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/mm.h>
32#include <linux/vmacache.h>
33#include <linux/nsproxy.h>
34#include <linux/capability.h>
35#include <linux/cpu.h>
36#include <linux/cgroup.h>
37#include <linux/security.h>
38#include <linux/hugetlb.h>
39#include <linux/seccomp.h>
40#include <linux/swap.h>
41#include <linux/syscalls.h>
42#include <linux/jiffies.h>
43#include <linux/futex.h>
44#include <linux/compat.h>
45#include <linux/kthread.h>
46#include <linux/task_io_accounting_ops.h>
47#include <linux/rcupdate.h>
48#include <linux/ptrace.h>
49#include <linux/mount.h>
50#include <linux/audit.h>
51#include <linux/memcontrol.h>
52#include <linux/ftrace.h>
53#include <linux/proc_fs.h>
54#include <linux/profile.h>
55#include <linux/rmap.h>
56#include <linux/ksm.h>
57#include <linux/acct.h>
58#include <linux/tsacct_kern.h>
59#include <linux/cn_proc.h>
60#include <linux/freezer.h>
61#include <linux/delayacct.h>
62#include <linux/taskstats_kern.h>
63#include <linux/random.h>
64#include <linux/tty.h>
65#include <linux/blkdev.h>
66#include <linux/fs_struct.h>
67#include <linux/magic.h>
68#include <linux/perf_event.h>
69#include <linux/posix-timers.h>
70#include <linux/user-return-notifier.h>
71#include <linux/oom.h>
72#include <linux/khugepaged.h>
73#include <linux/signalfd.h>
74#include <linux/uprobes.h>
75#include <linux/aio.h>
76#include <linux/compiler.h>
77#include <linux/sysctl.h>
78
79#include <asm/pgtable.h>
80#include <asm/pgalloc.h>
81#include <asm/uaccess.h>
82#include <asm/mmu_context.h>
83#include <asm/cacheflush.h>
84#include <asm/tlbflush.h>
85
86#include <trace/events/sched.h>
87
88#define CREATE_TRACE_POINTS
89#include <trace/events/task.h>
90
91/*
92 * Minimum number of threads to boot the kernel
93 */
94#define MIN_THREADS 20
95
96/*
97 * Maximum number of threads
98 */
99#define MAX_THREADS FUTEX_TID_MASK
100
101/*
102 * Protected counters by write_lock_irq(&tasklist_lock)
103 */
104unsigned long total_forks;	/* Handle normal Linux uptimes. */
105int nr_threads;			/* The idle threads do not count.. */
106
107int max_threads;		/* tunable limit on nr_threads */
108
109DEFINE_PER_CPU(unsigned long, process_counts) = 0;
110
111__cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
112
113#ifdef CONFIG_PROVE_RCU
114int lockdep_tasklist_lock_is_held(void)
115{
116	return lockdep_is_held(&tasklist_lock);
117}
118EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
119#endif /* #ifdef CONFIG_PROVE_RCU */
120
121int nr_processes(void)
122{
123	int cpu;
124	int total = 0;
125
126	for_each_possible_cpu(cpu)
127		total += per_cpu(process_counts, cpu);
128
129	return total;
130}
131
132void __weak arch_release_task_struct(struct task_struct *tsk)
133{
134}
135
136#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
137static struct kmem_cache *task_struct_cachep;
138
139static inline struct task_struct *alloc_task_struct_node(int node)
140{
141	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
142}
143
144static inline void free_task_struct(struct task_struct *tsk)
145{
146	kmem_cache_free(task_struct_cachep, tsk);
147}
148#endif
149
150void __weak arch_release_thread_info(struct thread_info *ti)
151{
152}
153
154#ifndef CONFIG_ARCH_THREAD_INFO_ALLOCATOR
155
156/*
157 * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
158 * kmemcache based allocator.
159 */
160# if THREAD_SIZE >= PAGE_SIZE
161static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
162						  int node)
163{
164	struct page *page = alloc_kmem_pages_node(node, THREADINFO_GFP,
165						  THREAD_SIZE_ORDER);
166
167	return page ? page_address(page) : NULL;
168}
169
170static inline void free_thread_info(struct thread_info *ti)
171{
172	free_kmem_pages((unsigned long)ti, THREAD_SIZE_ORDER);
173}
174# else
175static struct kmem_cache *thread_info_cache;
176
177static struct thread_info *alloc_thread_info_node(struct task_struct *tsk,
178						  int node)
179{
180	return kmem_cache_alloc_node(thread_info_cache, THREADINFO_GFP, node);
181}
182
183static void free_thread_info(struct thread_info *ti)
184{
185	kmem_cache_free(thread_info_cache, ti);
186}
187
188void thread_info_cache_init(void)
189{
190	thread_info_cache = kmem_cache_create("thread_info", THREAD_SIZE,
191					      THREAD_SIZE, 0, NULL);
192	BUG_ON(thread_info_cache == NULL);
193}
194# endif
195#endif
196
197/* SLAB cache for signal_struct structures (tsk->signal) */
198static struct kmem_cache *signal_cachep;
199
200/* SLAB cache for sighand_struct structures (tsk->sighand) */
201struct kmem_cache *sighand_cachep;
202
203/* SLAB cache for files_struct structures (tsk->files) */
204struct kmem_cache *files_cachep;
205
206/* SLAB cache for fs_struct structures (tsk->fs) */
207struct kmem_cache *fs_cachep;
208
209/* SLAB cache for vm_area_struct structures */
210struct kmem_cache *vm_area_cachep;
211
212/* SLAB cache for mm_struct structures (tsk->mm) */
213static struct kmem_cache *mm_cachep;
214
215static void account_kernel_stack(struct thread_info *ti, int account)
216{
217	struct zone *zone = page_zone(virt_to_page(ti));
218
219	mod_zone_page_state(zone, NR_KERNEL_STACK, account);
220}
221
222void free_task(struct task_struct *tsk)
223{
224	account_kernel_stack(tsk->stack, -1);
225	arch_release_thread_info(tsk->stack);
226	free_thread_info(tsk->stack);
227	rt_mutex_debug_task_free(tsk);
228	ftrace_graph_exit_task(tsk);
229	put_seccomp_filter(tsk);
230	arch_release_task_struct(tsk);
231	free_task_struct(tsk);
232}
233EXPORT_SYMBOL(free_task);
234
235static inline void free_signal_struct(struct signal_struct *sig)
236{
237	taskstats_tgid_free(sig);
238	sched_autogroup_exit(sig);
239	kmem_cache_free(signal_cachep, sig);
240}
241
242static inline void put_signal_struct(struct signal_struct *sig)
243{
244	if (atomic_dec_and_test(&sig->sigcnt))
245		free_signal_struct(sig);
246}
247
248void __put_task_struct(struct task_struct *tsk)
249{
250	WARN_ON(!tsk->exit_state);
251	WARN_ON(atomic_read(&tsk->usage));
252	WARN_ON(tsk == current);
253
254	task_numa_free(tsk);
255	security_task_free(tsk);
256	exit_creds(tsk);
257	delayacct_tsk_free(tsk);
258	put_signal_struct(tsk->signal);
259
260	if (!profile_handoff_task(tsk))
261		free_task(tsk);
262}
263EXPORT_SYMBOL_GPL(__put_task_struct);
264
265void __init __weak arch_task_cache_init(void) { }
266
267/*
268 * set_max_threads
269 */
270static void set_max_threads(unsigned int max_threads_suggested)
271{
272	u64 threads;
273
274	/*
275	 * The number of threads shall be limited such that the thread
276	 * structures may only consume a small part of the available memory.
277	 */
278	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
279		threads = MAX_THREADS;
280	else
281		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
282				    (u64) THREAD_SIZE * 8UL);
283
284	if (threads > max_threads_suggested)
285		threads = max_threads_suggested;
286
287	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
288}
289
290void __init fork_init(void)
291{
292#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
293#ifndef ARCH_MIN_TASKALIGN
294#define ARCH_MIN_TASKALIGN	L1_CACHE_BYTES
295#endif
296	/* create a slab on which task_structs can be allocated */
297	task_struct_cachep =
298		kmem_cache_create("task_struct", sizeof(struct task_struct),
299			ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
300#endif
301
302	/* do the arch specific task caches init */
303	arch_task_cache_init();
304
305	set_max_threads(MAX_THREADS);
306
307	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
308	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
309	init_task.signal->rlim[RLIMIT_SIGPENDING] =
310		init_task.signal->rlim[RLIMIT_NPROC];
311}
312
313int __weak arch_dup_task_struct(struct task_struct *dst,
314					       struct task_struct *src)
315{
316	*dst = *src;
317	return 0;
318}
319
320void set_task_stack_end_magic(struct task_struct *tsk)
321{
322	unsigned long *stackend;
323
324	stackend = end_of_stack(tsk);
325	*stackend = STACK_END_MAGIC;	/* for overflow detection */
326}
327
328static struct task_struct *dup_task_struct(struct task_struct *orig)
329{
330	struct task_struct *tsk;
331	struct thread_info *ti;
332	int node = tsk_fork_get_node(orig);
333	int err;
334
335	tsk = alloc_task_struct_node(node);
336	if (!tsk)
337		return NULL;
338
339	ti = alloc_thread_info_node(tsk, node);
340	if (!ti)
341		goto free_tsk;
342
343	err = arch_dup_task_struct(tsk, orig);
344	if (err)
345		goto free_ti;
346
347	tsk->stack = ti;
348#ifdef CONFIG_SECCOMP
349	/*
350	 * We must handle setting up seccomp filters once we're under
351	 * the sighand lock in case orig has changed between now and
352	 * then. Until then, filter must be NULL to avoid messing up
353	 * the usage counts on the error path calling free_task.
354	 */
355	tsk->seccomp.filter = NULL;
356#endif
357
358	setup_thread_stack(tsk, orig);
359	clear_user_return_notifier(tsk);
360	clear_tsk_need_resched(tsk);
361	set_task_stack_end_magic(tsk);
362
363#ifdef CONFIG_CC_STACKPROTECTOR
364	tsk->stack_canary = get_random_int();
365#endif
366
367	/*
368	 * One for us, one for whoever does the "release_task()" (usually
369	 * parent)
370	 */
371	atomic_set(&tsk->usage, 2);
372#ifdef CONFIG_BLK_DEV_IO_TRACE
373	tsk->btrace_seq = 0;
374#endif
375	tsk->splice_pipe = NULL;
376	tsk->task_frag.page = NULL;
377
378	account_kernel_stack(ti, 1);
379
380	return tsk;
381
382free_ti:
383	free_thread_info(ti);
384free_tsk:
385	free_task_struct(tsk);
386	return NULL;
387}
388
389#ifdef CONFIG_MMU
390static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
391{
392	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
393	struct rb_node **rb_link, *rb_parent;
394	int retval;
395	unsigned long charge;
396
397	uprobe_start_dup_mmap();
398	down_write(&oldmm->mmap_sem);
399	flush_cache_dup_mm(oldmm);
400	uprobe_dup_mmap(oldmm, mm);
401	/*
402	 * Not linked in yet - no deadlock potential:
403	 */
404	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
405
406	/* No ordering required: file already has been exposed. */
407	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
408
409	mm->total_vm = oldmm->total_vm;
410	mm->shared_vm = oldmm->shared_vm;
411	mm->exec_vm = oldmm->exec_vm;
412	mm->stack_vm = oldmm->stack_vm;
413
414	rb_link = &mm->mm_rb.rb_node;
415	rb_parent = NULL;
416	pprev = &mm->mmap;
417	retval = ksm_fork(mm, oldmm);
418	if (retval)
419		goto out;
420	retval = khugepaged_fork(mm, oldmm);
421	if (retval)
422		goto out;
423
424	prev = NULL;
425	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
426		struct file *file;
427
428		if (mpnt->vm_flags & VM_DONTCOPY) {
429			vm_stat_account(mm, mpnt->vm_flags, mpnt->vm_file,
430							-vma_pages(mpnt));
431			continue;
432		}
433		charge = 0;
434		if (mpnt->vm_flags & VM_ACCOUNT) {
435			unsigned long len = vma_pages(mpnt);
436
437			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
438				goto fail_nomem;
439			charge = len;
440		}
441		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
442		if (!tmp)
443			goto fail_nomem;
444		*tmp = *mpnt;
445		INIT_LIST_HEAD(&tmp->anon_vma_chain);
446		retval = vma_dup_policy(mpnt, tmp);
447		if (retval)
448			goto fail_nomem_policy;
449		tmp->vm_mm = mm;
450		if (anon_vma_fork(tmp, mpnt))
451			goto fail_nomem_anon_vma_fork;
452		tmp->vm_flags &= ~VM_LOCKED;
453		tmp->vm_next = tmp->vm_prev = NULL;
454		file = tmp->vm_file;
455		if (file) {
456			struct inode *inode = file_inode(file);
457			struct address_space *mapping = file->f_mapping;
458
459			get_file(file);
460			if (tmp->vm_flags & VM_DENYWRITE)
461				atomic_dec(&inode->i_writecount);
462			i_mmap_lock_write(mapping);
463			if (tmp->vm_flags & VM_SHARED)
464				atomic_inc(&mapping->i_mmap_writable);
465			flush_dcache_mmap_lock(mapping);
466			/* insert tmp into the share list, just after mpnt */
467			vma_interval_tree_insert_after(tmp, mpnt,
468					&mapping->i_mmap);
469			flush_dcache_mmap_unlock(mapping);
470			i_mmap_unlock_write(mapping);
471		}
472
473		/*
474		 * Clear hugetlb-related page reserves for children. This only
475		 * affects MAP_PRIVATE mappings. Faults generated by the child
476		 * are not guaranteed to succeed, even if read-only
477		 */
478		if (is_vm_hugetlb_page(tmp))
479			reset_vma_resv_huge_pages(tmp);
480
481		/*
482		 * Link in the new vma and copy the page table entries.
483		 */
484		*pprev = tmp;
485		pprev = &tmp->vm_next;
486		tmp->vm_prev = prev;
487		prev = tmp;
488
489		__vma_link_rb(mm, tmp, rb_link, rb_parent);
490		rb_link = &tmp->vm_rb.rb_right;
491		rb_parent = &tmp->vm_rb;
492
493		mm->map_count++;
494		retval = copy_page_range(mm, oldmm, mpnt);
495
496		if (tmp->vm_ops && tmp->vm_ops->open)
497			tmp->vm_ops->open(tmp);
498
499		if (retval)
500			goto out;
501	}
502	/* a new mm has just been created */
503	arch_dup_mmap(oldmm, mm);
504	retval = 0;
505out:
506	up_write(&mm->mmap_sem);
507	flush_tlb_mm(oldmm);
508	up_write(&oldmm->mmap_sem);
509	uprobe_end_dup_mmap();
510	return retval;
511fail_nomem_anon_vma_fork:
512	mpol_put(vma_policy(tmp));
513fail_nomem_policy:
514	kmem_cache_free(vm_area_cachep, tmp);
515fail_nomem:
516	retval = -ENOMEM;
517	vm_unacct_memory(charge);
518	goto out;
519}
520
521static inline int mm_alloc_pgd(struct mm_struct *mm)
522{
523	mm->pgd = pgd_alloc(mm);
524	if (unlikely(!mm->pgd))
525		return -ENOMEM;
526	return 0;
527}
528
529static inline void mm_free_pgd(struct mm_struct *mm)
530{
531	pgd_free(mm, mm->pgd);
532}
533#else
534static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
535{
536	down_write(&oldmm->mmap_sem);
537	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
538	up_write(&oldmm->mmap_sem);
539	return 0;
540}
541#define mm_alloc_pgd(mm)	(0)
542#define mm_free_pgd(mm)
543#endif /* CONFIG_MMU */
544
545__cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
546
547#define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
548#define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
549
550static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
551
552static int __init coredump_filter_setup(char *s)
553{
554	default_dump_filter =
555		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
556		MMF_DUMP_FILTER_MASK;
557	return 1;
558}
559
560__setup("coredump_filter=", coredump_filter_setup);
561
562#include <linux/init_task.h>
563
564static void mm_init_aio(struct mm_struct *mm)
565{
566#ifdef CONFIG_AIO
567	spin_lock_init(&mm->ioctx_lock);
568	mm->ioctx_table = NULL;
569#endif
570}
571
572static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
573{
574#ifdef CONFIG_MEMCG
575	mm->owner = p;
576#endif
577}
578
579static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
580{
581	mm->mmap = NULL;
582	mm->mm_rb = RB_ROOT;
583	mm->vmacache_seqnum = 0;
584	atomic_set(&mm->mm_users, 1);
585	atomic_set(&mm->mm_count, 1);
586	init_rwsem(&mm->mmap_sem);
587	INIT_LIST_HEAD(&mm->mmlist);
588	mm->core_state = NULL;
589	atomic_long_set(&mm->nr_ptes, 0);
590	mm_nr_pmds_init(mm);
591	mm->map_count = 0;
592	mm->locked_vm = 0;
593	mm->pinned_vm = 0;
594	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
595	spin_lock_init(&mm->page_table_lock);
596	mm_init_cpumask(mm);
597	mm_init_aio(mm);
598	mm_init_owner(mm, p);
599	mmu_notifier_mm_init(mm);
600	clear_tlb_flush_pending(mm);
601#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
602	mm->pmd_huge_pte = NULL;
603#endif
604
605	if (current->mm) {
606		mm->flags = current->mm->flags & MMF_INIT_MASK;
607		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
608	} else {
609		mm->flags = default_dump_filter;
610		mm->def_flags = 0;
611	}
612
613	if (mm_alloc_pgd(mm))
614		goto fail_nopgd;
615
616	if (init_new_context(p, mm))
617		goto fail_nocontext;
618
619	return mm;
620
621fail_nocontext:
622	mm_free_pgd(mm);
623fail_nopgd:
624	free_mm(mm);
625	return NULL;
626}
627
628static void check_mm(struct mm_struct *mm)
629{
630	int i;
631
632	for (i = 0; i < NR_MM_COUNTERS; i++) {
633		long x = atomic_long_read(&mm->rss_stat.count[i]);
634
635		if (unlikely(x))
636			printk(KERN_ALERT "BUG: Bad rss-counter state "
637					  "mm:%p idx:%d val:%ld\n", mm, i, x);
638	}
639
640	if (atomic_long_read(&mm->nr_ptes))
641		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
642				atomic_long_read(&mm->nr_ptes));
643	if (mm_nr_pmds(mm))
644		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
645				mm_nr_pmds(mm));
646
647#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
648	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
649#endif
650}
651
652/*
653 * Allocate and initialize an mm_struct.
654 */
655struct mm_struct *mm_alloc(void)
656{
657	struct mm_struct *mm;
658
659	mm = allocate_mm();
660	if (!mm)
661		return NULL;
662
663	memset(mm, 0, sizeof(*mm));
664	return mm_init(mm, current);
665}
666
667/*
668 * Called when the last reference to the mm
669 * is dropped: either by a lazy thread or by
670 * mmput. Free the page directory and the mm.
671 */
672void __mmdrop(struct mm_struct *mm)
673{
674	BUG_ON(mm == &init_mm);
675	mm_free_pgd(mm);
676	destroy_context(mm);
677	mmu_notifier_mm_destroy(mm);
678	check_mm(mm);
679	free_mm(mm);
680}
681EXPORT_SYMBOL_GPL(__mmdrop);
682
683/*
684 * Decrement the use count and release all resources for an mm.
685 */
686void mmput(struct mm_struct *mm)
687{
688	might_sleep();
689
690	if (atomic_dec_and_test(&mm->mm_users)) {
691		uprobe_clear_state(mm);
692		exit_aio(mm);
693		ksm_exit(mm);
694		khugepaged_exit(mm); /* must run before exit_mmap */
695		exit_mmap(mm);
696		set_mm_exe_file(mm, NULL);
697		if (!list_empty(&mm->mmlist)) {
698			spin_lock(&mmlist_lock);
699			list_del(&mm->mmlist);
700			spin_unlock(&mmlist_lock);
701		}
702		if (mm->binfmt)
703			module_put(mm->binfmt->module);
704		mmdrop(mm);
705	}
706}
707EXPORT_SYMBOL_GPL(mmput);
708
709/**
710 * set_mm_exe_file - change a reference to the mm's executable file
711 *
712 * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
713 *
714 * Main users are mmput() and sys_execve(). Callers prevent concurrent
715 * invocations: in mmput() nobody alive left, in execve task is single
716 * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
717 * mm->exe_file, but does so without using set_mm_exe_file() in order
718 * to do avoid the need for any locks.
719 */
720void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
721{
722	struct file *old_exe_file;
723
724	/*
725	 * It is safe to dereference the exe_file without RCU as
726	 * this function is only called if nobody else can access
727	 * this mm -- see comment above for justification.
728	 */
729	old_exe_file = rcu_dereference_raw(mm->exe_file);
730
731	if (new_exe_file)
732		get_file(new_exe_file);
733	rcu_assign_pointer(mm->exe_file, new_exe_file);
734	if (old_exe_file)
735		fput(old_exe_file);
736}
737
738/**
739 * get_mm_exe_file - acquire a reference to the mm's executable file
740 *
741 * Returns %NULL if mm has no associated executable file.
742 * User must release file via fput().
743 */
744struct file *get_mm_exe_file(struct mm_struct *mm)
745{
746	struct file *exe_file;
747
748	rcu_read_lock();
749	exe_file = rcu_dereference(mm->exe_file);
750	if (exe_file && !get_file_rcu(exe_file))
751		exe_file = NULL;
752	rcu_read_unlock();
753	return exe_file;
754}
755EXPORT_SYMBOL(get_mm_exe_file);
756
757/**
758 * get_task_mm - acquire a reference to the task's mm
759 *
760 * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
761 * this kernel workthread has transiently adopted a user mm with use_mm,
762 * to do its AIO) is not set and if so returns a reference to it, after
763 * bumping up the use count.  User must release the mm via mmput()
764 * after use.  Typically used by /proc and ptrace.
765 */
766struct mm_struct *get_task_mm(struct task_struct *task)
767{
768	struct mm_struct *mm;
769
770	task_lock(task);
771	mm = task->mm;
772	if (mm) {
773		if (task->flags & PF_KTHREAD)
774			mm = NULL;
775		else
776			atomic_inc(&mm->mm_users);
777	}
778	task_unlock(task);
779	return mm;
780}
781EXPORT_SYMBOL_GPL(get_task_mm);
782
783struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
784{
785	struct mm_struct *mm;
786	int err;
787
788	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
789	if (err)
790		return ERR_PTR(err);
791
792	mm = get_task_mm(task);
793	if (mm && mm != current->mm &&
794			!ptrace_may_access(task, mode)) {
795		mmput(mm);
796		mm = ERR_PTR(-EACCES);
797	}
798	mutex_unlock(&task->signal->cred_guard_mutex);
799
800	return mm;
801}
802
803static void complete_vfork_done(struct task_struct *tsk)
804{
805	struct completion *vfork;
806
807	task_lock(tsk);
808	vfork = tsk->vfork_done;
809	if (likely(vfork)) {
810		tsk->vfork_done = NULL;
811		complete(vfork);
812	}
813	task_unlock(tsk);
814}
815
816static int wait_for_vfork_done(struct task_struct *child,
817				struct completion *vfork)
818{
819	int killed;
820
821	freezer_do_not_count();
822	killed = wait_for_completion_killable(vfork);
823	freezer_count();
824
825	if (killed) {
826		task_lock(child);
827		child->vfork_done = NULL;
828		task_unlock(child);
829	}
830
831	put_task_struct(child);
832	return killed;
833}
834
835/* Please note the differences between mmput and mm_release.
836 * mmput is called whenever we stop holding onto a mm_struct,
837 * error success whatever.
838 *
839 * mm_release is called after a mm_struct has been removed
840 * from the current process.
841 *
842 * This difference is important for error handling, when we
843 * only half set up a mm_struct for a new process and need to restore
844 * the old one.  Because we mmput the new mm_struct before
845 * restoring the old one. . .
846 * Eric Biederman 10 January 1998
847 */
848void mm_release(struct task_struct *tsk, struct mm_struct *mm)
849{
850	/* Get rid of any futexes when releasing the mm */
851#ifdef CONFIG_FUTEX
852	if (unlikely(tsk->robust_list)) {
853		exit_robust_list(tsk);
854		tsk->robust_list = NULL;
855	}
856#ifdef CONFIG_COMPAT
857	if (unlikely(tsk->compat_robust_list)) {
858		compat_exit_robust_list(tsk);
859		tsk->compat_robust_list = NULL;
860	}
861#endif
862	if (unlikely(!list_empty(&tsk->pi_state_list)))
863		exit_pi_state_list(tsk);
864#endif
865
866	uprobe_free_utask(tsk);
867
868	/* Get rid of any cached register state */
869	deactivate_mm(tsk, mm);
870
871	/*
872	 * If we're exiting normally, clear a user-space tid field if
873	 * requested.  We leave this alone when dying by signal, to leave
874	 * the value intact in a core dump, and to save the unnecessary
875	 * trouble, say, a killed vfork parent shouldn't touch this mm.
876	 * Userland only wants this done for a sys_exit.
877	 */
878	if (tsk->clear_child_tid) {
879		if (!(tsk->flags & PF_SIGNALED) &&
880		    atomic_read(&mm->mm_users) > 1) {
881			/*
882			 * We don't check the error code - if userspace has
883			 * not set up a proper pointer then tough luck.
884			 */
885			put_user(0, tsk->clear_child_tid);
886			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
887					1, NULL, NULL, 0);
888		}
889		tsk->clear_child_tid = NULL;
890	}
891
892	/*
893	 * All done, finally we can wake up parent and return this mm to him.
894	 * Also kthread_stop() uses this completion for synchronization.
895	 */
896	if (tsk->vfork_done)
897		complete_vfork_done(tsk);
898}
899
900/*
901 * Allocate a new mm structure and copy contents from the
902 * mm structure of the passed in task structure.
903 */
904static struct mm_struct *dup_mm(struct task_struct *tsk)
905{
906	struct mm_struct *mm, *oldmm = current->mm;
907	int err;
908
909	mm = allocate_mm();
910	if (!mm)
911		goto fail_nomem;
912
913	memcpy(mm, oldmm, sizeof(*mm));
914
915	if (!mm_init(mm, tsk))
916		goto fail_nomem;
917
918	err = dup_mmap(mm, oldmm);
919	if (err)
920		goto free_pt;
921
922	mm->hiwater_rss = get_mm_rss(mm);
923	mm->hiwater_vm = mm->total_vm;
924
925	if (mm->binfmt && !try_module_get(mm->binfmt->module))
926		goto free_pt;
927
928	return mm;
929
930free_pt:
931	/* don't put binfmt in mmput, we haven't got module yet */
932	mm->binfmt = NULL;
933	mmput(mm);
934
935fail_nomem:
936	return NULL;
937}
938
939static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
940{
941	struct mm_struct *mm, *oldmm;
942	int retval;
943
944	tsk->min_flt = tsk->maj_flt = 0;
945	tsk->nvcsw = tsk->nivcsw = 0;
946#ifdef CONFIG_DETECT_HUNG_TASK
947	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
948#endif
949
950	tsk->mm = NULL;
951	tsk->active_mm = NULL;
952
953	/*
954	 * Are we cloning a kernel thread?
955	 *
956	 * We need to steal a active VM for that..
957	 */
958	oldmm = current->mm;
959	if (!oldmm)
960		return 0;
961
962	/* initialize the new vmacache entries */
963	vmacache_flush(tsk);
964
965	if (clone_flags & CLONE_VM) {
966		atomic_inc(&oldmm->mm_users);
967		mm = oldmm;
968		goto good_mm;
969	}
970
971	retval = -ENOMEM;
972	mm = dup_mm(tsk);
973	if (!mm)
974		goto fail_nomem;
975
976good_mm:
977	tsk->mm = mm;
978	tsk->active_mm = mm;
979	return 0;
980
981fail_nomem:
982	return retval;
983}
984
985static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
986{
987	struct fs_struct *fs = current->fs;
988	if (clone_flags & CLONE_FS) {
989		/* tsk->fs is already what we want */
990		spin_lock(&fs->lock);
991		if (fs->in_exec) {
992			spin_unlock(&fs->lock);
993			return -EAGAIN;
994		}
995		fs->users++;
996		spin_unlock(&fs->lock);
997		return 0;
998	}
999	tsk->fs = copy_fs_struct(fs);
1000	if (!tsk->fs)
1001		return -ENOMEM;
1002	return 0;
1003}
1004
1005static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1006{
1007	struct files_struct *oldf, *newf;
1008	int error = 0;
1009
1010	/*
1011	 * A background process may not have any files ...
1012	 */
1013	oldf = current->files;
1014	if (!oldf)
1015		goto out;
1016
1017	if (clone_flags & CLONE_FILES) {
1018		atomic_inc(&oldf->count);
1019		goto out;
1020	}
1021
1022	newf = dup_fd(oldf, &error);
1023	if (!newf)
1024		goto out;
1025
1026	tsk->files = newf;
1027	error = 0;
1028out:
1029	return error;
1030}
1031
1032static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1033{
1034#ifdef CONFIG_BLOCK
1035	struct io_context *ioc = current->io_context;
1036	struct io_context *new_ioc;
1037
1038	if (!ioc)
1039		return 0;
1040	/*
1041	 * Share io context with parent, if CLONE_IO is set
1042	 */
1043	if (clone_flags & CLONE_IO) {
1044		ioc_task_link(ioc);
1045		tsk->io_context = ioc;
1046	} else if (ioprio_valid(ioc->ioprio)) {
1047		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1048		if (unlikely(!new_ioc))
1049			return -ENOMEM;
1050
1051		new_ioc->ioprio = ioc->ioprio;
1052		put_io_context(new_ioc);
1053	}
1054#endif
1055	return 0;
1056}
1057
1058static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1059{
1060	struct sighand_struct *sig;
1061
1062	if (clone_flags & CLONE_SIGHAND) {
1063		atomic_inc(&current->sighand->count);
1064		return 0;
1065	}
1066	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1067	rcu_assign_pointer(tsk->sighand, sig);
1068	if (!sig)
1069		return -ENOMEM;
1070	atomic_set(&sig->count, 1);
1071	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1072	return 0;
1073}
1074
1075void __cleanup_sighand(struct sighand_struct *sighand)
1076{
1077	if (atomic_dec_and_test(&sighand->count)) {
1078		signalfd_cleanup(sighand);
1079		/*
1080		 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1081		 * without an RCU grace period, see __lock_task_sighand().
1082		 */
1083		kmem_cache_free(sighand_cachep, sighand);
1084	}
1085}
1086
1087/*
1088 * Initialize POSIX timer handling for a thread group.
1089 */
1090static void posix_cpu_timers_init_group(struct signal_struct *sig)
1091{
1092	unsigned long cpu_limit;
1093
1094	/* Thread group counters. */
1095	thread_group_cputime_init(sig);
1096
1097	cpu_limit = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1098	if (cpu_limit != RLIM_INFINITY) {
1099		sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1100		sig->cputimer.running = 1;
1101	}
1102
1103	/* The timer lists. */
1104	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1105	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1106	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1107}
1108
1109static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1110{
1111	struct signal_struct *sig;
1112
1113	if (clone_flags & CLONE_THREAD)
1114		return 0;
1115
1116	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1117	tsk->signal = sig;
1118	if (!sig)
1119		return -ENOMEM;
1120
1121	sig->nr_threads = 1;
1122	atomic_set(&sig->live, 1);
1123	atomic_set(&sig->sigcnt, 1);
1124
1125	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1126	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1127	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1128
1129	init_waitqueue_head(&sig->wait_chldexit);
1130	sig->curr_target = tsk;
1131	init_sigpending(&sig->shared_pending);
1132	INIT_LIST_HEAD(&sig->posix_timers);
1133	seqlock_init(&sig->stats_lock);
1134
1135	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1136	sig->real_timer.function = it_real_fn;
1137
1138	task_lock(current->group_leader);
1139	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1140	task_unlock(current->group_leader);
1141
1142	posix_cpu_timers_init_group(sig);
1143
1144	tty_audit_fork(sig);
1145	sched_autogroup_fork(sig);
1146
1147#ifdef CONFIG_CGROUPS
1148	init_rwsem(&sig->group_rwsem);
1149#endif
1150
1151	sig->oom_score_adj = current->signal->oom_score_adj;
1152	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1153
1154	sig->has_child_subreaper = current->signal->has_child_subreaper ||
1155				   current->signal->is_child_subreaper;
1156
1157	mutex_init(&sig->cred_guard_mutex);
1158
1159	return 0;
1160}
1161
1162static void copy_seccomp(struct task_struct *p)
1163{
1164#ifdef CONFIG_SECCOMP
1165	/*
1166	 * Must be called with sighand->lock held, which is common to
1167	 * all threads in the group. Holding cred_guard_mutex is not
1168	 * needed because this new task is not yet running and cannot
1169	 * be racing exec.
1170	 */
1171	assert_spin_locked(&current->sighand->siglock);
1172
1173	/* Ref-count the new filter user, and assign it. */
1174	get_seccomp_filter(current);
1175	p->seccomp = current->seccomp;
1176
1177	/*
1178	 * Explicitly enable no_new_privs here in case it got set
1179	 * between the task_struct being duplicated and holding the
1180	 * sighand lock. The seccomp state and nnp must be in sync.
1181	 */
1182	if (task_no_new_privs(current))
1183		task_set_no_new_privs(p);
1184
1185	/*
1186	 * If the parent gained a seccomp mode after copying thread
1187	 * flags and between before we held the sighand lock, we have
1188	 * to manually enable the seccomp thread flag here.
1189	 */
1190	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1191		set_tsk_thread_flag(p, TIF_SECCOMP);
1192#endif
1193}
1194
1195SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1196{
1197	current->clear_child_tid = tidptr;
1198
1199	return task_pid_vnr(current);
1200}
1201
1202static void rt_mutex_init_task(struct task_struct *p)
1203{
1204	raw_spin_lock_init(&p->pi_lock);
1205#ifdef CONFIG_RT_MUTEXES
1206	p->pi_waiters = RB_ROOT;
1207	p->pi_waiters_leftmost = NULL;
1208	p->pi_blocked_on = NULL;
1209#endif
1210}
1211
1212/*
1213 * Initialize POSIX timer handling for a single task.
1214 */
1215static void posix_cpu_timers_init(struct task_struct *tsk)
1216{
1217	tsk->cputime_expires.prof_exp = 0;
1218	tsk->cputime_expires.virt_exp = 0;
1219	tsk->cputime_expires.sched_exp = 0;
1220	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1221	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1222	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1223}
1224
1225static inline void
1226init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1227{
1228	 task->pids[type].pid = pid;
1229}
1230
1231/*
1232 * This creates a new process as a copy of the old one,
1233 * but does not actually start it yet.
1234 *
1235 * It copies the registers, and all the appropriate
1236 * parts of the process environment (as per the clone
1237 * flags). The actual kick-off is left to the caller.
1238 */
1239static struct task_struct *copy_process(unsigned long clone_flags,
1240					unsigned long stack_start,
1241					unsigned long stack_size,
1242					int __user *child_tidptr,
1243					struct pid *pid,
1244					int trace)
1245{
1246	int retval;
1247	struct task_struct *p;
1248
1249	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1250		return ERR_PTR(-EINVAL);
1251
1252	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1253		return ERR_PTR(-EINVAL);
1254
1255	/*
1256	 * Thread groups must share signals as well, and detached threads
1257	 * can only be started up within the thread group.
1258	 */
1259	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1260		return ERR_PTR(-EINVAL);
1261
1262	/*
1263	 * Shared signal handlers imply shared VM. By way of the above,
1264	 * thread groups also imply shared VM. Blocking this case allows
1265	 * for various simplifications in other code.
1266	 */
1267	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1268		return ERR_PTR(-EINVAL);
1269
1270	/*
1271	 * Siblings of global init remain as zombies on exit since they are
1272	 * not reaped by their parent (swapper). To solve this and to avoid
1273	 * multi-rooted process trees, prevent global and container-inits
1274	 * from creating siblings.
1275	 */
1276	if ((clone_flags & CLONE_PARENT) &&
1277				current->signal->flags & SIGNAL_UNKILLABLE)
1278		return ERR_PTR(-EINVAL);
1279
1280	/*
1281	 * If the new process will be in a different pid or user namespace
1282	 * do not allow it to share a thread group or signal handlers or
1283	 * parent with the forking task.
1284	 */
1285	if (clone_flags & CLONE_SIGHAND) {
1286		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1287		    (task_active_pid_ns(current) !=
1288				current->nsproxy->pid_ns_for_children))
1289			return ERR_PTR(-EINVAL);
1290	}
1291
1292	retval = security_task_create(clone_flags);
1293	if (retval)
1294		goto fork_out;
1295
1296	retval = -ENOMEM;
1297	p = dup_task_struct(current);
1298	if (!p)
1299		goto fork_out;
1300
1301	ftrace_graph_init_task(p);
1302
1303	rt_mutex_init_task(p);
1304
1305#ifdef CONFIG_PROVE_LOCKING
1306	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1307	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1308#endif
1309	retval = -EAGAIN;
1310	if (atomic_read(&p->real_cred->user->processes) >=
1311			task_rlimit(p, RLIMIT_NPROC)) {
1312		if (p->real_cred->user != INIT_USER &&
1313		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1314			goto bad_fork_free;
1315	}
1316	current->flags &= ~PF_NPROC_EXCEEDED;
1317
1318	retval = copy_creds(p, clone_flags);
1319	if (retval < 0)
1320		goto bad_fork_free;
1321
1322	/*
1323	 * If multiple threads are within copy_process(), then this check
1324	 * triggers too late. This doesn't hurt, the check is only there
1325	 * to stop root fork bombs.
1326	 */
1327	retval = -EAGAIN;
1328	if (nr_threads >= max_threads)
1329		goto bad_fork_cleanup_count;
1330
1331	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1332	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1333	p->flags |= PF_FORKNOEXEC;
1334	INIT_LIST_HEAD(&p->children);
1335	INIT_LIST_HEAD(&p->sibling);
1336	rcu_copy_process(p);
1337	p->vfork_done = NULL;
1338	spin_lock_init(&p->alloc_lock);
1339
1340	init_sigpending(&p->pending);
1341
1342	p->utime = p->stime = p->gtime = 0;
1343	p->utimescaled = p->stimescaled = 0;
1344#ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
1345	p->prev_cputime.utime = p->prev_cputime.stime = 0;
1346#endif
1347#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1348	seqlock_init(&p->vtime_seqlock);
1349	p->vtime_snap = 0;
1350	p->vtime_snap_whence = VTIME_SLEEPING;
1351#endif
1352
1353#if defined(SPLIT_RSS_COUNTING)
1354	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1355#endif
1356
1357	p->default_timer_slack_ns = current->timer_slack_ns;
1358
1359	task_io_accounting_init(&p->ioac);
1360	acct_clear_integrals(p);
1361
1362	posix_cpu_timers_init(p);
1363
1364	p->start_time = ktime_get_ns();
1365	p->real_start_time = ktime_get_boot_ns();
1366	p->io_context = NULL;
1367	p->audit_context = NULL;
1368	if (clone_flags & CLONE_THREAD)
1369		threadgroup_change_begin(current);
1370	cgroup_fork(p);
1371#ifdef CONFIG_NUMA
1372	p->mempolicy = mpol_dup(p->mempolicy);
1373	if (IS_ERR(p->mempolicy)) {
1374		retval = PTR_ERR(p->mempolicy);
1375		p->mempolicy = NULL;
1376		goto bad_fork_cleanup_threadgroup_lock;
1377	}
1378#endif
1379#ifdef CONFIG_CPUSETS
1380	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1381	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1382	seqcount_init(&p->mems_allowed_seq);
1383#endif
1384#ifdef CONFIG_TRACE_IRQFLAGS
1385	p->irq_events = 0;
1386	p->hardirqs_enabled = 0;
1387	p->hardirq_enable_ip = 0;
1388	p->hardirq_enable_event = 0;
1389	p->hardirq_disable_ip = _THIS_IP_;
1390	p->hardirq_disable_event = 0;
1391	p->softirqs_enabled = 1;
1392	p->softirq_enable_ip = _THIS_IP_;
1393	p->softirq_enable_event = 0;
1394	p->softirq_disable_ip = 0;
1395	p->softirq_disable_event = 0;
1396	p->hardirq_context = 0;
1397	p->softirq_context = 0;
1398#endif
1399#ifdef CONFIG_LOCKDEP
1400	p->lockdep_depth = 0; /* no locks held yet */
1401	p->curr_chain_key = 0;
1402	p->lockdep_recursion = 0;
1403#endif
1404
1405#ifdef CONFIG_DEBUG_MUTEXES
1406	p->blocked_on = NULL; /* not blocked yet */
1407#endif
1408#ifdef CONFIG_BCACHE
1409	p->sequential_io	= 0;
1410	p->sequential_io_avg	= 0;
1411#endif
1412
1413	/* Perform scheduler related setup. Assign this task to a CPU. */
1414	retval = sched_fork(clone_flags, p);
1415	if (retval)
1416		goto bad_fork_cleanup_policy;
1417
1418	retval = perf_event_init_task(p);
1419	if (retval)
1420		goto bad_fork_cleanup_policy;
1421	retval = audit_alloc(p);
1422	if (retval)
1423		goto bad_fork_cleanup_perf;
1424	/* copy all the process information */
1425	shm_init_task(p);
1426	retval = copy_semundo(clone_flags, p);
1427	if (retval)
1428		goto bad_fork_cleanup_audit;
1429	retval = copy_files(clone_flags, p);
1430	if (retval)
1431		goto bad_fork_cleanup_semundo;
1432	retval = copy_fs(clone_flags, p);
1433	if (retval)
1434		goto bad_fork_cleanup_files;
1435	retval = copy_sighand(clone_flags, p);
1436	if (retval)
1437		goto bad_fork_cleanup_fs;
1438	retval = copy_signal(clone_flags, p);
1439	if (retval)
1440		goto bad_fork_cleanup_sighand;
1441	retval = copy_mm(clone_flags, p);
1442	if (retval)
1443		goto bad_fork_cleanup_signal;
1444	retval = copy_namespaces(clone_flags, p);
1445	if (retval)
1446		goto bad_fork_cleanup_mm;
1447	retval = copy_io(clone_flags, p);
1448	if (retval)
1449		goto bad_fork_cleanup_namespaces;
1450	retval = copy_thread(clone_flags, stack_start, stack_size, p);
1451	if (retval)
1452		goto bad_fork_cleanup_io;
1453
1454	if (pid != &init_struct_pid) {
1455		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1456		if (IS_ERR(pid)) {
1457			retval = PTR_ERR(pid);
1458			goto bad_fork_cleanup_io;
1459		}
1460	}
1461
1462	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1463	/*
1464	 * Clear TID on mm_release()?
1465	 */
1466	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1467#ifdef CONFIG_BLOCK
1468	p->plug = NULL;
1469#endif
1470#ifdef CONFIG_FUTEX
1471	p->robust_list = NULL;
1472#ifdef CONFIG_COMPAT
1473	p->compat_robust_list = NULL;
1474#endif
1475	INIT_LIST_HEAD(&p->pi_state_list);
1476	p->pi_state_cache = NULL;
1477#endif
1478	/*
1479	 * sigaltstack should be cleared when sharing the same VM
1480	 */
1481	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1482		p->sas_ss_sp = p->sas_ss_size = 0;
1483
1484	/*
1485	 * Syscall tracing and stepping should be turned off in the
1486	 * child regardless of CLONE_PTRACE.
1487	 */
1488	user_disable_single_step(p);
1489	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1490#ifdef TIF_SYSCALL_EMU
1491	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1492#endif
1493	clear_all_latency_tracing(p);
1494
1495	/* ok, now we should be set up.. */
1496	p->pid = pid_nr(pid);
1497	if (clone_flags & CLONE_THREAD) {
1498		p->exit_signal = -1;
1499		p->group_leader = current->group_leader;
1500		p->tgid = current->tgid;
1501	} else {
1502		if (clone_flags & CLONE_PARENT)
1503			p->exit_signal = current->group_leader->exit_signal;
1504		else
1505			p->exit_signal = (clone_flags & CSIGNAL);
1506		p->group_leader = p;
1507		p->tgid = p->pid;
1508	}
1509
1510	p->nr_dirtied = 0;
1511	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1512	p->dirty_paused_when = 0;
1513
1514	p->pdeath_signal = 0;
1515	INIT_LIST_HEAD(&p->thread_group);
1516	p->task_works = NULL;
1517
1518	/*
1519	 * Make it visible to the rest of the system, but dont wake it up yet.
1520	 * Need tasklist lock for parent etc handling!
1521	 */
1522	write_lock_irq(&tasklist_lock);
1523
1524	/* CLONE_PARENT re-uses the old parent */
1525	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1526		p->real_parent = current->real_parent;
1527		p->parent_exec_id = current->parent_exec_id;
1528	} else {
1529		p->real_parent = current;
1530		p->parent_exec_id = current->self_exec_id;
1531	}
1532
1533	spin_lock(&current->sighand->siglock);
1534
1535	/*
1536	 * Copy seccomp details explicitly here, in case they were changed
1537	 * before holding sighand lock.
1538	 */
1539	copy_seccomp(p);
1540
1541	/*
1542	 * Process group and session signals need to be delivered to just the
1543	 * parent before the fork or both the parent and the child after the
1544	 * fork. Restart if a signal comes in before we add the new process to
1545	 * it's process group.
1546	 * A fatal signal pending means that current will exit, so the new
1547	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1548	*/
1549	recalc_sigpending();
1550	if (signal_pending(current)) {
1551		spin_unlock(&current->sighand->siglock);
1552		write_unlock_irq(&tasklist_lock);
1553		retval = -ERESTARTNOINTR;
1554		goto bad_fork_free_pid;
1555	}
1556
1557	if (likely(p->pid)) {
1558		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1559
1560		init_task_pid(p, PIDTYPE_PID, pid);
1561		if (thread_group_leader(p)) {
1562			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1563			init_task_pid(p, PIDTYPE_SID, task_session(current));
1564
1565			if (is_child_reaper(pid)) {
1566				ns_of_pid(pid)->child_reaper = p;
1567				p->signal->flags |= SIGNAL_UNKILLABLE;
1568			}
1569
1570			p->signal->leader_pid = pid;
1571			p->signal->tty = tty_kref_get(current->signal->tty);
1572			list_add_tail(&p->sibling, &p->real_parent->children);
1573			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1574			attach_pid(p, PIDTYPE_PGID);
1575			attach_pid(p, PIDTYPE_SID);
1576			__this_cpu_inc(process_counts);
1577		} else {
1578			current->signal->nr_threads++;
1579			atomic_inc(&current->signal->live);
1580			atomic_inc(&current->signal->sigcnt);
1581			list_add_tail_rcu(&p->thread_group,
1582					  &p->group_leader->thread_group);
1583			list_add_tail_rcu(&p->thread_node,
1584					  &p->signal->thread_head);
1585		}
1586		attach_pid(p, PIDTYPE_PID);
1587		nr_threads++;
1588	}
1589
1590	total_forks++;
1591	spin_unlock(&current->sighand->siglock);
1592	syscall_tracepoint_update(p);
1593	write_unlock_irq(&tasklist_lock);
1594
1595	proc_fork_connector(p);
1596	cgroup_post_fork(p);
1597	if (clone_flags & CLONE_THREAD)
1598		threadgroup_change_end(current);
1599	perf_event_fork(p);
1600
1601	trace_task_newtask(p, clone_flags);
1602	uprobe_copy_process(p, clone_flags);
1603
1604	return p;
1605
1606bad_fork_free_pid:
1607	if (pid != &init_struct_pid)
1608		free_pid(pid);
1609bad_fork_cleanup_io:
1610	if (p->io_context)
1611		exit_io_context(p);
1612bad_fork_cleanup_namespaces:
1613	exit_task_namespaces(p);
1614bad_fork_cleanup_mm:
1615	if (p->mm)
1616		mmput(p->mm);
1617bad_fork_cleanup_signal:
1618	if (!(clone_flags & CLONE_THREAD))
1619		free_signal_struct(p->signal);
1620bad_fork_cleanup_sighand:
1621	__cleanup_sighand(p->sighand);
1622bad_fork_cleanup_fs:
1623	exit_fs(p); /* blocking */
1624bad_fork_cleanup_files:
1625	exit_files(p); /* blocking */
1626bad_fork_cleanup_semundo:
1627	exit_sem(p);
1628bad_fork_cleanup_audit:
1629	audit_free(p);
1630bad_fork_cleanup_perf:
1631	perf_event_free_task(p);
1632bad_fork_cleanup_policy:
1633#ifdef CONFIG_NUMA
1634	mpol_put(p->mempolicy);
1635bad_fork_cleanup_threadgroup_lock:
1636#endif
1637	if (clone_flags & CLONE_THREAD)
1638		threadgroup_change_end(current);
1639	delayacct_tsk_free(p);
1640bad_fork_cleanup_count:
1641	atomic_dec(&p->cred->user->processes);
1642	exit_creds(p);
1643bad_fork_free:
1644	free_task(p);
1645fork_out:
1646	return ERR_PTR(retval);
1647}
1648
1649static inline void init_idle_pids(struct pid_link *links)
1650{
1651	enum pid_type type;
1652
1653	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1654		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1655		links[type].pid = &init_struct_pid;
1656	}
1657}
1658
1659struct task_struct *fork_idle(int cpu)
1660{
1661	struct task_struct *task;
1662	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0);
1663	if (!IS_ERR(task)) {
1664		init_idle_pids(task->pids);
1665		init_idle(task, cpu);
1666	}
1667
1668	return task;
1669}
1670
1671/*
1672 *  Ok, this is the main fork-routine.
1673 *
1674 * It copies the process, and if successful kick-starts
1675 * it and waits for it to finish using the VM if required.
1676 */
1677long do_fork(unsigned long clone_flags,
1678	      unsigned long stack_start,
1679	      unsigned long stack_size,
1680	      int __user *parent_tidptr,
1681	      int __user *child_tidptr)
1682{
1683	struct task_struct *p;
1684	int trace = 0;
1685	long nr;
1686
1687	/*
1688	 * Determine whether and which event to report to ptracer.  When
1689	 * called from kernel_thread or CLONE_UNTRACED is explicitly
1690	 * requested, no event is reported; otherwise, report if the event
1691	 * for the type of forking is enabled.
1692	 */
1693	if (!(clone_flags & CLONE_UNTRACED)) {
1694		if (clone_flags & CLONE_VFORK)
1695			trace = PTRACE_EVENT_VFORK;
1696		else if ((clone_flags & CSIGNAL) != SIGCHLD)
1697			trace = PTRACE_EVENT_CLONE;
1698		else
1699			trace = PTRACE_EVENT_FORK;
1700
1701		if (likely(!ptrace_event_enabled(current, trace)))
1702			trace = 0;
1703	}
1704
1705	p = copy_process(clone_flags, stack_start, stack_size,
1706			 child_tidptr, NULL, trace);
1707	/*
1708	 * Do this prior waking up the new thread - the thread pointer
1709	 * might get invalid after that point, if the thread exits quickly.
1710	 */
1711	if (!IS_ERR(p)) {
1712		struct completion vfork;
1713		struct pid *pid;
1714
1715		trace_sched_process_fork(current, p);
1716
1717		pid = get_task_pid(p, PIDTYPE_PID);
1718		nr = pid_vnr(pid);
1719
1720		if (clone_flags & CLONE_PARENT_SETTID)
1721			put_user(nr, parent_tidptr);
1722
1723		if (clone_flags & CLONE_VFORK) {
1724			p->vfork_done = &vfork;
1725			init_completion(&vfork);
1726			get_task_struct(p);
1727		}
1728
1729		wake_up_new_task(p);
1730
1731		/* forking complete and child started to run, tell ptracer */
1732		if (unlikely(trace))
1733			ptrace_event_pid(trace, pid);
1734
1735		if (clone_flags & CLONE_VFORK) {
1736			if (!wait_for_vfork_done(p, &vfork))
1737				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1738		}
1739
1740		put_pid(pid);
1741	} else {
1742		nr = PTR_ERR(p);
1743	}
1744	return nr;
1745}
1746
1747/*
1748 * Create a kernel thread.
1749 */
1750pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1751{
1752	return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1753		(unsigned long)arg, NULL, NULL);
1754}
1755
1756#ifdef __ARCH_WANT_SYS_FORK
1757SYSCALL_DEFINE0(fork)
1758{
1759#ifdef CONFIG_MMU
1760	return do_fork(SIGCHLD, 0, 0, NULL, NULL);
1761#else
1762	/* can not support in nommu mode */
1763	return -EINVAL;
1764#endif
1765}
1766#endif
1767
1768#ifdef __ARCH_WANT_SYS_VFORK
1769SYSCALL_DEFINE0(vfork)
1770{
1771	return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1772			0, NULL, NULL);
1773}
1774#endif
1775
1776#ifdef __ARCH_WANT_SYS_CLONE
1777#ifdef CONFIG_CLONE_BACKWARDS
1778SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1779		 int __user *, parent_tidptr,
1780		 int, tls_val,
1781		 int __user *, child_tidptr)
1782#elif defined(CONFIG_CLONE_BACKWARDS2)
1783SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1784		 int __user *, parent_tidptr,
1785		 int __user *, child_tidptr,
1786		 int, tls_val)
1787#elif defined(CONFIG_CLONE_BACKWARDS3)
1788SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1789		int, stack_size,
1790		int __user *, parent_tidptr,
1791		int __user *, child_tidptr,
1792		int, tls_val)
1793#else
1794SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1795		 int __user *, parent_tidptr,
1796		 int __user *, child_tidptr,
1797		 int, tls_val)
1798#endif
1799{
1800	return do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr);
1801}
1802#endif
1803
1804#ifndef ARCH_MIN_MMSTRUCT_ALIGN
1805#define ARCH_MIN_MMSTRUCT_ALIGN 0
1806#endif
1807
1808static void sighand_ctor(void *data)
1809{
1810	struct sighand_struct *sighand = data;
1811
1812	spin_lock_init(&sighand->siglock);
1813	init_waitqueue_head(&sighand->signalfd_wqh);
1814}
1815
1816void __init proc_caches_init(void)
1817{
1818	sighand_cachep = kmem_cache_create("sighand_cache",
1819			sizeof(struct sighand_struct), 0,
1820			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1821			SLAB_NOTRACK, sighand_ctor);
1822	signal_cachep = kmem_cache_create("signal_cache",
1823			sizeof(struct signal_struct), 0,
1824			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1825	files_cachep = kmem_cache_create("files_cache",
1826			sizeof(struct files_struct), 0,
1827			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1828	fs_cachep = kmem_cache_create("fs_cache",
1829			sizeof(struct fs_struct), 0,
1830			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1831	/*
1832	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1833	 * whole struct cpumask for the OFFSTACK case. We could change
1834	 * this to *only* allocate as much of it as required by the
1835	 * maximum number of CPU's we can ever have.  The cpumask_allocation
1836	 * is at the end of the structure, exactly for that reason.
1837	 */
1838	mm_cachep = kmem_cache_create("mm_struct",
1839			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1840			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
1841	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1842	mmap_init();
1843	nsproxy_cache_init();
1844}
1845
1846/*
1847 * Check constraints on flags passed to the unshare system call.
1848 */
1849static int check_unshare_flags(unsigned long unshare_flags)
1850{
1851	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1852				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1853				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1854				CLONE_NEWUSER|CLONE_NEWPID))
1855		return -EINVAL;
1856	/*
1857	 * Not implemented, but pretend it works if there is nothing
1858	 * to unshare.  Note that unsharing the address space or the
1859	 * signal handlers also need to unshare the signal queues (aka
1860	 * CLONE_THREAD).
1861	 */
1862	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1863		if (!thread_group_empty(current))
1864			return -EINVAL;
1865	}
1866	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1867		if (atomic_read(&current->sighand->count) > 1)
1868			return -EINVAL;
1869	}
1870	if (unshare_flags & CLONE_VM) {
1871		if (!current_is_single_threaded())
1872			return -EINVAL;
1873	}
1874
1875	return 0;
1876}
1877
1878/*
1879 * Unshare the filesystem structure if it is being shared
1880 */
1881static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1882{
1883	struct fs_struct *fs = current->fs;
1884
1885	if (!(unshare_flags & CLONE_FS) || !fs)
1886		return 0;
1887
1888	/* don't need lock here; in the worst case we'll do useless copy */
1889	if (fs->users == 1)
1890		return 0;
1891
1892	*new_fsp = copy_fs_struct(fs);
1893	if (!*new_fsp)
1894		return -ENOMEM;
1895
1896	return 0;
1897}
1898
1899/*
1900 * Unshare file descriptor table if it is being shared
1901 */
1902static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1903{
1904	struct files_struct *fd = current->files;
1905	int error = 0;
1906
1907	if ((unshare_flags & CLONE_FILES) &&
1908	    (fd && atomic_read(&fd->count) > 1)) {
1909		*new_fdp = dup_fd(fd, &error);
1910		if (!*new_fdp)
1911			return error;
1912	}
1913
1914	return 0;
1915}
1916
1917/*
1918 * unshare allows a process to 'unshare' part of the process
1919 * context which was originally shared using clone.  copy_*
1920 * functions used by do_fork() cannot be used here directly
1921 * because they modify an inactive task_struct that is being
1922 * constructed. Here we are modifying the current, active,
1923 * task_struct.
1924 */
1925SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1926{
1927	struct fs_struct *fs, *new_fs = NULL;
1928	struct files_struct *fd, *new_fd = NULL;
1929	struct cred *new_cred = NULL;
1930	struct nsproxy *new_nsproxy = NULL;
1931	int do_sysvsem = 0;
1932	int err;
1933
1934	/*
1935	 * If unsharing a user namespace must also unshare the thread.
1936	 */
1937	if (unshare_flags & CLONE_NEWUSER)
1938		unshare_flags |= CLONE_THREAD | CLONE_FS;
1939	/*
1940	 * If unsharing vm, must also unshare signal handlers.
1941	 */
1942	if (unshare_flags & CLONE_VM)
1943		unshare_flags |= CLONE_SIGHAND;
1944	/*
1945	 * If unsharing a signal handlers, must also unshare the signal queues.
1946	 */
1947	if (unshare_flags & CLONE_SIGHAND)
1948		unshare_flags |= CLONE_THREAD;
1949	/*
1950	 * If unsharing namespace, must also unshare filesystem information.
1951	 */
1952	if (unshare_flags & CLONE_NEWNS)
1953		unshare_flags |= CLONE_FS;
1954
1955	err = check_unshare_flags(unshare_flags);
1956	if (err)
1957		goto bad_unshare_out;
1958	/*
1959	 * CLONE_NEWIPC must also detach from the undolist: after switching
1960	 * to a new ipc namespace, the semaphore arrays from the old
1961	 * namespace are unreachable.
1962	 */
1963	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
1964		do_sysvsem = 1;
1965	err = unshare_fs(unshare_flags, &new_fs);
1966	if (err)
1967		goto bad_unshare_out;
1968	err = unshare_fd(unshare_flags, &new_fd);
1969	if (err)
1970		goto bad_unshare_cleanup_fs;
1971	err = unshare_userns(unshare_flags, &new_cred);
1972	if (err)
1973		goto bad_unshare_cleanup_fd;
1974	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
1975					 new_cred, new_fs);
1976	if (err)
1977		goto bad_unshare_cleanup_cred;
1978
1979	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
1980		if (do_sysvsem) {
1981			/*
1982			 * CLONE_SYSVSEM is equivalent to sys_exit().
1983			 */
1984			exit_sem(current);
1985		}
1986		if (unshare_flags & CLONE_NEWIPC) {
1987			/* Orphan segments in old ns (see sem above). */
1988			exit_shm(current);
1989			shm_init_task(current);
1990		}
1991
1992		if (new_nsproxy)
1993			switch_task_namespaces(current, new_nsproxy);
1994
1995		task_lock(current);
1996
1997		if (new_fs) {
1998			fs = current->fs;
1999			spin_lock(&fs->lock);
2000			current->fs = new_fs;
2001			if (--fs->users)
2002				new_fs = NULL;
2003			else
2004				new_fs = fs;
2005			spin_unlock(&fs->lock);
2006		}
2007
2008		if (new_fd) {
2009			fd = current->files;
2010			current->files = new_fd;
2011			new_fd = fd;
2012		}
2013
2014		task_unlock(current);
2015
2016		if (new_cred) {
2017			/* Install the new user namespace */
2018			commit_creds(new_cred);
2019			new_cred = NULL;
2020		}
2021	}
2022
2023bad_unshare_cleanup_cred:
2024	if (new_cred)
2025		put_cred(new_cred);
2026bad_unshare_cleanup_fd:
2027	if (new_fd)
2028		put_files_struct(new_fd);
2029
2030bad_unshare_cleanup_fs:
2031	if (new_fs)
2032		free_fs_struct(new_fs);
2033
2034bad_unshare_out:
2035	return err;
2036}
2037
2038/*
2039 *	Helper to unshare the files of the current task.
2040 *	We don't want to expose copy_files internals to
2041 *	the exec layer of the kernel.
2042 */
2043
2044int unshare_files(struct files_struct **displaced)
2045{
2046	struct task_struct *task = current;
2047	struct files_struct *copy = NULL;
2048	int error;
2049
2050	error = unshare_fd(CLONE_FILES, &copy);
2051	if (error || !copy) {
2052		*displaced = NULL;
2053		return error;
2054	}
2055	*displaced = task->files;
2056	task_lock(task);
2057	task->files = copy;
2058	task_unlock(task);
2059	return 0;
2060}
2061
2062int sysctl_max_threads(struct ctl_table *table, int write,
2063		       void __user *buffer, size_t *lenp, loff_t *ppos)
2064{
2065	struct ctl_table t;
2066	int ret;
2067	int threads = max_threads;
2068	int min = MIN_THREADS;
2069	int max = MAX_THREADS;
2070
2071	t = *table;
2072	t.data = &threads;
2073	t.extra1 = &min;
2074	t.extra2 = &max;
2075
2076	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2077	if (ret || !write)
2078		return ret;
2079
2080	set_max_threads(threads);
2081
2082	return 0;
2083}
2084