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