlikely(tsk->robust_list))
549                 exit_robust_list(tsk);
550 #ifdef CONFIG_COMPAT
551         if (unlikely(tsk->compat_robust_list))
552                 compat_exit_robust_list(tsk);
553 #endif
554 #endif
555
556         /* Get rid of any cached register state */
557         deactivate_mm(tsk, mm);
558
559         /* notify parent sleeping on vfork() */
560         if (vfork_done) {
561                 tsk->vfork_done = NULL;
562                 complete(vfork_done);
563         }
564
565         /*
566          * If we're exiting normally, clear a user-space tid field if
567          * requested.  We leave this alone when dying by signal, to leave
568          * the value intact in a core dump, and to save the unnecessary
569          * trouble otherwise.  Userland only wants this done for a sys_exit.
570          */
571         if (tsk->clear_child_tid) {
572                 if (!(tsk->flags & PF_SIGNALED) &&
573                     atomic_read(&mm->mm_users) > 1) {
574                         /*
575                          * We don't check the error code - if userspace has
576                          * not set up a proper pointer then tough luck.
577                          */
578                         put_user(0, tsk->clear_child_tid);
579                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
580                                         1, NULL, NULL, 0);
581                 }
582                 tsk->clear_child_tid = NULL;
583         }
584 }
585
586 /*
587  * Allocate a new mm structure and copy contents from the
588  * mm structure of the passed in task structure.
589  */
590 struct mm_struct *dup_mm(struct task_struct *tsk)
591 {
592         struct mm_struct *mm, *oldmm = current->mm;
593         int err;
594
595         if (!oldmm)
596                 return NULL;
597
598         mm = allocate_mm();
599         if (!mm)
600                 goto fail_nomem;
601
602         memcpy(mm, oldmm, sizeof(*mm));
603
604         /* Initializing for Swap token stuff */
605         mm->token_priority = 0;
606         mm->last_interval = 0;
607
608         if (!mm_init(mm, tsk))
609                 goto fail_nomem;
610
611         if (init_new_context(tsk, mm))
612                 goto fail_nocontext;
613
614         dup_mm_exe_file(oldmm, mm);
615
616         err = dup_mmap(mm, oldmm);
617         if (err)
618                 goto free_pt;
619
620         mm->hiwater_rss = get_mm_rss(mm);
621         mm->hiwater_vm = mm->total_vm;
622
623         return mm;
624
625 free_pt:
626         mmput(mm);
627
628 fail_nomem:
629         return NULL;
630
631 fail_nocontext:
632         /*
633          * If init_new_context() failed, we cannot use mmput() to free the mm
634          * because it calls destroy_context()
635          */
636         mm_free_pgd(mm);
637         free_mm(mm);
638         return NULL;
639 }
640
641 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
642 {
643         struct mm_struct * mm, *oldmm;
644         int retval;
645
646         tsk->min_flt = tsk->maj_flt = 0;
647         tsk->nvcsw = tsk->nivcsw = 0;
648 #ifdef CONFIG_DETECT_HUNG_TASK
649         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
650 #endif
651
652         tsk->mm = NULL;
653         tsk->active_mm = NULL;
654
655         /*
656          * Are we cloning a kernel thread?
657          *
658          * We need to steal a active VM for that..
659          */
660         oldmm = current->mm;
661         if (!oldmm)
662                 return 0;
663
664         if (clone_flags & CLONE_VM) {
665                 atomic_inc(&oldmm->mm_users);
666                 mm = oldmm;
667                 goto good_mm;
668         }
669
670         retval = -ENOMEM;
671         mm = dup_mm(tsk);
672         if (!mm)
673                 goto fail_nomem;
674
675 good_mm:
676         /* Initializing for Swap token stuff */
677         mm->token_priority = 0;
678         mm->last_interval = 0;
679
680         tsk->mm = mm;
681         tsk->active_mm = mm;
682         return 0;
683
684 fail_nomem:
685         return retval;
686 }
687
688 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
689 {
690         struct fs_struct *fs = current->fs;
691         if (clone_flags & CLONE_FS) {
692                 /* tsk->fs is already what we want */
693                 write_lock(&fs->lock);
694                 if (fs->in_exec) {
695                         write_unlock(&fs->lock);
696                         return -EAGAIN;
697                 }
698                 fs->users++;
699                 write_unlock(&fs->lock);
700                 return 0;
701         }
702         tsk->fs = copy_fs_struct(fs);
703         if (!tsk->fs)
704                 return -ENOMEM;
705         return 0;
706 }
707
708 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
709 {
710         struct files_struct *oldf, *newf;
711         int error = 0;
712
713         /*
714          * A background process may not have any files ...
715          */
716         oldf = current->files;
717         if (!oldf)
718                 goto out;
719
720         if (clone_flags & CLONE_FILES) {
721                 atomic_inc(&oldf->count);
722                 goto out;
723         }
724
725         newf = dup_fd(oldf, &error);
726         if (!newf)
727                 goto out;
728
729         tsk->files = newf;
730         error = 0;
731 out:
732         return error;
733 }
734
735 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
736 {
737 #ifdef CONFIG_BLOCK
738         struct io_context *ioc = current->io_context;
739
740         if (!ioc)
741                 return 0;
742         /*
743          * Share io context with parent, if CLONE_IO is set
744          */
745         if (clone_flags & CLONE_IO) {
746                 tsk->io_context = ioc_task_link(ioc);
747                 if (unlikely(!tsk->io_context))
748                         return -ENOMEM;
749         } else if (ioprio_valid(ioc->ioprio)) {
750                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
751                 if (unlikely(!tsk->io_context))
752                         return -ENOMEM;
753
754                 tsk->io_context->ioprio = ioc->ioprio;
755         }
756 #endif
757         return 0;
758 }
759
760 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
761 {
762         struct sighand_struct *sig;
763
764         if (clone_flags & CLONE_SIGHAND) {
765                 atomic_inc(&current->sighand->count);
766                 return 0;
767         }
768         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
769         rcu_assign_pointer(tsk->sighand, sig);
770         if (!sig)
771                 return -ENOMEM;
772         atomic_set(&sig->count, 1);
773         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
774         return 0;
775 }
776
777 void __cleanup_sighand(struct sighand_struct *sighand)
778 {
779         if (atomic_dec_and_test(&sighand->count))
780                 kmem_cache_free(sighand_cachep, sighand);
781 }
782
783
784 /*
785  * Initialize POSIX timer handling for a thread group.
786  */
787 static void posix_cpu_timers_init_group(struct signal_struct *sig)
788 {
789         /* Thread group counters. */
790         thread_group_cputime_init(sig);
791
792         /* Expiration times and increments. */
793         sig->it_virt_expires = cputime_zero;
794         sig->it_virt_incr = cputime_zero;
795         sig->it_prof_expires = cputime_zero;
796         sig->it_prof_incr = cputime_zero;
797
798         /* Cached expiration times. */
799         sig->cputime_expires.prof_exp = cputime_zero;
800         sig->cputime_expires.virt_exp = cputime_zero;
801         sig->cputime_expires.sched_exp = 0;
802
803         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
804                 sig->cputime_expires.prof_exp =
805                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
806                 sig->cputimer.running = 1;
807         }
808
809         /* The timer lists. */
810         INIT_LIST_HEAD(&sig->cpu_timers[0]);
811         INIT_LIST_HEAD(&sig->cpu_timers[1]);
812         INIT_LIST_HEAD(&sig->cpu_timers[2]);
813 }
814
815 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
816 {
817         struct signal_struct *sig;
818
819         if (clone_flags & CLONE_THREAD) {
820                 atomic_inc(&current->signal->count);
821                 atomic_inc(&current->signal->live);
822                 return 0;
823         }
824
825         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
826         tsk->signal = sig;
827         if (!sig)
828                 return -ENOMEM;
829
830         atomic_set(&sig->count, 1);
831         atomic_set(&sig->live, 1);
832         init_waitqueue_head(&sig->wait_chldexit);
833         sig->flags = 0;
834         if (clone_flags & CLONE_NEWPID)
835                 sig->flags |= SIGNAL_UNKILLABLE;
836         sig->group_exit_code = 0;
837         sig->group_exit_task = NULL;
838         sig->group_stop_count = 0;
839         sig->curr_target = tsk;
840         init_sigpending(&sig->shared_pending);
841         INIT_LIST_HEAD(&sig->posix_timers);
842
843         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
844         sig->it_real_incr.tv64 = 0;
845         sig->real_timer.function = it_real_fn;
846
847         sig->leader = 0;        /* session leadership doesn't inherit */
848         sig->tty_old_pgrp = NULL;
849         sig->tty = NULL;
850
851         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
852         sig->gtime = cputime_zero;
853         sig->cgtime = cputime_zero;
854         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
855         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
856         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
857         task_io_accounting_init(&sig->ioac);
858         sig->sum_sched_runtime = 0;
859         taskstats_tgid_init(sig);
860
861         task_lock(current->group_leader);
862         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
863         task_unlock(current->group_leader);
864
865         posix_cpu_timers_init_group(sig);
866
867         acct_init_pacct(&sig->pacct);
868
869         tty_audit_fork(sig);
870
871         return 0;
872 }
873
874 void __cleanup_signal(struct signal_struct *sig)
875 {
876         thread_group_cputime_free(sig);
877         tty_kref_put(sig->tty);
878         kmem_cache_free(signal_cachep, sig);
879 }
880
881 static void cleanup_signal(struct task_struct *tsk)
882 {
883         struct signal_struct *sig = tsk->signal;
884
885         atomic_dec(&sig->live);
886
887         if (atomic_dec_and_test(&sig->count))
888                 __cleanup_signal(sig);
889 }
890
891 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
892 {
893         unsigned long new_flags = p->flags;
894
895         new_flags &= ~PF_SUPERPRIV;
896         new_flags |= PF_FORKNOEXEC;
897         new_flags |= PF_STARTING;
898         p->flags = new_flags;
899         clear_freeze_flag(p);
900 }
901
902 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
903 {
904         current->clear_child_tid = tidptr;
905
906         return task_pid_vnr(current);
907 }
908
909 static void rt_mutex_init_task(struct task_struct *p)
910 {
911         spin_lock_init(&p->pi_lock);
912 #ifdef CONFIG_RT_MUTEXES
913         plist_head_init(&p->pi_waiters, &p->pi_lock);
914         p->pi_blocked_on = NULL;
915 #endif
916 }
917
918 #ifdef CONFIG_MM_OWNER
919 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
920 {
921         mm->owner = p;
922 }
923 #endif /* CONFIG_MM_OWNER */
924
925 /*
926  * Initialize POSIX timer handling for a single task.
927  */
928 static void posix_cpu_timers_init(struct task_struct *tsk)
929 {
930         tsk->cputime_expires.prof_exp = cputime_zero;
931         tsk->cputime_expires.virt_exp = cputime_zero;
932         tsk->cputime_expires.sched_exp = 0;
933         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
934         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
935         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
936 }
937
938 /*
939  * This creates a new process as a copy of the old one,
940  * but does not actually start it yet.
941  *
942  * It copies the registers, and all the appropriate
943  * parts of the process environment (as per the clone
944  * flags). The actual kick-off is left to the caller.
945  */
946 static struct task_struct *copy_process(unsigned long clone_flags,
947                                         unsigned long stack_start,
948                                         struct pt_regs *regs,
949                                         unsigned long stack_size,
950                                         int __user *child_tidptr,
951                                         struct pid *pid,
952                                         int trace)
953 {
954         int retval;
955         struct task_struct *p;
956         int cgroup_callbacks_done = 0;
957
958         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
959                 return ERR_PTR(-EINVAL);
960
961         /*
962          * Thread groups must share signals as well, and detached threads
963          * can only be started up within the thread group.
964          */
965         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
966                 return ERR_PTR(-EINVAL);
967
968         /*
969          * Shared signal handlers imply shared VM. By way of the above,
970          * thread groups also imply shared VM. Blocking this case allows
971          * for various simplifications in other code.
972          */
973         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
974                 return ERR_PTR(-EINVAL);
975
976         retval = security_task_create(clone_flags);
977         if (retval)
978                 goto fork_out;
979
980         retval = -ENOMEM;
981         p = dup_task_struct(current);
982         if (!p)
983                 goto fork_out;
984
985         ftrace_graph_init_task(p);
986
987         rt_mutex_init_task(p);
988
989 #ifdef CONFIG_PROVE_LOCKING
990         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
991         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
992 #endif
993         retval = -EAGAIN;
994         if (atomic_read(&p->real_cred->user->processes) >=
995                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
996                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
997                     p->real_cred->user != INIT_USER)
998                         goto bad_fork_free;
999         }
1000
1001         retval = copy_creds(p, clone_flags);
1002         if (retval < 0)
1003                 goto bad_fork_free;
1004
1005         /*
1006          * If multiple threads are within copy_process(), then this check
1007          * triggers too late. This doesn't hurt, the check is only there
1008          * to stop root fork bombs.
1009          */
1010         retval = -EAGAIN;
1011         if (nr_threads >= max_threads)
1012                 goto bad_fork_cleanup_count;
1013
1014         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1015                 goto bad_fork_cleanup_count;
1016
1017         if (p->binfmt && !try_module_get(p->binfmt->module))
1018                 goto bad_fork_cleanup_put_domain;
1019
1020         p->did_exec = 0;
1021         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1022         copy_flags(clone_flags, p);
1023         INIT_LIST_HEAD(&p->children);
1024         INIT_LIST_HEAD(&p->sibling);
1025 #ifdef CONFIG_PREEMPT_RCU
1026         p->rcu_read_lock_nesting = 0;
1027         p->rcu_flipctr_idx = 0;
1028 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1029         p->vfork_done = NULL;
1030         spin_lock_init(&p->alloc_lock);
1031
1032         init_sigpending(&p->pending);
1033
1034         p->utime = cputime_zero;
1035         p->stime = cputime_zero;
1036         p->gtime = cputime_zero;
1037         p->utimescaled = cputime_zero;
1038         p->stimescaled = cputime_zero;
1039         p->prev_utime = cputime_zero;
1040         p->prev_stime = cputime_zero;
1041
1042         p->default_timer_slack_ns = current->timer_slack_ns;
1043
1044         task_io_accounting_init(&p->ioac);
1045         acct_clear_integrals(p);
1046
1047         posix_cpu_timers_init(p);
1048
1049         p->lock_depth = -1;             /* -1 = no lock */
1050         do_posix_clock_monotonic_gettime(&p->start_time);
1051         p->real_start_time = p->start_time;
1052         monotonic_to_bootbased(&p->real_start_time);
1053         p->io_context = NULL;
1054         p->audit_context = NULL;
1055         cgroup_fork(p);
1056 #ifdef CONFIG_NUMA
1057         p->mempolicy = mpol_dup(p->mempolicy);
1058         if (IS_ERR(p->mempolicy)) {
1059                 retval = PTR_ERR(p->mempolicy);
1060                 p->mempolicy = NULL;
1061                 goto bad_fork_cleanup_cgroup;
1062         }
1063         mpol_fix_fork_child_flag(p);
1064 #endif
1065 #ifdef CONFIG_TRACE_IRQFLAGS
1066         p->irq_events = 0;
1067 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1068         p->hardirqs_enabled = 1;
1069 #else
1070         p->hardirqs_enabled = 0;
1071 #endif
1072         p->hardirq_enable_ip = 0;
1073         p->hardirq_enable_event = 0;
1074         p->hardirq_disable_ip = _THIS_IP_;
1075         p->hardirq_disable_event = 0;
1076         p->softirqs_enabled = 1;
1077         p->softirq_enable_ip = _THIS_IP_;
1078         p->softirq_enable_event = 0;
1079         p->softirq_disable_ip = 0;
1080         p->softirq_disable_event = 0;
1081         p->hardirq_context = 0;
1082         p->softirq_context = 0;
1083 #endif
1084 #ifdef CONFIG_LOCKDEP
1085         p->lockdep_depth = 0; /* no locks held yet */
1086         p->curr_chain_key = 0;
1087         p->lockdep_recursion = 0;
1088 #endif
1089
1090 #ifdef CONFIG_DEBUG_MUTEXES
1091         p->blocked_on = NULL; /* not blocked yet */
1092 #endif
1093
1094         p->bts = NULL;
1095
1096         /* Perform scheduler related setup. Assign this task to a CPU. */
1097         sched_fork(p, clone_flags);
1098
1099         retval = perf_counter_init_task(p);
1100         if (retval)
1101                 goto bad_fork_cleanup_policy;
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(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         p->pid = pid_nr(pid);
1140         p->tgid = p->pid;
1141         if (clone_flags & CLONE_THREAD)
1142                 p->tgid = current->tgid;
1143
1144         if (current->nsproxy != p->nsproxy) {
1145                 retval = ns_cgroup_clone(p, pid);
1146                 if (retval)
1147                         goto bad_fork_free_pid;
1148         }
1149
1150         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1151         /*
1152          * Clear TID on mm_release()?
1153          */
1154         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1155 #ifdef CONFIG_FUTEX
1156         p->robust_list = NULL;
1157 #ifdef CONFIG_COMPAT
1158         p->compat_robust_list = NULL;
1159 #endif
1160         INIT_LIST_HEAD(&p->pi_state_list);
1161         p->pi_state_cache = NULL;
1162 #endif
1163         /*
1164          * sigaltstack should be cleared when sharing the same VM
1165          */
1166         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1167                 p->sas_ss_sp = p->sas_ss_size = 0;
1168
1169         /*
1170          * Syscall tracing should be turned off in the child regardless
1171          * of CLONE_PTRACE.
1172          */
1173         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1174 #ifdef TIF_SYSCALL_EMU
1175         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1176 #endif
1177         clear_all_latency_tracing(p);
1178
1179         /* ok, now we should be set up.. */
1180         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1181         p->pdeath_signal = 0;
1182         p->exit_state = 0;
1183
1184         /*
1185          * Ok, make it visible to the rest of the system.
1186          * We dont wake it up yet.
1187          */
1188         p->group_leader = p;
1189         INIT_LIST_HEAD(&p->thread_group);
1190
1191         /* Now that the task is set up, run cgroup callbacks if
1192          * necessary. We need to run them before the task is visible
1193          * on the tasklist. */
1194         cgroup_fork_callbacks(p);
1195         cgroup_callbacks_done = 1;
1196
1197         /* Need tasklist lock for parent etc handling! */
1198         write_lock_irq(&tasklist_lock);
1199
1200         /*
1201          * The task hasn't been attached yet, so its cpus_allowed mask will
1202          * not be changed, nor will its assigned CPU.
1203          *
1204          * The cpus_allowed mask of the parent may have changed after it was
1205          * copied first time - so re-copy it here, then check the child's CPU
1206          * to ensure it is on a valid CPU (and if not, just force it back to
1207          * parent's CPU). This avoids alot of nasty races.
1208          */
1209         p->cpus_allowed = current->cpus_allowed;
1210         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1211         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1212                         !cpu_online(task_cpu(p))))
1213                 set_task_cpu(p, smp_processor_id());
1214
1215         /* CLONE_PARENT re-uses the old parent */
1216         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1217                 p->real_parent = current->real_parent;
1218                 p->parent_exec_id = current->parent_exec_id;
1219         } else {
1220                 p->real_parent = current;
1221                 p->parent_exec_id = current->self_exec_id;
1222         }
1223
1224         spin_lock(&current->sighand->siglock);
1225
1226         /*
1227          * Process group and session signals need to be delivered to just the
1228          * parent before the fork or both the parent and the child after the
1229          * fork. Restart if a signal comes in before we add the new process to
1230          * it's process group.
1231          * A fatal signal pending means that current will exit, so the new
1232          * thread can't slip out of an OOM kill (or normal SIGKILL).
1233          */
1234         recalc_sigpending();
1235         if (signal_pending(current)) {
1236                 spin_unlock(&current->sighand->siglock);
1237                 write_unlock_irq(&tasklist_lock);
1238                 retval = -ERESTARTNOINTR;
1239                 goto bad_fork_free_pid;
1240         }
1241
1242         if (clone_flags & CLONE_THREAD) {
1243                 p->group_leader = current->group_leader;
1244                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1245         }
1246
1247         if (likely(p->pid)) {
1248                 list_add_tail(&p->sibling, &p->real_parent->children);
1249                 tracehook_finish_clone(p, clone_flags, trace);
1250
1251                 if (thread_group_leader(p)) {
1252                         if (clone_flags & CLONE_NEWPID)
1253                                 p->nsproxy->pid_ns->child_reaper = p;
1254
1255                         p->signal->leader_pid = pid;
1256                         tty_kref_put(p->signal->tty);
1257                         p->signal->tty = tty_kref_get(current->signal->tty);
1258                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1259                         attach_pid(p, PIDTYPE_SID, task_session(current));
1260                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1261                         __get_cpu_var(process_counts)++;
1262                 }
1263                 attach_pid(p, PIDTYPE_PID, pid);
1264                 nr_threads++;
1265         }
1266
1267         total_forks++;
1268         spin_unlock(&current->sighand->siglock);
1269         write_unlock_irq(&tasklist_lock);
1270         proc_fork_connector(p);
1271         cgroup_post_fork(p);
1272         perf_counter_fork(p);
1273         return p;
1274
1275 bad_fork_free_pid:
1276         if (pid != &init_struct_pid)
1277                 free_pid(pid);
1278 bad_fork_cleanup_io:
1279         put_io_context(p->io_context);
1280 bad_fork_cleanup_namespaces:
1281         exit_task_namespaces(p);
1282 bad_fork_cleanup_mm:
1283         if (p->mm)
1284                 mmput(p->mm);
1285 bad_fork_cleanup_signal:
1286         cleanup_signal(p);
1287 bad_fork_cleanup_sighand:
1288         __cleanup_sighand(p->sighand);
1289 bad_fork_cleanup_fs:
1290         exit_fs(p); /* blocking */
1291 bad_fork_cleanup_files:
1292         exit_files(p); /* blocking */
1293 bad_fork_cleanup_semundo:
1294         exit_sem(p);
1295 bad_fork_cleanup_audit:
1296         audit_free(p);
1297 bad_fork_cleanup_policy:
1298         perf_counter_free_task(p);
1299 #ifdef CONFIG_NUMA
1300         mpol_put(p->mempolicy);
1301 bad_fork_cleanup_cgroup:
1302 #endif
1303         cgroup_exit(p, cgroup_callbacks_done);
1304         delayacct_tsk_free(p);
1305         if (p->binfmt)
1306                 module_put(p->binfmt->module);
1307 bad_fork_cleanup_put_domain:
1308         module_put(task_thread_info(p)->exec_domain->module);
1309 bad_fork_cleanup_count:
1310         atomic_dec(&p->cred->user->processes);
1311         put_cred(p->real_cred);
1312         put_cred(p->cred);
1313 bad_fork_free:
1314         free_task(p);
1315 fork_out:
1316         return ERR_PTR(retval);
1317 }
1318
1319 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1320 {
1321         memset(regs, 0, sizeof(struct pt_regs));
1322         return regs;
1323 }
1324
1325 struct task_struct * __cpuinit fork_idle(int cpu)
1326 {
1327         struct task_struct *task;
1328         struct pt_regs regs;
1329
1330         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1331                             &init_struct_pid, 0);
1332         if (!IS_ERR(task))
1333                 init_idle(task, cpu);
1334
1335         return task;
1336 }
1337
1338 /*
1339  *  Ok, this is the main fork-routine.
1340  *
1341  * It copies the process, and if successful kick-starts
1342  * it and waits for it to finish using the VM if required.
1343  */
1344 long do_fork(unsigned long clone_flags,
1345               unsigned long stack_start,
1346               struct pt_regs *regs,
1347               unsigned long stack_size,
1348               int __user *parent_tidptr,
1349               int __user *child_tidptr)
1350 {
1351         struct task_struct *p;
1352         int trace = 0;
1353         long nr;
1354
1355         /*
1356          * Do some preliminary argument and permissions checking before we
1357          * actually start allocating stuff
1358          */
1359         if (clone_flags & CLONE_NEWUSER) {
1360                 if (clone_flags & CLONE_THREAD)
1361                         return -EINVAL;
1362                 /* hopefully this check will go away when userns support is
1363                  * complete
1364                  */
1365                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1366                                 !capable(CAP_SETGID))
1367                         return -EPERM;
1368         }
1369
1370         /*
1371          * We hope to recycle these flags after 2.6.26
1372          */
1373         if (unlikely(clone_flags & CLONE_STOPPED)) {
1374                 static int __read_mostly count = 100;
1375
1376                 if (count > 0 && printk_ratelimit()) {
1377                         char comm[TASK_COMM_LEN];
1378
1379                         count--;
1380                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1381                                         "clone flags 0x%lx\n",
1382                                 get_task_comm(comm, current),
1383                                 clone_flags & CLONE_STOPPED);
1384                 }
1385         }
1386
1387         /*
1388          * When called from kernel_thread, don't do user tracing stuff.
1389          */
1390         if (likely(user_mode(regs)))
1391                 trace = tracehook_prepare_clone(clone_flags);
1392
1393         p = copy_process(clone_flags, stack_start, regs, stack_size,
1394                          child_tidptr, NULL, trace);
1395         /*
1396          * Do this prior waking up the new thread - the thread pointer
1397          * might get invalid after that point, if the thread exits quickly.
1398          */
1399         if (!IS_ERR(p)) {
1400                 struct completion vfork;
1401
1402                 trace_sched_process_fork(current, p);
1403
1404                 nr = task_pid_vnr(p);
1405
1406                 if (clone_flags & CLONE_PARENT_SETTID)
1407                         put_user(nr, parent_tidptr);
1408
1409                 if (clone_flags & CLONE_VFORK) {
1410                         p->vfork_done = &vfork;
1411                         init_completion(&vfork);
1412                 }
1413
1414                 audit_finish_fork(p);
1415                 tracehook_report_clone(regs, clone_flags, nr, p);
1416
1417                 /*
1418                  * We set PF_STARTING at creation in case tracing wants to
1419                  * use this to distinguish a fully live task from one that
1420                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1421                  * clear it and set the child going.
1422                  */
1423                 p->flags &= ~PF_STARTING;
1424
1425                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1426                         /*
1427                          * We'll start up with an immediate SIGSTOP.
1428                          */
1429                         sigaddset(&p->pending.signal, SIGSTOP);
1430                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1431                         __set_task_state(p, TASK_STOPPED);
1432                 } else {
1433                         wake_up_new_task(p, clone_flags);
1434                 }
1435
1436                 tracehook_report_clone_complete(trace, regs,
1437                                                 clone_flags, nr, p);
1438
1439                 if (clone_flags & CLONE_VFORK) {
1440                         freezer_do_not_count();
1441                         wait_for_completion(&vfork);
1442                         freezer_count();
1443                         tracehook_report_vfork_done(p, nr);
1444                 }
1445         } else {
1446                 nr = PTR_ERR(p);
1447         }
1448         return nr;
1449 }
1450
1451 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1452 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1453 #endif
1454
1455 static void sighand_ctor(void *data)
1456 {
1457         struct sighand_struct *sighand = data;
1458
1459         spin_lock_init(&sighand->siglock);
1460         init_waitqueue_head(&sighand->signalfd_wqh);
1461 }
1462
1463 void __init proc_caches_init(void)
1464 {
1465         sighand_cachep = kmem_cache_create("sighand_cache",
1466                         sizeof(struct sighand_struct), 0,
1467                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1468                         SLAB_NOTRACK, sighand_ctor);
1469         signal_cachep = kmem_cache_create("signal_cache",
1470                         sizeof(struct signal_struct), 0,
1471                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1472         files_cachep = kmem_cache_create("files_cache",
1473                         sizeof(struct files_struct), 0,
1474                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1475         fs_cachep = kmem_cache_create("fs_cache",
1476                         sizeof(struct fs_struct), 0,
1477                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1478         mm_cachep = kmem_cache_create("mm_struct",
1479                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1480                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1481         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1482         mmap_init();
1483 }
1484
1485 /*
1486  * Check constraints on flags passed to the unshare system call and
1487  * force unsharing of additional process context as appropriate.
1488  */
1489 static void check_unshare_flags(unsigned long *flags_ptr)
1490 {
1491         /*
1492          * If unsharing a thread from a thread group, must also
1493          * unshare vm.
1494          */
1495         if (*flags_ptr & CLONE_THREAD)
1496                 *flags_ptr |= CLONE_VM;
1497
1498         /*
1499          * If unsharing vm, must also unshare signal handlers.
1500          */
1501         if (*flags_ptr & CLONE_VM)
1502                 *flags_ptr |= CLONE_SIGHAND;
1503
1504         /*
1505          * If unsharing signal handlers and the task was created
1506          * using CLONE_THREAD, then must unshare the thread
1507          */
1508         if ((*flags_ptr & CLONE_SIGHAND) &&
1509             (atomic_read(&current->signal->count) > 1))
1510                 *flags_ptr |= CLONE_THREAD;
1511
1512         /*
1513          * If unsharing namespace, must also unshare filesystem information.
1514          */
1515         if (*flags_ptr & CLONE_NEWNS)
1516                 *flags_ptr |= CLONE_FS;
1517 }
1518
1519 /*
1520  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1521  */
1522 static int unshare_thread(unsigned long unshare_flags)
1523 {
1524         if (unshare_flags & CLONE_THREAD)
1525                 return -EINVAL;
1526
1527         return 0;
1528 }
1529
1530 /*
1531  * Unshare the filesystem structure if it is being shared
1532  */
1533 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1534 {
1535         struct fs_struct *fs = current->fs;
1536
1537         if (!(unshare_flags & CLONE_FS) || !fs)
1538                 return 0;
1539
1540         /* don't need lock here; in the worst case we'll do useless copy */
1541         if (fs->users == 1)
1542                 return 0;
1543
1544         *new_fsp = copy_fs_struct(fs);
1545         if (!*new_fsp)
1546                 return -ENOMEM;
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                         write_lock(&fs->lock);
1663                         current->fs = new_fs;
1664                         if (--fs->users)
1665                                 new_fs = NULL;
1666                         else
1667                                 new_fs = fs;
1668                         write_unlock(&fs->lock);
1669                 }
1670
1671                 if (new_mm) {
1672                         mm = current->mm;
1673                         active_mm = current->active_mm;
1674                         current->mm = new_mm;
1675                         current->active_mm = new_mm;
1676                         activate_mm(active_mm, new_mm);
1677                         new_mm = mm;
1678                 }
1679
1680                 if (new_fd) {
1681                         fd = current->files;
1682                         current->files = new_fd;
1683                         new_fd = fd;
1684                 }
1685
1686                 task_unlock(current);
1687         }
1688
1689         if (new_nsproxy)
1690                 put_nsproxy(new_nsproxy);
1691
1692 bad_unshare_cleanup_fd:
1693         if (new_fd)
1694                 put_files_struct(new_fd);
1695
1696 bad_unshare_cleanup_vm:
1697         if (new_mm)
1698                 mmput(new_mm);
1699
1700 bad_unshare_cleanup_sigh:
1701         if (new_sigh)
1702                 if (atomic_dec_and_test(&new_sigh->count))
1703                         kmem_cache_free(sighand_cachep, new_sigh);
1704
1705 bad_unshare_cleanup_fs:
1706         if (new_fs)
1707                 free_fs_struct(new_fs);
1708
1709 bad_unshare_cleanup_thread:
1710 bad_unshare_out:
1711         return err;
1712 }
1713
1714 /*
1715  *      Helper to unshare the files of the current task.
1716  *      We don't want to expose copy_files internals to
1717  *      the exec layer of the kernel.
1718  */
1719
1720 int unshare_files(struct files_struct **displaced)
1721 {
1722         struct task_struct *task = current;
1723         struct files_struct *copy = NULL;
1724         int error;
1725
1726         error = unshare_fd(CLONE_FILES, &copy);
1727         if (error || !copy) {
1728                 *displaced = NULL;
1729                 return error;
1730         }
1731         *displaced = task->files;
1732         task_lock(task);
1733         task->files = copy;
1734         task_unlock(task);
1735         return 0;
1736 }