tracing/events: move trace point headers into include/trace/events
[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 <linux/fs_struct.h>
64 #include <linux/magic.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, 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->core_state = NULL;
430         mm->nr_ptes = 0;
431         set_mm_counter(mm, file_rss, 0);
432         set_mm_counter(mm, anon_rss, 0);
433         spin_lock_init(&mm->page_table_lock);
434         spin_lock_init(&mm->ioctx_lock);
435         INIT_HLIST_HEAD(&mm->ioctx_list);
436         mm->free_area_cache = TASK_UNMAPPED_BASE;
437         mm->cached_hole_size = ~0UL;
438         mm_init_owner(mm, p);
439
440         if (likely(!mm_alloc_pgd(mm))) {
441                 mm->def_flags = 0;
442                 mmu_notifier_mm_init(mm);
443                 return mm;
444         }
445
446         free_mm(mm);
447         return NULL;
448 }
449
450 /*
451  * Allocate and initialize an mm_struct.
452  */
453 struct mm_struct * mm_alloc(void)
454 {
455         struct mm_struct * mm;
456
457         mm = allocate_mm();
458         if (mm) {
459                 memset(mm, 0, sizeof(*mm));
460                 mm = mm_init(mm, current);
461         }
462         return mm;
463 }
464
465 /*
466  * Called when the last reference to the mm
467  * is dropped: either by a lazy thread or by
468  * mmput. Free the page directory and the mm.
469  */
470 void __mmdrop(struct mm_struct *mm)
471 {
472         BUG_ON(mm == &init_mm);
473         mm_free_pgd(mm);
474         destroy_context(mm);
475         mmu_notifier_mm_destroy(mm);
476         free_mm(mm);
477 }
478 EXPORT_SYMBOL_GPL(__mmdrop);
479
480 /*
481  * Decrement the use count and release all resources for an mm.
482  */
483 void mmput(struct mm_struct *mm)
484 {
485         might_sleep();
486
487         if (atomic_dec_and_test(&mm->mm_users)) {
488                 exit_aio(mm);
489                 exit_mmap(mm);
490                 set_mm_exe_file(mm, NULL);
491                 if (!list_empty(&mm->mmlist)) {
492                         spin_lock(&mmlist_lock);
493                         list_del(&mm->mmlist);
494                         spin_unlock(&mmlist_lock);
495                 }
496                 put_swap_token(mm);
497                 mmdrop(mm);
498         }
499 }
500 EXPORT_SYMBOL_GPL(mmput);
501
502 /**
503  * get_task_mm - acquire a reference to the task's mm
504  *
505  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
506  * this kernel workthread has transiently adopted a user mm with use_mm,
507  * to do its AIO) is not set and if so returns a reference to it, after
508  * bumping up the use count.  User must release the mm via mmput()
509  * after use.  Typically used by /proc and ptrace.
510  */
511 struct mm_struct *get_task_mm(struct task_struct *task)
512 {
513         struct mm_struct *mm;
514
515         task_lock(task);
516         mm = task->mm;
517         if (mm) {
518                 if (task->flags & PF_KTHREAD)
519                         mm = NULL;
520                 else
521                         atomic_inc(&mm->mm_users);
522         }
523         task_unlock(task);
524         return mm;
525 }
526 EXPORT_SYMBOL_GPL(get_task_mm);
527
528 /* Please note the differences between mmput and mm_release.
529  * mmput is called whenever we stop holding onto a mm_struct,
530  * error success whatever.
531  *
532  * mm_release is called after a mm_struct has been removed
533  * from the current process.
534  *
535  * This difference is important for error handling, when we
536  * only half set up a mm_struct for a new process and need to restore
537  * the old one.  Because we mmput the new mm_struct before
538  * restoring the old one. . .
539  * Eric Biederman 10 January 1998
540  */
541 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
542 {
543         struct completion *vfork_done = tsk->vfork_done;
544
545         /* Get rid of any futexes when releasing the mm */
546 #ifdef CONFIG_FUTEX
547         if (unlikely(tsk->robust_list))
548                 exit_robust_list(tsk);
549 #ifdef CONFIG_COMPAT
550         if (unlikely(tsk->compat_robust_list))
551                 compat_exit_robust_list(tsk);
552 #endif
553 #endif
554
555         /* Get rid of any cached register state */
556         deactivate_mm(tsk, mm);
557
558         /* notify parent sleeping on vfork() */
559         if (vfork_done) {
560                 tsk->vfork_done = NULL;
561                 complete(vfork_done);
562         }
563
564         /*
565          * If we're exiting normally, clear a user-space tid field if
566          * requested.  We leave this alone when dying by signal, to leave
567          * the value intact in a core dump, and to save the unnecessary
568          * trouble otherwise.  Userland only wants this done for a sys_exit.
569          */
570         if (tsk->clear_child_tid
571             && !(tsk->flags & PF_SIGNALED)
572             && atomic_read(&mm->mm_users) > 1) {
573                 u32 __user * tidptr = tsk->clear_child_tid;
574                 tsk->clear_child_tid = NULL;
575
576                 /*
577                  * We don't check the error code - if userspace has
578                  * not set up a proper pointer then tough luck.
579                  */
580                 put_user(0, tidptr);
581                 sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
582         }
583 }
584
585 /*
586  * Allocate a new mm structure and copy contents from the
587  * mm structure of the passed in task structure.
588  */
589 struct mm_struct *dup_mm(struct task_struct *tsk)
590 {
591         struct mm_struct *mm, *oldmm = current->mm;
592         int err;
593
594         if (!oldmm)
595                 return NULL;
596
597         mm = allocate_mm();
598         if (!mm)
599                 goto fail_nomem;
600
601         memcpy(mm, oldmm, sizeof(*mm));
602
603         /* Initializing for Swap token stuff */
604         mm->token_priority = 0;
605         mm->last_interval = 0;
606
607         if (!mm_init(mm, tsk))
608                 goto fail_nomem;
609
610         if (init_new_context(tsk, mm))
611                 goto fail_nocontext;
612
613         dup_mm_exe_file(oldmm, mm);
614
615         err = dup_mmap(mm, oldmm);
616         if (err)
617                 goto free_pt;
618
619         mm->hiwater_rss = get_mm_rss(mm);
620         mm->hiwater_vm = mm->total_vm;
621
622         return mm;
623
624 free_pt:
625         mmput(mm);
626
627 fail_nomem:
628         return NULL;
629
630 fail_nocontext:
631         /*
632          * If init_new_context() failed, we cannot use mmput() to free the mm
633          * because it calls destroy_context()
634          */
635         mm_free_pgd(mm);
636         free_mm(mm);
637         return NULL;
638 }
639
640 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
641 {
642         struct mm_struct * mm, *oldmm;
643         int retval;
644
645         tsk->min_flt = tsk->maj_flt = 0;
646         tsk->nvcsw = tsk->nivcsw = 0;
647 #ifdef CONFIG_DETECT_HUNG_TASK
648         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
649 #endif
650
651         tsk->mm = NULL;
652         tsk->active_mm = NULL;
653
654         /*
655          * Are we cloning a kernel thread?
656          *
657          * We need to steal a active VM for that..
658          */
659         oldmm = current->mm;
660         if (!oldmm)
661                 return 0;
662
663         if (clone_flags & CLONE_VM) {
664                 atomic_inc(&oldmm->mm_users);
665                 mm = oldmm;
666                 goto good_mm;
667         }
668
669         retval = -ENOMEM;
670         mm = dup_mm(tsk);
671         if (!mm)
672                 goto fail_nomem;
673
674 good_mm:
675         /* Initializing for Swap token stuff */
676         mm->token_priority = 0;
677         mm->last_interval = 0;
678
679         tsk->mm = mm;
680         tsk->active_mm = mm;
681         return 0;
682
683 fail_nomem:
684         return retval;
685 }
686
687 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
688 {
689         struct fs_struct *fs = current->fs;
690         if (clone_flags & CLONE_FS) {
691                 /* tsk->fs is already what we want */
692                 write_lock(&fs->lock);
693                 if (fs->in_exec) {
694                         write_unlock(&fs->lock);
695                         return -EAGAIN;
696                 }
697                 fs->users++;
698                 write_unlock(&fs->lock);
699                 return 0;
700         }
701         tsk->fs = copy_fs_struct(fs);
702         if (!tsk->fs)
703                 return -ENOMEM;
704         return 0;
705 }
706
707 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
708 {
709         struct files_struct *oldf, *newf;
710         int error = 0;
711
712         /*
713          * A background process may not have any files ...
714          */
715         oldf = current->files;
716         if (!oldf)
717                 goto out;
718
719         if (clone_flags & CLONE_FILES) {
720                 atomic_inc(&oldf->count);
721                 goto out;
722         }
723
724         newf = dup_fd(oldf, &error);
725         if (!newf)
726                 goto out;
727
728         tsk->files = newf;
729         error = 0;
730 out:
731         return error;
732 }
733
734 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
735 {
736 #ifdef CONFIG_BLOCK
737         struct io_context *ioc = current->io_context;
738
739         if (!ioc)
740                 return 0;
741         /*
742          * Share io context with parent, if CLONE_IO is set
743          */
744         if (clone_flags & CLONE_IO) {
745                 tsk->io_context = ioc_task_link(ioc);
746                 if (unlikely(!tsk->io_context))
747                         return -ENOMEM;
748         } else if (ioprio_valid(ioc->ioprio)) {
749                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
750                 if (unlikely(!tsk->io_context))
751                         return -ENOMEM;
752
753                 tsk->io_context->ioprio = ioc->ioprio;
754         }
755 #endif
756         return 0;
757 }
758
759 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
760 {
761         struct sighand_struct *sig;
762
763         if (clone_flags & CLONE_SIGHAND) {
764                 atomic_inc(&current->sighand->count);
765                 return 0;
766         }
767         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
768         rcu_assign_pointer(tsk->sighand, sig);
769         if (!sig)
770                 return -ENOMEM;
771         atomic_set(&sig->count, 1);
772         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
773         return 0;
774 }
775
776 void __cleanup_sighand(struct sighand_struct *sighand)
777 {
778         if (atomic_dec_and_test(&sighand->count))
779                 kmem_cache_free(sighand_cachep, sighand);
780 }
781
782
783 /*
784  * Initialize POSIX timer handling for a thread group.
785  */
786 static void posix_cpu_timers_init_group(struct signal_struct *sig)
787 {
788         /* Thread group counters. */
789         thread_group_cputime_init(sig);
790
791         /* Expiration times and increments. */
792         sig->it_virt_expires = cputime_zero;
793         sig->it_virt_incr = cputime_zero;
794         sig->it_prof_expires = cputime_zero;
795         sig->it_prof_incr = cputime_zero;
796
797         /* Cached expiration times. */
798         sig->cputime_expires.prof_exp = cputime_zero;
799         sig->cputime_expires.virt_exp = cputime_zero;
800         sig->cputime_expires.sched_exp = 0;
801
802         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
803                 sig->cputime_expires.prof_exp =
804                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
805                 sig->cputimer.running = 1;
806         }
807
808         /* The timer lists. */
809         INIT_LIST_HEAD(&sig->cpu_timers[0]);
810         INIT_LIST_HEAD(&sig->cpu_timers[1]);
811         INIT_LIST_HEAD(&sig->cpu_timers[2]);
812 }
813
814 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
815 {
816         struct signal_struct *sig;
817
818         if (clone_flags & CLONE_THREAD) {
819                 atomic_inc(&current->signal->count);
820                 atomic_inc(&current->signal->live);
821                 return 0;
822         }
823
824         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
825         tsk->signal = sig;
826         if (!sig)
827                 return -ENOMEM;
828
829         atomic_set(&sig->count, 1);
830         atomic_set(&sig->live, 1);
831         init_waitqueue_head(&sig->wait_chldexit);
832         sig->flags = 0;
833         if (clone_flags & CLONE_NEWPID)
834                 sig->flags |= SIGNAL_UNKILLABLE;
835         sig->group_exit_code = 0;
836         sig->group_exit_task = NULL;
837         sig->group_stop_count = 0;
838         sig->curr_target = tsk;
839         init_sigpending(&sig->shared_pending);
840         INIT_LIST_HEAD(&sig->posix_timers);
841
842         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
843         sig->it_real_incr.tv64 = 0;
844         sig->real_timer.function = it_real_fn;
845
846         sig->leader = 0;        /* session leadership doesn't inherit */
847         sig->tty_old_pgrp = NULL;
848         sig->tty = NULL;
849
850         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
851         sig->gtime = cputime_zero;
852         sig->cgtime = cputime_zero;
853         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
854         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
855         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
856         task_io_accounting_init(&sig->ioac);
857         sig->sum_sched_runtime = 0;
858         taskstats_tgid_init(sig);
859
860         task_lock(current->group_leader);
861         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
862         task_unlock(current->group_leader);
863
864         posix_cpu_timers_init_group(sig);
865
866         acct_init_pacct(&sig->pacct);
867
868         tty_audit_fork(sig);
869
870         return 0;
871 }
872
873 void __cleanup_signal(struct signal_struct *sig)
874 {
875         thread_group_cputime_free(sig);
876         tty_kref_put(sig->tty);
877         kmem_cache_free(signal_cachep, sig);
878 }
879
880 static void cleanup_signal(struct task_struct *tsk)
881 {
882         struct signal_struct *sig = tsk->signal;
883
884         atomic_dec(&sig->live);
885
886         if (atomic_dec_and_test(&sig->count))
887                 __cleanup_signal(sig);
888 }
889
890 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
891 {
892         unsigned long new_flags = p->flags;
893
894         new_flags &= ~PF_SUPERPRIV;
895         new_flags |= PF_FORKNOEXEC;
896         new_flags |= PF_STARTING;
897         p->flags = new_flags;
898         clear_freeze_flag(p);
899 }
900
901 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
902 {
903         current->clear_child_tid = tidptr;
904
905         return task_pid_vnr(current);
906 }
907
908 static void rt_mutex_init_task(struct task_struct *p)
909 {
910         spin_lock_init(&p->pi_lock);
911 #ifdef CONFIG_RT_MUTEXES
912         plist_head_init(&p->pi_waiters, &p->pi_lock);
913         p->pi_blocked_on = NULL;
914 #endif
915 }
916
917 #ifdef CONFIG_MM_OWNER
918 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
919 {
920         mm->owner = p;
921 }
922 #endif /* CONFIG_MM_OWNER */
923
924 /*
925  * Initialize POSIX timer handling for a single task.
926  */
927 static void posix_cpu_timers_init(struct task_struct *tsk)
928 {
929         tsk->cputime_expires.prof_exp = cputime_zero;
930         tsk->cputime_expires.virt_exp = cputime_zero;
931         tsk->cputime_expires.sched_exp = 0;
932         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
933         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
934         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
935 }
936
937 /*
938  * This creates a new process as a copy of the old one,
939  * but does not actually start it yet.
940  *
941  * It copies the registers, and all the appropriate
942  * parts of the process environment (as per the clone
943  * flags). The actual kick-off is left to the caller.
944  */
945 static struct task_struct *copy_process(unsigned long clone_flags,
946                                         unsigned long stack_start,
947                                         struct pt_regs *regs,
948                                         unsigned long stack_size,
949                                         int __user *child_tidptr,
950                                         struct pid *pid,
951                                         int trace)
952 {
953         int retval;
954         struct task_struct *p;
955         int cgroup_callbacks_done = 0;
956
957         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
958                 return ERR_PTR(-EINVAL);
959
960         /*
961          * Thread groups must share signals as well, and detached threads
962          * can only be started up within the thread group.
963          */
964         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
965                 return ERR_PTR(-EINVAL);
966
967         /*
968          * Shared signal handlers imply shared VM. By way of the above,
969          * thread groups also imply shared VM. Blocking this case allows
970          * for various simplifications in other code.
971          */
972         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
973                 return ERR_PTR(-EINVAL);
974
975         retval = security_task_create(clone_flags);
976         if (retval)
977                 goto fork_out;
978
979         retval = -ENOMEM;
980         p = dup_task_struct(current);
981         if (!p)
982                 goto fork_out;
983
984         rt_mutex_init_task(p);
985
986 #ifdef CONFIG_PROVE_LOCKING
987         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
988         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
989 #endif
990         retval = -EAGAIN;
991         if (atomic_read(&p->real_cred->user->processes) >=
992                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
993                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
994                     p->real_cred->user != INIT_USER)
995                         goto bad_fork_free;
996         }
997
998         retval = copy_creds(p, clone_flags);
999         if (retval < 0)
1000                 goto bad_fork_free;
1001
1002         /*
1003          * If multiple threads are within copy_process(), then this check
1004          * triggers too late. This doesn't hurt, the check is only there
1005          * to stop root fork bombs.
1006          */
1007         retval = -EAGAIN;
1008         if (nr_threads >= max_threads)
1009                 goto bad_fork_cleanup_count;
1010
1011         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1012                 goto bad_fork_cleanup_count;
1013
1014         if (p->binfmt && !try_module_get(p->binfmt->module))
1015                 goto bad_fork_cleanup_put_domain;
1016
1017         p->did_exec = 0;
1018         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1019         copy_flags(clone_flags, p);
1020         INIT_LIST_HEAD(&p->children);
1021         INIT_LIST_HEAD(&p->sibling);
1022 #ifdef CONFIG_PREEMPT_RCU
1023         p->rcu_read_lock_nesting = 0;
1024         p->rcu_flipctr_idx = 0;
1025 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1026         p->vfork_done = NULL;
1027         spin_lock_init(&p->alloc_lock);
1028
1029         clear_tsk_thread_flag(p, TIF_SIGPENDING);
1030         init_sigpending(&p->pending);
1031
1032         p->utime = cputime_zero;
1033         p->stime = cputime_zero;
1034         p->gtime = cputime_zero;
1035         p->utimescaled = cputime_zero;
1036         p->stimescaled = cputime_zero;
1037         p->prev_utime = cputime_zero;
1038         p->prev_stime = cputime_zero;
1039
1040         p->default_timer_slack_ns = current->timer_slack_ns;
1041
1042         task_io_accounting_init(&p->ioac);
1043         acct_clear_integrals(p);
1044
1045         posix_cpu_timers_init(p);
1046
1047         p->lock_depth = -1;             /* -1 = no lock */
1048         do_posix_clock_monotonic_gettime(&p->start_time);
1049         p->real_start_time = p->start_time;
1050         monotonic_to_bootbased(&p->real_start_time);
1051         p->io_context = NULL;
1052         p->audit_context = NULL;
1053         cgroup_fork(p);
1054 #ifdef CONFIG_NUMA
1055         p->mempolicy = mpol_dup(p->mempolicy);
1056         if (IS_ERR(p->mempolicy)) {
1057                 retval = PTR_ERR(p->mempolicy);
1058                 p->mempolicy = NULL;
1059                 goto bad_fork_cleanup_cgroup;
1060         }
1061         mpol_fix_fork_child_flag(p);
1062 #endif
1063 #ifdef CONFIG_TRACE_IRQFLAGS
1064         p->irq_events = 0;
1065 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1066         p->hardirqs_enabled = 1;
1067 #else
1068         p->hardirqs_enabled = 0;
1069 #endif
1070         p->hardirq_enable_ip = 0;
1071         p->hardirq_enable_event = 0;
1072         p->hardirq_disable_ip = _THIS_IP_;
1073         p->hardirq_disable_event = 0;
1074         p->softirqs_enabled = 1;
1075         p->softirq_enable_ip = _THIS_IP_;
1076         p->softirq_enable_event = 0;
1077         p->softirq_disable_ip = 0;
1078         p->softirq_disable_event = 0;
1079         p->hardirq_context = 0;
1080         p->softirq_context = 0;
1081 #endif
1082 #ifdef CONFIG_LOCKDEP
1083         p->lockdep_depth = 0; /* no locks held yet */
1084         p->curr_chain_key = 0;
1085         p->lockdep_recursion = 0;
1086 #endif
1087
1088 #ifdef CONFIG_DEBUG_MUTEXES
1089         p->blocked_on = NULL; /* not blocked yet */
1090 #endif
1091         if (unlikely(current->ptrace))
1092                 ptrace_fork(p, clone_flags);
1093
1094         /* Perform scheduler related setup. Assign this task to a CPU. */
1095         sched_fork(p, clone_flags);
1096
1097         if ((retval = audit_alloc(p)))
1098                 goto bad_fork_cleanup_policy;
1099         /* copy all the process information */
1100         if ((retval = copy_semundo(clone_flags, p)))
1101                 goto bad_fork_cleanup_audit;
1102         if ((retval = copy_files(clone_flags, p)))
1103                 goto bad_fork_cleanup_semundo;
1104         if ((retval = copy_fs(clone_flags, p)))
1105                 goto bad_fork_cleanup_files;
1106         if ((retval = copy_sighand(clone_flags, p)))
1107                 goto bad_fork_cleanup_fs;
1108         if ((retval = copy_signal(clone_flags, p)))
1109                 goto bad_fork_cleanup_sighand;
1110         if ((retval = copy_mm(clone_flags, p)))
1111                 goto bad_fork_cleanup_signal;
1112         if ((retval = copy_namespaces(clone_flags, p)))
1113                 goto bad_fork_cleanup_mm;
1114         if ((retval = copy_io(clone_flags, p)))
1115                 goto bad_fork_cleanup_namespaces;
1116         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1117         if (retval)
1118                 goto bad_fork_cleanup_io;
1119
1120         if (pid != &init_struct_pid) {
1121                 retval = -ENOMEM;
1122                 pid = alloc_pid(p->nsproxy->pid_ns);
1123                 if (!pid)
1124                         goto bad_fork_cleanup_io;
1125
1126                 if (clone_flags & CLONE_NEWPID) {
1127                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1128                         if (retval < 0)
1129                                 goto bad_fork_free_pid;
1130                 }
1131         }
1132
1133         ftrace_graph_init_task(p);
1134
1135         p->pid = pid_nr(pid);
1136         p->tgid = p->pid;
1137         if (clone_flags & CLONE_THREAD)
1138                 p->tgid = current->tgid;
1139
1140         if (current->nsproxy != p->nsproxy) {
1141                 retval = ns_cgroup_clone(p, pid);
1142                 if (retval)
1143                         goto bad_fork_free_graph;
1144         }
1145
1146         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1147         /*
1148          * Clear TID on mm_release()?
1149          */
1150         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1151 #ifdef CONFIG_FUTEX
1152         p->robust_list = NULL;
1153 #ifdef CONFIG_COMPAT
1154         p->compat_robust_list = NULL;
1155 #endif
1156         INIT_LIST_HEAD(&p->pi_state_list);
1157         p->pi_state_cache = NULL;
1158 #endif
1159         /*
1160          * sigaltstack should be cleared when sharing the same VM
1161          */
1162         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1163                 p->sas_ss_sp = p->sas_ss_size = 0;
1164
1165         /*
1166          * Syscall tracing should be turned off in the child regardless
1167          * of CLONE_PTRACE.
1168          */
1169         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1170 #ifdef TIF_SYSCALL_EMU
1171         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1172 #endif
1173         clear_all_latency_tracing(p);
1174
1175         /* ok, now we should be set up.. */
1176         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1177         p->pdeath_signal = 0;
1178         p->exit_state = 0;
1179
1180         /*
1181          * Ok, make it visible to the rest of the system.
1182          * We dont wake it up yet.
1183          */
1184         p->group_leader = p;
1185         INIT_LIST_HEAD(&p->thread_group);
1186
1187         /* Now that the task is set up, run cgroup callbacks if
1188          * necessary. We need to run them before the task is visible
1189          * on the tasklist. */
1190         cgroup_fork_callbacks(p);
1191         cgroup_callbacks_done = 1;
1192
1193         /* Need tasklist lock for parent etc handling! */
1194         write_lock_irq(&tasklist_lock);
1195
1196         /*
1197          * The task hasn't been attached yet, so its cpus_allowed mask will
1198          * not be changed, nor will its assigned CPU.
1199          *
1200          * The cpus_allowed mask of the parent may have changed after it was
1201          * copied first time - so re-copy it here, then check the child's CPU
1202          * to ensure it is on a valid CPU (and if not, just force it back to
1203          * parent's CPU). This avoids alot of nasty races.
1204          */
1205         p->cpus_allowed = current->cpus_allowed;
1206         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1207         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1208                         !cpu_online(task_cpu(p))))
1209                 set_task_cpu(p, smp_processor_id());
1210
1211         /* CLONE_PARENT re-uses the old parent */
1212         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1213                 p->real_parent = current->real_parent;
1214                 p->parent_exec_id = current->parent_exec_id;
1215         } else {
1216                 p->real_parent = current;
1217                 p->parent_exec_id = current->self_exec_id;
1218         }
1219
1220         spin_lock(&current->sighand->siglock);
1221
1222         /*
1223          * Process group and session signals need to be delivered to just the
1224          * parent before the fork or both the parent and the child after the
1225          * fork. Restart if a signal comes in before we add the new process to
1226          * it's process group.
1227          * A fatal signal pending means that current will exit, so the new
1228          * thread can't slip out of an OOM kill (or normal SIGKILL).
1229          */
1230         recalc_sigpending();
1231         if (signal_pending(current)) {
1232                 spin_unlock(&current->sighand->siglock);
1233                 write_unlock_irq(&tasklist_lock);
1234                 retval = -ERESTARTNOINTR;
1235                 goto bad_fork_free_graph;
1236         }
1237
1238         if (clone_flags & CLONE_THREAD) {
1239                 p->group_leader = current->group_leader;
1240                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1241         }
1242
1243         if (likely(p->pid)) {
1244                 list_add_tail(&p->sibling, &p->real_parent->children);
1245                 tracehook_finish_clone(p, clone_flags, trace);
1246
1247                 if (thread_group_leader(p)) {
1248                         if (clone_flags & CLONE_NEWPID)
1249                                 p->nsproxy->pid_ns->child_reaper = p;
1250
1251                         p->signal->leader_pid = pid;
1252                         tty_kref_put(p->signal->tty);
1253                         p->signal->tty = tty_kref_get(current->signal->tty);
1254                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1255                         attach_pid(p, PIDTYPE_SID, task_session(current));
1256                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1257                         __get_cpu_var(process_counts)++;
1258                 }
1259                 attach_pid(p, PIDTYPE_PID, pid);
1260                 nr_threads++;
1261         }
1262
1263         total_forks++;
1264         spin_unlock(&current->sighand->siglock);
1265         write_unlock_irq(&tasklist_lock);
1266         proc_fork_connector(p);
1267         cgroup_post_fork(p);
1268         return p;
1269
1270 bad_fork_free_graph:
1271         ftrace_graph_exit_task(p);
1272 bad_fork_free_pid:
1273         if (pid != &init_struct_pid)
1274                 free_pid(pid);
1275 bad_fork_cleanup_io:
1276         put_io_context(p->io_context);
1277 bad_fork_cleanup_namespaces:
1278         exit_task_namespaces(p);
1279 bad_fork_cleanup_mm:
1280         if (p->mm)
1281                 mmput(p->mm);
1282 bad_fork_cleanup_signal:
1283         cleanup_signal(p);
1284 bad_fork_cleanup_sighand:
1285         __cleanup_sighand(p->sighand);
1286 bad_fork_cleanup_fs:
1287         exit_fs(p); /* blocking */
1288 bad_fork_cleanup_files:
1289         exit_files(p); /* blocking */
1290 bad_fork_cleanup_semundo:
1291         exit_sem(p);
1292 bad_fork_cleanup_audit:
1293         audit_free(p);
1294 bad_fork_cleanup_policy:
1295 #ifdef CONFIG_NUMA
1296         mpol_put(p->mempolicy);
1297 bad_fork_cleanup_cgroup:
1298 #endif
1299         cgroup_exit(p, cgroup_callbacks_done);
1300         delayacct_tsk_free(p);
1301         if (p->binfmt)
1302                 module_put(p->binfmt->module);
1303 bad_fork_cleanup_put_domain:
1304         module_put(task_thread_info(p)->exec_domain->module);
1305 bad_fork_cleanup_count:
1306         atomic_dec(&p->cred->user->processes);
1307         put_cred(p->real_cred);
1308         put_cred(p->cred);
1309 bad_fork_free:
1310         free_task(p);
1311 fork_out:
1312         return ERR_PTR(retval);
1313 }
1314
1315 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1316 {
1317         memset(regs, 0, sizeof(struct pt_regs));
1318         return regs;
1319 }
1320
1321 struct task_struct * __cpuinit fork_idle(int cpu)
1322 {
1323         struct task_struct *task;
1324         struct pt_regs regs;
1325
1326         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1327                             &init_struct_pid, 0);
1328         if (!IS_ERR(task))
1329                 init_idle(task, cpu);
1330
1331         return task;
1332 }
1333
1334 /*
1335  *  Ok, this is the main fork-routine.
1336  *
1337  * It copies the process, and if successful kick-starts
1338  * it and waits for it to finish using the VM if required.
1339  */
1340 long do_fork(unsigned long clone_flags,
1341               unsigned long stack_start,
1342               struct pt_regs *regs,
1343               unsigned long stack_size,
1344               int __user *parent_tidptr,
1345               int __user *child_tidptr)
1346 {
1347         struct task_struct *p;
1348         int trace = 0;
1349         long nr;
1350
1351         /*
1352          * Do some preliminary argument and permissions checking before we
1353          * actually start allocating stuff
1354          */
1355         if (clone_flags & CLONE_NEWUSER) {
1356                 if (clone_flags & CLONE_THREAD)
1357                         return -EINVAL;
1358                 /* hopefully this check will go away when userns support is
1359                  * complete
1360                  */
1361                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1362                                 !capable(CAP_SETGID))
1363                         return -EPERM;
1364         }
1365
1366         /*
1367          * We hope to recycle these flags after 2.6.26
1368          */
1369         if (unlikely(clone_flags & CLONE_STOPPED)) {
1370                 static int __read_mostly count = 100;
1371
1372                 if (count > 0 && printk_ratelimit()) {
1373                         char comm[TASK_COMM_LEN];
1374
1375                         count--;
1376                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1377                                         "clone flags 0x%lx\n",
1378                                 get_task_comm(comm, current),
1379                                 clone_flags & CLONE_STOPPED);
1380                 }
1381         }
1382
1383         /*
1384          * When called from kernel_thread, don't do user tracing stuff.
1385          */
1386         if (likely(user_mode(regs)))
1387                 trace = tracehook_prepare_clone(clone_flags);
1388
1389         p = copy_process(clone_flags, stack_start, regs, stack_size,
1390                          child_tidptr, NULL, trace);
1391         /*
1392          * Do this prior waking up the new thread - the thread pointer
1393          * might get invalid after that point, if the thread exits quickly.
1394          */
1395         if (!IS_ERR(p)) {
1396                 struct completion vfork;
1397
1398                 trace_sched_process_fork(current, p);
1399
1400                 nr = task_pid_vnr(p);
1401
1402                 if (clone_flags & CLONE_PARENT_SETTID)
1403                         put_user(nr, parent_tidptr);
1404
1405                 if (clone_flags & CLONE_VFORK) {
1406                         p->vfork_done = &vfork;
1407                         init_completion(&vfork);
1408                 }
1409
1410                 audit_finish_fork(p);
1411                 tracehook_report_clone(trace, regs, clone_flags, nr, p);
1412
1413                 /*
1414                  * We set PF_STARTING at creation in case tracing wants to
1415                  * use this to distinguish a fully live task from one that
1416                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1417                  * clear it and set the child going.
1418                  */
1419                 p->flags &= ~PF_STARTING;
1420
1421                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1422                         /*
1423                          * We'll start up with an immediate SIGSTOP.
1424                          */
1425                         sigaddset(&p->pending.signal, SIGSTOP);
1426                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1427                         __set_task_state(p, TASK_STOPPED);
1428                 } else {
1429                         wake_up_new_task(p, clone_flags);
1430                 }
1431
1432                 tracehook_report_clone_complete(trace, regs,
1433                                                 clone_flags, nr, p);
1434
1435                 if (clone_flags & CLONE_VFORK) {
1436                         freezer_do_not_count();
1437                         wait_for_completion(&vfork);
1438                         freezer_count();
1439                         tracehook_report_vfork_done(p, nr);
1440                 }
1441         } else {
1442                 nr = PTR_ERR(p);
1443         }
1444         return nr;
1445 }
1446
1447 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1448 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1449 #endif
1450
1451 static void sighand_ctor(void *data)
1452 {
1453         struct sighand_struct *sighand = data;
1454
1455         spin_lock_init(&sighand->siglock);
1456         init_waitqueue_head(&sighand->signalfd_wqh);
1457 }
1458
1459 void __init proc_caches_init(void)
1460 {
1461         sighand_cachep = kmem_cache_create("sighand_cache",
1462                         sizeof(struct sighand_struct), 0,
1463                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
1464                         sighand_ctor);
1465         signal_cachep = kmem_cache_create("signal_cache",
1466                         sizeof(struct signal_struct), 0,
1467                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1468         files_cachep = kmem_cache_create("files_cache",
1469                         sizeof(struct files_struct), 0,
1470                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1471         fs_cachep = kmem_cache_create("fs_cache",
1472                         sizeof(struct fs_struct), 0,
1473                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1474         mm_cachep = kmem_cache_create("mm_struct",
1475                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1476                         SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1477         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1478         mmap_init();
1479 }
1480
1481 /*
1482  * Check constraints on flags passed to the unshare system call and
1483  * force unsharing of additional process context as appropriate.
1484  */
1485 static void check_unshare_flags(unsigned long *flags_ptr)
1486 {
1487         /*
1488          * If unsharing a thread from a thread group, must also
1489          * unshare vm.
1490          */
1491         if (*flags_ptr & CLONE_THREAD)
1492                 *flags_ptr |= CLONE_VM;
1493
1494         /*
1495          * If unsharing vm, must also unshare signal handlers.
1496          */
1497         if (*flags_ptr & CLONE_VM)
1498                 *flags_ptr |= CLONE_SIGHAND;
1499
1500         /*
1501          * If unsharing signal handlers and the task was created
1502          * using CLONE_THREAD, then must unshare the thread
1503          */
1504         if ((*flags_ptr & CLONE_SIGHAND) &&
1505             (atomic_read(&current->signal->count) > 1))
1506                 *flags_ptr |= CLONE_THREAD;
1507
1508         /*
1509          * If unsharing namespace, must also unshare filesystem information.
1510          */
1511         if (*flags_ptr & CLONE_NEWNS)
1512                 *flags_ptr |= CLONE_FS;
1513 }
1514
1515 /*
1516  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1517  */
1518 static int unshare_thread(unsigned long unshare_flags)
1519 {
1520         if (unshare_flags & CLONE_THREAD)
1521                 return -EINVAL;
1522
1523         return 0;
1524 }
1525
1526 /*
1527  * Unshare the filesystem structure if it is being shared
1528  */
1529 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1530 {
1531         struct fs_struct *fs = current->fs;
1532
1533         if (!(unshare_flags & CLONE_FS) || !fs)
1534                 return 0;
1535
1536         /* don't need lock here; in the worst case we'll do useless copy */
1537         if (fs->users == 1)
1538                 return 0;
1539
1540         *new_fsp = copy_fs_struct(fs);
1541         if (!*new_fsp)
1542                 return -ENOMEM;
1543
1544         return 0;
1545 }
1546
1547 /*
1548  * Unsharing of sighand is not supported yet
1549  */
1550 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1551 {
1552         struct sighand_struct *sigh = current->sighand;
1553
1554         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1555                 return -EINVAL;
1556         else
1557                 return 0;
1558 }
1559
1560 /*
1561  * Unshare vm if it is being shared
1562  */
1563 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1564 {
1565         struct mm_struct *mm = current->mm;
1566
1567         if ((unshare_flags & CLONE_VM) &&
1568             (mm && atomic_read(&mm->mm_users) > 1)) {
1569                 return -EINVAL;
1570         }
1571
1572         return 0;
1573 }
1574
1575 /*
1576  * Unshare file descriptor table if it is being shared
1577  */
1578 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1579 {
1580         struct files_struct *fd = current->files;
1581         int error = 0;
1582
1583         if ((unshare_flags & CLONE_FILES) &&
1584             (fd && atomic_read(&fd->count) > 1)) {
1585                 *new_fdp = dup_fd(fd, &error);
1586                 if (!*new_fdp)
1587                         return error;
1588         }
1589
1590         return 0;
1591 }
1592
1593 /*
1594  * unshare allows a process to 'unshare' part of the process
1595  * context which was originally shared using clone.  copy_*
1596  * functions used by do_fork() cannot be used here directly
1597  * because they modify an inactive task_struct that is being
1598  * constructed. Here we are modifying the current, active,
1599  * task_struct.
1600  */
1601 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1602 {
1603         int err = 0;
1604         struct fs_struct *fs, *new_fs = NULL;
1605         struct sighand_struct *new_sigh = NULL;
1606         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1607         struct files_struct *fd, *new_fd = NULL;
1608         struct nsproxy *new_nsproxy = NULL;
1609         int do_sysvsem = 0;
1610
1611         check_unshare_flags(&unshare_flags);
1612
1613         /* Return -EINVAL for all unsupported flags */
1614         err = -EINVAL;
1615         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1616                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1617                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1618                 goto bad_unshare_out;
1619
1620         /*
1621          * CLONE_NEWIPC must also detach from the undolist: after switching
1622          * to a new ipc namespace, the semaphore arrays from the old
1623          * namespace are unreachable.
1624          */
1625         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1626                 do_sysvsem = 1;
1627         if ((err = unshare_thread(unshare_flags)))
1628                 goto bad_unshare_out;
1629         if ((err = unshare_fs(unshare_flags, &new_fs)))
1630                 goto bad_unshare_cleanup_thread;
1631         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1632                 goto bad_unshare_cleanup_fs;
1633         if ((err = unshare_vm(unshare_flags, &new_mm)))
1634                 goto bad_unshare_cleanup_sigh;
1635         if ((err = unshare_fd(unshare_flags, &new_fd)))
1636                 goto bad_unshare_cleanup_vm;
1637         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1638                         new_fs)))
1639                 goto bad_unshare_cleanup_fd;
1640
1641         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1642                 if (do_sysvsem) {
1643                         /*
1644                          * CLONE_SYSVSEM is equivalent to sys_exit().
1645                          */
1646                         exit_sem(current);
1647                 }
1648
1649                 if (new_nsproxy) {
1650                         switch_task_namespaces(current, new_nsproxy);
1651                         new_nsproxy = NULL;
1652                 }
1653
1654                 task_lock(current);
1655
1656                 if (new_fs) {
1657                         fs = current->fs;
1658                         write_lock(&fs->lock);
1659                         current->fs = new_fs;
1660                         if (--fs->users)
1661                                 new_fs = NULL;
1662                         else
1663                                 new_fs = fs;
1664                         write_unlock(&fs->lock);
1665                 }
1666
1667                 if (new_mm) {
1668                         mm = current->mm;
1669                         active_mm = current->active_mm;
1670                         current->mm = new_mm;
1671                         current->active_mm = new_mm;
1672                         activate_mm(active_mm, new_mm);
1673                         new_mm = mm;
1674                 }
1675
1676                 if (new_fd) {
1677                         fd = current->files;
1678                         current->files = new_fd;
1679                         new_fd = fd;
1680                 }
1681
1682                 task_unlock(current);
1683         }
1684
1685         if (new_nsproxy)
1686                 put_nsproxy(new_nsproxy);
1687
1688 bad_unshare_cleanup_fd:
1689         if (new_fd)
1690                 put_files_struct(new_fd);
1691
1692 bad_unshare_cleanup_vm:
1693         if (new_mm)
1694                 mmput(new_mm);
1695
1696 bad_unshare_cleanup_sigh:
1697         if (new_sigh)
1698                 if (atomic_dec_and_test(&new_sigh->count))
1699                         kmem_cache_free(sighand_cachep, new_sigh);
1700
1701 bad_unshare_cleanup_fs:
1702         if (new_fs)
1703                 free_fs_struct(new_fs);
1704
1705 bad_unshare_cleanup_thread:
1706 bad_unshare_out:
1707         return err;
1708 }
1709
1710 /*
1711  *      Helper to unshare the files of the current task.
1712  *      We don't want to expose copy_files internals to
1713  *      the exec layer of the kernel.
1714  */
1715
1716 int unshare_files(struct files_struct **displaced)
1717 {
1718         struct task_struct *task = current;
1719         struct files_struct *copy = NULL;
1720         int error;
1721
1722         error = unshare_fd(CLONE_FILES, &copy);
1723         if (error || !copy) {
1724                 *displaced = NULL;
1725                 return error;
1726         }
1727         *displaced = task->files;
1728         task_lock(task);
1729         task->files = copy;
1730         task_unlock(task);
1731         return 0;
1732 }