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