xref: /openbmc/linux/kernel/fork.c (revision 711aab1d)
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/hmm.h>
41 #include <linux/fs.h>
42 #include <linux/mm.h>
43 #include <linux/vmacache.h>
44 #include <linux/nsproxy.h>
45 #include <linux/capability.h>
46 #include <linux/cpu.h>
47 #include <linux/cgroup.h>
48 #include <linux/security.h>
49 #include <linux/hugetlb.h>
50 #include <linux/seccomp.h>
51 #include <linux/swap.h>
52 #include <linux/syscalls.h>
53 #include <linux/jiffies.h>
54 #include <linux/futex.h>
55 #include <linux/compat.h>
56 #include <linux/kthread.h>
57 #include <linux/task_io_accounting_ops.h>
58 #include <linux/rcupdate.h>
59 #include <linux/ptrace.h>
60 #include <linux/mount.h>
61 #include <linux/audit.h>
62 #include <linux/memcontrol.h>
63 #include <linux/ftrace.h>
64 #include <linux/proc_fs.h>
65 #include <linux/profile.h>
66 #include <linux/rmap.h>
67 #include <linux/ksm.h>
68 #include <linux/acct.h>
69 #include <linux/userfaultfd_k.h>
70 #include <linux/tsacct_kern.h>
71 #include <linux/cn_proc.h>
72 #include <linux/freezer.h>
73 #include <linux/delayacct.h>
74 #include <linux/taskstats_kern.h>
75 #include <linux/random.h>
76 #include <linux/tty.h>
77 #include <linux/blkdev.h>
78 #include <linux/fs_struct.h>
79 #include <linux/magic.h>
80 #include <linux/perf_event.h>
81 #include <linux/posix-timers.h>
82 #include <linux/user-return-notifier.h>
83 #include <linux/oom.h>
84 #include <linux/khugepaged.h>
85 #include <linux/signalfd.h>
86 #include <linux/uprobes.h>
87 #include <linux/aio.h>
88 #include <linux/compiler.h>
89 #include <linux/sysctl.h>
90 #include <linux/kcov.h>
91 #include <linux/livepatch.h>
92 #include <linux/thread_info.h>
93 
94 #include <asm/pgtable.h>
95 #include <asm/pgalloc.h>
96 #include <linux/uaccess.h>
97 #include <asm/mmu_context.h>
98 #include <asm/cacheflush.h>
99 #include <asm/tlbflush.h>
100 
101 #include <trace/events/sched.h>
102 
103 #define CREATE_TRACE_POINTS
104 #include <trace/events/task.h>
105 
106 /*
107  * Minimum number of threads to boot the kernel
108  */
109 #define MIN_THREADS 20
110 
111 /*
112  * Maximum number of threads
113  */
114 #define MAX_THREADS FUTEX_TID_MASK
115 
116 /*
117  * Protected counters by write_lock_irq(&tasklist_lock)
118  */
119 unsigned long total_forks;	/* Handle normal Linux uptimes. */
120 int nr_threads;			/* The idle threads do not count.. */
121 
122 int max_threads;		/* tunable limit on nr_threads */
123 
124 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
125 
126 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
127 
128 #ifdef CONFIG_PROVE_RCU
129 int lockdep_tasklist_lock_is_held(void)
130 {
131 	return lockdep_is_held(&tasklist_lock);
132 }
133 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
134 #endif /* #ifdef CONFIG_PROVE_RCU */
135 
136 int nr_processes(void)
137 {
138 	int cpu;
139 	int total = 0;
140 
141 	for_each_possible_cpu(cpu)
142 		total += per_cpu(process_counts, cpu);
143 
144 	return total;
145 }
146 
147 void __weak arch_release_task_struct(struct task_struct *tsk)
148 {
149 }
150 
151 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
152 static struct kmem_cache *task_struct_cachep;
153 
154 static inline struct task_struct *alloc_task_struct_node(int node)
155 {
156 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
157 }
158 
159 static inline void free_task_struct(struct task_struct *tsk)
160 {
161 	kmem_cache_free(task_struct_cachep, tsk);
162 }
163 #endif
164 
165 void __weak arch_release_thread_stack(unsigned long *stack)
166 {
167 }
168 
169 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
170 
171 /*
172  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
173  * kmemcache based allocator.
174  */
175 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
176 
177 #ifdef CONFIG_VMAP_STACK
178 /*
179  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
180  * flush.  Try to minimize the number of calls by caching stacks.
181  */
182 #define NR_CACHED_STACKS 2
183 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
184 
185 static int free_vm_stack_cache(unsigned int cpu)
186 {
187 	struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
188 	int i;
189 
190 	for (i = 0; i < NR_CACHED_STACKS; i++) {
191 		struct vm_struct *vm_stack = cached_vm_stacks[i];
192 
193 		if (!vm_stack)
194 			continue;
195 
196 		vfree(vm_stack->addr);
197 		cached_vm_stacks[i] = NULL;
198 	}
199 
200 	return 0;
201 }
202 #endif
203 
204 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
205 {
206 #ifdef CONFIG_VMAP_STACK
207 	void *stack;
208 	int i;
209 
210 	for (i = 0; i < NR_CACHED_STACKS; i++) {
211 		struct vm_struct *s;
212 
213 		s = this_cpu_xchg(cached_stacks[i], NULL);
214 
215 		if (!s)
216 			continue;
217 
218 		tsk->stack_vm_area = s;
219 		return s->addr;
220 	}
221 
222 	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
223 				     VMALLOC_START, VMALLOC_END,
224 				     THREADINFO_GFP,
225 				     PAGE_KERNEL,
226 				     0, node, __builtin_return_address(0));
227 
228 	/*
229 	 * We can't call find_vm_area() in interrupt context, and
230 	 * free_thread_stack() can be called in interrupt context,
231 	 * so cache the vm_struct.
232 	 */
233 	if (stack)
234 		tsk->stack_vm_area = find_vm_area(stack);
235 	return stack;
236 #else
237 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
238 					     THREAD_SIZE_ORDER);
239 
240 	return page ? page_address(page) : NULL;
241 #endif
242 }
243 
244 static inline void free_thread_stack(struct task_struct *tsk)
245 {
246 #ifdef CONFIG_VMAP_STACK
247 	if (task_stack_vm_area(tsk)) {
248 		int i;
249 
250 		for (i = 0; i < NR_CACHED_STACKS; i++) {
251 			if (this_cpu_cmpxchg(cached_stacks[i],
252 					NULL, tsk->stack_vm_area) != NULL)
253 				continue;
254 
255 			return;
256 		}
257 
258 		vfree_atomic(tsk->stack);
259 		return;
260 	}
261 #endif
262 
263 	__free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
264 }
265 # else
266 static struct kmem_cache *thread_stack_cache;
267 
268 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
269 						  int node)
270 {
271 	return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
272 }
273 
274 static void free_thread_stack(struct task_struct *tsk)
275 {
276 	kmem_cache_free(thread_stack_cache, tsk->stack);
277 }
278 
279 void thread_stack_cache_init(void)
280 {
281 	thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
282 					      THREAD_SIZE, 0, NULL);
283 	BUG_ON(thread_stack_cache == NULL);
284 }
285 # endif
286 #endif
287 
288 /* SLAB cache for signal_struct structures (tsk->signal) */
289 static struct kmem_cache *signal_cachep;
290 
291 /* SLAB cache for sighand_struct structures (tsk->sighand) */
292 struct kmem_cache *sighand_cachep;
293 
294 /* SLAB cache for files_struct structures (tsk->files) */
295 struct kmem_cache *files_cachep;
296 
297 /* SLAB cache for fs_struct structures (tsk->fs) */
298 struct kmem_cache *fs_cachep;
299 
300 /* SLAB cache for vm_area_struct structures */
301 struct kmem_cache *vm_area_cachep;
302 
303 /* SLAB cache for mm_struct structures (tsk->mm) */
304 static struct kmem_cache *mm_cachep;
305 
306 static void account_kernel_stack(struct task_struct *tsk, int account)
307 {
308 	void *stack = task_stack_page(tsk);
309 	struct vm_struct *vm = task_stack_vm_area(tsk);
310 
311 	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
312 
313 	if (vm) {
314 		int i;
315 
316 		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
317 
318 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
319 			mod_zone_page_state(page_zone(vm->pages[i]),
320 					    NR_KERNEL_STACK_KB,
321 					    PAGE_SIZE / 1024 * account);
322 		}
323 
324 		/* All stack pages belong to the same memcg. */
325 		mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
326 				     account * (THREAD_SIZE / 1024));
327 	} else {
328 		/*
329 		 * All stack pages are in the same zone and belong to the
330 		 * same memcg.
331 		 */
332 		struct page *first_page = virt_to_page(stack);
333 
334 		mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
335 				    THREAD_SIZE / 1024 * account);
336 
337 		mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
338 				     account * (THREAD_SIZE / 1024));
339 	}
340 }
341 
342 static void release_task_stack(struct task_struct *tsk)
343 {
344 	if (WARN_ON(tsk->state != TASK_DEAD))
345 		return;  /* Better to leak the stack than to free prematurely */
346 
347 	account_kernel_stack(tsk, -1);
348 	arch_release_thread_stack(tsk->stack);
349 	free_thread_stack(tsk);
350 	tsk->stack = NULL;
351 #ifdef CONFIG_VMAP_STACK
352 	tsk->stack_vm_area = NULL;
353 #endif
354 }
355 
356 #ifdef CONFIG_THREAD_INFO_IN_TASK
357 void put_task_stack(struct task_struct *tsk)
358 {
359 	if (atomic_dec_and_test(&tsk->stack_refcount))
360 		release_task_stack(tsk);
361 }
362 #endif
363 
364 void free_task(struct task_struct *tsk)
365 {
366 #ifndef CONFIG_THREAD_INFO_IN_TASK
367 	/*
368 	 * The task is finally done with both the stack and thread_info,
369 	 * so free both.
370 	 */
371 	release_task_stack(tsk);
372 #else
373 	/*
374 	 * If the task had a separate stack allocation, it should be gone
375 	 * by now.
376 	 */
377 	WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
378 #endif
379 	rt_mutex_debug_task_free(tsk);
380 	ftrace_graph_exit_task(tsk);
381 	put_seccomp_filter(tsk);
382 	arch_release_task_struct(tsk);
383 	if (tsk->flags & PF_KTHREAD)
384 		free_kthread_struct(tsk);
385 	free_task_struct(tsk);
386 }
387 EXPORT_SYMBOL(free_task);
388 
389 static inline void free_signal_struct(struct signal_struct *sig)
390 {
391 	taskstats_tgid_free(sig);
392 	sched_autogroup_exit(sig);
393 	/*
394 	 * __mmdrop is not safe to call from softirq context on x86 due to
395 	 * pgd_dtor so postpone it to the async context
396 	 */
397 	if (sig->oom_mm)
398 		mmdrop_async(sig->oom_mm);
399 	kmem_cache_free(signal_cachep, sig);
400 }
401 
402 static inline void put_signal_struct(struct signal_struct *sig)
403 {
404 	if (atomic_dec_and_test(&sig->sigcnt))
405 		free_signal_struct(sig);
406 }
407 
408 void __put_task_struct(struct task_struct *tsk)
409 {
410 	WARN_ON(!tsk->exit_state);
411 	WARN_ON(atomic_read(&tsk->usage));
412 	WARN_ON(tsk == current);
413 
414 	cgroup_free(tsk);
415 	task_numa_free(tsk);
416 	security_task_free(tsk);
417 	exit_creds(tsk);
418 	delayacct_tsk_free(tsk);
419 	put_signal_struct(tsk->signal);
420 
421 	if (!profile_handoff_task(tsk))
422 		free_task(tsk);
423 }
424 EXPORT_SYMBOL_GPL(__put_task_struct);
425 
426 void __init __weak arch_task_cache_init(void) { }
427 
428 /*
429  * set_max_threads
430  */
431 static void set_max_threads(unsigned int max_threads_suggested)
432 {
433 	u64 threads;
434 
435 	/*
436 	 * The number of threads shall be limited such that the thread
437 	 * structures may only consume a small part of the available memory.
438 	 */
439 	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
440 		threads = MAX_THREADS;
441 	else
442 		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
443 				    (u64) THREAD_SIZE * 8UL);
444 
445 	if (threads > max_threads_suggested)
446 		threads = max_threads_suggested;
447 
448 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
449 }
450 
451 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
452 /* Initialized by the architecture: */
453 int arch_task_struct_size __read_mostly;
454 #endif
455 
456 void __init fork_init(void)
457 {
458 	int i;
459 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
460 #ifndef ARCH_MIN_TASKALIGN
461 #define ARCH_MIN_TASKALIGN	0
462 #endif
463 	int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
464 
465 	/* create a slab on which task_structs can be allocated */
466 	task_struct_cachep = kmem_cache_create("task_struct",
467 			arch_task_struct_size, align,
468 			SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
469 #endif
470 
471 	/* do the arch specific task caches init */
472 	arch_task_cache_init();
473 
474 	set_max_threads(MAX_THREADS);
475 
476 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
477 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
478 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
479 		init_task.signal->rlim[RLIMIT_NPROC];
480 
481 	for (i = 0; i < UCOUNT_COUNTS; i++) {
482 		init_user_ns.ucount_max[i] = max_threads/2;
483 	}
484 
485 #ifdef CONFIG_VMAP_STACK
486 	cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
487 			  NULL, free_vm_stack_cache);
488 #endif
489 
490 	lockdep_init_task(&init_task);
491 }
492 
493 int __weak arch_dup_task_struct(struct task_struct *dst,
494 					       struct task_struct *src)
495 {
496 	*dst = *src;
497 	return 0;
498 }
499 
500 void set_task_stack_end_magic(struct task_struct *tsk)
501 {
502 	unsigned long *stackend;
503 
504 	stackend = end_of_stack(tsk);
505 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
506 }
507 
508 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
509 {
510 	struct task_struct *tsk;
511 	unsigned long *stack;
512 	struct vm_struct *stack_vm_area;
513 	int err;
514 
515 	if (node == NUMA_NO_NODE)
516 		node = tsk_fork_get_node(orig);
517 	tsk = alloc_task_struct_node(node);
518 	if (!tsk)
519 		return NULL;
520 
521 	stack = alloc_thread_stack_node(tsk, node);
522 	if (!stack)
523 		goto free_tsk;
524 
525 	stack_vm_area = task_stack_vm_area(tsk);
526 
527 	err = arch_dup_task_struct(tsk, orig);
528 
529 	/*
530 	 * arch_dup_task_struct() clobbers the stack-related fields.  Make
531 	 * sure they're properly initialized before using any stack-related
532 	 * functions again.
533 	 */
534 	tsk->stack = stack;
535 #ifdef CONFIG_VMAP_STACK
536 	tsk->stack_vm_area = stack_vm_area;
537 #endif
538 #ifdef CONFIG_THREAD_INFO_IN_TASK
539 	atomic_set(&tsk->stack_refcount, 1);
540 #endif
541 
542 	if (err)
543 		goto free_stack;
544 
545 #ifdef CONFIG_SECCOMP
546 	/*
547 	 * We must handle setting up seccomp filters once we're under
548 	 * the sighand lock in case orig has changed between now and
549 	 * then. Until then, filter must be NULL to avoid messing up
550 	 * the usage counts on the error path calling free_task.
551 	 */
552 	tsk->seccomp.filter = NULL;
553 #endif
554 
555 	setup_thread_stack(tsk, orig);
556 	clear_user_return_notifier(tsk);
557 	clear_tsk_need_resched(tsk);
558 	set_task_stack_end_magic(tsk);
559 
560 #ifdef CONFIG_CC_STACKPROTECTOR
561 	tsk->stack_canary = get_random_canary();
562 #endif
563 
564 	/*
565 	 * One for us, one for whoever does the "release_task()" (usually
566 	 * parent)
567 	 */
568 	atomic_set(&tsk->usage, 2);
569 #ifdef CONFIG_BLK_DEV_IO_TRACE
570 	tsk->btrace_seq = 0;
571 #endif
572 	tsk->splice_pipe = NULL;
573 	tsk->task_frag.page = NULL;
574 	tsk->wake_q.next = NULL;
575 
576 	account_kernel_stack(tsk, 1);
577 
578 	kcov_task_init(tsk);
579 
580 #ifdef CONFIG_FAULT_INJECTION
581 	tsk->fail_nth = 0;
582 #endif
583 
584 	return tsk;
585 
586 free_stack:
587 	free_thread_stack(tsk);
588 free_tsk:
589 	free_task_struct(tsk);
590 	return NULL;
591 }
592 
593 #ifdef CONFIG_MMU
594 static __latent_entropy int dup_mmap(struct mm_struct *mm,
595 					struct mm_struct *oldmm)
596 {
597 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
598 	struct rb_node **rb_link, *rb_parent;
599 	int retval;
600 	unsigned long charge;
601 	LIST_HEAD(uf);
602 
603 	uprobe_start_dup_mmap();
604 	if (down_write_killable(&oldmm->mmap_sem)) {
605 		retval = -EINTR;
606 		goto fail_uprobe_end;
607 	}
608 	flush_cache_dup_mm(oldmm);
609 	uprobe_dup_mmap(oldmm, mm);
610 	/*
611 	 * Not linked in yet - no deadlock potential:
612 	 */
613 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
614 
615 	/* No ordering required: file already has been exposed. */
616 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
617 
618 	mm->total_vm = oldmm->total_vm;
619 	mm->data_vm = oldmm->data_vm;
620 	mm->exec_vm = oldmm->exec_vm;
621 	mm->stack_vm = oldmm->stack_vm;
622 
623 	rb_link = &mm->mm_rb.rb_node;
624 	rb_parent = NULL;
625 	pprev = &mm->mmap;
626 	retval = ksm_fork(mm, oldmm);
627 	if (retval)
628 		goto out;
629 	retval = khugepaged_fork(mm, oldmm);
630 	if (retval)
631 		goto out;
632 
633 	prev = NULL;
634 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
635 		struct file *file;
636 
637 		if (mpnt->vm_flags & VM_DONTCOPY) {
638 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
639 			continue;
640 		}
641 		charge = 0;
642 		if (mpnt->vm_flags & VM_ACCOUNT) {
643 			unsigned long len = vma_pages(mpnt);
644 
645 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
646 				goto fail_nomem;
647 			charge = len;
648 		}
649 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
650 		if (!tmp)
651 			goto fail_nomem;
652 		*tmp = *mpnt;
653 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
654 		retval = vma_dup_policy(mpnt, tmp);
655 		if (retval)
656 			goto fail_nomem_policy;
657 		tmp->vm_mm = mm;
658 		retval = dup_userfaultfd(tmp, &uf);
659 		if (retval)
660 			goto fail_nomem_anon_vma_fork;
661 		if (tmp->vm_flags & VM_WIPEONFORK) {
662 			/* VM_WIPEONFORK gets a clean slate in the child. */
663 			tmp->anon_vma = NULL;
664 			if (anon_vma_prepare(tmp))
665 				goto fail_nomem_anon_vma_fork;
666 		} else if (anon_vma_fork(tmp, mpnt))
667 			goto fail_nomem_anon_vma_fork;
668 		tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
669 		tmp->vm_next = tmp->vm_prev = NULL;
670 		file = tmp->vm_file;
671 		if (file) {
672 			struct inode *inode = file_inode(file);
673 			struct address_space *mapping = file->f_mapping;
674 
675 			get_file(file);
676 			if (tmp->vm_flags & VM_DENYWRITE)
677 				atomic_dec(&inode->i_writecount);
678 			i_mmap_lock_write(mapping);
679 			if (tmp->vm_flags & VM_SHARED)
680 				atomic_inc(&mapping->i_mmap_writable);
681 			flush_dcache_mmap_lock(mapping);
682 			/* insert tmp into the share list, just after mpnt */
683 			vma_interval_tree_insert_after(tmp, mpnt,
684 					&mapping->i_mmap);
685 			flush_dcache_mmap_unlock(mapping);
686 			i_mmap_unlock_write(mapping);
687 		}
688 
689 		/*
690 		 * Clear hugetlb-related page reserves for children. This only
691 		 * affects MAP_PRIVATE mappings. Faults generated by the child
692 		 * are not guaranteed to succeed, even if read-only
693 		 */
694 		if (is_vm_hugetlb_page(tmp))
695 			reset_vma_resv_huge_pages(tmp);
696 
697 		/*
698 		 * Link in the new vma and copy the page table entries.
699 		 */
700 		*pprev = tmp;
701 		pprev = &tmp->vm_next;
702 		tmp->vm_prev = prev;
703 		prev = tmp;
704 
705 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
706 		rb_link = &tmp->vm_rb.rb_right;
707 		rb_parent = &tmp->vm_rb;
708 
709 		mm->map_count++;
710 		if (!(tmp->vm_flags & VM_WIPEONFORK))
711 			retval = copy_page_range(mm, oldmm, mpnt);
712 
713 		if (tmp->vm_ops && tmp->vm_ops->open)
714 			tmp->vm_ops->open(tmp);
715 
716 		if (retval)
717 			goto out;
718 	}
719 	/* a new mm has just been created */
720 	arch_dup_mmap(oldmm, mm);
721 	retval = 0;
722 out:
723 	up_write(&mm->mmap_sem);
724 	flush_tlb_mm(oldmm);
725 	up_write(&oldmm->mmap_sem);
726 	dup_userfaultfd_complete(&uf);
727 fail_uprobe_end:
728 	uprobe_end_dup_mmap();
729 	return retval;
730 fail_nomem_anon_vma_fork:
731 	mpol_put(vma_policy(tmp));
732 fail_nomem_policy:
733 	kmem_cache_free(vm_area_cachep, tmp);
734 fail_nomem:
735 	retval = -ENOMEM;
736 	vm_unacct_memory(charge);
737 	goto out;
738 }
739 
740 static inline int mm_alloc_pgd(struct mm_struct *mm)
741 {
742 	mm->pgd = pgd_alloc(mm);
743 	if (unlikely(!mm->pgd))
744 		return -ENOMEM;
745 	return 0;
746 }
747 
748 static inline void mm_free_pgd(struct mm_struct *mm)
749 {
750 	pgd_free(mm, mm->pgd);
751 }
752 #else
753 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
754 {
755 	down_write(&oldmm->mmap_sem);
756 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
757 	up_write(&oldmm->mmap_sem);
758 	return 0;
759 }
760 #define mm_alloc_pgd(mm)	(0)
761 #define mm_free_pgd(mm)
762 #endif /* CONFIG_MMU */
763 
764 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
765 
766 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
767 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
768 
769 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
770 
771 static int __init coredump_filter_setup(char *s)
772 {
773 	default_dump_filter =
774 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
775 		MMF_DUMP_FILTER_MASK;
776 	return 1;
777 }
778 
779 __setup("coredump_filter=", coredump_filter_setup);
780 
781 #include <linux/init_task.h>
782 
783 static void mm_init_aio(struct mm_struct *mm)
784 {
785 #ifdef CONFIG_AIO
786 	spin_lock_init(&mm->ioctx_lock);
787 	mm->ioctx_table = NULL;
788 #endif
789 }
790 
791 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
792 {
793 #ifdef CONFIG_MEMCG
794 	mm->owner = p;
795 #endif
796 }
797 
798 static void mm_init_uprobes_state(struct mm_struct *mm)
799 {
800 #ifdef CONFIG_UPROBES
801 	mm->uprobes_state.xol_area = NULL;
802 #endif
803 }
804 
805 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
806 	struct user_namespace *user_ns)
807 {
808 	mm->mmap = NULL;
809 	mm->mm_rb = RB_ROOT;
810 	mm->vmacache_seqnum = 0;
811 	atomic_set(&mm->mm_users, 1);
812 	atomic_set(&mm->mm_count, 1);
813 	init_rwsem(&mm->mmap_sem);
814 	INIT_LIST_HEAD(&mm->mmlist);
815 	mm->core_state = NULL;
816 	atomic_long_set(&mm->nr_ptes, 0);
817 	mm_nr_pmds_init(mm);
818 	mm->map_count = 0;
819 	mm->locked_vm = 0;
820 	mm->pinned_vm = 0;
821 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
822 	spin_lock_init(&mm->page_table_lock);
823 	mm_init_cpumask(mm);
824 	mm_init_aio(mm);
825 	mm_init_owner(mm, p);
826 	RCU_INIT_POINTER(mm->exe_file, NULL);
827 	mmu_notifier_mm_init(mm);
828 	hmm_mm_init(mm);
829 	init_tlb_flush_pending(mm);
830 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
831 	mm->pmd_huge_pte = NULL;
832 #endif
833 	mm_init_uprobes_state(mm);
834 
835 	if (current->mm) {
836 		mm->flags = current->mm->flags & MMF_INIT_MASK;
837 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
838 	} else {
839 		mm->flags = default_dump_filter;
840 		mm->def_flags = 0;
841 	}
842 
843 	if (mm_alloc_pgd(mm))
844 		goto fail_nopgd;
845 
846 	if (init_new_context(p, mm))
847 		goto fail_nocontext;
848 
849 	mm->user_ns = get_user_ns(user_ns);
850 	return mm;
851 
852 fail_nocontext:
853 	mm_free_pgd(mm);
854 fail_nopgd:
855 	free_mm(mm);
856 	return NULL;
857 }
858 
859 static void check_mm(struct mm_struct *mm)
860 {
861 	int i;
862 
863 	for (i = 0; i < NR_MM_COUNTERS; i++) {
864 		long x = atomic_long_read(&mm->rss_stat.count[i]);
865 
866 		if (unlikely(x))
867 			printk(KERN_ALERT "BUG: Bad rss-counter state "
868 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
869 	}
870 
871 	if (atomic_long_read(&mm->nr_ptes))
872 		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
873 				atomic_long_read(&mm->nr_ptes));
874 	if (mm_nr_pmds(mm))
875 		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
876 				mm_nr_pmds(mm));
877 
878 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
879 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
880 #endif
881 }
882 
883 /*
884  * Allocate and initialize an mm_struct.
885  */
886 struct mm_struct *mm_alloc(void)
887 {
888 	struct mm_struct *mm;
889 
890 	mm = allocate_mm();
891 	if (!mm)
892 		return NULL;
893 
894 	memset(mm, 0, sizeof(*mm));
895 	return mm_init(mm, current, current_user_ns());
896 }
897 
898 /*
899  * Called when the last reference to the mm
900  * is dropped: either by a lazy thread or by
901  * mmput. Free the page directory and the mm.
902  */
903 void __mmdrop(struct mm_struct *mm)
904 {
905 	BUG_ON(mm == &init_mm);
906 	mm_free_pgd(mm);
907 	destroy_context(mm);
908 	hmm_mm_destroy(mm);
909 	mmu_notifier_mm_destroy(mm);
910 	check_mm(mm);
911 	put_user_ns(mm->user_ns);
912 	free_mm(mm);
913 }
914 EXPORT_SYMBOL_GPL(__mmdrop);
915 
916 static inline void __mmput(struct mm_struct *mm)
917 {
918 	VM_BUG_ON(atomic_read(&mm->mm_users));
919 
920 	uprobe_clear_state(mm);
921 	exit_aio(mm);
922 	ksm_exit(mm);
923 	khugepaged_exit(mm); /* must run before exit_mmap */
924 	exit_mmap(mm);
925 	mm_put_huge_zero_page(mm);
926 	set_mm_exe_file(mm, NULL);
927 	if (!list_empty(&mm->mmlist)) {
928 		spin_lock(&mmlist_lock);
929 		list_del(&mm->mmlist);
930 		spin_unlock(&mmlist_lock);
931 	}
932 	if (mm->binfmt)
933 		module_put(mm->binfmt->module);
934 	mmdrop(mm);
935 }
936 
937 /*
938  * Decrement the use count and release all resources for an mm.
939  */
940 void mmput(struct mm_struct *mm)
941 {
942 	might_sleep();
943 
944 	if (atomic_dec_and_test(&mm->mm_users))
945 		__mmput(mm);
946 }
947 EXPORT_SYMBOL_GPL(mmput);
948 
949 /**
950  * set_mm_exe_file - change a reference to the mm's executable file
951  *
952  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
953  *
954  * Main users are mmput() and sys_execve(). Callers prevent concurrent
955  * invocations: in mmput() nobody alive left, in execve task is single
956  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
957  * mm->exe_file, but does so without using set_mm_exe_file() in order
958  * to do avoid the need for any locks.
959  */
960 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
961 {
962 	struct file *old_exe_file;
963 
964 	/*
965 	 * It is safe to dereference the exe_file without RCU as
966 	 * this function is only called if nobody else can access
967 	 * this mm -- see comment above for justification.
968 	 */
969 	old_exe_file = rcu_dereference_raw(mm->exe_file);
970 
971 	if (new_exe_file)
972 		get_file(new_exe_file);
973 	rcu_assign_pointer(mm->exe_file, new_exe_file);
974 	if (old_exe_file)
975 		fput(old_exe_file);
976 }
977 
978 /**
979  * get_mm_exe_file - acquire a reference to the mm's executable file
980  *
981  * Returns %NULL if mm has no associated executable file.
982  * User must release file via fput().
983  */
984 struct file *get_mm_exe_file(struct mm_struct *mm)
985 {
986 	struct file *exe_file;
987 
988 	rcu_read_lock();
989 	exe_file = rcu_dereference(mm->exe_file);
990 	if (exe_file && !get_file_rcu(exe_file))
991 		exe_file = NULL;
992 	rcu_read_unlock();
993 	return exe_file;
994 }
995 EXPORT_SYMBOL(get_mm_exe_file);
996 
997 /**
998  * get_task_exe_file - acquire a reference to the task's executable file
999  *
1000  * Returns %NULL if task's mm (if any) has no associated executable file or
1001  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1002  * User must release file via fput().
1003  */
1004 struct file *get_task_exe_file(struct task_struct *task)
1005 {
1006 	struct file *exe_file = NULL;
1007 	struct mm_struct *mm;
1008 
1009 	task_lock(task);
1010 	mm = task->mm;
1011 	if (mm) {
1012 		if (!(task->flags & PF_KTHREAD))
1013 			exe_file = get_mm_exe_file(mm);
1014 	}
1015 	task_unlock(task);
1016 	return exe_file;
1017 }
1018 EXPORT_SYMBOL(get_task_exe_file);
1019 
1020 /**
1021  * get_task_mm - acquire a reference to the task's mm
1022  *
1023  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1024  * this kernel workthread has transiently adopted a user mm with use_mm,
1025  * to do its AIO) is not set and if so returns a reference to it, after
1026  * bumping up the use count.  User must release the mm via mmput()
1027  * after use.  Typically used by /proc and ptrace.
1028  */
1029 struct mm_struct *get_task_mm(struct task_struct *task)
1030 {
1031 	struct mm_struct *mm;
1032 
1033 	task_lock(task);
1034 	mm = task->mm;
1035 	if (mm) {
1036 		if (task->flags & PF_KTHREAD)
1037 			mm = NULL;
1038 		else
1039 			mmget(mm);
1040 	}
1041 	task_unlock(task);
1042 	return mm;
1043 }
1044 EXPORT_SYMBOL_GPL(get_task_mm);
1045 
1046 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1047 {
1048 	struct mm_struct *mm;
1049 	int err;
1050 
1051 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1052 	if (err)
1053 		return ERR_PTR(err);
1054 
1055 	mm = get_task_mm(task);
1056 	if (mm && mm != current->mm &&
1057 			!ptrace_may_access(task, mode)) {
1058 		mmput(mm);
1059 		mm = ERR_PTR(-EACCES);
1060 	}
1061 	mutex_unlock(&task->signal->cred_guard_mutex);
1062 
1063 	return mm;
1064 }
1065 
1066 static void complete_vfork_done(struct task_struct *tsk)
1067 {
1068 	struct completion *vfork;
1069 
1070 	task_lock(tsk);
1071 	vfork = tsk->vfork_done;
1072 	if (likely(vfork)) {
1073 		tsk->vfork_done = NULL;
1074 		complete(vfork);
1075 	}
1076 	task_unlock(tsk);
1077 }
1078 
1079 static int wait_for_vfork_done(struct task_struct *child,
1080 				struct completion *vfork)
1081 {
1082 	int killed;
1083 
1084 	freezer_do_not_count();
1085 	killed = wait_for_completion_killable(vfork);
1086 	freezer_count();
1087 
1088 	if (killed) {
1089 		task_lock(child);
1090 		child->vfork_done = NULL;
1091 		task_unlock(child);
1092 	}
1093 
1094 	put_task_struct(child);
1095 	return killed;
1096 }
1097 
1098 /* Please note the differences between mmput and mm_release.
1099  * mmput is called whenever we stop holding onto a mm_struct,
1100  * error success whatever.
1101  *
1102  * mm_release is called after a mm_struct has been removed
1103  * from the current process.
1104  *
1105  * This difference is important for error handling, when we
1106  * only half set up a mm_struct for a new process and need to restore
1107  * the old one.  Because we mmput the new mm_struct before
1108  * restoring the old one. . .
1109  * Eric Biederman 10 January 1998
1110  */
1111 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1112 {
1113 	/* Get rid of any futexes when releasing the mm */
1114 #ifdef CONFIG_FUTEX
1115 	if (unlikely(tsk->robust_list)) {
1116 		exit_robust_list(tsk);
1117 		tsk->robust_list = NULL;
1118 	}
1119 #ifdef CONFIG_COMPAT
1120 	if (unlikely(tsk->compat_robust_list)) {
1121 		compat_exit_robust_list(tsk);
1122 		tsk->compat_robust_list = NULL;
1123 	}
1124 #endif
1125 	if (unlikely(!list_empty(&tsk->pi_state_list)))
1126 		exit_pi_state_list(tsk);
1127 #endif
1128 
1129 	uprobe_free_utask(tsk);
1130 
1131 	/* Get rid of any cached register state */
1132 	deactivate_mm(tsk, mm);
1133 
1134 	/*
1135 	 * Signal userspace if we're not exiting with a core dump
1136 	 * because we want to leave the value intact for debugging
1137 	 * purposes.
1138 	 */
1139 	if (tsk->clear_child_tid) {
1140 		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1141 		    atomic_read(&mm->mm_users) > 1) {
1142 			/*
1143 			 * We don't check the error code - if userspace has
1144 			 * not set up a proper pointer then tough luck.
1145 			 */
1146 			put_user(0, tsk->clear_child_tid);
1147 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1148 					1, NULL, NULL, 0);
1149 		}
1150 		tsk->clear_child_tid = NULL;
1151 	}
1152 
1153 	/*
1154 	 * All done, finally we can wake up parent and return this mm to him.
1155 	 * Also kthread_stop() uses this completion for synchronization.
1156 	 */
1157 	if (tsk->vfork_done)
1158 		complete_vfork_done(tsk);
1159 }
1160 
1161 /*
1162  * Allocate a new mm structure and copy contents from the
1163  * mm structure of the passed in task structure.
1164  */
1165 static struct mm_struct *dup_mm(struct task_struct *tsk)
1166 {
1167 	struct mm_struct *mm, *oldmm = current->mm;
1168 	int err;
1169 
1170 	mm = allocate_mm();
1171 	if (!mm)
1172 		goto fail_nomem;
1173 
1174 	memcpy(mm, oldmm, sizeof(*mm));
1175 
1176 	if (!mm_init(mm, tsk, mm->user_ns))
1177 		goto fail_nomem;
1178 
1179 	err = dup_mmap(mm, oldmm);
1180 	if (err)
1181 		goto free_pt;
1182 
1183 	mm->hiwater_rss = get_mm_rss(mm);
1184 	mm->hiwater_vm = mm->total_vm;
1185 
1186 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1187 		goto free_pt;
1188 
1189 	return mm;
1190 
1191 free_pt:
1192 	/* don't put binfmt in mmput, we haven't got module yet */
1193 	mm->binfmt = NULL;
1194 	mmput(mm);
1195 
1196 fail_nomem:
1197 	return NULL;
1198 }
1199 
1200 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1201 {
1202 	struct mm_struct *mm, *oldmm;
1203 	int retval;
1204 
1205 	tsk->min_flt = tsk->maj_flt = 0;
1206 	tsk->nvcsw = tsk->nivcsw = 0;
1207 #ifdef CONFIG_DETECT_HUNG_TASK
1208 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1209 #endif
1210 
1211 	tsk->mm = NULL;
1212 	tsk->active_mm = NULL;
1213 
1214 	/*
1215 	 * Are we cloning a kernel thread?
1216 	 *
1217 	 * We need to steal a active VM for that..
1218 	 */
1219 	oldmm = current->mm;
1220 	if (!oldmm)
1221 		return 0;
1222 
1223 	/* initialize the new vmacache entries */
1224 	vmacache_flush(tsk);
1225 
1226 	if (clone_flags & CLONE_VM) {
1227 		mmget(oldmm);
1228 		mm = oldmm;
1229 		goto good_mm;
1230 	}
1231 
1232 	retval = -ENOMEM;
1233 	mm = dup_mm(tsk);
1234 	if (!mm)
1235 		goto fail_nomem;
1236 
1237 good_mm:
1238 	tsk->mm = mm;
1239 	tsk->active_mm = mm;
1240 	return 0;
1241 
1242 fail_nomem:
1243 	return retval;
1244 }
1245 
1246 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1247 {
1248 	struct fs_struct *fs = current->fs;
1249 	if (clone_flags & CLONE_FS) {
1250 		/* tsk->fs is already what we want */
1251 		spin_lock(&fs->lock);
1252 		if (fs->in_exec) {
1253 			spin_unlock(&fs->lock);
1254 			return -EAGAIN;
1255 		}
1256 		fs->users++;
1257 		spin_unlock(&fs->lock);
1258 		return 0;
1259 	}
1260 	tsk->fs = copy_fs_struct(fs);
1261 	if (!tsk->fs)
1262 		return -ENOMEM;
1263 	return 0;
1264 }
1265 
1266 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1267 {
1268 	struct files_struct *oldf, *newf;
1269 	int error = 0;
1270 
1271 	/*
1272 	 * A background process may not have any files ...
1273 	 */
1274 	oldf = current->files;
1275 	if (!oldf)
1276 		goto out;
1277 
1278 	if (clone_flags & CLONE_FILES) {
1279 		atomic_inc(&oldf->count);
1280 		goto out;
1281 	}
1282 
1283 	newf = dup_fd(oldf, &error);
1284 	if (!newf)
1285 		goto out;
1286 
1287 	tsk->files = newf;
1288 	error = 0;
1289 out:
1290 	return error;
1291 }
1292 
1293 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1294 {
1295 #ifdef CONFIG_BLOCK
1296 	struct io_context *ioc = current->io_context;
1297 	struct io_context *new_ioc;
1298 
1299 	if (!ioc)
1300 		return 0;
1301 	/*
1302 	 * Share io context with parent, if CLONE_IO is set
1303 	 */
1304 	if (clone_flags & CLONE_IO) {
1305 		ioc_task_link(ioc);
1306 		tsk->io_context = ioc;
1307 	} else if (ioprio_valid(ioc->ioprio)) {
1308 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1309 		if (unlikely(!new_ioc))
1310 			return -ENOMEM;
1311 
1312 		new_ioc->ioprio = ioc->ioprio;
1313 		put_io_context(new_ioc);
1314 	}
1315 #endif
1316 	return 0;
1317 }
1318 
1319 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1320 {
1321 	struct sighand_struct *sig;
1322 
1323 	if (clone_flags & CLONE_SIGHAND) {
1324 		atomic_inc(&current->sighand->count);
1325 		return 0;
1326 	}
1327 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1328 	rcu_assign_pointer(tsk->sighand, sig);
1329 	if (!sig)
1330 		return -ENOMEM;
1331 
1332 	atomic_set(&sig->count, 1);
1333 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1334 	return 0;
1335 }
1336 
1337 void __cleanup_sighand(struct sighand_struct *sighand)
1338 {
1339 	if (atomic_dec_and_test(&sighand->count)) {
1340 		signalfd_cleanup(sighand);
1341 		/*
1342 		 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1343 		 * without an RCU grace period, see __lock_task_sighand().
1344 		 */
1345 		kmem_cache_free(sighand_cachep, sighand);
1346 	}
1347 }
1348 
1349 #ifdef CONFIG_POSIX_TIMERS
1350 /*
1351  * Initialize POSIX timer handling for a thread group.
1352  */
1353 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1354 {
1355 	unsigned long cpu_limit;
1356 
1357 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1358 	if (cpu_limit != RLIM_INFINITY) {
1359 		sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1360 		sig->cputimer.running = true;
1361 	}
1362 
1363 	/* The timer lists. */
1364 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1365 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1366 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1367 }
1368 #else
1369 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1370 #endif
1371 
1372 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1373 {
1374 	struct signal_struct *sig;
1375 
1376 	if (clone_flags & CLONE_THREAD)
1377 		return 0;
1378 
1379 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1380 	tsk->signal = sig;
1381 	if (!sig)
1382 		return -ENOMEM;
1383 
1384 	sig->nr_threads = 1;
1385 	atomic_set(&sig->live, 1);
1386 	atomic_set(&sig->sigcnt, 1);
1387 
1388 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1389 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1390 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1391 
1392 	init_waitqueue_head(&sig->wait_chldexit);
1393 	sig->curr_target = tsk;
1394 	init_sigpending(&sig->shared_pending);
1395 	seqlock_init(&sig->stats_lock);
1396 	prev_cputime_init(&sig->prev_cputime);
1397 
1398 #ifdef CONFIG_POSIX_TIMERS
1399 	INIT_LIST_HEAD(&sig->posix_timers);
1400 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1401 	sig->real_timer.function = it_real_fn;
1402 #endif
1403 
1404 	task_lock(current->group_leader);
1405 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1406 	task_unlock(current->group_leader);
1407 
1408 	posix_cpu_timers_init_group(sig);
1409 
1410 	tty_audit_fork(sig);
1411 	sched_autogroup_fork(sig);
1412 
1413 	sig->oom_score_adj = current->signal->oom_score_adj;
1414 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1415 
1416 	mutex_init(&sig->cred_guard_mutex);
1417 
1418 	return 0;
1419 }
1420 
1421 static void copy_seccomp(struct task_struct *p)
1422 {
1423 #ifdef CONFIG_SECCOMP
1424 	/*
1425 	 * Must be called with sighand->lock held, which is common to
1426 	 * all threads in the group. Holding cred_guard_mutex is not
1427 	 * needed because this new task is not yet running and cannot
1428 	 * be racing exec.
1429 	 */
1430 	assert_spin_locked(&current->sighand->siglock);
1431 
1432 	/* Ref-count the new filter user, and assign it. */
1433 	get_seccomp_filter(current);
1434 	p->seccomp = current->seccomp;
1435 
1436 	/*
1437 	 * Explicitly enable no_new_privs here in case it got set
1438 	 * between the task_struct being duplicated and holding the
1439 	 * sighand lock. The seccomp state and nnp must be in sync.
1440 	 */
1441 	if (task_no_new_privs(current))
1442 		task_set_no_new_privs(p);
1443 
1444 	/*
1445 	 * If the parent gained a seccomp mode after copying thread
1446 	 * flags and between before we held the sighand lock, we have
1447 	 * to manually enable the seccomp thread flag here.
1448 	 */
1449 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1450 		set_tsk_thread_flag(p, TIF_SECCOMP);
1451 #endif
1452 }
1453 
1454 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1455 {
1456 	current->clear_child_tid = tidptr;
1457 
1458 	return task_pid_vnr(current);
1459 }
1460 
1461 static void rt_mutex_init_task(struct task_struct *p)
1462 {
1463 	raw_spin_lock_init(&p->pi_lock);
1464 #ifdef CONFIG_RT_MUTEXES
1465 	p->pi_waiters = RB_ROOT_CACHED;
1466 	p->pi_top_task = NULL;
1467 	p->pi_blocked_on = NULL;
1468 #endif
1469 }
1470 
1471 #ifdef CONFIG_POSIX_TIMERS
1472 /*
1473  * Initialize POSIX timer handling for a single task.
1474  */
1475 static void posix_cpu_timers_init(struct task_struct *tsk)
1476 {
1477 	tsk->cputime_expires.prof_exp = 0;
1478 	tsk->cputime_expires.virt_exp = 0;
1479 	tsk->cputime_expires.sched_exp = 0;
1480 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1481 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1482 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1483 }
1484 #else
1485 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1486 #endif
1487 
1488 static inline void
1489 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1490 {
1491 	 task->pids[type].pid = pid;
1492 }
1493 
1494 static inline void rcu_copy_process(struct task_struct *p)
1495 {
1496 #ifdef CONFIG_PREEMPT_RCU
1497 	p->rcu_read_lock_nesting = 0;
1498 	p->rcu_read_unlock_special.s = 0;
1499 	p->rcu_blocked_node = NULL;
1500 	INIT_LIST_HEAD(&p->rcu_node_entry);
1501 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1502 #ifdef CONFIG_TASKS_RCU
1503 	p->rcu_tasks_holdout = false;
1504 	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1505 	p->rcu_tasks_idle_cpu = -1;
1506 #endif /* #ifdef CONFIG_TASKS_RCU */
1507 }
1508 
1509 /*
1510  * This creates a new process as a copy of the old one,
1511  * but does not actually start it yet.
1512  *
1513  * It copies the registers, and all the appropriate
1514  * parts of the process environment (as per the clone
1515  * flags). The actual kick-off is left to the caller.
1516  */
1517 static __latent_entropy struct task_struct *copy_process(
1518 					unsigned long clone_flags,
1519 					unsigned long stack_start,
1520 					unsigned long stack_size,
1521 					int __user *child_tidptr,
1522 					struct pid *pid,
1523 					int trace,
1524 					unsigned long tls,
1525 					int node)
1526 {
1527 	int retval;
1528 	struct task_struct *p;
1529 
1530 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1531 		return ERR_PTR(-EINVAL);
1532 
1533 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1534 		return ERR_PTR(-EINVAL);
1535 
1536 	/*
1537 	 * Thread groups must share signals as well, and detached threads
1538 	 * can only be started up within the thread group.
1539 	 */
1540 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1541 		return ERR_PTR(-EINVAL);
1542 
1543 	/*
1544 	 * Shared signal handlers imply shared VM. By way of the above,
1545 	 * thread groups also imply shared VM. Blocking this case allows
1546 	 * for various simplifications in other code.
1547 	 */
1548 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1549 		return ERR_PTR(-EINVAL);
1550 
1551 	/*
1552 	 * Siblings of global init remain as zombies on exit since they are
1553 	 * not reaped by their parent (swapper). To solve this and to avoid
1554 	 * multi-rooted process trees, prevent global and container-inits
1555 	 * from creating siblings.
1556 	 */
1557 	if ((clone_flags & CLONE_PARENT) &&
1558 				current->signal->flags & SIGNAL_UNKILLABLE)
1559 		return ERR_PTR(-EINVAL);
1560 
1561 	/*
1562 	 * If the new process will be in a different pid or user namespace
1563 	 * do not allow it to share a thread group with the forking task.
1564 	 */
1565 	if (clone_flags & CLONE_THREAD) {
1566 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1567 		    (task_active_pid_ns(current) !=
1568 				current->nsproxy->pid_ns_for_children))
1569 			return ERR_PTR(-EINVAL);
1570 	}
1571 
1572 	retval = -ENOMEM;
1573 	p = dup_task_struct(current, node);
1574 	if (!p)
1575 		goto fork_out;
1576 
1577 	/*
1578 	 * This _must_ happen before we call free_task(), i.e. before we jump
1579 	 * to any of the bad_fork_* labels. This is to avoid freeing
1580 	 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1581 	 * kernel threads (PF_KTHREAD).
1582 	 */
1583 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1584 	/*
1585 	 * Clear TID on mm_release()?
1586 	 */
1587 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1588 
1589 	ftrace_graph_init_task(p);
1590 
1591 	rt_mutex_init_task(p);
1592 
1593 #ifdef CONFIG_PROVE_LOCKING
1594 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1595 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1596 #endif
1597 	retval = -EAGAIN;
1598 	if (atomic_read(&p->real_cred->user->processes) >=
1599 			task_rlimit(p, RLIMIT_NPROC)) {
1600 		if (p->real_cred->user != INIT_USER &&
1601 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1602 			goto bad_fork_free;
1603 	}
1604 	current->flags &= ~PF_NPROC_EXCEEDED;
1605 
1606 	retval = copy_creds(p, clone_flags);
1607 	if (retval < 0)
1608 		goto bad_fork_free;
1609 
1610 	/*
1611 	 * If multiple threads are within copy_process(), then this check
1612 	 * triggers too late. This doesn't hurt, the check is only there
1613 	 * to stop root fork bombs.
1614 	 */
1615 	retval = -EAGAIN;
1616 	if (nr_threads >= max_threads)
1617 		goto bad_fork_cleanup_count;
1618 
1619 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1620 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1621 	p->flags |= PF_FORKNOEXEC;
1622 	INIT_LIST_HEAD(&p->children);
1623 	INIT_LIST_HEAD(&p->sibling);
1624 	rcu_copy_process(p);
1625 	p->vfork_done = NULL;
1626 	spin_lock_init(&p->alloc_lock);
1627 
1628 	init_sigpending(&p->pending);
1629 
1630 	p->utime = p->stime = p->gtime = 0;
1631 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1632 	p->utimescaled = p->stimescaled = 0;
1633 #endif
1634 	prev_cputime_init(&p->prev_cputime);
1635 
1636 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1637 	seqcount_init(&p->vtime.seqcount);
1638 	p->vtime.starttime = 0;
1639 	p->vtime.state = VTIME_INACTIVE;
1640 #endif
1641 
1642 #if defined(SPLIT_RSS_COUNTING)
1643 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1644 #endif
1645 
1646 	p->default_timer_slack_ns = current->timer_slack_ns;
1647 
1648 	task_io_accounting_init(&p->ioac);
1649 	acct_clear_integrals(p);
1650 
1651 	posix_cpu_timers_init(p);
1652 
1653 	p->start_time = ktime_get_ns();
1654 	p->real_start_time = ktime_get_boot_ns();
1655 	p->io_context = NULL;
1656 	p->audit_context = NULL;
1657 	cgroup_fork(p);
1658 #ifdef CONFIG_NUMA
1659 	p->mempolicy = mpol_dup(p->mempolicy);
1660 	if (IS_ERR(p->mempolicy)) {
1661 		retval = PTR_ERR(p->mempolicy);
1662 		p->mempolicy = NULL;
1663 		goto bad_fork_cleanup_threadgroup_lock;
1664 	}
1665 #endif
1666 #ifdef CONFIG_CPUSETS
1667 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1668 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1669 	seqcount_init(&p->mems_allowed_seq);
1670 #endif
1671 #ifdef CONFIG_TRACE_IRQFLAGS
1672 	p->irq_events = 0;
1673 	p->hardirqs_enabled = 0;
1674 	p->hardirq_enable_ip = 0;
1675 	p->hardirq_enable_event = 0;
1676 	p->hardirq_disable_ip = _THIS_IP_;
1677 	p->hardirq_disable_event = 0;
1678 	p->softirqs_enabled = 1;
1679 	p->softirq_enable_ip = _THIS_IP_;
1680 	p->softirq_enable_event = 0;
1681 	p->softirq_disable_ip = 0;
1682 	p->softirq_disable_event = 0;
1683 	p->hardirq_context = 0;
1684 	p->softirq_context = 0;
1685 #endif
1686 
1687 	p->pagefault_disabled = 0;
1688 
1689 #ifdef CONFIG_LOCKDEP
1690 	p->lockdep_depth = 0; /* no locks held yet */
1691 	p->curr_chain_key = 0;
1692 	p->lockdep_recursion = 0;
1693 	lockdep_init_task(p);
1694 #endif
1695 
1696 #ifdef CONFIG_DEBUG_MUTEXES
1697 	p->blocked_on = NULL; /* not blocked yet */
1698 #endif
1699 #ifdef CONFIG_BCACHE
1700 	p->sequential_io	= 0;
1701 	p->sequential_io_avg	= 0;
1702 #endif
1703 
1704 	/* Perform scheduler related setup. Assign this task to a CPU. */
1705 	retval = sched_fork(clone_flags, p);
1706 	if (retval)
1707 		goto bad_fork_cleanup_policy;
1708 
1709 	retval = perf_event_init_task(p);
1710 	if (retval)
1711 		goto bad_fork_cleanup_policy;
1712 	retval = audit_alloc(p);
1713 	if (retval)
1714 		goto bad_fork_cleanup_perf;
1715 	/* copy all the process information */
1716 	shm_init_task(p);
1717 	retval = security_task_alloc(p, clone_flags);
1718 	if (retval)
1719 		goto bad_fork_cleanup_audit;
1720 	retval = copy_semundo(clone_flags, p);
1721 	if (retval)
1722 		goto bad_fork_cleanup_security;
1723 	retval = copy_files(clone_flags, p);
1724 	if (retval)
1725 		goto bad_fork_cleanup_semundo;
1726 	retval = copy_fs(clone_flags, p);
1727 	if (retval)
1728 		goto bad_fork_cleanup_files;
1729 	retval = copy_sighand(clone_flags, p);
1730 	if (retval)
1731 		goto bad_fork_cleanup_fs;
1732 	retval = copy_signal(clone_flags, p);
1733 	if (retval)
1734 		goto bad_fork_cleanup_sighand;
1735 	retval = copy_mm(clone_flags, p);
1736 	if (retval)
1737 		goto bad_fork_cleanup_signal;
1738 	retval = copy_namespaces(clone_flags, p);
1739 	if (retval)
1740 		goto bad_fork_cleanup_mm;
1741 	retval = copy_io(clone_flags, p);
1742 	if (retval)
1743 		goto bad_fork_cleanup_namespaces;
1744 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1745 	if (retval)
1746 		goto bad_fork_cleanup_io;
1747 
1748 	if (pid != &init_struct_pid) {
1749 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1750 		if (IS_ERR(pid)) {
1751 			retval = PTR_ERR(pid);
1752 			goto bad_fork_cleanup_thread;
1753 		}
1754 	}
1755 
1756 #ifdef CONFIG_BLOCK
1757 	p->plug = NULL;
1758 #endif
1759 #ifdef CONFIG_FUTEX
1760 	p->robust_list = NULL;
1761 #ifdef CONFIG_COMPAT
1762 	p->compat_robust_list = NULL;
1763 #endif
1764 	INIT_LIST_HEAD(&p->pi_state_list);
1765 	p->pi_state_cache = NULL;
1766 #endif
1767 	/*
1768 	 * sigaltstack should be cleared when sharing the same VM
1769 	 */
1770 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1771 		sas_ss_reset(p);
1772 
1773 	/*
1774 	 * Syscall tracing and stepping should be turned off in the
1775 	 * child regardless of CLONE_PTRACE.
1776 	 */
1777 	user_disable_single_step(p);
1778 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1779 #ifdef TIF_SYSCALL_EMU
1780 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1781 #endif
1782 	clear_all_latency_tracing(p);
1783 
1784 	/* ok, now we should be set up.. */
1785 	p->pid = pid_nr(pid);
1786 	if (clone_flags & CLONE_THREAD) {
1787 		p->exit_signal = -1;
1788 		p->group_leader = current->group_leader;
1789 		p->tgid = current->tgid;
1790 	} else {
1791 		if (clone_flags & CLONE_PARENT)
1792 			p->exit_signal = current->group_leader->exit_signal;
1793 		else
1794 			p->exit_signal = (clone_flags & CSIGNAL);
1795 		p->group_leader = p;
1796 		p->tgid = p->pid;
1797 	}
1798 
1799 	p->nr_dirtied = 0;
1800 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1801 	p->dirty_paused_when = 0;
1802 
1803 	p->pdeath_signal = 0;
1804 	INIT_LIST_HEAD(&p->thread_group);
1805 	p->task_works = NULL;
1806 
1807 	cgroup_threadgroup_change_begin(current);
1808 	/*
1809 	 * Ensure that the cgroup subsystem policies allow the new process to be
1810 	 * forked. It should be noted the the new process's css_set can be changed
1811 	 * between here and cgroup_post_fork() if an organisation operation is in
1812 	 * progress.
1813 	 */
1814 	retval = cgroup_can_fork(p);
1815 	if (retval)
1816 		goto bad_fork_free_pid;
1817 
1818 	/*
1819 	 * Make it visible to the rest of the system, but dont wake it up yet.
1820 	 * Need tasklist lock for parent etc handling!
1821 	 */
1822 	write_lock_irq(&tasklist_lock);
1823 
1824 	/* CLONE_PARENT re-uses the old parent */
1825 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1826 		p->real_parent = current->real_parent;
1827 		p->parent_exec_id = current->parent_exec_id;
1828 	} else {
1829 		p->real_parent = current;
1830 		p->parent_exec_id = current->self_exec_id;
1831 	}
1832 
1833 	klp_copy_process(p);
1834 
1835 	spin_lock(&current->sighand->siglock);
1836 
1837 	/*
1838 	 * Copy seccomp details explicitly here, in case they were changed
1839 	 * before holding sighand lock.
1840 	 */
1841 	copy_seccomp(p);
1842 
1843 	/*
1844 	 * Process group and session signals need to be delivered to just the
1845 	 * parent before the fork or both the parent and the child after the
1846 	 * fork. Restart if a signal comes in before we add the new process to
1847 	 * it's process group.
1848 	 * A fatal signal pending means that current will exit, so the new
1849 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1850 	*/
1851 	recalc_sigpending();
1852 	if (signal_pending(current)) {
1853 		retval = -ERESTARTNOINTR;
1854 		goto bad_fork_cancel_cgroup;
1855 	}
1856 	if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1857 		retval = -ENOMEM;
1858 		goto bad_fork_cancel_cgroup;
1859 	}
1860 
1861 	if (likely(p->pid)) {
1862 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1863 
1864 		init_task_pid(p, PIDTYPE_PID, pid);
1865 		if (thread_group_leader(p)) {
1866 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1867 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1868 
1869 			if (is_child_reaper(pid)) {
1870 				ns_of_pid(pid)->child_reaper = p;
1871 				p->signal->flags |= SIGNAL_UNKILLABLE;
1872 			}
1873 
1874 			p->signal->leader_pid = pid;
1875 			p->signal->tty = tty_kref_get(current->signal->tty);
1876 			/*
1877 			 * Inherit has_child_subreaper flag under the same
1878 			 * tasklist_lock with adding child to the process tree
1879 			 * for propagate_has_child_subreaper optimization.
1880 			 */
1881 			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1882 							 p->real_parent->signal->is_child_subreaper;
1883 			list_add_tail(&p->sibling, &p->real_parent->children);
1884 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1885 			attach_pid(p, PIDTYPE_PGID);
1886 			attach_pid(p, PIDTYPE_SID);
1887 			__this_cpu_inc(process_counts);
1888 		} else {
1889 			current->signal->nr_threads++;
1890 			atomic_inc(&current->signal->live);
1891 			atomic_inc(&current->signal->sigcnt);
1892 			list_add_tail_rcu(&p->thread_group,
1893 					  &p->group_leader->thread_group);
1894 			list_add_tail_rcu(&p->thread_node,
1895 					  &p->signal->thread_head);
1896 		}
1897 		attach_pid(p, PIDTYPE_PID);
1898 		nr_threads++;
1899 	}
1900 
1901 	total_forks++;
1902 	spin_unlock(&current->sighand->siglock);
1903 	syscall_tracepoint_update(p);
1904 	write_unlock_irq(&tasklist_lock);
1905 
1906 	proc_fork_connector(p);
1907 	cgroup_post_fork(p);
1908 	cgroup_threadgroup_change_end(current);
1909 	perf_event_fork(p);
1910 
1911 	trace_task_newtask(p, clone_flags);
1912 	uprobe_copy_process(p, clone_flags);
1913 
1914 	return p;
1915 
1916 bad_fork_cancel_cgroup:
1917 	spin_unlock(&current->sighand->siglock);
1918 	write_unlock_irq(&tasklist_lock);
1919 	cgroup_cancel_fork(p);
1920 bad_fork_free_pid:
1921 	cgroup_threadgroup_change_end(current);
1922 	if (pid != &init_struct_pid)
1923 		free_pid(pid);
1924 bad_fork_cleanup_thread:
1925 	exit_thread(p);
1926 bad_fork_cleanup_io:
1927 	if (p->io_context)
1928 		exit_io_context(p);
1929 bad_fork_cleanup_namespaces:
1930 	exit_task_namespaces(p);
1931 bad_fork_cleanup_mm:
1932 	if (p->mm)
1933 		mmput(p->mm);
1934 bad_fork_cleanup_signal:
1935 	if (!(clone_flags & CLONE_THREAD))
1936 		free_signal_struct(p->signal);
1937 bad_fork_cleanup_sighand:
1938 	__cleanup_sighand(p->sighand);
1939 bad_fork_cleanup_fs:
1940 	exit_fs(p); /* blocking */
1941 bad_fork_cleanup_files:
1942 	exit_files(p); /* blocking */
1943 bad_fork_cleanup_semundo:
1944 	exit_sem(p);
1945 bad_fork_cleanup_security:
1946 	security_task_free(p);
1947 bad_fork_cleanup_audit:
1948 	audit_free(p);
1949 bad_fork_cleanup_perf:
1950 	perf_event_free_task(p);
1951 bad_fork_cleanup_policy:
1952 	lockdep_free_task(p);
1953 #ifdef CONFIG_NUMA
1954 	mpol_put(p->mempolicy);
1955 bad_fork_cleanup_threadgroup_lock:
1956 #endif
1957 	delayacct_tsk_free(p);
1958 bad_fork_cleanup_count:
1959 	atomic_dec(&p->cred->user->processes);
1960 	exit_creds(p);
1961 bad_fork_free:
1962 	p->state = TASK_DEAD;
1963 	put_task_stack(p);
1964 	free_task(p);
1965 fork_out:
1966 	return ERR_PTR(retval);
1967 }
1968 
1969 static inline void init_idle_pids(struct pid_link *links)
1970 {
1971 	enum pid_type type;
1972 
1973 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1974 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1975 		links[type].pid = &init_struct_pid;
1976 	}
1977 }
1978 
1979 struct task_struct *fork_idle(int cpu)
1980 {
1981 	struct task_struct *task;
1982 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1983 			    cpu_to_node(cpu));
1984 	if (!IS_ERR(task)) {
1985 		init_idle_pids(task->pids);
1986 		init_idle(task, cpu);
1987 	}
1988 
1989 	return task;
1990 }
1991 
1992 /*
1993  *  Ok, this is the main fork-routine.
1994  *
1995  * It copies the process, and if successful kick-starts
1996  * it and waits for it to finish using the VM if required.
1997  */
1998 long _do_fork(unsigned long clone_flags,
1999 	      unsigned long stack_start,
2000 	      unsigned long stack_size,
2001 	      int __user *parent_tidptr,
2002 	      int __user *child_tidptr,
2003 	      unsigned long tls)
2004 {
2005 	struct task_struct *p;
2006 	int trace = 0;
2007 	long nr;
2008 
2009 	/*
2010 	 * Determine whether and which event to report to ptracer.  When
2011 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
2012 	 * requested, no event is reported; otherwise, report if the event
2013 	 * for the type of forking is enabled.
2014 	 */
2015 	if (!(clone_flags & CLONE_UNTRACED)) {
2016 		if (clone_flags & CLONE_VFORK)
2017 			trace = PTRACE_EVENT_VFORK;
2018 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
2019 			trace = PTRACE_EVENT_CLONE;
2020 		else
2021 			trace = PTRACE_EVENT_FORK;
2022 
2023 		if (likely(!ptrace_event_enabled(current, trace)))
2024 			trace = 0;
2025 	}
2026 
2027 	p = copy_process(clone_flags, stack_start, stack_size,
2028 			 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2029 	add_latent_entropy();
2030 	/*
2031 	 * Do this prior waking up the new thread - the thread pointer
2032 	 * might get invalid after that point, if the thread exits quickly.
2033 	 */
2034 	if (!IS_ERR(p)) {
2035 		struct completion vfork;
2036 		struct pid *pid;
2037 
2038 		trace_sched_process_fork(current, p);
2039 
2040 		pid = get_task_pid(p, PIDTYPE_PID);
2041 		nr = pid_vnr(pid);
2042 
2043 		if (clone_flags & CLONE_PARENT_SETTID)
2044 			put_user(nr, parent_tidptr);
2045 
2046 		if (clone_flags & CLONE_VFORK) {
2047 			p->vfork_done = &vfork;
2048 			init_completion(&vfork);
2049 			get_task_struct(p);
2050 		}
2051 
2052 		wake_up_new_task(p);
2053 
2054 		/* forking complete and child started to run, tell ptracer */
2055 		if (unlikely(trace))
2056 			ptrace_event_pid(trace, pid);
2057 
2058 		if (clone_flags & CLONE_VFORK) {
2059 			if (!wait_for_vfork_done(p, &vfork))
2060 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2061 		}
2062 
2063 		put_pid(pid);
2064 	} else {
2065 		nr = PTR_ERR(p);
2066 	}
2067 	return nr;
2068 }
2069 
2070 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2071 /* For compatibility with architectures that call do_fork directly rather than
2072  * using the syscall entry points below. */
2073 long do_fork(unsigned long clone_flags,
2074 	      unsigned long stack_start,
2075 	      unsigned long stack_size,
2076 	      int __user *parent_tidptr,
2077 	      int __user *child_tidptr)
2078 {
2079 	return _do_fork(clone_flags, stack_start, stack_size,
2080 			parent_tidptr, child_tidptr, 0);
2081 }
2082 #endif
2083 
2084 /*
2085  * Create a kernel thread.
2086  */
2087 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2088 {
2089 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2090 		(unsigned long)arg, NULL, NULL, 0);
2091 }
2092 
2093 #ifdef __ARCH_WANT_SYS_FORK
2094 SYSCALL_DEFINE0(fork)
2095 {
2096 #ifdef CONFIG_MMU
2097 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2098 #else
2099 	/* can not support in nommu mode */
2100 	return -EINVAL;
2101 #endif
2102 }
2103 #endif
2104 
2105 #ifdef __ARCH_WANT_SYS_VFORK
2106 SYSCALL_DEFINE0(vfork)
2107 {
2108 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2109 			0, NULL, NULL, 0);
2110 }
2111 #endif
2112 
2113 #ifdef __ARCH_WANT_SYS_CLONE
2114 #ifdef CONFIG_CLONE_BACKWARDS
2115 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2116 		 int __user *, parent_tidptr,
2117 		 unsigned long, tls,
2118 		 int __user *, child_tidptr)
2119 #elif defined(CONFIG_CLONE_BACKWARDS2)
2120 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2121 		 int __user *, parent_tidptr,
2122 		 int __user *, child_tidptr,
2123 		 unsigned long, tls)
2124 #elif defined(CONFIG_CLONE_BACKWARDS3)
2125 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2126 		int, stack_size,
2127 		int __user *, parent_tidptr,
2128 		int __user *, child_tidptr,
2129 		unsigned long, tls)
2130 #else
2131 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2132 		 int __user *, parent_tidptr,
2133 		 int __user *, child_tidptr,
2134 		 unsigned long, tls)
2135 #endif
2136 {
2137 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2138 }
2139 #endif
2140 
2141 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2142 {
2143 	struct task_struct *leader, *parent, *child;
2144 	int res;
2145 
2146 	read_lock(&tasklist_lock);
2147 	leader = top = top->group_leader;
2148 down:
2149 	for_each_thread(leader, parent) {
2150 		list_for_each_entry(child, &parent->children, sibling) {
2151 			res = visitor(child, data);
2152 			if (res) {
2153 				if (res < 0)
2154 					goto out;
2155 				leader = child;
2156 				goto down;
2157 			}
2158 up:
2159 			;
2160 		}
2161 	}
2162 
2163 	if (leader != top) {
2164 		child = leader;
2165 		parent = child->real_parent;
2166 		leader = parent->group_leader;
2167 		goto up;
2168 	}
2169 out:
2170 	read_unlock(&tasklist_lock);
2171 }
2172 
2173 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2174 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2175 #endif
2176 
2177 static void sighand_ctor(void *data)
2178 {
2179 	struct sighand_struct *sighand = data;
2180 
2181 	spin_lock_init(&sighand->siglock);
2182 	init_waitqueue_head(&sighand->signalfd_wqh);
2183 }
2184 
2185 void __init proc_caches_init(void)
2186 {
2187 	sighand_cachep = kmem_cache_create("sighand_cache",
2188 			sizeof(struct sighand_struct), 0,
2189 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2190 			SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2191 	signal_cachep = kmem_cache_create("signal_cache",
2192 			sizeof(struct signal_struct), 0,
2193 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2194 			NULL);
2195 	files_cachep = kmem_cache_create("files_cache",
2196 			sizeof(struct files_struct), 0,
2197 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2198 			NULL);
2199 	fs_cachep = kmem_cache_create("fs_cache",
2200 			sizeof(struct fs_struct), 0,
2201 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2202 			NULL);
2203 	/*
2204 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2205 	 * whole struct cpumask for the OFFSTACK case. We could change
2206 	 * this to *only* allocate as much of it as required by the
2207 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
2208 	 * is at the end of the structure, exactly for that reason.
2209 	 */
2210 	mm_cachep = kmem_cache_create("mm_struct",
2211 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2212 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2213 			NULL);
2214 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2215 	mmap_init();
2216 	nsproxy_cache_init();
2217 }
2218 
2219 /*
2220  * Check constraints on flags passed to the unshare system call.
2221  */
2222 static int check_unshare_flags(unsigned long unshare_flags)
2223 {
2224 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2225 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2226 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2227 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2228 		return -EINVAL;
2229 	/*
2230 	 * Not implemented, but pretend it works if there is nothing
2231 	 * to unshare.  Note that unsharing the address space or the
2232 	 * signal handlers also need to unshare the signal queues (aka
2233 	 * CLONE_THREAD).
2234 	 */
2235 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2236 		if (!thread_group_empty(current))
2237 			return -EINVAL;
2238 	}
2239 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2240 		if (atomic_read(&current->sighand->count) > 1)
2241 			return -EINVAL;
2242 	}
2243 	if (unshare_flags & CLONE_VM) {
2244 		if (!current_is_single_threaded())
2245 			return -EINVAL;
2246 	}
2247 
2248 	return 0;
2249 }
2250 
2251 /*
2252  * Unshare the filesystem structure if it is being shared
2253  */
2254 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2255 {
2256 	struct fs_struct *fs = current->fs;
2257 
2258 	if (!(unshare_flags & CLONE_FS) || !fs)
2259 		return 0;
2260 
2261 	/* don't need lock here; in the worst case we'll do useless copy */
2262 	if (fs->users == 1)
2263 		return 0;
2264 
2265 	*new_fsp = copy_fs_struct(fs);
2266 	if (!*new_fsp)
2267 		return -ENOMEM;
2268 
2269 	return 0;
2270 }
2271 
2272 /*
2273  * Unshare file descriptor table if it is being shared
2274  */
2275 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2276 {
2277 	struct files_struct *fd = current->files;
2278 	int error = 0;
2279 
2280 	if ((unshare_flags & CLONE_FILES) &&
2281 	    (fd && atomic_read(&fd->count) > 1)) {
2282 		*new_fdp = dup_fd(fd, &error);
2283 		if (!*new_fdp)
2284 			return error;
2285 	}
2286 
2287 	return 0;
2288 }
2289 
2290 /*
2291  * unshare allows a process to 'unshare' part of the process
2292  * context which was originally shared using clone.  copy_*
2293  * functions used by do_fork() cannot be used here directly
2294  * because they modify an inactive task_struct that is being
2295  * constructed. Here we are modifying the current, active,
2296  * task_struct.
2297  */
2298 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2299 {
2300 	struct fs_struct *fs, *new_fs = NULL;
2301 	struct files_struct *fd, *new_fd = NULL;
2302 	struct cred *new_cred = NULL;
2303 	struct nsproxy *new_nsproxy = NULL;
2304 	int do_sysvsem = 0;
2305 	int err;
2306 
2307 	/*
2308 	 * If unsharing a user namespace must also unshare the thread group
2309 	 * and unshare the filesystem root and working directories.
2310 	 */
2311 	if (unshare_flags & CLONE_NEWUSER)
2312 		unshare_flags |= CLONE_THREAD | CLONE_FS;
2313 	/*
2314 	 * If unsharing vm, must also unshare signal handlers.
2315 	 */
2316 	if (unshare_flags & CLONE_VM)
2317 		unshare_flags |= CLONE_SIGHAND;
2318 	/*
2319 	 * If unsharing a signal handlers, must also unshare the signal queues.
2320 	 */
2321 	if (unshare_flags & CLONE_SIGHAND)
2322 		unshare_flags |= CLONE_THREAD;
2323 	/*
2324 	 * If unsharing namespace, must also unshare filesystem information.
2325 	 */
2326 	if (unshare_flags & CLONE_NEWNS)
2327 		unshare_flags |= CLONE_FS;
2328 
2329 	err = check_unshare_flags(unshare_flags);
2330 	if (err)
2331 		goto bad_unshare_out;
2332 	/*
2333 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2334 	 * to a new ipc namespace, the semaphore arrays from the old
2335 	 * namespace are unreachable.
2336 	 */
2337 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2338 		do_sysvsem = 1;
2339 	err = unshare_fs(unshare_flags, &new_fs);
2340 	if (err)
2341 		goto bad_unshare_out;
2342 	err = unshare_fd(unshare_flags, &new_fd);
2343 	if (err)
2344 		goto bad_unshare_cleanup_fs;
2345 	err = unshare_userns(unshare_flags, &new_cred);
2346 	if (err)
2347 		goto bad_unshare_cleanup_fd;
2348 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2349 					 new_cred, new_fs);
2350 	if (err)
2351 		goto bad_unshare_cleanup_cred;
2352 
2353 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2354 		if (do_sysvsem) {
2355 			/*
2356 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2357 			 */
2358 			exit_sem(current);
2359 		}
2360 		if (unshare_flags & CLONE_NEWIPC) {
2361 			/* Orphan segments in old ns (see sem above). */
2362 			exit_shm(current);
2363 			shm_init_task(current);
2364 		}
2365 
2366 		if (new_nsproxy)
2367 			switch_task_namespaces(current, new_nsproxy);
2368 
2369 		task_lock(current);
2370 
2371 		if (new_fs) {
2372 			fs = current->fs;
2373 			spin_lock(&fs->lock);
2374 			current->fs = new_fs;
2375 			if (--fs->users)
2376 				new_fs = NULL;
2377 			else
2378 				new_fs = fs;
2379 			spin_unlock(&fs->lock);
2380 		}
2381 
2382 		if (new_fd) {
2383 			fd = current->files;
2384 			current->files = new_fd;
2385 			new_fd = fd;
2386 		}
2387 
2388 		task_unlock(current);
2389 
2390 		if (new_cred) {
2391 			/* Install the new user namespace */
2392 			commit_creds(new_cred);
2393 			new_cred = NULL;
2394 		}
2395 	}
2396 
2397 	perf_event_namespaces(current);
2398 
2399 bad_unshare_cleanup_cred:
2400 	if (new_cred)
2401 		put_cred(new_cred);
2402 bad_unshare_cleanup_fd:
2403 	if (new_fd)
2404 		put_files_struct(new_fd);
2405 
2406 bad_unshare_cleanup_fs:
2407 	if (new_fs)
2408 		free_fs_struct(new_fs);
2409 
2410 bad_unshare_out:
2411 	return err;
2412 }
2413 
2414 /*
2415  *	Helper to unshare the files of the current task.
2416  *	We don't want to expose copy_files internals to
2417  *	the exec layer of the kernel.
2418  */
2419 
2420 int unshare_files(struct files_struct **displaced)
2421 {
2422 	struct task_struct *task = current;
2423 	struct files_struct *copy = NULL;
2424 	int error;
2425 
2426 	error = unshare_fd(CLONE_FILES, &copy);
2427 	if (error || !copy) {
2428 		*displaced = NULL;
2429 		return error;
2430 	}
2431 	*displaced = task->files;
2432 	task_lock(task);
2433 	task->files = copy;
2434 	task_unlock(task);
2435 	return 0;
2436 }
2437 
2438 int sysctl_max_threads(struct ctl_table *table, int write,
2439 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2440 {
2441 	struct ctl_table t;
2442 	int ret;
2443 	int threads = max_threads;
2444 	int min = MIN_THREADS;
2445 	int max = MAX_THREADS;
2446 
2447 	t = *table;
2448 	t.data = &threads;
2449 	t.extra1 = &min;
2450 	t.extra2 = &max;
2451 
2452 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2453 	if (ret || !write)
2454 		return ret;
2455 
2456 	set_max_threads(threads);
2457 
2458 	return 0;
2459 }
2460