sched: Convert pi_lock to raw_spinlock
[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/ksm.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 #include <linux/perf_event.h>
66 #include <linux/posix-timers.h>
67 #include <linux/user-return-notifier.h>
68
69 #include <asm/pgtable.h>
70 #include <asm/pgalloc.h>
71 #include <asm/uaccess.h>
72 #include <asm/mmu_context.h>
73 #include <asm/cacheflush.h>
74 #include <asm/tlbflush.h>
75
76 #include <trace/events/sched.h>
77
78 /*
79  * Protected counters by write_lock_irq(&tasklist_lock)
80  */
81 unsigned long total_forks;      /* Handle normal Linux uptimes. */
82 int nr_threads;                 /* The idle threads do not count.. */
83
84 int max_threads;                /* tunable limit on nr_threads */
85
86 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
87
88 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
89
90 int nr_processes(void)
91 {
92         int cpu;
93         int total = 0;
94
95         for_each_possible_cpu(cpu)
96                 total += per_cpu(process_counts, cpu);
97
98         return total;
99 }
100
101 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
102 # define alloc_task_struct()    kmem_cache_alloc(task_struct_cachep, GFP_KERNEL)
103 # define free_task_struct(tsk)  kmem_cache_free(task_struct_cachep, (tsk))
104 static struct kmem_cache *task_struct_cachep;
105 #endif
106
107 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
108 static inline struct thread_info *alloc_thread_info(struct task_struct *tsk)
109 {
110 #ifdef CONFIG_DEBUG_STACK_USAGE
111         gfp_t mask = GFP_KERNEL | __GFP_ZERO;
112 #else
113         gfp_t mask = GFP_KERNEL;
114 #endif
115         return (struct thread_info *)__get_free_pages(mask, THREAD_SIZE_ORDER);
116 }
117
118 static inline void free_thread_info(struct thread_info *ti)
119 {
120         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
121 }
122 #endif
123
124 /* SLAB cache for signal_struct structures (tsk->signal) */
125 static struct kmem_cache *signal_cachep;
126
127 /* SLAB cache for sighand_struct structures (tsk->sighand) */
128 struct kmem_cache *sighand_cachep;
129
130 /* SLAB cache for files_struct structures (tsk->files) */
131 struct kmem_cache *files_cachep;
132
133 /* SLAB cache for fs_struct structures (tsk->fs) */
134 struct kmem_cache *fs_cachep;
135
136 /* SLAB cache for vm_area_struct structures */
137 struct kmem_cache *vm_area_cachep;
138
139 /* SLAB cache for mm_struct structures (tsk->mm) */
140 static struct kmem_cache *mm_cachep;
141
142 static void account_kernel_stack(struct thread_info *ti, int account)
143 {
144         struct zone *zone = page_zone(virt_to_page(ti));
145
146         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
147 }
148
149 void free_task(struct task_struct *tsk)
150 {
151         prop_local_destroy_single(&tsk->dirties);
152         account_kernel_stack(tsk->stack, -1);
153         free_thread_info(tsk->stack);
154         rt_mutex_debug_task_free(tsk);
155         ftrace_graph_exit_task(tsk);
156         free_task_struct(tsk);
157 }
158 EXPORT_SYMBOL(free_task);
159
160 void __put_task_struct(struct task_struct *tsk)
161 {
162         WARN_ON(!tsk->exit_state);
163         WARN_ON(atomic_read(&tsk->usage));
164         WARN_ON(tsk == current);
165
166         exit_creds(tsk);
167         delayacct_tsk_free(tsk);
168
169         if (!profile_handoff_task(tsk))
170                 free_task(tsk);
171 }
172
173 /*
174  * macro override instead of weak attribute alias, to workaround
175  * gcc 4.1.0 and 4.1.1 bugs with weak attribute and empty functions.
176  */
177 #ifndef arch_task_cache_init
178 #define arch_task_cache_init()
179 #endif
180
181 void __init fork_init(unsigned long mempages)
182 {
183 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
184 #ifndef ARCH_MIN_TASKALIGN
185 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
186 #endif
187         /* create a slab on which task_structs can be allocated */
188         task_struct_cachep =
189                 kmem_cache_create("task_struct", sizeof(struct task_struct),
190                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
191 #endif
192
193         /* do the arch specific task caches init */
194         arch_task_cache_init();
195
196         /*
197          * The default maximum number of threads is set to a safe
198          * value: the thread structures can take up at most half
199          * of memory.
200          */
201         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
202
203         /*
204          * we need to allow at least 20 threads to boot a system
205          */
206         if(max_threads < 20)
207                 max_threads = 20;
208
209         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
210         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
211         init_task.signal->rlim[RLIMIT_SIGPENDING] =
212                 init_task.signal->rlim[RLIMIT_NPROC];
213 }
214
215 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
216                                                struct task_struct *src)
217 {
218         *dst = *src;
219         return 0;
220 }
221
222 static struct task_struct *dup_task_struct(struct task_struct *orig)
223 {
224         struct task_struct *tsk;
225         struct thread_info *ti;
226         unsigned long *stackend;
227
228         int err;
229
230         prepare_to_copy(orig);
231
232         tsk = alloc_task_struct();
233         if (!tsk)
234                 return NULL;
235
236         ti = alloc_thread_info(tsk);
237         if (!ti) {
238                 free_task_struct(tsk);
239                 return NULL;
240         }
241
242         err = arch_dup_task_struct(tsk, orig);
243         if (err)
244                 goto out;
245
246         tsk->stack = ti;
247
248         err = prop_local_init_single(&tsk->dirties);
249         if (err)
250                 goto out;
251
252         setup_thread_stack(tsk, orig);
253         clear_user_return_notifier(tsk);
254         stackend = end_of_stack(tsk);
255         *stackend = STACK_END_MAGIC;    /* for overflow detection */
256
257 #ifdef CONFIG_CC_STACKPROTECTOR
258         tsk->stack_canary = get_random_int();
259 #endif
260
261         /* One for us, one for whoever does the "release_task()" (usually parent) */
262         atomic_set(&tsk->usage,2);
263         atomic_set(&tsk->fs_excl, 0);
264 #ifdef CONFIG_BLK_DEV_IO_TRACE
265         tsk->btrace_seq = 0;
266 #endif
267         tsk->splice_pipe = NULL;
268
269         account_kernel_stack(ti, 1);
270
271         return tsk;
272
273 out:
274         free_thread_info(ti);
275         free_task_struct(tsk);
276         return NULL;
277 }
278
279 #ifdef CONFIG_MMU
280 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
281 {
282         struct vm_area_struct *mpnt, *tmp, **pprev;
283         struct rb_node **rb_link, *rb_parent;
284         int retval;
285         unsigned long charge;
286         struct mempolicy *pol;
287
288         down_write(&oldmm->mmap_sem);
289         flush_cache_dup_mm(oldmm);
290         /*
291          * Not linked in yet - no deadlock potential:
292          */
293         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
294
295         mm->locked_vm = 0;
296         mm->mmap = NULL;
297         mm->mmap_cache = NULL;
298         mm->free_area_cache = oldmm->mmap_base;
299         mm->cached_hole_size = ~0UL;
300         mm->map_count = 0;
301         cpumask_clear(mm_cpumask(mm));
302         mm->mm_rb = RB_ROOT;
303         rb_link = &mm->mm_rb.rb_node;
304         rb_parent = NULL;
305         pprev = &mm->mmap;
306         retval = ksm_fork(mm, oldmm);
307         if (retval)
308                 goto out;
309
310         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
311                 struct file *file;
312
313                 if (mpnt->vm_flags & VM_DONTCOPY) {
314                         long pages = vma_pages(mpnt);
315                         mm->total_vm -= pages;
316                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
317                                                                 -pages);
318                         continue;
319                 }
320                 charge = 0;
321                 if (mpnt->vm_flags & VM_ACCOUNT) {
322                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
323                         if (security_vm_enough_memory(len))
324                                 goto fail_nomem;
325                         charge = len;
326                 }
327                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
328                 if (!tmp)
329                         goto fail_nomem;
330                 *tmp = *mpnt;
331                 pol = mpol_dup(vma_policy(mpnt));
332                 retval = PTR_ERR(pol);
333                 if (IS_ERR(pol))
334                         goto fail_nomem_policy;
335                 vma_set_policy(tmp, pol);
336                 tmp->vm_flags &= ~VM_LOCKED;
337                 tmp->vm_mm = mm;
338                 tmp->vm_next = NULL;
339                 anon_vma_link(tmp);
340                 file = tmp->vm_file;
341                 if (file) {
342                         struct inode *inode = file->f_path.dentry->d_inode;
343                         struct address_space *mapping = file->f_mapping;
344
345                         get_file(file);
346                         if (tmp->vm_flags & VM_DENYWRITE)
347                                 atomic_dec(&inode->i_writecount);
348                         spin_lock(&mapping->i_mmap_lock);
349                         if (tmp->vm_flags & VM_SHARED)
350                                 mapping->i_mmap_writable++;
351                         tmp->vm_truncate_count = mpnt->vm_truncate_count;
352                         flush_dcache_mmap_lock(mapping);
353                         /* insert tmp into the share list, just after mpnt */
354                         vma_prio_tree_add(tmp, mpnt);
355                         flush_dcache_mmap_unlock(mapping);
356                         spin_unlock(&mapping->i_mmap_lock);
357                 }
358
359                 /*
360                  * Clear hugetlb-related page reserves for children. This only
361                  * affects MAP_PRIVATE mappings. Faults generated by the child
362                  * are not guaranteed to succeed, even if read-only
363                  */
364                 if (is_vm_hugetlb_page(tmp))
365                         reset_vma_resv_huge_pages(tmp);
366
367                 /*
368                  * Link in the new vma and copy the page table entries.
369                  */
370                 *pprev = tmp;
371                 pprev = &tmp->vm_next;
372
373                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
374                 rb_link = &tmp->vm_rb.rb_right;
375                 rb_parent = &tmp->vm_rb;
376
377                 mm->map_count++;
378                 retval = copy_page_range(mm, oldmm, mpnt);
379
380                 if (tmp->vm_ops && tmp->vm_ops->open)
381                         tmp->vm_ops->open(tmp);
382
383                 if (retval)
384                         goto out;
385         }
386         /* a new mm has just been created */
387         arch_dup_mmap(oldmm, mm);
388         retval = 0;
389 out:
390         up_write(&mm->mmap_sem);
391         flush_tlb_mm(oldmm);
392         up_write(&oldmm->mmap_sem);
393         return retval;
394 fail_nomem_policy:
395         kmem_cache_free(vm_area_cachep, tmp);
396 fail_nomem:
397         retval = -ENOMEM;
398         vm_unacct_memory(charge);
399         goto out;
400 }
401
402 static inline int mm_alloc_pgd(struct mm_struct * mm)
403 {
404         mm->pgd = pgd_alloc(mm);
405         if (unlikely(!mm->pgd))
406                 return -ENOMEM;
407         return 0;
408 }
409
410 static inline void mm_free_pgd(struct mm_struct * mm)
411 {
412         pgd_free(mm, mm->pgd);
413 }
414 #else
415 #define dup_mmap(mm, oldmm)     (0)
416 #define mm_alloc_pgd(mm)        (0)
417 #define mm_free_pgd(mm)
418 #endif /* CONFIG_MMU */
419
420 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
421
422 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
423 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
424
425 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
426
427 static int __init coredump_filter_setup(char *s)
428 {
429         default_dump_filter =
430                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
431                 MMF_DUMP_FILTER_MASK;
432         return 1;
433 }
434
435 __setup("coredump_filter=", coredump_filter_setup);
436
437 #include <linux/init_task.h>
438
439 static void mm_init_aio(struct mm_struct *mm)
440 {
441 #ifdef CONFIG_AIO
442         spin_lock_init(&mm->ioctx_lock);
443         INIT_HLIST_HEAD(&mm->ioctx_list);
444 #endif
445 }
446
447 static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
448 {
449         atomic_set(&mm->mm_users, 1);
450         atomic_set(&mm->mm_count, 1);
451         init_rwsem(&mm->mmap_sem);
452         INIT_LIST_HEAD(&mm->mmlist);
453         mm->flags = (current->mm) ?
454                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
455         mm->core_state = NULL;
456         mm->nr_ptes = 0;
457         set_mm_counter(mm, file_rss, 0);
458         set_mm_counter(mm, anon_rss, 0);
459         spin_lock_init(&mm->page_table_lock);
460         mm->free_area_cache = TASK_UNMAPPED_BASE;
461         mm->cached_hole_size = ~0UL;
462         mm_init_aio(mm);
463         mm_init_owner(mm, p);
464
465         if (likely(!mm_alloc_pgd(mm))) {
466                 mm->def_flags = 0;
467                 mmu_notifier_mm_init(mm);
468                 return mm;
469         }
470
471         free_mm(mm);
472         return NULL;
473 }
474
475 /*
476  * Allocate and initialize an mm_struct.
477  */
478 struct mm_struct * mm_alloc(void)
479 {
480         struct mm_struct * mm;
481
482         mm = allocate_mm();
483         if (mm) {
484                 memset(mm, 0, sizeof(*mm));
485                 mm = mm_init(mm, current);
486         }
487         return mm;
488 }
489
490 /*
491  * Called when the last reference to the mm
492  * is dropped: either by a lazy thread or by
493  * mmput. Free the page directory and the mm.
494  */
495 void __mmdrop(struct mm_struct *mm)
496 {
497         BUG_ON(mm == &init_mm);
498         mm_free_pgd(mm);
499         destroy_context(mm);
500         mmu_notifier_mm_destroy(mm);
501         free_mm(mm);
502 }
503 EXPORT_SYMBOL_GPL(__mmdrop);
504
505 /*
506  * Decrement the use count and release all resources for an mm.
507  */
508 void mmput(struct mm_struct *mm)
509 {
510         might_sleep();
511
512         if (atomic_dec_and_test(&mm->mm_users)) {
513                 exit_aio(mm);
514                 ksm_exit(mm);
515                 exit_mmap(mm);
516                 set_mm_exe_file(mm, NULL);
517                 if (!list_empty(&mm->mmlist)) {
518                         spin_lock(&mmlist_lock);
519                         list_del(&mm->mmlist);
520                         spin_unlock(&mmlist_lock);
521                 }
522                 put_swap_token(mm);
523                 if (mm->binfmt)
524                         module_put(mm->binfmt->module);
525                 mmdrop(mm);
526         }
527 }
528 EXPORT_SYMBOL_GPL(mmput);
529
530 /**
531  * get_task_mm - acquire a reference to the task's mm
532  *
533  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
534  * this kernel workthread has transiently adopted a user mm with use_mm,
535  * to do its AIO) is not set and if so returns a reference to it, after
536  * bumping up the use count.  User must release the mm via mmput()
537  * after use.  Typically used by /proc and ptrace.
538  */
539 struct mm_struct *get_task_mm(struct task_struct *task)
540 {
541         struct mm_struct *mm;
542
543         task_lock(task);
544         mm = task->mm;
545         if (mm) {
546                 if (task->flags & PF_KTHREAD)
547                         mm = NULL;
548                 else
549                         atomic_inc(&mm->mm_users);
550         }
551         task_unlock(task);
552         return mm;
553 }
554 EXPORT_SYMBOL_GPL(get_task_mm);
555
556 /* Please note the differences between mmput and mm_release.
557  * mmput is called whenever we stop holding onto a mm_struct,
558  * error success whatever.
559  *
560  * mm_release is called after a mm_struct has been removed
561  * from the current process.
562  *
563  * This difference is important for error handling, when we
564  * only half set up a mm_struct for a new process and need to restore
565  * the old one.  Because we mmput the new mm_struct before
566  * restoring the old one. . .
567  * Eric Biederman 10 January 1998
568  */
569 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
570 {
571         struct completion *vfork_done = tsk->vfork_done;
572
573         /* Get rid of any futexes when releasing the mm */
574 #ifdef CONFIG_FUTEX
575         if (unlikely(tsk->robust_list)) {
576                 exit_robust_list(tsk);
577                 tsk->robust_list = NULL;
578         }
579 #ifdef CONFIG_COMPAT
580         if (unlikely(tsk->compat_robust_list)) {
581                 compat_exit_robust_list(tsk);
582                 tsk->compat_robust_list = NULL;
583         }
584 #endif
585         if (unlikely(!list_empty(&tsk->pi_state_list)))
586                 exit_pi_state_list(tsk);
587 #endif
588
589         /* Get rid of any cached register state */
590         deactivate_mm(tsk, mm);
591
592         /* notify parent sleeping on vfork() */
593         if (vfork_done) {
594                 tsk->vfork_done = NULL;
595                 complete(vfork_done);
596         }
597
598         /*
599          * If we're exiting normally, clear a user-space tid field if
600          * requested.  We leave this alone when dying by signal, to leave
601          * the value intact in a core dump, and to save the unnecessary
602          * trouble otherwise.  Userland only wants this done for a sys_exit.
603          */
604         if (tsk->clear_child_tid) {
605                 if (!(tsk->flags & PF_SIGNALED) &&
606                     atomic_read(&mm->mm_users) > 1) {
607                         /*
608                          * We don't check the error code - if userspace has
609                          * not set up a proper pointer then tough luck.
610                          */
611                         put_user(0, tsk->clear_child_tid);
612                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
613                                         1, NULL, NULL, 0);
614                 }
615                 tsk->clear_child_tid = NULL;
616         }
617 }
618
619 /*
620  * Allocate a new mm structure and copy contents from the
621  * mm structure of the passed in task structure.
622  */
623 struct mm_struct *dup_mm(struct task_struct *tsk)
624 {
625         struct mm_struct *mm, *oldmm = current->mm;
626         int err;
627
628         if (!oldmm)
629                 return NULL;
630
631         mm = allocate_mm();
632         if (!mm)
633                 goto fail_nomem;
634
635         memcpy(mm, oldmm, sizeof(*mm));
636
637         /* Initializing for Swap token stuff */
638         mm->token_priority = 0;
639         mm->last_interval = 0;
640
641         if (!mm_init(mm, tsk))
642                 goto fail_nomem;
643
644         if (init_new_context(tsk, mm))
645                 goto fail_nocontext;
646
647         dup_mm_exe_file(oldmm, mm);
648
649         err = dup_mmap(mm, oldmm);
650         if (err)
651                 goto free_pt;
652
653         mm->hiwater_rss = get_mm_rss(mm);
654         mm->hiwater_vm = mm->total_vm;
655
656         if (mm->binfmt && !try_module_get(mm->binfmt->module))
657                 goto free_pt;
658
659         return mm;
660
661 free_pt:
662         /* don't put binfmt in mmput, we haven't got module yet */
663         mm->binfmt = NULL;
664         mmput(mm);
665
666 fail_nomem:
667         return NULL;
668
669 fail_nocontext:
670         /*
671          * If init_new_context() failed, we cannot use mmput() to free the mm
672          * because it calls destroy_context()
673          */
674         mm_free_pgd(mm);
675         free_mm(mm);
676         return NULL;
677 }
678
679 static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
680 {
681         struct mm_struct * mm, *oldmm;
682         int retval;
683
684         tsk->min_flt = tsk->maj_flt = 0;
685         tsk->nvcsw = tsk->nivcsw = 0;
686 #ifdef CONFIG_DETECT_HUNG_TASK
687         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
688 #endif
689
690         tsk->mm = NULL;
691         tsk->active_mm = NULL;
692
693         /*
694          * Are we cloning a kernel thread?
695          *
696          * We need to steal a active VM for that..
697          */
698         oldmm = current->mm;
699         if (!oldmm)
700                 return 0;
701
702         if (clone_flags & CLONE_VM) {
703                 atomic_inc(&oldmm->mm_users);
704                 mm = oldmm;
705                 goto good_mm;
706         }
707
708         retval = -ENOMEM;
709         mm = dup_mm(tsk);
710         if (!mm)
711                 goto fail_nomem;
712
713 good_mm:
714         /* Initializing for Swap token stuff */
715         mm->token_priority = 0;
716         mm->last_interval = 0;
717
718         tsk->mm = mm;
719         tsk->active_mm = mm;
720         return 0;
721
722 fail_nomem:
723         return retval;
724 }
725
726 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
727 {
728         struct fs_struct *fs = current->fs;
729         if (clone_flags & CLONE_FS) {
730                 /* tsk->fs is already what we want */
731                 write_lock(&fs->lock);
732                 if (fs->in_exec) {
733                         write_unlock(&fs->lock);
734                         return -EAGAIN;
735                 }
736                 fs->users++;
737                 write_unlock(&fs->lock);
738                 return 0;
739         }
740         tsk->fs = copy_fs_struct(fs);
741         if (!tsk->fs)
742                 return -ENOMEM;
743         return 0;
744 }
745
746 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
747 {
748         struct files_struct *oldf, *newf;
749         int error = 0;
750
751         /*
752          * A background process may not have any files ...
753          */
754         oldf = current->files;
755         if (!oldf)
756                 goto out;
757
758         if (clone_flags & CLONE_FILES) {
759                 atomic_inc(&oldf->count);
760                 goto out;
761         }
762
763         newf = dup_fd(oldf, &error);
764         if (!newf)
765                 goto out;
766
767         tsk->files = newf;
768         error = 0;
769 out:
770         return error;
771 }
772
773 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
774 {
775 #ifdef CONFIG_BLOCK
776         struct io_context *ioc = current->io_context;
777
778         if (!ioc)
779                 return 0;
780         /*
781          * Share io context with parent, if CLONE_IO is set
782          */
783         if (clone_flags & CLONE_IO) {
784                 tsk->io_context = ioc_task_link(ioc);
785                 if (unlikely(!tsk->io_context))
786                         return -ENOMEM;
787         } else if (ioprio_valid(ioc->ioprio)) {
788                 tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
789                 if (unlikely(!tsk->io_context))
790                         return -ENOMEM;
791
792                 tsk->io_context->ioprio = ioc->ioprio;
793         }
794 #endif
795         return 0;
796 }
797
798 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
799 {
800         struct sighand_struct *sig;
801
802         if (clone_flags & CLONE_SIGHAND) {
803                 atomic_inc(&current->sighand->count);
804                 return 0;
805         }
806         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
807         rcu_assign_pointer(tsk->sighand, sig);
808         if (!sig)
809                 return -ENOMEM;
810         atomic_set(&sig->count, 1);
811         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
812         return 0;
813 }
814
815 void __cleanup_sighand(struct sighand_struct *sighand)
816 {
817         if (atomic_dec_and_test(&sighand->count))
818                 kmem_cache_free(sighand_cachep, sighand);
819 }
820
821
822 /*
823  * Initialize POSIX timer handling for a thread group.
824  */
825 static void posix_cpu_timers_init_group(struct signal_struct *sig)
826 {
827         /* Thread group counters. */
828         thread_group_cputime_init(sig);
829
830         /* Expiration times and increments. */
831         sig->it[CPUCLOCK_PROF].expires = cputime_zero;
832         sig->it[CPUCLOCK_PROF].incr = cputime_zero;
833         sig->it[CPUCLOCK_VIRT].expires = cputime_zero;
834         sig->it[CPUCLOCK_VIRT].incr = cputime_zero;
835
836         /* Cached expiration times. */
837         sig->cputime_expires.prof_exp = cputime_zero;
838         sig->cputime_expires.virt_exp = cputime_zero;
839         sig->cputime_expires.sched_exp = 0;
840
841         if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
842                 sig->cputime_expires.prof_exp =
843                         secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
844                 sig->cputimer.running = 1;
845         }
846
847         /* The timer lists. */
848         INIT_LIST_HEAD(&sig->cpu_timers[0]);
849         INIT_LIST_HEAD(&sig->cpu_timers[1]);
850         INIT_LIST_HEAD(&sig->cpu_timers[2]);
851 }
852
853 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
854 {
855         struct signal_struct *sig;
856
857         if (clone_flags & CLONE_THREAD)
858                 return 0;
859
860         sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
861         tsk->signal = sig;
862         if (!sig)
863                 return -ENOMEM;
864
865         atomic_set(&sig->count, 1);
866         atomic_set(&sig->live, 1);
867         init_waitqueue_head(&sig->wait_chldexit);
868         sig->flags = 0;
869         if (clone_flags & CLONE_NEWPID)
870                 sig->flags |= SIGNAL_UNKILLABLE;
871         sig->group_exit_code = 0;
872         sig->group_exit_task = NULL;
873         sig->group_stop_count = 0;
874         sig->curr_target = tsk;
875         init_sigpending(&sig->shared_pending);
876         INIT_LIST_HEAD(&sig->posix_timers);
877
878         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
879         sig->it_real_incr.tv64 = 0;
880         sig->real_timer.function = it_real_fn;
881
882         sig->leader = 0;        /* session leadership doesn't inherit */
883         sig->tty_old_pgrp = NULL;
884         sig->tty = NULL;
885
886         sig->utime = sig->stime = sig->cutime = sig->cstime = cputime_zero;
887         sig->gtime = cputime_zero;
888         sig->cgtime = cputime_zero;
889 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
890         sig->prev_utime = sig->prev_stime = cputime_zero;
891 #endif
892         sig->nvcsw = sig->nivcsw = sig->cnvcsw = sig->cnivcsw = 0;
893         sig->min_flt = sig->maj_flt = sig->cmin_flt = sig->cmaj_flt = 0;
894         sig->inblock = sig->oublock = sig->cinblock = sig->coublock = 0;
895         sig->maxrss = sig->cmaxrss = 0;
896         task_io_accounting_init(&sig->ioac);
897         sig->sum_sched_runtime = 0;
898         taskstats_tgid_init(sig);
899
900         task_lock(current->group_leader);
901         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
902         task_unlock(current->group_leader);
903
904         posix_cpu_timers_init_group(sig);
905
906         acct_init_pacct(&sig->pacct);
907
908         tty_audit_fork(sig);
909
910         sig->oom_adj = current->signal->oom_adj;
911
912         return 0;
913 }
914
915 void __cleanup_signal(struct signal_struct *sig)
916 {
917         thread_group_cputime_free(sig);
918         tty_kref_put(sig->tty);
919         kmem_cache_free(signal_cachep, sig);
920 }
921
922 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
923 {
924         unsigned long new_flags = p->flags;
925
926         new_flags &= ~PF_SUPERPRIV;
927         new_flags |= PF_FORKNOEXEC;
928         new_flags |= PF_STARTING;
929         p->flags = new_flags;
930         clear_freeze_flag(p);
931 }
932
933 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
934 {
935         current->clear_child_tid = tidptr;
936
937         return task_pid_vnr(current);
938 }
939
940 static void rt_mutex_init_task(struct task_struct *p)
941 {
942         raw_spin_lock_init(&p->pi_lock);
943 #ifdef CONFIG_RT_MUTEXES
944         plist_head_init_raw(&p->pi_waiters, &p->pi_lock);
945         p->pi_blocked_on = NULL;
946 #endif
947 }
948
949 #ifdef CONFIG_MM_OWNER
950 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
951 {
952         mm->owner = p;
953 }
954 #endif /* CONFIG_MM_OWNER */
955
956 /*
957  * Initialize POSIX timer handling for a single task.
958  */
959 static void posix_cpu_timers_init(struct task_struct *tsk)
960 {
961         tsk->cputime_expires.prof_exp = cputime_zero;
962         tsk->cputime_expires.virt_exp = cputime_zero;
963         tsk->cputime_expires.sched_exp = 0;
964         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
965         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
966         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
967 }
968
969 /*
970  * This creates a new process as a copy of the old one,
971  * but does not actually start it yet.
972  *
973  * It copies the registers, and all the appropriate
974  * parts of the process environment (as per the clone
975  * flags). The actual kick-off is left to the caller.
976  */
977 static struct task_struct *copy_process(unsigned long clone_flags,
978                                         unsigned long stack_start,
979                                         struct pt_regs *regs,
980                                         unsigned long stack_size,
981                                         int __user *child_tidptr,
982                                         struct pid *pid,
983                                         int trace)
984 {
985         int retval;
986         struct task_struct *p;
987         int cgroup_callbacks_done = 0;
988
989         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
990                 return ERR_PTR(-EINVAL);
991
992         /*
993          * Thread groups must share signals as well, and detached threads
994          * can only be started up within the thread group.
995          */
996         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
997                 return ERR_PTR(-EINVAL);
998
999         /*
1000          * Shared signal handlers imply shared VM. By way of the above,
1001          * thread groups also imply shared VM. Blocking this case allows
1002          * for various simplifications in other code.
1003          */
1004         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1005                 return ERR_PTR(-EINVAL);
1006
1007         /*
1008          * Siblings of global init remain as zombies on exit since they are
1009          * not reaped by their parent (swapper). To solve this and to avoid
1010          * multi-rooted process trees, prevent global and container-inits
1011          * from creating siblings.
1012          */
1013         if ((clone_flags & CLONE_PARENT) &&
1014                                 current->signal->flags & SIGNAL_UNKILLABLE)
1015                 return ERR_PTR(-EINVAL);
1016
1017         retval = security_task_create(clone_flags);
1018         if (retval)
1019                 goto fork_out;
1020
1021         retval = -ENOMEM;
1022         p = dup_task_struct(current);
1023         if (!p)
1024                 goto fork_out;
1025
1026         ftrace_graph_init_task(p);
1027
1028         rt_mutex_init_task(p);
1029
1030 #ifdef CONFIG_PROVE_LOCKING
1031         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1032         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1033 #endif
1034         retval = -EAGAIN;
1035         if (atomic_read(&p->real_cred->user->processes) >=
1036                         p->signal->rlim[RLIMIT_NPROC].rlim_cur) {
1037                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1038                     p->real_cred->user != INIT_USER)
1039                         goto bad_fork_free;
1040         }
1041
1042         retval = copy_creds(p, clone_flags);
1043         if (retval < 0)
1044                 goto bad_fork_free;
1045
1046         /*
1047          * If multiple threads are within copy_process(), then this check
1048          * triggers too late. This doesn't hurt, the check is only there
1049          * to stop root fork bombs.
1050          */
1051         retval = -EAGAIN;
1052         if (nr_threads >= max_threads)
1053                 goto bad_fork_cleanup_count;
1054
1055         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1056                 goto bad_fork_cleanup_count;
1057
1058         p->did_exec = 0;
1059         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1060         copy_flags(clone_flags, p);
1061         INIT_LIST_HEAD(&p->children);
1062         INIT_LIST_HEAD(&p->sibling);
1063         rcu_copy_process(p);
1064         p->vfork_done = NULL;
1065         spin_lock_init(&p->alloc_lock);
1066
1067         init_sigpending(&p->pending);
1068
1069         p->utime = cputime_zero;
1070         p->stime = cputime_zero;
1071         p->gtime = cputime_zero;
1072         p->utimescaled = cputime_zero;
1073         p->stimescaled = cputime_zero;
1074 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1075         p->prev_utime = cputime_zero;
1076         p->prev_stime = cputime_zero;
1077 #endif
1078
1079         p->default_timer_slack_ns = current->timer_slack_ns;
1080
1081         task_io_accounting_init(&p->ioac);
1082         acct_clear_integrals(p);
1083
1084         posix_cpu_timers_init(p);
1085
1086         p->lock_depth = -1;             /* -1 = no lock */
1087         do_posix_clock_monotonic_gettime(&p->start_time);
1088         p->real_start_time = p->start_time;
1089         monotonic_to_bootbased(&p->real_start_time);
1090         p->io_context = NULL;
1091         p->audit_context = NULL;
1092         cgroup_fork(p);
1093 #ifdef CONFIG_NUMA
1094         p->mempolicy = mpol_dup(p->mempolicy);
1095         if (IS_ERR(p->mempolicy)) {
1096                 retval = PTR_ERR(p->mempolicy);
1097                 p->mempolicy = NULL;
1098                 goto bad_fork_cleanup_cgroup;
1099         }
1100         mpol_fix_fork_child_flag(p);
1101 #endif
1102 #ifdef CONFIG_TRACE_IRQFLAGS
1103         p->irq_events = 0;
1104 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1105         p->hardirqs_enabled = 1;
1106 #else
1107         p->hardirqs_enabled = 0;
1108 #endif
1109         p->hardirq_enable_ip = 0;
1110         p->hardirq_enable_event = 0;
1111         p->hardirq_disable_ip = _THIS_IP_;
1112         p->hardirq_disable_event = 0;
1113         p->softirqs_enabled = 1;
1114         p->softirq_enable_ip = _THIS_IP_;
1115         p->softirq_enable_event = 0;
1116         p->softirq_disable_ip = 0;
1117         p->softirq_disable_event = 0;
1118         p->hardirq_context = 0;
1119         p->softirq_context = 0;
1120 #endif
1121 #ifdef CONFIG_LOCKDEP
1122         p->lockdep_depth = 0; /* no locks held yet */
1123         p->curr_chain_key = 0;
1124         p->lockdep_recursion = 0;
1125 #endif
1126
1127 #ifdef CONFIG_DEBUG_MUTEXES
1128         p->blocked_on = NULL; /* not blocked yet */
1129 #endif
1130
1131         p->bts = NULL;
1132
1133         p->stack_start = stack_start;
1134
1135         /* Perform scheduler related setup. Assign this task to a CPU. */
1136         sched_fork(p, clone_flags);
1137
1138         retval = perf_event_init_task(p);
1139         if (retval)
1140                 goto bad_fork_cleanup_policy;
1141
1142         if ((retval = audit_alloc(p)))
1143                 goto bad_fork_cleanup_policy;
1144         /* copy all the process information */
1145         if ((retval = copy_semundo(clone_flags, p)))
1146                 goto bad_fork_cleanup_audit;
1147         if ((retval = copy_files(clone_flags, p)))
1148                 goto bad_fork_cleanup_semundo;
1149         if ((retval = copy_fs(clone_flags, p)))
1150                 goto bad_fork_cleanup_files;
1151         if ((retval = copy_sighand(clone_flags, p)))
1152                 goto bad_fork_cleanup_fs;
1153         if ((retval = copy_signal(clone_flags, p)))
1154                 goto bad_fork_cleanup_sighand;
1155         if ((retval = copy_mm(clone_flags, p)))
1156                 goto bad_fork_cleanup_signal;
1157         if ((retval = copy_namespaces(clone_flags, p)))
1158                 goto bad_fork_cleanup_mm;
1159         if ((retval = copy_io(clone_flags, p)))
1160                 goto bad_fork_cleanup_namespaces;
1161         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1162         if (retval)
1163                 goto bad_fork_cleanup_io;
1164
1165         if (pid != &init_struct_pid) {
1166                 retval = -ENOMEM;
1167                 pid = alloc_pid(p->nsproxy->pid_ns);
1168                 if (!pid)
1169                         goto bad_fork_cleanup_io;
1170
1171                 if (clone_flags & CLONE_NEWPID) {
1172                         retval = pid_ns_prepare_proc(p->nsproxy->pid_ns);
1173                         if (retval < 0)
1174                                 goto bad_fork_free_pid;
1175                 }
1176         }
1177
1178         p->pid = pid_nr(pid);
1179         p->tgid = p->pid;
1180         if (clone_flags & CLONE_THREAD)
1181                 p->tgid = current->tgid;
1182
1183         if (current->nsproxy != p->nsproxy) {
1184                 retval = ns_cgroup_clone(p, pid);
1185                 if (retval)
1186                         goto bad_fork_free_pid;
1187         }
1188
1189         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1190         /*
1191          * Clear TID on mm_release()?
1192          */
1193         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr: NULL;
1194 #ifdef CONFIG_FUTEX
1195         p->robust_list = NULL;
1196 #ifdef CONFIG_COMPAT
1197         p->compat_robust_list = NULL;
1198 #endif
1199         INIT_LIST_HEAD(&p->pi_state_list);
1200         p->pi_state_cache = NULL;
1201 #endif
1202         /*
1203          * sigaltstack should be cleared when sharing the same VM
1204          */
1205         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1206                 p->sas_ss_sp = p->sas_ss_size = 0;
1207
1208         /*
1209          * Syscall tracing should be turned off in the child regardless
1210          * of CLONE_PTRACE.
1211          */
1212         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1213 #ifdef TIF_SYSCALL_EMU
1214         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1215 #endif
1216         clear_all_latency_tracing(p);
1217
1218         /* ok, now we should be set up.. */
1219         p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1220         p->pdeath_signal = 0;
1221         p->exit_state = 0;
1222
1223         /*
1224          * Ok, make it visible to the rest of the system.
1225          * We dont wake it up yet.
1226          */
1227         p->group_leader = p;
1228         INIT_LIST_HEAD(&p->thread_group);
1229
1230         /* Now that the task is set up, run cgroup callbacks if
1231          * necessary. We need to run them before the task is visible
1232          * on the tasklist. */
1233         cgroup_fork_callbacks(p);
1234         cgroup_callbacks_done = 1;
1235
1236         /* Need tasklist lock for parent etc handling! */
1237         write_lock_irq(&tasklist_lock);
1238
1239         /*
1240          * The task hasn't been attached yet, so its cpus_allowed mask will
1241          * not be changed, nor will its assigned CPU.
1242          *
1243          * The cpus_allowed mask of the parent may have changed after it was
1244          * copied first time - so re-copy it here, then check the child's CPU
1245          * to ensure it is on a valid CPU (and if not, just force it back to
1246          * parent's CPU). This avoids alot of nasty races.
1247          */
1248         p->cpus_allowed = current->cpus_allowed;
1249         p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
1250         if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
1251                         !cpu_online(task_cpu(p))))
1252                 set_task_cpu(p, smp_processor_id());
1253
1254         /* CLONE_PARENT re-uses the old parent */
1255         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1256                 p->real_parent = current->real_parent;
1257                 p->parent_exec_id = current->parent_exec_id;
1258         } else {
1259                 p->real_parent = current;
1260                 p->parent_exec_id = current->self_exec_id;
1261         }
1262
1263         spin_lock(&current->sighand->siglock);
1264
1265         /*
1266          * Process group and session signals need to be delivered to just the
1267          * parent before the fork or both the parent and the child after the
1268          * fork. Restart if a signal comes in before we add the new process to
1269          * it's process group.
1270          * A fatal signal pending means that current will exit, so the new
1271          * thread can't slip out of an OOM kill (or normal SIGKILL).
1272          */
1273         recalc_sigpending();
1274         if (signal_pending(current)) {
1275                 spin_unlock(&current->sighand->siglock);
1276                 write_unlock_irq(&tasklist_lock);
1277                 retval = -ERESTARTNOINTR;
1278                 goto bad_fork_free_pid;
1279         }
1280
1281         if (clone_flags & CLONE_THREAD) {
1282                 atomic_inc(&current->signal->count);
1283                 atomic_inc(&current->signal->live);
1284                 p->group_leader = current->group_leader;
1285                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1286         }
1287
1288         if (likely(p->pid)) {
1289                 list_add_tail(&p->sibling, &p->real_parent->children);
1290                 tracehook_finish_clone(p, clone_flags, trace);
1291
1292                 if (thread_group_leader(p)) {
1293                         if (clone_flags & CLONE_NEWPID)
1294                                 p->nsproxy->pid_ns->child_reaper = p;
1295
1296                         p->signal->leader_pid = pid;
1297                         tty_kref_put(p->signal->tty);
1298                         p->signal->tty = tty_kref_get(current->signal->tty);
1299                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1300                         attach_pid(p, PIDTYPE_SID, task_session(current));
1301                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1302                         __get_cpu_var(process_counts)++;
1303                 }
1304                 attach_pid(p, PIDTYPE_PID, pid);
1305                 nr_threads++;
1306         }
1307
1308         total_forks++;
1309         spin_unlock(&current->sighand->siglock);
1310         write_unlock_irq(&tasklist_lock);
1311         proc_fork_connector(p);
1312         cgroup_post_fork(p);
1313         perf_event_fork(p);
1314         return p;
1315
1316 bad_fork_free_pid:
1317         if (pid != &init_struct_pid)
1318                 free_pid(pid);
1319 bad_fork_cleanup_io:
1320         if (p->io_context)
1321                 exit_io_context(p);
1322 bad_fork_cleanup_namespaces:
1323         exit_task_namespaces(p);
1324 bad_fork_cleanup_mm:
1325         if (p->mm)
1326                 mmput(p->mm);
1327 bad_fork_cleanup_signal:
1328         if (!(clone_flags & CLONE_THREAD))
1329                 __cleanup_signal(p->signal);
1330 bad_fork_cleanup_sighand:
1331         __cleanup_sighand(p->sighand);
1332 bad_fork_cleanup_fs:
1333         exit_fs(p); /* blocking */
1334 bad_fork_cleanup_files:
1335         exit_files(p); /* blocking */
1336 bad_fork_cleanup_semundo:
1337         exit_sem(p);
1338 bad_fork_cleanup_audit:
1339         audit_free(p);
1340 bad_fork_cleanup_policy:
1341         perf_event_free_task(p);
1342 #ifdef CONFIG_NUMA
1343         mpol_put(p->mempolicy);
1344 bad_fork_cleanup_cgroup:
1345 #endif
1346         cgroup_exit(p, cgroup_callbacks_done);
1347         delayacct_tsk_free(p);
1348         module_put(task_thread_info(p)->exec_domain->module);
1349 bad_fork_cleanup_count:
1350         atomic_dec(&p->cred->user->processes);
1351         exit_creds(p);
1352 bad_fork_free:
1353         free_task(p);
1354 fork_out:
1355         return ERR_PTR(retval);
1356 }
1357
1358 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1359 {
1360         memset(regs, 0, sizeof(struct pt_regs));
1361         return regs;
1362 }
1363
1364 struct task_struct * __cpuinit fork_idle(int cpu)
1365 {
1366         struct task_struct *task;
1367         struct pt_regs regs;
1368
1369         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1370                             &init_struct_pid, 0);
1371         if (!IS_ERR(task))
1372                 init_idle(task, cpu);
1373
1374         return task;
1375 }
1376
1377 /*
1378  *  Ok, this is the main fork-routine.
1379  *
1380  * It copies the process, and if successful kick-starts
1381  * it and waits for it to finish using the VM if required.
1382  */
1383 long do_fork(unsigned long clone_flags,
1384               unsigned long stack_start,
1385               struct pt_regs *regs,
1386               unsigned long stack_size,
1387               int __user *parent_tidptr,
1388               int __user *child_tidptr)
1389 {
1390         struct task_struct *p;
1391         int trace = 0;
1392         long nr;
1393
1394         /*
1395          * Do some preliminary argument and permissions checking before we
1396          * actually start allocating stuff
1397          */
1398         if (clone_flags & CLONE_NEWUSER) {
1399                 if (clone_flags & CLONE_THREAD)
1400                         return -EINVAL;
1401                 /* hopefully this check will go away when userns support is
1402                  * complete
1403                  */
1404                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1405                                 !capable(CAP_SETGID))
1406                         return -EPERM;
1407         }
1408
1409         /*
1410          * We hope to recycle these flags after 2.6.26
1411          */
1412         if (unlikely(clone_flags & CLONE_STOPPED)) {
1413                 static int __read_mostly count = 100;
1414
1415                 if (count > 0 && printk_ratelimit()) {
1416                         char comm[TASK_COMM_LEN];
1417
1418                         count--;
1419                         printk(KERN_INFO "fork(): process `%s' used deprecated "
1420                                         "clone flags 0x%lx\n",
1421                                 get_task_comm(comm, current),
1422                                 clone_flags & CLONE_STOPPED);
1423                 }
1424         }
1425
1426         /*
1427          * When called from kernel_thread, don't do user tracing stuff.
1428          */
1429         if (likely(user_mode(regs)))
1430                 trace = tracehook_prepare_clone(clone_flags);
1431
1432         p = copy_process(clone_flags, stack_start, regs, stack_size,
1433                          child_tidptr, NULL, trace);
1434         /*
1435          * Do this prior waking up the new thread - the thread pointer
1436          * might get invalid after that point, if the thread exits quickly.
1437          */
1438         if (!IS_ERR(p)) {
1439                 struct completion vfork;
1440
1441                 trace_sched_process_fork(current, p);
1442
1443                 nr = task_pid_vnr(p);
1444
1445                 if (clone_flags & CLONE_PARENT_SETTID)
1446                         put_user(nr, parent_tidptr);
1447
1448                 if (clone_flags & CLONE_VFORK) {
1449                         p->vfork_done = &vfork;
1450                         init_completion(&vfork);
1451                 }
1452
1453                 audit_finish_fork(p);
1454                 tracehook_report_clone(regs, clone_flags, nr, p);
1455
1456                 /*
1457                  * We set PF_STARTING at creation in case tracing wants to
1458                  * use this to distinguish a fully live task from one that
1459                  * hasn't gotten to tracehook_report_clone() yet.  Now we
1460                  * clear it and set the child going.
1461                  */
1462                 p->flags &= ~PF_STARTING;
1463
1464                 if (unlikely(clone_flags & CLONE_STOPPED)) {
1465                         /*
1466                          * We'll start up with an immediate SIGSTOP.
1467                          */
1468                         sigaddset(&p->pending.signal, SIGSTOP);
1469                         set_tsk_thread_flag(p, TIF_SIGPENDING);
1470                         __set_task_state(p, TASK_STOPPED);
1471                 } else {
1472                         wake_up_new_task(p, clone_flags);
1473                 }
1474
1475                 tracehook_report_clone_complete(trace, regs,
1476                                                 clone_flags, nr, p);
1477
1478                 if (clone_flags & CLONE_VFORK) {
1479                         freezer_do_not_count();
1480                         wait_for_completion(&vfork);
1481                         freezer_count();
1482                         tracehook_report_vfork_done(p, nr);
1483                 }
1484         } else {
1485                 nr = PTR_ERR(p);
1486         }
1487         return nr;
1488 }
1489
1490 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1491 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1492 #endif
1493
1494 static void sighand_ctor(void *data)
1495 {
1496         struct sighand_struct *sighand = data;
1497
1498         spin_lock_init(&sighand->siglock);
1499         init_waitqueue_head(&sighand->signalfd_wqh);
1500 }
1501
1502 void __init proc_caches_init(void)
1503 {
1504         sighand_cachep = kmem_cache_create("sighand_cache",
1505                         sizeof(struct sighand_struct), 0,
1506                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1507                         SLAB_NOTRACK, sighand_ctor);
1508         signal_cachep = kmem_cache_create("signal_cache",
1509                         sizeof(struct signal_struct), 0,
1510                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1511         files_cachep = kmem_cache_create("files_cache",
1512                         sizeof(struct files_struct), 0,
1513                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1514         fs_cachep = kmem_cache_create("fs_cache",
1515                         sizeof(struct fs_struct), 0,
1516                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1517         mm_cachep = kmem_cache_create("mm_struct",
1518                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1519                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1520         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1521         mmap_init();
1522 }
1523
1524 /*
1525  * Check constraints on flags passed to the unshare system call and
1526  * force unsharing of additional process context as appropriate.
1527  */
1528 static void check_unshare_flags(unsigned long *flags_ptr)
1529 {
1530         /*
1531          * If unsharing a thread from a thread group, must also
1532          * unshare vm.
1533          */
1534         if (*flags_ptr & CLONE_THREAD)
1535                 *flags_ptr |= CLONE_VM;
1536
1537         /*
1538          * If unsharing vm, must also unshare signal handlers.
1539          */
1540         if (*flags_ptr & CLONE_VM)
1541                 *flags_ptr |= CLONE_SIGHAND;
1542
1543         /*
1544          * If unsharing signal handlers and the task was created
1545          * using CLONE_THREAD, then must unshare the thread
1546          */
1547         if ((*flags_ptr & CLONE_SIGHAND) &&
1548             (atomic_read(&current->signal->count) > 1))
1549                 *flags_ptr |= CLONE_THREAD;
1550
1551         /*
1552          * If unsharing namespace, must also unshare filesystem information.
1553          */
1554         if (*flags_ptr & CLONE_NEWNS)
1555                 *flags_ptr |= CLONE_FS;
1556 }
1557
1558 /*
1559  * Unsharing of tasks created with CLONE_THREAD is not supported yet
1560  */
1561 static int unshare_thread(unsigned long unshare_flags)
1562 {
1563         if (unshare_flags & CLONE_THREAD)
1564                 return -EINVAL;
1565
1566         return 0;
1567 }
1568
1569 /*
1570  * Unshare the filesystem structure if it is being shared
1571  */
1572 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1573 {
1574         struct fs_struct *fs = current->fs;
1575
1576         if (!(unshare_flags & CLONE_FS) || !fs)
1577                 return 0;
1578
1579         /* don't need lock here; in the worst case we'll do useless copy */
1580         if (fs->users == 1)
1581                 return 0;
1582
1583         *new_fsp = copy_fs_struct(fs);
1584         if (!*new_fsp)
1585                 return -ENOMEM;
1586
1587         return 0;
1588 }
1589
1590 /*
1591  * Unsharing of sighand is not supported yet
1592  */
1593 static int unshare_sighand(unsigned long unshare_flags, struct sighand_struct **new_sighp)
1594 {
1595         struct sighand_struct *sigh = current->sighand;
1596
1597         if ((unshare_flags & CLONE_SIGHAND) && atomic_read(&sigh->count) > 1)
1598                 return -EINVAL;
1599         else
1600                 return 0;
1601 }
1602
1603 /*
1604  * Unshare vm if it is being shared
1605  */
1606 static int unshare_vm(unsigned long unshare_flags, struct mm_struct **new_mmp)
1607 {
1608         struct mm_struct *mm = current->mm;
1609
1610         if ((unshare_flags & CLONE_VM) &&
1611             (mm && atomic_read(&mm->mm_users) > 1)) {
1612                 return -EINVAL;
1613         }
1614
1615         return 0;
1616 }
1617
1618 /*
1619  * Unshare file descriptor table if it is being shared
1620  */
1621 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1622 {
1623         struct files_struct *fd = current->files;
1624         int error = 0;
1625
1626         if ((unshare_flags & CLONE_FILES) &&
1627             (fd && atomic_read(&fd->count) > 1)) {
1628                 *new_fdp = dup_fd(fd, &error);
1629                 if (!*new_fdp)
1630                         return error;
1631         }
1632
1633         return 0;
1634 }
1635
1636 /*
1637  * unshare allows a process to 'unshare' part of the process
1638  * context which was originally shared using clone.  copy_*
1639  * functions used by do_fork() cannot be used here directly
1640  * because they modify an inactive task_struct that is being
1641  * constructed. Here we are modifying the current, active,
1642  * task_struct.
1643  */
1644 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1645 {
1646         int err = 0;
1647         struct fs_struct *fs, *new_fs = NULL;
1648         struct sighand_struct *new_sigh = NULL;
1649         struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
1650         struct files_struct *fd, *new_fd = NULL;
1651         struct nsproxy *new_nsproxy = NULL;
1652         int do_sysvsem = 0;
1653
1654         check_unshare_flags(&unshare_flags);
1655
1656         /* Return -EINVAL for all unsupported flags */
1657         err = -EINVAL;
1658         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1659                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1660                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1661                 goto bad_unshare_out;
1662
1663         /*
1664          * CLONE_NEWIPC must also detach from the undolist: after switching
1665          * to a new ipc namespace, the semaphore arrays from the old
1666          * namespace are unreachable.
1667          */
1668         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1669                 do_sysvsem = 1;
1670         if ((err = unshare_thread(unshare_flags)))
1671                 goto bad_unshare_out;
1672         if ((err = unshare_fs(unshare_flags, &new_fs)))
1673                 goto bad_unshare_cleanup_thread;
1674         if ((err = unshare_sighand(unshare_flags, &new_sigh)))
1675                 goto bad_unshare_cleanup_fs;
1676         if ((err = unshare_vm(unshare_flags, &new_mm)))
1677                 goto bad_unshare_cleanup_sigh;
1678         if ((err = unshare_fd(unshare_flags, &new_fd)))
1679                 goto bad_unshare_cleanup_vm;
1680         if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1681                         new_fs)))
1682                 goto bad_unshare_cleanup_fd;
1683
1684         if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
1685                 if (do_sysvsem) {
1686                         /*
1687                          * CLONE_SYSVSEM is equivalent to sys_exit().
1688                          */
1689                         exit_sem(current);
1690                 }
1691
1692                 if (new_nsproxy) {
1693                         switch_task_namespaces(current, new_nsproxy);
1694                         new_nsproxy = NULL;
1695                 }
1696
1697                 task_lock(current);
1698
1699                 if (new_fs) {
1700                         fs = current->fs;
1701                         write_lock(&fs->lock);
1702                         current->fs = new_fs;
1703                         if (--fs->users)
1704                                 new_fs = NULL;
1705                         else
1706                                 new_fs = fs;
1707                         write_unlock(&fs->lock);
1708                 }
1709
1710                 if (new_mm) {
1711                         mm = current->mm;
1712                         active_mm = current->active_mm;
1713                         current->mm = new_mm;
1714                         current->active_mm = new_mm;
1715                         activate_mm(active_mm, new_mm);
1716                         new_mm = mm;
1717                 }
1718
1719                 if (new_fd) {
1720                         fd = current->files;
1721                         current->files = new_fd;
1722                         new_fd = fd;
1723                 }
1724
1725                 task_unlock(current);
1726         }
1727
1728         if (new_nsproxy)
1729                 put_nsproxy(new_nsproxy);
1730
1731 bad_unshare_cleanup_fd:
1732         if (new_fd)
1733                 put_files_struct(new_fd);
1734
1735 bad_unshare_cleanup_vm:
1736         if (new_mm)
1737                 mmput(new_mm);
1738
1739 bad_unshare_cleanup_sigh:
1740         if (new_sigh)
1741                 if (atomic_dec_and_test(&new_sigh->count))
1742                         kmem_cache_free(sighand_cachep, new_sigh);
1743
1744 bad_unshare_cleanup_fs:
1745         if (new_fs)
1746                 free_fs_struct(new_fs);
1747
1748 bad_unshare_cleanup_thread:
1749 bad_unshare_out:
1750         return err;
1751 }
1752
1753 /*
1754  *      Helper to unshare the files of the current task.
1755  *      We don't want to expose copy_files internals to
1756  *      the exec layer of the kernel.
1757  */
1758
1759 int unshare_files(struct files_struct **displaced)
1760 {
1761         struct task_struct *task = current;
1762         struct files_struct *copy = NULL;
1763         int error;
1764
1765         error = unshare_fd(CLONE_FILES, &copy);
1766         if (error || !copy) {
1767                 *displaced = NULL;
1768                 return error;
1769         }
1770         *displaced = task->files;
1771         task_lock(task);
1772         task->files = copy;
1773         task_unlock(task);
1774         return 0;
1775 }