xref: /linux/kernel/fork.c (revision 9263969a46fc899092ba4f8c4206fa2340c9a64e)
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/sched/user.h>
19 #include <linux/sched/numa_balancing.h>
20 #include <linux/sched/stat.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/task_stack.h>
23 #include <linux/sched/cputime.h>
24 #include <linux/rtmutex.h>
25 #include <linux/init.h>
26 #include <linux/unistd.h>
27 #include <linux/module.h>
28 #include <linux/vmalloc.h>
29 #include <linux/completion.h>
30 #include <linux/personality.h>
31 #include <linux/mempolicy.h>
32 #include <linux/sem.h>
33 #include <linux/file.h>
34 #include <linux/fdtable.h>
35 #include <linux/iocontext.h>
36 #include <linux/key.h>
37 #include <linux/binfmts.h>
38 #include <linux/mman.h>
39 #include <linux/mmu_notifier.h>
40 #include <linux/fs.h>
41 #include <linux/mm.h>
42 #include <linux/vmacache.h>
43 #include <linux/nsproxy.h>
44 #include <linux/capability.h>
45 #include <linux/cpu.h>
46 #include <linux/cgroup.h>
47 #include <linux/security.h>
48 #include <linux/hugetlb.h>
49 #include <linux/seccomp.h>
50 #include <linux/swap.h>
51 #include <linux/syscalls.h>
52 #include <linux/jiffies.h>
53 #include <linux/futex.h>
54 #include <linux/compat.h>
55 #include <linux/kthread.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/rcupdate.h>
58 #include <linux/ptrace.h>
59 #include <linux/mount.h>
60 #include <linux/audit.h>
61 #include <linux/memcontrol.h>
62 #include <linux/ftrace.h>
63 #include <linux/proc_fs.h>
64 #include <linux/profile.h>
65 #include <linux/rmap.h>
66 #include <linux/ksm.h>
67 #include <linux/acct.h>
68 #include <linux/userfaultfd_k.h>
69 #include <linux/tsacct_kern.h>
70 #include <linux/cn_proc.h>
71 #include <linux/freezer.h>
72 #include <linux/delayacct.h>
73 #include <linux/taskstats_kern.h>
74 #include <linux/random.h>
75 #include <linux/tty.h>
76 #include <linux/blkdev.h>
77 #include <linux/fs_struct.h>
78 #include <linux/magic.h>
79 #include <linux/perf_event.h>
80 #include <linux/posix-timers.h>
81 #include <linux/user-return-notifier.h>
82 #include <linux/oom.h>
83 #include <linux/khugepaged.h>
84 #include <linux/signalfd.h>
85 #include <linux/uprobes.h>
86 #include <linux/aio.h>
87 #include <linux/compiler.h>
88 #include <linux/sysctl.h>
89 #include <linux/kcov.h>
90 #include <linux/livepatch.h>
91 
92 #include <asm/pgtable.h>
93 #include <asm/pgalloc.h>
94 #include <linux/uaccess.h>
95 #include <asm/mmu_context.h>
96 #include <asm/cacheflush.h>
97 #include <asm/tlbflush.h>
98 
99 #include <trace/events/sched.h>
100 
101 #define CREATE_TRACE_POINTS
102 #include <trace/events/task.h>
103 
104 /*
105  * Minimum number of threads to boot the kernel
106  */
107 #define MIN_THREADS 20
108 
109 /*
110  * Maximum number of threads
111  */
112 #define MAX_THREADS FUTEX_TID_MASK
113 
114 /*
115  * Protected counters by write_lock_irq(&tasklist_lock)
116  */
117 unsigned long total_forks;	/* Handle normal Linux uptimes. */
118 int nr_threads;			/* The idle threads do not count.. */
119 
120 int max_threads;		/* tunable limit on nr_threads */
121 
122 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
123 
124 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
125 
126 #ifdef CONFIG_PROVE_RCU
127 int lockdep_tasklist_lock_is_held(void)
128 {
129 	return lockdep_is_held(&tasklist_lock);
130 }
131 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
132 #endif /* #ifdef CONFIG_PROVE_RCU */
133 
134 int nr_processes(void)
135 {
136 	int cpu;
137 	int total = 0;
138 
139 	for_each_possible_cpu(cpu)
140 		total += per_cpu(process_counts, cpu);
141 
142 	return total;
143 }
144 
145 void __weak arch_release_task_struct(struct task_struct *tsk)
146 {
147 }
148 
149 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
150 static struct kmem_cache *task_struct_cachep;
151 
152 static inline struct task_struct *alloc_task_struct_node(int node)
153 {
154 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
155 }
156 
157 static inline void free_task_struct(struct task_struct *tsk)
158 {
159 	kmem_cache_free(task_struct_cachep, tsk);
160 }
161 #endif
162 
163 void __weak arch_release_thread_stack(unsigned long *stack)
164 {
165 }
166 
167 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
168 
169 /*
170  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
171  * kmemcache based allocator.
172  */
173 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
174 
175 #ifdef CONFIG_VMAP_STACK
176 /*
177  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
178  * flush.  Try to minimize the number of calls by caching stacks.
179  */
180 #define NR_CACHED_STACKS 2
181 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
182 
183 static int free_vm_stack_cache(unsigned int cpu)
184 {
185 	struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
186 	int i;
187 
188 	for (i = 0; i < NR_CACHED_STACKS; i++) {
189 		struct vm_struct *vm_stack = cached_vm_stacks[i];
190 
191 		if (!vm_stack)
192 			continue;
193 
194 		vfree(vm_stack->addr);
195 		cached_vm_stacks[i] = NULL;
196 	}
197 
198 	return 0;
199 }
200 #endif
201 
202 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
203 {
204 #ifdef CONFIG_VMAP_STACK
205 	void *stack;
206 	int i;
207 
208 	for (i = 0; i < NR_CACHED_STACKS; i++) {
209 		struct vm_struct *s;
210 
211 		s = this_cpu_xchg(cached_stacks[i], NULL);
212 
213 		if (!s)
214 			continue;
215 
216 		tsk->stack_vm_area = s;
217 		return s->addr;
218 	}
219 
220 	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
221 				     VMALLOC_START, VMALLOC_END,
222 				     THREADINFO_GFP,
223 				     PAGE_KERNEL,
224 				     0, node, __builtin_return_address(0));
225 
226 	/*
227 	 * We can't call find_vm_area() in interrupt context, and
228 	 * free_thread_stack() can be called in interrupt context,
229 	 * so cache the vm_struct.
230 	 */
231 	if (stack)
232 		tsk->stack_vm_area = find_vm_area(stack);
233 	return stack;
234 #else
235 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
236 					     THREAD_SIZE_ORDER);
237 
238 	return page ? page_address(page) : NULL;
239 #endif
240 }
241 
242 static inline void free_thread_stack(struct task_struct *tsk)
243 {
244 #ifdef CONFIG_VMAP_STACK
245 	if (task_stack_vm_area(tsk)) {
246 		int i;
247 
248 		for (i = 0; i < NR_CACHED_STACKS; i++) {
249 			if (this_cpu_cmpxchg(cached_stacks[i],
250 					NULL, tsk->stack_vm_area) != NULL)
251 				continue;
252 
253 			return;
254 		}
255 
256 		vfree_atomic(tsk->stack);
257 		return;
258 	}
259 #endif
260 
261 	__free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
262 }
263 # else
264 static struct kmem_cache *thread_stack_cache;
265 
266 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
267 						  int node)
268 {
269 	return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
270 }
271 
272 static void free_thread_stack(struct task_struct *tsk)
273 {
274 	kmem_cache_free(thread_stack_cache, tsk->stack);
275 }
276 
277 void thread_stack_cache_init(void)
278 {
279 	thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
280 					      THREAD_SIZE, 0, NULL);
281 	BUG_ON(thread_stack_cache == NULL);
282 }
283 # endif
284 #endif
285 
286 /* SLAB cache for signal_struct structures (tsk->signal) */
287 static struct kmem_cache *signal_cachep;
288 
289 /* SLAB cache for sighand_struct structures (tsk->sighand) */
290 struct kmem_cache *sighand_cachep;
291 
292 /* SLAB cache for files_struct structures (tsk->files) */
293 struct kmem_cache *files_cachep;
294 
295 /* SLAB cache for fs_struct structures (tsk->fs) */
296 struct kmem_cache *fs_cachep;
297 
298 /* SLAB cache for vm_area_struct structures */
299 struct kmem_cache *vm_area_cachep;
300 
301 /* SLAB cache for mm_struct structures (tsk->mm) */
302 static struct kmem_cache *mm_cachep;
303 
304 static void account_kernel_stack(struct task_struct *tsk, int account)
305 {
306 	void *stack = task_stack_page(tsk);
307 	struct vm_struct *vm = task_stack_vm_area(tsk);
308 
309 	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
310 
311 	if (vm) {
312 		int i;
313 
314 		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
315 
316 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
317 			mod_zone_page_state(page_zone(vm->pages[i]),
318 					    NR_KERNEL_STACK_KB,
319 					    PAGE_SIZE / 1024 * account);
320 		}
321 
322 		/* All stack pages belong to the same memcg. */
323 		mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
324 				     account * (THREAD_SIZE / 1024));
325 	} else {
326 		/*
327 		 * All stack pages are in the same zone and belong to the
328 		 * same memcg.
329 		 */
330 		struct page *first_page = virt_to_page(stack);
331 
332 		mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
333 				    THREAD_SIZE / 1024 * account);
334 
335 		mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
336 				     account * (THREAD_SIZE / 1024));
337 	}
338 }
339 
340 static void release_task_stack(struct task_struct *tsk)
341 {
342 	if (WARN_ON(tsk->state != TASK_DEAD))
343 		return;  /* Better to leak the stack than to free prematurely */
344 
345 	account_kernel_stack(tsk, -1);
346 	arch_release_thread_stack(tsk->stack);
347 	free_thread_stack(tsk);
348 	tsk->stack = NULL;
349 #ifdef CONFIG_VMAP_STACK
350 	tsk->stack_vm_area = NULL;
351 #endif
352 }
353 
354 #ifdef CONFIG_THREAD_INFO_IN_TASK
355 void put_task_stack(struct task_struct *tsk)
356 {
357 	if (atomic_dec_and_test(&tsk->stack_refcount))
358 		release_task_stack(tsk);
359 }
360 #endif
361 
362 void free_task(struct task_struct *tsk)
363 {
364 #ifndef CONFIG_THREAD_INFO_IN_TASK
365 	/*
366 	 * The task is finally done with both the stack and thread_info,
367 	 * so free both.
368 	 */
369 	release_task_stack(tsk);
370 #else
371 	/*
372 	 * If the task had a separate stack allocation, it should be gone
373 	 * by now.
374 	 */
375 	WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
376 #endif
377 	rt_mutex_debug_task_free(tsk);
378 	ftrace_graph_exit_task(tsk);
379 	put_seccomp_filter(tsk);
380 	arch_release_task_struct(tsk);
381 	if (tsk->flags & PF_KTHREAD)
382 		free_kthread_struct(tsk);
383 	free_task_struct(tsk);
384 }
385 EXPORT_SYMBOL(free_task);
386 
387 static inline void free_signal_struct(struct signal_struct *sig)
388 {
389 	taskstats_tgid_free(sig);
390 	sched_autogroup_exit(sig);
391 	/*
392 	 * __mmdrop is not safe to call from softirq context on x86 due to
393 	 * pgd_dtor so postpone it to the async context
394 	 */
395 	if (sig->oom_mm)
396 		mmdrop_async(sig->oom_mm);
397 	kmem_cache_free(signal_cachep, sig);
398 }
399 
400 static inline void put_signal_struct(struct signal_struct *sig)
401 {
402 	if (atomic_dec_and_test(&sig->sigcnt))
403 		free_signal_struct(sig);
404 }
405 
406 void __put_task_struct(struct task_struct *tsk)
407 {
408 	WARN_ON(!tsk->exit_state);
409 	WARN_ON(atomic_read(&tsk->usage));
410 	WARN_ON(tsk == current);
411 
412 	cgroup_free(tsk);
413 	task_numa_free(tsk);
414 	security_task_free(tsk);
415 	exit_creds(tsk);
416 	delayacct_tsk_free(tsk);
417 	put_signal_struct(tsk->signal);
418 
419 	if (!profile_handoff_task(tsk))
420 		free_task(tsk);
421 }
422 EXPORT_SYMBOL_GPL(__put_task_struct);
423 
424 void __init __weak arch_task_cache_init(void) { }
425 
426 /*
427  * set_max_threads
428  */
429 static void set_max_threads(unsigned int max_threads_suggested)
430 {
431 	u64 threads;
432 
433 	/*
434 	 * The number of threads shall be limited such that the thread
435 	 * structures may only consume a small part of the available memory.
436 	 */
437 	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
438 		threads = MAX_THREADS;
439 	else
440 		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
441 				    (u64) THREAD_SIZE * 8UL);
442 
443 	if (threads > max_threads_suggested)
444 		threads = max_threads_suggested;
445 
446 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
447 }
448 
449 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
450 /* Initialized by the architecture: */
451 int arch_task_struct_size __read_mostly;
452 #endif
453 
454 void __init fork_init(void)
455 {
456 	int i;
457 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
458 #ifndef ARCH_MIN_TASKALIGN
459 #define ARCH_MIN_TASKALIGN	0
460 #endif
461 	int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
462 
463 	/* create a slab on which task_structs can be allocated */
464 	task_struct_cachep = kmem_cache_create("task_struct",
465 			arch_task_struct_size, align,
466 			SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
467 #endif
468 
469 	/* do the arch specific task caches init */
470 	arch_task_cache_init();
471 
472 	set_max_threads(MAX_THREADS);
473 
474 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
475 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
476 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
477 		init_task.signal->rlim[RLIMIT_NPROC];
478 
479 	for (i = 0; i < UCOUNT_COUNTS; i++) {
480 		init_user_ns.ucount_max[i] = max_threads/2;
481 	}
482 
483 #ifdef CONFIG_VMAP_STACK
484 	cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
485 			  NULL, free_vm_stack_cache);
486 #endif
487 }
488 
489 int __weak arch_dup_task_struct(struct task_struct *dst,
490 					       struct task_struct *src)
491 {
492 	*dst = *src;
493 	return 0;
494 }
495 
496 void set_task_stack_end_magic(struct task_struct *tsk)
497 {
498 	unsigned long *stackend;
499 
500 	stackend = end_of_stack(tsk);
501 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
502 }
503 
504 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
505 {
506 	struct task_struct *tsk;
507 	unsigned long *stack;
508 	struct vm_struct *stack_vm_area;
509 	int err;
510 
511 	if (node == NUMA_NO_NODE)
512 		node = tsk_fork_get_node(orig);
513 	tsk = alloc_task_struct_node(node);
514 	if (!tsk)
515 		return NULL;
516 
517 	stack = alloc_thread_stack_node(tsk, node);
518 	if (!stack)
519 		goto free_tsk;
520 
521 	stack_vm_area = task_stack_vm_area(tsk);
522 
523 	err = arch_dup_task_struct(tsk, orig);
524 
525 	/*
526 	 * arch_dup_task_struct() clobbers the stack-related fields.  Make
527 	 * sure they're properly initialized before using any stack-related
528 	 * functions again.
529 	 */
530 	tsk->stack = stack;
531 #ifdef CONFIG_VMAP_STACK
532 	tsk->stack_vm_area = stack_vm_area;
533 #endif
534 #ifdef CONFIG_THREAD_INFO_IN_TASK
535 	atomic_set(&tsk->stack_refcount, 1);
536 #endif
537 
538 	if (err)
539 		goto free_stack;
540 
541 #ifdef CONFIG_SECCOMP
542 	/*
543 	 * We must handle setting up seccomp filters once we're under
544 	 * the sighand lock in case orig has changed between now and
545 	 * then. Until then, filter must be NULL to avoid messing up
546 	 * the usage counts on the error path calling free_task.
547 	 */
548 	tsk->seccomp.filter = NULL;
549 #endif
550 
551 	setup_thread_stack(tsk, orig);
552 	clear_user_return_notifier(tsk);
553 	clear_tsk_need_resched(tsk);
554 	set_task_stack_end_magic(tsk);
555 
556 #ifdef CONFIG_CC_STACKPROTECTOR
557 	tsk->stack_canary = get_random_long();
558 #endif
559 
560 	/*
561 	 * One for us, one for whoever does the "release_task()" (usually
562 	 * parent)
563 	 */
564 	atomic_set(&tsk->usage, 2);
565 #ifdef CONFIG_BLK_DEV_IO_TRACE
566 	tsk->btrace_seq = 0;
567 #endif
568 	tsk->splice_pipe = NULL;
569 	tsk->task_frag.page = NULL;
570 	tsk->wake_q.next = NULL;
571 
572 	account_kernel_stack(tsk, 1);
573 
574 	kcov_task_init(tsk);
575 
576 	return tsk;
577 
578 free_stack:
579 	free_thread_stack(tsk);
580 free_tsk:
581 	free_task_struct(tsk);
582 	return NULL;
583 }
584 
585 #ifdef CONFIG_MMU
586 static __latent_entropy int dup_mmap(struct mm_struct *mm,
587 					struct mm_struct *oldmm)
588 {
589 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
590 	struct rb_node **rb_link, *rb_parent;
591 	int retval;
592 	unsigned long charge;
593 	LIST_HEAD(uf);
594 
595 	uprobe_start_dup_mmap();
596 	if (down_write_killable(&oldmm->mmap_sem)) {
597 		retval = -EINTR;
598 		goto fail_uprobe_end;
599 	}
600 	flush_cache_dup_mm(oldmm);
601 	uprobe_dup_mmap(oldmm, mm);
602 	/*
603 	 * Not linked in yet - no deadlock potential:
604 	 */
605 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
606 
607 	/* No ordering required: file already has been exposed. */
608 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
609 
610 	mm->total_vm = oldmm->total_vm;
611 	mm->data_vm = oldmm->data_vm;
612 	mm->exec_vm = oldmm->exec_vm;
613 	mm->stack_vm = oldmm->stack_vm;
614 
615 	rb_link = &mm->mm_rb.rb_node;
616 	rb_parent = NULL;
617 	pprev = &mm->mmap;
618 	retval = ksm_fork(mm, oldmm);
619 	if (retval)
620 		goto out;
621 	retval = khugepaged_fork(mm, oldmm);
622 	if (retval)
623 		goto out;
624 
625 	prev = NULL;
626 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
627 		struct file *file;
628 
629 		if (mpnt->vm_flags & VM_DONTCOPY) {
630 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
631 			continue;
632 		}
633 		charge = 0;
634 		if (mpnt->vm_flags & VM_ACCOUNT) {
635 			unsigned long len = vma_pages(mpnt);
636 
637 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
638 				goto fail_nomem;
639 			charge = len;
640 		}
641 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
642 		if (!tmp)
643 			goto fail_nomem;
644 		*tmp = *mpnt;
645 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
646 		retval = vma_dup_policy(mpnt, tmp);
647 		if (retval)
648 			goto fail_nomem_policy;
649 		tmp->vm_mm = mm;
650 		retval = dup_userfaultfd(tmp, &uf);
651 		if (retval)
652 			goto fail_nomem_anon_vma_fork;
653 		if (anon_vma_fork(tmp, mpnt))
654 			goto fail_nomem_anon_vma_fork;
655 		tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
656 		tmp->vm_next = tmp->vm_prev = NULL;
657 		file = tmp->vm_file;
658 		if (file) {
659 			struct inode *inode = file_inode(file);
660 			struct address_space *mapping = file->f_mapping;
661 
662 			get_file(file);
663 			if (tmp->vm_flags & VM_DENYWRITE)
664 				atomic_dec(&inode->i_writecount);
665 			i_mmap_lock_write(mapping);
666 			if (tmp->vm_flags & VM_SHARED)
667 				atomic_inc(&mapping->i_mmap_writable);
668 			flush_dcache_mmap_lock(mapping);
669 			/* insert tmp into the share list, just after mpnt */
670 			vma_interval_tree_insert_after(tmp, mpnt,
671 					&mapping->i_mmap);
672 			flush_dcache_mmap_unlock(mapping);
673 			i_mmap_unlock_write(mapping);
674 		}
675 
676 		/*
677 		 * Clear hugetlb-related page reserves for children. This only
678 		 * affects MAP_PRIVATE mappings. Faults generated by the child
679 		 * are not guaranteed to succeed, even if read-only
680 		 */
681 		if (is_vm_hugetlb_page(tmp))
682 			reset_vma_resv_huge_pages(tmp);
683 
684 		/*
685 		 * Link in the new vma and copy the page table entries.
686 		 */
687 		*pprev = tmp;
688 		pprev = &tmp->vm_next;
689 		tmp->vm_prev = prev;
690 		prev = tmp;
691 
692 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
693 		rb_link = &tmp->vm_rb.rb_right;
694 		rb_parent = &tmp->vm_rb;
695 
696 		mm->map_count++;
697 		retval = copy_page_range(mm, oldmm, mpnt);
698 
699 		if (tmp->vm_ops && tmp->vm_ops->open)
700 			tmp->vm_ops->open(tmp);
701 
702 		if (retval)
703 			goto out;
704 	}
705 	/* a new mm has just been created */
706 	arch_dup_mmap(oldmm, mm);
707 	retval = 0;
708 out:
709 	up_write(&mm->mmap_sem);
710 	flush_tlb_mm(oldmm);
711 	up_write(&oldmm->mmap_sem);
712 	dup_userfaultfd_complete(&uf);
713 fail_uprobe_end:
714 	uprobe_end_dup_mmap();
715 	return retval;
716 fail_nomem_anon_vma_fork:
717 	mpol_put(vma_policy(tmp));
718 fail_nomem_policy:
719 	kmem_cache_free(vm_area_cachep, tmp);
720 fail_nomem:
721 	retval = -ENOMEM;
722 	vm_unacct_memory(charge);
723 	goto out;
724 }
725 
726 static inline int mm_alloc_pgd(struct mm_struct *mm)
727 {
728 	mm->pgd = pgd_alloc(mm);
729 	if (unlikely(!mm->pgd))
730 		return -ENOMEM;
731 	return 0;
732 }
733 
734 static inline void mm_free_pgd(struct mm_struct *mm)
735 {
736 	pgd_free(mm, mm->pgd);
737 }
738 #else
739 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
740 {
741 	down_write(&oldmm->mmap_sem);
742 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
743 	up_write(&oldmm->mmap_sem);
744 	return 0;
745 }
746 #define mm_alloc_pgd(mm)	(0)
747 #define mm_free_pgd(mm)
748 #endif /* CONFIG_MMU */
749 
750 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
751 
752 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
753 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
754 
755 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
756 
757 static int __init coredump_filter_setup(char *s)
758 {
759 	default_dump_filter =
760 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
761 		MMF_DUMP_FILTER_MASK;
762 	return 1;
763 }
764 
765 __setup("coredump_filter=", coredump_filter_setup);
766 
767 #include <linux/init_task.h>
768 
769 static void mm_init_aio(struct mm_struct *mm)
770 {
771 #ifdef CONFIG_AIO
772 	spin_lock_init(&mm->ioctx_lock);
773 	mm->ioctx_table = NULL;
774 #endif
775 }
776 
777 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
778 {
779 #ifdef CONFIG_MEMCG
780 	mm->owner = p;
781 #endif
782 }
783 
784 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
785 	struct user_namespace *user_ns)
786 {
787 	mm->mmap = NULL;
788 	mm->mm_rb = RB_ROOT;
789 	mm->vmacache_seqnum = 0;
790 	atomic_set(&mm->mm_users, 1);
791 	atomic_set(&mm->mm_count, 1);
792 	init_rwsem(&mm->mmap_sem);
793 	INIT_LIST_HEAD(&mm->mmlist);
794 	mm->core_state = NULL;
795 	atomic_long_set(&mm->nr_ptes, 0);
796 	mm_nr_pmds_init(mm);
797 	mm->map_count = 0;
798 	mm->locked_vm = 0;
799 	mm->pinned_vm = 0;
800 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
801 	spin_lock_init(&mm->page_table_lock);
802 	mm_init_cpumask(mm);
803 	mm_init_aio(mm);
804 	mm_init_owner(mm, p);
805 	mmu_notifier_mm_init(mm);
806 	clear_tlb_flush_pending(mm);
807 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
808 	mm->pmd_huge_pte = NULL;
809 #endif
810 
811 	if (current->mm) {
812 		mm->flags = current->mm->flags & MMF_INIT_MASK;
813 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
814 	} else {
815 		mm->flags = default_dump_filter;
816 		mm->def_flags = 0;
817 	}
818 
819 	if (mm_alloc_pgd(mm))
820 		goto fail_nopgd;
821 
822 	if (init_new_context(p, mm))
823 		goto fail_nocontext;
824 
825 	mm->user_ns = get_user_ns(user_ns);
826 	return mm;
827 
828 fail_nocontext:
829 	mm_free_pgd(mm);
830 fail_nopgd:
831 	free_mm(mm);
832 	return NULL;
833 }
834 
835 static void check_mm(struct mm_struct *mm)
836 {
837 	int i;
838 
839 	for (i = 0; i < NR_MM_COUNTERS; i++) {
840 		long x = atomic_long_read(&mm->rss_stat.count[i]);
841 
842 		if (unlikely(x))
843 			printk(KERN_ALERT "BUG: Bad rss-counter state "
844 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
845 	}
846 
847 	if (atomic_long_read(&mm->nr_ptes))
848 		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
849 				atomic_long_read(&mm->nr_ptes));
850 	if (mm_nr_pmds(mm))
851 		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
852 				mm_nr_pmds(mm));
853 
854 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
855 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
856 #endif
857 }
858 
859 /*
860  * Allocate and initialize an mm_struct.
861  */
862 struct mm_struct *mm_alloc(void)
863 {
864 	struct mm_struct *mm;
865 
866 	mm = allocate_mm();
867 	if (!mm)
868 		return NULL;
869 
870 	memset(mm, 0, sizeof(*mm));
871 	return mm_init(mm, current, current_user_ns());
872 }
873 
874 /*
875  * Called when the last reference to the mm
876  * is dropped: either by a lazy thread or by
877  * mmput. Free the page directory and the mm.
878  */
879 void __mmdrop(struct mm_struct *mm)
880 {
881 	BUG_ON(mm == &init_mm);
882 	mm_free_pgd(mm);
883 	destroy_context(mm);
884 	mmu_notifier_mm_destroy(mm);
885 	check_mm(mm);
886 	put_user_ns(mm->user_ns);
887 	free_mm(mm);
888 }
889 EXPORT_SYMBOL_GPL(__mmdrop);
890 
891 static inline void __mmput(struct mm_struct *mm)
892 {
893 	VM_BUG_ON(atomic_read(&mm->mm_users));
894 
895 	uprobe_clear_state(mm);
896 	exit_aio(mm);
897 	ksm_exit(mm);
898 	khugepaged_exit(mm); /* must run before exit_mmap */
899 	exit_mmap(mm);
900 	mm_put_huge_zero_page(mm);
901 	set_mm_exe_file(mm, NULL);
902 	if (!list_empty(&mm->mmlist)) {
903 		spin_lock(&mmlist_lock);
904 		list_del(&mm->mmlist);
905 		spin_unlock(&mmlist_lock);
906 	}
907 	if (mm->binfmt)
908 		module_put(mm->binfmt->module);
909 	set_bit(MMF_OOM_SKIP, &mm->flags);
910 	mmdrop(mm);
911 }
912 
913 /*
914  * Decrement the use count and release all resources for an mm.
915  */
916 void mmput(struct mm_struct *mm)
917 {
918 	might_sleep();
919 
920 	if (atomic_dec_and_test(&mm->mm_users))
921 		__mmput(mm);
922 }
923 EXPORT_SYMBOL_GPL(mmput);
924 
925 #ifdef CONFIG_MMU
926 static void mmput_async_fn(struct work_struct *work)
927 {
928 	struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
929 	__mmput(mm);
930 }
931 
932 void mmput_async(struct mm_struct *mm)
933 {
934 	if (atomic_dec_and_test(&mm->mm_users)) {
935 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
936 		schedule_work(&mm->async_put_work);
937 	}
938 }
939 #endif
940 
941 /**
942  * set_mm_exe_file - change a reference to the mm's executable file
943  *
944  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
945  *
946  * Main users are mmput() and sys_execve(). Callers prevent concurrent
947  * invocations: in mmput() nobody alive left, in execve task is single
948  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
949  * mm->exe_file, but does so without using set_mm_exe_file() in order
950  * to do avoid the need for any locks.
951  */
952 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
953 {
954 	struct file *old_exe_file;
955 
956 	/*
957 	 * It is safe to dereference the exe_file without RCU as
958 	 * this function is only called if nobody else can access
959 	 * this mm -- see comment above for justification.
960 	 */
961 	old_exe_file = rcu_dereference_raw(mm->exe_file);
962 
963 	if (new_exe_file)
964 		get_file(new_exe_file);
965 	rcu_assign_pointer(mm->exe_file, new_exe_file);
966 	if (old_exe_file)
967 		fput(old_exe_file);
968 }
969 
970 /**
971  * get_mm_exe_file - acquire a reference to the mm's executable file
972  *
973  * Returns %NULL if mm has no associated executable file.
974  * User must release file via fput().
975  */
976 struct file *get_mm_exe_file(struct mm_struct *mm)
977 {
978 	struct file *exe_file;
979 
980 	rcu_read_lock();
981 	exe_file = rcu_dereference(mm->exe_file);
982 	if (exe_file && !get_file_rcu(exe_file))
983 		exe_file = NULL;
984 	rcu_read_unlock();
985 	return exe_file;
986 }
987 EXPORT_SYMBOL(get_mm_exe_file);
988 
989 /**
990  * get_task_exe_file - acquire a reference to the task's executable file
991  *
992  * Returns %NULL if task's mm (if any) has no associated executable file or
993  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
994  * User must release file via fput().
995  */
996 struct file *get_task_exe_file(struct task_struct *task)
997 {
998 	struct file *exe_file = NULL;
999 	struct mm_struct *mm;
1000 
1001 	task_lock(task);
1002 	mm = task->mm;
1003 	if (mm) {
1004 		if (!(task->flags & PF_KTHREAD))
1005 			exe_file = get_mm_exe_file(mm);
1006 	}
1007 	task_unlock(task);
1008 	return exe_file;
1009 }
1010 EXPORT_SYMBOL(get_task_exe_file);
1011 
1012 /**
1013  * get_task_mm - acquire a reference to the task's mm
1014  *
1015  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1016  * this kernel workthread has transiently adopted a user mm with use_mm,
1017  * to do its AIO) is not set and if so returns a reference to it, after
1018  * bumping up the use count.  User must release the mm via mmput()
1019  * after use.  Typically used by /proc and ptrace.
1020  */
1021 struct mm_struct *get_task_mm(struct task_struct *task)
1022 {
1023 	struct mm_struct *mm;
1024 
1025 	task_lock(task);
1026 	mm = task->mm;
1027 	if (mm) {
1028 		if (task->flags & PF_KTHREAD)
1029 			mm = NULL;
1030 		else
1031 			mmget(mm);
1032 	}
1033 	task_unlock(task);
1034 	return mm;
1035 }
1036 EXPORT_SYMBOL_GPL(get_task_mm);
1037 
1038 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1039 {
1040 	struct mm_struct *mm;
1041 	int err;
1042 
1043 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1044 	if (err)
1045 		return ERR_PTR(err);
1046 
1047 	mm = get_task_mm(task);
1048 	if (mm && mm != current->mm &&
1049 			!ptrace_may_access(task, mode)) {
1050 		mmput(mm);
1051 		mm = ERR_PTR(-EACCES);
1052 	}
1053 	mutex_unlock(&task->signal->cred_guard_mutex);
1054 
1055 	return mm;
1056 }
1057 
1058 static void complete_vfork_done(struct task_struct *tsk)
1059 {
1060 	struct completion *vfork;
1061 
1062 	task_lock(tsk);
1063 	vfork = tsk->vfork_done;
1064 	if (likely(vfork)) {
1065 		tsk->vfork_done = NULL;
1066 		complete(vfork);
1067 	}
1068 	task_unlock(tsk);
1069 }
1070 
1071 static int wait_for_vfork_done(struct task_struct *child,
1072 				struct completion *vfork)
1073 {
1074 	int killed;
1075 
1076 	freezer_do_not_count();
1077 	killed = wait_for_completion_killable(vfork);
1078 	freezer_count();
1079 
1080 	if (killed) {
1081 		task_lock(child);
1082 		child->vfork_done = NULL;
1083 		task_unlock(child);
1084 	}
1085 
1086 	put_task_struct(child);
1087 	return killed;
1088 }
1089 
1090 /* Please note the differences between mmput and mm_release.
1091  * mmput is called whenever we stop holding onto a mm_struct,
1092  * error success whatever.
1093  *
1094  * mm_release is called after a mm_struct has been removed
1095  * from the current process.
1096  *
1097  * This difference is important for error handling, when we
1098  * only half set up a mm_struct for a new process and need to restore
1099  * the old one.  Because we mmput the new mm_struct before
1100  * restoring the old one. . .
1101  * Eric Biederman 10 January 1998
1102  */
1103 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1104 {
1105 	/* Get rid of any futexes when releasing the mm */
1106 #ifdef CONFIG_FUTEX
1107 	if (unlikely(tsk->robust_list)) {
1108 		exit_robust_list(tsk);
1109 		tsk->robust_list = NULL;
1110 	}
1111 #ifdef CONFIG_COMPAT
1112 	if (unlikely(tsk->compat_robust_list)) {
1113 		compat_exit_robust_list(tsk);
1114 		tsk->compat_robust_list = NULL;
1115 	}
1116 #endif
1117 	if (unlikely(!list_empty(&tsk->pi_state_list)))
1118 		exit_pi_state_list(tsk);
1119 #endif
1120 
1121 	uprobe_free_utask(tsk);
1122 
1123 	/* Get rid of any cached register state */
1124 	deactivate_mm(tsk, mm);
1125 
1126 	/*
1127 	 * Signal userspace if we're not exiting with a core dump
1128 	 * because we want to leave the value intact for debugging
1129 	 * purposes.
1130 	 */
1131 	if (tsk->clear_child_tid) {
1132 		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1133 		    atomic_read(&mm->mm_users) > 1) {
1134 			/*
1135 			 * We don't check the error code - if userspace has
1136 			 * not set up a proper pointer then tough luck.
1137 			 */
1138 			put_user(0, tsk->clear_child_tid);
1139 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1140 					1, NULL, NULL, 0);
1141 		}
1142 		tsk->clear_child_tid = NULL;
1143 	}
1144 
1145 	/*
1146 	 * All done, finally we can wake up parent and return this mm to him.
1147 	 * Also kthread_stop() uses this completion for synchronization.
1148 	 */
1149 	if (tsk->vfork_done)
1150 		complete_vfork_done(tsk);
1151 }
1152 
1153 /*
1154  * Allocate a new mm structure and copy contents from the
1155  * mm structure of the passed in task structure.
1156  */
1157 static struct mm_struct *dup_mm(struct task_struct *tsk)
1158 {
1159 	struct mm_struct *mm, *oldmm = current->mm;
1160 	int err;
1161 
1162 	mm = allocate_mm();
1163 	if (!mm)
1164 		goto fail_nomem;
1165 
1166 	memcpy(mm, oldmm, sizeof(*mm));
1167 
1168 	if (!mm_init(mm, tsk, mm->user_ns))
1169 		goto fail_nomem;
1170 
1171 	err = dup_mmap(mm, oldmm);
1172 	if (err)
1173 		goto free_pt;
1174 
1175 	mm->hiwater_rss = get_mm_rss(mm);
1176 	mm->hiwater_vm = mm->total_vm;
1177 
1178 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1179 		goto free_pt;
1180 
1181 	return mm;
1182 
1183 free_pt:
1184 	/* don't put binfmt in mmput, we haven't got module yet */
1185 	mm->binfmt = NULL;
1186 	mmput(mm);
1187 
1188 fail_nomem:
1189 	return NULL;
1190 }
1191 
1192 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1193 {
1194 	struct mm_struct *mm, *oldmm;
1195 	int retval;
1196 
1197 	tsk->min_flt = tsk->maj_flt = 0;
1198 	tsk->nvcsw = tsk->nivcsw = 0;
1199 #ifdef CONFIG_DETECT_HUNG_TASK
1200 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1201 #endif
1202 
1203 	tsk->mm = NULL;
1204 	tsk->active_mm = NULL;
1205 
1206 	/*
1207 	 * Are we cloning a kernel thread?
1208 	 *
1209 	 * We need to steal a active VM for that..
1210 	 */
1211 	oldmm = current->mm;
1212 	if (!oldmm)
1213 		return 0;
1214 
1215 	/* initialize the new vmacache entries */
1216 	vmacache_flush(tsk);
1217 
1218 	if (clone_flags & CLONE_VM) {
1219 		mmget(oldmm);
1220 		mm = oldmm;
1221 		goto good_mm;
1222 	}
1223 
1224 	retval = -ENOMEM;
1225 	mm = dup_mm(tsk);
1226 	if (!mm)
1227 		goto fail_nomem;
1228 
1229 good_mm:
1230 	tsk->mm = mm;
1231 	tsk->active_mm = mm;
1232 	return 0;
1233 
1234 fail_nomem:
1235 	return retval;
1236 }
1237 
1238 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1239 {
1240 	struct fs_struct *fs = current->fs;
1241 	if (clone_flags & CLONE_FS) {
1242 		/* tsk->fs is already what we want */
1243 		spin_lock(&fs->lock);
1244 		if (fs->in_exec) {
1245 			spin_unlock(&fs->lock);
1246 			return -EAGAIN;
1247 		}
1248 		fs->users++;
1249 		spin_unlock(&fs->lock);
1250 		return 0;
1251 	}
1252 	tsk->fs = copy_fs_struct(fs);
1253 	if (!tsk->fs)
1254 		return -ENOMEM;
1255 	return 0;
1256 }
1257 
1258 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1259 {
1260 	struct files_struct *oldf, *newf;
1261 	int error = 0;
1262 
1263 	/*
1264 	 * A background process may not have any files ...
1265 	 */
1266 	oldf = current->files;
1267 	if (!oldf)
1268 		goto out;
1269 
1270 	if (clone_flags & CLONE_FILES) {
1271 		atomic_inc(&oldf->count);
1272 		goto out;
1273 	}
1274 
1275 	newf = dup_fd(oldf, &error);
1276 	if (!newf)
1277 		goto out;
1278 
1279 	tsk->files = newf;
1280 	error = 0;
1281 out:
1282 	return error;
1283 }
1284 
1285 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1286 {
1287 #ifdef CONFIG_BLOCK
1288 	struct io_context *ioc = current->io_context;
1289 	struct io_context *new_ioc;
1290 
1291 	if (!ioc)
1292 		return 0;
1293 	/*
1294 	 * Share io context with parent, if CLONE_IO is set
1295 	 */
1296 	if (clone_flags & CLONE_IO) {
1297 		ioc_task_link(ioc);
1298 		tsk->io_context = ioc;
1299 	} else if (ioprio_valid(ioc->ioprio)) {
1300 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1301 		if (unlikely(!new_ioc))
1302 			return -ENOMEM;
1303 
1304 		new_ioc->ioprio = ioc->ioprio;
1305 		put_io_context(new_ioc);
1306 	}
1307 #endif
1308 	return 0;
1309 }
1310 
1311 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1312 {
1313 	struct sighand_struct *sig;
1314 
1315 	if (clone_flags & CLONE_SIGHAND) {
1316 		atomic_inc(&current->sighand->count);
1317 		return 0;
1318 	}
1319 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1320 	rcu_assign_pointer(tsk->sighand, sig);
1321 	if (!sig)
1322 		return -ENOMEM;
1323 
1324 	atomic_set(&sig->count, 1);
1325 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1326 	return 0;
1327 }
1328 
1329 void __cleanup_sighand(struct sighand_struct *sighand)
1330 {
1331 	if (atomic_dec_and_test(&sighand->count)) {
1332 		signalfd_cleanup(sighand);
1333 		/*
1334 		 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1335 		 * without an RCU grace period, see __lock_task_sighand().
1336 		 */
1337 		kmem_cache_free(sighand_cachep, sighand);
1338 	}
1339 }
1340 
1341 #ifdef CONFIG_POSIX_TIMERS
1342 /*
1343  * Initialize POSIX timer handling for a thread group.
1344  */
1345 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1346 {
1347 	unsigned long cpu_limit;
1348 
1349 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1350 	if (cpu_limit != RLIM_INFINITY) {
1351 		sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1352 		sig->cputimer.running = true;
1353 	}
1354 
1355 	/* The timer lists. */
1356 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1357 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1358 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1359 }
1360 #else
1361 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1362 #endif
1363 
1364 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1365 {
1366 	struct signal_struct *sig;
1367 
1368 	if (clone_flags & CLONE_THREAD)
1369 		return 0;
1370 
1371 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1372 	tsk->signal = sig;
1373 	if (!sig)
1374 		return -ENOMEM;
1375 
1376 	sig->nr_threads = 1;
1377 	atomic_set(&sig->live, 1);
1378 	atomic_set(&sig->sigcnt, 1);
1379 
1380 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1381 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1382 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1383 
1384 	init_waitqueue_head(&sig->wait_chldexit);
1385 	sig->curr_target = tsk;
1386 	init_sigpending(&sig->shared_pending);
1387 	seqlock_init(&sig->stats_lock);
1388 	prev_cputime_init(&sig->prev_cputime);
1389 
1390 #ifdef CONFIG_POSIX_TIMERS
1391 	INIT_LIST_HEAD(&sig->posix_timers);
1392 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1393 	sig->real_timer.function = it_real_fn;
1394 #endif
1395 
1396 	task_lock(current->group_leader);
1397 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1398 	task_unlock(current->group_leader);
1399 
1400 	posix_cpu_timers_init_group(sig);
1401 
1402 	tty_audit_fork(sig);
1403 	sched_autogroup_fork(sig);
1404 
1405 	sig->oom_score_adj = current->signal->oom_score_adj;
1406 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1407 
1408 	mutex_init(&sig->cred_guard_mutex);
1409 
1410 	return 0;
1411 }
1412 
1413 static void copy_seccomp(struct task_struct *p)
1414 {
1415 #ifdef CONFIG_SECCOMP
1416 	/*
1417 	 * Must be called with sighand->lock held, which is common to
1418 	 * all threads in the group. Holding cred_guard_mutex is not
1419 	 * needed because this new task is not yet running and cannot
1420 	 * be racing exec.
1421 	 */
1422 	assert_spin_locked(&current->sighand->siglock);
1423 
1424 	/* Ref-count the new filter user, and assign it. */
1425 	get_seccomp_filter(current);
1426 	p->seccomp = current->seccomp;
1427 
1428 	/*
1429 	 * Explicitly enable no_new_privs here in case it got set
1430 	 * between the task_struct being duplicated and holding the
1431 	 * sighand lock. The seccomp state and nnp must be in sync.
1432 	 */
1433 	if (task_no_new_privs(current))
1434 		task_set_no_new_privs(p);
1435 
1436 	/*
1437 	 * If the parent gained a seccomp mode after copying thread
1438 	 * flags and between before we held the sighand lock, we have
1439 	 * to manually enable the seccomp thread flag here.
1440 	 */
1441 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1442 		set_tsk_thread_flag(p, TIF_SECCOMP);
1443 #endif
1444 }
1445 
1446 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1447 {
1448 	current->clear_child_tid = tidptr;
1449 
1450 	return task_pid_vnr(current);
1451 }
1452 
1453 static void rt_mutex_init_task(struct task_struct *p)
1454 {
1455 	raw_spin_lock_init(&p->pi_lock);
1456 #ifdef CONFIG_RT_MUTEXES
1457 	p->pi_waiters = RB_ROOT;
1458 	p->pi_waiters_leftmost = NULL;
1459 	p->pi_top_task = NULL;
1460 	p->pi_blocked_on = NULL;
1461 #endif
1462 }
1463 
1464 #ifdef CONFIG_POSIX_TIMERS
1465 /*
1466  * Initialize POSIX timer handling for a single task.
1467  */
1468 static void posix_cpu_timers_init(struct task_struct *tsk)
1469 {
1470 	tsk->cputime_expires.prof_exp = 0;
1471 	tsk->cputime_expires.virt_exp = 0;
1472 	tsk->cputime_expires.sched_exp = 0;
1473 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1474 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1475 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1476 }
1477 #else
1478 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1479 #endif
1480 
1481 static inline void
1482 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1483 {
1484 	 task->pids[type].pid = pid;
1485 }
1486 
1487 static inline void rcu_copy_process(struct task_struct *p)
1488 {
1489 #ifdef CONFIG_PREEMPT_RCU
1490 	p->rcu_read_lock_nesting = 0;
1491 	p->rcu_read_unlock_special.s = 0;
1492 	p->rcu_blocked_node = NULL;
1493 	INIT_LIST_HEAD(&p->rcu_node_entry);
1494 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1495 #ifdef CONFIG_TASKS_RCU
1496 	p->rcu_tasks_holdout = false;
1497 	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1498 	p->rcu_tasks_idle_cpu = -1;
1499 #endif /* #ifdef CONFIG_TASKS_RCU */
1500 }
1501 
1502 /*
1503  * This creates a new process as a copy of the old one,
1504  * but does not actually start it yet.
1505  *
1506  * It copies the registers, and all the appropriate
1507  * parts of the process environment (as per the clone
1508  * flags). The actual kick-off is left to the caller.
1509  */
1510 static __latent_entropy struct task_struct *copy_process(
1511 					unsigned long clone_flags,
1512 					unsigned long stack_start,
1513 					unsigned long stack_size,
1514 					int __user *child_tidptr,
1515 					struct pid *pid,
1516 					int trace,
1517 					unsigned long tls,
1518 					int node)
1519 {
1520 	int retval;
1521 	struct task_struct *p;
1522 
1523 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1524 		return ERR_PTR(-EINVAL);
1525 
1526 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1527 		return ERR_PTR(-EINVAL);
1528 
1529 	/*
1530 	 * Thread groups must share signals as well, and detached threads
1531 	 * can only be started up within the thread group.
1532 	 */
1533 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1534 		return ERR_PTR(-EINVAL);
1535 
1536 	/*
1537 	 * Shared signal handlers imply shared VM. By way of the above,
1538 	 * thread groups also imply shared VM. Blocking this case allows
1539 	 * for various simplifications in other code.
1540 	 */
1541 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1542 		return ERR_PTR(-EINVAL);
1543 
1544 	/*
1545 	 * Siblings of global init remain as zombies on exit since they are
1546 	 * not reaped by their parent (swapper). To solve this and to avoid
1547 	 * multi-rooted process trees, prevent global and container-inits
1548 	 * from creating siblings.
1549 	 */
1550 	if ((clone_flags & CLONE_PARENT) &&
1551 				current->signal->flags & SIGNAL_UNKILLABLE)
1552 		return ERR_PTR(-EINVAL);
1553 
1554 	/*
1555 	 * If the new process will be in a different pid or user namespace
1556 	 * do not allow it to share a thread group with the forking task.
1557 	 */
1558 	if (clone_flags & CLONE_THREAD) {
1559 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1560 		    (task_active_pid_ns(current) !=
1561 				current->nsproxy->pid_ns_for_children))
1562 			return ERR_PTR(-EINVAL);
1563 	}
1564 
1565 	retval = security_task_create(clone_flags);
1566 	if (retval)
1567 		goto fork_out;
1568 
1569 	retval = -ENOMEM;
1570 	p = dup_task_struct(current, node);
1571 	if (!p)
1572 		goto fork_out;
1573 
1574 	/*
1575 	 * This _must_ happen before we call free_task(), i.e. before we jump
1576 	 * to any of the bad_fork_* labels. This is to avoid freeing
1577 	 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1578 	 * kernel threads (PF_KTHREAD).
1579 	 */
1580 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1581 	/*
1582 	 * Clear TID on mm_release()?
1583 	 */
1584 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1585 
1586 	ftrace_graph_init_task(p);
1587 
1588 	rt_mutex_init_task(p);
1589 
1590 #ifdef CONFIG_PROVE_LOCKING
1591 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1592 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1593 #endif
1594 	retval = -EAGAIN;
1595 	if (atomic_read(&p->real_cred->user->processes) >=
1596 			task_rlimit(p, RLIMIT_NPROC)) {
1597 		if (p->real_cred->user != INIT_USER &&
1598 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1599 			goto bad_fork_free;
1600 	}
1601 	current->flags &= ~PF_NPROC_EXCEEDED;
1602 
1603 	retval = copy_creds(p, clone_flags);
1604 	if (retval < 0)
1605 		goto bad_fork_free;
1606 
1607 	/*
1608 	 * If multiple threads are within copy_process(), then this check
1609 	 * triggers too late. This doesn't hurt, the check is only there
1610 	 * to stop root fork bombs.
1611 	 */
1612 	retval = -EAGAIN;
1613 	if (nr_threads >= max_threads)
1614 		goto bad_fork_cleanup_count;
1615 
1616 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1617 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1618 	p->flags |= PF_FORKNOEXEC;
1619 	INIT_LIST_HEAD(&p->children);
1620 	INIT_LIST_HEAD(&p->sibling);
1621 	rcu_copy_process(p);
1622 	p->vfork_done = NULL;
1623 	spin_lock_init(&p->alloc_lock);
1624 
1625 	init_sigpending(&p->pending);
1626 
1627 	p->utime = p->stime = p->gtime = 0;
1628 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1629 	p->utimescaled = p->stimescaled = 0;
1630 #endif
1631 	prev_cputime_init(&p->prev_cputime);
1632 
1633 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1634 	seqcount_init(&p->vtime.seqcount);
1635 	p->vtime.starttime = 0;
1636 	p->vtime.state = VTIME_INACTIVE;
1637 #endif
1638 
1639 #if defined(SPLIT_RSS_COUNTING)
1640 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1641 #endif
1642 
1643 	p->default_timer_slack_ns = current->timer_slack_ns;
1644 
1645 	task_io_accounting_init(&p->ioac);
1646 	acct_clear_integrals(p);
1647 
1648 	posix_cpu_timers_init(p);
1649 
1650 	p->start_time = ktime_get_ns();
1651 	p->real_start_time = ktime_get_boot_ns();
1652 	p->io_context = NULL;
1653 	p->audit_context = NULL;
1654 	cgroup_fork(p);
1655 #ifdef CONFIG_NUMA
1656 	p->mempolicy = mpol_dup(p->mempolicy);
1657 	if (IS_ERR(p->mempolicy)) {
1658 		retval = PTR_ERR(p->mempolicy);
1659 		p->mempolicy = NULL;
1660 		goto bad_fork_cleanup_threadgroup_lock;
1661 	}
1662 #endif
1663 #ifdef CONFIG_CPUSETS
1664 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1665 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1666 	seqcount_init(&p->mems_allowed_seq);
1667 #endif
1668 #ifdef CONFIG_TRACE_IRQFLAGS
1669 	p->irq_events = 0;
1670 	p->hardirqs_enabled = 0;
1671 	p->hardirq_enable_ip = 0;
1672 	p->hardirq_enable_event = 0;
1673 	p->hardirq_disable_ip = _THIS_IP_;
1674 	p->hardirq_disable_event = 0;
1675 	p->softirqs_enabled = 1;
1676 	p->softirq_enable_ip = _THIS_IP_;
1677 	p->softirq_enable_event = 0;
1678 	p->softirq_disable_ip = 0;
1679 	p->softirq_disable_event = 0;
1680 	p->hardirq_context = 0;
1681 	p->softirq_context = 0;
1682 #endif
1683 
1684 	p->pagefault_disabled = 0;
1685 
1686 #ifdef CONFIG_LOCKDEP
1687 	p->lockdep_depth = 0; /* no locks held yet */
1688 	p->curr_chain_key = 0;
1689 	p->lockdep_recursion = 0;
1690 #endif
1691 
1692 #ifdef CONFIG_DEBUG_MUTEXES
1693 	p->blocked_on = NULL; /* not blocked yet */
1694 #endif
1695 #ifdef CONFIG_BCACHE
1696 	p->sequential_io	= 0;
1697 	p->sequential_io_avg	= 0;
1698 #endif
1699 
1700 	/* Perform scheduler related setup. Assign this task to a CPU. */
1701 	retval = sched_fork(clone_flags, p);
1702 	if (retval)
1703 		goto bad_fork_cleanup_policy;
1704 
1705 	retval = perf_event_init_task(p);
1706 	if (retval)
1707 		goto bad_fork_cleanup_policy;
1708 	retval = audit_alloc(p);
1709 	if (retval)
1710 		goto bad_fork_cleanup_perf;
1711 	/* copy all the process information */
1712 	shm_init_task(p);
1713 	retval = security_task_alloc(p, clone_flags);
1714 	if (retval)
1715 		goto bad_fork_cleanup_audit;
1716 	retval = copy_semundo(clone_flags, p);
1717 	if (retval)
1718 		goto bad_fork_cleanup_security;
1719 	retval = copy_files(clone_flags, p);
1720 	if (retval)
1721 		goto bad_fork_cleanup_semundo;
1722 	retval = copy_fs(clone_flags, p);
1723 	if (retval)
1724 		goto bad_fork_cleanup_files;
1725 	retval = copy_sighand(clone_flags, p);
1726 	if (retval)
1727 		goto bad_fork_cleanup_fs;
1728 	retval = copy_signal(clone_flags, p);
1729 	if (retval)
1730 		goto bad_fork_cleanup_sighand;
1731 	retval = copy_mm(clone_flags, p);
1732 	if (retval)
1733 		goto bad_fork_cleanup_signal;
1734 	retval = copy_namespaces(clone_flags, p);
1735 	if (retval)
1736 		goto bad_fork_cleanup_mm;
1737 	retval = copy_io(clone_flags, p);
1738 	if (retval)
1739 		goto bad_fork_cleanup_namespaces;
1740 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1741 	if (retval)
1742 		goto bad_fork_cleanup_io;
1743 
1744 	if (pid != &init_struct_pid) {
1745 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1746 		if (IS_ERR(pid)) {
1747 			retval = PTR_ERR(pid);
1748 			goto bad_fork_cleanup_thread;
1749 		}
1750 	}
1751 
1752 #ifdef CONFIG_BLOCK
1753 	p->plug = NULL;
1754 #endif
1755 #ifdef CONFIG_FUTEX
1756 	p->robust_list = NULL;
1757 #ifdef CONFIG_COMPAT
1758 	p->compat_robust_list = NULL;
1759 #endif
1760 	INIT_LIST_HEAD(&p->pi_state_list);
1761 	p->pi_state_cache = NULL;
1762 #endif
1763 	/*
1764 	 * sigaltstack should be cleared when sharing the same VM
1765 	 */
1766 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1767 		sas_ss_reset(p);
1768 
1769 	/*
1770 	 * Syscall tracing and stepping should be turned off in the
1771 	 * child regardless of CLONE_PTRACE.
1772 	 */
1773 	user_disable_single_step(p);
1774 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1775 #ifdef TIF_SYSCALL_EMU
1776 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1777 #endif
1778 	clear_all_latency_tracing(p);
1779 
1780 	/* ok, now we should be set up.. */
1781 	p->pid = pid_nr(pid);
1782 	if (clone_flags & CLONE_THREAD) {
1783 		p->exit_signal = -1;
1784 		p->group_leader = current->group_leader;
1785 		p->tgid = current->tgid;
1786 	} else {
1787 		if (clone_flags & CLONE_PARENT)
1788 			p->exit_signal = current->group_leader->exit_signal;
1789 		else
1790 			p->exit_signal = (clone_flags & CSIGNAL);
1791 		p->group_leader = p;
1792 		p->tgid = p->pid;
1793 	}
1794 
1795 	p->nr_dirtied = 0;
1796 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1797 	p->dirty_paused_when = 0;
1798 
1799 	p->pdeath_signal = 0;
1800 	INIT_LIST_HEAD(&p->thread_group);
1801 	p->task_works = NULL;
1802 
1803 	cgroup_threadgroup_change_begin(current);
1804 	/*
1805 	 * Ensure that the cgroup subsystem policies allow the new process to be
1806 	 * forked. It should be noted the the new process's css_set can be changed
1807 	 * between here and cgroup_post_fork() if an organisation operation is in
1808 	 * progress.
1809 	 */
1810 	retval = cgroup_can_fork(p);
1811 	if (retval)
1812 		goto bad_fork_free_pid;
1813 
1814 	/*
1815 	 * Make it visible to the rest of the system, but dont wake it up yet.
1816 	 * Need tasklist lock for parent etc handling!
1817 	 */
1818 	write_lock_irq(&tasklist_lock);
1819 
1820 	/* CLONE_PARENT re-uses the old parent */
1821 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1822 		p->real_parent = current->real_parent;
1823 		p->parent_exec_id = current->parent_exec_id;
1824 	} else {
1825 		p->real_parent = current;
1826 		p->parent_exec_id = current->self_exec_id;
1827 	}
1828 
1829 	klp_copy_process(p);
1830 
1831 	spin_lock(&current->sighand->siglock);
1832 
1833 	/*
1834 	 * Copy seccomp details explicitly here, in case they were changed
1835 	 * before holding sighand lock.
1836 	 */
1837 	copy_seccomp(p);
1838 
1839 	/*
1840 	 * Process group and session signals need to be delivered to just the
1841 	 * parent before the fork or both the parent and the child after the
1842 	 * fork. Restart if a signal comes in before we add the new process to
1843 	 * it's process group.
1844 	 * A fatal signal pending means that current will exit, so the new
1845 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1846 	*/
1847 	recalc_sigpending();
1848 	if (signal_pending(current)) {
1849 		retval = -ERESTARTNOINTR;
1850 		goto bad_fork_cancel_cgroup;
1851 	}
1852 	if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1853 		retval = -ENOMEM;
1854 		goto bad_fork_cancel_cgroup;
1855 	}
1856 
1857 	if (likely(p->pid)) {
1858 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1859 
1860 		init_task_pid(p, PIDTYPE_PID, pid);
1861 		if (thread_group_leader(p)) {
1862 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1863 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1864 
1865 			if (is_child_reaper(pid)) {
1866 				ns_of_pid(pid)->child_reaper = p;
1867 				p->signal->flags |= SIGNAL_UNKILLABLE;
1868 			}
1869 
1870 			p->signal->leader_pid = pid;
1871 			p->signal->tty = tty_kref_get(current->signal->tty);
1872 			/*
1873 			 * Inherit has_child_subreaper flag under the same
1874 			 * tasklist_lock with adding child to the process tree
1875 			 * for propagate_has_child_subreaper optimization.
1876 			 */
1877 			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1878 							 p->real_parent->signal->is_child_subreaper;
1879 			list_add_tail(&p->sibling, &p->real_parent->children);
1880 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1881 			attach_pid(p, PIDTYPE_PGID);
1882 			attach_pid(p, PIDTYPE_SID);
1883 			__this_cpu_inc(process_counts);
1884 		} else {
1885 			current->signal->nr_threads++;
1886 			atomic_inc(&current->signal->live);
1887 			atomic_inc(&current->signal->sigcnt);
1888 			list_add_tail_rcu(&p->thread_group,
1889 					  &p->group_leader->thread_group);
1890 			list_add_tail_rcu(&p->thread_node,
1891 					  &p->signal->thread_head);
1892 		}
1893 		attach_pid(p, PIDTYPE_PID);
1894 		nr_threads++;
1895 	}
1896 
1897 	total_forks++;
1898 	spin_unlock(&current->sighand->siglock);
1899 	syscall_tracepoint_update(p);
1900 	write_unlock_irq(&tasklist_lock);
1901 
1902 	proc_fork_connector(p);
1903 	cgroup_post_fork(p);
1904 	cgroup_threadgroup_change_end(current);
1905 	perf_event_fork(p);
1906 
1907 	trace_task_newtask(p, clone_flags);
1908 	uprobe_copy_process(p, clone_flags);
1909 
1910 	return p;
1911 
1912 bad_fork_cancel_cgroup:
1913 	spin_unlock(&current->sighand->siglock);
1914 	write_unlock_irq(&tasklist_lock);
1915 	cgroup_cancel_fork(p);
1916 bad_fork_free_pid:
1917 	cgroup_threadgroup_change_end(current);
1918 	if (pid != &init_struct_pid)
1919 		free_pid(pid);
1920 bad_fork_cleanup_thread:
1921 	exit_thread(p);
1922 bad_fork_cleanup_io:
1923 	if (p->io_context)
1924 		exit_io_context(p);
1925 bad_fork_cleanup_namespaces:
1926 	exit_task_namespaces(p);
1927 bad_fork_cleanup_mm:
1928 	if (p->mm)
1929 		mmput(p->mm);
1930 bad_fork_cleanup_signal:
1931 	if (!(clone_flags & CLONE_THREAD))
1932 		free_signal_struct(p->signal);
1933 bad_fork_cleanup_sighand:
1934 	__cleanup_sighand(p->sighand);
1935 bad_fork_cleanup_fs:
1936 	exit_fs(p); /* blocking */
1937 bad_fork_cleanup_files:
1938 	exit_files(p); /* blocking */
1939 bad_fork_cleanup_semundo:
1940 	exit_sem(p);
1941 bad_fork_cleanup_security:
1942 	security_task_free(p);
1943 bad_fork_cleanup_audit:
1944 	audit_free(p);
1945 bad_fork_cleanup_perf:
1946 	perf_event_free_task(p);
1947 bad_fork_cleanup_policy:
1948 #ifdef CONFIG_NUMA
1949 	mpol_put(p->mempolicy);
1950 bad_fork_cleanup_threadgroup_lock:
1951 #endif
1952 	delayacct_tsk_free(p);
1953 bad_fork_cleanup_count:
1954 	atomic_dec(&p->cred->user->processes);
1955 	exit_creds(p);
1956 bad_fork_free:
1957 	p->state = TASK_DEAD;
1958 	put_task_stack(p);
1959 	free_task(p);
1960 fork_out:
1961 	return ERR_PTR(retval);
1962 }
1963 
1964 static inline void init_idle_pids(struct pid_link *links)
1965 {
1966 	enum pid_type type;
1967 
1968 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1969 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1970 		links[type].pid = &init_struct_pid;
1971 	}
1972 }
1973 
1974 struct task_struct *fork_idle(int cpu)
1975 {
1976 	struct task_struct *task;
1977 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1978 			    cpu_to_node(cpu));
1979 	if (!IS_ERR(task)) {
1980 		init_idle_pids(task->pids);
1981 		init_idle(task, cpu);
1982 	}
1983 
1984 	return task;
1985 }
1986 
1987 /*
1988  *  Ok, this is the main fork-routine.
1989  *
1990  * It copies the process, and if successful kick-starts
1991  * it and waits for it to finish using the VM if required.
1992  */
1993 long _do_fork(unsigned long clone_flags,
1994 	      unsigned long stack_start,
1995 	      unsigned long stack_size,
1996 	      int __user *parent_tidptr,
1997 	      int __user *child_tidptr,
1998 	      unsigned long tls)
1999 {
2000 	struct task_struct *p;
2001 	int trace = 0;
2002 	long nr;
2003 
2004 	/*
2005 	 * Determine whether and which event to report to ptracer.  When
2006 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
2007 	 * requested, no event is reported; otherwise, report if the event
2008 	 * for the type of forking is enabled.
2009 	 */
2010 	if (!(clone_flags & CLONE_UNTRACED)) {
2011 		if (clone_flags & CLONE_VFORK)
2012 			trace = PTRACE_EVENT_VFORK;
2013 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
2014 			trace = PTRACE_EVENT_CLONE;
2015 		else
2016 			trace = PTRACE_EVENT_FORK;
2017 
2018 		if (likely(!ptrace_event_enabled(current, trace)))
2019 			trace = 0;
2020 	}
2021 
2022 	p = copy_process(clone_flags, stack_start, stack_size,
2023 			 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2024 	add_latent_entropy();
2025 	/*
2026 	 * Do this prior waking up the new thread - the thread pointer
2027 	 * might get invalid after that point, if the thread exits quickly.
2028 	 */
2029 	if (!IS_ERR(p)) {
2030 		struct completion vfork;
2031 		struct pid *pid;
2032 
2033 		trace_sched_process_fork(current, p);
2034 
2035 		pid = get_task_pid(p, PIDTYPE_PID);
2036 		nr = pid_vnr(pid);
2037 
2038 		if (clone_flags & CLONE_PARENT_SETTID)
2039 			put_user(nr, parent_tidptr);
2040 
2041 		if (clone_flags & CLONE_VFORK) {
2042 			p->vfork_done = &vfork;
2043 			init_completion(&vfork);
2044 			get_task_struct(p);
2045 		}
2046 
2047 		wake_up_new_task(p);
2048 
2049 		/* forking complete and child started to run, tell ptracer */
2050 		if (unlikely(trace))
2051 			ptrace_event_pid(trace, pid);
2052 
2053 		if (clone_flags & CLONE_VFORK) {
2054 			if (!wait_for_vfork_done(p, &vfork))
2055 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2056 		}
2057 
2058 		put_pid(pid);
2059 	} else {
2060 		nr = PTR_ERR(p);
2061 	}
2062 	return nr;
2063 }
2064 
2065 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2066 /* For compatibility with architectures that call do_fork directly rather than
2067  * using the syscall entry points below. */
2068 long do_fork(unsigned long clone_flags,
2069 	      unsigned long stack_start,
2070 	      unsigned long stack_size,
2071 	      int __user *parent_tidptr,
2072 	      int __user *child_tidptr)
2073 {
2074 	return _do_fork(clone_flags, stack_start, stack_size,
2075 			parent_tidptr, child_tidptr, 0);
2076 }
2077 #endif
2078 
2079 /*
2080  * Create a kernel thread.
2081  */
2082 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2083 {
2084 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2085 		(unsigned long)arg, NULL, NULL, 0);
2086 }
2087 
2088 #ifdef __ARCH_WANT_SYS_FORK
2089 SYSCALL_DEFINE0(fork)
2090 {
2091 #ifdef CONFIG_MMU
2092 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2093 #else
2094 	/* can not support in nommu mode */
2095 	return -EINVAL;
2096 #endif
2097 }
2098 #endif
2099 
2100 #ifdef __ARCH_WANT_SYS_VFORK
2101 SYSCALL_DEFINE0(vfork)
2102 {
2103 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2104 			0, NULL, NULL, 0);
2105 }
2106 #endif
2107 
2108 #ifdef __ARCH_WANT_SYS_CLONE
2109 #ifdef CONFIG_CLONE_BACKWARDS
2110 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2111 		 int __user *, parent_tidptr,
2112 		 unsigned long, tls,
2113 		 int __user *, child_tidptr)
2114 #elif defined(CONFIG_CLONE_BACKWARDS2)
2115 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2116 		 int __user *, parent_tidptr,
2117 		 int __user *, child_tidptr,
2118 		 unsigned long, tls)
2119 #elif defined(CONFIG_CLONE_BACKWARDS3)
2120 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2121 		int, stack_size,
2122 		int __user *, parent_tidptr,
2123 		int __user *, child_tidptr,
2124 		unsigned long, tls)
2125 #else
2126 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2127 		 int __user *, parent_tidptr,
2128 		 int __user *, child_tidptr,
2129 		 unsigned long, tls)
2130 #endif
2131 {
2132 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2133 }
2134 #endif
2135 
2136 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2137 {
2138 	struct task_struct *leader, *parent, *child;
2139 	int res;
2140 
2141 	read_lock(&tasklist_lock);
2142 	leader = top = top->group_leader;
2143 down:
2144 	for_each_thread(leader, parent) {
2145 		list_for_each_entry(child, &parent->children, sibling) {
2146 			res = visitor(child, data);
2147 			if (res) {
2148 				if (res < 0)
2149 					goto out;
2150 				leader = child;
2151 				goto down;
2152 			}
2153 up:
2154 			;
2155 		}
2156 	}
2157 
2158 	if (leader != top) {
2159 		child = leader;
2160 		parent = child->real_parent;
2161 		leader = parent->group_leader;
2162 		goto up;
2163 	}
2164 out:
2165 	read_unlock(&tasklist_lock);
2166 }
2167 
2168 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2169 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2170 #endif
2171 
2172 static void sighand_ctor(void *data)
2173 {
2174 	struct sighand_struct *sighand = data;
2175 
2176 	spin_lock_init(&sighand->siglock);
2177 	init_waitqueue_head(&sighand->signalfd_wqh);
2178 }
2179 
2180 void __init proc_caches_init(void)
2181 {
2182 	sighand_cachep = kmem_cache_create("sighand_cache",
2183 			sizeof(struct sighand_struct), 0,
2184 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2185 			SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2186 	signal_cachep = kmem_cache_create("signal_cache",
2187 			sizeof(struct signal_struct), 0,
2188 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2189 			NULL);
2190 	files_cachep = kmem_cache_create("files_cache",
2191 			sizeof(struct files_struct), 0,
2192 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2193 			NULL);
2194 	fs_cachep = kmem_cache_create("fs_cache",
2195 			sizeof(struct fs_struct), 0,
2196 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2197 			NULL);
2198 	/*
2199 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2200 	 * whole struct cpumask for the OFFSTACK case. We could change
2201 	 * this to *only* allocate as much of it as required by the
2202 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
2203 	 * is at the end of the structure, exactly for that reason.
2204 	 */
2205 	mm_cachep = kmem_cache_create("mm_struct",
2206 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2207 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2208 			NULL);
2209 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2210 	mmap_init();
2211 	nsproxy_cache_init();
2212 }
2213 
2214 /*
2215  * Check constraints on flags passed to the unshare system call.
2216  */
2217 static int check_unshare_flags(unsigned long unshare_flags)
2218 {
2219 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2220 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2221 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2222 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2223 		return -EINVAL;
2224 	/*
2225 	 * Not implemented, but pretend it works if there is nothing
2226 	 * to unshare.  Note that unsharing the address space or the
2227 	 * signal handlers also need to unshare the signal queues (aka
2228 	 * CLONE_THREAD).
2229 	 */
2230 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2231 		if (!thread_group_empty(current))
2232 			return -EINVAL;
2233 	}
2234 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2235 		if (atomic_read(&current->sighand->count) > 1)
2236 			return -EINVAL;
2237 	}
2238 	if (unshare_flags & CLONE_VM) {
2239 		if (!current_is_single_threaded())
2240 			return -EINVAL;
2241 	}
2242 
2243 	return 0;
2244 }
2245 
2246 /*
2247  * Unshare the filesystem structure if it is being shared
2248  */
2249 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2250 {
2251 	struct fs_struct *fs = current->fs;
2252 
2253 	if (!(unshare_flags & CLONE_FS) || !fs)
2254 		return 0;
2255 
2256 	/* don't need lock here; in the worst case we'll do useless copy */
2257 	if (fs->users == 1)
2258 		return 0;
2259 
2260 	*new_fsp = copy_fs_struct(fs);
2261 	if (!*new_fsp)
2262 		return -ENOMEM;
2263 
2264 	return 0;
2265 }
2266 
2267 /*
2268  * Unshare file descriptor table if it is being shared
2269  */
2270 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2271 {
2272 	struct files_struct *fd = current->files;
2273 	int error = 0;
2274 
2275 	if ((unshare_flags & CLONE_FILES) &&
2276 	    (fd && atomic_read(&fd->count) > 1)) {
2277 		*new_fdp = dup_fd(fd, &error);
2278 		if (!*new_fdp)
2279 			return error;
2280 	}
2281 
2282 	return 0;
2283 }
2284 
2285 /*
2286  * unshare allows a process to 'unshare' part of the process
2287  * context which was originally shared using clone.  copy_*
2288  * functions used by do_fork() cannot be used here directly
2289  * because they modify an inactive task_struct that is being
2290  * constructed. Here we are modifying the current, active,
2291  * task_struct.
2292  */
2293 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2294 {
2295 	struct fs_struct *fs, *new_fs = NULL;
2296 	struct files_struct *fd, *new_fd = NULL;
2297 	struct cred *new_cred = NULL;
2298 	struct nsproxy *new_nsproxy = NULL;
2299 	int do_sysvsem = 0;
2300 	int err;
2301 
2302 	/*
2303 	 * If unsharing a user namespace must also unshare the thread group
2304 	 * and unshare the filesystem root and working directories.
2305 	 */
2306 	if (unshare_flags & CLONE_NEWUSER)
2307 		unshare_flags |= CLONE_THREAD | CLONE_FS;
2308 	/*
2309 	 * If unsharing vm, must also unshare signal handlers.
2310 	 */
2311 	if (unshare_flags & CLONE_VM)
2312 		unshare_flags |= CLONE_SIGHAND;
2313 	/*
2314 	 * If unsharing a signal handlers, must also unshare the signal queues.
2315 	 */
2316 	if (unshare_flags & CLONE_SIGHAND)
2317 		unshare_flags |= CLONE_THREAD;
2318 	/*
2319 	 * If unsharing namespace, must also unshare filesystem information.
2320 	 */
2321 	if (unshare_flags & CLONE_NEWNS)
2322 		unshare_flags |= CLONE_FS;
2323 
2324 	err = check_unshare_flags(unshare_flags);
2325 	if (err)
2326 		goto bad_unshare_out;
2327 	/*
2328 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2329 	 * to a new ipc namespace, the semaphore arrays from the old
2330 	 * namespace are unreachable.
2331 	 */
2332 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2333 		do_sysvsem = 1;
2334 	err = unshare_fs(unshare_flags, &new_fs);
2335 	if (err)
2336 		goto bad_unshare_out;
2337 	err = unshare_fd(unshare_flags, &new_fd);
2338 	if (err)
2339 		goto bad_unshare_cleanup_fs;
2340 	err = unshare_userns(unshare_flags, &new_cred);
2341 	if (err)
2342 		goto bad_unshare_cleanup_fd;
2343 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2344 					 new_cred, new_fs);
2345 	if (err)
2346 		goto bad_unshare_cleanup_cred;
2347 
2348 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2349 		if (do_sysvsem) {
2350 			/*
2351 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2352 			 */
2353 			exit_sem(current);
2354 		}
2355 		if (unshare_flags & CLONE_NEWIPC) {
2356 			/* Orphan segments in old ns (see sem above). */
2357 			exit_shm(current);
2358 			shm_init_task(current);
2359 		}
2360 
2361 		if (new_nsproxy)
2362 			switch_task_namespaces(current, new_nsproxy);
2363 
2364 		task_lock(current);
2365 
2366 		if (new_fs) {
2367 			fs = current->fs;
2368 			spin_lock(&fs->lock);
2369 			current->fs = new_fs;
2370 			if (--fs->users)
2371 				new_fs = NULL;
2372 			else
2373 				new_fs = fs;
2374 			spin_unlock(&fs->lock);
2375 		}
2376 
2377 		if (new_fd) {
2378 			fd = current->files;
2379 			current->files = new_fd;
2380 			new_fd = fd;
2381 		}
2382 
2383 		task_unlock(current);
2384 
2385 		if (new_cred) {
2386 			/* Install the new user namespace */
2387 			commit_creds(new_cred);
2388 			new_cred = NULL;
2389 		}
2390 	}
2391 
2392 	perf_event_namespaces(current);
2393 
2394 bad_unshare_cleanup_cred:
2395 	if (new_cred)
2396 		put_cred(new_cred);
2397 bad_unshare_cleanup_fd:
2398 	if (new_fd)
2399 		put_files_struct(new_fd);
2400 
2401 bad_unshare_cleanup_fs:
2402 	if (new_fs)
2403 		free_fs_struct(new_fs);
2404 
2405 bad_unshare_out:
2406 	return err;
2407 }
2408 
2409 /*
2410  *	Helper to unshare the files of the current task.
2411  *	We don't want to expose copy_files internals to
2412  *	the exec layer of the kernel.
2413  */
2414 
2415 int unshare_files(struct files_struct **displaced)
2416 {
2417 	struct task_struct *task = current;
2418 	struct files_struct *copy = NULL;
2419 	int error;
2420 
2421 	error = unshare_fd(CLONE_FILES, &copy);
2422 	if (error || !copy) {
2423 		*displaced = NULL;
2424 		return error;
2425 	}
2426 	*displaced = task->files;
2427 	task_lock(task);
2428 	task->files = copy;
2429 	task_unlock(task);
2430 	return 0;
2431 }
2432 
2433 int sysctl_max_threads(struct ctl_table *table, int write,
2434 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2435 {
2436 	struct ctl_table t;
2437 	int ret;
2438 	int threads = max_threads;
2439 	int min = MIN_THREADS;
2440 	int max = MAX_THREADS;
2441 
2442 	t = *table;
2443 	t.data = &threads;
2444 	t.extra1 = &min;
2445 	t.extra2 = &max;
2446 
2447 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2448 	if (ret || !write)
2449 		return ret;
2450 
2451 	set_max_threads(threads);
2452 
2453 	return 0;
2454 }
2455