xref: /linux-6.15/kernel/fork.c (revision 3f07c014)
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/sched/autogroup.h>
16 #include <linux/sched/mm.h>
17 #include <linux/sched/coredump.h>
18 #include <linux/init.h>
19 #include <linux/unistd.h>
20 #include <linux/module.h>
21 #include <linux/vmalloc.h>
22 #include <linux/completion.h>
23 #include <linux/personality.h>
24 #include <linux/mempolicy.h>
25 #include <linux/sem.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/iocontext.h>
29 #include <linux/key.h>
30 #include <linux/binfmts.h>
31 #include <linux/mman.h>
32 #include <linux/mmu_notifier.h>
33 #include <linux/fs.h>
34 #include <linux/mm.h>
35 #include <linux/vmacache.h>
36 #include <linux/nsproxy.h>
37 #include <linux/capability.h>
38 #include <linux/cpu.h>
39 #include <linux/cgroup.h>
40 #include <linux/security.h>
41 #include <linux/hugetlb.h>
42 #include <linux/seccomp.h>
43 #include <linux/swap.h>
44 #include <linux/syscalls.h>
45 #include <linux/jiffies.h>
46 #include <linux/futex.h>
47 #include <linux/compat.h>
48 #include <linux/kthread.h>
49 #include <linux/task_io_accounting_ops.h>
50 #include <linux/rcupdate.h>
51 #include <linux/ptrace.h>
52 #include <linux/mount.h>
53 #include <linux/audit.h>
54 #include <linux/memcontrol.h>
55 #include <linux/ftrace.h>
56 #include <linux/proc_fs.h>
57 #include <linux/profile.h>
58 #include <linux/rmap.h>
59 #include <linux/ksm.h>
60 #include <linux/acct.h>
61 #include <linux/userfaultfd_k.h>
62 #include <linux/tsacct_kern.h>
63 #include <linux/cn_proc.h>
64 #include <linux/freezer.h>
65 #include <linux/delayacct.h>
66 #include <linux/taskstats_kern.h>
67 #include <linux/random.h>
68 #include <linux/tty.h>
69 #include <linux/blkdev.h>
70 #include <linux/fs_struct.h>
71 #include <linux/magic.h>
72 #include <linux/perf_event.h>
73 #include <linux/posix-timers.h>
74 #include <linux/user-return-notifier.h>
75 #include <linux/oom.h>
76 #include <linux/khugepaged.h>
77 #include <linux/signalfd.h>
78 #include <linux/uprobes.h>
79 #include <linux/aio.h>
80 #include <linux/compiler.h>
81 #include <linux/sysctl.h>
82 #include <linux/kcov.h>
83 
84 #include <asm/pgtable.h>
85 #include <asm/pgalloc.h>
86 #include <linux/uaccess.h>
87 #include <asm/mmu_context.h>
88 #include <asm/cacheflush.h>
89 #include <asm/tlbflush.h>
90 
91 #include <trace/events/sched.h>
92 
93 #define CREATE_TRACE_POINTS
94 #include <trace/events/task.h>
95 
96 /*
97  * Minimum number of threads to boot the kernel
98  */
99 #define MIN_THREADS 20
100 
101 /*
102  * Maximum number of threads
103  */
104 #define MAX_THREADS FUTEX_TID_MASK
105 
106 /*
107  * Protected counters by write_lock_irq(&tasklist_lock)
108  */
109 unsigned long total_forks;	/* Handle normal Linux uptimes. */
110 int nr_threads;			/* The idle threads do not count.. */
111 
112 int max_threads;		/* tunable limit on nr_threads */
113 
114 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
115 
116 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
117 
118 #ifdef CONFIG_PROVE_RCU
119 int lockdep_tasklist_lock_is_held(void)
120 {
121 	return lockdep_is_held(&tasklist_lock);
122 }
123 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
124 #endif /* #ifdef CONFIG_PROVE_RCU */
125 
126 int nr_processes(void)
127 {
128 	int cpu;
129 	int total = 0;
130 
131 	for_each_possible_cpu(cpu)
132 		total += per_cpu(process_counts, cpu);
133 
134 	return total;
135 }
136 
137 void __weak arch_release_task_struct(struct task_struct *tsk)
138 {
139 }
140 
141 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
142 static struct kmem_cache *task_struct_cachep;
143 
144 static inline struct task_struct *alloc_task_struct_node(int node)
145 {
146 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
147 }
148 
149 static inline void free_task_struct(struct task_struct *tsk)
150 {
151 	kmem_cache_free(task_struct_cachep, tsk);
152 }
153 #endif
154 
155 void __weak arch_release_thread_stack(unsigned long *stack)
156 {
157 }
158 
159 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
160 
161 /*
162  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
163  * kmemcache based allocator.
164  */
165 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
166 
167 #ifdef CONFIG_VMAP_STACK
168 /*
169  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
170  * flush.  Try to minimize the number of calls by caching stacks.
171  */
172 #define NR_CACHED_STACKS 2
173 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
174 #endif
175 
176 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
177 {
178 #ifdef CONFIG_VMAP_STACK
179 	void *stack;
180 	int i;
181 
182 	local_irq_disable();
183 	for (i = 0; i < NR_CACHED_STACKS; i++) {
184 		struct vm_struct *s = this_cpu_read(cached_stacks[i]);
185 
186 		if (!s)
187 			continue;
188 		this_cpu_write(cached_stacks[i], NULL);
189 
190 		tsk->stack_vm_area = s;
191 		local_irq_enable();
192 		return s->addr;
193 	}
194 	local_irq_enable();
195 
196 	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
197 				     VMALLOC_START, VMALLOC_END,
198 				     THREADINFO_GFP | __GFP_HIGHMEM,
199 				     PAGE_KERNEL,
200 				     0, node, __builtin_return_address(0));
201 
202 	/*
203 	 * We can't call find_vm_area() in interrupt context, and
204 	 * free_thread_stack() can be called in interrupt context,
205 	 * so cache the vm_struct.
206 	 */
207 	if (stack)
208 		tsk->stack_vm_area = find_vm_area(stack);
209 	return stack;
210 #else
211 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
212 					     THREAD_SIZE_ORDER);
213 
214 	return page ? page_address(page) : NULL;
215 #endif
216 }
217 
218 static inline void free_thread_stack(struct task_struct *tsk)
219 {
220 #ifdef CONFIG_VMAP_STACK
221 	if (task_stack_vm_area(tsk)) {
222 		unsigned long flags;
223 		int i;
224 
225 		local_irq_save(flags);
226 		for (i = 0; i < NR_CACHED_STACKS; i++) {
227 			if (this_cpu_read(cached_stacks[i]))
228 				continue;
229 
230 			this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
231 			local_irq_restore(flags);
232 			return;
233 		}
234 		local_irq_restore(flags);
235 
236 		vfree_atomic(tsk->stack);
237 		return;
238 	}
239 #endif
240 
241 	__free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
242 }
243 # else
244 static struct kmem_cache *thread_stack_cache;
245 
246 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
247 						  int node)
248 {
249 	return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
250 }
251 
252 static void free_thread_stack(struct task_struct *tsk)
253 {
254 	kmem_cache_free(thread_stack_cache, tsk->stack);
255 }
256 
257 void thread_stack_cache_init(void)
258 {
259 	thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
260 					      THREAD_SIZE, 0, NULL);
261 	BUG_ON(thread_stack_cache == NULL);
262 }
263 # endif
264 #endif
265 
266 /* SLAB cache for signal_struct structures (tsk->signal) */
267 static struct kmem_cache *signal_cachep;
268 
269 /* SLAB cache for sighand_struct structures (tsk->sighand) */
270 struct kmem_cache *sighand_cachep;
271 
272 /* SLAB cache for files_struct structures (tsk->files) */
273 struct kmem_cache *files_cachep;
274 
275 /* SLAB cache for fs_struct structures (tsk->fs) */
276 struct kmem_cache *fs_cachep;
277 
278 /* SLAB cache for vm_area_struct structures */
279 struct kmem_cache *vm_area_cachep;
280 
281 /* SLAB cache for mm_struct structures (tsk->mm) */
282 static struct kmem_cache *mm_cachep;
283 
284 static void account_kernel_stack(struct task_struct *tsk, int account)
285 {
286 	void *stack = task_stack_page(tsk);
287 	struct vm_struct *vm = task_stack_vm_area(tsk);
288 
289 	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
290 
291 	if (vm) {
292 		int i;
293 
294 		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
295 
296 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
297 			mod_zone_page_state(page_zone(vm->pages[i]),
298 					    NR_KERNEL_STACK_KB,
299 					    PAGE_SIZE / 1024 * account);
300 		}
301 
302 		/* All stack pages belong to the same memcg. */
303 		memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
304 					    account * (THREAD_SIZE / 1024));
305 	} else {
306 		/*
307 		 * All stack pages are in the same zone and belong to the
308 		 * same memcg.
309 		 */
310 		struct page *first_page = virt_to_page(stack);
311 
312 		mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
313 				    THREAD_SIZE / 1024 * account);
314 
315 		memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
316 					    account * (THREAD_SIZE / 1024));
317 	}
318 }
319 
320 static void release_task_stack(struct task_struct *tsk)
321 {
322 	if (WARN_ON(tsk->state != TASK_DEAD))
323 		return;  /* Better to leak the stack than to free prematurely */
324 
325 	account_kernel_stack(tsk, -1);
326 	arch_release_thread_stack(tsk->stack);
327 	free_thread_stack(tsk);
328 	tsk->stack = NULL;
329 #ifdef CONFIG_VMAP_STACK
330 	tsk->stack_vm_area = NULL;
331 #endif
332 }
333 
334 #ifdef CONFIG_THREAD_INFO_IN_TASK
335 void put_task_stack(struct task_struct *tsk)
336 {
337 	if (atomic_dec_and_test(&tsk->stack_refcount))
338 		release_task_stack(tsk);
339 }
340 #endif
341 
342 void free_task(struct task_struct *tsk)
343 {
344 #ifndef CONFIG_THREAD_INFO_IN_TASK
345 	/*
346 	 * The task is finally done with both the stack and thread_info,
347 	 * so free both.
348 	 */
349 	release_task_stack(tsk);
350 #else
351 	/*
352 	 * If the task had a separate stack allocation, it should be gone
353 	 * by now.
354 	 */
355 	WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
356 #endif
357 	rt_mutex_debug_task_free(tsk);
358 	ftrace_graph_exit_task(tsk);
359 	put_seccomp_filter(tsk);
360 	arch_release_task_struct(tsk);
361 	if (tsk->flags & PF_KTHREAD)
362 		free_kthread_struct(tsk);
363 	free_task_struct(tsk);
364 }
365 EXPORT_SYMBOL(free_task);
366 
367 static inline void free_signal_struct(struct signal_struct *sig)
368 {
369 	taskstats_tgid_free(sig);
370 	sched_autogroup_exit(sig);
371 	/*
372 	 * __mmdrop is not safe to call from softirq context on x86 due to
373 	 * pgd_dtor so postpone it to the async context
374 	 */
375 	if (sig->oom_mm)
376 		mmdrop_async(sig->oom_mm);
377 	kmem_cache_free(signal_cachep, sig);
378 }
379 
380 static inline void put_signal_struct(struct signal_struct *sig)
381 {
382 	if (atomic_dec_and_test(&sig->sigcnt))
383 		free_signal_struct(sig);
384 }
385 
386 void __put_task_struct(struct task_struct *tsk)
387 {
388 	WARN_ON(!tsk->exit_state);
389 	WARN_ON(atomic_read(&tsk->usage));
390 	WARN_ON(tsk == current);
391 
392 	cgroup_free(tsk);
393 	task_numa_free(tsk);
394 	security_task_free(tsk);
395 	exit_creds(tsk);
396 	delayacct_tsk_free(tsk);
397 	put_signal_struct(tsk->signal);
398 
399 	if (!profile_handoff_task(tsk))
400 		free_task(tsk);
401 }
402 EXPORT_SYMBOL_GPL(__put_task_struct);
403 
404 void __init __weak arch_task_cache_init(void) { }
405 
406 /*
407  * set_max_threads
408  */
409 static void set_max_threads(unsigned int max_threads_suggested)
410 {
411 	u64 threads;
412 
413 	/*
414 	 * The number of threads shall be limited such that the thread
415 	 * structures may only consume a small part of the available memory.
416 	 */
417 	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
418 		threads = MAX_THREADS;
419 	else
420 		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
421 				    (u64) THREAD_SIZE * 8UL);
422 
423 	if (threads > max_threads_suggested)
424 		threads = max_threads_suggested;
425 
426 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
427 }
428 
429 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
430 /* Initialized by the architecture: */
431 int arch_task_struct_size __read_mostly;
432 #endif
433 
434 void __init fork_init(void)
435 {
436 	int i;
437 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
438 #ifndef ARCH_MIN_TASKALIGN
439 #define ARCH_MIN_TASKALIGN	0
440 #endif
441 	int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
442 
443 	/* create a slab on which task_structs can be allocated */
444 	task_struct_cachep = kmem_cache_create("task_struct",
445 			arch_task_struct_size, align,
446 			SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
447 #endif
448 
449 	/* do the arch specific task caches init */
450 	arch_task_cache_init();
451 
452 	set_max_threads(MAX_THREADS);
453 
454 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
455 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
456 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
457 		init_task.signal->rlim[RLIMIT_NPROC];
458 
459 	for (i = 0; i < UCOUNT_COUNTS; i++) {
460 		init_user_ns.ucount_max[i] = max_threads/2;
461 	}
462 }
463 
464 int __weak arch_dup_task_struct(struct task_struct *dst,
465 					       struct task_struct *src)
466 {
467 	*dst = *src;
468 	return 0;
469 }
470 
471 void set_task_stack_end_magic(struct task_struct *tsk)
472 {
473 	unsigned long *stackend;
474 
475 	stackend = end_of_stack(tsk);
476 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
477 }
478 
479 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
480 {
481 	struct task_struct *tsk;
482 	unsigned long *stack;
483 	struct vm_struct *stack_vm_area;
484 	int err;
485 
486 	if (node == NUMA_NO_NODE)
487 		node = tsk_fork_get_node(orig);
488 	tsk = alloc_task_struct_node(node);
489 	if (!tsk)
490 		return NULL;
491 
492 	stack = alloc_thread_stack_node(tsk, node);
493 	if (!stack)
494 		goto free_tsk;
495 
496 	stack_vm_area = task_stack_vm_area(tsk);
497 
498 	err = arch_dup_task_struct(tsk, orig);
499 
500 	/*
501 	 * arch_dup_task_struct() clobbers the stack-related fields.  Make
502 	 * sure they're properly initialized before using any stack-related
503 	 * functions again.
504 	 */
505 	tsk->stack = stack;
506 #ifdef CONFIG_VMAP_STACK
507 	tsk->stack_vm_area = stack_vm_area;
508 #endif
509 #ifdef CONFIG_THREAD_INFO_IN_TASK
510 	atomic_set(&tsk->stack_refcount, 1);
511 #endif
512 
513 	if (err)
514 		goto free_stack;
515 
516 #ifdef CONFIG_SECCOMP
517 	/*
518 	 * We must handle setting up seccomp filters once we're under
519 	 * the sighand lock in case orig has changed between now and
520 	 * then. Until then, filter must be NULL to avoid messing up
521 	 * the usage counts on the error path calling free_task.
522 	 */
523 	tsk->seccomp.filter = NULL;
524 #endif
525 
526 	setup_thread_stack(tsk, orig);
527 	clear_user_return_notifier(tsk);
528 	clear_tsk_need_resched(tsk);
529 	set_task_stack_end_magic(tsk);
530 
531 #ifdef CONFIG_CC_STACKPROTECTOR
532 	tsk->stack_canary = get_random_int();
533 #endif
534 
535 	/*
536 	 * One for us, one for whoever does the "release_task()" (usually
537 	 * parent)
538 	 */
539 	atomic_set(&tsk->usage, 2);
540 #ifdef CONFIG_BLK_DEV_IO_TRACE
541 	tsk->btrace_seq = 0;
542 #endif
543 	tsk->splice_pipe = NULL;
544 	tsk->task_frag.page = NULL;
545 	tsk->wake_q.next = NULL;
546 
547 	account_kernel_stack(tsk, 1);
548 
549 	kcov_task_init(tsk);
550 
551 	return tsk;
552 
553 free_stack:
554 	free_thread_stack(tsk);
555 free_tsk:
556 	free_task_struct(tsk);
557 	return NULL;
558 }
559 
560 #ifdef CONFIG_MMU
561 static __latent_entropy int dup_mmap(struct mm_struct *mm,
562 					struct mm_struct *oldmm)
563 {
564 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
565 	struct rb_node **rb_link, *rb_parent;
566 	int retval;
567 	unsigned long charge;
568 	LIST_HEAD(uf);
569 
570 	uprobe_start_dup_mmap();
571 	if (down_write_killable(&oldmm->mmap_sem)) {
572 		retval = -EINTR;
573 		goto fail_uprobe_end;
574 	}
575 	flush_cache_dup_mm(oldmm);
576 	uprobe_dup_mmap(oldmm, mm);
577 	/*
578 	 * Not linked in yet - no deadlock potential:
579 	 */
580 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
581 
582 	/* No ordering required: file already has been exposed. */
583 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
584 
585 	mm->total_vm = oldmm->total_vm;
586 	mm->data_vm = oldmm->data_vm;
587 	mm->exec_vm = oldmm->exec_vm;
588 	mm->stack_vm = oldmm->stack_vm;
589 
590 	rb_link = &mm->mm_rb.rb_node;
591 	rb_parent = NULL;
592 	pprev = &mm->mmap;
593 	retval = ksm_fork(mm, oldmm);
594 	if (retval)
595 		goto out;
596 	retval = khugepaged_fork(mm, oldmm);
597 	if (retval)
598 		goto out;
599 
600 	prev = NULL;
601 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
602 		struct file *file;
603 
604 		if (mpnt->vm_flags & VM_DONTCOPY) {
605 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
606 			continue;
607 		}
608 		charge = 0;
609 		if (mpnt->vm_flags & VM_ACCOUNT) {
610 			unsigned long len = vma_pages(mpnt);
611 
612 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
613 				goto fail_nomem;
614 			charge = len;
615 		}
616 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
617 		if (!tmp)
618 			goto fail_nomem;
619 		*tmp = *mpnt;
620 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
621 		retval = vma_dup_policy(mpnt, tmp);
622 		if (retval)
623 			goto fail_nomem_policy;
624 		tmp->vm_mm = mm;
625 		retval = dup_userfaultfd(tmp, &uf);
626 		if (retval)
627 			goto fail_nomem_anon_vma_fork;
628 		if (anon_vma_fork(tmp, mpnt))
629 			goto fail_nomem_anon_vma_fork;
630 		tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
631 		tmp->vm_next = tmp->vm_prev = NULL;
632 		file = tmp->vm_file;
633 		if (file) {
634 			struct inode *inode = file_inode(file);
635 			struct address_space *mapping = file->f_mapping;
636 
637 			get_file(file);
638 			if (tmp->vm_flags & VM_DENYWRITE)
639 				atomic_dec(&inode->i_writecount);
640 			i_mmap_lock_write(mapping);
641 			if (tmp->vm_flags & VM_SHARED)
642 				atomic_inc(&mapping->i_mmap_writable);
643 			flush_dcache_mmap_lock(mapping);
644 			/* insert tmp into the share list, just after mpnt */
645 			vma_interval_tree_insert_after(tmp, mpnt,
646 					&mapping->i_mmap);
647 			flush_dcache_mmap_unlock(mapping);
648 			i_mmap_unlock_write(mapping);
649 		}
650 
651 		/*
652 		 * Clear hugetlb-related page reserves for children. This only
653 		 * affects MAP_PRIVATE mappings. Faults generated by the child
654 		 * are not guaranteed to succeed, even if read-only
655 		 */
656 		if (is_vm_hugetlb_page(tmp))
657 			reset_vma_resv_huge_pages(tmp);
658 
659 		/*
660 		 * Link in the new vma and copy the page table entries.
661 		 */
662 		*pprev = tmp;
663 		pprev = &tmp->vm_next;
664 		tmp->vm_prev = prev;
665 		prev = tmp;
666 
667 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
668 		rb_link = &tmp->vm_rb.rb_right;
669 		rb_parent = &tmp->vm_rb;
670 
671 		mm->map_count++;
672 		retval = copy_page_range(mm, oldmm, mpnt);
673 
674 		if (tmp->vm_ops && tmp->vm_ops->open)
675 			tmp->vm_ops->open(tmp);
676 
677 		if (retval)
678 			goto out;
679 	}
680 	/* a new mm has just been created */
681 	arch_dup_mmap(oldmm, mm);
682 	retval = 0;
683 out:
684 	up_write(&mm->mmap_sem);
685 	flush_tlb_mm(oldmm);
686 	up_write(&oldmm->mmap_sem);
687 	dup_userfaultfd_complete(&uf);
688 fail_uprobe_end:
689 	uprobe_end_dup_mmap();
690 	return retval;
691 fail_nomem_anon_vma_fork:
692 	mpol_put(vma_policy(tmp));
693 fail_nomem_policy:
694 	kmem_cache_free(vm_area_cachep, tmp);
695 fail_nomem:
696 	retval = -ENOMEM;
697 	vm_unacct_memory(charge);
698 	goto out;
699 }
700 
701 static inline int mm_alloc_pgd(struct mm_struct *mm)
702 {
703 	mm->pgd = pgd_alloc(mm);
704 	if (unlikely(!mm->pgd))
705 		return -ENOMEM;
706 	return 0;
707 }
708 
709 static inline void mm_free_pgd(struct mm_struct *mm)
710 {
711 	pgd_free(mm, mm->pgd);
712 }
713 #else
714 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
715 {
716 	down_write(&oldmm->mmap_sem);
717 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
718 	up_write(&oldmm->mmap_sem);
719 	return 0;
720 }
721 #define mm_alloc_pgd(mm)	(0)
722 #define mm_free_pgd(mm)
723 #endif /* CONFIG_MMU */
724 
725 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
726 
727 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
728 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
729 
730 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
731 
732 static int __init coredump_filter_setup(char *s)
733 {
734 	default_dump_filter =
735 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
736 		MMF_DUMP_FILTER_MASK;
737 	return 1;
738 }
739 
740 __setup("coredump_filter=", coredump_filter_setup);
741 
742 #include <linux/init_task.h>
743 
744 static void mm_init_aio(struct mm_struct *mm)
745 {
746 #ifdef CONFIG_AIO
747 	spin_lock_init(&mm->ioctx_lock);
748 	mm->ioctx_table = NULL;
749 #endif
750 }
751 
752 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
753 {
754 #ifdef CONFIG_MEMCG
755 	mm->owner = p;
756 #endif
757 }
758 
759 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
760 	struct user_namespace *user_ns)
761 {
762 	mm->mmap = NULL;
763 	mm->mm_rb = RB_ROOT;
764 	mm->vmacache_seqnum = 0;
765 	atomic_set(&mm->mm_users, 1);
766 	atomic_set(&mm->mm_count, 1);
767 	init_rwsem(&mm->mmap_sem);
768 	INIT_LIST_HEAD(&mm->mmlist);
769 	mm->core_state = NULL;
770 	atomic_long_set(&mm->nr_ptes, 0);
771 	mm_nr_pmds_init(mm);
772 	mm->map_count = 0;
773 	mm->locked_vm = 0;
774 	mm->pinned_vm = 0;
775 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
776 	spin_lock_init(&mm->page_table_lock);
777 	mm_init_cpumask(mm);
778 	mm_init_aio(mm);
779 	mm_init_owner(mm, p);
780 	mmu_notifier_mm_init(mm);
781 	clear_tlb_flush_pending(mm);
782 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
783 	mm->pmd_huge_pte = NULL;
784 #endif
785 
786 	if (current->mm) {
787 		mm->flags = current->mm->flags & MMF_INIT_MASK;
788 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
789 	} else {
790 		mm->flags = default_dump_filter;
791 		mm->def_flags = 0;
792 	}
793 
794 	if (mm_alloc_pgd(mm))
795 		goto fail_nopgd;
796 
797 	if (init_new_context(p, mm))
798 		goto fail_nocontext;
799 
800 	mm->user_ns = get_user_ns(user_ns);
801 	return mm;
802 
803 fail_nocontext:
804 	mm_free_pgd(mm);
805 fail_nopgd:
806 	free_mm(mm);
807 	return NULL;
808 }
809 
810 static void check_mm(struct mm_struct *mm)
811 {
812 	int i;
813 
814 	for (i = 0; i < NR_MM_COUNTERS; i++) {
815 		long x = atomic_long_read(&mm->rss_stat.count[i]);
816 
817 		if (unlikely(x))
818 			printk(KERN_ALERT "BUG: Bad rss-counter state "
819 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
820 	}
821 
822 	if (atomic_long_read(&mm->nr_ptes))
823 		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
824 				atomic_long_read(&mm->nr_ptes));
825 	if (mm_nr_pmds(mm))
826 		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
827 				mm_nr_pmds(mm));
828 
829 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
830 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
831 #endif
832 }
833 
834 /*
835  * Allocate and initialize an mm_struct.
836  */
837 struct mm_struct *mm_alloc(void)
838 {
839 	struct mm_struct *mm;
840 
841 	mm = allocate_mm();
842 	if (!mm)
843 		return NULL;
844 
845 	memset(mm, 0, sizeof(*mm));
846 	return mm_init(mm, current, current_user_ns());
847 }
848 
849 /*
850  * Called when the last reference to the mm
851  * is dropped: either by a lazy thread or by
852  * mmput. Free the page directory and the mm.
853  */
854 void __mmdrop(struct mm_struct *mm)
855 {
856 	BUG_ON(mm == &init_mm);
857 	mm_free_pgd(mm);
858 	destroy_context(mm);
859 	mmu_notifier_mm_destroy(mm);
860 	check_mm(mm);
861 	put_user_ns(mm->user_ns);
862 	free_mm(mm);
863 }
864 EXPORT_SYMBOL_GPL(__mmdrop);
865 
866 static inline void __mmput(struct mm_struct *mm)
867 {
868 	VM_BUG_ON(atomic_read(&mm->mm_users));
869 
870 	uprobe_clear_state(mm);
871 	exit_aio(mm);
872 	ksm_exit(mm);
873 	khugepaged_exit(mm); /* must run before exit_mmap */
874 	exit_mmap(mm);
875 	mm_put_huge_zero_page(mm);
876 	set_mm_exe_file(mm, NULL);
877 	if (!list_empty(&mm->mmlist)) {
878 		spin_lock(&mmlist_lock);
879 		list_del(&mm->mmlist);
880 		spin_unlock(&mmlist_lock);
881 	}
882 	if (mm->binfmt)
883 		module_put(mm->binfmt->module);
884 	set_bit(MMF_OOM_SKIP, &mm->flags);
885 	mmdrop(mm);
886 }
887 
888 /*
889  * Decrement the use count and release all resources for an mm.
890  */
891 void mmput(struct mm_struct *mm)
892 {
893 	might_sleep();
894 
895 	if (atomic_dec_and_test(&mm->mm_users))
896 		__mmput(mm);
897 }
898 EXPORT_SYMBOL_GPL(mmput);
899 
900 #ifdef CONFIG_MMU
901 static void mmput_async_fn(struct work_struct *work)
902 {
903 	struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
904 	__mmput(mm);
905 }
906 
907 void mmput_async(struct mm_struct *mm)
908 {
909 	if (atomic_dec_and_test(&mm->mm_users)) {
910 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
911 		schedule_work(&mm->async_put_work);
912 	}
913 }
914 #endif
915 
916 /**
917  * set_mm_exe_file - change a reference to the mm's executable file
918  *
919  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
920  *
921  * Main users are mmput() and sys_execve(). Callers prevent concurrent
922  * invocations: in mmput() nobody alive left, in execve task is single
923  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
924  * mm->exe_file, but does so without using set_mm_exe_file() in order
925  * to do avoid the need for any locks.
926  */
927 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
928 {
929 	struct file *old_exe_file;
930 
931 	/*
932 	 * It is safe to dereference the exe_file without RCU as
933 	 * this function is only called if nobody else can access
934 	 * this mm -- see comment above for justification.
935 	 */
936 	old_exe_file = rcu_dereference_raw(mm->exe_file);
937 
938 	if (new_exe_file)
939 		get_file(new_exe_file);
940 	rcu_assign_pointer(mm->exe_file, new_exe_file);
941 	if (old_exe_file)
942 		fput(old_exe_file);
943 }
944 
945 /**
946  * get_mm_exe_file - acquire a reference to the mm's executable file
947  *
948  * Returns %NULL if mm has no associated executable file.
949  * User must release file via fput().
950  */
951 struct file *get_mm_exe_file(struct mm_struct *mm)
952 {
953 	struct file *exe_file;
954 
955 	rcu_read_lock();
956 	exe_file = rcu_dereference(mm->exe_file);
957 	if (exe_file && !get_file_rcu(exe_file))
958 		exe_file = NULL;
959 	rcu_read_unlock();
960 	return exe_file;
961 }
962 EXPORT_SYMBOL(get_mm_exe_file);
963 
964 /**
965  * get_task_exe_file - acquire a reference to the task's executable file
966  *
967  * Returns %NULL if task's mm (if any) has no associated executable file or
968  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
969  * User must release file via fput().
970  */
971 struct file *get_task_exe_file(struct task_struct *task)
972 {
973 	struct file *exe_file = NULL;
974 	struct mm_struct *mm;
975 
976 	task_lock(task);
977 	mm = task->mm;
978 	if (mm) {
979 		if (!(task->flags & PF_KTHREAD))
980 			exe_file = get_mm_exe_file(mm);
981 	}
982 	task_unlock(task);
983 	return exe_file;
984 }
985 EXPORT_SYMBOL(get_task_exe_file);
986 
987 /**
988  * get_task_mm - acquire a reference to the task's mm
989  *
990  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
991  * this kernel workthread has transiently adopted a user mm with use_mm,
992  * to do its AIO) is not set and if so returns a reference to it, after
993  * bumping up the use count.  User must release the mm via mmput()
994  * after use.  Typically used by /proc and ptrace.
995  */
996 struct mm_struct *get_task_mm(struct task_struct *task)
997 {
998 	struct mm_struct *mm;
999 
1000 	task_lock(task);
1001 	mm = task->mm;
1002 	if (mm) {
1003 		if (task->flags & PF_KTHREAD)
1004 			mm = NULL;
1005 		else
1006 			mmget(mm);
1007 	}
1008 	task_unlock(task);
1009 	return mm;
1010 }
1011 EXPORT_SYMBOL_GPL(get_task_mm);
1012 
1013 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1014 {
1015 	struct mm_struct *mm;
1016 	int err;
1017 
1018 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1019 	if (err)
1020 		return ERR_PTR(err);
1021 
1022 	mm = get_task_mm(task);
1023 	if (mm && mm != current->mm &&
1024 			!ptrace_may_access(task, mode)) {
1025 		mmput(mm);
1026 		mm = ERR_PTR(-EACCES);
1027 	}
1028 	mutex_unlock(&task->signal->cred_guard_mutex);
1029 
1030 	return mm;
1031 }
1032 
1033 static void complete_vfork_done(struct task_struct *tsk)
1034 {
1035 	struct completion *vfork;
1036 
1037 	task_lock(tsk);
1038 	vfork = tsk->vfork_done;
1039 	if (likely(vfork)) {
1040 		tsk->vfork_done = NULL;
1041 		complete(vfork);
1042 	}
1043 	task_unlock(tsk);
1044 }
1045 
1046 static int wait_for_vfork_done(struct task_struct *child,
1047 				struct completion *vfork)
1048 {
1049 	int killed;
1050 
1051 	freezer_do_not_count();
1052 	killed = wait_for_completion_killable(vfork);
1053 	freezer_count();
1054 
1055 	if (killed) {
1056 		task_lock(child);
1057 		child->vfork_done = NULL;
1058 		task_unlock(child);
1059 	}
1060 
1061 	put_task_struct(child);
1062 	return killed;
1063 }
1064 
1065 /* Please note the differences between mmput and mm_release.
1066  * mmput is called whenever we stop holding onto a mm_struct,
1067  * error success whatever.
1068  *
1069  * mm_release is called after a mm_struct has been removed
1070  * from the current process.
1071  *
1072  * This difference is important for error handling, when we
1073  * only half set up a mm_struct for a new process and need to restore
1074  * the old one.  Because we mmput the new mm_struct before
1075  * restoring the old one. . .
1076  * Eric Biederman 10 January 1998
1077  */
1078 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1079 {
1080 	/* Get rid of any futexes when releasing the mm */
1081 #ifdef CONFIG_FUTEX
1082 	if (unlikely(tsk->robust_list)) {
1083 		exit_robust_list(tsk);
1084 		tsk->robust_list = NULL;
1085 	}
1086 #ifdef CONFIG_COMPAT
1087 	if (unlikely(tsk->compat_robust_list)) {
1088 		compat_exit_robust_list(tsk);
1089 		tsk->compat_robust_list = NULL;
1090 	}
1091 #endif
1092 	if (unlikely(!list_empty(&tsk->pi_state_list)))
1093 		exit_pi_state_list(tsk);
1094 #endif
1095 
1096 	uprobe_free_utask(tsk);
1097 
1098 	/* Get rid of any cached register state */
1099 	deactivate_mm(tsk, mm);
1100 
1101 	/*
1102 	 * Signal userspace if we're not exiting with a core dump
1103 	 * because we want to leave the value intact for debugging
1104 	 * purposes.
1105 	 */
1106 	if (tsk->clear_child_tid) {
1107 		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1108 		    atomic_read(&mm->mm_users) > 1) {
1109 			/*
1110 			 * We don't check the error code - if userspace has
1111 			 * not set up a proper pointer then tough luck.
1112 			 */
1113 			put_user(0, tsk->clear_child_tid);
1114 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1115 					1, NULL, NULL, 0);
1116 		}
1117 		tsk->clear_child_tid = NULL;
1118 	}
1119 
1120 	/*
1121 	 * All done, finally we can wake up parent and return this mm to him.
1122 	 * Also kthread_stop() uses this completion for synchronization.
1123 	 */
1124 	if (tsk->vfork_done)
1125 		complete_vfork_done(tsk);
1126 }
1127 
1128 /*
1129  * Allocate a new mm structure and copy contents from the
1130  * mm structure of the passed in task structure.
1131  */
1132 static struct mm_struct *dup_mm(struct task_struct *tsk)
1133 {
1134 	struct mm_struct *mm, *oldmm = current->mm;
1135 	int err;
1136 
1137 	mm = allocate_mm();
1138 	if (!mm)
1139 		goto fail_nomem;
1140 
1141 	memcpy(mm, oldmm, sizeof(*mm));
1142 
1143 	if (!mm_init(mm, tsk, mm->user_ns))
1144 		goto fail_nomem;
1145 
1146 	err = dup_mmap(mm, oldmm);
1147 	if (err)
1148 		goto free_pt;
1149 
1150 	mm->hiwater_rss = get_mm_rss(mm);
1151 	mm->hiwater_vm = mm->total_vm;
1152 
1153 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1154 		goto free_pt;
1155 
1156 	return mm;
1157 
1158 free_pt:
1159 	/* don't put binfmt in mmput, we haven't got module yet */
1160 	mm->binfmt = NULL;
1161 	mmput(mm);
1162 
1163 fail_nomem:
1164 	return NULL;
1165 }
1166 
1167 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1168 {
1169 	struct mm_struct *mm, *oldmm;
1170 	int retval;
1171 
1172 	tsk->min_flt = tsk->maj_flt = 0;
1173 	tsk->nvcsw = tsk->nivcsw = 0;
1174 #ifdef CONFIG_DETECT_HUNG_TASK
1175 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1176 #endif
1177 
1178 	tsk->mm = NULL;
1179 	tsk->active_mm = NULL;
1180 
1181 	/*
1182 	 * Are we cloning a kernel thread?
1183 	 *
1184 	 * We need to steal a active VM for that..
1185 	 */
1186 	oldmm = current->mm;
1187 	if (!oldmm)
1188 		return 0;
1189 
1190 	/* initialize the new vmacache entries */
1191 	vmacache_flush(tsk);
1192 
1193 	if (clone_flags & CLONE_VM) {
1194 		mmget(oldmm);
1195 		mm = oldmm;
1196 		goto good_mm;
1197 	}
1198 
1199 	retval = -ENOMEM;
1200 	mm = dup_mm(tsk);
1201 	if (!mm)
1202 		goto fail_nomem;
1203 
1204 good_mm:
1205 	tsk->mm = mm;
1206 	tsk->active_mm = mm;
1207 	return 0;
1208 
1209 fail_nomem:
1210 	return retval;
1211 }
1212 
1213 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1214 {
1215 	struct fs_struct *fs = current->fs;
1216 	if (clone_flags & CLONE_FS) {
1217 		/* tsk->fs is already what we want */
1218 		spin_lock(&fs->lock);
1219 		if (fs->in_exec) {
1220 			spin_unlock(&fs->lock);
1221 			return -EAGAIN;
1222 		}
1223 		fs->users++;
1224 		spin_unlock(&fs->lock);
1225 		return 0;
1226 	}
1227 	tsk->fs = copy_fs_struct(fs);
1228 	if (!tsk->fs)
1229 		return -ENOMEM;
1230 	return 0;
1231 }
1232 
1233 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1234 {
1235 	struct files_struct *oldf, *newf;
1236 	int error = 0;
1237 
1238 	/*
1239 	 * A background process may not have any files ...
1240 	 */
1241 	oldf = current->files;
1242 	if (!oldf)
1243 		goto out;
1244 
1245 	if (clone_flags & CLONE_FILES) {
1246 		atomic_inc(&oldf->count);
1247 		goto out;
1248 	}
1249 
1250 	newf = dup_fd(oldf, &error);
1251 	if (!newf)
1252 		goto out;
1253 
1254 	tsk->files = newf;
1255 	error = 0;
1256 out:
1257 	return error;
1258 }
1259 
1260 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1261 {
1262 #ifdef CONFIG_BLOCK
1263 	struct io_context *ioc = current->io_context;
1264 	struct io_context *new_ioc;
1265 
1266 	if (!ioc)
1267 		return 0;
1268 	/*
1269 	 * Share io context with parent, if CLONE_IO is set
1270 	 */
1271 	if (clone_flags & CLONE_IO) {
1272 		ioc_task_link(ioc);
1273 		tsk->io_context = ioc;
1274 	} else if (ioprio_valid(ioc->ioprio)) {
1275 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1276 		if (unlikely(!new_ioc))
1277 			return -ENOMEM;
1278 
1279 		new_ioc->ioprio = ioc->ioprio;
1280 		put_io_context(new_ioc);
1281 	}
1282 #endif
1283 	return 0;
1284 }
1285 
1286 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1287 {
1288 	struct sighand_struct *sig;
1289 
1290 	if (clone_flags & CLONE_SIGHAND) {
1291 		atomic_inc(&current->sighand->count);
1292 		return 0;
1293 	}
1294 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1295 	rcu_assign_pointer(tsk->sighand, sig);
1296 	if (!sig)
1297 		return -ENOMEM;
1298 
1299 	atomic_set(&sig->count, 1);
1300 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1301 	return 0;
1302 }
1303 
1304 void __cleanup_sighand(struct sighand_struct *sighand)
1305 {
1306 	if (atomic_dec_and_test(&sighand->count)) {
1307 		signalfd_cleanup(sighand);
1308 		/*
1309 		 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1310 		 * without an RCU grace period, see __lock_task_sighand().
1311 		 */
1312 		kmem_cache_free(sighand_cachep, sighand);
1313 	}
1314 }
1315 
1316 #ifdef CONFIG_POSIX_TIMERS
1317 /*
1318  * Initialize POSIX timer handling for a thread group.
1319  */
1320 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1321 {
1322 	unsigned long cpu_limit;
1323 
1324 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1325 	if (cpu_limit != RLIM_INFINITY) {
1326 		sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1327 		sig->cputimer.running = true;
1328 	}
1329 
1330 	/* The timer lists. */
1331 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1332 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1333 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1334 }
1335 #else
1336 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1337 #endif
1338 
1339 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1340 {
1341 	struct signal_struct *sig;
1342 
1343 	if (clone_flags & CLONE_THREAD)
1344 		return 0;
1345 
1346 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1347 	tsk->signal = sig;
1348 	if (!sig)
1349 		return -ENOMEM;
1350 
1351 	sig->nr_threads = 1;
1352 	atomic_set(&sig->live, 1);
1353 	atomic_set(&sig->sigcnt, 1);
1354 
1355 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1356 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1357 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1358 
1359 	init_waitqueue_head(&sig->wait_chldexit);
1360 	sig->curr_target = tsk;
1361 	init_sigpending(&sig->shared_pending);
1362 	seqlock_init(&sig->stats_lock);
1363 	prev_cputime_init(&sig->prev_cputime);
1364 
1365 #ifdef CONFIG_POSIX_TIMERS
1366 	INIT_LIST_HEAD(&sig->posix_timers);
1367 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1368 	sig->real_timer.function = it_real_fn;
1369 #endif
1370 
1371 	task_lock(current->group_leader);
1372 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1373 	task_unlock(current->group_leader);
1374 
1375 	posix_cpu_timers_init_group(sig);
1376 
1377 	tty_audit_fork(sig);
1378 	sched_autogroup_fork(sig);
1379 
1380 	sig->oom_score_adj = current->signal->oom_score_adj;
1381 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1382 
1383 	mutex_init(&sig->cred_guard_mutex);
1384 
1385 	return 0;
1386 }
1387 
1388 static void copy_seccomp(struct task_struct *p)
1389 {
1390 #ifdef CONFIG_SECCOMP
1391 	/*
1392 	 * Must be called with sighand->lock held, which is common to
1393 	 * all threads in the group. Holding cred_guard_mutex is not
1394 	 * needed because this new task is not yet running and cannot
1395 	 * be racing exec.
1396 	 */
1397 	assert_spin_locked(&current->sighand->siglock);
1398 
1399 	/* Ref-count the new filter user, and assign it. */
1400 	get_seccomp_filter(current);
1401 	p->seccomp = current->seccomp;
1402 
1403 	/*
1404 	 * Explicitly enable no_new_privs here in case it got set
1405 	 * between the task_struct being duplicated and holding the
1406 	 * sighand lock. The seccomp state and nnp must be in sync.
1407 	 */
1408 	if (task_no_new_privs(current))
1409 		task_set_no_new_privs(p);
1410 
1411 	/*
1412 	 * If the parent gained a seccomp mode after copying thread
1413 	 * flags and between before we held the sighand lock, we have
1414 	 * to manually enable the seccomp thread flag here.
1415 	 */
1416 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1417 		set_tsk_thread_flag(p, TIF_SECCOMP);
1418 #endif
1419 }
1420 
1421 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1422 {
1423 	current->clear_child_tid = tidptr;
1424 
1425 	return task_pid_vnr(current);
1426 }
1427 
1428 static void rt_mutex_init_task(struct task_struct *p)
1429 {
1430 	raw_spin_lock_init(&p->pi_lock);
1431 #ifdef CONFIG_RT_MUTEXES
1432 	p->pi_waiters = RB_ROOT;
1433 	p->pi_waiters_leftmost = NULL;
1434 	p->pi_blocked_on = NULL;
1435 #endif
1436 }
1437 
1438 #ifdef CONFIG_POSIX_TIMERS
1439 /*
1440  * Initialize POSIX timer handling for a single task.
1441  */
1442 static void posix_cpu_timers_init(struct task_struct *tsk)
1443 {
1444 	tsk->cputime_expires.prof_exp = 0;
1445 	tsk->cputime_expires.virt_exp = 0;
1446 	tsk->cputime_expires.sched_exp = 0;
1447 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1448 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1449 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1450 }
1451 #else
1452 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1453 #endif
1454 
1455 static inline void
1456 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1457 {
1458 	 task->pids[type].pid = pid;
1459 }
1460 
1461 /*
1462  * This creates a new process as a copy of the old one,
1463  * but does not actually start it yet.
1464  *
1465  * It copies the registers, and all the appropriate
1466  * parts of the process environment (as per the clone
1467  * flags). The actual kick-off is left to the caller.
1468  */
1469 static __latent_entropy struct task_struct *copy_process(
1470 					unsigned long clone_flags,
1471 					unsigned long stack_start,
1472 					unsigned long stack_size,
1473 					int __user *child_tidptr,
1474 					struct pid *pid,
1475 					int trace,
1476 					unsigned long tls,
1477 					int node)
1478 {
1479 	int retval;
1480 	struct task_struct *p;
1481 
1482 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1483 		return ERR_PTR(-EINVAL);
1484 
1485 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1486 		return ERR_PTR(-EINVAL);
1487 
1488 	/*
1489 	 * Thread groups must share signals as well, and detached threads
1490 	 * can only be started up within the thread group.
1491 	 */
1492 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1493 		return ERR_PTR(-EINVAL);
1494 
1495 	/*
1496 	 * Shared signal handlers imply shared VM. By way of the above,
1497 	 * thread groups also imply shared VM. Blocking this case allows
1498 	 * for various simplifications in other code.
1499 	 */
1500 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1501 		return ERR_PTR(-EINVAL);
1502 
1503 	/*
1504 	 * Siblings of global init remain as zombies on exit since they are
1505 	 * not reaped by their parent (swapper). To solve this and to avoid
1506 	 * multi-rooted process trees, prevent global and container-inits
1507 	 * from creating siblings.
1508 	 */
1509 	if ((clone_flags & CLONE_PARENT) &&
1510 				current->signal->flags & SIGNAL_UNKILLABLE)
1511 		return ERR_PTR(-EINVAL);
1512 
1513 	/*
1514 	 * If the new process will be in a different pid or user namespace
1515 	 * do not allow it to share a thread group with the forking task.
1516 	 */
1517 	if (clone_flags & CLONE_THREAD) {
1518 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1519 		    (task_active_pid_ns(current) !=
1520 				current->nsproxy->pid_ns_for_children))
1521 			return ERR_PTR(-EINVAL);
1522 	}
1523 
1524 	retval = security_task_create(clone_flags);
1525 	if (retval)
1526 		goto fork_out;
1527 
1528 	retval = -ENOMEM;
1529 	p = dup_task_struct(current, node);
1530 	if (!p)
1531 		goto fork_out;
1532 
1533 	ftrace_graph_init_task(p);
1534 
1535 	rt_mutex_init_task(p);
1536 
1537 #ifdef CONFIG_PROVE_LOCKING
1538 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1539 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1540 #endif
1541 	retval = -EAGAIN;
1542 	if (atomic_read(&p->real_cred->user->processes) >=
1543 			task_rlimit(p, RLIMIT_NPROC)) {
1544 		if (p->real_cred->user != INIT_USER &&
1545 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1546 			goto bad_fork_free;
1547 	}
1548 	current->flags &= ~PF_NPROC_EXCEEDED;
1549 
1550 	retval = copy_creds(p, clone_flags);
1551 	if (retval < 0)
1552 		goto bad_fork_free;
1553 
1554 	/*
1555 	 * If multiple threads are within copy_process(), then this check
1556 	 * triggers too late. This doesn't hurt, the check is only there
1557 	 * to stop root fork bombs.
1558 	 */
1559 	retval = -EAGAIN;
1560 	if (nr_threads >= max_threads)
1561 		goto bad_fork_cleanup_count;
1562 
1563 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1564 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1565 	p->flags |= PF_FORKNOEXEC;
1566 	INIT_LIST_HEAD(&p->children);
1567 	INIT_LIST_HEAD(&p->sibling);
1568 	rcu_copy_process(p);
1569 	p->vfork_done = NULL;
1570 	spin_lock_init(&p->alloc_lock);
1571 
1572 	init_sigpending(&p->pending);
1573 
1574 	p->utime = p->stime = p->gtime = 0;
1575 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1576 	p->utimescaled = p->stimescaled = 0;
1577 #endif
1578 	prev_cputime_init(&p->prev_cputime);
1579 
1580 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1581 	seqcount_init(&p->vtime_seqcount);
1582 	p->vtime_snap = 0;
1583 	p->vtime_snap_whence = VTIME_INACTIVE;
1584 #endif
1585 
1586 #if defined(SPLIT_RSS_COUNTING)
1587 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1588 #endif
1589 
1590 	p->default_timer_slack_ns = current->timer_slack_ns;
1591 
1592 	task_io_accounting_init(&p->ioac);
1593 	acct_clear_integrals(p);
1594 
1595 	posix_cpu_timers_init(p);
1596 
1597 	p->start_time = ktime_get_ns();
1598 	p->real_start_time = ktime_get_boot_ns();
1599 	p->io_context = NULL;
1600 	p->audit_context = NULL;
1601 	cgroup_fork(p);
1602 #ifdef CONFIG_NUMA
1603 	p->mempolicy = mpol_dup(p->mempolicy);
1604 	if (IS_ERR(p->mempolicy)) {
1605 		retval = PTR_ERR(p->mempolicy);
1606 		p->mempolicy = NULL;
1607 		goto bad_fork_cleanup_threadgroup_lock;
1608 	}
1609 #endif
1610 #ifdef CONFIG_CPUSETS
1611 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1612 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1613 	seqcount_init(&p->mems_allowed_seq);
1614 #endif
1615 #ifdef CONFIG_TRACE_IRQFLAGS
1616 	p->irq_events = 0;
1617 	p->hardirqs_enabled = 0;
1618 	p->hardirq_enable_ip = 0;
1619 	p->hardirq_enable_event = 0;
1620 	p->hardirq_disable_ip = _THIS_IP_;
1621 	p->hardirq_disable_event = 0;
1622 	p->softirqs_enabled = 1;
1623 	p->softirq_enable_ip = _THIS_IP_;
1624 	p->softirq_enable_event = 0;
1625 	p->softirq_disable_ip = 0;
1626 	p->softirq_disable_event = 0;
1627 	p->hardirq_context = 0;
1628 	p->softirq_context = 0;
1629 #endif
1630 
1631 	p->pagefault_disabled = 0;
1632 
1633 #ifdef CONFIG_LOCKDEP
1634 	p->lockdep_depth = 0; /* no locks held yet */
1635 	p->curr_chain_key = 0;
1636 	p->lockdep_recursion = 0;
1637 #endif
1638 
1639 #ifdef CONFIG_DEBUG_MUTEXES
1640 	p->blocked_on = NULL; /* not blocked yet */
1641 #endif
1642 #ifdef CONFIG_BCACHE
1643 	p->sequential_io	= 0;
1644 	p->sequential_io_avg	= 0;
1645 #endif
1646 
1647 	/* Perform scheduler related setup. Assign this task to a CPU. */
1648 	retval = sched_fork(clone_flags, p);
1649 	if (retval)
1650 		goto bad_fork_cleanup_policy;
1651 
1652 	retval = perf_event_init_task(p);
1653 	if (retval)
1654 		goto bad_fork_cleanup_policy;
1655 	retval = audit_alloc(p);
1656 	if (retval)
1657 		goto bad_fork_cleanup_perf;
1658 	/* copy all the process information */
1659 	shm_init_task(p);
1660 	retval = copy_semundo(clone_flags, p);
1661 	if (retval)
1662 		goto bad_fork_cleanup_audit;
1663 	retval = copy_files(clone_flags, p);
1664 	if (retval)
1665 		goto bad_fork_cleanup_semundo;
1666 	retval = copy_fs(clone_flags, p);
1667 	if (retval)
1668 		goto bad_fork_cleanup_files;
1669 	retval = copy_sighand(clone_flags, p);
1670 	if (retval)
1671 		goto bad_fork_cleanup_fs;
1672 	retval = copy_signal(clone_flags, p);
1673 	if (retval)
1674 		goto bad_fork_cleanup_sighand;
1675 	retval = copy_mm(clone_flags, p);
1676 	if (retval)
1677 		goto bad_fork_cleanup_signal;
1678 	retval = copy_namespaces(clone_flags, p);
1679 	if (retval)
1680 		goto bad_fork_cleanup_mm;
1681 	retval = copy_io(clone_flags, p);
1682 	if (retval)
1683 		goto bad_fork_cleanup_namespaces;
1684 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1685 	if (retval)
1686 		goto bad_fork_cleanup_io;
1687 
1688 	if (pid != &init_struct_pid) {
1689 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1690 		if (IS_ERR(pid)) {
1691 			retval = PTR_ERR(pid);
1692 			goto bad_fork_cleanup_thread;
1693 		}
1694 	}
1695 
1696 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1697 	/*
1698 	 * Clear TID on mm_release()?
1699 	 */
1700 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1701 #ifdef CONFIG_BLOCK
1702 	p->plug = NULL;
1703 #endif
1704 #ifdef CONFIG_FUTEX
1705 	p->robust_list = NULL;
1706 #ifdef CONFIG_COMPAT
1707 	p->compat_robust_list = NULL;
1708 #endif
1709 	INIT_LIST_HEAD(&p->pi_state_list);
1710 	p->pi_state_cache = NULL;
1711 #endif
1712 	/*
1713 	 * sigaltstack should be cleared when sharing the same VM
1714 	 */
1715 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1716 		sas_ss_reset(p);
1717 
1718 	/*
1719 	 * Syscall tracing and stepping should be turned off in the
1720 	 * child regardless of CLONE_PTRACE.
1721 	 */
1722 	user_disable_single_step(p);
1723 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1724 #ifdef TIF_SYSCALL_EMU
1725 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1726 #endif
1727 	clear_all_latency_tracing(p);
1728 
1729 	/* ok, now we should be set up.. */
1730 	p->pid = pid_nr(pid);
1731 	if (clone_flags & CLONE_THREAD) {
1732 		p->exit_signal = -1;
1733 		p->group_leader = current->group_leader;
1734 		p->tgid = current->tgid;
1735 	} else {
1736 		if (clone_flags & CLONE_PARENT)
1737 			p->exit_signal = current->group_leader->exit_signal;
1738 		else
1739 			p->exit_signal = (clone_flags & CSIGNAL);
1740 		p->group_leader = p;
1741 		p->tgid = p->pid;
1742 	}
1743 
1744 	p->nr_dirtied = 0;
1745 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1746 	p->dirty_paused_when = 0;
1747 
1748 	p->pdeath_signal = 0;
1749 	INIT_LIST_HEAD(&p->thread_group);
1750 	p->task_works = NULL;
1751 
1752 	cgroup_threadgroup_change_begin(current);
1753 	/*
1754 	 * Ensure that the cgroup subsystem policies allow the new process to be
1755 	 * forked. It should be noted the the new process's css_set can be changed
1756 	 * between here and cgroup_post_fork() if an organisation operation is in
1757 	 * progress.
1758 	 */
1759 	retval = cgroup_can_fork(p);
1760 	if (retval)
1761 		goto bad_fork_free_pid;
1762 
1763 	/*
1764 	 * Make it visible to the rest of the system, but dont wake it up yet.
1765 	 * Need tasklist lock for parent etc handling!
1766 	 */
1767 	write_lock_irq(&tasklist_lock);
1768 
1769 	/* CLONE_PARENT re-uses the old parent */
1770 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1771 		p->real_parent = current->real_parent;
1772 		p->parent_exec_id = current->parent_exec_id;
1773 	} else {
1774 		p->real_parent = current;
1775 		p->parent_exec_id = current->self_exec_id;
1776 	}
1777 
1778 	spin_lock(&current->sighand->siglock);
1779 
1780 	/*
1781 	 * Copy seccomp details explicitly here, in case they were changed
1782 	 * before holding sighand lock.
1783 	 */
1784 	copy_seccomp(p);
1785 
1786 	/*
1787 	 * Process group and session signals need to be delivered to just the
1788 	 * parent before the fork or both the parent and the child after the
1789 	 * fork. Restart if a signal comes in before we add the new process to
1790 	 * it's process group.
1791 	 * A fatal signal pending means that current will exit, so the new
1792 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1793 	*/
1794 	recalc_sigpending();
1795 	if (signal_pending(current)) {
1796 		spin_unlock(&current->sighand->siglock);
1797 		write_unlock_irq(&tasklist_lock);
1798 		retval = -ERESTARTNOINTR;
1799 		goto bad_fork_cancel_cgroup;
1800 	}
1801 
1802 	if (likely(p->pid)) {
1803 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1804 
1805 		init_task_pid(p, PIDTYPE_PID, pid);
1806 		if (thread_group_leader(p)) {
1807 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1808 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1809 
1810 			if (is_child_reaper(pid)) {
1811 				ns_of_pid(pid)->child_reaper = p;
1812 				p->signal->flags |= SIGNAL_UNKILLABLE;
1813 			}
1814 
1815 			p->signal->leader_pid = pid;
1816 			p->signal->tty = tty_kref_get(current->signal->tty);
1817 			/*
1818 			 * Inherit has_child_subreaper flag under the same
1819 			 * tasklist_lock with adding child to the process tree
1820 			 * for propagate_has_child_subreaper optimization.
1821 			 */
1822 			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1823 							 p->real_parent->signal->is_child_subreaper;
1824 			list_add_tail(&p->sibling, &p->real_parent->children);
1825 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1826 			attach_pid(p, PIDTYPE_PGID);
1827 			attach_pid(p, PIDTYPE_SID);
1828 			__this_cpu_inc(process_counts);
1829 		} else {
1830 			current->signal->nr_threads++;
1831 			atomic_inc(&current->signal->live);
1832 			atomic_inc(&current->signal->sigcnt);
1833 			list_add_tail_rcu(&p->thread_group,
1834 					  &p->group_leader->thread_group);
1835 			list_add_tail_rcu(&p->thread_node,
1836 					  &p->signal->thread_head);
1837 		}
1838 		attach_pid(p, PIDTYPE_PID);
1839 		nr_threads++;
1840 	}
1841 
1842 	total_forks++;
1843 	spin_unlock(&current->sighand->siglock);
1844 	syscall_tracepoint_update(p);
1845 	write_unlock_irq(&tasklist_lock);
1846 
1847 	proc_fork_connector(p);
1848 	cgroup_post_fork(p);
1849 	cgroup_threadgroup_change_end(current);
1850 	perf_event_fork(p);
1851 
1852 	trace_task_newtask(p, clone_flags);
1853 	uprobe_copy_process(p, clone_flags);
1854 
1855 	return p;
1856 
1857 bad_fork_cancel_cgroup:
1858 	cgroup_cancel_fork(p);
1859 bad_fork_free_pid:
1860 	cgroup_threadgroup_change_end(current);
1861 	if (pid != &init_struct_pid)
1862 		free_pid(pid);
1863 bad_fork_cleanup_thread:
1864 	exit_thread(p);
1865 bad_fork_cleanup_io:
1866 	if (p->io_context)
1867 		exit_io_context(p);
1868 bad_fork_cleanup_namespaces:
1869 	exit_task_namespaces(p);
1870 bad_fork_cleanup_mm:
1871 	if (p->mm)
1872 		mmput(p->mm);
1873 bad_fork_cleanup_signal:
1874 	if (!(clone_flags & CLONE_THREAD))
1875 		free_signal_struct(p->signal);
1876 bad_fork_cleanup_sighand:
1877 	__cleanup_sighand(p->sighand);
1878 bad_fork_cleanup_fs:
1879 	exit_fs(p); /* blocking */
1880 bad_fork_cleanup_files:
1881 	exit_files(p); /* blocking */
1882 bad_fork_cleanup_semundo:
1883 	exit_sem(p);
1884 bad_fork_cleanup_audit:
1885 	audit_free(p);
1886 bad_fork_cleanup_perf:
1887 	perf_event_free_task(p);
1888 bad_fork_cleanup_policy:
1889 #ifdef CONFIG_NUMA
1890 	mpol_put(p->mempolicy);
1891 bad_fork_cleanup_threadgroup_lock:
1892 #endif
1893 	delayacct_tsk_free(p);
1894 bad_fork_cleanup_count:
1895 	atomic_dec(&p->cred->user->processes);
1896 	exit_creds(p);
1897 bad_fork_free:
1898 	p->state = TASK_DEAD;
1899 	put_task_stack(p);
1900 	free_task(p);
1901 fork_out:
1902 	return ERR_PTR(retval);
1903 }
1904 
1905 static inline void init_idle_pids(struct pid_link *links)
1906 {
1907 	enum pid_type type;
1908 
1909 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1910 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1911 		links[type].pid = &init_struct_pid;
1912 	}
1913 }
1914 
1915 struct task_struct *fork_idle(int cpu)
1916 {
1917 	struct task_struct *task;
1918 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1919 			    cpu_to_node(cpu));
1920 	if (!IS_ERR(task)) {
1921 		init_idle_pids(task->pids);
1922 		init_idle(task, cpu);
1923 	}
1924 
1925 	return task;
1926 }
1927 
1928 /*
1929  *  Ok, this is the main fork-routine.
1930  *
1931  * It copies the process, and if successful kick-starts
1932  * it and waits for it to finish using the VM if required.
1933  */
1934 long _do_fork(unsigned long clone_flags,
1935 	      unsigned long stack_start,
1936 	      unsigned long stack_size,
1937 	      int __user *parent_tidptr,
1938 	      int __user *child_tidptr,
1939 	      unsigned long tls)
1940 {
1941 	struct task_struct *p;
1942 	int trace = 0;
1943 	long nr;
1944 
1945 	/*
1946 	 * Determine whether and which event to report to ptracer.  When
1947 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
1948 	 * requested, no event is reported; otherwise, report if the event
1949 	 * for the type of forking is enabled.
1950 	 */
1951 	if (!(clone_flags & CLONE_UNTRACED)) {
1952 		if (clone_flags & CLONE_VFORK)
1953 			trace = PTRACE_EVENT_VFORK;
1954 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
1955 			trace = PTRACE_EVENT_CLONE;
1956 		else
1957 			trace = PTRACE_EVENT_FORK;
1958 
1959 		if (likely(!ptrace_event_enabled(current, trace)))
1960 			trace = 0;
1961 	}
1962 
1963 	p = copy_process(clone_flags, stack_start, stack_size,
1964 			 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1965 	add_latent_entropy();
1966 	/*
1967 	 * Do this prior waking up the new thread - the thread pointer
1968 	 * might get invalid after that point, if the thread exits quickly.
1969 	 */
1970 	if (!IS_ERR(p)) {
1971 		struct completion vfork;
1972 		struct pid *pid;
1973 
1974 		trace_sched_process_fork(current, p);
1975 
1976 		pid = get_task_pid(p, PIDTYPE_PID);
1977 		nr = pid_vnr(pid);
1978 
1979 		if (clone_flags & CLONE_PARENT_SETTID)
1980 			put_user(nr, parent_tidptr);
1981 
1982 		if (clone_flags & CLONE_VFORK) {
1983 			p->vfork_done = &vfork;
1984 			init_completion(&vfork);
1985 			get_task_struct(p);
1986 		}
1987 
1988 		wake_up_new_task(p);
1989 
1990 		/* forking complete and child started to run, tell ptracer */
1991 		if (unlikely(trace))
1992 			ptrace_event_pid(trace, pid);
1993 
1994 		if (clone_flags & CLONE_VFORK) {
1995 			if (!wait_for_vfork_done(p, &vfork))
1996 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1997 		}
1998 
1999 		put_pid(pid);
2000 	} else {
2001 		nr = PTR_ERR(p);
2002 	}
2003 	return nr;
2004 }
2005 
2006 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2007 /* For compatibility with architectures that call do_fork directly rather than
2008  * using the syscall entry points below. */
2009 long do_fork(unsigned long clone_flags,
2010 	      unsigned long stack_start,
2011 	      unsigned long stack_size,
2012 	      int __user *parent_tidptr,
2013 	      int __user *child_tidptr)
2014 {
2015 	return _do_fork(clone_flags, stack_start, stack_size,
2016 			parent_tidptr, child_tidptr, 0);
2017 }
2018 #endif
2019 
2020 /*
2021  * Create a kernel thread.
2022  */
2023 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2024 {
2025 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2026 		(unsigned long)arg, NULL, NULL, 0);
2027 }
2028 
2029 #ifdef __ARCH_WANT_SYS_FORK
2030 SYSCALL_DEFINE0(fork)
2031 {
2032 #ifdef CONFIG_MMU
2033 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2034 #else
2035 	/* can not support in nommu mode */
2036 	return -EINVAL;
2037 #endif
2038 }
2039 #endif
2040 
2041 #ifdef __ARCH_WANT_SYS_VFORK
2042 SYSCALL_DEFINE0(vfork)
2043 {
2044 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2045 			0, NULL, NULL, 0);
2046 }
2047 #endif
2048 
2049 #ifdef __ARCH_WANT_SYS_CLONE
2050 #ifdef CONFIG_CLONE_BACKWARDS
2051 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2052 		 int __user *, parent_tidptr,
2053 		 unsigned long, tls,
2054 		 int __user *, child_tidptr)
2055 #elif defined(CONFIG_CLONE_BACKWARDS2)
2056 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2057 		 int __user *, parent_tidptr,
2058 		 int __user *, child_tidptr,
2059 		 unsigned long, tls)
2060 #elif defined(CONFIG_CLONE_BACKWARDS3)
2061 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2062 		int, stack_size,
2063 		int __user *, parent_tidptr,
2064 		int __user *, child_tidptr,
2065 		unsigned long, tls)
2066 #else
2067 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2068 		 int __user *, parent_tidptr,
2069 		 int __user *, child_tidptr,
2070 		 unsigned long, tls)
2071 #endif
2072 {
2073 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2074 }
2075 #endif
2076 
2077 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2078 {
2079 	struct task_struct *leader, *parent, *child;
2080 	int res;
2081 
2082 	read_lock(&tasklist_lock);
2083 	leader = top = top->group_leader;
2084 down:
2085 	for_each_thread(leader, parent) {
2086 		list_for_each_entry(child, &parent->children, sibling) {
2087 			res = visitor(child, data);
2088 			if (res) {
2089 				if (res < 0)
2090 					goto out;
2091 				leader = child;
2092 				goto down;
2093 			}
2094 up:
2095 			;
2096 		}
2097 	}
2098 
2099 	if (leader != top) {
2100 		child = leader;
2101 		parent = child->real_parent;
2102 		leader = parent->group_leader;
2103 		goto up;
2104 	}
2105 out:
2106 	read_unlock(&tasklist_lock);
2107 }
2108 
2109 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2110 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2111 #endif
2112 
2113 static void sighand_ctor(void *data)
2114 {
2115 	struct sighand_struct *sighand = data;
2116 
2117 	spin_lock_init(&sighand->siglock);
2118 	init_waitqueue_head(&sighand->signalfd_wqh);
2119 }
2120 
2121 void __init proc_caches_init(void)
2122 {
2123 	sighand_cachep = kmem_cache_create("sighand_cache",
2124 			sizeof(struct sighand_struct), 0,
2125 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2126 			SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2127 	signal_cachep = kmem_cache_create("signal_cache",
2128 			sizeof(struct signal_struct), 0,
2129 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2130 			NULL);
2131 	files_cachep = kmem_cache_create("files_cache",
2132 			sizeof(struct files_struct), 0,
2133 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2134 			NULL);
2135 	fs_cachep = kmem_cache_create("fs_cache",
2136 			sizeof(struct fs_struct), 0,
2137 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2138 			NULL);
2139 	/*
2140 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2141 	 * whole struct cpumask for the OFFSTACK case. We could change
2142 	 * this to *only* allocate as much of it as required by the
2143 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
2144 	 * is at the end of the structure, exactly for that reason.
2145 	 */
2146 	mm_cachep = kmem_cache_create("mm_struct",
2147 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2148 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2149 			NULL);
2150 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2151 	mmap_init();
2152 	nsproxy_cache_init();
2153 }
2154 
2155 /*
2156  * Check constraints on flags passed to the unshare system call.
2157  */
2158 static int check_unshare_flags(unsigned long unshare_flags)
2159 {
2160 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2161 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2162 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2163 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2164 		return -EINVAL;
2165 	/*
2166 	 * Not implemented, but pretend it works if there is nothing
2167 	 * to unshare.  Note that unsharing the address space or the
2168 	 * signal handlers also need to unshare the signal queues (aka
2169 	 * CLONE_THREAD).
2170 	 */
2171 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2172 		if (!thread_group_empty(current))
2173 			return -EINVAL;
2174 	}
2175 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2176 		if (atomic_read(&current->sighand->count) > 1)
2177 			return -EINVAL;
2178 	}
2179 	if (unshare_flags & CLONE_VM) {
2180 		if (!current_is_single_threaded())
2181 			return -EINVAL;
2182 	}
2183 
2184 	return 0;
2185 }
2186 
2187 /*
2188  * Unshare the filesystem structure if it is being shared
2189  */
2190 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2191 {
2192 	struct fs_struct *fs = current->fs;
2193 
2194 	if (!(unshare_flags & CLONE_FS) || !fs)
2195 		return 0;
2196 
2197 	/* don't need lock here; in the worst case we'll do useless copy */
2198 	if (fs->users == 1)
2199 		return 0;
2200 
2201 	*new_fsp = copy_fs_struct(fs);
2202 	if (!*new_fsp)
2203 		return -ENOMEM;
2204 
2205 	return 0;
2206 }
2207 
2208 /*
2209  * Unshare file descriptor table if it is being shared
2210  */
2211 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2212 {
2213 	struct files_struct *fd = current->files;
2214 	int error = 0;
2215 
2216 	if ((unshare_flags & CLONE_FILES) &&
2217 	    (fd && atomic_read(&fd->count) > 1)) {
2218 		*new_fdp = dup_fd(fd, &error);
2219 		if (!*new_fdp)
2220 			return error;
2221 	}
2222 
2223 	return 0;
2224 }
2225 
2226 /*
2227  * unshare allows a process to 'unshare' part of the process
2228  * context which was originally shared using clone.  copy_*
2229  * functions used by do_fork() cannot be used here directly
2230  * because they modify an inactive task_struct that is being
2231  * constructed. Here we are modifying the current, active,
2232  * task_struct.
2233  */
2234 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2235 {
2236 	struct fs_struct *fs, *new_fs = NULL;
2237 	struct files_struct *fd, *new_fd = NULL;
2238 	struct cred *new_cred = NULL;
2239 	struct nsproxy *new_nsproxy = NULL;
2240 	int do_sysvsem = 0;
2241 	int err;
2242 
2243 	/*
2244 	 * If unsharing a user namespace must also unshare the thread group
2245 	 * and unshare the filesystem root and working directories.
2246 	 */
2247 	if (unshare_flags & CLONE_NEWUSER)
2248 		unshare_flags |= CLONE_THREAD | CLONE_FS;
2249 	/*
2250 	 * If unsharing vm, must also unshare signal handlers.
2251 	 */
2252 	if (unshare_flags & CLONE_VM)
2253 		unshare_flags |= CLONE_SIGHAND;
2254 	/*
2255 	 * If unsharing a signal handlers, must also unshare the signal queues.
2256 	 */
2257 	if (unshare_flags & CLONE_SIGHAND)
2258 		unshare_flags |= CLONE_THREAD;
2259 	/*
2260 	 * If unsharing namespace, must also unshare filesystem information.
2261 	 */
2262 	if (unshare_flags & CLONE_NEWNS)
2263 		unshare_flags |= CLONE_FS;
2264 
2265 	err = check_unshare_flags(unshare_flags);
2266 	if (err)
2267 		goto bad_unshare_out;
2268 	/*
2269 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2270 	 * to a new ipc namespace, the semaphore arrays from the old
2271 	 * namespace are unreachable.
2272 	 */
2273 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2274 		do_sysvsem = 1;
2275 	err = unshare_fs(unshare_flags, &new_fs);
2276 	if (err)
2277 		goto bad_unshare_out;
2278 	err = unshare_fd(unshare_flags, &new_fd);
2279 	if (err)
2280 		goto bad_unshare_cleanup_fs;
2281 	err = unshare_userns(unshare_flags, &new_cred);
2282 	if (err)
2283 		goto bad_unshare_cleanup_fd;
2284 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2285 					 new_cred, new_fs);
2286 	if (err)
2287 		goto bad_unshare_cleanup_cred;
2288 
2289 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2290 		if (do_sysvsem) {
2291 			/*
2292 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2293 			 */
2294 			exit_sem(current);
2295 		}
2296 		if (unshare_flags & CLONE_NEWIPC) {
2297 			/* Orphan segments in old ns (see sem above). */
2298 			exit_shm(current);
2299 			shm_init_task(current);
2300 		}
2301 
2302 		if (new_nsproxy)
2303 			switch_task_namespaces(current, new_nsproxy);
2304 
2305 		task_lock(current);
2306 
2307 		if (new_fs) {
2308 			fs = current->fs;
2309 			spin_lock(&fs->lock);
2310 			current->fs = new_fs;
2311 			if (--fs->users)
2312 				new_fs = NULL;
2313 			else
2314 				new_fs = fs;
2315 			spin_unlock(&fs->lock);
2316 		}
2317 
2318 		if (new_fd) {
2319 			fd = current->files;
2320 			current->files = new_fd;
2321 			new_fd = fd;
2322 		}
2323 
2324 		task_unlock(current);
2325 
2326 		if (new_cred) {
2327 			/* Install the new user namespace */
2328 			commit_creds(new_cred);
2329 			new_cred = NULL;
2330 		}
2331 	}
2332 
2333 bad_unshare_cleanup_cred:
2334 	if (new_cred)
2335 		put_cred(new_cred);
2336 bad_unshare_cleanup_fd:
2337 	if (new_fd)
2338 		put_files_struct(new_fd);
2339 
2340 bad_unshare_cleanup_fs:
2341 	if (new_fs)
2342 		free_fs_struct(new_fs);
2343 
2344 bad_unshare_out:
2345 	return err;
2346 }
2347 
2348 /*
2349  *	Helper to unshare the files of the current task.
2350  *	We don't want to expose copy_files internals to
2351  *	the exec layer of the kernel.
2352  */
2353 
2354 int unshare_files(struct files_struct **displaced)
2355 {
2356 	struct task_struct *task = current;
2357 	struct files_struct *copy = NULL;
2358 	int error;
2359 
2360 	error = unshare_fd(CLONE_FILES, &copy);
2361 	if (error || !copy) {
2362 		*displaced = NULL;
2363 		return error;
2364 	}
2365 	*displaced = task->files;
2366 	task_lock(task);
2367 	task->files = copy;
2368 	task_unlock(task);
2369 	return 0;
2370 }
2371 
2372 int sysctl_max_threads(struct ctl_table *table, int write,
2373 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2374 {
2375 	struct ctl_table t;
2376 	int ret;
2377 	int threads = max_threads;
2378 	int min = MIN_THREADS;
2379 	int max = MAX_THREADS;
2380 
2381 	t = *table;
2382 	t.data = &threads;
2383 	t.extra1 = &min;
2384 	t.extra2 = &max;
2385 
2386 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2387 	if (ret || !write)
2388 		return ret;
2389 
2390 	set_max_threads(threads);
2391 
2392 	return 0;
2393 }
2394