xref: /openbmc/linux/kernel/fork.c (revision 174cd4b1)
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/rtmutex.h>
21 #include <linux/init.h>
22 #include <linux/unistd.h>
23 #include <linux/module.h>
24 #include <linux/vmalloc.h>
25 #include <linux/completion.h>
26 #include <linux/personality.h>
27 #include <linux/mempolicy.h>
28 #include <linux/sem.h>
29 #include <linux/file.h>
30 #include <linux/fdtable.h>
31 #include <linux/iocontext.h>
32 #include <linux/key.h>
33 #include <linux/binfmts.h>
34 #include <linux/mman.h>
35 #include <linux/mmu_notifier.h>
36 #include <linux/fs.h>
37 #include <linux/mm.h>
38 #include <linux/vmacache.h>
39 #include <linux/nsproxy.h>
40 #include <linux/capability.h>
41 #include <linux/cpu.h>
42 #include <linux/cgroup.h>
43 #include <linux/security.h>
44 #include <linux/hugetlb.h>
45 #include <linux/seccomp.h>
46 #include <linux/swap.h>
47 #include <linux/syscalls.h>
48 #include <linux/jiffies.h>
49 #include <linux/futex.h>
50 #include <linux/compat.h>
51 #include <linux/kthread.h>
52 #include <linux/task_io_accounting_ops.h>
53 #include <linux/rcupdate.h>
54 #include <linux/ptrace.h>
55 #include <linux/mount.h>
56 #include <linux/audit.h>
57 #include <linux/memcontrol.h>
58 #include <linux/ftrace.h>
59 #include <linux/proc_fs.h>
60 #include <linux/profile.h>
61 #include <linux/rmap.h>
62 #include <linux/ksm.h>
63 #include <linux/acct.h>
64 #include <linux/userfaultfd_k.h>
65 #include <linux/tsacct_kern.h>
66 #include <linux/cn_proc.h>
67 #include <linux/freezer.h>
68 #include <linux/delayacct.h>
69 #include <linux/taskstats_kern.h>
70 #include <linux/random.h>
71 #include <linux/tty.h>
72 #include <linux/blkdev.h>
73 #include <linux/fs_struct.h>
74 #include <linux/magic.h>
75 #include <linux/perf_event.h>
76 #include <linux/posix-timers.h>
77 #include <linux/user-return-notifier.h>
78 #include <linux/oom.h>
79 #include <linux/khugepaged.h>
80 #include <linux/signalfd.h>
81 #include <linux/uprobes.h>
82 #include <linux/aio.h>
83 #include <linux/compiler.h>
84 #include <linux/sysctl.h>
85 #include <linux/kcov.h>
86 
87 #include <asm/pgtable.h>
88 #include <asm/pgalloc.h>
89 #include <linux/uaccess.h>
90 #include <asm/mmu_context.h>
91 #include <asm/cacheflush.h>
92 #include <asm/tlbflush.h>
93 
94 #include <trace/events/sched.h>
95 
96 #define CREATE_TRACE_POINTS
97 #include <trace/events/task.h>
98 
99 /*
100  * Minimum number of threads to boot the kernel
101  */
102 #define MIN_THREADS 20
103 
104 /*
105  * Maximum number of threads
106  */
107 #define MAX_THREADS FUTEX_TID_MASK
108 
109 /*
110  * Protected counters by write_lock_irq(&tasklist_lock)
111  */
112 unsigned long total_forks;	/* Handle normal Linux uptimes. */
113 int nr_threads;			/* The idle threads do not count.. */
114 
115 int max_threads;		/* tunable limit on nr_threads */
116 
117 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
118 
119 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
120 
121 #ifdef CONFIG_PROVE_RCU
122 int lockdep_tasklist_lock_is_held(void)
123 {
124 	return lockdep_is_held(&tasklist_lock);
125 }
126 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
127 #endif /* #ifdef CONFIG_PROVE_RCU */
128 
129 int nr_processes(void)
130 {
131 	int cpu;
132 	int total = 0;
133 
134 	for_each_possible_cpu(cpu)
135 		total += per_cpu(process_counts, cpu);
136 
137 	return total;
138 }
139 
140 void __weak arch_release_task_struct(struct task_struct *tsk)
141 {
142 }
143 
144 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
145 static struct kmem_cache *task_struct_cachep;
146 
147 static inline struct task_struct *alloc_task_struct_node(int node)
148 {
149 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
150 }
151 
152 static inline void free_task_struct(struct task_struct *tsk)
153 {
154 	kmem_cache_free(task_struct_cachep, tsk);
155 }
156 #endif
157 
158 void __weak arch_release_thread_stack(unsigned long *stack)
159 {
160 }
161 
162 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
163 
164 /*
165  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
166  * kmemcache based allocator.
167  */
168 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
169 
170 #ifdef CONFIG_VMAP_STACK
171 /*
172  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
173  * flush.  Try to minimize the number of calls by caching stacks.
174  */
175 #define NR_CACHED_STACKS 2
176 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
177 #endif
178 
179 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
180 {
181 #ifdef CONFIG_VMAP_STACK
182 	void *stack;
183 	int i;
184 
185 	local_irq_disable();
186 	for (i = 0; i < NR_CACHED_STACKS; i++) {
187 		struct vm_struct *s = this_cpu_read(cached_stacks[i]);
188 
189 		if (!s)
190 			continue;
191 		this_cpu_write(cached_stacks[i], NULL);
192 
193 		tsk->stack_vm_area = s;
194 		local_irq_enable();
195 		return s->addr;
196 	}
197 	local_irq_enable();
198 
199 	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
200 				     VMALLOC_START, VMALLOC_END,
201 				     THREADINFO_GFP | __GFP_HIGHMEM,
202 				     PAGE_KERNEL,
203 				     0, node, __builtin_return_address(0));
204 
205 	/*
206 	 * We can't call find_vm_area() in interrupt context, and
207 	 * free_thread_stack() can be called in interrupt context,
208 	 * so cache the vm_struct.
209 	 */
210 	if (stack)
211 		tsk->stack_vm_area = find_vm_area(stack);
212 	return stack;
213 #else
214 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
215 					     THREAD_SIZE_ORDER);
216 
217 	return page ? page_address(page) : NULL;
218 #endif
219 }
220 
221 static inline void free_thread_stack(struct task_struct *tsk)
222 {
223 #ifdef CONFIG_VMAP_STACK
224 	if (task_stack_vm_area(tsk)) {
225 		unsigned long flags;
226 		int i;
227 
228 		local_irq_save(flags);
229 		for (i = 0; i < NR_CACHED_STACKS; i++) {
230 			if (this_cpu_read(cached_stacks[i]))
231 				continue;
232 
233 			this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
234 			local_irq_restore(flags);
235 			return;
236 		}
237 		local_irq_restore(flags);
238 
239 		vfree_atomic(tsk->stack);
240 		return;
241 	}
242 #endif
243 
244 	__free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
245 }
246 # else
247 static struct kmem_cache *thread_stack_cache;
248 
249 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
250 						  int node)
251 {
252 	return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
253 }
254 
255 static void free_thread_stack(struct task_struct *tsk)
256 {
257 	kmem_cache_free(thread_stack_cache, tsk->stack);
258 }
259 
260 void thread_stack_cache_init(void)
261 {
262 	thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
263 					      THREAD_SIZE, 0, NULL);
264 	BUG_ON(thread_stack_cache == NULL);
265 }
266 # endif
267 #endif
268 
269 /* SLAB cache for signal_struct structures (tsk->signal) */
270 static struct kmem_cache *signal_cachep;
271 
272 /* SLAB cache for sighand_struct structures (tsk->sighand) */
273 struct kmem_cache *sighand_cachep;
274 
275 /* SLAB cache for files_struct structures (tsk->files) */
276 struct kmem_cache *files_cachep;
277 
278 /* SLAB cache for fs_struct structures (tsk->fs) */
279 struct kmem_cache *fs_cachep;
280 
281 /* SLAB cache for vm_area_struct structures */
282 struct kmem_cache *vm_area_cachep;
283 
284 /* SLAB cache for mm_struct structures (tsk->mm) */
285 static struct kmem_cache *mm_cachep;
286 
287 static void account_kernel_stack(struct task_struct *tsk, int account)
288 {
289 	void *stack = task_stack_page(tsk);
290 	struct vm_struct *vm = task_stack_vm_area(tsk);
291 
292 	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
293 
294 	if (vm) {
295 		int i;
296 
297 		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
298 
299 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
300 			mod_zone_page_state(page_zone(vm->pages[i]),
301 					    NR_KERNEL_STACK_KB,
302 					    PAGE_SIZE / 1024 * account);
303 		}
304 
305 		/* All stack pages belong to the same memcg. */
306 		memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
307 					    account * (THREAD_SIZE / 1024));
308 	} else {
309 		/*
310 		 * All stack pages are in the same zone and belong to the
311 		 * same memcg.
312 		 */
313 		struct page *first_page = virt_to_page(stack);
314 
315 		mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
316 				    THREAD_SIZE / 1024 * account);
317 
318 		memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
319 					    account * (THREAD_SIZE / 1024));
320 	}
321 }
322 
323 static void release_task_stack(struct task_struct *tsk)
324 {
325 	if (WARN_ON(tsk->state != TASK_DEAD))
326 		return;  /* Better to leak the stack than to free prematurely */
327 
328 	account_kernel_stack(tsk, -1);
329 	arch_release_thread_stack(tsk->stack);
330 	free_thread_stack(tsk);
331 	tsk->stack = NULL;
332 #ifdef CONFIG_VMAP_STACK
333 	tsk->stack_vm_area = NULL;
334 #endif
335 }
336 
337 #ifdef CONFIG_THREAD_INFO_IN_TASK
338 void put_task_stack(struct task_struct *tsk)
339 {
340 	if (atomic_dec_and_test(&tsk->stack_refcount))
341 		release_task_stack(tsk);
342 }
343 #endif
344 
345 void free_task(struct task_struct *tsk)
346 {
347 #ifndef CONFIG_THREAD_INFO_IN_TASK
348 	/*
349 	 * The task is finally done with both the stack and thread_info,
350 	 * so free both.
351 	 */
352 	release_task_stack(tsk);
353 #else
354 	/*
355 	 * If the task had a separate stack allocation, it should be gone
356 	 * by now.
357 	 */
358 	WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
359 #endif
360 	rt_mutex_debug_task_free(tsk);
361 	ftrace_graph_exit_task(tsk);
362 	put_seccomp_filter(tsk);
363 	arch_release_task_struct(tsk);
364 	if (tsk->flags & PF_KTHREAD)
365 		free_kthread_struct(tsk);
366 	free_task_struct(tsk);
367 }
368 EXPORT_SYMBOL(free_task);
369 
370 static inline void free_signal_struct(struct signal_struct *sig)
371 {
372 	taskstats_tgid_free(sig);
373 	sched_autogroup_exit(sig);
374 	/*
375 	 * __mmdrop is not safe to call from softirq context on x86 due to
376 	 * pgd_dtor so postpone it to the async context
377 	 */
378 	if (sig->oom_mm)
379 		mmdrop_async(sig->oom_mm);
380 	kmem_cache_free(signal_cachep, sig);
381 }
382 
383 static inline void put_signal_struct(struct signal_struct *sig)
384 {
385 	if (atomic_dec_and_test(&sig->sigcnt))
386 		free_signal_struct(sig);
387 }
388 
389 void __put_task_struct(struct task_struct *tsk)
390 {
391 	WARN_ON(!tsk->exit_state);
392 	WARN_ON(atomic_read(&tsk->usage));
393 	WARN_ON(tsk == current);
394 
395 	cgroup_free(tsk);
396 	task_numa_free(tsk);
397 	security_task_free(tsk);
398 	exit_creds(tsk);
399 	delayacct_tsk_free(tsk);
400 	put_signal_struct(tsk->signal);
401 
402 	if (!profile_handoff_task(tsk))
403 		free_task(tsk);
404 }
405 EXPORT_SYMBOL_GPL(__put_task_struct);
406 
407 void __init __weak arch_task_cache_init(void) { }
408 
409 /*
410  * set_max_threads
411  */
412 static void set_max_threads(unsigned int max_threads_suggested)
413 {
414 	u64 threads;
415 
416 	/*
417 	 * The number of threads shall be limited such that the thread
418 	 * structures may only consume a small part of the available memory.
419 	 */
420 	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
421 		threads = MAX_THREADS;
422 	else
423 		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
424 				    (u64) THREAD_SIZE * 8UL);
425 
426 	if (threads > max_threads_suggested)
427 		threads = max_threads_suggested;
428 
429 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
430 }
431 
432 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
433 /* Initialized by the architecture: */
434 int arch_task_struct_size __read_mostly;
435 #endif
436 
437 void __init fork_init(void)
438 {
439 	int i;
440 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
441 #ifndef ARCH_MIN_TASKALIGN
442 #define ARCH_MIN_TASKALIGN	0
443 #endif
444 	int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
445 
446 	/* create a slab on which task_structs can be allocated */
447 	task_struct_cachep = kmem_cache_create("task_struct",
448 			arch_task_struct_size, align,
449 			SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
450 #endif
451 
452 	/* do the arch specific task caches init */
453 	arch_task_cache_init();
454 
455 	set_max_threads(MAX_THREADS);
456 
457 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
458 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
459 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
460 		init_task.signal->rlim[RLIMIT_NPROC];
461 
462 	for (i = 0; i < UCOUNT_COUNTS; i++) {
463 		init_user_ns.ucount_max[i] = max_threads/2;
464 	}
465 }
466 
467 int __weak arch_dup_task_struct(struct task_struct *dst,
468 					       struct task_struct *src)
469 {
470 	*dst = *src;
471 	return 0;
472 }
473 
474 void set_task_stack_end_magic(struct task_struct *tsk)
475 {
476 	unsigned long *stackend;
477 
478 	stackend = end_of_stack(tsk);
479 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
480 }
481 
482 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
483 {
484 	struct task_struct *tsk;
485 	unsigned long *stack;
486 	struct vm_struct *stack_vm_area;
487 	int err;
488 
489 	if (node == NUMA_NO_NODE)
490 		node = tsk_fork_get_node(orig);
491 	tsk = alloc_task_struct_node(node);
492 	if (!tsk)
493 		return NULL;
494 
495 	stack = alloc_thread_stack_node(tsk, node);
496 	if (!stack)
497 		goto free_tsk;
498 
499 	stack_vm_area = task_stack_vm_area(tsk);
500 
501 	err = arch_dup_task_struct(tsk, orig);
502 
503 	/*
504 	 * arch_dup_task_struct() clobbers the stack-related fields.  Make
505 	 * sure they're properly initialized before using any stack-related
506 	 * functions again.
507 	 */
508 	tsk->stack = stack;
509 #ifdef CONFIG_VMAP_STACK
510 	tsk->stack_vm_area = stack_vm_area;
511 #endif
512 #ifdef CONFIG_THREAD_INFO_IN_TASK
513 	atomic_set(&tsk->stack_refcount, 1);
514 #endif
515 
516 	if (err)
517 		goto free_stack;
518 
519 #ifdef CONFIG_SECCOMP
520 	/*
521 	 * We must handle setting up seccomp filters once we're under
522 	 * the sighand lock in case orig has changed between now and
523 	 * then. Until then, filter must be NULL to avoid messing up
524 	 * the usage counts on the error path calling free_task.
525 	 */
526 	tsk->seccomp.filter = NULL;
527 #endif
528 
529 	setup_thread_stack(tsk, orig);
530 	clear_user_return_notifier(tsk);
531 	clear_tsk_need_resched(tsk);
532 	set_task_stack_end_magic(tsk);
533 
534 #ifdef CONFIG_CC_STACKPROTECTOR
535 	tsk->stack_canary = get_random_int();
536 #endif
537 
538 	/*
539 	 * One for us, one for whoever does the "release_task()" (usually
540 	 * parent)
541 	 */
542 	atomic_set(&tsk->usage, 2);
543 #ifdef CONFIG_BLK_DEV_IO_TRACE
544 	tsk->btrace_seq = 0;
545 #endif
546 	tsk->splice_pipe = NULL;
547 	tsk->task_frag.page = NULL;
548 	tsk->wake_q.next = NULL;
549 
550 	account_kernel_stack(tsk, 1);
551 
552 	kcov_task_init(tsk);
553 
554 	return tsk;
555 
556 free_stack:
557 	free_thread_stack(tsk);
558 free_tsk:
559 	free_task_struct(tsk);
560 	return NULL;
561 }
562 
563 #ifdef CONFIG_MMU
564 static __latent_entropy int dup_mmap(struct mm_struct *mm,
565 					struct mm_struct *oldmm)
566 {
567 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
568 	struct rb_node **rb_link, *rb_parent;
569 	int retval;
570 	unsigned long charge;
571 	LIST_HEAD(uf);
572 
573 	uprobe_start_dup_mmap();
574 	if (down_write_killable(&oldmm->mmap_sem)) {
575 		retval = -EINTR;
576 		goto fail_uprobe_end;
577 	}
578 	flush_cache_dup_mm(oldmm);
579 	uprobe_dup_mmap(oldmm, mm);
580 	/*
581 	 * Not linked in yet - no deadlock potential:
582 	 */
583 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
584 
585 	/* No ordering required: file already has been exposed. */
586 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
587 
588 	mm->total_vm = oldmm->total_vm;
589 	mm->data_vm = oldmm->data_vm;
590 	mm->exec_vm = oldmm->exec_vm;
591 	mm->stack_vm = oldmm->stack_vm;
592 
593 	rb_link = &mm->mm_rb.rb_node;
594 	rb_parent = NULL;
595 	pprev = &mm->mmap;
596 	retval = ksm_fork(mm, oldmm);
597 	if (retval)
598 		goto out;
599 	retval = khugepaged_fork(mm, oldmm);
600 	if (retval)
601 		goto out;
602 
603 	prev = NULL;
604 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
605 		struct file *file;
606 
607 		if (mpnt->vm_flags & VM_DONTCOPY) {
608 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
609 			continue;
610 		}
611 		charge = 0;
612 		if (mpnt->vm_flags & VM_ACCOUNT) {
613 			unsigned long len = vma_pages(mpnt);
614 
615 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
616 				goto fail_nomem;
617 			charge = len;
618 		}
619 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
620 		if (!tmp)
621 			goto fail_nomem;
622 		*tmp = *mpnt;
623 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
624 		retval = vma_dup_policy(mpnt, tmp);
625 		if (retval)
626 			goto fail_nomem_policy;
627 		tmp->vm_mm = mm;
628 		retval = dup_userfaultfd(tmp, &uf);
629 		if (retval)
630 			goto fail_nomem_anon_vma_fork;
631 		if (anon_vma_fork(tmp, mpnt))
632 			goto fail_nomem_anon_vma_fork;
633 		tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
634 		tmp->vm_next = tmp->vm_prev = NULL;
635 		file = tmp->vm_file;
636 		if (file) {
637 			struct inode *inode = file_inode(file);
638 			struct address_space *mapping = file->f_mapping;
639 
640 			get_file(file);
641 			if (tmp->vm_flags & VM_DENYWRITE)
642 				atomic_dec(&inode->i_writecount);
643 			i_mmap_lock_write(mapping);
644 			if (tmp->vm_flags & VM_SHARED)
645 				atomic_inc(&mapping->i_mmap_writable);
646 			flush_dcache_mmap_lock(mapping);
647 			/* insert tmp into the share list, just after mpnt */
648 			vma_interval_tree_insert_after(tmp, mpnt,
649 					&mapping->i_mmap);
650 			flush_dcache_mmap_unlock(mapping);
651 			i_mmap_unlock_write(mapping);
652 		}
653 
654 		/*
655 		 * Clear hugetlb-related page reserves for children. This only
656 		 * affects MAP_PRIVATE mappings. Faults generated by the child
657 		 * are not guaranteed to succeed, even if read-only
658 		 */
659 		if (is_vm_hugetlb_page(tmp))
660 			reset_vma_resv_huge_pages(tmp);
661 
662 		/*
663 		 * Link in the new vma and copy the page table entries.
664 		 */
665 		*pprev = tmp;
666 		pprev = &tmp->vm_next;
667 		tmp->vm_prev = prev;
668 		prev = tmp;
669 
670 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
671 		rb_link = &tmp->vm_rb.rb_right;
672 		rb_parent = &tmp->vm_rb;
673 
674 		mm->map_count++;
675 		retval = copy_page_range(mm, oldmm, mpnt);
676 
677 		if (tmp->vm_ops && tmp->vm_ops->open)
678 			tmp->vm_ops->open(tmp);
679 
680 		if (retval)
681 			goto out;
682 	}
683 	/* a new mm has just been created */
684 	arch_dup_mmap(oldmm, mm);
685 	retval = 0;
686 out:
687 	up_write(&mm->mmap_sem);
688 	flush_tlb_mm(oldmm);
689 	up_write(&oldmm->mmap_sem);
690 	dup_userfaultfd_complete(&uf);
691 fail_uprobe_end:
692 	uprobe_end_dup_mmap();
693 	return retval;
694 fail_nomem_anon_vma_fork:
695 	mpol_put(vma_policy(tmp));
696 fail_nomem_policy:
697 	kmem_cache_free(vm_area_cachep, tmp);
698 fail_nomem:
699 	retval = -ENOMEM;
700 	vm_unacct_memory(charge);
701 	goto out;
702 }
703 
704 static inline int mm_alloc_pgd(struct mm_struct *mm)
705 {
706 	mm->pgd = pgd_alloc(mm);
707 	if (unlikely(!mm->pgd))
708 		return -ENOMEM;
709 	return 0;
710 }
711 
712 static inline void mm_free_pgd(struct mm_struct *mm)
713 {
714 	pgd_free(mm, mm->pgd);
715 }
716 #else
717 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
718 {
719 	down_write(&oldmm->mmap_sem);
720 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
721 	up_write(&oldmm->mmap_sem);
722 	return 0;
723 }
724 #define mm_alloc_pgd(mm)	(0)
725 #define mm_free_pgd(mm)
726 #endif /* CONFIG_MMU */
727 
728 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
729 
730 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
731 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
732 
733 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
734 
735 static int __init coredump_filter_setup(char *s)
736 {
737 	default_dump_filter =
738 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
739 		MMF_DUMP_FILTER_MASK;
740 	return 1;
741 }
742 
743 __setup("coredump_filter=", coredump_filter_setup);
744 
745 #include <linux/init_task.h>
746 
747 static void mm_init_aio(struct mm_struct *mm)
748 {
749 #ifdef CONFIG_AIO
750 	spin_lock_init(&mm->ioctx_lock);
751 	mm->ioctx_table = NULL;
752 #endif
753 }
754 
755 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
756 {
757 #ifdef CONFIG_MEMCG
758 	mm->owner = p;
759 #endif
760 }
761 
762 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
763 	struct user_namespace *user_ns)
764 {
765 	mm->mmap = NULL;
766 	mm->mm_rb = RB_ROOT;
767 	mm->vmacache_seqnum = 0;
768 	atomic_set(&mm->mm_users, 1);
769 	atomic_set(&mm->mm_count, 1);
770 	init_rwsem(&mm->mmap_sem);
771 	INIT_LIST_HEAD(&mm->mmlist);
772 	mm->core_state = NULL;
773 	atomic_long_set(&mm->nr_ptes, 0);
774 	mm_nr_pmds_init(mm);
775 	mm->map_count = 0;
776 	mm->locked_vm = 0;
777 	mm->pinned_vm = 0;
778 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
779 	spin_lock_init(&mm->page_table_lock);
780 	mm_init_cpumask(mm);
781 	mm_init_aio(mm);
782 	mm_init_owner(mm, p);
783 	mmu_notifier_mm_init(mm);
784 	clear_tlb_flush_pending(mm);
785 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
786 	mm->pmd_huge_pte = NULL;
787 #endif
788 
789 	if (current->mm) {
790 		mm->flags = current->mm->flags & MMF_INIT_MASK;
791 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
792 	} else {
793 		mm->flags = default_dump_filter;
794 		mm->def_flags = 0;
795 	}
796 
797 	if (mm_alloc_pgd(mm))
798 		goto fail_nopgd;
799 
800 	if (init_new_context(p, mm))
801 		goto fail_nocontext;
802 
803 	mm->user_ns = get_user_ns(user_ns);
804 	return mm;
805 
806 fail_nocontext:
807 	mm_free_pgd(mm);
808 fail_nopgd:
809 	free_mm(mm);
810 	return NULL;
811 }
812 
813 static void check_mm(struct mm_struct *mm)
814 {
815 	int i;
816 
817 	for (i = 0; i < NR_MM_COUNTERS; i++) {
818 		long x = atomic_long_read(&mm->rss_stat.count[i]);
819 
820 		if (unlikely(x))
821 			printk(KERN_ALERT "BUG: Bad rss-counter state "
822 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
823 	}
824 
825 	if (atomic_long_read(&mm->nr_ptes))
826 		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
827 				atomic_long_read(&mm->nr_ptes));
828 	if (mm_nr_pmds(mm))
829 		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
830 				mm_nr_pmds(mm));
831 
832 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
833 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
834 #endif
835 }
836 
837 /*
838  * Allocate and initialize an mm_struct.
839  */
840 struct mm_struct *mm_alloc(void)
841 {
842 	struct mm_struct *mm;
843 
844 	mm = allocate_mm();
845 	if (!mm)
846 		return NULL;
847 
848 	memset(mm, 0, sizeof(*mm));
849 	return mm_init(mm, current, current_user_ns());
850 }
851 
852 /*
853  * Called when the last reference to the mm
854  * is dropped: either by a lazy thread or by
855  * mmput. Free the page directory and the mm.
856  */
857 void __mmdrop(struct mm_struct *mm)
858 {
859 	BUG_ON(mm == &init_mm);
860 	mm_free_pgd(mm);
861 	destroy_context(mm);
862 	mmu_notifier_mm_destroy(mm);
863 	check_mm(mm);
864 	put_user_ns(mm->user_ns);
865 	free_mm(mm);
866 }
867 EXPORT_SYMBOL_GPL(__mmdrop);
868 
869 static inline void __mmput(struct mm_struct *mm)
870 {
871 	VM_BUG_ON(atomic_read(&mm->mm_users));
872 
873 	uprobe_clear_state(mm);
874 	exit_aio(mm);
875 	ksm_exit(mm);
876 	khugepaged_exit(mm); /* must run before exit_mmap */
877 	exit_mmap(mm);
878 	mm_put_huge_zero_page(mm);
879 	set_mm_exe_file(mm, NULL);
880 	if (!list_empty(&mm->mmlist)) {
881 		spin_lock(&mmlist_lock);
882 		list_del(&mm->mmlist);
883 		spin_unlock(&mmlist_lock);
884 	}
885 	if (mm->binfmt)
886 		module_put(mm->binfmt->module);
887 	set_bit(MMF_OOM_SKIP, &mm->flags);
888 	mmdrop(mm);
889 }
890 
891 /*
892  * Decrement the use count and release all resources for an mm.
893  */
894 void mmput(struct mm_struct *mm)
895 {
896 	might_sleep();
897 
898 	if (atomic_dec_and_test(&mm->mm_users))
899 		__mmput(mm);
900 }
901 EXPORT_SYMBOL_GPL(mmput);
902 
903 #ifdef CONFIG_MMU
904 static void mmput_async_fn(struct work_struct *work)
905 {
906 	struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
907 	__mmput(mm);
908 }
909 
910 void mmput_async(struct mm_struct *mm)
911 {
912 	if (atomic_dec_and_test(&mm->mm_users)) {
913 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
914 		schedule_work(&mm->async_put_work);
915 	}
916 }
917 #endif
918 
919 /**
920  * set_mm_exe_file - change a reference to the mm's executable file
921  *
922  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
923  *
924  * Main users are mmput() and sys_execve(). Callers prevent concurrent
925  * invocations: in mmput() nobody alive left, in execve task is single
926  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
927  * mm->exe_file, but does so without using set_mm_exe_file() in order
928  * to do avoid the need for any locks.
929  */
930 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
931 {
932 	struct file *old_exe_file;
933 
934 	/*
935 	 * It is safe to dereference the exe_file without RCU as
936 	 * this function is only called if nobody else can access
937 	 * this mm -- see comment above for justification.
938 	 */
939 	old_exe_file = rcu_dereference_raw(mm->exe_file);
940 
941 	if (new_exe_file)
942 		get_file(new_exe_file);
943 	rcu_assign_pointer(mm->exe_file, new_exe_file);
944 	if (old_exe_file)
945 		fput(old_exe_file);
946 }
947 
948 /**
949  * get_mm_exe_file - acquire a reference to the mm's executable file
950  *
951  * Returns %NULL if mm has no associated executable file.
952  * User must release file via fput().
953  */
954 struct file *get_mm_exe_file(struct mm_struct *mm)
955 {
956 	struct file *exe_file;
957 
958 	rcu_read_lock();
959 	exe_file = rcu_dereference(mm->exe_file);
960 	if (exe_file && !get_file_rcu(exe_file))
961 		exe_file = NULL;
962 	rcu_read_unlock();
963 	return exe_file;
964 }
965 EXPORT_SYMBOL(get_mm_exe_file);
966 
967 /**
968  * get_task_exe_file - acquire a reference to the task's executable file
969  *
970  * Returns %NULL if task's mm (if any) has no associated executable file or
971  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
972  * User must release file via fput().
973  */
974 struct file *get_task_exe_file(struct task_struct *task)
975 {
976 	struct file *exe_file = NULL;
977 	struct mm_struct *mm;
978 
979 	task_lock(task);
980 	mm = task->mm;
981 	if (mm) {
982 		if (!(task->flags & PF_KTHREAD))
983 			exe_file = get_mm_exe_file(mm);
984 	}
985 	task_unlock(task);
986 	return exe_file;
987 }
988 EXPORT_SYMBOL(get_task_exe_file);
989 
990 /**
991  * get_task_mm - acquire a reference to the task's mm
992  *
993  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
994  * this kernel workthread has transiently adopted a user mm with use_mm,
995  * to do its AIO) is not set and if so returns a reference to it, after
996  * bumping up the use count.  User must release the mm via mmput()
997  * after use.  Typically used by /proc and ptrace.
998  */
999 struct mm_struct *get_task_mm(struct task_struct *task)
1000 {
1001 	struct mm_struct *mm;
1002 
1003 	task_lock(task);
1004 	mm = task->mm;
1005 	if (mm) {
1006 		if (task->flags & PF_KTHREAD)
1007 			mm = NULL;
1008 		else
1009 			mmget(mm);
1010 	}
1011 	task_unlock(task);
1012 	return mm;
1013 }
1014 EXPORT_SYMBOL_GPL(get_task_mm);
1015 
1016 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1017 {
1018 	struct mm_struct *mm;
1019 	int err;
1020 
1021 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1022 	if (err)
1023 		return ERR_PTR(err);
1024 
1025 	mm = get_task_mm(task);
1026 	if (mm && mm != current->mm &&
1027 			!ptrace_may_access(task, mode)) {
1028 		mmput(mm);
1029 		mm = ERR_PTR(-EACCES);
1030 	}
1031 	mutex_unlock(&task->signal->cred_guard_mutex);
1032 
1033 	return mm;
1034 }
1035 
1036 static void complete_vfork_done(struct task_struct *tsk)
1037 {
1038 	struct completion *vfork;
1039 
1040 	task_lock(tsk);
1041 	vfork = tsk->vfork_done;
1042 	if (likely(vfork)) {
1043 		tsk->vfork_done = NULL;
1044 		complete(vfork);
1045 	}
1046 	task_unlock(tsk);
1047 }
1048 
1049 static int wait_for_vfork_done(struct task_struct *child,
1050 				struct completion *vfork)
1051 {
1052 	int killed;
1053 
1054 	freezer_do_not_count();
1055 	killed = wait_for_completion_killable(vfork);
1056 	freezer_count();
1057 
1058 	if (killed) {
1059 		task_lock(child);
1060 		child->vfork_done = NULL;
1061 		task_unlock(child);
1062 	}
1063 
1064 	put_task_struct(child);
1065 	return killed;
1066 }
1067 
1068 /* Please note the differences between mmput and mm_release.
1069  * mmput is called whenever we stop holding onto a mm_struct,
1070  * error success whatever.
1071  *
1072  * mm_release is called after a mm_struct has been removed
1073  * from the current process.
1074  *
1075  * This difference is important for error handling, when we
1076  * only half set up a mm_struct for a new process and need to restore
1077  * the old one.  Because we mmput the new mm_struct before
1078  * restoring the old one. . .
1079  * Eric Biederman 10 January 1998
1080  */
1081 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1082 {
1083 	/* Get rid of any futexes when releasing the mm */
1084 #ifdef CONFIG_FUTEX
1085 	if (unlikely(tsk->robust_list)) {
1086 		exit_robust_list(tsk);
1087 		tsk->robust_list = NULL;
1088 	}
1089 #ifdef CONFIG_COMPAT
1090 	if (unlikely(tsk->compat_robust_list)) {
1091 		compat_exit_robust_list(tsk);
1092 		tsk->compat_robust_list = NULL;
1093 	}
1094 #endif
1095 	if (unlikely(!list_empty(&tsk->pi_state_list)))
1096 		exit_pi_state_list(tsk);
1097 #endif
1098 
1099 	uprobe_free_utask(tsk);
1100 
1101 	/* Get rid of any cached register state */
1102 	deactivate_mm(tsk, mm);
1103 
1104 	/*
1105 	 * Signal userspace if we're not exiting with a core dump
1106 	 * because we want to leave the value intact for debugging
1107 	 * purposes.
1108 	 */
1109 	if (tsk->clear_child_tid) {
1110 		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1111 		    atomic_read(&mm->mm_users) > 1) {
1112 			/*
1113 			 * We don't check the error code - if userspace has
1114 			 * not set up a proper pointer then tough luck.
1115 			 */
1116 			put_user(0, tsk->clear_child_tid);
1117 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1118 					1, NULL, NULL, 0);
1119 		}
1120 		tsk->clear_child_tid = NULL;
1121 	}
1122 
1123 	/*
1124 	 * All done, finally we can wake up parent and return this mm to him.
1125 	 * Also kthread_stop() uses this completion for synchronization.
1126 	 */
1127 	if (tsk->vfork_done)
1128 		complete_vfork_done(tsk);
1129 }
1130 
1131 /*
1132  * Allocate a new mm structure and copy contents from the
1133  * mm structure of the passed in task structure.
1134  */
1135 static struct mm_struct *dup_mm(struct task_struct *tsk)
1136 {
1137 	struct mm_struct *mm, *oldmm = current->mm;
1138 	int err;
1139 
1140 	mm = allocate_mm();
1141 	if (!mm)
1142 		goto fail_nomem;
1143 
1144 	memcpy(mm, oldmm, sizeof(*mm));
1145 
1146 	if (!mm_init(mm, tsk, mm->user_ns))
1147 		goto fail_nomem;
1148 
1149 	err = dup_mmap(mm, oldmm);
1150 	if (err)
1151 		goto free_pt;
1152 
1153 	mm->hiwater_rss = get_mm_rss(mm);
1154 	mm->hiwater_vm = mm->total_vm;
1155 
1156 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1157 		goto free_pt;
1158 
1159 	return mm;
1160 
1161 free_pt:
1162 	/* don't put binfmt in mmput, we haven't got module yet */
1163 	mm->binfmt = NULL;
1164 	mmput(mm);
1165 
1166 fail_nomem:
1167 	return NULL;
1168 }
1169 
1170 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1171 {
1172 	struct mm_struct *mm, *oldmm;
1173 	int retval;
1174 
1175 	tsk->min_flt = tsk->maj_flt = 0;
1176 	tsk->nvcsw = tsk->nivcsw = 0;
1177 #ifdef CONFIG_DETECT_HUNG_TASK
1178 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1179 #endif
1180 
1181 	tsk->mm = NULL;
1182 	tsk->active_mm = NULL;
1183 
1184 	/*
1185 	 * Are we cloning a kernel thread?
1186 	 *
1187 	 * We need to steal a active VM for that..
1188 	 */
1189 	oldmm = current->mm;
1190 	if (!oldmm)
1191 		return 0;
1192 
1193 	/* initialize the new vmacache entries */
1194 	vmacache_flush(tsk);
1195 
1196 	if (clone_flags & CLONE_VM) {
1197 		mmget(oldmm);
1198 		mm = oldmm;
1199 		goto good_mm;
1200 	}
1201 
1202 	retval = -ENOMEM;
1203 	mm = dup_mm(tsk);
1204 	if (!mm)
1205 		goto fail_nomem;
1206 
1207 good_mm:
1208 	tsk->mm = mm;
1209 	tsk->active_mm = mm;
1210 	return 0;
1211 
1212 fail_nomem:
1213 	return retval;
1214 }
1215 
1216 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1217 {
1218 	struct fs_struct *fs = current->fs;
1219 	if (clone_flags & CLONE_FS) {
1220 		/* tsk->fs is already what we want */
1221 		spin_lock(&fs->lock);
1222 		if (fs->in_exec) {
1223 			spin_unlock(&fs->lock);
1224 			return -EAGAIN;
1225 		}
1226 		fs->users++;
1227 		spin_unlock(&fs->lock);
1228 		return 0;
1229 	}
1230 	tsk->fs = copy_fs_struct(fs);
1231 	if (!tsk->fs)
1232 		return -ENOMEM;
1233 	return 0;
1234 }
1235 
1236 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1237 {
1238 	struct files_struct *oldf, *newf;
1239 	int error = 0;
1240 
1241 	/*
1242 	 * A background process may not have any files ...
1243 	 */
1244 	oldf = current->files;
1245 	if (!oldf)
1246 		goto out;
1247 
1248 	if (clone_flags & CLONE_FILES) {
1249 		atomic_inc(&oldf->count);
1250 		goto out;
1251 	}
1252 
1253 	newf = dup_fd(oldf, &error);
1254 	if (!newf)
1255 		goto out;
1256 
1257 	tsk->files = newf;
1258 	error = 0;
1259 out:
1260 	return error;
1261 }
1262 
1263 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1264 {
1265 #ifdef CONFIG_BLOCK
1266 	struct io_context *ioc = current->io_context;
1267 	struct io_context *new_ioc;
1268 
1269 	if (!ioc)
1270 		return 0;
1271 	/*
1272 	 * Share io context with parent, if CLONE_IO is set
1273 	 */
1274 	if (clone_flags & CLONE_IO) {
1275 		ioc_task_link(ioc);
1276 		tsk->io_context = ioc;
1277 	} else if (ioprio_valid(ioc->ioprio)) {
1278 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1279 		if (unlikely(!new_ioc))
1280 			return -ENOMEM;
1281 
1282 		new_ioc->ioprio = ioc->ioprio;
1283 		put_io_context(new_ioc);
1284 	}
1285 #endif
1286 	return 0;
1287 }
1288 
1289 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1290 {
1291 	struct sighand_struct *sig;
1292 
1293 	if (clone_flags & CLONE_SIGHAND) {
1294 		atomic_inc(&current->sighand->count);
1295 		return 0;
1296 	}
1297 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1298 	rcu_assign_pointer(tsk->sighand, sig);
1299 	if (!sig)
1300 		return -ENOMEM;
1301 
1302 	atomic_set(&sig->count, 1);
1303 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1304 	return 0;
1305 }
1306 
1307 void __cleanup_sighand(struct sighand_struct *sighand)
1308 {
1309 	if (atomic_dec_and_test(&sighand->count)) {
1310 		signalfd_cleanup(sighand);
1311 		/*
1312 		 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1313 		 * without an RCU grace period, see __lock_task_sighand().
1314 		 */
1315 		kmem_cache_free(sighand_cachep, sighand);
1316 	}
1317 }
1318 
1319 #ifdef CONFIG_POSIX_TIMERS
1320 /*
1321  * Initialize POSIX timer handling for a thread group.
1322  */
1323 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1324 {
1325 	unsigned long cpu_limit;
1326 
1327 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1328 	if (cpu_limit != RLIM_INFINITY) {
1329 		sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1330 		sig->cputimer.running = true;
1331 	}
1332 
1333 	/* The timer lists. */
1334 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1335 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1336 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1337 }
1338 #else
1339 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1340 #endif
1341 
1342 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1343 {
1344 	struct signal_struct *sig;
1345 
1346 	if (clone_flags & CLONE_THREAD)
1347 		return 0;
1348 
1349 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1350 	tsk->signal = sig;
1351 	if (!sig)
1352 		return -ENOMEM;
1353 
1354 	sig->nr_threads = 1;
1355 	atomic_set(&sig->live, 1);
1356 	atomic_set(&sig->sigcnt, 1);
1357 
1358 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1359 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1360 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1361 
1362 	init_waitqueue_head(&sig->wait_chldexit);
1363 	sig->curr_target = tsk;
1364 	init_sigpending(&sig->shared_pending);
1365 	seqlock_init(&sig->stats_lock);
1366 	prev_cputime_init(&sig->prev_cputime);
1367 
1368 #ifdef CONFIG_POSIX_TIMERS
1369 	INIT_LIST_HEAD(&sig->posix_timers);
1370 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1371 	sig->real_timer.function = it_real_fn;
1372 #endif
1373 
1374 	task_lock(current->group_leader);
1375 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1376 	task_unlock(current->group_leader);
1377 
1378 	posix_cpu_timers_init_group(sig);
1379 
1380 	tty_audit_fork(sig);
1381 	sched_autogroup_fork(sig);
1382 
1383 	sig->oom_score_adj = current->signal->oom_score_adj;
1384 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1385 
1386 	mutex_init(&sig->cred_guard_mutex);
1387 
1388 	return 0;
1389 }
1390 
1391 static void copy_seccomp(struct task_struct *p)
1392 {
1393 #ifdef CONFIG_SECCOMP
1394 	/*
1395 	 * Must be called with sighand->lock held, which is common to
1396 	 * all threads in the group. Holding cred_guard_mutex is not
1397 	 * needed because this new task is not yet running and cannot
1398 	 * be racing exec.
1399 	 */
1400 	assert_spin_locked(&current->sighand->siglock);
1401 
1402 	/* Ref-count the new filter user, and assign it. */
1403 	get_seccomp_filter(current);
1404 	p->seccomp = current->seccomp;
1405 
1406 	/*
1407 	 * Explicitly enable no_new_privs here in case it got set
1408 	 * between the task_struct being duplicated and holding the
1409 	 * sighand lock. The seccomp state and nnp must be in sync.
1410 	 */
1411 	if (task_no_new_privs(current))
1412 		task_set_no_new_privs(p);
1413 
1414 	/*
1415 	 * If the parent gained a seccomp mode after copying thread
1416 	 * flags and between before we held the sighand lock, we have
1417 	 * to manually enable the seccomp thread flag here.
1418 	 */
1419 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1420 		set_tsk_thread_flag(p, TIF_SECCOMP);
1421 #endif
1422 }
1423 
1424 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1425 {
1426 	current->clear_child_tid = tidptr;
1427 
1428 	return task_pid_vnr(current);
1429 }
1430 
1431 static void rt_mutex_init_task(struct task_struct *p)
1432 {
1433 	raw_spin_lock_init(&p->pi_lock);
1434 #ifdef CONFIG_RT_MUTEXES
1435 	p->pi_waiters = RB_ROOT;
1436 	p->pi_waiters_leftmost = NULL;
1437 	p->pi_blocked_on = NULL;
1438 #endif
1439 }
1440 
1441 #ifdef CONFIG_POSIX_TIMERS
1442 /*
1443  * Initialize POSIX timer handling for a single task.
1444  */
1445 static void posix_cpu_timers_init(struct task_struct *tsk)
1446 {
1447 	tsk->cputime_expires.prof_exp = 0;
1448 	tsk->cputime_expires.virt_exp = 0;
1449 	tsk->cputime_expires.sched_exp = 0;
1450 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1451 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1452 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1453 }
1454 #else
1455 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1456 #endif
1457 
1458 static inline void
1459 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1460 {
1461 	 task->pids[type].pid = pid;
1462 }
1463 
1464 /*
1465  * This creates a new process as a copy of the old one,
1466  * but does not actually start it yet.
1467  *
1468  * It copies the registers, and all the appropriate
1469  * parts of the process environment (as per the clone
1470  * flags). The actual kick-off is left to the caller.
1471  */
1472 static __latent_entropy struct task_struct *copy_process(
1473 					unsigned long clone_flags,
1474 					unsigned long stack_start,
1475 					unsigned long stack_size,
1476 					int __user *child_tidptr,
1477 					struct pid *pid,
1478 					int trace,
1479 					unsigned long tls,
1480 					int node)
1481 {
1482 	int retval;
1483 	struct task_struct *p;
1484 
1485 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1486 		return ERR_PTR(-EINVAL);
1487 
1488 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1489 		return ERR_PTR(-EINVAL);
1490 
1491 	/*
1492 	 * Thread groups must share signals as well, and detached threads
1493 	 * can only be started up within the thread group.
1494 	 */
1495 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1496 		return ERR_PTR(-EINVAL);
1497 
1498 	/*
1499 	 * Shared signal handlers imply shared VM. By way of the above,
1500 	 * thread groups also imply shared VM. Blocking this case allows
1501 	 * for various simplifications in other code.
1502 	 */
1503 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1504 		return ERR_PTR(-EINVAL);
1505 
1506 	/*
1507 	 * Siblings of global init remain as zombies on exit since they are
1508 	 * not reaped by their parent (swapper). To solve this and to avoid
1509 	 * multi-rooted process trees, prevent global and container-inits
1510 	 * from creating siblings.
1511 	 */
1512 	if ((clone_flags & CLONE_PARENT) &&
1513 				current->signal->flags & SIGNAL_UNKILLABLE)
1514 		return ERR_PTR(-EINVAL);
1515 
1516 	/*
1517 	 * If the new process will be in a different pid or user namespace
1518 	 * do not allow it to share a thread group with the forking task.
1519 	 */
1520 	if (clone_flags & CLONE_THREAD) {
1521 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1522 		    (task_active_pid_ns(current) !=
1523 				current->nsproxy->pid_ns_for_children))
1524 			return ERR_PTR(-EINVAL);
1525 	}
1526 
1527 	retval = security_task_create(clone_flags);
1528 	if (retval)
1529 		goto fork_out;
1530 
1531 	retval = -ENOMEM;
1532 	p = dup_task_struct(current, node);
1533 	if (!p)
1534 		goto fork_out;
1535 
1536 	ftrace_graph_init_task(p);
1537 
1538 	rt_mutex_init_task(p);
1539 
1540 #ifdef CONFIG_PROVE_LOCKING
1541 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1542 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1543 #endif
1544 	retval = -EAGAIN;
1545 	if (atomic_read(&p->real_cred->user->processes) >=
1546 			task_rlimit(p, RLIMIT_NPROC)) {
1547 		if (p->real_cred->user != INIT_USER &&
1548 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1549 			goto bad_fork_free;
1550 	}
1551 	current->flags &= ~PF_NPROC_EXCEEDED;
1552 
1553 	retval = copy_creds(p, clone_flags);
1554 	if (retval < 0)
1555 		goto bad_fork_free;
1556 
1557 	/*
1558 	 * If multiple threads are within copy_process(), then this check
1559 	 * triggers too late. This doesn't hurt, the check is only there
1560 	 * to stop root fork bombs.
1561 	 */
1562 	retval = -EAGAIN;
1563 	if (nr_threads >= max_threads)
1564 		goto bad_fork_cleanup_count;
1565 
1566 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1567 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1568 	p->flags |= PF_FORKNOEXEC;
1569 	INIT_LIST_HEAD(&p->children);
1570 	INIT_LIST_HEAD(&p->sibling);
1571 	rcu_copy_process(p);
1572 	p->vfork_done = NULL;
1573 	spin_lock_init(&p->alloc_lock);
1574 
1575 	init_sigpending(&p->pending);
1576 
1577 	p->utime = p->stime = p->gtime = 0;
1578 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1579 	p->utimescaled = p->stimescaled = 0;
1580 #endif
1581 	prev_cputime_init(&p->prev_cputime);
1582 
1583 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1584 	seqcount_init(&p->vtime_seqcount);
1585 	p->vtime_snap = 0;
1586 	p->vtime_snap_whence = VTIME_INACTIVE;
1587 #endif
1588 
1589 #if defined(SPLIT_RSS_COUNTING)
1590 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1591 #endif
1592 
1593 	p->default_timer_slack_ns = current->timer_slack_ns;
1594 
1595 	task_io_accounting_init(&p->ioac);
1596 	acct_clear_integrals(p);
1597 
1598 	posix_cpu_timers_init(p);
1599 
1600 	p->start_time = ktime_get_ns();
1601 	p->real_start_time = ktime_get_boot_ns();
1602 	p->io_context = NULL;
1603 	p->audit_context = NULL;
1604 	cgroup_fork(p);
1605 #ifdef CONFIG_NUMA
1606 	p->mempolicy = mpol_dup(p->mempolicy);
1607 	if (IS_ERR(p->mempolicy)) {
1608 		retval = PTR_ERR(p->mempolicy);
1609 		p->mempolicy = NULL;
1610 		goto bad_fork_cleanup_threadgroup_lock;
1611 	}
1612 #endif
1613 #ifdef CONFIG_CPUSETS
1614 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1615 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1616 	seqcount_init(&p->mems_allowed_seq);
1617 #endif
1618 #ifdef CONFIG_TRACE_IRQFLAGS
1619 	p->irq_events = 0;
1620 	p->hardirqs_enabled = 0;
1621 	p->hardirq_enable_ip = 0;
1622 	p->hardirq_enable_event = 0;
1623 	p->hardirq_disable_ip = _THIS_IP_;
1624 	p->hardirq_disable_event = 0;
1625 	p->softirqs_enabled = 1;
1626 	p->softirq_enable_ip = _THIS_IP_;
1627 	p->softirq_enable_event = 0;
1628 	p->softirq_disable_ip = 0;
1629 	p->softirq_disable_event = 0;
1630 	p->hardirq_context = 0;
1631 	p->softirq_context = 0;
1632 #endif
1633 
1634 	p->pagefault_disabled = 0;
1635 
1636 #ifdef CONFIG_LOCKDEP
1637 	p->lockdep_depth = 0; /* no locks held yet */
1638 	p->curr_chain_key = 0;
1639 	p->lockdep_recursion = 0;
1640 #endif
1641 
1642 #ifdef CONFIG_DEBUG_MUTEXES
1643 	p->blocked_on = NULL; /* not blocked yet */
1644 #endif
1645 #ifdef CONFIG_BCACHE
1646 	p->sequential_io	= 0;
1647 	p->sequential_io_avg	= 0;
1648 #endif
1649 
1650 	/* Perform scheduler related setup. Assign this task to a CPU. */
1651 	retval = sched_fork(clone_flags, p);
1652 	if (retval)
1653 		goto bad_fork_cleanup_policy;
1654 
1655 	retval = perf_event_init_task(p);
1656 	if (retval)
1657 		goto bad_fork_cleanup_policy;
1658 	retval = audit_alloc(p);
1659 	if (retval)
1660 		goto bad_fork_cleanup_perf;
1661 	/* copy all the process information */
1662 	shm_init_task(p);
1663 	retval = copy_semundo(clone_flags, p);
1664 	if (retval)
1665 		goto bad_fork_cleanup_audit;
1666 	retval = copy_files(clone_flags, p);
1667 	if (retval)
1668 		goto bad_fork_cleanup_semundo;
1669 	retval = copy_fs(clone_flags, p);
1670 	if (retval)
1671 		goto bad_fork_cleanup_files;
1672 	retval = copy_sighand(clone_flags, p);
1673 	if (retval)
1674 		goto bad_fork_cleanup_fs;
1675 	retval = copy_signal(clone_flags, p);
1676 	if (retval)
1677 		goto bad_fork_cleanup_sighand;
1678 	retval = copy_mm(clone_flags, p);
1679 	if (retval)
1680 		goto bad_fork_cleanup_signal;
1681 	retval = copy_namespaces(clone_flags, p);
1682 	if (retval)
1683 		goto bad_fork_cleanup_mm;
1684 	retval = copy_io(clone_flags, p);
1685 	if (retval)
1686 		goto bad_fork_cleanup_namespaces;
1687 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1688 	if (retval)
1689 		goto bad_fork_cleanup_io;
1690 
1691 	if (pid != &init_struct_pid) {
1692 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1693 		if (IS_ERR(pid)) {
1694 			retval = PTR_ERR(pid);
1695 			goto bad_fork_cleanup_thread;
1696 		}
1697 	}
1698 
1699 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1700 	/*
1701 	 * Clear TID on mm_release()?
1702 	 */
1703 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1704 #ifdef CONFIG_BLOCK
1705 	p->plug = NULL;
1706 #endif
1707 #ifdef CONFIG_FUTEX
1708 	p->robust_list = NULL;
1709 #ifdef CONFIG_COMPAT
1710 	p->compat_robust_list = NULL;
1711 #endif
1712 	INIT_LIST_HEAD(&p->pi_state_list);
1713 	p->pi_state_cache = NULL;
1714 #endif
1715 	/*
1716 	 * sigaltstack should be cleared when sharing the same VM
1717 	 */
1718 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1719 		sas_ss_reset(p);
1720 
1721 	/*
1722 	 * Syscall tracing and stepping should be turned off in the
1723 	 * child regardless of CLONE_PTRACE.
1724 	 */
1725 	user_disable_single_step(p);
1726 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1727 #ifdef TIF_SYSCALL_EMU
1728 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1729 #endif
1730 	clear_all_latency_tracing(p);
1731 
1732 	/* ok, now we should be set up.. */
1733 	p->pid = pid_nr(pid);
1734 	if (clone_flags & CLONE_THREAD) {
1735 		p->exit_signal = -1;
1736 		p->group_leader = current->group_leader;
1737 		p->tgid = current->tgid;
1738 	} else {
1739 		if (clone_flags & CLONE_PARENT)
1740 			p->exit_signal = current->group_leader->exit_signal;
1741 		else
1742 			p->exit_signal = (clone_flags & CSIGNAL);
1743 		p->group_leader = p;
1744 		p->tgid = p->pid;
1745 	}
1746 
1747 	p->nr_dirtied = 0;
1748 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1749 	p->dirty_paused_when = 0;
1750 
1751 	p->pdeath_signal = 0;
1752 	INIT_LIST_HEAD(&p->thread_group);
1753 	p->task_works = NULL;
1754 
1755 	cgroup_threadgroup_change_begin(current);
1756 	/*
1757 	 * Ensure that the cgroup subsystem policies allow the new process to be
1758 	 * forked. It should be noted the the new process's css_set can be changed
1759 	 * between here and cgroup_post_fork() if an organisation operation is in
1760 	 * progress.
1761 	 */
1762 	retval = cgroup_can_fork(p);
1763 	if (retval)
1764 		goto bad_fork_free_pid;
1765 
1766 	/*
1767 	 * Make it visible to the rest of the system, but dont wake it up yet.
1768 	 * Need tasklist lock for parent etc handling!
1769 	 */
1770 	write_lock_irq(&tasklist_lock);
1771 
1772 	/* CLONE_PARENT re-uses the old parent */
1773 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1774 		p->real_parent = current->real_parent;
1775 		p->parent_exec_id = current->parent_exec_id;
1776 	} else {
1777 		p->real_parent = current;
1778 		p->parent_exec_id = current->self_exec_id;
1779 	}
1780 
1781 	spin_lock(&current->sighand->siglock);
1782 
1783 	/*
1784 	 * Copy seccomp details explicitly here, in case they were changed
1785 	 * before holding sighand lock.
1786 	 */
1787 	copy_seccomp(p);
1788 
1789 	/*
1790 	 * Process group and session signals need to be delivered to just the
1791 	 * parent before the fork or both the parent and the child after the
1792 	 * fork. Restart if a signal comes in before we add the new process to
1793 	 * it's process group.
1794 	 * A fatal signal pending means that current will exit, so the new
1795 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1796 	*/
1797 	recalc_sigpending();
1798 	if (signal_pending(current)) {
1799 		spin_unlock(&current->sighand->siglock);
1800 		write_unlock_irq(&tasklist_lock);
1801 		retval = -ERESTARTNOINTR;
1802 		goto bad_fork_cancel_cgroup;
1803 	}
1804 
1805 	if (likely(p->pid)) {
1806 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1807 
1808 		init_task_pid(p, PIDTYPE_PID, pid);
1809 		if (thread_group_leader(p)) {
1810 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1811 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1812 
1813 			if (is_child_reaper(pid)) {
1814 				ns_of_pid(pid)->child_reaper = p;
1815 				p->signal->flags |= SIGNAL_UNKILLABLE;
1816 			}
1817 
1818 			p->signal->leader_pid = pid;
1819 			p->signal->tty = tty_kref_get(current->signal->tty);
1820 			/*
1821 			 * Inherit has_child_subreaper flag under the same
1822 			 * tasklist_lock with adding child to the process tree
1823 			 * for propagate_has_child_subreaper optimization.
1824 			 */
1825 			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1826 							 p->real_parent->signal->is_child_subreaper;
1827 			list_add_tail(&p->sibling, &p->real_parent->children);
1828 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1829 			attach_pid(p, PIDTYPE_PGID);
1830 			attach_pid(p, PIDTYPE_SID);
1831 			__this_cpu_inc(process_counts);
1832 		} else {
1833 			current->signal->nr_threads++;
1834 			atomic_inc(&current->signal->live);
1835 			atomic_inc(&current->signal->sigcnt);
1836 			list_add_tail_rcu(&p->thread_group,
1837 					  &p->group_leader->thread_group);
1838 			list_add_tail_rcu(&p->thread_node,
1839 					  &p->signal->thread_head);
1840 		}
1841 		attach_pid(p, PIDTYPE_PID);
1842 		nr_threads++;
1843 	}
1844 
1845 	total_forks++;
1846 	spin_unlock(&current->sighand->siglock);
1847 	syscall_tracepoint_update(p);
1848 	write_unlock_irq(&tasklist_lock);
1849 
1850 	proc_fork_connector(p);
1851 	cgroup_post_fork(p);
1852 	cgroup_threadgroup_change_end(current);
1853 	perf_event_fork(p);
1854 
1855 	trace_task_newtask(p, clone_flags);
1856 	uprobe_copy_process(p, clone_flags);
1857 
1858 	return p;
1859 
1860 bad_fork_cancel_cgroup:
1861 	cgroup_cancel_fork(p);
1862 bad_fork_free_pid:
1863 	cgroup_threadgroup_change_end(current);
1864 	if (pid != &init_struct_pid)
1865 		free_pid(pid);
1866 bad_fork_cleanup_thread:
1867 	exit_thread(p);
1868 bad_fork_cleanup_io:
1869 	if (p->io_context)
1870 		exit_io_context(p);
1871 bad_fork_cleanup_namespaces:
1872 	exit_task_namespaces(p);
1873 bad_fork_cleanup_mm:
1874 	if (p->mm)
1875 		mmput(p->mm);
1876 bad_fork_cleanup_signal:
1877 	if (!(clone_flags & CLONE_THREAD))
1878 		free_signal_struct(p->signal);
1879 bad_fork_cleanup_sighand:
1880 	__cleanup_sighand(p->sighand);
1881 bad_fork_cleanup_fs:
1882 	exit_fs(p); /* blocking */
1883 bad_fork_cleanup_files:
1884 	exit_files(p); /* blocking */
1885 bad_fork_cleanup_semundo:
1886 	exit_sem(p);
1887 bad_fork_cleanup_audit:
1888 	audit_free(p);
1889 bad_fork_cleanup_perf:
1890 	perf_event_free_task(p);
1891 bad_fork_cleanup_policy:
1892 #ifdef CONFIG_NUMA
1893 	mpol_put(p->mempolicy);
1894 bad_fork_cleanup_threadgroup_lock:
1895 #endif
1896 	delayacct_tsk_free(p);
1897 bad_fork_cleanup_count:
1898 	atomic_dec(&p->cred->user->processes);
1899 	exit_creds(p);
1900 bad_fork_free:
1901 	p->state = TASK_DEAD;
1902 	put_task_stack(p);
1903 	free_task(p);
1904 fork_out:
1905 	return ERR_PTR(retval);
1906 }
1907 
1908 static inline void init_idle_pids(struct pid_link *links)
1909 {
1910 	enum pid_type type;
1911 
1912 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1913 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1914 		links[type].pid = &init_struct_pid;
1915 	}
1916 }
1917 
1918 struct task_struct *fork_idle(int cpu)
1919 {
1920 	struct task_struct *task;
1921 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1922 			    cpu_to_node(cpu));
1923 	if (!IS_ERR(task)) {
1924 		init_idle_pids(task->pids);
1925 		init_idle(task, cpu);
1926 	}
1927 
1928 	return task;
1929 }
1930 
1931 /*
1932  *  Ok, this is the main fork-routine.
1933  *
1934  * It copies the process, and if successful kick-starts
1935  * it and waits for it to finish using the VM if required.
1936  */
1937 long _do_fork(unsigned long clone_flags,
1938 	      unsigned long stack_start,
1939 	      unsigned long stack_size,
1940 	      int __user *parent_tidptr,
1941 	      int __user *child_tidptr,
1942 	      unsigned long tls)
1943 {
1944 	struct task_struct *p;
1945 	int trace = 0;
1946 	long nr;
1947 
1948 	/*
1949 	 * Determine whether and which event to report to ptracer.  When
1950 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
1951 	 * requested, no event is reported; otherwise, report if the event
1952 	 * for the type of forking is enabled.
1953 	 */
1954 	if (!(clone_flags & CLONE_UNTRACED)) {
1955 		if (clone_flags & CLONE_VFORK)
1956 			trace = PTRACE_EVENT_VFORK;
1957 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
1958 			trace = PTRACE_EVENT_CLONE;
1959 		else
1960 			trace = PTRACE_EVENT_FORK;
1961 
1962 		if (likely(!ptrace_event_enabled(current, trace)))
1963 			trace = 0;
1964 	}
1965 
1966 	p = copy_process(clone_flags, stack_start, stack_size,
1967 			 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1968 	add_latent_entropy();
1969 	/*
1970 	 * Do this prior waking up the new thread - the thread pointer
1971 	 * might get invalid after that point, if the thread exits quickly.
1972 	 */
1973 	if (!IS_ERR(p)) {
1974 		struct completion vfork;
1975 		struct pid *pid;
1976 
1977 		trace_sched_process_fork(current, p);
1978 
1979 		pid = get_task_pid(p, PIDTYPE_PID);
1980 		nr = pid_vnr(pid);
1981 
1982 		if (clone_flags & CLONE_PARENT_SETTID)
1983 			put_user(nr, parent_tidptr);
1984 
1985 		if (clone_flags & CLONE_VFORK) {
1986 			p->vfork_done = &vfork;
1987 			init_completion(&vfork);
1988 			get_task_struct(p);
1989 		}
1990 
1991 		wake_up_new_task(p);
1992 
1993 		/* forking complete and child started to run, tell ptracer */
1994 		if (unlikely(trace))
1995 			ptrace_event_pid(trace, pid);
1996 
1997 		if (clone_flags & CLONE_VFORK) {
1998 			if (!wait_for_vfork_done(p, &vfork))
1999 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2000 		}
2001 
2002 		put_pid(pid);
2003 	} else {
2004 		nr = PTR_ERR(p);
2005 	}
2006 	return nr;
2007 }
2008 
2009 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2010 /* For compatibility with architectures that call do_fork directly rather than
2011  * using the syscall entry points below. */
2012 long do_fork(unsigned long clone_flags,
2013 	      unsigned long stack_start,
2014 	      unsigned long stack_size,
2015 	      int __user *parent_tidptr,
2016 	      int __user *child_tidptr)
2017 {
2018 	return _do_fork(clone_flags, stack_start, stack_size,
2019 			parent_tidptr, child_tidptr, 0);
2020 }
2021 #endif
2022 
2023 /*
2024  * Create a kernel thread.
2025  */
2026 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2027 {
2028 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2029 		(unsigned long)arg, NULL, NULL, 0);
2030 }
2031 
2032 #ifdef __ARCH_WANT_SYS_FORK
2033 SYSCALL_DEFINE0(fork)
2034 {
2035 #ifdef CONFIG_MMU
2036 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2037 #else
2038 	/* can not support in nommu mode */
2039 	return -EINVAL;
2040 #endif
2041 }
2042 #endif
2043 
2044 #ifdef __ARCH_WANT_SYS_VFORK
2045 SYSCALL_DEFINE0(vfork)
2046 {
2047 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2048 			0, NULL, NULL, 0);
2049 }
2050 #endif
2051 
2052 #ifdef __ARCH_WANT_SYS_CLONE
2053 #ifdef CONFIG_CLONE_BACKWARDS
2054 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2055 		 int __user *, parent_tidptr,
2056 		 unsigned long, tls,
2057 		 int __user *, child_tidptr)
2058 #elif defined(CONFIG_CLONE_BACKWARDS2)
2059 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2060 		 int __user *, parent_tidptr,
2061 		 int __user *, child_tidptr,
2062 		 unsigned long, tls)
2063 #elif defined(CONFIG_CLONE_BACKWARDS3)
2064 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2065 		int, stack_size,
2066 		int __user *, parent_tidptr,
2067 		int __user *, child_tidptr,
2068 		unsigned long, tls)
2069 #else
2070 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2071 		 int __user *, parent_tidptr,
2072 		 int __user *, child_tidptr,
2073 		 unsigned long, tls)
2074 #endif
2075 {
2076 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2077 }
2078 #endif
2079 
2080 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2081 {
2082 	struct task_struct *leader, *parent, *child;
2083 	int res;
2084 
2085 	read_lock(&tasklist_lock);
2086 	leader = top = top->group_leader;
2087 down:
2088 	for_each_thread(leader, parent) {
2089 		list_for_each_entry(child, &parent->children, sibling) {
2090 			res = visitor(child, data);
2091 			if (res) {
2092 				if (res < 0)
2093 					goto out;
2094 				leader = child;
2095 				goto down;
2096 			}
2097 up:
2098 			;
2099 		}
2100 	}
2101 
2102 	if (leader != top) {
2103 		child = leader;
2104 		parent = child->real_parent;
2105 		leader = parent->group_leader;
2106 		goto up;
2107 	}
2108 out:
2109 	read_unlock(&tasklist_lock);
2110 }
2111 
2112 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2113 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2114 #endif
2115 
2116 static void sighand_ctor(void *data)
2117 {
2118 	struct sighand_struct *sighand = data;
2119 
2120 	spin_lock_init(&sighand->siglock);
2121 	init_waitqueue_head(&sighand->signalfd_wqh);
2122 }
2123 
2124 void __init proc_caches_init(void)
2125 {
2126 	sighand_cachep = kmem_cache_create("sighand_cache",
2127 			sizeof(struct sighand_struct), 0,
2128 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2129 			SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2130 	signal_cachep = kmem_cache_create("signal_cache",
2131 			sizeof(struct signal_struct), 0,
2132 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2133 			NULL);
2134 	files_cachep = kmem_cache_create("files_cache",
2135 			sizeof(struct files_struct), 0,
2136 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2137 			NULL);
2138 	fs_cachep = kmem_cache_create("fs_cache",
2139 			sizeof(struct fs_struct), 0,
2140 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2141 			NULL);
2142 	/*
2143 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2144 	 * whole struct cpumask for the OFFSTACK case. We could change
2145 	 * this to *only* allocate as much of it as required by the
2146 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
2147 	 * is at the end of the structure, exactly for that reason.
2148 	 */
2149 	mm_cachep = kmem_cache_create("mm_struct",
2150 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2151 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2152 			NULL);
2153 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2154 	mmap_init();
2155 	nsproxy_cache_init();
2156 }
2157 
2158 /*
2159  * Check constraints on flags passed to the unshare system call.
2160  */
2161 static int check_unshare_flags(unsigned long unshare_flags)
2162 {
2163 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2164 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2165 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2166 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2167 		return -EINVAL;
2168 	/*
2169 	 * Not implemented, but pretend it works if there is nothing
2170 	 * to unshare.  Note that unsharing the address space or the
2171 	 * signal handlers also need to unshare the signal queues (aka
2172 	 * CLONE_THREAD).
2173 	 */
2174 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2175 		if (!thread_group_empty(current))
2176 			return -EINVAL;
2177 	}
2178 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2179 		if (atomic_read(&current->sighand->count) > 1)
2180 			return -EINVAL;
2181 	}
2182 	if (unshare_flags & CLONE_VM) {
2183 		if (!current_is_single_threaded())
2184 			return -EINVAL;
2185 	}
2186 
2187 	return 0;
2188 }
2189 
2190 /*
2191  * Unshare the filesystem structure if it is being shared
2192  */
2193 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2194 {
2195 	struct fs_struct *fs = current->fs;
2196 
2197 	if (!(unshare_flags & CLONE_FS) || !fs)
2198 		return 0;
2199 
2200 	/* don't need lock here; in the worst case we'll do useless copy */
2201 	if (fs->users == 1)
2202 		return 0;
2203 
2204 	*new_fsp = copy_fs_struct(fs);
2205 	if (!*new_fsp)
2206 		return -ENOMEM;
2207 
2208 	return 0;
2209 }
2210 
2211 /*
2212  * Unshare file descriptor table if it is being shared
2213  */
2214 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2215 {
2216 	struct files_struct *fd = current->files;
2217 	int error = 0;
2218 
2219 	if ((unshare_flags & CLONE_FILES) &&
2220 	    (fd && atomic_read(&fd->count) > 1)) {
2221 		*new_fdp = dup_fd(fd, &error);
2222 		if (!*new_fdp)
2223 			return error;
2224 	}
2225 
2226 	return 0;
2227 }
2228 
2229 /*
2230  * unshare allows a process to 'unshare' part of the process
2231  * context which was originally shared using clone.  copy_*
2232  * functions used by do_fork() cannot be used here directly
2233  * because they modify an inactive task_struct that is being
2234  * constructed. Here we are modifying the current, active,
2235  * task_struct.
2236  */
2237 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2238 {
2239 	struct fs_struct *fs, *new_fs = NULL;
2240 	struct files_struct *fd, *new_fd = NULL;
2241 	struct cred *new_cred = NULL;
2242 	struct nsproxy *new_nsproxy = NULL;
2243 	int do_sysvsem = 0;
2244 	int err;
2245 
2246 	/*
2247 	 * If unsharing a user namespace must also unshare the thread group
2248 	 * and unshare the filesystem root and working directories.
2249 	 */
2250 	if (unshare_flags & CLONE_NEWUSER)
2251 		unshare_flags |= CLONE_THREAD | CLONE_FS;
2252 	/*
2253 	 * If unsharing vm, must also unshare signal handlers.
2254 	 */
2255 	if (unshare_flags & CLONE_VM)
2256 		unshare_flags |= CLONE_SIGHAND;
2257 	/*
2258 	 * If unsharing a signal handlers, must also unshare the signal queues.
2259 	 */
2260 	if (unshare_flags & CLONE_SIGHAND)
2261 		unshare_flags |= CLONE_THREAD;
2262 	/*
2263 	 * If unsharing namespace, must also unshare filesystem information.
2264 	 */
2265 	if (unshare_flags & CLONE_NEWNS)
2266 		unshare_flags |= CLONE_FS;
2267 
2268 	err = check_unshare_flags(unshare_flags);
2269 	if (err)
2270 		goto bad_unshare_out;
2271 	/*
2272 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2273 	 * to a new ipc namespace, the semaphore arrays from the old
2274 	 * namespace are unreachable.
2275 	 */
2276 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2277 		do_sysvsem = 1;
2278 	err = unshare_fs(unshare_flags, &new_fs);
2279 	if (err)
2280 		goto bad_unshare_out;
2281 	err = unshare_fd(unshare_flags, &new_fd);
2282 	if (err)
2283 		goto bad_unshare_cleanup_fs;
2284 	err = unshare_userns(unshare_flags, &new_cred);
2285 	if (err)
2286 		goto bad_unshare_cleanup_fd;
2287 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2288 					 new_cred, new_fs);
2289 	if (err)
2290 		goto bad_unshare_cleanup_cred;
2291 
2292 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2293 		if (do_sysvsem) {
2294 			/*
2295 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2296 			 */
2297 			exit_sem(current);
2298 		}
2299 		if (unshare_flags & CLONE_NEWIPC) {
2300 			/* Orphan segments in old ns (see sem above). */
2301 			exit_shm(current);
2302 			shm_init_task(current);
2303 		}
2304 
2305 		if (new_nsproxy)
2306 			switch_task_namespaces(current, new_nsproxy);
2307 
2308 		task_lock(current);
2309 
2310 		if (new_fs) {
2311 			fs = current->fs;
2312 			spin_lock(&fs->lock);
2313 			current->fs = new_fs;
2314 			if (--fs->users)
2315 				new_fs = NULL;
2316 			else
2317 				new_fs = fs;
2318 			spin_unlock(&fs->lock);
2319 		}
2320 
2321 		if (new_fd) {
2322 			fd = current->files;
2323 			current->files = new_fd;
2324 			new_fd = fd;
2325 		}
2326 
2327 		task_unlock(current);
2328 
2329 		if (new_cred) {
2330 			/* Install the new user namespace */
2331 			commit_creds(new_cred);
2332 			new_cred = NULL;
2333 		}
2334 	}
2335 
2336 bad_unshare_cleanup_cred:
2337 	if (new_cred)
2338 		put_cred(new_cred);
2339 bad_unshare_cleanup_fd:
2340 	if (new_fd)
2341 		put_files_struct(new_fd);
2342 
2343 bad_unshare_cleanup_fs:
2344 	if (new_fs)
2345 		free_fs_struct(new_fs);
2346 
2347 bad_unshare_out:
2348 	return err;
2349 }
2350 
2351 /*
2352  *	Helper to unshare the files of the current task.
2353  *	We don't want to expose copy_files internals to
2354  *	the exec layer of the kernel.
2355  */
2356 
2357 int unshare_files(struct files_struct **displaced)
2358 {
2359 	struct task_struct *task = current;
2360 	struct files_struct *copy = NULL;
2361 	int error;
2362 
2363 	error = unshare_fd(CLONE_FILES, &copy);
2364 	if (error || !copy) {
2365 		*displaced = NULL;
2366 		return error;
2367 	}
2368 	*displaced = task->files;
2369 	task_lock(task);
2370 	task->files = copy;
2371 	task_unlock(task);
2372 	return 0;
2373 }
2374 
2375 int sysctl_max_threads(struct ctl_table *table, int write,
2376 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2377 {
2378 	struct ctl_table t;
2379 	int ret;
2380 	int threads = max_threads;
2381 	int min = MIN_THREADS;
2382 	int max = MAX_THREADS;
2383 
2384 	t = *table;
2385 	t.data = &threads;
2386 	t.extra1 = &min;
2387 	t.extra2 = &max;
2388 
2389 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2390 	if (ret || !write)
2391 		return ret;
2392 
2393 	set_max_threads(threads);
2394 
2395 	return 0;
2396 }
2397