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