987b28a1f01b6c6ce5d554eb22d9c89e1cde1e7b
[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/seccomp.h>
38 #include <linux/swap.h>
39 #include <linux/syscalls.h>
40 #include <linux/jiffies.h>
41 #include <linux/futex.h>
42 #include <linux/compat.h>
43 #include <linux/kthread.h>
44 #include <linux/task_io_accounting_ops.h>
45 #include <linux/rcupdate.h>
46 #include <linux/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/audit.h>
49 #include <linux/memcontrol.h>
50 #include <linux/ftrace.h>
51 #include <linux/proc_fs.h>
52 #include <linux/profile.h>
53 #include <linux/rmap.h>
54 #include <linux/ksm.h>
55 #include <linux/acct.h>
56 #include <linux/tsacct_kern.h>
57 #include <linux/cn_proc.h>
58 #include <linux/freezer.h>
59 #include <linux/delayacct.h>
60 #include <linux/taskstats_kern.h>
61 #include <linux/random.h>
62 #include <linux/tty.h>
63 #include <linux/blkdev.h>
64 #include <linux/fs_struct.h>
65 #include <linux/magic.h>
66 #include <linux/perf_event.h>
67 #include <linux/posix-timers.h>
68 #include <linux/user-return-notifier.h>
69 #include <linux/oom.h>
70 #include <linux/khugepaged.h>
71 #include <linux/signalfd.h>
72 #include <linux/uprobes.h>
73 #include <linux/aio.h>
74
75 #include <asm/pgtable.h>
76 #include <asm/pgalloc.h>
77 #include <asm/uaccess.h>
78 #include <asm/mmu_context.h>
79 #include <asm/cacheflush.h>
80 #include <asm/tlbflush.h>
81
82 #include <trace/events/sched.h>
83
84 #define CREATE_TRACE_POINTS
85 #include <trace/events/task.h>
86
87 /*
88  * Protected counters by write_lock_irq(&tasklist_lock)
89  */
90 unsigned long total_forks;      /* Handle normal Linux uptimes. */
91 int nr_threads;                 /* The idle threads do not count.. */
92
93 int max_threads;                /* tunable limit on nr_threads */
94
95 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
96
97 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
98
99 #ifdef CONFIG_PROVE_RCU
100 int lockdep_tasklist_lock_is_held(void)
101 {
102         return lockdep_is_held(&tasklist_lock);
103 }
104 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
105 #endif /* #ifdef CONFIG_PROVE_RCU */
106
107 int nr_processes(void)
108 {
109         int cpu;
110         int total = 0;
111
112         for_each_possible_cpu(cpu)
113                 total += per_cpu(process_counts, cpu);
114
115         return total;
116 }
117
118 void __weak arch_release_task_struct(struct task_struct *tsk)
119 {
120 }
121
122 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
123 static struct kmem_cache *task_struct_cachep;
124
125 static inline struct task_struct *alloc_task_struct_node(int node)
126 {
127         return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
128 }
129
130 static inline void free_task_struct(struct task_struct *tsk)
131 {
132         kmem_cache_free(task_struct_cachep, tsk);
133 }
134 #endif
135
136 void __weak arch_release_thread_info(struct thread_info *ti)
137 {
138 }
139
140 #ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
141
142 /*
143  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
144  * kmemcache based allocator.
145  */
146 # if THREAD_SIZE >= PAGE_SIZE
147 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
148                                                   int node)
149 {
150         struct page *page = alloc_pages_node(node, THREADINFO_GFP_ACCOUNTED,
151                                              THREAD_SIZE_ORDER);
152
153         return page ? page_address(page) : NULL;
154 }
155
156 static inline void free_thread_info(struct thread_info *ti)
157 {
158         free_memcg_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
159 }
160 # else
161 static struct kmem_cache *thread_info_cache;
162
163 static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
164                                                   int node)
165 {
166         return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
167 }
168
169 static void free_thread_info(struct thread_info *ti)
170 {
171         kmem_cache_free(thread_info_cache, ti);
172 }
173
174 void thread_info_cache_init(void)
175 {
176         thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
177                                               THREAD_SIZE, 0, NULL);
178         BUG_ON(thread_info_cache == NULL);
179 }
180 # endif
181 #endif
182
183 /* SLAB cache for signal_struct structures (tsk->signal) */
184 static struct kmem_cache *signal_cachep;
185
186 /* SLAB cache for sighand_struct structures (tsk->sighand) */
187 struct kmem_cache *sighand_cachep;
188
189 /* SLAB cache for files_struct structures (tsk->files) */
190 struct kmem_cache *files_cachep;
191
192 /* SLAB cache for fs_struct structures (tsk->fs) */
193 struct kmem_cache *fs_cachep;
194
195 /* SLAB cache for vm_area_struct structures */
196 struct kmem_cache *vm_area_cachep;
197
198 /* SLAB cache for mm_struct structures (tsk->mm) */
199 static struct kmem_cache *mm_cachep;
200
201 static void account_kernel_stack(struct thread_info *ti, int account)
202 {
203         struct zone *zone = page_zone(virt_to_page(ti));
204
205         mod_zone_page_state(zone, NR_KERNEL_STACK, account);
206 }
207
208 void free_task(struct task_struct *tsk)
209 {
210         account_kernel_stack(tsk->stack, -1);
211         arch_release_thread_info(tsk->stack);
212         free_thread_info(tsk->stack);
213         rt_mutex_debug_task_free(tsk);
214         ftrace_graph_exit_task(tsk);
215         put_seccomp_filter(tsk);
216         arch_release_task_struct(tsk);
217         free_task_struct(tsk);
218 }
219 EXPORT_SYMBOL(free_task);
220
221 static inline void free_signal_struct(struct signal_struct *sig)
222 {
223         taskstats_tgid_free(sig);
224         sched_autogroup_exit(sig);
225         kmem_cache_free(signal_cachep, sig);
226 }
227
228 static inline void put_signal_struct(struct signal_struct *sig)
229 {
230         if (atomic_dec_and_test(&sig->sigcnt))
231                 free_signal_struct(sig);
232 }
233
234 void __put_task_struct(struct task_struct *tsk)
235 {
236         WARN_ON(!tsk->exit_state);
237         WARN_ON(atomic_read(&tsk->usage));
238         WARN_ON(tsk == current);
239
240         security_task_free(tsk);
241         exit_creds(tsk);
242         delayacct_tsk_free(tsk);
243         put_signal_struct(tsk->signal);
244
245         if (!profile_handoff_task(tsk))
246                 free_task(tsk);
247 }
248 EXPORT_SYMBOL_GPL(__put_task_struct);
249
250 void __init __weak arch_task_cache_init(void) { }
251
252 void __init fork_init(unsigned long mempages)
253 {
254 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
255 #ifndef ARCH_MIN_TASKALIGN
256 #define ARCH_MIN_TASKALIGN      L1_CACHE_BYTES
257 #endif
258         /* create a slab on which task_structs can be allocated */
259         task_struct_cachep =
260                 kmem_cache_create("task_struct", sizeof(struct task_struct),
261                         ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
262 #endif
263
264         /* do the arch specific task caches init */
265         arch_task_cache_init();
266
267         /*
268          * The default maximum number of threads is set to a safe
269          * value: the thread structures can take up at most half
270          * of memory.
271          */
272         max_threads = mempages / (8 * THREAD_SIZE / PAGE_SIZE);
273
274         /*
275          * we need to allow at least 20 threads to boot a system
276          */
277         if (max_threads < 20)
278                 max_threads = 20;
279
280         init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
281         init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
282         init_task.signal->rlim[RLIMIT_SIGPENDING] =
283                 init_task.signal->rlim[RLIMIT_NPROC];
284 }
285
286 int __attribute__((weak)) arch_dup_task_struct(struct task_struct *dst,
287                                                struct task_struct *src)
288 {
289         *dst = *src;
290         return 0;
291 }
292
293 static struct task_struct *dup_task_struct(struct task_struct *orig)
294 {
295         struct task_struct *tsk;
296         struct thread_info *ti;
297         unsigned long *stackend;
298         int node = tsk_fork_get_node(orig);
299         int err;
300
301         tsk = alloc_task_struct_node(node);
302         if (!tsk)
303                 return NULL;
304
305         ti = alloc_thread_info_node(tsk, node);
306         if (!ti)
307                 goto free_tsk;
308
309         err = arch_dup_task_struct(tsk, orig);
310         if (err)
311                 goto free_ti;
312
313         tsk->stack = ti;
314
315         setup_thread_stack(tsk, orig);
316         clear_user_return_notifier(tsk);
317         clear_tsk_need_resched(tsk);
318         stackend = end_of_stack(tsk);
319         *stackend = STACK_END_MAGIC;    /* for overflow detection */
320
321 #ifdef CONFIG_CC_STACKPROTECTOR
322         tsk->stack_canary = get_random_int();
323 #endif
324
325         /*
326          * One for us, one for whoever does the "release_task()" (usually
327          * parent)
328          */
329         atomic_set(&tsk->usage, 2);
330 #ifdef CONFIG_BLK_DEV_IO_TRACE
331         tsk->btrace_seq = 0;
332 #endif
333         tsk->splice_pipe = NULL;
334         tsk->task_frag.page = NULL;
335
336         account_kernel_stack(ti, 1);
337
338         return tsk;
339
340 free_ti:
341         free_thread_info(ti);
342 free_tsk:
343         free_task_struct(tsk);
344         return NULL;
345 }
346
347 #ifdef CONFIG_MMU
348 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
349 {
350         struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
351         struct rb_node **rb_link, *rb_parent;
352         int retval;
353         unsigned long charge;
354         struct mempolicy *pol;
355
356         uprobe_start_dup_mmap();
357         down_write(&oldmm->mmap_sem);
358         flush_cache_dup_mm(oldmm);
359         uprobe_dup_mmap(oldmm, mm);
360         /*
361          * Not linked in yet - no deadlock potential:
362          */
363         down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
364
365         mm->locked_vm = 0;
366         mm->mmap = NULL;
367         mm->mmap_cache = NULL;
368         mm->free_area_cache = oldmm->mmap_base;
369         mm->cached_hole_size = ~0UL;
370         mm->map_count = 0;
371         cpumask_clear(mm_cpumask(mm));
372         mm->mm_rb = RB_ROOT;
373         rb_link = &mm->mm_rb.rb_node;
374         rb_parent = NULL;
375         pprev = &mm->mmap;
376         retval = ksm_fork(mm, oldmm);
377         if (retval)
378                 goto out;
379         retval = khugepaged_fork(mm, oldmm);
380         if (retval)
381                 goto out;
382
383         prev = NULL;
384         for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
385                 struct file *file;
386
387                 if (mpnt->vm_flags & VM_DONTCOPY) {
388                         vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
389                                                         -vma_pages(mpnt));
390                         continue;
391                 }
392                 charge = 0;
393                 if (mpnt->vm_flags & VM_ACCOUNT) {
394                         unsigned long len = vma_pages(mpnt);
395
396                         if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
397                                 goto fail_nomem;
398                         charge = len;
399                 }
400                 tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
401                 if (!tmp)
402                         goto fail_nomem;
403                 *tmp = *mpnt;
404                 INIT_LIST_HEAD(&tmp->anon_vma_chain);
405                 pol = mpol_dup(vma_policy(mpnt));
406                 retval = PTR_ERR(pol);
407                 if (IS_ERR(pol))
408                         goto fail_nomem_policy;
409                 vma_set_policy(tmp, pol);
410                 tmp->vm_mm = mm;
411                 if (anon_vma_fork(tmp, mpnt))
412                         goto fail_nomem_anon_vma_fork;
413                 tmp->vm_flags &= ~VM_LOCKED;
414                 tmp->vm_next = tmp->vm_prev = NULL;
415                 file = tmp->vm_file;
416                 if (file) {
417                         struct inode *inode = file_inode(file);
418                         struct address_space *mapping = file->f_mapping;
419
420                         get_file(file);
421                         if (tmp->vm_flags & VM_DENYWRITE)
422                                 atomic_dec(&inode->i_writecount);
423                         mutex_lock(&mapping->i_mmap_mutex);
424                         if (tmp->vm_flags & VM_SHARED)
425                                 mapping->i_mmap_writable++;
426                         flush_dcache_mmap_lock(mapping);
427                         /* insert tmp into the share list, just after mpnt */
428                         if (unlikely(tmp->vm_flags & VM_NONLINEAR))
429                                 vma_nonlinear_insert(tmp,
430                                                 &mapping->i_mmap_nonlinear);
431                         else
432                                 vma_interval_tree_insert_after(tmp, mpnt,
433                                                         &mapping->i_mmap);
434                         flush_dcache_mmap_unlock(mapping);
435                         mutex_unlock(&mapping->i_mmap_mutex);
436                 }
437
438                 /*
439                  * Clear hugetlb-related page reserves for children. This only
440                  * affects MAP_PRIVATE mappings. Faults generated by the child
441                  * are not guaranteed to succeed, even if read-only
442                  */
443                 if (is_vm_hugetlb_page(tmp))
444                         reset_vma_resv_huge_pages(tmp);
445
446                 /*
447                  * Link in the new vma and copy the page table entries.
448                  */
449                 *pprev = tmp;
450                 pprev = &tmp->vm_next;
451                 tmp->vm_prev = prev;
452                 prev = tmp;
453
454                 __vma_link_rb(mm, tmp, rb_link, rb_parent);
455                 rb_link = &tmp->vm_rb.rb_right;
456                 rb_parent = &tmp->vm_rb;
457
458                 mm->map_count++;
459                 retval = copy_page_range(mm, oldmm, mpnt);
460
461                 if (tmp->vm_ops && tmp->vm_ops->open)
462                         tmp->vm_ops->open(tmp);
463
464                 if (retval)
465                         goto out;
466         }
467         /* a new mm has just been created */
468         arch_dup_mmap(oldmm, mm);
469         retval = 0;
470 out:
471         up_write(&mm->mmap_sem);
472         flush_tlb_mm(oldmm);
473         up_write(&oldmm->mmap_sem);
474         uprobe_end_dup_mmap();
475         return retval;
476 fail_nomem_anon_vma_fork:
477         mpol_put(pol);
478 fail_nomem_policy:
479         kmem_cache_free(vm_area_cachep, tmp);
480 fail_nomem:
481         retval = -ENOMEM;
482         vm_unacct_memory(charge);
483         goto out;
484 }
485
486 static inline int mm_alloc_pgd(struct mm_struct *mm)
487 {
488         mm->pgd = pgd_alloc(mm);
489         if (unlikely(!mm->pgd))
490                 return -ENOMEM;
491         return 0;
492 }
493
494 static inline void mm_free_pgd(struct mm_struct *mm)
495 {
496         pgd_free(mm, mm->pgd);
497 }
498 #else
499 #define dup_mmap(mm, oldmm)     (0)
500 #define mm_alloc_pgd(mm)        (0)
501 #define mm_free_pgd(mm)
502 #endif /* CONFIG_MMU */
503
504 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
505
506 #define allocate_mm()   (kmem_cache_alloc(mm_cachep, GFP_KERNEL))
507 #define free_mm(mm)     (kmem_cache_free(mm_cachep, (mm)))
508
509 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
510
511 static int __init coredump_filter_setup(char *s)
512 {
513         default_dump_filter =
514                 (simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
515                 MMF_DUMP_FILTER_MASK;
516         return 1;
517 }
518
519 __setup("coredump_filter=", coredump_filter_setup);
520
521 #include <linux/init_task.h>
522
523 static void mm_init_aio(struct mm_struct *mm)
524 {
525 #ifdef CONFIG_AIO
526         spin_lock_init(&mm->ioctx_lock);
527         INIT_HLIST_HEAD(&mm->ioctx_list);
528 #endif
529 }
530
531 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
532 {
533         atomic_set(&mm->mm_users, 1);
534         atomic_set(&mm->mm_count, 1);
535         init_rwsem(&mm->mmap_sem);
536         INIT_LIST_HEAD(&mm->mmlist);
537         mm->flags = (current->mm) ?
538                 (current->mm->flags & MMF_INIT_MASK) : default_dump_filter;
539         mm->core_state = NULL;
540         mm->nr_ptes = 0;
541         memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
542         spin_lock_init(&mm->page_table_lock);
543         mm->free_area_cache = TASK_UNMAPPED_BASE;
544         mm->cached_hole_size = ~0UL;
545         mm_init_aio(mm);
546         mm_init_owner(mm, p);
547
548         if (likely(!mm_alloc_pgd(mm))) {
549                 mm->def_flags = 0;
550                 mmu_notifier_mm_init(mm);
551                 return mm;
552         }
553
554         free_mm(mm);
555         return NULL;
556 }
557
558 static void check_mm(struct mm_struct *mm)
559 {
560         int i;
561
562         for (i = 0; i < NR_MM_COUNTERS; i++) {
563                 long x = atomic_long_read(&mm->rss_stat.count[i]);
564
565                 if (unlikely(x))
566                         printk(KERN_ALERT "BUG: Bad rss-counter state "
567                                           "mm:%p idx:%d val:%ld\n", mm, i, x);
568         }
569
570 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
571         VM_BUG_ON(mm->pmd_huge_pte);
572 #endif
573 }
574
575 /*
576  * Allocate and initialize an mm_struct.
577  */
578 struct mm_struct *mm_alloc(void)
579 {
580         struct mm_struct *mm;
581
582         mm = allocate_mm();
583         if (!mm)
584                 return NULL;
585
586         memset(mm, 0, sizeof(*mm));
587         mm_init_cpumask(mm);
588         return mm_init(mm, current);
589 }
590
591 /*
592  * Called when the last reference to the mm
593  * is dropped: either by a lazy thread or by
594  * mmput. Free the page directory and the mm.
595  */
596 void __mmdrop(struct mm_struct *mm)
597 {
598         BUG_ON(mm == &init_mm);
599         mm_free_pgd(mm);
600         destroy_context(mm);
601         mmu_notifier_mm_destroy(mm);
602         check_mm(mm);
603         free_mm(mm);
604 }
605 EXPORT_SYMBOL_GPL(__mmdrop);
606
607 /*
608  * Decrement the use count and release all resources for an mm.
609  */
610 void mmput(struct mm_struct *mm)
611 {
612         might_sleep();
613
614         if (atomic_dec_and_test(&mm->mm_users)) {
615                 uprobe_clear_state(mm);
616                 exit_aio(mm);
617                 ksm_exit(mm);
618                 khugepaged_exit(mm); /* must run before exit_mmap */
619                 exit_mmap(mm);
620                 set_mm_exe_file(mm, NULL);
621                 if (!list_empty(&mm->mmlist)) {
622                         spin_lock(&mmlist_lock);
623                         list_del(&mm->mmlist);
624                         spin_unlock(&mmlist_lock);
625                 }
626                 if (mm->binfmt)
627                         module_put(mm->binfmt->module);
628                 mmdrop(mm);
629         }
630 }
631 EXPORT_SYMBOL_GPL(mmput);
632
633 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
634 {
635         if (new_exe_file)
636                 get_file(new_exe_file);
637         if (mm->exe_file)
638                 fput(mm->exe_file);
639         mm->exe_file = new_exe_file;
640 }
641
642 struct file *get_mm_exe_file(struct mm_struct *mm)
643 {
644         struct file *exe_file;
645
646         /* We need mmap_sem to protect against races with removal of exe_file */
647         down_read(&mm->mmap_sem);
648         exe_file = mm->exe_file;
649         if (exe_file)
650                 get_file(exe_file);
651         up_read(&mm->mmap_sem);
652         return exe_file;
653 }
654
655 static void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
656 {
657         /* It's safe to write the exe_file pointer without exe_file_lock because
658          * this is called during fork when the task is not yet in /proc */
659         newmm->exe_file = get_mm_exe_file(oldmm);
660 }
661
662 /**
663  * get_task_mm - acquire a reference to the task's mm
664  *
665  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
666  * this kernel workthread has transiently adopted a user mm with use_mm,
667  * to do its AIO) is not set and if so returns a reference to it, after
668  * bumping up the use count.  User must release the mm via mmput()
669  * after use.  Typically used by /proc and ptrace.
670  */
671 struct mm_struct *get_task_mm(struct task_struct *task)
672 {
673         struct mm_struct *mm;
674
675         task_lock(task);
676         mm = task->mm;
677         if (mm) {
678                 if (task->flags & PF_KTHREAD)
679                         mm = NULL;
680                 else
681                         atomic_inc(&mm->mm_users);
682         }
683         task_unlock(task);
684         return mm;
685 }
686 EXPORT_SYMBOL_GPL(get_task_mm);
687
688 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
689 {
690         struct mm_struct *mm;
691         int err;
692
693         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
694         if (err)
695                 return ERR_PTR(err);
696
697         mm = get_task_mm(task);
698         if (mm && mm != current->mm &&
699                         !ptrace_may_access(task, mode)) {
700                 mmput(mm);
701                 mm = ERR_PTR(-EACCES);
702         }
703         mutex_unlock(&task->signal->cred_guard_mutex);
704
705         return mm;
706 }
707
708 static void complete_vfork_done(struct task_struct *tsk)
709 {
710         struct completion *vfork;
711
712         task_lock(tsk);
713         vfork = tsk->vfork_done;
714         if (likely(vfork)) {
715                 tsk->vfork_done = NULL;
716                 complete(vfork);
717         }
718         task_unlock(tsk);
719 }
720
721 static int wait_for_vfork_done(struct task_struct *child,
722                                 struct completion *vfork)
723 {
724         int killed;
725
726         freezer_do_not_count();
727         killed = wait_for_completion_killable(vfork);
728         freezer_count();
729
730         if (killed) {
731                 task_lock(child);
732                 child->vfork_done = NULL;
733                 task_unlock(child);
734         }
735
736         put_task_struct(child);
737         return killed;
738 }
739
740 /* Please note the differences between mmput and mm_release.
741  * mmput is called whenever we stop holding onto a mm_struct,
742  * error success whatever.
743  *
744  * mm_release is called after a mm_struct has been removed
745  * from the current process.
746  *
747  * This difference is important for error handling, when we
748  * only half set up a mm_struct for a new process and need to restore
749  * the old one.  Because we mmput the new mm_struct before
750  * restoring the old one. . .
751  * Eric Biederman 10 January 1998
752  */
753 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
754 {
755         /* Get rid of any futexes when releasing the mm */
756 #ifdef CONFIG_FUTEX
757         if (unlikely(tsk->robust_list)) {
758                 exit_robust_list(tsk);
759                 tsk->robust_list = NULL;
760         }
761 #ifdef CONFIG_COMPAT
762         if (unlikely(tsk->compat_robust_list)) {
763                 compat_exit_robust_list(tsk);
764                 tsk->compat_robust_list = NULL;
765         }
766 #endif
767         if (unlikely(!list_empty(&tsk->pi_state_list)))
768                 exit_pi_state_list(tsk);
769 #endif
770
771         uprobe_free_utask(tsk);
772
773         /* Get rid of any cached register state */
774         deactivate_mm(tsk, mm);
775
776         /*
777          * If we're exiting normally, clear a user-space tid field if
778          * requested.  We leave this alone when dying by signal, to leave
779          * the value intact in a core dump, and to save the unnecessary
780          * trouble, say, a killed vfork parent shouldn't touch this mm.
781          * Userland only wants this done for a sys_exit.
782          */
783         if (tsk->clear_child_tid) {
784                 if (!(tsk->flags & PF_SIGNALED) &&
785                     atomic_read(&mm->mm_users) > 1) {
786                         /*
787                          * We don't check the error code - if userspace has
788                          * not set up a proper pointer then tough luck.
789                          */
790                         put_user(0, tsk->clear_child_tid);
791                         sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
792                                         1, NULL, NULL, 0);
793                 }
794                 tsk->clear_child_tid = NULL;
795         }
796
797         /*
798          * All done, finally we can wake up parent and return this mm to him.
799          * Also kthread_stop() uses this completion for synchronization.
800          */
801         if (tsk->vfork_done)
802                 complete_vfork_done(tsk);
803 }
804
805 /*
806  * Allocate a new mm structure and copy contents from the
807  * mm structure of the passed in task structure.
808  */
809 struct mm_struct *dup_mm(struct task_struct *tsk)
810 {
811         struct mm_struct *mm, *oldmm = current->mm;
812         int err;
813
814         if (!oldmm)
815                 return NULL;
816
817         mm = allocate_mm();
818         if (!mm)
819                 goto fail_nomem;
820
821         memcpy(mm, oldmm, sizeof(*mm));
822         mm_init_cpumask(mm);
823
824 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
825         mm->pmd_huge_pte = NULL;
826 #endif
827 #ifdef CONFIG_NUMA_BALANCING
828         mm->first_nid = NUMA_PTE_SCAN_INIT;
829 #endif
830         if (!mm_init(mm, tsk))
831                 goto fail_nomem;
832
833         if (init_new_context(tsk, mm))
834                 goto fail_nocontext;
835
836         dup_mm_exe_file(oldmm, mm);
837
838         err = dup_mmap(mm, oldmm);
839         if (err)
840                 goto free_pt;
841
842         mm->hiwater_rss = get_mm_rss(mm);
843         mm->hiwater_vm = mm->total_vm;
844
845         if (mm->binfmt && !try_module_get(mm->binfmt->module))
846                 goto free_pt;
847
848         return mm;
849
850 free_pt:
851         /* don't put binfmt in mmput, we haven't got module yet */
852         mm->binfmt = NULL;
853         mmput(mm);
854
855 fail_nomem:
856         return NULL;
857
858 fail_nocontext:
859         /*
860          * If init_new_context() failed, we cannot use mmput() to free the mm
861          * because it calls destroy_context()
862          */
863         mm_free_pgd(mm);
864         free_mm(mm);
865         return NULL;
866 }
867
868 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
869 {
870         struct mm_struct *mm, *oldmm;
871         int retval;
872
873         tsk->min_flt = tsk->maj_flt = 0;
874         tsk->nvcsw = tsk->nivcsw = 0;
875 #ifdef CONFIG_DETECT_HUNG_TASK
876         tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
877 #endif
878
879         tsk->mm = NULL;
880         tsk->active_mm = NULL;
881
882         /*
883          * Are we cloning a kernel thread?
884          *
885          * We need to steal a active VM for that..
886          */
887         oldmm = current->mm;
888         if (!oldmm)
889                 return 0;
890
891         if (clone_flags & CLONE_VM) {
892                 atomic_inc(&oldmm->mm_users);
893                 mm = oldmm;
894                 goto good_mm;
895         }
896
897         retval = -ENOMEM;
898         mm = dup_mm(tsk);
899         if (!mm)
900                 goto fail_nomem;
901
902 good_mm:
903         tsk->mm = mm;
904         tsk->active_mm = mm;
905         return 0;
906
907 fail_nomem:
908         return retval;
909 }
910
911 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
912 {
913         struct fs_struct *fs = current->fs;
914         if (clone_flags & CLONE_FS) {
915                 /* tsk->fs is already what we want */
916                 spin_lock(&fs->lock);
917                 if (fs->in_exec) {
918                         spin_unlock(&fs->lock);
919                         return -EAGAIN;
920                 }
921                 fs->users++;
922                 spin_unlock(&fs->lock);
923                 return 0;
924         }
925         tsk->fs = copy_fs_struct(fs);
926         if (!tsk->fs)
927                 return -ENOMEM;
928         return 0;
929 }
930
931 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
932 {
933         struct files_struct *oldf, *newf;
934         int error = 0;
935
936         /*
937          * A background process may not have any files ...
938          */
939         oldf = current->files;
940         if (!oldf)
941                 goto out;
942
943         if (clone_flags & CLONE_FILES) {
944                 atomic_inc(&oldf->count);
945                 goto out;
946         }
947
948         newf = dup_fd(oldf, &error);
949         if (!newf)
950                 goto out;
951
952         tsk->files = newf;
953         error = 0;
954 out:
955         return error;
956 }
957
958 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
959 {
960 #ifdef CONFIG_BLOCK
961         struct io_context *ioc = current->io_context;
962         struct io_context *new_ioc;
963
964         if (!ioc)
965                 return 0;
966         /*
967          * Share io context with parent, if CLONE_IO is set
968          */
969         if (clone_flags & CLONE_IO) {
970                 ioc_task_link(ioc);
971                 tsk->io_context = ioc;
972         } else if (ioprio_valid(ioc->ioprio)) {
973                 new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
974                 if (unlikely(!new_ioc))
975                         return -ENOMEM;
976
977                 new_ioc->ioprio = ioc->ioprio;
978                 put_io_context(new_ioc);
979         }
980 #endif
981         return 0;
982 }
983
984 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
985 {
986         struct sighand_struct *sig;
987
988         if (clone_flags & CLONE_SIGHAND) {
989                 atomic_inc(&current->sighand->count);
990                 return 0;
991         }
992         sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
993         rcu_assign_pointer(tsk->sighand, sig);
994         if (!sig)
995                 return -ENOMEM;
996         atomic_set(&sig->count, 1);
997         memcpy(sig->action, current->sighand->action, sizeof(sig->action));
998         return 0;
999 }
1000
1001 void __cleanup_sighand(struct sighand_struct *sighand)
1002 {
1003         if (atomic_dec_and_test(&sighand->count)) {
1004                 signalfd_cleanup(sighand);
1005                 kmem_cache_free(sighand_cachep, sighand);
1006         }
1007 }
1008
1009
1010 /*
1011  * Initialize POSIX timer handling for a thread group.
1012  */
1013 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1014 {
1015         unsigned long cpu_limit;
1016
1017         /* Thread group counters. */
1018         thread_group_cputime_init(sig);
1019
1020         cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1021         if (cpu_limit != RLIM_INFINITY) {
1022                 sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1023                 sig->cputimer.running = 1;
1024         }
1025
1026         /* The timer lists. */
1027         INIT_LIST_HEAD(&sig->cpu_timers[0]);
1028         INIT_LIST_HEAD(&sig->cpu_timers[1]);
1029         INIT_LIST_HEAD(&sig->cpu_timers[2]);
1030 }
1031
1032 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1033 {
1034         struct signal_struct *sig;
1035
1036         if (clone_flags & CLONE_THREAD)
1037                 return 0;
1038
1039         sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1040         tsk->signal = sig;
1041         if (!sig)
1042                 return -ENOMEM;
1043
1044         sig->nr_threads = 1;
1045         atomic_set(&sig->live, 1);
1046         atomic_set(&sig->sigcnt, 1);
1047         init_waitqueue_head(&sig->wait_chldexit);
1048         sig->curr_target = tsk;
1049         init_sigpending(&sig->shared_pending);
1050         INIT_LIST_HEAD(&sig->posix_timers);
1051
1052         hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1053         sig->real_timer.function = it_real_fn;
1054
1055         task_lock(current->group_leader);
1056         memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1057         task_unlock(current->group_leader);
1058
1059         posix_cpu_timers_init_group(sig);
1060
1061         tty_audit_fork(sig);
1062         sched_autogroup_fork(sig);
1063
1064 #ifdef CONFIG_CGROUPS
1065         init_rwsem(&sig->group_rwsem);
1066 #endif
1067
1068         sig->oom_score_adj = current->signal->oom_score_adj;
1069         sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1070
1071         sig->has_child_subreaper = current->signal->has_child_subreaper ||
1072                                    current->signal->is_child_subreaper;
1073
1074         mutex_init(&sig->cred_guard_mutex);
1075
1076         return 0;
1077 }
1078
1079 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
1080 {
1081         unsigned long new_flags = p->flags;
1082
1083         new_flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1084         new_flags |= PF_FORKNOEXEC;
1085         p->flags = new_flags;
1086 }
1087
1088 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1089 {
1090         current->clear_child_tid = tidptr;
1091
1092         return task_pid_vnr(current);
1093 }
1094
1095 static void rt_mutex_init_task(struct task_struct *p)
1096 {
1097         raw_spin_lock_init(&p->pi_lock);
1098 #ifdef CONFIG_RT_MUTEXES
1099         plist_head_init(&p->pi_waiters);
1100         p->pi_blocked_on = NULL;
1101 #endif
1102 }
1103
1104 #ifdef CONFIG_MM_OWNER
1105 void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1106 {
1107         mm->owner = p;
1108 }
1109 #endif /* CONFIG_MM_OWNER */
1110
1111 /*
1112  * Initialize POSIX timer handling for a single task.
1113  */
1114 static void posix_cpu_timers_init(struct task_struct *tsk)
1115 {
1116         tsk->cputime_expires.prof_exp = 0;
1117         tsk->cputime_expires.virt_exp = 0;
1118         tsk->cputime_expires.sched_exp = 0;
1119         INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1120         INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1121         INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1122 }
1123
1124 /*
1125  * This creates a new process as a copy of the old one,
1126  * but does not actually start it yet.
1127  *
1128  * It copies the registers, and all the appropriate
1129  * parts of the process environment (as per the clone
1130  * flags). The actual kick-off is left to the caller.
1131  */
1132 static struct task_struct *copy_process(unsigned long clone_flags,
1133                                         unsigned long stack_start,
1134                                         unsigned long stack_size,
1135                                         int __user *child_tidptr,
1136                                         struct pid *pid,
1137                                         int trace)
1138 {
1139         int retval;
1140         struct task_struct *p;
1141
1142         if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1143                 return ERR_PTR(-EINVAL);
1144
1145         if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1146                 return ERR_PTR(-EINVAL);
1147
1148         /*
1149          * Thread groups must share signals as well, and detached threads
1150          * can only be started up within the thread group.
1151          */
1152         if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1153                 return ERR_PTR(-EINVAL);
1154
1155         /*
1156          * Shared signal handlers imply shared VM. By way of the above,
1157          * thread groups also imply shared VM. Blocking this case allows
1158          * for various simplifications in other code.
1159          */
1160         if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1161                 return ERR_PTR(-EINVAL);
1162
1163         /*
1164          * Siblings of global init remain as zombies on exit since they are
1165          * not reaped by their parent (swapper). To solve this and to avoid
1166          * multi-rooted process trees, prevent global and container-inits
1167          * from creating siblings.
1168          */
1169         if ((clone_flags & CLONE_PARENT) &&
1170                                 current->signal->flags & SIGNAL_UNKILLABLE)
1171                 return ERR_PTR(-EINVAL);
1172
1173         /*
1174          * If the new process will be in a different pid namespace
1175          * don't allow the creation of threads.
1176          */
1177         if ((clone_flags & (CLONE_VM|CLONE_NEWPID)) &&
1178             (task_active_pid_ns(current) != current->nsproxy->pid_ns))
1179                 return ERR_PTR(-EINVAL);
1180
1181         retval = security_task_create(clone_flags);
1182         if (retval)
1183                 goto fork_out;
1184
1185         retval = -ENOMEM;
1186         p = dup_task_struct(current);
1187         if (!p)
1188                 goto fork_out;
1189
1190         ftrace_graph_init_task(p);
1191         get_seccomp_filter(p);
1192
1193         rt_mutex_init_task(p);
1194
1195 #ifdef CONFIG_PROVE_LOCKING
1196         DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1197         DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1198 #endif
1199         retval = -EAGAIN;
1200         if (atomic_read(&p->real_cred->user->processes) >=
1201                         task_rlimit(p, RLIMIT_NPROC)) {
1202                 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_RESOURCE) &&
1203                     p->real_cred->user != INIT_USER)
1204                         goto bad_fork_free;
1205         }
1206         current->flags &= ~PF_NPROC_EXCEEDED;
1207
1208         retval = copy_creds(p, clone_flags);
1209         if (retval < 0)
1210                 goto bad_fork_free;
1211
1212         /*
1213          * If multiple threads are within copy_process(), then this check
1214          * triggers too late. This doesn't hurt, the check is only there
1215          * to stop root fork bombs.
1216          */
1217         retval = -EAGAIN;
1218         if (nr_threads >= max_threads)
1219                 goto bad_fork_cleanup_count;
1220
1221         if (!try_module_get(task_thread_info(p)->exec_domain->module))
1222                 goto bad_fork_cleanup_count;
1223
1224         p->did_exec = 0;
1225         delayacct_tsk_init(p);  /* Must remain after dup_task_struct() */
1226         copy_flags(clone_flags, p);
1227         INIT_LIST_HEAD(&p->children);
1228         INIT_LIST_HEAD(&p->sibling);
1229         rcu_copy_process(p);
1230         p->vfork_done = NULL;
1231         spin_lock_init(&p->alloc_lock);
1232
1233         init_sigpending(&p->pending);
1234
1235         p->utime = p->stime = p->gtime = 0;
1236         p->utimescaled = p->stimescaled = 0;
1237 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1238         p->prev_cputime.utime = p->prev_cputime.stime = 0;
1239 #endif
1240 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1241         seqlock_init(&p->vtime_seqlock);
1242         p->vtime_snap = 0;
1243         p->vtime_snap_whence = VTIME_SLEEPING;
1244 #endif
1245
1246 #if defined(SPLIT_RSS_COUNTING)
1247         memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1248 #endif
1249
1250         p->default_timer_slack_ns = current->timer_slack_ns;
1251
1252         task_io_accounting_init(&p->ioac);
1253         acct_clear_integrals(p);
1254
1255         posix_cpu_timers_init(p);
1256
1257         do_posix_clock_monotonic_gettime(&p->start_time);
1258         p->real_start_time = p->start_time;
1259         monotonic_to_bootbased(&p->real_start_time);
1260         p->io_context = NULL;
1261         p->audit_context = NULL;
1262         if (clone_flags & CLONE_THREAD)
1263                 threadgroup_change_begin(current);
1264         cgroup_fork(p);
1265 #ifdef CONFIG_NUMA
1266         p->mempolicy = mpol_dup(p->mempolicy);
1267         if (IS_ERR(p->mempolicy)) {
1268                 retval = PTR_ERR(p->mempolicy);
1269                 p->mempolicy = NULL;
1270                 goto bad_fork_cleanup_cgroup;
1271         }
1272         mpol_fix_fork_child_flag(p);
1273 #endif
1274 #ifdef CONFIG_CPUSETS
1275         p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1276         p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1277         seqcount_init(&p->mems_allowed_seq);
1278 #endif
1279 #ifdef CONFIG_TRACE_IRQFLAGS
1280         p->irq_events = 0;
1281         p->hardirqs_enabled = 0;
1282         p->hardirq_enable_ip = 0;
1283         p->hardirq_enable_event = 0;
1284         p->hardirq_disable_ip = _THIS_IP_;
1285         p->hardirq_disable_event = 0;
1286         p->softirqs_enabled = 1;
1287         p->softirq_enable_ip = _THIS_IP_;
1288         p->softirq_enable_event = 0;
1289         p->softirq_disable_ip = 0;
1290         p->softirq_disable_event = 0;
1291         p->hardirq_context = 0;
1292         p->softirq_context = 0;
1293 #endif
1294 #ifdef CONFIG_LOCKDEP
1295         p->lockdep_depth = 0; /* no locks held yet */
1296         p->curr_chain_key = 0;
1297         p->lockdep_recursion = 0;
1298 #endif
1299
1300 #ifdef CONFIG_DEBUG_MUTEXES
1301         p->blocked_on = NULL; /* not blocked yet */
1302 #endif
1303 #ifdef CONFIG_MEMCG
1304         p->memcg_batch.do_batch = 0;
1305         p->memcg_batch.memcg = NULL;
1306 #endif
1307 #ifdef CONFIG_BCACHE
1308         p->sequential_io        = 0;
1309         p->sequential_io_avg    = 0;
1310 #endif
1311
1312         /* Perform scheduler related setup. Assign this task to a CPU. */
1313         sched_fork(p);
1314
1315         retval = perf_event_init_task(p);
1316         if (retval)
1317                 goto bad_fork_cleanup_policy;
1318         retval = audit_alloc(p);
1319         if (retval)
1320                 goto bad_fork_cleanup_policy;
1321         /* copy all the process information */
1322         retval = copy_semundo(clone_flags, p);
1323         if (retval)
1324                 goto bad_fork_cleanup_audit;
1325         retval = copy_files(clone_flags, p);
1326         if (retval)
1327                 goto bad_fork_cleanup_semundo;
1328         retval = copy_fs(clone_flags, p);
1329         if (retval)
1330                 goto bad_fork_cleanup_files;
1331         retval = copy_sighand(clone_flags, p);
1332         if (retval)
1333                 goto bad_fork_cleanup_fs;
1334         retval = copy_signal(clone_flags, p);
1335         if (retval)
1336                 goto bad_fork_cleanup_sighand;
1337         retval = copy_mm(clone_flags, p);
1338         if (retval)
1339                 goto bad_fork_cleanup_signal;
1340         retval = copy_namespaces(clone_flags, p);
1341         if (retval)
1342                 goto bad_fork_cleanup_mm;
1343         retval = copy_io(clone_flags, p);
1344         if (retval)
1345                 goto bad_fork_cleanup_namespaces;
1346         retval = copy_thread(clone_flags, stack_start, stack_size, p);
1347         if (retval)
1348                 goto bad_fork_cleanup_io;
1349
1350         if (pid != &init_struct_pid) {
1351                 retval = -ENOMEM;
1352                 pid = alloc_pid(p->nsproxy->pid_ns);
1353                 if (!pid)
1354                         goto bad_fork_cleanup_io;
1355         }
1356
1357         p->pid = pid_nr(pid);
1358         p->tgid = p->pid;
1359         if (clone_flags & CLONE_THREAD)
1360                 p->tgid = current->tgid;
1361
1362         p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1363         /*
1364          * Clear TID on mm_release()?
1365          */
1366         p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1367 #ifdef CONFIG_BLOCK
1368         p->plug = NULL;
1369 #endif
1370 #ifdef CONFIG_FUTEX
1371         p->robust_list = NULL;
1372 #ifdef CONFIG_COMPAT
1373         p->compat_robust_list = NULL;
1374 #endif
1375         INIT_LIST_HEAD(&p->pi_state_list);
1376         p->pi_state_cache = NULL;
1377 #endif
1378         uprobe_copy_process(p);
1379         /*
1380          * sigaltstack should be cleared when sharing the same VM
1381          */
1382         if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1383                 p->sas_ss_sp = p->sas_ss_size = 0;
1384
1385         /*
1386          * Syscall tracing and stepping should be turned off in the
1387          * child regardless of CLONE_PTRACE.
1388          */
1389         user_disable_single_step(p);
1390         clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1391 #ifdef TIF_SYSCALL_EMU
1392         clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1393 #endif
1394         clear_all_latency_tracing(p);
1395
1396         /* ok, now we should be set up.. */
1397         if (clone_flags & CLONE_THREAD)
1398                 p->exit_signal = -1;
1399         else if (clone_flags & CLONE_PARENT)
1400                 p->exit_signal = current->group_leader->exit_signal;
1401         else
1402                 p->exit_signal = (clone_flags & CSIGNAL);
1403
1404         p->pdeath_signal = 0;
1405         p->exit_state = 0;
1406
1407         p->nr_dirtied = 0;
1408         p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1409         p->dirty_paused_when = 0;
1410
1411         /*
1412          * Ok, make it visible to the rest of the system.
1413          * We dont wake it up yet.
1414          */
1415         p->group_leader = p;
1416         INIT_LIST_HEAD(&p->thread_group);
1417         p->task_works = NULL;
1418
1419         /* Need tasklist lock for parent etc handling! */
1420         write_lock_irq(&tasklist_lock);
1421
1422         /* CLONE_PARENT re-uses the old parent */
1423         if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1424                 p->real_parent = current->real_parent;
1425                 p->parent_exec_id = current->parent_exec_id;
1426         } else {
1427                 p->real_parent = current;
1428                 p->parent_exec_id = current->self_exec_id;
1429         }
1430
1431         spin_lock(&current->sighand->siglock);
1432
1433         /*
1434          * Process group and session signals need to be delivered to just the
1435          * parent before the fork or both the parent and the child after the
1436          * fork. Restart if a signal comes in before we add the new process to
1437          * it's process group.
1438          * A fatal signal pending means that current will exit, so the new
1439          * thread can't slip out of an OOM kill (or normal SIGKILL).
1440         */
1441         recalc_sigpending();
1442         if (signal_pending(current)) {
1443                 spin_unlock(&current->sighand->siglock);
1444                 write_unlock_irq(&tasklist_lock);
1445                 retval = -ERESTARTNOINTR;
1446                 goto bad_fork_free_pid;
1447         }
1448
1449         if (clone_flags & CLONE_THREAD) {
1450                 current->signal->nr_threads++;
1451                 atomic_inc(&current->signal->live);
1452                 atomic_inc(&current->signal->sigcnt);
1453                 p->group_leader = current->group_leader;
1454                 list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
1455         }
1456
1457         if (likely(p->pid)) {
1458                 ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1459
1460                 if (thread_group_leader(p)) {
1461                         if (is_child_reaper(pid)) {
1462                                 ns_of_pid(pid)->child_reaper = p;
1463                                 p->signal->flags |= SIGNAL_UNKILLABLE;
1464                         }
1465
1466                         p->signal->leader_pid = pid;
1467                         p->signal->tty = tty_kref_get(current->signal->tty);
1468                         attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1469                         attach_pid(p, PIDTYPE_SID, task_session(current));
1470                         list_add_tail(&p->sibling, &p->real_parent->children);
1471                         list_add_tail_rcu(&p->tasks, &init_task.tasks);
1472                         __this_cpu_inc(process_counts);
1473                 }
1474                 attach_pid(p, PIDTYPE_PID, pid);
1475                 nr_threads++;
1476         }
1477
1478         total_forks++;
1479         spin_unlock(&current->sighand->siglock);
1480         write_unlock_irq(&tasklist_lock);
1481         proc_fork_connector(p);
1482         cgroup_post_fork(p);
1483         if (clone_flags & CLONE_THREAD)
1484                 threadgroup_change_end(current);
1485         perf_event_fork(p);
1486
1487         trace_task_newtask(p, clone_flags);
1488
1489         return p;
1490
1491 bad_fork_free_pid:
1492         if (pid != &init_struct_pid)
1493                 free_pid(pid);
1494 bad_fork_cleanup_io:
1495         if (p->io_context)
1496                 exit_io_context(p);
1497 bad_fork_cleanup_namespaces:
1498         exit_task_namespaces(p);
1499 bad_fork_cleanup_mm:
1500         if (p->mm)
1501                 mmput(p->mm);
1502 bad_fork_cleanup_signal:
1503         if (!(clone_flags & CLONE_THREAD))
1504                 free_signal_struct(p->signal);
1505 bad_fork_cleanup_sighand:
1506         __cleanup_sighand(p->sighand);
1507 bad_fork_cleanup_fs:
1508         exit_fs(p); /* blocking */
1509 bad_fork_cleanup_files:
1510         exit_files(p); /* blocking */
1511 bad_fork_cleanup_semundo:
1512         exit_sem(p);
1513 bad_fork_cleanup_audit:
1514         audit_free(p);
1515 bad_fork_cleanup_policy:
1516         perf_event_free_task(p);
1517 #ifdef CONFIG_NUMA
1518         mpol_put(p->mempolicy);
1519 bad_fork_cleanup_cgroup:
1520 #endif
1521         if (clone_flags & CLONE_THREAD)
1522                 threadgroup_change_end(current);
1523         cgroup_exit(p, 0);
1524         delayacct_tsk_free(p);
1525         module_put(task_thread_info(p)->exec_domain->module);
1526 bad_fork_cleanup_count:
1527         atomic_dec(&p->cred->user->processes);
1528         exit_creds(p);
1529 bad_fork_free:
1530         free_task(p);
1531 fork_out:
1532         return ERR_PTR(retval);
1533 }
1534
1535 static inline void init_idle_pids(struct pid_link *links)
1536 {
1537         enum pid_type type;
1538
1539         for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1540                 INIT_HLIST_NODE(&links[type].node); /* not really needed */
1541                 links[type].pid = &init_struct_pid;
1542         }
1543 }
1544
1545 struct task_struct * __cpuinit fork_idle(int cpu)
1546 {
1547         struct task_struct *task;
1548         task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1549         if (!IS_ERR(task)) {
1550                 init_idle_pids(task->pids);
1551                 init_idle(task, cpu);
1552         }
1553
1554         return task;
1555 }
1556
1557 /*
1558  *  Ok, this is the main fork-routine.
1559  *
1560  * It copies the process, and if successful kick-starts
1561  * it and waits for it to finish using the VM if required.
1562  */
1563 long do_fork(unsigned long clone_flags,
1564               unsigned long stack_start,
1565               unsigned long stack_size,
1566               int __user *parent_tidptr,
1567               int __user *child_tidptr)
1568 {
1569         struct task_struct *p;
1570         int trace = 0;
1571         long nr;
1572
1573         /*
1574          * Do some preliminary argument and permissions checking before we
1575          * actually start allocating stuff
1576          */
1577         if (clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) {
1578                 if (clone_flags & (CLONE_THREAD|CLONE_PARENT))
1579                         return -EINVAL;
1580         }
1581
1582         /*
1583          * Determine whether and which event to report to ptracer.  When
1584          * called from kernel_thread or CLONE_UNTRACED is explicitly
1585          * requested, no event is reported; otherwise, report if the event
1586          * for the type of forking is enabled.
1587          */
1588         if (!(clone_flags & CLONE_UNTRACED)) {
1589                 if (clone_flags & CLONE_VFORK)
1590                         trace = PTRACE_EVENT_VFORK;
1591                 else if ((clone_flags & CSIGNAL) != SIGCHLD)
1592                         trace = PTRACE_EVENT_CLONE;
1593                 else
1594                         trace = PTRACE_EVENT_FORK;
1595
1596                 if (likely(!ptrace_event_enabled(current, trace)))
1597                         trace = 0;
1598         }
1599
1600         p = copy_process(clone_flags, stack_start, stack_size,
1601                          child_tidptr, NULL, trace);
1602         /*
1603          * Do this prior waking up the new thread - the thread pointer
1604          * might get invalid after that point, if the thread exits quickly.
1605          */
1606         if (!IS_ERR(p)) {
1607                 struct completion vfork;
1608
1609                 trace_sched_process_fork(current, p);
1610
1611                 nr = task_pid_vnr(p);
1612
1613                 if (clone_flags & CLONE_PARENT_SETTID)
1614                         put_user(nr, parent_tidptr);
1615
1616                 if (clone_flags & CLONE_VFORK) {
1617                         p->vfork_done = &vfork;
1618                         init_completion(&vfork);
1619                         get_task_struct(p);
1620                 }
1621
1622                 wake_up_new_task(p);
1623
1624                 /* forking complete and child started to run, tell ptracer */
1625                 if (unlikely(trace))
1626                         ptrace_event(trace, nr);
1627
1628                 if (clone_flags & CLONE_VFORK) {
1629                         if (!wait_for_vfork_done(p, &vfork))
1630                                 ptrace_event(PTRACE_EVENT_VFORK_DONE, nr);
1631                 }
1632         } else {
1633                 nr = PTR_ERR(p);
1634         }
1635         return nr;
1636 }
1637
1638 /*
1639  * Create a kernel thread.
1640  */
1641 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1642 {
1643         return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1644                 (unsigned long)arg, NULL, NULL);
1645 }
1646
1647 #ifdef __ARCH_WANT_SYS_FORK
1648 SYSCALL_DEFINE0(fork)
1649 {
1650 #ifdef CONFIG_MMU
1651         return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1652 #else
1653         /* can not support in nommu mode */
1654         return(-EINVAL);
1655 #endif
1656 }
1657 #endif
1658
1659 #ifdef __ARCH_WANT_SYS_VFORK
1660 SYSCALL_DEFINE0(vfork)
1661 {
1662         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0, 
1663                         0, NULL, NULL);
1664 }
1665 #endif
1666
1667 #ifdef __ARCH_WANT_SYS_CLONE
1668 #ifdef CONFIG_CLONE_BACKWARDS
1669 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1670                  int __user *, parent_tidptr,
1671                  int, tls_val,
1672                  int __user *, child_tidptr)
1673 #elif defined(CONFIG_CLONE_BACKWARDS2)
1674 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1675                  int __user *, parent_tidptr,
1676                  int __user *, child_tidptr,
1677                  int, tls_val)
1678 #else
1679 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1680                  int __user *, parent_tidptr,
1681                  int __user *, child_tidptr,
1682                  int, tls_val)
1683 #endif
1684 {
1685         return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1686 }
1687 #endif
1688
1689 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1690 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1691 #endif
1692
1693 static void sighand_ctor(void *data)
1694 {
1695         struct sighand_struct *sighand = data;
1696
1697         spin_lock_init(&sighand->siglock);
1698         init_waitqueue_head(&sighand->signalfd_wqh);
1699 }
1700
1701 void __init proc_caches_init(void)
1702 {
1703         sighand_cachep = kmem_cache_create("sighand_cache",
1704                         sizeof(struct sighand_struct), 0,
1705                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1706                         SLAB_NOTRACK, sighand_ctor);
1707         signal_cachep = kmem_cache_create("signal_cache",
1708                         sizeof(struct signal_struct), 0,
1709                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1710         files_cachep = kmem_cache_create("files_cache",
1711                         sizeof(struct files_struct), 0,
1712                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1713         fs_cachep = kmem_cache_create("fs_cache",
1714                         sizeof(struct fs_struct), 0,
1715                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1716         /*
1717          * FIXME! The "sizeof(struct mm_struct)" currently includes the
1718          * whole struct cpumask for the OFFSTACK case. We could change
1719          * this to *only* allocate as much of it as required by the
1720          * maximum number of CPU's we can ever have.  The cpumask_allocation
1721          * is at the end of the structure, exactly for that reason.
1722          */
1723         mm_cachep = kmem_cache_create("mm_struct",
1724                         sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1725                         SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1726         vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1727         mmap_init();
1728         nsproxy_cache_init();
1729 }
1730
1731 /*
1732  * Check constraints on flags passed to the unshare system call.
1733  */
1734 static int check_unshare_flags(unsigned long unshare_flags)
1735 {
1736         if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1737                                 CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1738                                 CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1739                                 CLONE_NEWUSER|CLONE_NEWPID))
1740                 return -EINVAL;
1741         /*
1742          * Not implemented, but pretend it works if there is nothing to
1743          * unshare. Note that unsharing CLONE_THREAD or CLONE_SIGHAND
1744          * needs to unshare vm.
1745          */
1746         if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1747                 /* FIXME: get_task_mm() increments ->mm_users */
1748                 if (atomic_read(&current->mm->mm_users) > 1)
1749                         return -EINVAL;
1750         }
1751
1752         return 0;
1753 }
1754
1755 /*
1756  * Unshare the filesystem structure if it is being shared
1757  */
1758 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1759 {
1760         struct fs_struct *fs = current->fs;
1761
1762         if (!(unshare_flags & CLONE_FS) || !fs)
1763                 return 0;
1764
1765         /* don't need lock here; in the worst case we'll do useless copy */
1766         if (fs->users == 1)
1767                 return 0;
1768
1769         *new_fsp = copy_fs_struct(fs);
1770         if (!*new_fsp)
1771                 return -ENOMEM;
1772
1773         return 0;
1774 }
1775
1776 /*
1777  * Unshare file descriptor table if it is being shared
1778  */
1779 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1780 {
1781         struct files_struct *fd = current->files;
1782         int error = 0;
1783
1784         if ((unshare_flags & CLONE_FILES) &&
1785             (fd && atomic_read(&fd->count) > 1)) {
1786                 *new_fdp = dup_fd(fd, &error);
1787                 if (!*new_fdp)
1788                         return error;
1789         }
1790
1791         return 0;
1792 }
1793
1794 /*
1795  * unshare allows a process to 'unshare' part of the process
1796  * context which was originally shared using clone.  copy_*
1797  * functions used by do_fork() cannot be used here directly
1798  * because they modify an inactive task_struct that is being
1799  * constructed. Here we are modifying the current, active,
1800  * task_struct.
1801  */
1802 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1803 {
1804         struct fs_struct *fs, *new_fs = NULL;
1805         struct files_struct *fd, *new_fd = NULL;
1806         struct cred *new_cred = NULL;
1807         struct nsproxy *new_nsproxy = NULL;
1808         int do_sysvsem = 0;
1809         int err;
1810
1811         /*
1812          * If unsharing a user namespace must also unshare the thread.
1813          */
1814         if (unshare_flags & CLONE_NEWUSER)
1815                 unshare_flags |= CLONE_THREAD | CLONE_FS;
1816         /*
1817          * If unsharing a pid namespace must also unshare the thread.
1818          */
1819         if (unshare_flags & CLONE_NEWPID)
1820                 unshare_flags |= CLONE_THREAD;
1821         /*
1822          * If unsharing a thread from a thread group, must also unshare vm.
1823          */
1824         if (unshare_flags & CLONE_THREAD)
1825                 unshare_flags |= CLONE_VM;
1826         /*
1827          * If unsharing vm, must also unshare signal handlers.
1828          */
1829         if (unshare_flags & CLONE_VM)
1830                 unshare_flags |= CLONE_SIGHAND;
1831         /*
1832          * If unsharing namespace, must also unshare filesystem information.
1833          */
1834         if (unshare_flags & CLONE_NEWNS)
1835                 unshare_flags |= CLONE_FS;
1836
1837         err = check_unshare_flags(unshare_flags);
1838         if (err)
1839                 goto bad_unshare_out;
1840         /*
1841          * CLONE_NEWIPC must also detach from the undolist: after switching
1842          * to a new ipc namespace, the semaphore arrays from the old
1843          * namespace are unreachable.
1844          */
1845         if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1846                 do_sysvsem = 1;
1847         err = unshare_fs(unshare_flags, &new_fs);
1848         if (err)
1849                 goto bad_unshare_out;
1850         err = unshare_fd(unshare_flags, &new_fd);
1851         if (err)
1852                 goto bad_unshare_cleanup_fs;
1853         err = unshare_userns(unshare_flags, &new_cred);
1854         if (err)
1855                 goto bad_unshare_cleanup_fd;
1856         err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1857                                          new_cred, new_fs);
1858         if (err)
1859                 goto bad_unshare_cleanup_cred;
1860
1861         if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1862                 if (do_sysvsem) {
1863                         /*
1864                          * CLONE_SYSVSEM is equivalent to sys_exit().
1865                          */
1866                         exit_sem(current);
1867                 }
1868
1869                 if (new_nsproxy)
1870                         switch_task_namespaces(current, new_nsproxy);
1871
1872                 task_lock(current);
1873
1874                 if (new_fs) {
1875                         fs = current->fs;
1876                         spin_lock(&fs->lock);
1877                         current->fs = new_fs;
1878                         if (--fs->users)
1879                                 new_fs = NULL;
1880                         else
1881                                 new_fs = fs;
1882                         spin_unlock(&fs->lock);
1883                 }
1884
1885                 if (new_fd) {
1886                         fd = current->files;
1887                         current->files = new_fd;
1888                         new_fd = fd;
1889                 }
1890
1891                 task_unlock(current);
1892
1893                 if (new_cred) {
1894                         /* Install the new user namespace */
1895                         commit_creds(new_cred);
1896                         new_cred = NULL;
1897                 }
1898         }
1899
1900 bad_unshare_cleanup_cred:
1901         if (new_cred)
1902                 put_cred(new_cred);
1903 bad_unshare_cleanup_fd:
1904         if (new_fd)
1905                 put_files_struct(new_fd);
1906
1907 bad_unshare_cleanup_fs:
1908         if (new_fs)
1909                 free_fs_struct(new_fs);
1910
1911 bad_unshare_out:
1912         return err;
1913 }
1914
1915 /*
1916  *      Helper to unshare the files of the current task.
1917  *      We don't want to expose copy_files internals to
1918  *      the exec layer of the kernel.
1919  */
1920
1921 int unshare_files(struct files_struct **displaced)
1922 {
1923         struct task_struct *task = current;
1924         struct files_struct *copy = NULL;
1925         int error;
1926
1927         error = unshare_fd(CLONE_FILES, &copy);
1928         if (error || !copy) {
1929                 *displaced = NULL;
1930                 return error;
1931         }
1932         *displaced = task->files;
1933         task_lock(task);
1934         task->files = copy;
1935         task_unlock(task);
1936         return 0;
1937 }