5d22b9b8cf7b5f156ace2de279ff4a6ee6f9b133
[linux-3.10.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/futex.h>
41 #include <linux/compat.h>
42 #include <linux/kthread.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/blkdev.h>
62 #include <linux/fs_struct.h>
63 #include <linux/magic.h>
64 #include <linux/perf_event.h>
65 #include <linux/posix-timers.h>
66 #include <linux/user-return-notifier.h>
67 #include <linux/oom.h>
68 #include <linux/khugepaged.h>
69 #include <linux/signalfd.h>
70
71 #include <asm/pgtable.h>
72 #include <asm/pgalloc.h>
73 #include <asm/uaccess.h>
74 #include <asm/mmu_context.h>
75 #include <asm/cacheflush.h>
76 #include <asm/tlbflush.h>
77
78 #include <trace/events/sched.h>
79
80 #define CREATE_TRACE_POINTS
81 #include <trace/events/task.h>
82
83 /*
84  * Protected counters by write_lock_irq(&tasklist_lock)
85  */
86 unsigned long total_forks;      /* Handle normal Linux uptimes. */
87 int nr_threads;                 /* The idle threads do not count.. */
88
89 int max_threads;                /* tunable limit on nr_threads */
90
91 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
92
93 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
94
95 #ifdef CONFIG_PROVE_RCU
96 int lockdep_tasklist_lock_is_held(void)
97 {
98         return lockdep_is_held(&tasklist_lock);
99 }
100 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
101 #endif /* #ifdef CONFIG_PROVE_RCU */
102
103 int nr_processes(void)
104 {
105         int cpu;
106         int total = 0;
107
108         for_each_possible_cpu(cpu)
109                 total += per_cpu(process_counts, cpu);
110
111         return total;
112 }
113
114 #ifndef __HAVE_ARCH_TASK_STRUCT_ALLOCATOR
115 # define alloc_task_struct_node(node)           \
116                 kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node)
117 # define free_task_struct(tsk)                  \
118                 kmem_cache_free(task_struct_cachep, (tsk))
119 static struct kmem_cache *task_struct_cachep;
120 #endif
121
122 #ifndef __HAVE_ARCH_THREAD_INFO_ALLOCATOR
123 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
124                                                   int node)
125 {
126         struct page *page = alloc_pages_node(node, THREADINFO_GFP,
127                                              THREAD_SIZE_ORDER);
128
129         return page ? page_address(page) : NULL;
130 }
131
132 static inline void free_thread_info(struct thread_info *ti)
133 {
134         free_pages((unsigned long)ti, THREAD_SIZE_ORDER);
135 }
136 #endif
137
138 /* SLAB cache for signal_struct structures (tsk->signal) */
139 static struct kmem_cache *signal_cachep;
140
141 /* SLAB cache for sighand_struct structures (tsk->sighand) */
142 struct kmem_cache *sighand_cachep;
143
144 /* SLAB cache for files_struct structures (tsk->files) */
145 struct kmem_cache *files_cachep;
146
147 /* SLAB cache for fs_struct structures (tsk->fs) */
148 struct kmem_cache *fs_cachep;
149
150 /* SLAB cache for vm_area_struct structures */
151 struct kmem_cache *vm_area_cachep;
152
153 /* SLAB cache for mm_struct structures (tsk->mm) */
154 static struct kmem_cache *mm_cachep;
155
156 static void account_kernel_stack(struct thread_info *ti, int account)
157 {
158         struct zone *zone = page_zone(virt_to_page(ti));
159
160         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
161 }
162
163 void free_task(struct task_struct *tsk)
164 {
165         account_kernel_stack(tsk->stack, -1);
166         free_thread_info(tsk->stack);
167         rt_mutex_debug_task_free(tsk);
168         ftrace_graph_exit_task(tsk);
169         free_task_struct(tsk);
170 }
171 EXPORT_SYMBOL(free_task);
172
173 static inline void free_signal_struct(struct signal_struct *sig)
174 {
175         taskstats_tgid_free(sig);
176         sched_autogroup_exit(sig);
177         kmem_cache_free(signal_cachep, sig);
178 }
179
180 static inline void put_signal_struct(struct signal_struct *sig)
181 {
182         if (atomic_dec_and_test(&sig->sigcnt))
183                 free_signal_struct(sig);
184 }
185
186 void __put_task_struct(struct task_struct *tsk)
187 {
188         WARN_ON(!tsk->exit_state);
189         WARN_ON(atomic_read(&tsk->usage));
190         WARN_ON(tsk == current);
191
192         security_task_free(tsk);
193         exit_creds(tsk);
194         delayacct_tsk_free(tsk);
195         put_signal_struct(tsk->signal);
196
197         if (!profile_handoff_task(tsk))
198                 free_task(tsk);
199 }
200 EXPORT_SYMBOL_GPL(__put_task_struct);
201
202 void __init __weak arch_task_cache_init(void) { }
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         int node = tsk_fork_get_node(orig);
251         int err;
252
253         prepare_to_copy(orig);
254
255         tsk = alloc_task_struct_node(node);
256         if (!tsk)
257                 return NULL;
258
259         ti = alloc_thread_info_node(tsk, node);
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         setup_thread_stack(tsk, orig);
272         clear_user_return_notifier(tsk);
273         clear_tsk_need_resched(tsk);
274         stackend = end_of_stack(tsk);
275         *stackend = STACK_END_MAGIC;    /* for overflow detection */
276
277 #ifdef CONFIG_CC_STACKPROTECTOR
278         tsk->stack_canary = get_random_int();
279 #endif
280
281         /*
282          * One for us, one for whoever does the "release_task()" (usually
283          * parent)
284          */
285         atomic_set(&tsk->usage, 2);
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         retval = khugepaged_fork(mm, oldmm);
332         if (retval)
333                 goto out;
334
335         prev = NULL;
336         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
337                 struct file *file;
338
339                 if (mpnt->vm_flags & VM_DONTCOPY) {
340                         long pages = vma_pages(mpnt);
341                         mm->total_vm -= pages;
342                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
343                                                                 -pages);
344                         continue;
345                 }
346                 charge = 0;
347                 if (mpnt->vm_flags & VM_ACCOUNT) {
348                         unsigned int len = (mpnt->vm_end - mpnt->vm_start) >> PAGE_SHIFT;
349                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
350                                 goto fail_nomem;
351                         charge = len;
352                 }
353                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
354                 if (!tmp)
355                         goto fail_nomem;
356                 *tmp = *mpnt;
357                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
358                 pol = mpol_dup(vma_policy(mpnt));
359                 retval = PTR_ERR(pol);
360                 if (IS_ERR(pol))
361                         goto fail_nomem_policy;
362                 vma_set_policy(tmp, pol);
363                 tmp->vm_mm = mm;
364                 if (anon_vma_fork(tmp, mpnt))
365                         goto fail_nomem_anon_vma_fork;
366                 tmp->vm_flags &= ~VM_LOCKED;
367                 tmp->vm_next = tmp->vm_prev = NULL;
368                 file = tmp->vm_file;
369                 if (file) {
370                         struct inode *inode = file->f_path.dentry->d_inode;
371                         struct address_space *mapping = file->f_mapping;
372
373                         get_file(file);
374                         if (tmp->vm_flags & VM_DENYWRITE)
375                                 atomic_dec(&inode->i_writecount);
376                         mutex_lock(&mapping->i_mmap_mutex);
377                         if (tmp->vm_flags & VM_SHARED)
378                                 mapping->i_mmap_writable++;
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                         mutex_unlock(&mapping->i_mmap_mutex);
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
495         if (likely(!mm_alloc_pgd(mm))) {
496                 mm->def_flags = 0;
497                 mmu_notifier_mm_init(mm);
498                 return mm;
499         }
500
501         free_mm(mm);
502         return NULL;
503 }
504
505 static void check_mm(struct mm_struct *mm)
506 {
507         int i;
508
509         for (i = 0; i < NR_MM_COUNTERS; i++) {
510                 long x = atomic_long_read(&mm->rss_stat.count[i]);
511
512                 if (unlikely(x))
513                         printk(KERN_ALERT "BUG: Bad rss-counter state "
514                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
515         }
516
517 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
518         VM_BUG_ON(mm->pmd_huge_pte);
519 #endif
520 }
521
522 /*
523  * Allocate and initialize an mm_struct.
524  */
525 struct mm_struct *mm_alloc(void)
526 {
527         struct mm_struct *mm;
528
529         mm = allocate_mm();
530         if (!mm)
531                 return NULL;
532
533         memset(mm, 0, sizeof(*mm));
534         mm_init_cpumask(mm);
535         return mm_init(mm, current);
536 }
537
538 /*
539  * Called when the last reference to the mm
540  * is dropped: either by a lazy thread or by
541  * mmput. Free the page directory and the mm.
542  */
543 void __mmdrop(struct mm_struct *mm)
544 {
545         BUG_ON(mm == &init_mm);
546         mm_free_pgd(mm);
547         destroy_context(mm);
548         mmu_notifier_mm_destroy(mm);
549         check_mm(mm);
550         free_mm(mm);
551 }
552 EXPORT_SYMBOL_GPL(__mmdrop);
553
554 /*
555  * Decrement the use count and release all resources for an mm.
556  */
557 void mmput(struct mm_struct *mm)
558 {
559         might_sleep();
560
561         if (atomic_dec_and_test(&mm->mm_users)) {
562                 exit_aio(mm);
563                 ksm_exit(mm);
564                 khugepaged_exit(mm); /* must run before exit_mmap */
565                 exit_mmap(mm);
566                 set_mm_exe_file(mm, NULL);
567                 if (!list_empty(&mm->mmlist)) {
568                         spin_lock(&mmlist_lock);
569                         list_del(&mm->mmlist);
570                         spin_unlock(&mmlist_lock);
571                 }
572                 put_swap_token(mm);
573                 if (mm->binfmt)
574                         module_put(mm->binfmt->module);
575                 mmdrop(mm);
576         }
577 }
578 EXPORT_SYMBOL_GPL(mmput);
579
580 /*
581  * We added or removed a vma mapping the executable. The vmas are only mapped
582  * during exec and are not mapped with the mmap system call.
583  * Callers must hold down_write() on the mm's mmap_sem for these
584  */
585 void added_exe_file_vma(struct mm_struct *mm)
586 {
587         mm->num_exe_file_vmas++;
588 }
589
590 void removed_exe_file_vma(struct mm_struct *mm)
591 {
592         mm->num_exe_file_vmas--;
593         if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
594                 fput(mm->exe_file);
595                 mm->exe_file = NULL;
596         }
597
598 }
599
600 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
601 {
602         if (new_exe_file)
603                 get_file(new_exe_file);
604         if (mm->exe_file)
605                 fput(mm->exe_file);
606         mm->exe_file = new_exe_file;
607         mm->num_exe_file_vmas = 0;
608 }
609
610 struct file *get_mm_exe_file(struct mm_struct *mm)
611 {
612         struct file *exe_file;
613
614         /* We need mmap_sem to protect against races with removal of
615          * VM_EXECUTABLE vmas */
616         down_read(&mm->mmap_sem);
617         exe_file = mm->exe_file;
618         if (exe_file)
619                 get_file(exe_file);
620         up_read(&mm->mmap_sem);
621         return exe_file;
622 }
623
624 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
625 {
626         /* It's safe to write the exe_file pointer without exe_file_lock because
627          * this is called during fork when the task is not yet in /proc */
628         newmm->exe_file = get_mm_exe_file(oldmm);
629 }
630
631 /**
632  * get_task_mm - acquire a reference to the task's mm
633  *
634  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
635  * this kernel workthread has transiently adopted a user mm with use_mm,
636  * to do its AIO) is not set and if so returns a reference to it, after
637  * bumping up the use count.  User must release the mm via mmput()
638  * after use.  Typically used by /proc and ptrace.
639  */
640 struct mm_struct *get_task_mm(struct task_struct *task)
641 {
642         struct mm_struct *mm;
643
644         task_lock(task);
645         mm = task->mm;
646         if (mm) {
647                 if (task->flags & PF_KTHREAD)
648                         mm = NULL;
649                 else
650                         atomic_inc(&mm->mm_users);
651         }
652         task_unlock(task);
653         return mm;
654 }
655 EXPORT_SYMBOL_GPL(get_task_mm);
656
657 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
658 {
659         struct mm_struct *mm;
660         int err;
661
662         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
663         if (err)
664                 return ERR_PTR(err);
665
666         mm = get_task_mm(task);
667         if (mm && mm != current->mm &&
668                         !ptrace_may_access(task, mode)) {
669                 mmput(mm);
670                 mm = ERR_PTR(-EACCES);
671         }
672         mutex_unlock(&task->signal->cred_guard_mutex);
673
674         return mm;
675 }
676
677 static void complete_vfork_done(struct task_struct *tsk)
678 {
679         struct completion *vfork;
680
681         task_lock(tsk);
682         vfork = tsk->vfork_done;
683         if (likely(vfork)) {
684                 tsk->vfork_done = NULL;
685                 complete(vfork);
686         }
687         task_unlock(tsk);
688 }
689
690 static int wait_for_vfork_done(struct task_struct *child,
691                                 struct completion *vfork)
692 {
693         int killed;
694
695         freezer_do_not_count();
696         killed = wait_for_completion_killable(vfork);
697         freezer_count();
698
699         if (killed) {
700                 task_lock(child);
701                 child->vfork_done = NULL;
702                 task_unlock(child);
703         }
704
705         put_task_struct(child);
706         return killed;
707 }
708
709 /* Please note the differences between mmput and mm_release.
710  * mmput is called whenever we stop holding onto a mm_struct,
711  * error success whatever.
712  *
713  * mm_release is called after a mm_struct has been removed
714  * from the current process.
715  *
716  * This difference is important for error handling, when we
717  * only half set up a mm_struct for a new process and need to restore
718  * the old one.  Because we mmput the new mm_struct before
719  * restoring the old one. . .
720  * Eric Biederman 10 January 1998
721  */
722 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
723 {
724         /* Get rid of any futexes when releasing the mm */
725 #ifdef CONFIG_FUTEX
726         if (unlikely(tsk->robust_list)) {
727                 exit_robust_list(tsk);
728                 tsk->robust_list = NULL;
729         }
730 #ifdef CONFIG_COMPAT
731         if (unlikely(tsk->compat_robust_list)) {
732                 compat_exit_robust_list(tsk);
733                 tsk->compat_robust_list = NULL;
734         }
735 #endif
736         if (unlikely(!list_empty(&tsk->pi_state_list)))
737                 exit_pi_state_list(tsk);
738 #endif
739
740         /* Get rid of any cached register state */
741         deactivate_mm(tsk, mm);
742
743         if (tsk->vfork_done)
744                 complete_vfork_done(tsk);
745
746         /*
747          * If we're exiting normally, clear a user-space tid field if
748          * requested.  We leave this alone when dying by signal, to leave
749          * the value intact in a core dump, and to save the unnecessary
750          * trouble, say, a killed vfork parent shouldn't touch this mm.
751          * Userland only wants this done for a sys_exit.
752          */
753         if (tsk->clear_child_tid) {
754                 if (!(tsk->flags & PF_SIGNALED) &&
755                     atomic_read(&mm->mm_users) > 1) {
756                         /*
757                          * We don't check the error code - if userspace has
758                          * not set up a proper pointer then tough luck.
759                          */
760                         put_user(0, tsk->clear_child_tid);
761                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
762                                         1, NULL, NULL, 0);
763                 }
764                 tsk->clear_child_tid = NULL;
765         }
766 }
767
768 /*
769  * Allocate a new mm structure and copy contents from the
770  * mm structure of the passed in task structure.
771  */
772 struct mm_struct *dup_mm(struct task_struct *tsk)
773 {
774         struct mm_struct *mm, *oldmm = current->mm;
775         int err;
776
777         if (!oldmm)
778                 return NULL;
779
780         mm = allocate_mm();
781         if (!mm)
782                 goto fail_nomem;
783
784         memcpy(mm, oldmm, sizeof(*mm));
785         mm_init_cpumask(mm);
786
787         /* Initializing for Swap token stuff */
788         mm->token_priority = 0;
789         mm->last_interval = 0;
790
791 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
792         mm->pmd_huge_pte = NULL;
793 #endif
794
795         if (!mm_init(mm, tsk))
796                 goto fail_nomem;
797
798         if (init_new_context(tsk, mm))
799                 goto fail_nocontext;
800
801         dup_mm_exe_file(oldmm, mm);
802
803         err = dup_mmap(mm, oldmm);
804         if (err)
805                 goto free_pt;
806
807         mm->hiwater_rss = get_mm_rss(mm);
808         mm->hiwater_vm = mm->total_vm;
809
810         if (mm->binfmt && !try_module_get(mm->binfmt->module))
811                 goto free_pt;
812
813         return mm;
814
815 free_pt:
816         /* don't put binfmt in mmput, we haven't got module yet */
817         mm->binfmt = NULL;
818         mmput(mm);
819
820 fail_nomem:
821         return NULL;
822
823 fail_nocontext:
824         /*
825          * If init_new_context() failed, we cannot use mmput() to free the mm
826          * because it calls destroy_context()
827          */
828         mm_free_pgd(mm);
829         free_mm(mm);
830         return NULL;
831 }
832
833 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
834 {
835         struct mm_struct *mm, *oldmm;
836         int retval;
837
838         tsk->min_flt = tsk->maj_flt = 0;
839         tsk->nvcsw = tsk->nivcsw = 0;
840 #ifdef CONFIG_DETECT_HUNG_TASK
841         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
842 #endif
843
844         tsk->mm = NULL;
845         tsk->active_mm = NULL;
846
847         /*
848          * Are we cloning a kernel thread?
849          *
850          * We need to steal a active VM for that..
851          */
852         oldmm = current->mm;
853         if (!oldmm)
854                 return 0;
855
856         if (clone_flags & CLONE_VM) {
857                 atomic_inc(&oldmm->mm_users);
858                 mm = oldmm;
859                 goto good_mm;
860         }
861
862         retval = -ENOMEM;
863         mm = dup_mm(tsk);
864         if (!mm)
865                 goto fail_nomem;
866
867 good_mm:
868         /* Initializing for Swap token stuff */
869         mm->token_priority = 0;
870         mm->last_interval = 0;
871
872         tsk->mm = mm;
873         tsk->active_mm = mm;
874         return 0;
875
876 fail_nomem:
877         return retval;
878 }
879
880 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
881 {
882         struct fs_struct *fs = current->fs;
883         if (clone_flags & CLONE_FS) {
884                 /* tsk->fs is already what we want */
885                 spin_lock(&fs->lock);
886                 if (fs->in_exec) {
887                         spin_unlock(&fs->lock);
888                         return -EAGAIN;
889                 }
890                 fs->users++;
891                 spin_unlock(&fs->lock);
892                 return 0;
893         }
894         tsk->fs = copy_fs_struct(fs);
895         if (!tsk->fs)
896                 return -ENOMEM;
897         return 0;
898 }
899
900 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
901 {
902         struct files_struct *oldf, *newf;
903         int error = 0;
904
905         /*
906          * A background process may not have any files ...
907          */
908         oldf = current->files;
909         if (!oldf)
910                 goto out;
911
912         if (clone_flags & CLONE_FILES) {
913                 atomic_inc(&oldf->count);
914                 goto out;
915         }
916
917         newf = dup_fd(oldf, &error);
918         if (!newf)
919                 goto out;
920
921         tsk->files = newf;
922         error = 0;
923 out:
924         return error;
925 }
926
927 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
928 {
929 #ifdef CONFIG_BLOCK
930         struct io_context *ioc = current->io_context;
931         struct io_context *new_ioc;
932
933         if (!ioc)
934                 return 0;
935         /*
936          * Share io context with parent, if CLONE_IO is set
937          */
938         if (clone_flags & CLONE_IO) {
939                 tsk->io_context = ioc_task_link(ioc);
940                 if (unlikely(!tsk->io_context))
941                         return -ENOMEM;
942         } else if (ioprio_valid(ioc->ioprio)) {
943                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
944                 if (unlikely(!new_ioc))
945                         return -ENOMEM;
946
947                 new_ioc->ioprio = ioc->ioprio;
948                 put_io_context(new_ioc);
949         }
950 #endif
951         return 0;
952 }
953
954 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
955 {
956         struct sighand_struct *sig;
957
958         if (clone_flags & CLONE_SIGHAND) {
959                 atomic_inc(&current->sighand->count);
960                 return 0;
961         }
962         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
963         rcu_assign_pointer(tsk->sighand, sig);
964         if (!sig)
965                 return -ENOMEM;
966         atomic_set(&sig->count, 1);
967         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
968         return 0;
969 }
970
971 void __cleanup_sighand(struct sighand_struct *sighand)
972 {
973         if (atomic_dec_and_test(&sighand->count)) {
974                 signalfd_cleanup(sighand);
975                 kmem_cache_free(sighand_cachep, sighand);
976         }
977 }
978
979
980 /*
981  * Initialize POSIX timer handling for a thread group.
982  */
983 static void posix_cpu_timers_init_group(struct signal_struct *sig)
984 {
985         unsigned long cpu_limit;
986
987         /* Thread group counters. */
988         thread_group_cputime_init(sig);
989
990         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
991         if (cpu_limit != RLIM_INFINITY) {
992                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
993                 sig->cputimer.running = 1;
994         }
995
996         /* The timer lists. */
997         INIT_LIST_HEAD(&sig->cpu_timers[0]);
998         INIT_LIST_HEAD(&sig->cpu_timers[1]);
999         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1000 }
1001
1002 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1003 {
1004         struct signal_struct *sig;
1005
1006         if (clone_flags & CLONE_THREAD)
1007                 return 0;
1008
1009         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1010         tsk->signal = sig;
1011         if (!sig)
1012                 return -ENOMEM;
1013
1014         sig->nr_threads = 1;
1015         atomic_set(&sig->live, 1);
1016         atomic_set(&sig->sigcnt, 1);
1017         init_waitqueue_head(&sig->wait_chldexit);
1018         if (clone_flags & CLONE_NEWPID)
1019                 sig->flags |= SIGNAL_UNKILLABLE;
1020         sig->curr_target = tsk;
1021         init_sigpending(&sig->shared_pending);
1022         INIT_LIST_HEAD(&sig->posix_timers);
1023
1024         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1025         sig->real_timer.function = it_real_fn;
1026
1027         task_lock(current->group_leader);
1028         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1029         task_unlock(current->group_leader);
1030
1031         posix_cpu_timers_init_group(sig);
1032
1033         tty_audit_fork(sig);
1034         sched_autogroup_fork(sig);
1035
1036 #ifdef CONFIG_CGROUPS
1037         init_rwsem(&sig->group_rwsem);
1038 #endif
1039
1040         sig->oom_adj = current->signal->oom_adj;
1041         sig->oom_score_adj = current->signal->oom_score_adj;
1042         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1043
1044         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1045                                    current->signal->is_child_subreaper;
1046
1047         mutex_init(&sig->cred_guard_mutex);
1048
1049         return 0;
1050 }
1051
1052 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1053 {
1054         unsigned long new_flags = p->flags;
1055
1056         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1057         new_flags |= PF_FORKNOEXEC;
1058         p->flags = new_flags;
1059 }
1060
1061 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1062 {
1063         current->clear_child_tid = tidptr;
1064
1065         return task_pid_vnr(current);
1066 }
1067
1068 static void rt_mutex_init_task(struct task_struct *p)
1069 {
1070         raw_spin_lock_init(&p->pi_lock);
1071 #ifdef CONFIG_RT_MUTEXES
1072         plist_head_init(&p->pi_waiters);
1073         p->pi_blocked_on = NULL;
1074 #endif
1075 }
1076
1077 #ifdef CONFIG_MM_OWNER
1078 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1079 {
1080         mm->owner = p;
1081 }
1082 #endif /* CONFIG_MM_OWNER */
1083
1084 /*
1085  * Initialize POSIX timer handling for a single task.
1086  */
1087 static void posix_cpu_timers_init(struct task_struct *tsk)
1088 {
1089         tsk->cputime_expires.prof_exp = 0;
1090         tsk->cputime_expires.virt_exp = 0;
1091         tsk->cputime_expires.sched_exp = 0;
1092         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1093         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1094         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1095 }
1096
1097 /*
1098  * This creates a new process as a copy of the old one,
1099  * but does not actually start it yet.
1100  *
1101  * It copies the registers, and all the appropriate
1102  * parts of the process environment (as per the clone
1103  * flags). The actual kick-off is left to the caller.
1104  */
1105 static struct task_struct *copy_process(unsigned long clone_flags,
1106                                         unsigned long stack_start,
1107                                         struct pt_regs *regs,
1108                                         unsigned long stack_size,
1109                                         int __user *child_tidptr,
1110                                         struct pid *pid,
1111                                         int trace)
1112 {
1113         int retval;
1114         struct task_struct *p;
1115         int cgroup_callbacks_done = 0;
1116
1117         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1118                 return ERR_PTR(-EINVAL);
1119
1120         /*
1121          * Thread groups must share signals as well, and detached threads
1122          * can only be started up within the thread group.
1123          */
1124         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1125                 return ERR_PTR(-EINVAL);
1126
1127         /*
1128          * Shared signal handlers imply shared VM. By way of the above,
1129          * thread groups also imply shared VM. Blocking this case allows
1130          * for various simplifications in other code.
1131          */
1132         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1133                 return ERR_PTR(-EINVAL);
1134
1135         /*
1136          * Siblings of global init remain as zombies on exit since they are
1137          * not reaped by their parent (swapper). To solve this and to avoid
1138          * multi-rooted process trees, prevent global and container-inits
1139          * from creating siblings.
1140          */
1141         if ((clone_flags & CLONE_PARENT) &&
1142                                 current->signal->flags & SIGNAL_UNKILLABLE)
1143                 return ERR_PTR(-EINVAL);
1144
1145         retval = security_task_create(clone_flags);
1146         if (retval)
1147                 goto fork_out;
1148
1149         retval = -ENOMEM;
1150         p = dup_task_struct(current);
1151         if (!p)
1152                 goto fork_out;
1153
1154         ftrace_graph_init_task(p);
1155
1156         rt_mutex_init_task(p);
1157
1158 #ifdef CONFIG_PROVE_LOCKING
1159         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1160         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1161 #endif
1162         retval = -EAGAIN;
1163         if (atomic_read(&p->real_cred->user->processes) >=
1164                         task_rlimit(p, RLIMIT_NPROC)) {
1165                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1166                     p->real_cred->user != INIT_USER)
1167                         goto bad_fork_free;
1168         }
1169         current->flags &= ~PF_NPROC_EXCEEDED;
1170
1171         retval = copy_creds(p, clone_flags);
1172         if (retval < 0)
1173                 goto bad_fork_free;
1174
1175         /*
1176          * If multiple threads are within copy_process(), then this check
1177          * triggers too late. This doesn't hurt, the check is only there
1178          * to stop root fork bombs.
1179          */
1180         retval = -EAGAIN;
1181         if (nr_threads >= max_threads)
1182                 goto bad_fork_cleanup_count;
1183
1184         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1185                 goto bad_fork_cleanup_count;
1186
1187         p->did_exec = 0;
1188         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1189         copy_flags(clone_flags, p);
1190         INIT_LIST_HEAD(&p->children);
1191         INIT_LIST_HEAD(&p->sibling);
1192         rcu_copy_process(p);
1193         p->vfork_done = NULL;
1194         spin_lock_init(&p->alloc_lock);
1195
1196         init_sigpending(&p->pending);
1197
1198         p->utime = p->stime = p->gtime = 0;
1199         p->utimescaled = p->stimescaled = 0;
1200 #ifndef CONFIG_VIRT_CPU_ACCOUNTING
1201         p->prev_utime = p->prev_stime = 0;
1202 #endif
1203 #if defined(SPLIT_RSS_COUNTING)
1204         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1205 #endif
1206
1207         p->default_timer_slack_ns = current->timer_slack_ns;
1208
1209         task_io_accounting_init(&p->ioac);
1210         acct_clear_integrals(p);
1211
1212         posix_cpu_timers_init(p);
1213
1214         do_posix_clock_monotonic_gettime(&p->start_time);
1215         p->real_start_time = p->start_time;
1216         monotonic_to_bootbased(&p->real_start_time);
1217         p->io_context = NULL;
1218         p->audit_context = NULL;
1219         if (clone_flags & CLONE_THREAD)
1220                 threadgroup_change_begin(current);
1221         cgroup_fork(p);
1222 #ifdef CONFIG_NUMA
1223         p->mempolicy = mpol_dup(p->mempolicy);
1224         if (IS_ERR(p->mempolicy)) {
1225                 retval = PTR_ERR(p->mempolicy);
1226                 p->mempolicy = NULL;
1227                 goto bad_fork_cleanup_cgroup;
1228         }
1229         mpol_fix_fork_child_flag(p);
1230 #endif
1231 #ifdef CONFIG_CPUSETS
1232         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1233         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1234         seqcount_init(&p->mems_allowed_seq);
1235 #endif
1236 #ifdef CONFIG_TRACE_IRQFLAGS
1237         p->irq_events = 0;
1238 #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1239         p->hardirqs_enabled = 1;
1240 #else
1241         p->hardirqs_enabled = 0;
1242 #endif
1243         p->hardirq_enable_ip = 0;
1244         p->hardirq_enable_event = 0;
1245         p->hardirq_disable_ip = _THIS_IP_;
1246         p->hardirq_disable_event = 0;
1247         p->softirqs_enabled = 1;
1248         p->softirq_enable_ip = _THIS_IP_;
1249         p->softirq_enable_event = 0;
1250         p->softirq_disable_ip = 0;
1251         p->softirq_disable_event = 0;
1252         p->hardirq_context = 0;
1253         p->softirq_context = 0;
1254 #endif
1255 #ifdef CONFIG_LOCKDEP
1256         p->lockdep_depth = 0; /* no locks held yet */
1257         p->curr_chain_key = 0;
1258         p->lockdep_recursion = 0;
1259 #endif
1260
1261 #ifdef CONFIG_DEBUG_MUTEXES
1262         p->blocked_on = NULL; /* not blocked yet */
1263 #endif
1264 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
1265         p->memcg_batch.do_batch = 0;
1266         p->memcg_batch.memcg = NULL;
1267 #endif
1268
1269         /* Perform scheduler related setup. Assign this task to a CPU. */
1270         sched_fork(p);
1271
1272         retval = perf_event_init_task(p);
1273         if (retval)
1274                 goto bad_fork_cleanup_policy;
1275         retval = audit_alloc(p);
1276         if (retval)
1277                 goto bad_fork_cleanup_policy;
1278         /* copy all the process information */
1279         retval = copy_semundo(clone_flags, p);
1280         if (retval)
1281                 goto bad_fork_cleanup_audit;
1282         retval = copy_files(clone_flags, p);
1283         if (retval)
1284                 goto bad_fork_cleanup_semundo;
1285         retval = copy_fs(clone_flags, p);
1286         if (retval)
1287                 goto bad_fork_cleanup_files;
1288         retval = copy_sighand(clone_flags, p);
1289         if (retval)
1290                 goto bad_fork_cleanup_fs;
1291         retval = copy_signal(clone_flags, p);
1292         if (retval)
1293                 goto bad_fork_cleanup_sighand;
1294         retval = copy_mm(clone_flags, p);
1295         if (retval)
1296                 goto bad_fork_cleanup_signal;
1297         retval = copy_namespaces(clone_flags, p);
1298         if (retval)
1299                 goto bad_fork_cleanup_mm;
1300         retval = copy_io(clone_flags, p);
1301         if (retval)
1302                 goto bad_fork_cleanup_namespaces;
1303         retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1304         if (retval)
1305                 goto bad_fork_cleanup_io;
1306
1307         if (pid != &init_struct_pid) {
1308                 retval = -ENOMEM;
1309                 pid = alloc_pid(p->nsproxy->pid_ns);
1310                 if (!pid)
1311                         goto bad_fork_cleanup_io;
1312         }
1313
1314         p->pid = pid_nr(pid);
1315         p->tgid = p->pid;
1316         if (clone_flags & CLONE_THREAD)
1317                 p->tgid = current->tgid;
1318
1319         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1320         /*
1321          * Clear TID on mm_release()?
1322          */
1323         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1324 #ifdef CONFIG_BLOCK
1325         p->plug = NULL;
1326 #endif
1327 #ifdef CONFIG_FUTEX
1328         p->robust_list = NULL;
1329 #ifdef CONFIG_COMPAT
1330         p->compat_robust_list = NULL;
1331 #endif
1332         INIT_LIST_HEAD(&p->pi_state_list);
1333         p->pi_state_cache = NULL;
1334 #endif
1335         /*
1336          * sigaltstack should be cleared when sharing the same VM
1337          */
1338         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1339                 p->sas_ss_sp = p->sas_ss_size = 0;
1340
1341         /*
1342          * Syscall tracing and stepping should be turned off in the
1343          * child regardless of CLONE_PTRACE.
1344          */
1345         user_disable_single_step(p);
1346         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1347 #ifdef TIF_SYSCALL_EMU
1348         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1349 #endif
1350         clear_all_latency_tracing(p);
1351
1352         /* ok, now we should be set up.. */
1353         if (clone_flags & CLONE_THREAD)
1354                 p->exit_signal = -1;
1355         else if (clone_flags & CLONE_PARENT)
1356                 p->exit_signal = current->group_leader->exit_signal;
1357         else
1358                 p->exit_signal = (clone_flags & CSIGNAL);
1359
1360         p->pdeath_signal = 0;
1361         p->exit_state = 0;
1362
1363         p->nr_dirtied = 0;
1364         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1365         p->dirty_paused_when = 0;
1366
1367         /*
1368          * Ok, make it visible to the rest of the system.
1369          * We dont wake it up yet.
1370          */
1371         p->group_leader = p;
1372         INIT_LIST_HEAD(&p->thread_group);
1373
1374         /* Now that the task is set up, run cgroup callbacks if
1375          * necessary. We need to run them before the task is visible
1376          * on the tasklist. */
1377         cgroup_fork_callbacks(p);
1378         cgroup_callbacks_done = 1;
1379
1380         /* Need tasklist lock for parent etc handling! */
1381         write_lock_irq(&tasklist_lock);
1382
1383         /* CLONE_PARENT re-uses the old parent */
1384         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1385                 p->real_parent = current->real_parent;
1386                 p->parent_exec_id = current->parent_exec_id;
1387         } else {
1388                 p->real_parent = current;
1389                 p->parent_exec_id = current->self_exec_id;
1390         }
1391
1392         spin_lock(&current->sighand->siglock);
1393
1394         /*
1395          * Process group and session signals need to be delivered to just the
1396          * parent before the fork or both the parent and the child after the
1397          * fork. Restart if a signal comes in before we add the new process to
1398          * it's process group.
1399          * A fatal signal pending means that current will exit, so the new
1400          * thread can't slip out of an OOM kill (or normal SIGKILL).
1401         */
1402         recalc_sigpending();
1403         if (signal_pending(current)) {
1404                 spin_unlock(&current->sighand->siglock);
1405                 write_unlock_irq(&tasklist_lock);
1406                 retval = -ERESTARTNOINTR;
1407                 goto bad_fork_free_pid;
1408         }
1409
1410         if (clone_flags & CLONE_THREAD) {
1411                 current->signal->nr_threads++;
1412                 atomic_inc(&current->signal->live);
1413                 atomic_inc(&current->signal->sigcnt);
1414                 p->group_leader = current->group_leader;
1415                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1416         }
1417
1418         if (likely(p->pid)) {
1419                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1420
1421                 if (thread_group_leader(p)) {
1422                         if (is_child_reaper(pid))
1423                                 p->nsproxy->pid_ns->child_reaper = p;
1424
1425                         p->signal->leader_pid = pid;
1426                         p->signal->tty = tty_kref_get(current->signal->tty);
1427                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1428                         attach_pid(p, PIDTYPE_SID, task_session(current));
1429                         list_add_tail(&p->sibling, &p->real_parent->children);
1430                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1431                         __this_cpu_inc(process_counts);
1432                 }
1433                 attach_pid(p, PIDTYPE_PID, pid);
1434                 nr_threads++;
1435         }
1436
1437         total_forks++;
1438         spin_unlock(&current->sighand->siglock);
1439         write_unlock_irq(&tasklist_lock);
1440         proc_fork_connector(p);
1441         cgroup_post_fork(p);
1442         if (clone_flags & CLONE_THREAD)
1443                 threadgroup_change_end(current);
1444         perf_event_fork(p);
1445
1446         trace_task_newtask(p, clone_flags);
1447
1448         return p;
1449
1450 bad_fork_free_pid:
1451         if (pid != &init_struct_pid)
1452                 free_pid(pid);
1453 bad_fork_cleanup_io:
1454         if (p->io_context)
1455                 exit_io_context(p);
1456 bad_fork_cleanup_namespaces:
1457         exit_task_namespaces(p);
1458 bad_fork_cleanup_mm:
1459         if (p->mm)
1460                 mmput(p->mm);
1461 bad_fork_cleanup_signal:
1462         if (!(clone_flags & CLONE_THREAD))
1463                 free_signal_struct(p->signal);
1464 bad_fork_cleanup_sighand:
1465         __cleanup_sighand(p->sighand);
1466 bad_fork_cleanup_fs:
1467         exit_fs(p); /* blocking */
1468 bad_fork_cleanup_files:
1469         exit_files(p); /* blocking */
1470 bad_fork_cleanup_semundo:
1471         exit_sem(p);
1472 bad_fork_cleanup_audit:
1473         audit_free(p);
1474 bad_fork_cleanup_policy:
1475         perf_event_free_task(p);
1476 #ifdef CONFIG_NUMA
1477         mpol_put(p->mempolicy);
1478 bad_fork_cleanup_cgroup:
1479 #endif
1480         if (clone_flags & CLONE_THREAD)
1481                 threadgroup_change_end(current);
1482         cgroup_exit(p, cgroup_callbacks_done);
1483         delayacct_tsk_free(p);
1484         module_put(task_thread_info(p)->exec_domain->module);
1485 bad_fork_cleanup_count:
1486         atomic_dec(&p->cred->user->processes);
1487         exit_creds(p);
1488 bad_fork_free:
1489         free_task(p);
1490 fork_out:
1491         return ERR_PTR(retval);
1492 }
1493
1494 noinline struct pt_regs * __cpuinit __attribute__((weak)) idle_regs(struct pt_regs *regs)
1495 {
1496         memset(regs, 0, sizeof(struct pt_regs));
1497         return regs;
1498 }
1499
1500 static inline void init_idle_pids(struct pid_link *links)
1501 {
1502         enum pid_type type;
1503
1504         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1505                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1506                 links[type].pid = &init_struct_pid;
1507         }
1508 }
1509
1510 struct task_struct * __cpuinit fork_idle(int cpu)
1511 {
1512         struct task_struct *task;
1513         struct pt_regs regs;
1514
1515         task = copy_process(CLONE_VM, 0, idle_regs(&regs), 0, NULL,
1516                             &init_struct_pid, 0);
1517         if (!IS_ERR(task)) {
1518                 init_idle_pids(task->pids);
1519                 init_idle(task, cpu);
1520         }
1521
1522         return task;
1523 }
1524
1525 /*
1526  *  Ok, this is the main fork-routine.
1527  *
1528  * It copies the process, and if successful kick-starts
1529  * it and waits for it to finish using the VM if required.
1530  */
1531 long do_fork(unsigned long clone_flags,
1532               unsigned long stack_start,
1533               struct pt_regs *regs,
1534               unsigned long stack_size,
1535               int __user *parent_tidptr,
1536               int __user *child_tidptr)
1537 {
1538         struct task_struct *p;
1539         int trace = 0;
1540         long nr;
1541
1542         /*
1543          * Do some preliminary argument and permissions checking before we
1544          * actually start allocating stuff
1545          */
1546         if (clone_flags & CLONE_NEWUSER) {
1547                 if (clone_flags & CLONE_THREAD)
1548                         return -EINVAL;
1549                 /* hopefully this check will go away when userns support is
1550                  * complete
1551                  */
1552                 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SETUID) ||
1553                                 !capable(CAP_SETGID))
1554                         return -EPERM;
1555         }
1556
1557         /*
1558          * Determine whether and which event to report to ptracer.  When
1559          * called from kernel_thread or CLONE_UNTRACED is explicitly
1560          * requested, no event is reported; otherwise, report if the event
1561          * for the type of forking is enabled.
1562          */
1563         if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) {
1564                 if (clone_flags & CLONE_VFORK)
1565                         trace = PTRACE_EVENT_VFORK;
1566                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1567                         trace = PTRACE_EVENT_CLONE;
1568                 else
1569                         trace = PTRACE_EVENT_FORK;
1570
1571                 if (likely(!ptrace_event_enabled(current, trace)))
1572                         trace = 0;
1573         }
1574
1575         p = copy_process(clone_flags, stack_start, regs, stack_size,
1576                          child_tidptr, NULL, trace);
1577         /*
1578          * Do this prior waking up the new thread - the thread pointer
1579          * might get invalid after that point, if the thread exits quickly.
1580          */
1581         if (!IS_ERR(p)) {
1582                 struct completion vfork;
1583
1584                 trace_sched_process_fork(current, p);
1585
1586                 nr = task_pid_vnr(p);
1587
1588                 if (clone_flags & CLONE_PARENT_SETTID)
1589                         put_user(nr, parent_tidptr);
1590
1591                 if (clone_flags & CLONE_VFORK) {
1592                         p->vfork_done = &vfork;
1593                         init_completion(&vfork);
1594                         get_task_struct(p);
1595                 }
1596
1597                 wake_up_new_task(p);
1598
1599                 /* forking complete and child started to run, tell ptracer */
1600                 if (unlikely(trace))
1601                         ptrace_event(trace, nr);
1602
1603                 if (clone_flags & CLONE_VFORK) {
1604                         if (!wait_for_vfork_done(p, &vfork))
1605                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1606                 }
1607         } else {
1608                 nr = PTR_ERR(p);
1609         }
1610         return nr;
1611 }
1612
1613 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1614 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1615 #endif
1616
1617 static void sighand_ctor(void *data)
1618 {
1619         struct sighand_struct *sighand = data;
1620
1621         spin_lock_init(&sighand->siglock);
1622         init_waitqueue_head(&sighand->signalfd_wqh);
1623 }
1624
1625 void __init proc_caches_init(void)
1626 {
1627         sighand_cachep = kmem_cache_create("sighand_cache",
1628                         sizeof(struct sighand_struct), 0,
1629                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1630                         SLAB_NOTRACK, sighand_ctor);
1631         signal_cachep = kmem_cache_create("signal_cache",
1632                         sizeof(struct signal_struct), 0,
1633                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1634         files_cachep = kmem_cache_create("files_cache",
1635                         sizeof(struct files_struct), 0,
1636                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1637         fs_cachep = kmem_cache_create("fs_cache",
1638                         sizeof(struct fs_struct), 0,
1639                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1640         /*
1641          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1642          * whole struct cpumask for the OFFSTACK case. We could change
1643          * this to *only* allocate as much of it as required by the
1644          * maximum number of CPU's we can ever have.  The cpumask_allocation
1645          * is at the end of the structure, exactly for that reason.
1646          */
1647         mm_cachep = kmem_cache_create("mm_struct",
1648                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1649                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1650         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1651         mmap_init();
1652         nsproxy_cache_init();
1653 }
1654
1655 /*
1656  * Check constraints on flags passed to the unshare system call.
1657  */
1658 static int check_unshare_flags(unsigned long unshare_flags)
1659 {
1660         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1661                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1662                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET))
1663                 return -EINVAL;
1664         /*
1665          * Not implemented, but pretend it works if there is nothing to
1666          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1667          * needs to unshare vm.
1668          */
1669         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1670                 /* FIXME: get_task_mm() increments ->mm_users */
1671                 if (atomic_read(&current->mm->mm_users) > 1)
1672                         return -EINVAL;
1673         }
1674
1675         return 0;
1676 }
1677
1678 /*
1679  * Unshare the filesystem structure if it is being shared
1680  */
1681 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1682 {
1683         struct fs_struct *fs = current->fs;
1684
1685         if (!(unshare_flags & CLONE_FS) || !fs)
1686                 return 0;
1687
1688         /* don't need lock here; in the worst case we'll do useless copy */
1689         if (fs->users == 1)
1690                 return 0;
1691
1692         *new_fsp = copy_fs_struct(fs);
1693         if (!*new_fsp)
1694                 return -ENOMEM;
1695
1696         return 0;
1697 }
1698
1699 /*
1700  * Unshare file descriptor table if it is being shared
1701  */
1702 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1703 {
1704         struct files_struct *fd = current->files;
1705         int error = 0;
1706
1707         if ((unshare_flags & CLONE_FILES) &&
1708             (fd && atomic_read(&fd->count) > 1)) {
1709                 *new_fdp = dup_fd(fd, &error);
1710                 if (!*new_fdp)
1711                         return error;
1712         }
1713
1714         return 0;
1715 }
1716
1717 /*
1718  * unshare allows a process to 'unshare' part of the process
1719  * context which was originally shared using clone.  copy_*
1720  * functions used by do_fork() cannot be used here directly
1721  * because they modify an inactive task_struct that is being
1722  * constructed. Here we are modifying the current, active,
1723  * task_struct.
1724  */
1725 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1726 {
1727         struct fs_struct *fs, *new_fs = NULL;
1728         struct files_struct *fd, *new_fd = NULL;
1729         struct nsproxy *new_nsproxy = NULL;
1730         int do_sysvsem = 0;
1731         int err;
1732
1733         err = check_unshare_flags(unshare_flags);
1734         if (err)
1735                 goto bad_unshare_out;
1736
1737         /*
1738          * If unsharing namespace, must also unshare filesystem information.
1739          */
1740         if (unshare_flags & CLONE_NEWNS)
1741                 unshare_flags |= CLONE_FS;
1742         /*
1743          * CLONE_NEWIPC must also detach from the undolist: after switching
1744          * to a new ipc namespace, the semaphore arrays from the old
1745          * namespace are unreachable.
1746          */
1747         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1748                 do_sysvsem = 1;
1749         err = unshare_fs(unshare_flags, &new_fs);
1750         if (err)
1751                 goto bad_unshare_out;
1752         err = unshare_fd(unshare_flags, &new_fd);
1753         if (err)
1754                 goto bad_unshare_cleanup_fs;
1755         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy, new_fs);
1756         if (err)
1757                 goto bad_unshare_cleanup_fd;
1758
1759         if (new_fs || new_fd || do_sysvsem || new_nsproxy) {
1760                 if (do_sysvsem) {
1761                         /*
1762                          * CLONE_SYSVSEM is equivalent to sys_exit().
1763                          */
1764                         exit_sem(current);
1765                 }
1766
1767                 if (new_nsproxy) {
1768                         switch_task_namespaces(current, new_nsproxy);
1769                         new_nsproxy = NULL;
1770                 }
1771
1772                 task_lock(current);
1773
1774                 if (new_fs) {
1775                         fs = current->fs;
1776                         spin_lock(&fs->lock);
1777                         current->fs = new_fs;
1778                         if (--fs->users)
1779                                 new_fs = NULL;
1780                         else
1781                                 new_fs = fs;
1782                         spin_unlock(&fs->lock);
1783                 }
1784
1785                 if (new_fd) {
1786                         fd = current->files;
1787                         current->files = new_fd;
1788                         new_fd = fd;
1789                 }
1790
1791                 task_unlock(current);
1792         }
1793
1794         if (new_nsproxy)
1795                 put_nsproxy(new_nsproxy);
1796
1797 bad_unshare_cleanup_fd:
1798         if (new_fd)
1799                 put_files_struct(new_fd);
1800
1801 bad_unshare_cleanup_fs:
1802         if (new_fs)
1803                 free_fs_struct(new_fs);
1804
1805 bad_unshare_out:
1806         return err;
1807 }
1808
1809 /*
1810  *      Helper to unshare the files of the current task.
1811  *      We don't want to expose copy_files internals to
1812  *      the exec layer of the kernel.
1813  */
1814
1815 int unshare_files(struct files_struct **displaced)
1816 {
1817         struct task_struct *task = current;
1818         struct files_struct *copy = NULL;
1819         int error;
1820
1821         error = unshare_fd(CLONE_FILES, &copy);
1822         if (error || !copy) {
1823                 *displaced = NULL;
1824                 return error;
1825         }
1826         *displaced = task->files;
1827         task_lock(task);
1828         task->files = copy;
1829         task_unlock(task);
1830         return 0;
1831 }