xref: /linux/kernel/fork.c (revision 9efac6798b20ae6b63ce244b569755f3b60d80aa)
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 #include <linux/kcov.h>
79 
80 #include <asm/pgtable.h>
81 #include <asm/pgalloc.h>
82 #include <asm/uaccess.h>
83 #include <asm/mmu_context.h>
84 #include <asm/cacheflush.h>
85 #include <asm/tlbflush.h>
86 
87 #include <trace/events/sched.h>
88 
89 #define CREATE_TRACE_POINTS
90 #include <trace/events/task.h>
91 
92 /*
93  * Minimum number of threads to boot the kernel
94  */
95 #define MIN_THREADS 20
96 
97 /*
98  * Maximum number of threads
99  */
100 #define MAX_THREADS FUTEX_TID_MASK
101 
102 /*
103  * Protected counters by write_lock_irq(&tasklist_lock)
104  */
105 unsigned long total_forks;	/* Handle normal Linux uptimes. */
106 int nr_threads;			/* The idle threads do not count.. */
107 
108 int max_threads;		/* tunable limit on nr_threads */
109 
110 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
111 
112 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
113 
114 #ifdef CONFIG_PROVE_RCU
115 int lockdep_tasklist_lock_is_held(void)
116 {
117 	return lockdep_is_held(&tasklist_lock);
118 }
119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
120 #endif /* #ifdef CONFIG_PROVE_RCU */
121 
122 int nr_processes(void)
123 {
124 	int cpu;
125 	int total = 0;
126 
127 	for_each_possible_cpu(cpu)
128 		total += per_cpu(process_counts, cpu);
129 
130 	return total;
131 }
132 
133 void __weak arch_release_task_struct(struct task_struct *tsk)
134 {
135 }
136 
137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
138 static struct kmem_cache *task_struct_cachep;
139 
140 static inline struct task_struct *alloc_task_struct_node(int node)
141 {
142 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
143 }
144 
145 static inline void free_task_struct(struct task_struct *tsk)
146 {
147 	kmem_cache_free(task_struct_cachep, tsk);
148 }
149 #endif
150 
151 void __weak arch_release_thread_stack(unsigned long *stack)
152 {
153 }
154 
155 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
156 
157 /*
158  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
159  * kmemcache based allocator.
160  */
161 # if THREAD_SIZE >= PAGE_SIZE
162 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
163 						  int node)
164 {
165 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
166 					     THREAD_SIZE_ORDER);
167 
168 	return page ? page_address(page) : NULL;
169 }
170 
171 static inline void free_thread_stack(unsigned long *stack)
172 {
173 	__free_pages(virt_to_page(stack), THREAD_SIZE_ORDER);
174 }
175 # else
176 static struct kmem_cache *thread_stack_cache;
177 
178 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
179 						  int node)
180 {
181 	return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
182 }
183 
184 static void free_thread_stack(unsigned long *stack)
185 {
186 	kmem_cache_free(thread_stack_cache, stack);
187 }
188 
189 void thread_stack_cache_init(void)
190 {
191 	thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
192 					      THREAD_SIZE, 0, NULL);
193 	BUG_ON(thread_stack_cache == NULL);
194 }
195 # endif
196 #endif
197 
198 /* SLAB cache for signal_struct structures (tsk->signal) */
199 static struct kmem_cache *signal_cachep;
200 
201 /* SLAB cache for sighand_struct structures (tsk->sighand) */
202 struct kmem_cache *sighand_cachep;
203 
204 /* SLAB cache for files_struct structures (tsk->files) */
205 struct kmem_cache *files_cachep;
206 
207 /* SLAB cache for fs_struct structures (tsk->fs) */
208 struct kmem_cache *fs_cachep;
209 
210 /* SLAB cache for vm_area_struct structures */
211 struct kmem_cache *vm_area_cachep;
212 
213 /* SLAB cache for mm_struct structures (tsk->mm) */
214 static struct kmem_cache *mm_cachep;
215 
216 static void account_kernel_stack(unsigned long *stack, int account)
217 {
218 	/* All stack pages are in the same zone and belong to the same memcg. */
219 	struct page *first_page = virt_to_page(stack);
220 
221 	mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
222 			    THREAD_SIZE / 1024 * account);
223 
224 	memcg_kmem_update_page_stat(
225 		first_page, MEMCG_KERNEL_STACK_KB,
226 		account * (THREAD_SIZE / 1024));
227 }
228 
229 void free_task(struct task_struct *tsk)
230 {
231 	account_kernel_stack(tsk->stack, -1);
232 	arch_release_thread_stack(tsk->stack);
233 	free_thread_stack(tsk->stack);
234 	rt_mutex_debug_task_free(tsk);
235 	ftrace_graph_exit_task(tsk);
236 	put_seccomp_filter(tsk);
237 	arch_release_task_struct(tsk);
238 	free_task_struct(tsk);
239 }
240 EXPORT_SYMBOL(free_task);
241 
242 static inline void free_signal_struct(struct signal_struct *sig)
243 {
244 	taskstats_tgid_free(sig);
245 	sched_autogroup_exit(sig);
246 	kmem_cache_free(signal_cachep, sig);
247 }
248 
249 static inline void put_signal_struct(struct signal_struct *sig)
250 {
251 	if (atomic_dec_and_test(&sig->sigcnt))
252 		free_signal_struct(sig);
253 }
254 
255 void __put_task_struct(struct task_struct *tsk)
256 {
257 	WARN_ON(!tsk->exit_state);
258 	WARN_ON(atomic_read(&tsk->usage));
259 	WARN_ON(tsk == current);
260 
261 	cgroup_free(tsk);
262 	task_numa_free(tsk);
263 	security_task_free(tsk);
264 	exit_creds(tsk);
265 	delayacct_tsk_free(tsk);
266 	put_signal_struct(tsk->signal);
267 
268 	if (!profile_handoff_task(tsk))
269 		free_task(tsk);
270 }
271 EXPORT_SYMBOL_GPL(__put_task_struct);
272 
273 void __init __weak arch_task_cache_init(void) { }
274 
275 /*
276  * set_max_threads
277  */
278 static void set_max_threads(unsigned int max_threads_suggested)
279 {
280 	u64 threads;
281 
282 	/*
283 	 * The number of threads shall be limited such that the thread
284 	 * structures may only consume a small part of the available memory.
285 	 */
286 	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
287 		threads = MAX_THREADS;
288 	else
289 		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
290 				    (u64) THREAD_SIZE * 8UL);
291 
292 	if (threads > max_threads_suggested)
293 		threads = max_threads_suggested;
294 
295 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
296 }
297 
298 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
299 /* Initialized by the architecture: */
300 int arch_task_struct_size __read_mostly;
301 #endif
302 
303 void __init fork_init(void)
304 {
305 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
306 #ifndef ARCH_MIN_TASKALIGN
307 #define ARCH_MIN_TASKALIGN	L1_CACHE_BYTES
308 #endif
309 	/* create a slab on which task_structs can be allocated */
310 	task_struct_cachep = kmem_cache_create("task_struct",
311 			arch_task_struct_size, ARCH_MIN_TASKALIGN,
312 			SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
313 #endif
314 
315 	/* do the arch specific task caches init */
316 	arch_task_cache_init();
317 
318 	set_max_threads(MAX_THREADS);
319 
320 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
321 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
322 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
323 		init_task.signal->rlim[RLIMIT_NPROC];
324 }
325 
326 int __weak arch_dup_task_struct(struct task_struct *dst,
327 					       struct task_struct *src)
328 {
329 	*dst = *src;
330 	return 0;
331 }
332 
333 void set_task_stack_end_magic(struct task_struct *tsk)
334 {
335 	unsigned long *stackend;
336 
337 	stackend = end_of_stack(tsk);
338 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
339 }
340 
341 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
342 {
343 	struct task_struct *tsk;
344 	unsigned long *stack;
345 	int err;
346 
347 	if (node == NUMA_NO_NODE)
348 		node = tsk_fork_get_node(orig);
349 	tsk = alloc_task_struct_node(node);
350 	if (!tsk)
351 		return NULL;
352 
353 	stack = alloc_thread_stack_node(tsk, node);
354 	if (!stack)
355 		goto free_tsk;
356 
357 	err = arch_dup_task_struct(tsk, orig);
358 	if (err)
359 		goto free_stack;
360 
361 	tsk->stack = stack;
362 #ifdef CONFIG_SECCOMP
363 	/*
364 	 * We must handle setting up seccomp filters once we're under
365 	 * the sighand lock in case orig has changed between now and
366 	 * then. Until then, filter must be NULL to avoid messing up
367 	 * the usage counts on the error path calling free_task.
368 	 */
369 	tsk->seccomp.filter = NULL;
370 #endif
371 
372 	setup_thread_stack(tsk, orig);
373 	clear_user_return_notifier(tsk);
374 	clear_tsk_need_resched(tsk);
375 	set_task_stack_end_magic(tsk);
376 
377 #ifdef CONFIG_CC_STACKPROTECTOR
378 	tsk->stack_canary = get_random_int();
379 #endif
380 
381 	/*
382 	 * One for us, one for whoever does the "release_task()" (usually
383 	 * parent)
384 	 */
385 	atomic_set(&tsk->usage, 2);
386 #ifdef CONFIG_BLK_DEV_IO_TRACE
387 	tsk->btrace_seq = 0;
388 #endif
389 	tsk->splice_pipe = NULL;
390 	tsk->task_frag.page = NULL;
391 	tsk->wake_q.next = NULL;
392 
393 	account_kernel_stack(stack, 1);
394 
395 	kcov_task_init(tsk);
396 
397 	return tsk;
398 
399 free_stack:
400 	free_thread_stack(stack);
401 free_tsk:
402 	free_task_struct(tsk);
403 	return NULL;
404 }
405 
406 #ifdef CONFIG_MMU
407 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
408 {
409 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
410 	struct rb_node **rb_link, *rb_parent;
411 	int retval;
412 	unsigned long charge;
413 
414 	uprobe_start_dup_mmap();
415 	if (down_write_killable(&oldmm->mmap_sem)) {
416 		retval = -EINTR;
417 		goto fail_uprobe_end;
418 	}
419 	flush_cache_dup_mm(oldmm);
420 	uprobe_dup_mmap(oldmm, mm);
421 	/*
422 	 * Not linked in yet - no deadlock potential:
423 	 */
424 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
425 
426 	/* No ordering required: file already has been exposed. */
427 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
428 
429 	mm->total_vm = oldmm->total_vm;
430 	mm->data_vm = oldmm->data_vm;
431 	mm->exec_vm = oldmm->exec_vm;
432 	mm->stack_vm = oldmm->stack_vm;
433 
434 	rb_link = &mm->mm_rb.rb_node;
435 	rb_parent = NULL;
436 	pprev = &mm->mmap;
437 	retval = ksm_fork(mm, oldmm);
438 	if (retval)
439 		goto out;
440 	retval = khugepaged_fork(mm, oldmm);
441 	if (retval)
442 		goto out;
443 
444 	prev = NULL;
445 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
446 		struct file *file;
447 
448 		if (mpnt->vm_flags & VM_DONTCOPY) {
449 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
450 			continue;
451 		}
452 		charge = 0;
453 		if (mpnt->vm_flags & VM_ACCOUNT) {
454 			unsigned long len = vma_pages(mpnt);
455 
456 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
457 				goto fail_nomem;
458 			charge = len;
459 		}
460 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
461 		if (!tmp)
462 			goto fail_nomem;
463 		*tmp = *mpnt;
464 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
465 		retval = vma_dup_policy(mpnt, tmp);
466 		if (retval)
467 			goto fail_nomem_policy;
468 		tmp->vm_mm = mm;
469 		if (anon_vma_fork(tmp, mpnt))
470 			goto fail_nomem_anon_vma_fork;
471 		tmp->vm_flags &=
472 			~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
473 		tmp->vm_next = tmp->vm_prev = NULL;
474 		tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
475 		file = tmp->vm_file;
476 		if (file) {
477 			struct inode *inode = file_inode(file);
478 			struct address_space *mapping = file->f_mapping;
479 
480 			get_file(file);
481 			if (tmp->vm_flags & VM_DENYWRITE)
482 				atomic_dec(&inode->i_writecount);
483 			i_mmap_lock_write(mapping);
484 			if (tmp->vm_flags & VM_SHARED)
485 				atomic_inc(&mapping->i_mmap_writable);
486 			flush_dcache_mmap_lock(mapping);
487 			/* insert tmp into the share list, just after mpnt */
488 			vma_interval_tree_insert_after(tmp, mpnt,
489 					&mapping->i_mmap);
490 			flush_dcache_mmap_unlock(mapping);
491 			i_mmap_unlock_write(mapping);
492 		}
493 
494 		/*
495 		 * Clear hugetlb-related page reserves for children. This only
496 		 * affects MAP_PRIVATE mappings. Faults generated by the child
497 		 * are not guaranteed to succeed, even if read-only
498 		 */
499 		if (is_vm_hugetlb_page(tmp))
500 			reset_vma_resv_huge_pages(tmp);
501 
502 		/*
503 		 * Link in the new vma and copy the page table entries.
504 		 */
505 		*pprev = tmp;
506 		pprev = &tmp->vm_next;
507 		tmp->vm_prev = prev;
508 		prev = tmp;
509 
510 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
511 		rb_link = &tmp->vm_rb.rb_right;
512 		rb_parent = &tmp->vm_rb;
513 
514 		mm->map_count++;
515 		retval = copy_page_range(mm, oldmm, mpnt);
516 
517 		if (tmp->vm_ops && tmp->vm_ops->open)
518 			tmp->vm_ops->open(tmp);
519 
520 		if (retval)
521 			goto out;
522 	}
523 	/* a new mm has just been created */
524 	arch_dup_mmap(oldmm, mm);
525 	retval = 0;
526 out:
527 	up_write(&mm->mmap_sem);
528 	flush_tlb_mm(oldmm);
529 	up_write(&oldmm->mmap_sem);
530 fail_uprobe_end:
531 	uprobe_end_dup_mmap();
532 	return retval;
533 fail_nomem_anon_vma_fork:
534 	mpol_put(vma_policy(tmp));
535 fail_nomem_policy:
536 	kmem_cache_free(vm_area_cachep, tmp);
537 fail_nomem:
538 	retval = -ENOMEM;
539 	vm_unacct_memory(charge);
540 	goto out;
541 }
542 
543 static inline int mm_alloc_pgd(struct mm_struct *mm)
544 {
545 	mm->pgd = pgd_alloc(mm);
546 	if (unlikely(!mm->pgd))
547 		return -ENOMEM;
548 	return 0;
549 }
550 
551 static inline void mm_free_pgd(struct mm_struct *mm)
552 {
553 	pgd_free(mm, mm->pgd);
554 }
555 #else
556 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
557 {
558 	down_write(&oldmm->mmap_sem);
559 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
560 	up_write(&oldmm->mmap_sem);
561 	return 0;
562 }
563 #define mm_alloc_pgd(mm)	(0)
564 #define mm_free_pgd(mm)
565 #endif /* CONFIG_MMU */
566 
567 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
568 
569 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
570 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
571 
572 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
573 
574 static int __init coredump_filter_setup(char *s)
575 {
576 	default_dump_filter =
577 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
578 		MMF_DUMP_FILTER_MASK;
579 	return 1;
580 }
581 
582 __setup("coredump_filter=", coredump_filter_setup);
583 
584 #include <linux/init_task.h>
585 
586 static void mm_init_aio(struct mm_struct *mm)
587 {
588 #ifdef CONFIG_AIO
589 	spin_lock_init(&mm->ioctx_lock);
590 	mm->ioctx_table = NULL;
591 #endif
592 }
593 
594 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
595 {
596 #ifdef CONFIG_MEMCG
597 	mm->owner = p;
598 #endif
599 }
600 
601 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
602 {
603 	mm->mmap = NULL;
604 	mm->mm_rb = RB_ROOT;
605 	mm->vmacache_seqnum = 0;
606 	atomic_set(&mm->mm_users, 1);
607 	atomic_set(&mm->mm_count, 1);
608 	init_rwsem(&mm->mmap_sem);
609 	INIT_LIST_HEAD(&mm->mmlist);
610 	mm->core_state = NULL;
611 	atomic_long_set(&mm->nr_ptes, 0);
612 	mm_nr_pmds_init(mm);
613 	mm->map_count = 0;
614 	mm->locked_vm = 0;
615 	mm->pinned_vm = 0;
616 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
617 	spin_lock_init(&mm->page_table_lock);
618 	mm_init_cpumask(mm);
619 	mm_init_aio(mm);
620 	mm_init_owner(mm, p);
621 	mmu_notifier_mm_init(mm);
622 	clear_tlb_flush_pending(mm);
623 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
624 	mm->pmd_huge_pte = NULL;
625 #endif
626 
627 	if (current->mm) {
628 		mm->flags = current->mm->flags & MMF_INIT_MASK;
629 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
630 	} else {
631 		mm->flags = default_dump_filter;
632 		mm->def_flags = 0;
633 	}
634 
635 	if (mm_alloc_pgd(mm))
636 		goto fail_nopgd;
637 
638 	if (init_new_context(p, mm))
639 		goto fail_nocontext;
640 
641 	return mm;
642 
643 fail_nocontext:
644 	mm_free_pgd(mm);
645 fail_nopgd:
646 	free_mm(mm);
647 	return NULL;
648 }
649 
650 static void check_mm(struct mm_struct *mm)
651 {
652 	int i;
653 
654 	for (i = 0; i < NR_MM_COUNTERS; i++) {
655 		long x = atomic_long_read(&mm->rss_stat.count[i]);
656 
657 		if (unlikely(x))
658 			printk(KERN_ALERT "BUG: Bad rss-counter state "
659 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
660 	}
661 
662 	if (atomic_long_read(&mm->nr_ptes))
663 		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
664 				atomic_long_read(&mm->nr_ptes));
665 	if (mm_nr_pmds(mm))
666 		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
667 				mm_nr_pmds(mm));
668 
669 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
670 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
671 #endif
672 }
673 
674 /*
675  * Allocate and initialize an mm_struct.
676  */
677 struct mm_struct *mm_alloc(void)
678 {
679 	struct mm_struct *mm;
680 
681 	mm = allocate_mm();
682 	if (!mm)
683 		return NULL;
684 
685 	memset(mm, 0, sizeof(*mm));
686 	return mm_init(mm, current);
687 }
688 
689 /*
690  * Called when the last reference to the mm
691  * is dropped: either by a lazy thread or by
692  * mmput. Free the page directory and the mm.
693  */
694 void __mmdrop(struct mm_struct *mm)
695 {
696 	BUG_ON(mm == &init_mm);
697 	mm_free_pgd(mm);
698 	destroy_context(mm);
699 	mmu_notifier_mm_destroy(mm);
700 	check_mm(mm);
701 	free_mm(mm);
702 }
703 EXPORT_SYMBOL_GPL(__mmdrop);
704 
705 static inline void __mmput(struct mm_struct *mm)
706 {
707 	VM_BUG_ON(atomic_read(&mm->mm_users));
708 
709 	uprobe_clear_state(mm);
710 	exit_aio(mm);
711 	ksm_exit(mm);
712 	khugepaged_exit(mm); /* must run before exit_mmap */
713 	exit_mmap(mm);
714 	set_mm_exe_file(mm, NULL);
715 	if (!list_empty(&mm->mmlist)) {
716 		spin_lock(&mmlist_lock);
717 		list_del(&mm->mmlist);
718 		spin_unlock(&mmlist_lock);
719 	}
720 	if (mm->binfmt)
721 		module_put(mm->binfmt->module);
722 	mmdrop(mm);
723 }
724 
725 /*
726  * Decrement the use count and release all resources for an mm.
727  */
728 void mmput(struct mm_struct *mm)
729 {
730 	might_sleep();
731 
732 	if (atomic_dec_and_test(&mm->mm_users))
733 		__mmput(mm);
734 }
735 EXPORT_SYMBOL_GPL(mmput);
736 
737 #ifdef CONFIG_MMU
738 static void mmput_async_fn(struct work_struct *work)
739 {
740 	struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
741 	__mmput(mm);
742 }
743 
744 void mmput_async(struct mm_struct *mm)
745 {
746 	if (atomic_dec_and_test(&mm->mm_users)) {
747 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
748 		schedule_work(&mm->async_put_work);
749 	}
750 }
751 #endif
752 
753 /**
754  * set_mm_exe_file - change a reference to the mm's executable file
755  *
756  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
757  *
758  * Main users are mmput() and sys_execve(). Callers prevent concurrent
759  * invocations: in mmput() nobody alive left, in execve task is single
760  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
761  * mm->exe_file, but does so without using set_mm_exe_file() in order
762  * to do avoid the need for any locks.
763  */
764 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
765 {
766 	struct file *old_exe_file;
767 
768 	/*
769 	 * It is safe to dereference the exe_file without RCU as
770 	 * this function is only called if nobody else can access
771 	 * this mm -- see comment above for justification.
772 	 */
773 	old_exe_file = rcu_dereference_raw(mm->exe_file);
774 
775 	if (new_exe_file)
776 		get_file(new_exe_file);
777 	rcu_assign_pointer(mm->exe_file, new_exe_file);
778 	if (old_exe_file)
779 		fput(old_exe_file);
780 }
781 
782 /**
783  * get_mm_exe_file - acquire a reference to the mm's executable file
784  *
785  * Returns %NULL if mm has no associated executable file.
786  * User must release file via fput().
787  */
788 struct file *get_mm_exe_file(struct mm_struct *mm)
789 {
790 	struct file *exe_file;
791 
792 	rcu_read_lock();
793 	exe_file = rcu_dereference(mm->exe_file);
794 	if (exe_file && !get_file_rcu(exe_file))
795 		exe_file = NULL;
796 	rcu_read_unlock();
797 	return exe_file;
798 }
799 EXPORT_SYMBOL(get_mm_exe_file);
800 
801 /**
802  * get_task_exe_file - acquire a reference to the task's executable file
803  *
804  * Returns %NULL if task's mm (if any) has no associated executable file or
805  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
806  * User must release file via fput().
807  */
808 struct file *get_task_exe_file(struct task_struct *task)
809 {
810 	struct file *exe_file = NULL;
811 	struct mm_struct *mm;
812 
813 	task_lock(task);
814 	mm = task->mm;
815 	if (mm) {
816 		if (!(task->flags & PF_KTHREAD))
817 			exe_file = get_mm_exe_file(mm);
818 	}
819 	task_unlock(task);
820 	return exe_file;
821 }
822 EXPORT_SYMBOL(get_task_exe_file);
823 
824 /**
825  * get_task_mm - acquire a reference to the task's mm
826  *
827  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
828  * this kernel workthread has transiently adopted a user mm with use_mm,
829  * to do its AIO) is not set and if so returns a reference to it, after
830  * bumping up the use count.  User must release the mm via mmput()
831  * after use.  Typically used by /proc and ptrace.
832  */
833 struct mm_struct *get_task_mm(struct task_struct *task)
834 {
835 	struct mm_struct *mm;
836 
837 	task_lock(task);
838 	mm = task->mm;
839 	if (mm) {
840 		if (task->flags & PF_KTHREAD)
841 			mm = NULL;
842 		else
843 			atomic_inc(&mm->mm_users);
844 	}
845 	task_unlock(task);
846 	return mm;
847 }
848 EXPORT_SYMBOL_GPL(get_task_mm);
849 
850 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
851 {
852 	struct mm_struct *mm;
853 	int err;
854 
855 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
856 	if (err)
857 		return ERR_PTR(err);
858 
859 	mm = get_task_mm(task);
860 	if (mm && mm != current->mm &&
861 			!ptrace_may_access(task, mode)) {
862 		mmput(mm);
863 		mm = ERR_PTR(-EACCES);
864 	}
865 	mutex_unlock(&task->signal->cred_guard_mutex);
866 
867 	return mm;
868 }
869 
870 static void complete_vfork_done(struct task_struct *tsk)
871 {
872 	struct completion *vfork;
873 
874 	task_lock(tsk);
875 	vfork = tsk->vfork_done;
876 	if (likely(vfork)) {
877 		tsk->vfork_done = NULL;
878 		complete(vfork);
879 	}
880 	task_unlock(tsk);
881 }
882 
883 static int wait_for_vfork_done(struct task_struct *child,
884 				struct completion *vfork)
885 {
886 	int killed;
887 
888 	freezer_do_not_count();
889 	killed = wait_for_completion_killable(vfork);
890 	freezer_count();
891 
892 	if (killed) {
893 		task_lock(child);
894 		child->vfork_done = NULL;
895 		task_unlock(child);
896 	}
897 
898 	put_task_struct(child);
899 	return killed;
900 }
901 
902 /* Please note the differences between mmput and mm_release.
903  * mmput is called whenever we stop holding onto a mm_struct,
904  * error success whatever.
905  *
906  * mm_release is called after a mm_struct has been removed
907  * from the current process.
908  *
909  * This difference is important for error handling, when we
910  * only half set up a mm_struct for a new process and need to restore
911  * the old one.  Because we mmput the new mm_struct before
912  * restoring the old one. . .
913  * Eric Biederman 10 January 1998
914  */
915 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
916 {
917 	/* Get rid of any futexes when releasing the mm */
918 #ifdef CONFIG_FUTEX
919 	if (unlikely(tsk->robust_list)) {
920 		exit_robust_list(tsk);
921 		tsk->robust_list = NULL;
922 	}
923 #ifdef CONFIG_COMPAT
924 	if (unlikely(tsk->compat_robust_list)) {
925 		compat_exit_robust_list(tsk);
926 		tsk->compat_robust_list = NULL;
927 	}
928 #endif
929 	if (unlikely(!list_empty(&tsk->pi_state_list)))
930 		exit_pi_state_list(tsk);
931 #endif
932 
933 	uprobe_free_utask(tsk);
934 
935 	/* Get rid of any cached register state */
936 	deactivate_mm(tsk, mm);
937 
938 	/*
939 	 * Signal userspace if we're not exiting with a core dump
940 	 * because we want to leave the value intact for debugging
941 	 * purposes.
942 	 */
943 	if (tsk->clear_child_tid) {
944 		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
945 		    atomic_read(&mm->mm_users) > 1) {
946 			/*
947 			 * We don't check the error code - if userspace has
948 			 * not set up a proper pointer then tough luck.
949 			 */
950 			put_user(0, tsk->clear_child_tid);
951 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
952 					1, NULL, NULL, 0);
953 		}
954 		tsk->clear_child_tid = NULL;
955 	}
956 
957 	/*
958 	 * All done, finally we can wake up parent and return this mm to him.
959 	 * Also kthread_stop() uses this completion for synchronization.
960 	 */
961 	if (tsk->vfork_done)
962 		complete_vfork_done(tsk);
963 }
964 
965 /*
966  * Allocate a new mm structure and copy contents from the
967  * mm structure of the passed in task structure.
968  */
969 static struct mm_struct *dup_mm(struct task_struct *tsk)
970 {
971 	struct mm_struct *mm, *oldmm = current->mm;
972 	int err;
973 
974 	mm = allocate_mm();
975 	if (!mm)
976 		goto fail_nomem;
977 
978 	memcpy(mm, oldmm, sizeof(*mm));
979 
980 	if (!mm_init(mm, tsk))
981 		goto fail_nomem;
982 
983 	err = dup_mmap(mm, oldmm);
984 	if (err)
985 		goto free_pt;
986 
987 	mm->hiwater_rss = get_mm_rss(mm);
988 	mm->hiwater_vm = mm->total_vm;
989 
990 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
991 		goto free_pt;
992 
993 	return mm;
994 
995 free_pt:
996 	/* don't put binfmt in mmput, we haven't got module yet */
997 	mm->binfmt = NULL;
998 	mmput(mm);
999 
1000 fail_nomem:
1001 	return NULL;
1002 }
1003 
1004 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1005 {
1006 	struct mm_struct *mm, *oldmm;
1007 	int retval;
1008 
1009 	tsk->min_flt = tsk->maj_flt = 0;
1010 	tsk->nvcsw = tsk->nivcsw = 0;
1011 #ifdef CONFIG_DETECT_HUNG_TASK
1012 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1013 #endif
1014 
1015 	tsk->mm = NULL;
1016 	tsk->active_mm = NULL;
1017 
1018 	/*
1019 	 * Are we cloning a kernel thread?
1020 	 *
1021 	 * We need to steal a active VM for that..
1022 	 */
1023 	oldmm = current->mm;
1024 	if (!oldmm)
1025 		return 0;
1026 
1027 	/* initialize the new vmacache entries */
1028 	vmacache_flush(tsk);
1029 
1030 	if (clone_flags & CLONE_VM) {
1031 		atomic_inc(&oldmm->mm_users);
1032 		mm = oldmm;
1033 		goto good_mm;
1034 	}
1035 
1036 	retval = -ENOMEM;
1037 	mm = dup_mm(tsk);
1038 	if (!mm)
1039 		goto fail_nomem;
1040 
1041 good_mm:
1042 	tsk->mm = mm;
1043 	tsk->active_mm = mm;
1044 	return 0;
1045 
1046 fail_nomem:
1047 	return retval;
1048 }
1049 
1050 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1051 {
1052 	struct fs_struct *fs = current->fs;
1053 	if (clone_flags & CLONE_FS) {
1054 		/* tsk->fs is already what we want */
1055 		spin_lock(&fs->lock);
1056 		if (fs->in_exec) {
1057 			spin_unlock(&fs->lock);
1058 			return -EAGAIN;
1059 		}
1060 		fs->users++;
1061 		spin_unlock(&fs->lock);
1062 		return 0;
1063 	}
1064 	tsk->fs = copy_fs_struct(fs);
1065 	if (!tsk->fs)
1066 		return -ENOMEM;
1067 	return 0;
1068 }
1069 
1070 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1071 {
1072 	struct files_struct *oldf, *newf;
1073 	int error = 0;
1074 
1075 	/*
1076 	 * A background process may not have any files ...
1077 	 */
1078 	oldf = current->files;
1079 	if (!oldf)
1080 		goto out;
1081 
1082 	if (clone_flags & CLONE_FILES) {
1083 		atomic_inc(&oldf->count);
1084 		goto out;
1085 	}
1086 
1087 	newf = dup_fd(oldf, &error);
1088 	if (!newf)
1089 		goto out;
1090 
1091 	tsk->files = newf;
1092 	error = 0;
1093 out:
1094 	return error;
1095 }
1096 
1097 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1098 {
1099 #ifdef CONFIG_BLOCK
1100 	struct io_context *ioc = current->io_context;
1101 	struct io_context *new_ioc;
1102 
1103 	if (!ioc)
1104 		return 0;
1105 	/*
1106 	 * Share io context with parent, if CLONE_IO is set
1107 	 */
1108 	if (clone_flags & CLONE_IO) {
1109 		ioc_task_link(ioc);
1110 		tsk->io_context = ioc;
1111 	} else if (ioprio_valid(ioc->ioprio)) {
1112 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1113 		if (unlikely(!new_ioc))
1114 			return -ENOMEM;
1115 
1116 		new_ioc->ioprio = ioc->ioprio;
1117 		put_io_context(new_ioc);
1118 	}
1119 #endif
1120 	return 0;
1121 }
1122 
1123 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1124 {
1125 	struct sighand_struct *sig;
1126 
1127 	if (clone_flags & CLONE_SIGHAND) {
1128 		atomic_inc(&current->sighand->count);
1129 		return 0;
1130 	}
1131 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1132 	rcu_assign_pointer(tsk->sighand, sig);
1133 	if (!sig)
1134 		return -ENOMEM;
1135 
1136 	atomic_set(&sig->count, 1);
1137 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1138 	return 0;
1139 }
1140 
1141 void __cleanup_sighand(struct sighand_struct *sighand)
1142 {
1143 	if (atomic_dec_and_test(&sighand->count)) {
1144 		signalfd_cleanup(sighand);
1145 		/*
1146 		 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1147 		 * without an RCU grace period, see __lock_task_sighand().
1148 		 */
1149 		kmem_cache_free(sighand_cachep, sighand);
1150 	}
1151 }
1152 
1153 /*
1154  * Initialize POSIX timer handling for a thread group.
1155  */
1156 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1157 {
1158 	unsigned long cpu_limit;
1159 
1160 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1161 	if (cpu_limit != RLIM_INFINITY) {
1162 		sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1163 		sig->cputimer.running = true;
1164 	}
1165 
1166 	/* The timer lists. */
1167 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1168 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1169 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1170 }
1171 
1172 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1173 {
1174 	struct signal_struct *sig;
1175 
1176 	if (clone_flags & CLONE_THREAD)
1177 		return 0;
1178 
1179 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1180 	tsk->signal = sig;
1181 	if (!sig)
1182 		return -ENOMEM;
1183 
1184 	sig->nr_threads = 1;
1185 	atomic_set(&sig->live, 1);
1186 	atomic_set(&sig->sigcnt, 1);
1187 
1188 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1189 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1190 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1191 
1192 	init_waitqueue_head(&sig->wait_chldexit);
1193 	sig->curr_target = tsk;
1194 	init_sigpending(&sig->shared_pending);
1195 	INIT_LIST_HEAD(&sig->posix_timers);
1196 	seqlock_init(&sig->stats_lock);
1197 	prev_cputime_init(&sig->prev_cputime);
1198 
1199 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1200 	sig->real_timer.function = it_real_fn;
1201 
1202 	task_lock(current->group_leader);
1203 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1204 	task_unlock(current->group_leader);
1205 
1206 	posix_cpu_timers_init_group(sig);
1207 
1208 	tty_audit_fork(sig);
1209 	sched_autogroup_fork(sig);
1210 
1211 	sig->oom_score_adj = current->signal->oom_score_adj;
1212 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1213 
1214 	sig->has_child_subreaper = current->signal->has_child_subreaper ||
1215 				   current->signal->is_child_subreaper;
1216 
1217 	mutex_init(&sig->cred_guard_mutex);
1218 
1219 	return 0;
1220 }
1221 
1222 static void copy_seccomp(struct task_struct *p)
1223 {
1224 #ifdef CONFIG_SECCOMP
1225 	/*
1226 	 * Must be called with sighand->lock held, which is common to
1227 	 * all threads in the group. Holding cred_guard_mutex is not
1228 	 * needed because this new task is not yet running and cannot
1229 	 * be racing exec.
1230 	 */
1231 	assert_spin_locked(&current->sighand->siglock);
1232 
1233 	/* Ref-count the new filter user, and assign it. */
1234 	get_seccomp_filter(current);
1235 	p->seccomp = current->seccomp;
1236 
1237 	/*
1238 	 * Explicitly enable no_new_privs here in case it got set
1239 	 * between the task_struct being duplicated and holding the
1240 	 * sighand lock. The seccomp state and nnp must be in sync.
1241 	 */
1242 	if (task_no_new_privs(current))
1243 		task_set_no_new_privs(p);
1244 
1245 	/*
1246 	 * If the parent gained a seccomp mode after copying thread
1247 	 * flags and between before we held the sighand lock, we have
1248 	 * to manually enable the seccomp thread flag here.
1249 	 */
1250 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1251 		set_tsk_thread_flag(p, TIF_SECCOMP);
1252 #endif
1253 }
1254 
1255 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1256 {
1257 	current->clear_child_tid = tidptr;
1258 
1259 	return task_pid_vnr(current);
1260 }
1261 
1262 static void rt_mutex_init_task(struct task_struct *p)
1263 {
1264 	raw_spin_lock_init(&p->pi_lock);
1265 #ifdef CONFIG_RT_MUTEXES
1266 	p->pi_waiters = RB_ROOT;
1267 	p->pi_waiters_leftmost = NULL;
1268 	p->pi_blocked_on = NULL;
1269 #endif
1270 }
1271 
1272 /*
1273  * Initialize POSIX timer handling for a single task.
1274  */
1275 static void posix_cpu_timers_init(struct task_struct *tsk)
1276 {
1277 	tsk->cputime_expires.prof_exp = 0;
1278 	tsk->cputime_expires.virt_exp = 0;
1279 	tsk->cputime_expires.sched_exp = 0;
1280 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1281 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1282 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1283 }
1284 
1285 static inline void
1286 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1287 {
1288 	 task->pids[type].pid = pid;
1289 }
1290 
1291 /*
1292  * This creates a new process as a copy of the old one,
1293  * but does not actually start it yet.
1294  *
1295  * It copies the registers, and all the appropriate
1296  * parts of the process environment (as per the clone
1297  * flags). The actual kick-off is left to the caller.
1298  */
1299 static struct task_struct *copy_process(unsigned long clone_flags,
1300 					unsigned long stack_start,
1301 					unsigned long stack_size,
1302 					int __user *child_tidptr,
1303 					struct pid *pid,
1304 					int trace,
1305 					unsigned long tls,
1306 					int node)
1307 {
1308 	int retval;
1309 	struct task_struct *p;
1310 
1311 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1312 		return ERR_PTR(-EINVAL);
1313 
1314 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1315 		return ERR_PTR(-EINVAL);
1316 
1317 	/*
1318 	 * Thread groups must share signals as well, and detached threads
1319 	 * can only be started up within the thread group.
1320 	 */
1321 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1322 		return ERR_PTR(-EINVAL);
1323 
1324 	/*
1325 	 * Shared signal handlers imply shared VM. By way of the above,
1326 	 * thread groups also imply shared VM. Blocking this case allows
1327 	 * for various simplifications in other code.
1328 	 */
1329 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1330 		return ERR_PTR(-EINVAL);
1331 
1332 	/*
1333 	 * Siblings of global init remain as zombies on exit since they are
1334 	 * not reaped by their parent (swapper). To solve this and to avoid
1335 	 * multi-rooted process trees, prevent global and container-inits
1336 	 * from creating siblings.
1337 	 */
1338 	if ((clone_flags & CLONE_PARENT) &&
1339 				current->signal->flags & SIGNAL_UNKILLABLE)
1340 		return ERR_PTR(-EINVAL);
1341 
1342 	/*
1343 	 * If the new process will be in a different pid or user namespace
1344 	 * do not allow it to share a thread group with the forking task.
1345 	 */
1346 	if (clone_flags & CLONE_THREAD) {
1347 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1348 		    (task_active_pid_ns(current) !=
1349 				current->nsproxy->pid_ns_for_children))
1350 			return ERR_PTR(-EINVAL);
1351 	}
1352 
1353 	retval = security_task_create(clone_flags);
1354 	if (retval)
1355 		goto fork_out;
1356 
1357 	retval = -ENOMEM;
1358 	p = dup_task_struct(current, node);
1359 	if (!p)
1360 		goto fork_out;
1361 
1362 	ftrace_graph_init_task(p);
1363 
1364 	rt_mutex_init_task(p);
1365 
1366 #ifdef CONFIG_PROVE_LOCKING
1367 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1368 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1369 #endif
1370 	retval = -EAGAIN;
1371 	if (atomic_read(&p->real_cred->user->processes) >=
1372 			task_rlimit(p, RLIMIT_NPROC)) {
1373 		if (p->real_cred->user != INIT_USER &&
1374 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1375 			goto bad_fork_free;
1376 	}
1377 	current->flags &= ~PF_NPROC_EXCEEDED;
1378 
1379 	retval = copy_creds(p, clone_flags);
1380 	if (retval < 0)
1381 		goto bad_fork_free;
1382 
1383 	/*
1384 	 * If multiple threads are within copy_process(), then this check
1385 	 * triggers too late. This doesn't hurt, the check is only there
1386 	 * to stop root fork bombs.
1387 	 */
1388 	retval = -EAGAIN;
1389 	if (nr_threads >= max_threads)
1390 		goto bad_fork_cleanup_count;
1391 
1392 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1393 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1394 	p->flags |= PF_FORKNOEXEC;
1395 	INIT_LIST_HEAD(&p->children);
1396 	INIT_LIST_HEAD(&p->sibling);
1397 	rcu_copy_process(p);
1398 	p->vfork_done = NULL;
1399 	spin_lock_init(&p->alloc_lock);
1400 
1401 	init_sigpending(&p->pending);
1402 
1403 	p->utime = p->stime = p->gtime = 0;
1404 	p->utimescaled = p->stimescaled = 0;
1405 	prev_cputime_init(&p->prev_cputime);
1406 
1407 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1408 	seqcount_init(&p->vtime_seqcount);
1409 	p->vtime_snap = 0;
1410 	p->vtime_snap_whence = VTIME_INACTIVE;
1411 #endif
1412 
1413 #if defined(SPLIT_RSS_COUNTING)
1414 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1415 #endif
1416 
1417 	p->default_timer_slack_ns = current->timer_slack_ns;
1418 
1419 	task_io_accounting_init(&p->ioac);
1420 	acct_clear_integrals(p);
1421 
1422 	posix_cpu_timers_init(p);
1423 
1424 	p->start_time = ktime_get_ns();
1425 	p->real_start_time = ktime_get_boot_ns();
1426 	p->io_context = NULL;
1427 	p->audit_context = NULL;
1428 	cgroup_fork(p);
1429 #ifdef CONFIG_NUMA
1430 	p->mempolicy = mpol_dup(p->mempolicy);
1431 	if (IS_ERR(p->mempolicy)) {
1432 		retval = PTR_ERR(p->mempolicy);
1433 		p->mempolicy = NULL;
1434 		goto bad_fork_cleanup_threadgroup_lock;
1435 	}
1436 #endif
1437 #ifdef CONFIG_CPUSETS
1438 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1439 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1440 	seqcount_init(&p->mems_allowed_seq);
1441 #endif
1442 #ifdef CONFIG_TRACE_IRQFLAGS
1443 	p->irq_events = 0;
1444 	p->hardirqs_enabled = 0;
1445 	p->hardirq_enable_ip = 0;
1446 	p->hardirq_enable_event = 0;
1447 	p->hardirq_disable_ip = _THIS_IP_;
1448 	p->hardirq_disable_event = 0;
1449 	p->softirqs_enabled = 1;
1450 	p->softirq_enable_ip = _THIS_IP_;
1451 	p->softirq_enable_event = 0;
1452 	p->softirq_disable_ip = 0;
1453 	p->softirq_disable_event = 0;
1454 	p->hardirq_context = 0;
1455 	p->softirq_context = 0;
1456 #endif
1457 
1458 	p->pagefault_disabled = 0;
1459 
1460 #ifdef CONFIG_LOCKDEP
1461 	p->lockdep_depth = 0; /* no locks held yet */
1462 	p->curr_chain_key = 0;
1463 	p->lockdep_recursion = 0;
1464 #endif
1465 
1466 #ifdef CONFIG_DEBUG_MUTEXES
1467 	p->blocked_on = NULL; /* not blocked yet */
1468 #endif
1469 #ifdef CONFIG_BCACHE
1470 	p->sequential_io	= 0;
1471 	p->sequential_io_avg	= 0;
1472 #endif
1473 
1474 	/* Perform scheduler related setup. Assign this task to a CPU. */
1475 	retval = sched_fork(clone_flags, p);
1476 	if (retval)
1477 		goto bad_fork_cleanup_policy;
1478 
1479 	retval = perf_event_init_task(p);
1480 	if (retval)
1481 		goto bad_fork_cleanup_policy;
1482 	retval = audit_alloc(p);
1483 	if (retval)
1484 		goto bad_fork_cleanup_perf;
1485 	/* copy all the process information */
1486 	shm_init_task(p);
1487 	retval = copy_semundo(clone_flags, p);
1488 	if (retval)
1489 		goto bad_fork_cleanup_audit;
1490 	retval = copy_files(clone_flags, p);
1491 	if (retval)
1492 		goto bad_fork_cleanup_semundo;
1493 	retval = copy_fs(clone_flags, p);
1494 	if (retval)
1495 		goto bad_fork_cleanup_files;
1496 	retval = copy_sighand(clone_flags, p);
1497 	if (retval)
1498 		goto bad_fork_cleanup_fs;
1499 	retval = copy_signal(clone_flags, p);
1500 	if (retval)
1501 		goto bad_fork_cleanup_sighand;
1502 	retval = copy_mm(clone_flags, p);
1503 	if (retval)
1504 		goto bad_fork_cleanup_signal;
1505 	retval = copy_namespaces(clone_flags, p);
1506 	if (retval)
1507 		goto bad_fork_cleanup_mm;
1508 	retval = copy_io(clone_flags, p);
1509 	if (retval)
1510 		goto bad_fork_cleanup_namespaces;
1511 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1512 	if (retval)
1513 		goto bad_fork_cleanup_io;
1514 
1515 	if (pid != &init_struct_pid) {
1516 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1517 		if (IS_ERR(pid)) {
1518 			retval = PTR_ERR(pid);
1519 			goto bad_fork_cleanup_thread;
1520 		}
1521 	}
1522 
1523 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1524 	/*
1525 	 * Clear TID on mm_release()?
1526 	 */
1527 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1528 #ifdef CONFIG_BLOCK
1529 	p->plug = NULL;
1530 #endif
1531 #ifdef CONFIG_FUTEX
1532 	p->robust_list = NULL;
1533 #ifdef CONFIG_COMPAT
1534 	p->compat_robust_list = NULL;
1535 #endif
1536 	INIT_LIST_HEAD(&p->pi_state_list);
1537 	p->pi_state_cache = NULL;
1538 #endif
1539 	/*
1540 	 * sigaltstack should be cleared when sharing the same VM
1541 	 */
1542 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1543 		sas_ss_reset(p);
1544 
1545 	/*
1546 	 * Syscall tracing and stepping should be turned off in the
1547 	 * child regardless of CLONE_PTRACE.
1548 	 */
1549 	user_disable_single_step(p);
1550 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1551 #ifdef TIF_SYSCALL_EMU
1552 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1553 #endif
1554 	clear_all_latency_tracing(p);
1555 
1556 	/* ok, now we should be set up.. */
1557 	p->pid = pid_nr(pid);
1558 	if (clone_flags & CLONE_THREAD) {
1559 		p->exit_signal = -1;
1560 		p->group_leader = current->group_leader;
1561 		p->tgid = current->tgid;
1562 	} else {
1563 		if (clone_flags & CLONE_PARENT)
1564 			p->exit_signal = current->group_leader->exit_signal;
1565 		else
1566 			p->exit_signal = (clone_flags & CSIGNAL);
1567 		p->group_leader = p;
1568 		p->tgid = p->pid;
1569 	}
1570 
1571 	p->nr_dirtied = 0;
1572 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1573 	p->dirty_paused_when = 0;
1574 
1575 	p->pdeath_signal = 0;
1576 	INIT_LIST_HEAD(&p->thread_group);
1577 	p->task_works = NULL;
1578 
1579 	threadgroup_change_begin(current);
1580 	/*
1581 	 * Ensure that the cgroup subsystem policies allow the new process to be
1582 	 * forked. It should be noted the the new process's css_set can be changed
1583 	 * between here and cgroup_post_fork() if an organisation operation is in
1584 	 * progress.
1585 	 */
1586 	retval = cgroup_can_fork(p);
1587 	if (retval)
1588 		goto bad_fork_free_pid;
1589 
1590 	/*
1591 	 * Make it visible to the rest of the system, but dont wake it up yet.
1592 	 * Need tasklist lock for parent etc handling!
1593 	 */
1594 	write_lock_irq(&tasklist_lock);
1595 
1596 	/* CLONE_PARENT re-uses the old parent */
1597 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1598 		p->real_parent = current->real_parent;
1599 		p->parent_exec_id = current->parent_exec_id;
1600 	} else {
1601 		p->real_parent = current;
1602 		p->parent_exec_id = current->self_exec_id;
1603 	}
1604 
1605 	spin_lock(&current->sighand->siglock);
1606 
1607 	/*
1608 	 * Copy seccomp details explicitly here, in case they were changed
1609 	 * before holding sighand lock.
1610 	 */
1611 	copy_seccomp(p);
1612 
1613 	/*
1614 	 * Process group and session signals need to be delivered to just the
1615 	 * parent before the fork or both the parent and the child after the
1616 	 * fork. Restart if a signal comes in before we add the new process to
1617 	 * it's process group.
1618 	 * A fatal signal pending means that current will exit, so the new
1619 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1620 	*/
1621 	recalc_sigpending();
1622 	if (signal_pending(current)) {
1623 		spin_unlock(&current->sighand->siglock);
1624 		write_unlock_irq(&tasklist_lock);
1625 		retval = -ERESTARTNOINTR;
1626 		goto bad_fork_cancel_cgroup;
1627 	}
1628 
1629 	if (likely(p->pid)) {
1630 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1631 
1632 		init_task_pid(p, PIDTYPE_PID, pid);
1633 		if (thread_group_leader(p)) {
1634 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1635 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1636 
1637 			if (is_child_reaper(pid)) {
1638 				ns_of_pid(pid)->child_reaper = p;
1639 				p->signal->flags |= SIGNAL_UNKILLABLE;
1640 			}
1641 
1642 			p->signal->leader_pid = pid;
1643 			p->signal->tty = tty_kref_get(current->signal->tty);
1644 			list_add_tail(&p->sibling, &p->real_parent->children);
1645 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1646 			attach_pid(p, PIDTYPE_PGID);
1647 			attach_pid(p, PIDTYPE_SID);
1648 			__this_cpu_inc(process_counts);
1649 		} else {
1650 			current->signal->nr_threads++;
1651 			atomic_inc(&current->signal->live);
1652 			atomic_inc(&current->signal->sigcnt);
1653 			list_add_tail_rcu(&p->thread_group,
1654 					  &p->group_leader->thread_group);
1655 			list_add_tail_rcu(&p->thread_node,
1656 					  &p->signal->thread_head);
1657 		}
1658 		attach_pid(p, PIDTYPE_PID);
1659 		nr_threads++;
1660 	}
1661 
1662 	total_forks++;
1663 	spin_unlock(&current->sighand->siglock);
1664 	syscall_tracepoint_update(p);
1665 	write_unlock_irq(&tasklist_lock);
1666 
1667 	proc_fork_connector(p);
1668 	cgroup_post_fork(p);
1669 	threadgroup_change_end(current);
1670 	perf_event_fork(p);
1671 
1672 	trace_task_newtask(p, clone_flags);
1673 	uprobe_copy_process(p, clone_flags);
1674 
1675 	return p;
1676 
1677 bad_fork_cancel_cgroup:
1678 	cgroup_cancel_fork(p);
1679 bad_fork_free_pid:
1680 	threadgroup_change_end(current);
1681 	if (pid != &init_struct_pid)
1682 		free_pid(pid);
1683 bad_fork_cleanup_thread:
1684 	exit_thread(p);
1685 bad_fork_cleanup_io:
1686 	if (p->io_context)
1687 		exit_io_context(p);
1688 bad_fork_cleanup_namespaces:
1689 	exit_task_namespaces(p);
1690 bad_fork_cleanup_mm:
1691 	if (p->mm)
1692 		mmput(p->mm);
1693 bad_fork_cleanup_signal:
1694 	if (!(clone_flags & CLONE_THREAD))
1695 		free_signal_struct(p->signal);
1696 bad_fork_cleanup_sighand:
1697 	__cleanup_sighand(p->sighand);
1698 bad_fork_cleanup_fs:
1699 	exit_fs(p); /* blocking */
1700 bad_fork_cleanup_files:
1701 	exit_files(p); /* blocking */
1702 bad_fork_cleanup_semundo:
1703 	exit_sem(p);
1704 bad_fork_cleanup_audit:
1705 	audit_free(p);
1706 bad_fork_cleanup_perf:
1707 	perf_event_free_task(p);
1708 bad_fork_cleanup_policy:
1709 #ifdef CONFIG_NUMA
1710 	mpol_put(p->mempolicy);
1711 bad_fork_cleanup_threadgroup_lock:
1712 #endif
1713 	delayacct_tsk_free(p);
1714 bad_fork_cleanup_count:
1715 	atomic_dec(&p->cred->user->processes);
1716 	exit_creds(p);
1717 bad_fork_free:
1718 	free_task(p);
1719 fork_out:
1720 	return ERR_PTR(retval);
1721 }
1722 
1723 static inline void init_idle_pids(struct pid_link *links)
1724 {
1725 	enum pid_type type;
1726 
1727 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1728 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1729 		links[type].pid = &init_struct_pid;
1730 	}
1731 }
1732 
1733 struct task_struct *fork_idle(int cpu)
1734 {
1735 	struct task_struct *task;
1736 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1737 			    cpu_to_node(cpu));
1738 	if (!IS_ERR(task)) {
1739 		init_idle_pids(task->pids);
1740 		init_idle(task, cpu);
1741 	}
1742 
1743 	return task;
1744 }
1745 
1746 /*
1747  *  Ok, this is the main fork-routine.
1748  *
1749  * It copies the process, and if successful kick-starts
1750  * it and waits for it to finish using the VM if required.
1751  */
1752 long _do_fork(unsigned long clone_flags,
1753 	      unsigned long stack_start,
1754 	      unsigned long stack_size,
1755 	      int __user *parent_tidptr,
1756 	      int __user *child_tidptr,
1757 	      unsigned long tls)
1758 {
1759 	struct task_struct *p;
1760 	int trace = 0;
1761 	long nr;
1762 
1763 	/*
1764 	 * Determine whether and which event to report to ptracer.  When
1765 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
1766 	 * requested, no event is reported; otherwise, report if the event
1767 	 * for the type of forking is enabled.
1768 	 */
1769 	if (!(clone_flags & CLONE_UNTRACED)) {
1770 		if (clone_flags & CLONE_VFORK)
1771 			trace = PTRACE_EVENT_VFORK;
1772 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
1773 			trace = PTRACE_EVENT_CLONE;
1774 		else
1775 			trace = PTRACE_EVENT_FORK;
1776 
1777 		if (likely(!ptrace_event_enabled(current, trace)))
1778 			trace = 0;
1779 	}
1780 
1781 	p = copy_process(clone_flags, stack_start, stack_size,
1782 			 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1783 	/*
1784 	 * Do this prior waking up the new thread - the thread pointer
1785 	 * might get invalid after that point, if the thread exits quickly.
1786 	 */
1787 	if (!IS_ERR(p)) {
1788 		struct completion vfork;
1789 		struct pid *pid;
1790 
1791 		trace_sched_process_fork(current, p);
1792 
1793 		pid = get_task_pid(p, PIDTYPE_PID);
1794 		nr = pid_vnr(pid);
1795 
1796 		if (clone_flags & CLONE_PARENT_SETTID)
1797 			put_user(nr, parent_tidptr);
1798 
1799 		if (clone_flags & CLONE_VFORK) {
1800 			p->vfork_done = &vfork;
1801 			init_completion(&vfork);
1802 			get_task_struct(p);
1803 		}
1804 
1805 		wake_up_new_task(p);
1806 
1807 		/* forking complete and child started to run, tell ptracer */
1808 		if (unlikely(trace))
1809 			ptrace_event_pid(trace, pid);
1810 
1811 		if (clone_flags & CLONE_VFORK) {
1812 			if (!wait_for_vfork_done(p, &vfork))
1813 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1814 		}
1815 
1816 		put_pid(pid);
1817 	} else {
1818 		nr = PTR_ERR(p);
1819 	}
1820 	return nr;
1821 }
1822 
1823 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1824 /* For compatibility with architectures that call do_fork directly rather than
1825  * using the syscall entry points below. */
1826 long do_fork(unsigned long clone_flags,
1827 	      unsigned long stack_start,
1828 	      unsigned long stack_size,
1829 	      int __user *parent_tidptr,
1830 	      int __user *child_tidptr)
1831 {
1832 	return _do_fork(clone_flags, stack_start, stack_size,
1833 			parent_tidptr, child_tidptr, 0);
1834 }
1835 #endif
1836 
1837 /*
1838  * Create a kernel thread.
1839  */
1840 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1841 {
1842 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1843 		(unsigned long)arg, NULL, NULL, 0);
1844 }
1845 
1846 #ifdef __ARCH_WANT_SYS_FORK
1847 SYSCALL_DEFINE0(fork)
1848 {
1849 #ifdef CONFIG_MMU
1850 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1851 #else
1852 	/* can not support in nommu mode */
1853 	return -EINVAL;
1854 #endif
1855 }
1856 #endif
1857 
1858 #ifdef __ARCH_WANT_SYS_VFORK
1859 SYSCALL_DEFINE0(vfork)
1860 {
1861 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
1862 			0, NULL, NULL, 0);
1863 }
1864 #endif
1865 
1866 #ifdef __ARCH_WANT_SYS_CLONE
1867 #ifdef CONFIG_CLONE_BACKWARDS
1868 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1869 		 int __user *, parent_tidptr,
1870 		 unsigned long, tls,
1871 		 int __user *, child_tidptr)
1872 #elif defined(CONFIG_CLONE_BACKWARDS2)
1873 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
1874 		 int __user *, parent_tidptr,
1875 		 int __user *, child_tidptr,
1876 		 unsigned long, tls)
1877 #elif defined(CONFIG_CLONE_BACKWARDS3)
1878 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
1879 		int, stack_size,
1880 		int __user *, parent_tidptr,
1881 		int __user *, child_tidptr,
1882 		unsigned long, tls)
1883 #else
1884 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
1885 		 int __user *, parent_tidptr,
1886 		 int __user *, child_tidptr,
1887 		 unsigned long, tls)
1888 #endif
1889 {
1890 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
1891 }
1892 #endif
1893 
1894 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
1895 #define ARCH_MIN_MMSTRUCT_ALIGN 0
1896 #endif
1897 
1898 static void sighand_ctor(void *data)
1899 {
1900 	struct sighand_struct *sighand = data;
1901 
1902 	spin_lock_init(&sighand->siglock);
1903 	init_waitqueue_head(&sighand->signalfd_wqh);
1904 }
1905 
1906 void __init proc_caches_init(void)
1907 {
1908 	sighand_cachep = kmem_cache_create("sighand_cache",
1909 			sizeof(struct sighand_struct), 0,
1910 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
1911 			SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
1912 	signal_cachep = kmem_cache_create("signal_cache",
1913 			sizeof(struct signal_struct), 0,
1914 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1915 			NULL);
1916 	files_cachep = kmem_cache_create("files_cache",
1917 			sizeof(struct files_struct), 0,
1918 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1919 			NULL);
1920 	fs_cachep = kmem_cache_create("fs_cache",
1921 			sizeof(struct fs_struct), 0,
1922 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1923 			NULL);
1924 	/*
1925 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
1926 	 * whole struct cpumask for the OFFSTACK case. We could change
1927 	 * this to *only* allocate as much of it as required by the
1928 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
1929 	 * is at the end of the structure, exactly for that reason.
1930 	 */
1931 	mm_cachep = kmem_cache_create("mm_struct",
1932 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1933 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
1934 			NULL);
1935 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
1936 	mmap_init();
1937 	nsproxy_cache_init();
1938 }
1939 
1940 /*
1941  * Check constraints on flags passed to the unshare system call.
1942  */
1943 static int check_unshare_flags(unsigned long unshare_flags)
1944 {
1945 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
1946 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
1947 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
1948 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
1949 		return -EINVAL;
1950 	/*
1951 	 * Not implemented, but pretend it works if there is nothing
1952 	 * to unshare.  Note that unsharing the address space or the
1953 	 * signal handlers also need to unshare the signal queues (aka
1954 	 * CLONE_THREAD).
1955 	 */
1956 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
1957 		if (!thread_group_empty(current))
1958 			return -EINVAL;
1959 	}
1960 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
1961 		if (atomic_read(&current->sighand->count) > 1)
1962 			return -EINVAL;
1963 	}
1964 	if (unshare_flags & CLONE_VM) {
1965 		if (!current_is_single_threaded())
1966 			return -EINVAL;
1967 	}
1968 
1969 	return 0;
1970 }
1971 
1972 /*
1973  * Unshare the filesystem structure if it is being shared
1974  */
1975 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1976 {
1977 	struct fs_struct *fs = current->fs;
1978 
1979 	if (!(unshare_flags & CLONE_FS) || !fs)
1980 		return 0;
1981 
1982 	/* don't need lock here; in the worst case we'll do useless copy */
1983 	if (fs->users == 1)
1984 		return 0;
1985 
1986 	*new_fsp = copy_fs_struct(fs);
1987 	if (!*new_fsp)
1988 		return -ENOMEM;
1989 
1990 	return 0;
1991 }
1992 
1993 /*
1994  * Unshare file descriptor table if it is being shared
1995  */
1996 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
1997 {
1998 	struct files_struct *fd = current->files;
1999 	int error = 0;
2000 
2001 	if ((unshare_flags & CLONE_FILES) &&
2002 	    (fd && atomic_read(&fd->count) > 1)) {
2003 		*new_fdp = dup_fd(fd, &error);
2004 		if (!*new_fdp)
2005 			return error;
2006 	}
2007 
2008 	return 0;
2009 }
2010 
2011 /*
2012  * unshare allows a process to 'unshare' part of the process
2013  * context which was originally shared using clone.  copy_*
2014  * functions used by do_fork() cannot be used here directly
2015  * because they modify an inactive task_struct that is being
2016  * constructed. Here we are modifying the current, active,
2017  * task_struct.
2018  */
2019 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2020 {
2021 	struct fs_struct *fs, *new_fs = NULL;
2022 	struct files_struct *fd, *new_fd = NULL;
2023 	struct cred *new_cred = NULL;
2024 	struct nsproxy *new_nsproxy = NULL;
2025 	int do_sysvsem = 0;
2026 	int err;
2027 
2028 	/*
2029 	 * If unsharing a user namespace must also unshare the thread group
2030 	 * and unshare the filesystem root and working directories.
2031 	 */
2032 	if (unshare_flags & CLONE_NEWUSER)
2033 		unshare_flags |= CLONE_THREAD | CLONE_FS;
2034 	/*
2035 	 * If unsharing vm, must also unshare signal handlers.
2036 	 */
2037 	if (unshare_flags & CLONE_VM)
2038 		unshare_flags |= CLONE_SIGHAND;
2039 	/*
2040 	 * If unsharing a signal handlers, must also unshare the signal queues.
2041 	 */
2042 	if (unshare_flags & CLONE_SIGHAND)
2043 		unshare_flags |= CLONE_THREAD;
2044 	/*
2045 	 * If unsharing namespace, must also unshare filesystem information.
2046 	 */
2047 	if (unshare_flags & CLONE_NEWNS)
2048 		unshare_flags |= CLONE_FS;
2049 
2050 	err = check_unshare_flags(unshare_flags);
2051 	if (err)
2052 		goto bad_unshare_out;
2053 	/*
2054 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2055 	 * to a new ipc namespace, the semaphore arrays from the old
2056 	 * namespace are unreachable.
2057 	 */
2058 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2059 		do_sysvsem = 1;
2060 	err = unshare_fs(unshare_flags, &new_fs);
2061 	if (err)
2062 		goto bad_unshare_out;
2063 	err = unshare_fd(unshare_flags, &new_fd);
2064 	if (err)
2065 		goto bad_unshare_cleanup_fs;
2066 	err = unshare_userns(unshare_flags, &new_cred);
2067 	if (err)
2068 		goto bad_unshare_cleanup_fd;
2069 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2070 					 new_cred, new_fs);
2071 	if (err)
2072 		goto bad_unshare_cleanup_cred;
2073 
2074 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2075 		if (do_sysvsem) {
2076 			/*
2077 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2078 			 */
2079 			exit_sem(current);
2080 		}
2081 		if (unshare_flags & CLONE_NEWIPC) {
2082 			/* Orphan segments in old ns (see sem above). */
2083 			exit_shm(current);
2084 			shm_init_task(current);
2085 		}
2086 
2087 		if (new_nsproxy)
2088 			switch_task_namespaces(current, new_nsproxy);
2089 
2090 		task_lock(current);
2091 
2092 		if (new_fs) {
2093 			fs = current->fs;
2094 			spin_lock(&fs->lock);
2095 			current->fs = new_fs;
2096 			if (--fs->users)
2097 				new_fs = NULL;
2098 			else
2099 				new_fs = fs;
2100 			spin_unlock(&fs->lock);
2101 		}
2102 
2103 		if (new_fd) {
2104 			fd = current->files;
2105 			current->files = new_fd;
2106 			new_fd = fd;
2107 		}
2108 
2109 		task_unlock(current);
2110 
2111 		if (new_cred) {
2112 			/* Install the new user namespace */
2113 			commit_creds(new_cred);
2114 			new_cred = NULL;
2115 		}
2116 	}
2117 
2118 bad_unshare_cleanup_cred:
2119 	if (new_cred)
2120 		put_cred(new_cred);
2121 bad_unshare_cleanup_fd:
2122 	if (new_fd)
2123 		put_files_struct(new_fd);
2124 
2125 bad_unshare_cleanup_fs:
2126 	if (new_fs)
2127 		free_fs_struct(new_fs);
2128 
2129 bad_unshare_out:
2130 	return err;
2131 }
2132 
2133 /*
2134  *	Helper to unshare the files of the current task.
2135  *	We don't want to expose copy_files internals to
2136  *	the exec layer of the kernel.
2137  */
2138 
2139 int unshare_files(struct files_struct **displaced)
2140 {
2141 	struct task_struct *task = current;
2142 	struct files_struct *copy = NULL;
2143 	int error;
2144 
2145 	error = unshare_fd(CLONE_FILES, &copy);
2146 	if (error || !copy) {
2147 		*displaced = NULL;
2148 		return error;
2149 	}
2150 	*displaced = task->files;
2151 	task_lock(task);
2152 	task->files = copy;
2153 	task_unlock(task);
2154 	return 0;
2155 }
2156 
2157 int sysctl_max_threads(struct ctl_table *table, int write,
2158 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2159 {
2160 	struct ctl_table t;
2161 	int ret;
2162 	int threads = max_threads;
2163 	int min = MIN_THREADS;
2164 	int max = MAX_THREADS;
2165 
2166 	t = *table;
2167 	t.data = &threads;
2168 	t.extra1 = &min;
2169 	t.extra2 = &max;
2170 
2171 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2172 	if (ret || !write)
2173 		return ret;
2174 
2175 	set_max_threads(threads);
2176 
2177 	return 0;
2178 }
2179