xref: /linux/kernel/fork.c (revision 4f58e6dceb0e44ca8f21568ed81e1df24e55964c)
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13 
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/unistd.h>
17 #include <linux/module.h>
18 #include <linux/vmalloc.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/mempolicy.h>
22 #include <linux/sem.h>
23 #include <linux/file.h>
24 #include <linux/fdtable.h>
25 #include <linux/iocontext.h>
26 #include <linux/key.h>
27 #include <linux/binfmts.h>
28 #include <linux/mman.h>
29 #include <linux/mmu_notifier.h>
30 #include <linux/fs.h>
31 #include <linux/mm.h>
32 #include <linux/vmacache.h>
33 #include <linux/nsproxy.h>
34 #include <linux/capability.h>
35 #include <linux/cpu.h>
36 #include <linux/cgroup.h>
37 #include <linux/security.h>
38 #include <linux/hugetlb.h>
39 #include <linux/seccomp.h>
40 #include <linux/swap.h>
41 #include <linux/syscalls.h>
42 #include <linux/jiffies.h>
43 #include <linux/futex.h>
44 #include <linux/compat.h>
45 #include <linux/kthread.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/rcupdate.h>
48 #include <linux/ptrace.h>
49 #include <linux/mount.h>
50 #include <linux/audit.h>
51 #include <linux/memcontrol.h>
52 #include <linux/ftrace.h>
53 #include <linux/proc_fs.h>
54 #include <linux/profile.h>
55 #include <linux/rmap.h>
56 #include <linux/ksm.h>
57 #include <linux/acct.h>
58 #include <linux/tsacct_kern.h>
59 #include <linux/cn_proc.h>
60 #include <linux/freezer.h>
61 #include <linux/delayacct.h>
62 #include <linux/taskstats_kern.h>
63 #include <linux/random.h>
64 #include <linux/tty.h>
65 #include <linux/blkdev.h>
66 #include <linux/fs_struct.h>
67 #include <linux/magic.h>
68 #include <linux/perf_event.h>
69 #include <linux/posix-timers.h>
70 #include <linux/user-return-notifier.h>
71 #include <linux/oom.h>
72 #include <linux/khugepaged.h>
73 #include <linux/signalfd.h>
74 #include <linux/uprobes.h>
75 #include <linux/aio.h>
76 #include <linux/compiler.h>
77 #include <linux/sysctl.h>
78 #include <linux/kcov.h>
79 
80 #include <asm/pgtable.h>
81 #include <asm/pgalloc.h>
82 #include <asm/uaccess.h>
83 #include <asm/mmu_context.h>
84 #include <asm/cacheflush.h>
85 #include <asm/tlbflush.h>
86 
87 #include <trace/events/sched.h>
88 
89 #define CREATE_TRACE_POINTS
90 #include <trace/events/task.h>
91 
92 /*
93  * Minimum number of threads to boot the kernel
94  */
95 #define MIN_THREADS 20
96 
97 /*
98  * Maximum number of threads
99  */
100 #define MAX_THREADS FUTEX_TID_MASK
101 
102 /*
103  * Protected counters by write_lock_irq(&tasklist_lock)
104  */
105 unsigned long total_forks;	/* Handle normal Linux uptimes. */
106 int nr_threads;			/* The idle threads do not count.. */
107 
108 int max_threads;		/* tunable limit on nr_threads */
109 
110 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
111 
112 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
113 
114 #ifdef CONFIG_PROVE_RCU
115 int lockdep_tasklist_lock_is_held(void)
116 {
117 	return lockdep_is_held(&tasklist_lock);
118 }
119 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
120 #endif /* #ifdef CONFIG_PROVE_RCU */
121 
122 int nr_processes(void)
123 {
124 	int cpu;
125 	int total = 0;
126 
127 	for_each_possible_cpu(cpu)
128 		total += per_cpu(process_counts, cpu);
129 
130 	return total;
131 }
132 
133 void __weak arch_release_task_struct(struct task_struct *tsk)
134 {
135 }
136 
137 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
138 static struct kmem_cache *task_struct_cachep;
139 
140 static inline struct task_struct *alloc_task_struct_node(int node)
141 {
142 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
143 }
144 
145 static inline void free_task_struct(struct task_struct *tsk)
146 {
147 	kmem_cache_free(task_struct_cachep, tsk);
148 }
149 #endif
150 
151 void __weak arch_release_thread_stack(unsigned long *stack)
152 {
153 }
154 
155 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
156 
157 /*
158  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
159  * kmemcache based allocator.
160  */
161 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
162 
163 #ifdef CONFIG_VMAP_STACK
164 /*
165  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
166  * flush.  Try to minimize the number of calls by caching stacks.
167  */
168 #define NR_CACHED_STACKS 2
169 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
170 #endif
171 
172 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
173 {
174 #ifdef CONFIG_VMAP_STACK
175 	void *stack;
176 	int i;
177 
178 	local_irq_disable();
179 	for (i = 0; i < NR_CACHED_STACKS; i++) {
180 		struct vm_struct *s = this_cpu_read(cached_stacks[i]);
181 
182 		if (!s)
183 			continue;
184 		this_cpu_write(cached_stacks[i], NULL);
185 
186 		tsk->stack_vm_area = s;
187 		local_irq_enable();
188 		return s->addr;
189 	}
190 	local_irq_enable();
191 
192 	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
193 				     VMALLOC_START, VMALLOC_END,
194 				     THREADINFO_GFP | __GFP_HIGHMEM,
195 				     PAGE_KERNEL,
196 				     0, node, __builtin_return_address(0));
197 
198 	/*
199 	 * We can't call find_vm_area() in interrupt context, and
200 	 * free_thread_stack() can be called in interrupt context,
201 	 * so cache the vm_struct.
202 	 */
203 	if (stack)
204 		tsk->stack_vm_area = find_vm_area(stack);
205 	return stack;
206 #else
207 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
208 					     THREAD_SIZE_ORDER);
209 
210 	return page ? page_address(page) : NULL;
211 #endif
212 }
213 
214 static inline void free_thread_stack(struct task_struct *tsk)
215 {
216 #ifdef CONFIG_VMAP_STACK
217 	if (task_stack_vm_area(tsk)) {
218 		unsigned long flags;
219 		int i;
220 
221 		local_irq_save(flags);
222 		for (i = 0; i < NR_CACHED_STACKS; i++) {
223 			if (this_cpu_read(cached_stacks[i]))
224 				continue;
225 
226 			this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
227 			local_irq_restore(flags);
228 			return;
229 		}
230 		local_irq_restore(flags);
231 
232 		vfree(tsk->stack);
233 		return;
234 	}
235 #endif
236 
237 	__free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
238 }
239 # else
240 static struct kmem_cache *thread_stack_cache;
241 
242 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
243 						  int node)
244 {
245 	return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
246 }
247 
248 static void free_thread_stack(struct task_struct *tsk)
249 {
250 	kmem_cache_free(thread_stack_cache, tsk->stack);
251 }
252 
253 void thread_stack_cache_init(void)
254 {
255 	thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
256 					      THREAD_SIZE, 0, NULL);
257 	BUG_ON(thread_stack_cache == NULL);
258 }
259 # endif
260 #endif
261 
262 /* SLAB cache for signal_struct structures (tsk->signal) */
263 static struct kmem_cache *signal_cachep;
264 
265 /* SLAB cache for sighand_struct structures (tsk->sighand) */
266 struct kmem_cache *sighand_cachep;
267 
268 /* SLAB cache for files_struct structures (tsk->files) */
269 struct kmem_cache *files_cachep;
270 
271 /* SLAB cache for fs_struct structures (tsk->fs) */
272 struct kmem_cache *fs_cachep;
273 
274 /* SLAB cache for vm_area_struct structures */
275 struct kmem_cache *vm_area_cachep;
276 
277 /* SLAB cache for mm_struct structures (tsk->mm) */
278 static struct kmem_cache *mm_cachep;
279 
280 static void account_kernel_stack(struct task_struct *tsk, int account)
281 {
282 	void *stack = task_stack_page(tsk);
283 	struct vm_struct *vm = task_stack_vm_area(tsk);
284 
285 	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
286 
287 	if (vm) {
288 		int i;
289 
290 		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
291 
292 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
293 			mod_zone_page_state(page_zone(vm->pages[i]),
294 					    NR_KERNEL_STACK_KB,
295 					    PAGE_SIZE / 1024 * account);
296 		}
297 
298 		/* All stack pages belong to the same memcg. */
299 		memcg_kmem_update_page_stat(vm->pages[0], MEMCG_KERNEL_STACK_KB,
300 					    account * (THREAD_SIZE / 1024));
301 	} else {
302 		/*
303 		 * All stack pages are in the same zone and belong to the
304 		 * same memcg.
305 		 */
306 		struct page *first_page = virt_to_page(stack);
307 
308 		mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
309 				    THREAD_SIZE / 1024 * account);
310 
311 		memcg_kmem_update_page_stat(first_page, MEMCG_KERNEL_STACK_KB,
312 					    account * (THREAD_SIZE / 1024));
313 	}
314 }
315 
316 static void release_task_stack(struct task_struct *tsk)
317 {
318 	account_kernel_stack(tsk, -1);
319 	arch_release_thread_stack(tsk->stack);
320 	free_thread_stack(tsk);
321 	tsk->stack = NULL;
322 #ifdef CONFIG_VMAP_STACK
323 	tsk->stack_vm_area = NULL;
324 #endif
325 }
326 
327 #ifdef CONFIG_THREAD_INFO_IN_TASK
328 void put_task_stack(struct task_struct *tsk)
329 {
330 	if (atomic_dec_and_test(&tsk->stack_refcount))
331 		release_task_stack(tsk);
332 }
333 #endif
334 
335 void free_task(struct task_struct *tsk)
336 {
337 #ifndef CONFIG_THREAD_INFO_IN_TASK
338 	/*
339 	 * The task is finally done with both the stack and thread_info,
340 	 * so free both.
341 	 */
342 	release_task_stack(tsk);
343 #else
344 	/*
345 	 * If the task had a separate stack allocation, it should be gone
346 	 * by now.
347 	 */
348 	WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
349 #endif
350 	rt_mutex_debug_task_free(tsk);
351 	ftrace_graph_exit_task(tsk);
352 	put_seccomp_filter(tsk);
353 	arch_release_task_struct(tsk);
354 	free_task_struct(tsk);
355 }
356 EXPORT_SYMBOL(free_task);
357 
358 static inline void free_signal_struct(struct signal_struct *sig)
359 {
360 	taskstats_tgid_free(sig);
361 	sched_autogroup_exit(sig);
362 	/*
363 	 * __mmdrop is not safe to call from softirq context on x86 due to
364 	 * pgd_dtor so postpone it to the async context
365 	 */
366 	if (sig->oom_mm)
367 		mmdrop_async(sig->oom_mm);
368 	kmem_cache_free(signal_cachep, sig);
369 }
370 
371 static inline void put_signal_struct(struct signal_struct *sig)
372 {
373 	if (atomic_dec_and_test(&sig->sigcnt))
374 		free_signal_struct(sig);
375 }
376 
377 void __put_task_struct(struct task_struct *tsk)
378 {
379 	WARN_ON(!tsk->exit_state);
380 	WARN_ON(atomic_read(&tsk->usage));
381 	WARN_ON(tsk == current);
382 
383 	cgroup_free(tsk);
384 	task_numa_free(tsk);
385 	security_task_free(tsk);
386 	exit_creds(tsk);
387 	delayacct_tsk_free(tsk);
388 	put_signal_struct(tsk->signal);
389 
390 	if (!profile_handoff_task(tsk))
391 		free_task(tsk);
392 }
393 EXPORT_SYMBOL_GPL(__put_task_struct);
394 
395 void __init __weak arch_task_cache_init(void) { }
396 
397 /*
398  * set_max_threads
399  */
400 static void set_max_threads(unsigned int max_threads_suggested)
401 {
402 	u64 threads;
403 
404 	/*
405 	 * The number of threads shall be limited such that the thread
406 	 * structures may only consume a small part of the available memory.
407 	 */
408 	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
409 		threads = MAX_THREADS;
410 	else
411 		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
412 				    (u64) THREAD_SIZE * 8UL);
413 
414 	if (threads > max_threads_suggested)
415 		threads = max_threads_suggested;
416 
417 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
418 }
419 
420 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
421 /* Initialized by the architecture: */
422 int arch_task_struct_size __read_mostly;
423 #endif
424 
425 void __init fork_init(void)
426 {
427 	int i;
428 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
429 #ifndef ARCH_MIN_TASKALIGN
430 #define ARCH_MIN_TASKALIGN	L1_CACHE_BYTES
431 #endif
432 	/* create a slab on which task_structs can be allocated */
433 	task_struct_cachep = kmem_cache_create("task_struct",
434 			arch_task_struct_size, ARCH_MIN_TASKALIGN,
435 			SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
436 #endif
437 
438 	/* do the arch specific task caches init */
439 	arch_task_cache_init();
440 
441 	set_max_threads(MAX_THREADS);
442 
443 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
444 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
445 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
446 		init_task.signal->rlim[RLIMIT_NPROC];
447 
448 	for (i = 0; i < UCOUNT_COUNTS; i++) {
449 		init_user_ns.ucount_max[i] = max_threads/2;
450 	}
451 }
452 
453 int __weak arch_dup_task_struct(struct task_struct *dst,
454 					       struct task_struct *src)
455 {
456 	*dst = *src;
457 	return 0;
458 }
459 
460 void set_task_stack_end_magic(struct task_struct *tsk)
461 {
462 	unsigned long *stackend;
463 
464 	stackend = end_of_stack(tsk);
465 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
466 }
467 
468 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
469 {
470 	struct task_struct *tsk;
471 	unsigned long *stack;
472 	struct vm_struct *stack_vm_area;
473 	int err;
474 
475 	if (node == NUMA_NO_NODE)
476 		node = tsk_fork_get_node(orig);
477 	tsk = alloc_task_struct_node(node);
478 	if (!tsk)
479 		return NULL;
480 
481 	stack = alloc_thread_stack_node(tsk, node);
482 	if (!stack)
483 		goto free_tsk;
484 
485 	stack_vm_area = task_stack_vm_area(tsk);
486 
487 	err = arch_dup_task_struct(tsk, orig);
488 
489 	/*
490 	 * arch_dup_task_struct() clobbers the stack-related fields.  Make
491 	 * sure they're properly initialized before using any stack-related
492 	 * functions again.
493 	 */
494 	tsk->stack = stack;
495 #ifdef CONFIG_VMAP_STACK
496 	tsk->stack_vm_area = stack_vm_area;
497 #endif
498 #ifdef CONFIG_THREAD_INFO_IN_TASK
499 	atomic_set(&tsk->stack_refcount, 1);
500 #endif
501 
502 	if (err)
503 		goto free_stack;
504 
505 #ifdef CONFIG_SECCOMP
506 	/*
507 	 * We must handle setting up seccomp filters once we're under
508 	 * the sighand lock in case orig has changed between now and
509 	 * then. Until then, filter must be NULL to avoid messing up
510 	 * the usage counts on the error path calling free_task.
511 	 */
512 	tsk->seccomp.filter = NULL;
513 #endif
514 
515 	setup_thread_stack(tsk, orig);
516 	clear_user_return_notifier(tsk);
517 	clear_tsk_need_resched(tsk);
518 	set_task_stack_end_magic(tsk);
519 
520 #ifdef CONFIG_CC_STACKPROTECTOR
521 	tsk->stack_canary = get_random_int();
522 #endif
523 
524 	/*
525 	 * One for us, one for whoever does the "release_task()" (usually
526 	 * parent)
527 	 */
528 	atomic_set(&tsk->usage, 2);
529 #ifdef CONFIG_BLK_DEV_IO_TRACE
530 	tsk->btrace_seq = 0;
531 #endif
532 	tsk->splice_pipe = NULL;
533 	tsk->task_frag.page = NULL;
534 	tsk->wake_q.next = NULL;
535 
536 	account_kernel_stack(tsk, 1);
537 
538 	kcov_task_init(tsk);
539 
540 	return tsk;
541 
542 free_stack:
543 	free_thread_stack(tsk);
544 free_tsk:
545 	free_task_struct(tsk);
546 	return NULL;
547 }
548 
549 #ifdef CONFIG_MMU
550 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
551 {
552 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
553 	struct rb_node **rb_link, *rb_parent;
554 	int retval;
555 	unsigned long charge;
556 
557 	uprobe_start_dup_mmap();
558 	if (down_write_killable(&oldmm->mmap_sem)) {
559 		retval = -EINTR;
560 		goto fail_uprobe_end;
561 	}
562 	flush_cache_dup_mm(oldmm);
563 	uprobe_dup_mmap(oldmm, mm);
564 	/*
565 	 * Not linked in yet - no deadlock potential:
566 	 */
567 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
568 
569 	/* No ordering required: file already has been exposed. */
570 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
571 
572 	mm->total_vm = oldmm->total_vm;
573 	mm->data_vm = oldmm->data_vm;
574 	mm->exec_vm = oldmm->exec_vm;
575 	mm->stack_vm = oldmm->stack_vm;
576 
577 	rb_link = &mm->mm_rb.rb_node;
578 	rb_parent = NULL;
579 	pprev = &mm->mmap;
580 	retval = ksm_fork(mm, oldmm);
581 	if (retval)
582 		goto out;
583 	retval = khugepaged_fork(mm, oldmm);
584 	if (retval)
585 		goto out;
586 
587 	prev = NULL;
588 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
589 		struct file *file;
590 
591 		if (mpnt->vm_flags & VM_DONTCOPY) {
592 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
593 			continue;
594 		}
595 		charge = 0;
596 		if (mpnt->vm_flags & VM_ACCOUNT) {
597 			unsigned long len = vma_pages(mpnt);
598 
599 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
600 				goto fail_nomem;
601 			charge = len;
602 		}
603 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
604 		if (!tmp)
605 			goto fail_nomem;
606 		*tmp = *mpnt;
607 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
608 		retval = vma_dup_policy(mpnt, tmp);
609 		if (retval)
610 			goto fail_nomem_policy;
611 		tmp->vm_mm = mm;
612 		if (anon_vma_fork(tmp, mpnt))
613 			goto fail_nomem_anon_vma_fork;
614 		tmp->vm_flags &=
615 			~(VM_LOCKED|VM_LOCKONFAULT|VM_UFFD_MISSING|VM_UFFD_WP);
616 		tmp->vm_next = tmp->vm_prev = NULL;
617 		tmp->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
618 		file = tmp->vm_file;
619 		if (file) {
620 			struct inode *inode = file_inode(file);
621 			struct address_space *mapping = file->f_mapping;
622 
623 			get_file(file);
624 			if (tmp->vm_flags & VM_DENYWRITE)
625 				atomic_dec(&inode->i_writecount);
626 			i_mmap_lock_write(mapping);
627 			if (tmp->vm_flags & VM_SHARED)
628 				atomic_inc(&mapping->i_mmap_writable);
629 			flush_dcache_mmap_lock(mapping);
630 			/* insert tmp into the share list, just after mpnt */
631 			vma_interval_tree_insert_after(tmp, mpnt,
632 					&mapping->i_mmap);
633 			flush_dcache_mmap_unlock(mapping);
634 			i_mmap_unlock_write(mapping);
635 		}
636 
637 		/*
638 		 * Clear hugetlb-related page reserves for children. This only
639 		 * affects MAP_PRIVATE mappings. Faults generated by the child
640 		 * are not guaranteed to succeed, even if read-only
641 		 */
642 		if (is_vm_hugetlb_page(tmp))
643 			reset_vma_resv_huge_pages(tmp);
644 
645 		/*
646 		 * Link in the new vma and copy the page table entries.
647 		 */
648 		*pprev = tmp;
649 		pprev = &tmp->vm_next;
650 		tmp->vm_prev = prev;
651 		prev = tmp;
652 
653 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
654 		rb_link = &tmp->vm_rb.rb_right;
655 		rb_parent = &tmp->vm_rb;
656 
657 		mm->map_count++;
658 		retval = copy_page_range(mm, oldmm, mpnt);
659 
660 		if (tmp->vm_ops && tmp->vm_ops->open)
661 			tmp->vm_ops->open(tmp);
662 
663 		if (retval)
664 			goto out;
665 	}
666 	/* a new mm has just been created */
667 	arch_dup_mmap(oldmm, mm);
668 	retval = 0;
669 out:
670 	up_write(&mm->mmap_sem);
671 	flush_tlb_mm(oldmm);
672 	up_write(&oldmm->mmap_sem);
673 fail_uprobe_end:
674 	uprobe_end_dup_mmap();
675 	return retval;
676 fail_nomem_anon_vma_fork:
677 	mpol_put(vma_policy(tmp));
678 fail_nomem_policy:
679 	kmem_cache_free(vm_area_cachep, tmp);
680 fail_nomem:
681 	retval = -ENOMEM;
682 	vm_unacct_memory(charge);
683 	goto out;
684 }
685 
686 static inline int mm_alloc_pgd(struct mm_struct *mm)
687 {
688 	mm->pgd = pgd_alloc(mm);
689 	if (unlikely(!mm->pgd))
690 		return -ENOMEM;
691 	return 0;
692 }
693 
694 static inline void mm_free_pgd(struct mm_struct *mm)
695 {
696 	pgd_free(mm, mm->pgd);
697 }
698 #else
699 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
700 {
701 	down_write(&oldmm->mmap_sem);
702 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
703 	up_write(&oldmm->mmap_sem);
704 	return 0;
705 }
706 #define mm_alloc_pgd(mm)	(0)
707 #define mm_free_pgd(mm)
708 #endif /* CONFIG_MMU */
709 
710 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
711 
712 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
713 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
714 
715 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
716 
717 static int __init coredump_filter_setup(char *s)
718 {
719 	default_dump_filter =
720 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
721 		MMF_DUMP_FILTER_MASK;
722 	return 1;
723 }
724 
725 __setup("coredump_filter=", coredump_filter_setup);
726 
727 #include <linux/init_task.h>
728 
729 static void mm_init_aio(struct mm_struct *mm)
730 {
731 #ifdef CONFIG_AIO
732 	spin_lock_init(&mm->ioctx_lock);
733 	mm->ioctx_table = NULL;
734 #endif
735 }
736 
737 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
738 {
739 #ifdef CONFIG_MEMCG
740 	mm->owner = p;
741 #endif
742 }
743 
744 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
745 {
746 	mm->mmap = NULL;
747 	mm->mm_rb = RB_ROOT;
748 	mm->vmacache_seqnum = 0;
749 	atomic_set(&mm->mm_users, 1);
750 	atomic_set(&mm->mm_count, 1);
751 	init_rwsem(&mm->mmap_sem);
752 	INIT_LIST_HEAD(&mm->mmlist);
753 	mm->core_state = NULL;
754 	atomic_long_set(&mm->nr_ptes, 0);
755 	mm_nr_pmds_init(mm);
756 	mm->map_count = 0;
757 	mm->locked_vm = 0;
758 	mm->pinned_vm = 0;
759 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
760 	spin_lock_init(&mm->page_table_lock);
761 	mm_init_cpumask(mm);
762 	mm_init_aio(mm);
763 	mm_init_owner(mm, p);
764 	mmu_notifier_mm_init(mm);
765 	clear_tlb_flush_pending(mm);
766 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
767 	mm->pmd_huge_pte = NULL;
768 #endif
769 
770 	if (current->mm) {
771 		mm->flags = current->mm->flags & MMF_INIT_MASK;
772 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
773 	} else {
774 		mm->flags = default_dump_filter;
775 		mm->def_flags = 0;
776 	}
777 
778 	if (mm_alloc_pgd(mm))
779 		goto fail_nopgd;
780 
781 	if (init_new_context(p, mm))
782 		goto fail_nocontext;
783 
784 	return mm;
785 
786 fail_nocontext:
787 	mm_free_pgd(mm);
788 fail_nopgd:
789 	free_mm(mm);
790 	return NULL;
791 }
792 
793 static void check_mm(struct mm_struct *mm)
794 {
795 	int i;
796 
797 	for (i = 0; i < NR_MM_COUNTERS; i++) {
798 		long x = atomic_long_read(&mm->rss_stat.count[i]);
799 
800 		if (unlikely(x))
801 			printk(KERN_ALERT "BUG: Bad rss-counter state "
802 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
803 	}
804 
805 	if (atomic_long_read(&mm->nr_ptes))
806 		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
807 				atomic_long_read(&mm->nr_ptes));
808 	if (mm_nr_pmds(mm))
809 		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
810 				mm_nr_pmds(mm));
811 
812 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
813 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
814 #endif
815 }
816 
817 /*
818  * Allocate and initialize an mm_struct.
819  */
820 struct mm_struct *mm_alloc(void)
821 {
822 	struct mm_struct *mm;
823 
824 	mm = allocate_mm();
825 	if (!mm)
826 		return NULL;
827 
828 	memset(mm, 0, sizeof(*mm));
829 	return mm_init(mm, current);
830 }
831 
832 /*
833  * Called when the last reference to the mm
834  * is dropped: either by a lazy thread or by
835  * mmput. Free the page directory and the mm.
836  */
837 void __mmdrop(struct mm_struct *mm)
838 {
839 	BUG_ON(mm == &init_mm);
840 	mm_free_pgd(mm);
841 	destroy_context(mm);
842 	mmu_notifier_mm_destroy(mm);
843 	check_mm(mm);
844 	free_mm(mm);
845 }
846 EXPORT_SYMBOL_GPL(__mmdrop);
847 
848 static inline void __mmput(struct mm_struct *mm)
849 {
850 	VM_BUG_ON(atomic_read(&mm->mm_users));
851 
852 	uprobe_clear_state(mm);
853 	exit_aio(mm);
854 	ksm_exit(mm);
855 	khugepaged_exit(mm); /* must run before exit_mmap */
856 	exit_mmap(mm);
857 	mm_put_huge_zero_page(mm);
858 	set_mm_exe_file(mm, NULL);
859 	if (!list_empty(&mm->mmlist)) {
860 		spin_lock(&mmlist_lock);
861 		list_del(&mm->mmlist);
862 		spin_unlock(&mmlist_lock);
863 	}
864 	if (mm->binfmt)
865 		module_put(mm->binfmt->module);
866 	set_bit(MMF_OOM_SKIP, &mm->flags);
867 	mmdrop(mm);
868 }
869 
870 /*
871  * Decrement the use count and release all resources for an mm.
872  */
873 void mmput(struct mm_struct *mm)
874 {
875 	might_sleep();
876 
877 	if (atomic_dec_and_test(&mm->mm_users))
878 		__mmput(mm);
879 }
880 EXPORT_SYMBOL_GPL(mmput);
881 
882 #ifdef CONFIG_MMU
883 static void mmput_async_fn(struct work_struct *work)
884 {
885 	struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
886 	__mmput(mm);
887 }
888 
889 void mmput_async(struct mm_struct *mm)
890 {
891 	if (atomic_dec_and_test(&mm->mm_users)) {
892 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
893 		schedule_work(&mm->async_put_work);
894 	}
895 }
896 #endif
897 
898 /**
899  * set_mm_exe_file - change a reference to the mm's executable file
900  *
901  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
902  *
903  * Main users are mmput() and sys_execve(). Callers prevent concurrent
904  * invocations: in mmput() nobody alive left, in execve task is single
905  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
906  * mm->exe_file, but does so without using set_mm_exe_file() in order
907  * to do avoid the need for any locks.
908  */
909 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
910 {
911 	struct file *old_exe_file;
912 
913 	/*
914 	 * It is safe to dereference the exe_file without RCU as
915 	 * this function is only called if nobody else can access
916 	 * this mm -- see comment above for justification.
917 	 */
918 	old_exe_file = rcu_dereference_raw(mm->exe_file);
919 
920 	if (new_exe_file)
921 		get_file(new_exe_file);
922 	rcu_assign_pointer(mm->exe_file, new_exe_file);
923 	if (old_exe_file)
924 		fput(old_exe_file);
925 }
926 
927 /**
928  * get_mm_exe_file - acquire a reference to the mm's executable file
929  *
930  * Returns %NULL if mm has no associated executable file.
931  * User must release file via fput().
932  */
933 struct file *get_mm_exe_file(struct mm_struct *mm)
934 {
935 	struct file *exe_file;
936 
937 	rcu_read_lock();
938 	exe_file = rcu_dereference(mm->exe_file);
939 	if (exe_file && !get_file_rcu(exe_file))
940 		exe_file = NULL;
941 	rcu_read_unlock();
942 	return exe_file;
943 }
944 EXPORT_SYMBOL(get_mm_exe_file);
945 
946 /**
947  * get_task_exe_file - acquire a reference to the task's executable file
948  *
949  * Returns %NULL if task's mm (if any) has no associated executable file or
950  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
951  * User must release file via fput().
952  */
953 struct file *get_task_exe_file(struct task_struct *task)
954 {
955 	struct file *exe_file = NULL;
956 	struct mm_struct *mm;
957 
958 	task_lock(task);
959 	mm = task->mm;
960 	if (mm) {
961 		if (!(task->flags & PF_KTHREAD))
962 			exe_file = get_mm_exe_file(mm);
963 	}
964 	task_unlock(task);
965 	return exe_file;
966 }
967 EXPORT_SYMBOL(get_task_exe_file);
968 
969 /**
970  * get_task_mm - acquire a reference to the task's mm
971  *
972  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
973  * this kernel workthread has transiently adopted a user mm with use_mm,
974  * to do its AIO) is not set and if so returns a reference to it, after
975  * bumping up the use count.  User must release the mm via mmput()
976  * after use.  Typically used by /proc and ptrace.
977  */
978 struct mm_struct *get_task_mm(struct task_struct *task)
979 {
980 	struct mm_struct *mm;
981 
982 	task_lock(task);
983 	mm = task->mm;
984 	if (mm) {
985 		if (task->flags & PF_KTHREAD)
986 			mm = NULL;
987 		else
988 			atomic_inc(&mm->mm_users);
989 	}
990 	task_unlock(task);
991 	return mm;
992 }
993 EXPORT_SYMBOL_GPL(get_task_mm);
994 
995 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
996 {
997 	struct mm_struct *mm;
998 	int err;
999 
1000 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1001 	if (err)
1002 		return ERR_PTR(err);
1003 
1004 	mm = get_task_mm(task);
1005 	if (mm && mm != current->mm &&
1006 			!ptrace_may_access(task, mode)) {
1007 		mmput(mm);
1008 		mm = ERR_PTR(-EACCES);
1009 	}
1010 	mutex_unlock(&task->signal->cred_guard_mutex);
1011 
1012 	return mm;
1013 }
1014 
1015 static void complete_vfork_done(struct task_struct *tsk)
1016 {
1017 	struct completion *vfork;
1018 
1019 	task_lock(tsk);
1020 	vfork = tsk->vfork_done;
1021 	if (likely(vfork)) {
1022 		tsk->vfork_done = NULL;
1023 		complete(vfork);
1024 	}
1025 	task_unlock(tsk);
1026 }
1027 
1028 static int wait_for_vfork_done(struct task_struct *child,
1029 				struct completion *vfork)
1030 {
1031 	int killed;
1032 
1033 	freezer_do_not_count();
1034 	killed = wait_for_completion_killable(vfork);
1035 	freezer_count();
1036 
1037 	if (killed) {
1038 		task_lock(child);
1039 		child->vfork_done = NULL;
1040 		task_unlock(child);
1041 	}
1042 
1043 	put_task_struct(child);
1044 	return killed;
1045 }
1046 
1047 /* Please note the differences between mmput and mm_release.
1048  * mmput is called whenever we stop holding onto a mm_struct,
1049  * error success whatever.
1050  *
1051  * mm_release is called after a mm_struct has been removed
1052  * from the current process.
1053  *
1054  * This difference is important for error handling, when we
1055  * only half set up a mm_struct for a new process and need to restore
1056  * the old one.  Because we mmput the new mm_struct before
1057  * restoring the old one. . .
1058  * Eric Biederman 10 January 1998
1059  */
1060 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1061 {
1062 	/* Get rid of any futexes when releasing the mm */
1063 #ifdef CONFIG_FUTEX
1064 	if (unlikely(tsk->robust_list)) {
1065 		exit_robust_list(tsk);
1066 		tsk->robust_list = NULL;
1067 	}
1068 #ifdef CONFIG_COMPAT
1069 	if (unlikely(tsk->compat_robust_list)) {
1070 		compat_exit_robust_list(tsk);
1071 		tsk->compat_robust_list = NULL;
1072 	}
1073 #endif
1074 	if (unlikely(!list_empty(&tsk->pi_state_list)))
1075 		exit_pi_state_list(tsk);
1076 #endif
1077 
1078 	uprobe_free_utask(tsk);
1079 
1080 	/* Get rid of any cached register state */
1081 	deactivate_mm(tsk, mm);
1082 
1083 	/*
1084 	 * Signal userspace if we're not exiting with a core dump
1085 	 * because we want to leave the value intact for debugging
1086 	 * purposes.
1087 	 */
1088 	if (tsk->clear_child_tid) {
1089 		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1090 		    atomic_read(&mm->mm_users) > 1) {
1091 			/*
1092 			 * We don't check the error code - if userspace has
1093 			 * not set up a proper pointer then tough luck.
1094 			 */
1095 			put_user(0, tsk->clear_child_tid);
1096 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1097 					1, NULL, NULL, 0);
1098 		}
1099 		tsk->clear_child_tid = NULL;
1100 	}
1101 
1102 	/*
1103 	 * All done, finally we can wake up parent and return this mm to him.
1104 	 * Also kthread_stop() uses this completion for synchronization.
1105 	 */
1106 	if (tsk->vfork_done)
1107 		complete_vfork_done(tsk);
1108 }
1109 
1110 /*
1111  * Allocate a new mm structure and copy contents from the
1112  * mm structure of the passed in task structure.
1113  */
1114 static struct mm_struct *dup_mm(struct task_struct *tsk)
1115 {
1116 	struct mm_struct *mm, *oldmm = current->mm;
1117 	int err;
1118 
1119 	mm = allocate_mm();
1120 	if (!mm)
1121 		goto fail_nomem;
1122 
1123 	memcpy(mm, oldmm, sizeof(*mm));
1124 
1125 	if (!mm_init(mm, tsk))
1126 		goto fail_nomem;
1127 
1128 	err = dup_mmap(mm, oldmm);
1129 	if (err)
1130 		goto free_pt;
1131 
1132 	mm->hiwater_rss = get_mm_rss(mm);
1133 	mm->hiwater_vm = mm->total_vm;
1134 
1135 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1136 		goto free_pt;
1137 
1138 	return mm;
1139 
1140 free_pt:
1141 	/* don't put binfmt in mmput, we haven't got module yet */
1142 	mm->binfmt = NULL;
1143 	mmput(mm);
1144 
1145 fail_nomem:
1146 	return NULL;
1147 }
1148 
1149 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1150 {
1151 	struct mm_struct *mm, *oldmm;
1152 	int retval;
1153 
1154 	tsk->min_flt = tsk->maj_flt = 0;
1155 	tsk->nvcsw = tsk->nivcsw = 0;
1156 #ifdef CONFIG_DETECT_HUNG_TASK
1157 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1158 #endif
1159 
1160 	tsk->mm = NULL;
1161 	tsk->active_mm = NULL;
1162 
1163 	/*
1164 	 * Are we cloning a kernel thread?
1165 	 *
1166 	 * We need to steal a active VM for that..
1167 	 */
1168 	oldmm = current->mm;
1169 	if (!oldmm)
1170 		return 0;
1171 
1172 	/* initialize the new vmacache entries */
1173 	vmacache_flush(tsk);
1174 
1175 	if (clone_flags & CLONE_VM) {
1176 		atomic_inc(&oldmm->mm_users);
1177 		mm = oldmm;
1178 		goto good_mm;
1179 	}
1180 
1181 	retval = -ENOMEM;
1182 	mm = dup_mm(tsk);
1183 	if (!mm)
1184 		goto fail_nomem;
1185 
1186 good_mm:
1187 	tsk->mm = mm;
1188 	tsk->active_mm = mm;
1189 	return 0;
1190 
1191 fail_nomem:
1192 	return retval;
1193 }
1194 
1195 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1196 {
1197 	struct fs_struct *fs = current->fs;
1198 	if (clone_flags & CLONE_FS) {
1199 		/* tsk->fs is already what we want */
1200 		spin_lock(&fs->lock);
1201 		if (fs->in_exec) {
1202 			spin_unlock(&fs->lock);
1203 			return -EAGAIN;
1204 		}
1205 		fs->users++;
1206 		spin_unlock(&fs->lock);
1207 		return 0;
1208 	}
1209 	tsk->fs = copy_fs_struct(fs);
1210 	if (!tsk->fs)
1211 		return -ENOMEM;
1212 	return 0;
1213 }
1214 
1215 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1216 {
1217 	struct files_struct *oldf, *newf;
1218 	int error = 0;
1219 
1220 	/*
1221 	 * A background process may not have any files ...
1222 	 */
1223 	oldf = current->files;
1224 	if (!oldf)
1225 		goto out;
1226 
1227 	if (clone_flags & CLONE_FILES) {
1228 		atomic_inc(&oldf->count);
1229 		goto out;
1230 	}
1231 
1232 	newf = dup_fd(oldf, &error);
1233 	if (!newf)
1234 		goto out;
1235 
1236 	tsk->files = newf;
1237 	error = 0;
1238 out:
1239 	return error;
1240 }
1241 
1242 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1243 {
1244 #ifdef CONFIG_BLOCK
1245 	struct io_context *ioc = current->io_context;
1246 	struct io_context *new_ioc;
1247 
1248 	if (!ioc)
1249 		return 0;
1250 	/*
1251 	 * Share io context with parent, if CLONE_IO is set
1252 	 */
1253 	if (clone_flags & CLONE_IO) {
1254 		ioc_task_link(ioc);
1255 		tsk->io_context = ioc;
1256 	} else if (ioprio_valid(ioc->ioprio)) {
1257 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1258 		if (unlikely(!new_ioc))
1259 			return -ENOMEM;
1260 
1261 		new_ioc->ioprio = ioc->ioprio;
1262 		put_io_context(new_ioc);
1263 	}
1264 #endif
1265 	return 0;
1266 }
1267 
1268 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1269 {
1270 	struct sighand_struct *sig;
1271 
1272 	if (clone_flags & CLONE_SIGHAND) {
1273 		atomic_inc(&current->sighand->count);
1274 		return 0;
1275 	}
1276 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1277 	rcu_assign_pointer(tsk->sighand, sig);
1278 	if (!sig)
1279 		return -ENOMEM;
1280 
1281 	atomic_set(&sig->count, 1);
1282 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1283 	return 0;
1284 }
1285 
1286 void __cleanup_sighand(struct sighand_struct *sighand)
1287 {
1288 	if (atomic_dec_and_test(&sighand->count)) {
1289 		signalfd_cleanup(sighand);
1290 		/*
1291 		 * sighand_cachep is SLAB_DESTROY_BY_RCU so we can free it
1292 		 * without an RCU grace period, see __lock_task_sighand().
1293 		 */
1294 		kmem_cache_free(sighand_cachep, sighand);
1295 	}
1296 }
1297 
1298 /*
1299  * Initialize POSIX timer handling for a thread group.
1300  */
1301 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1302 {
1303 	unsigned long cpu_limit;
1304 
1305 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1306 	if (cpu_limit != RLIM_INFINITY) {
1307 		sig->cputime_expires.prof_exp = secs_to_cputime(cpu_limit);
1308 		sig->cputimer.running = true;
1309 	}
1310 
1311 	/* The timer lists. */
1312 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1313 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1314 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1315 }
1316 
1317 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1318 {
1319 	struct signal_struct *sig;
1320 
1321 	if (clone_flags & CLONE_THREAD)
1322 		return 0;
1323 
1324 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1325 	tsk->signal = sig;
1326 	if (!sig)
1327 		return -ENOMEM;
1328 
1329 	sig->nr_threads = 1;
1330 	atomic_set(&sig->live, 1);
1331 	atomic_set(&sig->sigcnt, 1);
1332 
1333 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1334 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1335 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1336 
1337 	init_waitqueue_head(&sig->wait_chldexit);
1338 	sig->curr_target = tsk;
1339 	init_sigpending(&sig->shared_pending);
1340 	INIT_LIST_HEAD(&sig->posix_timers);
1341 	seqlock_init(&sig->stats_lock);
1342 	prev_cputime_init(&sig->prev_cputime);
1343 
1344 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1345 	sig->real_timer.function = it_real_fn;
1346 
1347 	task_lock(current->group_leader);
1348 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1349 	task_unlock(current->group_leader);
1350 
1351 	posix_cpu_timers_init_group(sig);
1352 
1353 	tty_audit_fork(sig);
1354 	sched_autogroup_fork(sig);
1355 
1356 	sig->oom_score_adj = current->signal->oom_score_adj;
1357 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1358 
1359 	sig->has_child_subreaper = current->signal->has_child_subreaper ||
1360 				   current->signal->is_child_subreaper;
1361 
1362 	mutex_init(&sig->cred_guard_mutex);
1363 
1364 	return 0;
1365 }
1366 
1367 static void copy_seccomp(struct task_struct *p)
1368 {
1369 #ifdef CONFIG_SECCOMP
1370 	/*
1371 	 * Must be called with sighand->lock held, which is common to
1372 	 * all threads in the group. Holding cred_guard_mutex is not
1373 	 * needed because this new task is not yet running and cannot
1374 	 * be racing exec.
1375 	 */
1376 	assert_spin_locked(&current->sighand->siglock);
1377 
1378 	/* Ref-count the new filter user, and assign it. */
1379 	get_seccomp_filter(current);
1380 	p->seccomp = current->seccomp;
1381 
1382 	/*
1383 	 * Explicitly enable no_new_privs here in case it got set
1384 	 * between the task_struct being duplicated and holding the
1385 	 * sighand lock. The seccomp state and nnp must be in sync.
1386 	 */
1387 	if (task_no_new_privs(current))
1388 		task_set_no_new_privs(p);
1389 
1390 	/*
1391 	 * If the parent gained a seccomp mode after copying thread
1392 	 * flags and between before we held the sighand lock, we have
1393 	 * to manually enable the seccomp thread flag here.
1394 	 */
1395 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1396 		set_tsk_thread_flag(p, TIF_SECCOMP);
1397 #endif
1398 }
1399 
1400 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1401 {
1402 	current->clear_child_tid = tidptr;
1403 
1404 	return task_pid_vnr(current);
1405 }
1406 
1407 static void rt_mutex_init_task(struct task_struct *p)
1408 {
1409 	raw_spin_lock_init(&p->pi_lock);
1410 #ifdef CONFIG_RT_MUTEXES
1411 	p->pi_waiters = RB_ROOT;
1412 	p->pi_waiters_leftmost = NULL;
1413 	p->pi_blocked_on = NULL;
1414 #endif
1415 }
1416 
1417 /*
1418  * Initialize POSIX timer handling for a single task.
1419  */
1420 static void posix_cpu_timers_init(struct task_struct *tsk)
1421 {
1422 	tsk->cputime_expires.prof_exp = 0;
1423 	tsk->cputime_expires.virt_exp = 0;
1424 	tsk->cputime_expires.sched_exp = 0;
1425 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1426 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1427 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1428 }
1429 
1430 static inline void
1431 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1432 {
1433 	 task->pids[type].pid = pid;
1434 }
1435 
1436 /*
1437  * This creates a new process as a copy of the old one,
1438  * but does not actually start it yet.
1439  *
1440  * It copies the registers, and all the appropriate
1441  * parts of the process environment (as per the clone
1442  * flags). The actual kick-off is left to the caller.
1443  */
1444 static struct task_struct *copy_process(unsigned long clone_flags,
1445 					unsigned long stack_start,
1446 					unsigned long stack_size,
1447 					int __user *child_tidptr,
1448 					struct pid *pid,
1449 					int trace,
1450 					unsigned long tls,
1451 					int node)
1452 {
1453 	int retval;
1454 	struct task_struct *p;
1455 
1456 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1457 		return ERR_PTR(-EINVAL);
1458 
1459 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1460 		return ERR_PTR(-EINVAL);
1461 
1462 	/*
1463 	 * Thread groups must share signals as well, and detached threads
1464 	 * can only be started up within the thread group.
1465 	 */
1466 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1467 		return ERR_PTR(-EINVAL);
1468 
1469 	/*
1470 	 * Shared signal handlers imply shared VM. By way of the above,
1471 	 * thread groups also imply shared VM. Blocking this case allows
1472 	 * for various simplifications in other code.
1473 	 */
1474 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1475 		return ERR_PTR(-EINVAL);
1476 
1477 	/*
1478 	 * Siblings of global init remain as zombies on exit since they are
1479 	 * not reaped by their parent (swapper). To solve this and to avoid
1480 	 * multi-rooted process trees, prevent global and container-inits
1481 	 * from creating siblings.
1482 	 */
1483 	if ((clone_flags & CLONE_PARENT) &&
1484 				current->signal->flags & SIGNAL_UNKILLABLE)
1485 		return ERR_PTR(-EINVAL);
1486 
1487 	/*
1488 	 * If the new process will be in a different pid or user namespace
1489 	 * do not allow it to share a thread group with the forking task.
1490 	 */
1491 	if (clone_flags & CLONE_THREAD) {
1492 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1493 		    (task_active_pid_ns(current) !=
1494 				current->nsproxy->pid_ns_for_children))
1495 			return ERR_PTR(-EINVAL);
1496 	}
1497 
1498 	retval = security_task_create(clone_flags);
1499 	if (retval)
1500 		goto fork_out;
1501 
1502 	retval = -ENOMEM;
1503 	p = dup_task_struct(current, node);
1504 	if (!p)
1505 		goto fork_out;
1506 
1507 	ftrace_graph_init_task(p);
1508 
1509 	rt_mutex_init_task(p);
1510 
1511 #ifdef CONFIG_PROVE_LOCKING
1512 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1513 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1514 #endif
1515 	retval = -EAGAIN;
1516 	if (atomic_read(&p->real_cred->user->processes) >=
1517 			task_rlimit(p, RLIMIT_NPROC)) {
1518 		if (p->real_cred->user != INIT_USER &&
1519 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1520 			goto bad_fork_free;
1521 	}
1522 	current->flags &= ~PF_NPROC_EXCEEDED;
1523 
1524 	retval = copy_creds(p, clone_flags);
1525 	if (retval < 0)
1526 		goto bad_fork_free;
1527 
1528 	/*
1529 	 * If multiple threads are within copy_process(), then this check
1530 	 * triggers too late. This doesn't hurt, the check is only there
1531 	 * to stop root fork bombs.
1532 	 */
1533 	retval = -EAGAIN;
1534 	if (nr_threads >= max_threads)
1535 		goto bad_fork_cleanup_count;
1536 
1537 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1538 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER);
1539 	p->flags |= PF_FORKNOEXEC;
1540 	INIT_LIST_HEAD(&p->children);
1541 	INIT_LIST_HEAD(&p->sibling);
1542 	rcu_copy_process(p);
1543 	p->vfork_done = NULL;
1544 	spin_lock_init(&p->alloc_lock);
1545 
1546 	init_sigpending(&p->pending);
1547 
1548 	p->utime = p->stime = p->gtime = 0;
1549 	p->utimescaled = p->stimescaled = 0;
1550 	prev_cputime_init(&p->prev_cputime);
1551 
1552 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1553 	seqcount_init(&p->vtime_seqcount);
1554 	p->vtime_snap = 0;
1555 	p->vtime_snap_whence = VTIME_INACTIVE;
1556 #endif
1557 
1558 #if defined(SPLIT_RSS_COUNTING)
1559 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1560 #endif
1561 
1562 	p->default_timer_slack_ns = current->timer_slack_ns;
1563 
1564 	task_io_accounting_init(&p->ioac);
1565 	acct_clear_integrals(p);
1566 
1567 	posix_cpu_timers_init(p);
1568 
1569 	p->start_time = ktime_get_ns();
1570 	p->real_start_time = ktime_get_boot_ns();
1571 	p->io_context = NULL;
1572 	p->audit_context = NULL;
1573 	cgroup_fork(p);
1574 #ifdef CONFIG_NUMA
1575 	p->mempolicy = mpol_dup(p->mempolicy);
1576 	if (IS_ERR(p->mempolicy)) {
1577 		retval = PTR_ERR(p->mempolicy);
1578 		p->mempolicy = NULL;
1579 		goto bad_fork_cleanup_threadgroup_lock;
1580 	}
1581 #endif
1582 #ifdef CONFIG_CPUSETS
1583 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1584 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1585 	seqcount_init(&p->mems_allowed_seq);
1586 #endif
1587 #ifdef CONFIG_TRACE_IRQFLAGS
1588 	p->irq_events = 0;
1589 	p->hardirqs_enabled = 0;
1590 	p->hardirq_enable_ip = 0;
1591 	p->hardirq_enable_event = 0;
1592 	p->hardirq_disable_ip = _THIS_IP_;
1593 	p->hardirq_disable_event = 0;
1594 	p->softirqs_enabled = 1;
1595 	p->softirq_enable_ip = _THIS_IP_;
1596 	p->softirq_enable_event = 0;
1597 	p->softirq_disable_ip = 0;
1598 	p->softirq_disable_event = 0;
1599 	p->hardirq_context = 0;
1600 	p->softirq_context = 0;
1601 #endif
1602 
1603 	p->pagefault_disabled = 0;
1604 
1605 #ifdef CONFIG_LOCKDEP
1606 	p->lockdep_depth = 0; /* no locks held yet */
1607 	p->curr_chain_key = 0;
1608 	p->lockdep_recursion = 0;
1609 #endif
1610 
1611 #ifdef CONFIG_DEBUG_MUTEXES
1612 	p->blocked_on = NULL; /* not blocked yet */
1613 #endif
1614 #ifdef CONFIG_BCACHE
1615 	p->sequential_io	= 0;
1616 	p->sequential_io_avg	= 0;
1617 #endif
1618 
1619 	/* Perform scheduler related setup. Assign this task to a CPU. */
1620 	retval = sched_fork(clone_flags, p);
1621 	if (retval)
1622 		goto bad_fork_cleanup_policy;
1623 
1624 	retval = perf_event_init_task(p);
1625 	if (retval)
1626 		goto bad_fork_cleanup_policy;
1627 	retval = audit_alloc(p);
1628 	if (retval)
1629 		goto bad_fork_cleanup_perf;
1630 	/* copy all the process information */
1631 	shm_init_task(p);
1632 	retval = copy_semundo(clone_flags, p);
1633 	if (retval)
1634 		goto bad_fork_cleanup_audit;
1635 	retval = copy_files(clone_flags, p);
1636 	if (retval)
1637 		goto bad_fork_cleanup_semundo;
1638 	retval = copy_fs(clone_flags, p);
1639 	if (retval)
1640 		goto bad_fork_cleanup_files;
1641 	retval = copy_sighand(clone_flags, p);
1642 	if (retval)
1643 		goto bad_fork_cleanup_fs;
1644 	retval = copy_signal(clone_flags, p);
1645 	if (retval)
1646 		goto bad_fork_cleanup_sighand;
1647 	retval = copy_mm(clone_flags, p);
1648 	if (retval)
1649 		goto bad_fork_cleanup_signal;
1650 	retval = copy_namespaces(clone_flags, p);
1651 	if (retval)
1652 		goto bad_fork_cleanup_mm;
1653 	retval = copy_io(clone_flags, p);
1654 	if (retval)
1655 		goto bad_fork_cleanup_namespaces;
1656 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1657 	if (retval)
1658 		goto bad_fork_cleanup_io;
1659 
1660 	if (pid != &init_struct_pid) {
1661 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1662 		if (IS_ERR(pid)) {
1663 			retval = PTR_ERR(pid);
1664 			goto bad_fork_cleanup_thread;
1665 		}
1666 	}
1667 
1668 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1669 	/*
1670 	 * Clear TID on mm_release()?
1671 	 */
1672 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1673 #ifdef CONFIG_BLOCK
1674 	p->plug = NULL;
1675 #endif
1676 #ifdef CONFIG_FUTEX
1677 	p->robust_list = NULL;
1678 #ifdef CONFIG_COMPAT
1679 	p->compat_robust_list = NULL;
1680 #endif
1681 	INIT_LIST_HEAD(&p->pi_state_list);
1682 	p->pi_state_cache = NULL;
1683 #endif
1684 	/*
1685 	 * sigaltstack should be cleared when sharing the same VM
1686 	 */
1687 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1688 		sas_ss_reset(p);
1689 
1690 	/*
1691 	 * Syscall tracing and stepping should be turned off in the
1692 	 * child regardless of CLONE_PTRACE.
1693 	 */
1694 	user_disable_single_step(p);
1695 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1696 #ifdef TIF_SYSCALL_EMU
1697 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1698 #endif
1699 	clear_all_latency_tracing(p);
1700 
1701 	/* ok, now we should be set up.. */
1702 	p->pid = pid_nr(pid);
1703 	if (clone_flags & CLONE_THREAD) {
1704 		p->exit_signal = -1;
1705 		p->group_leader = current->group_leader;
1706 		p->tgid = current->tgid;
1707 	} else {
1708 		if (clone_flags & CLONE_PARENT)
1709 			p->exit_signal = current->group_leader->exit_signal;
1710 		else
1711 			p->exit_signal = (clone_flags & CSIGNAL);
1712 		p->group_leader = p;
1713 		p->tgid = p->pid;
1714 	}
1715 
1716 	p->nr_dirtied = 0;
1717 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1718 	p->dirty_paused_when = 0;
1719 
1720 	p->pdeath_signal = 0;
1721 	INIT_LIST_HEAD(&p->thread_group);
1722 	p->task_works = NULL;
1723 
1724 	threadgroup_change_begin(current);
1725 	/*
1726 	 * Ensure that the cgroup subsystem policies allow the new process to be
1727 	 * forked. It should be noted the the new process's css_set can be changed
1728 	 * between here and cgroup_post_fork() if an organisation operation is in
1729 	 * progress.
1730 	 */
1731 	retval = cgroup_can_fork(p);
1732 	if (retval)
1733 		goto bad_fork_free_pid;
1734 
1735 	/*
1736 	 * Make it visible to the rest of the system, but dont wake it up yet.
1737 	 * Need tasklist lock for parent etc handling!
1738 	 */
1739 	write_lock_irq(&tasklist_lock);
1740 
1741 	/* CLONE_PARENT re-uses the old parent */
1742 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1743 		p->real_parent = current->real_parent;
1744 		p->parent_exec_id = current->parent_exec_id;
1745 	} else {
1746 		p->real_parent = current;
1747 		p->parent_exec_id = current->self_exec_id;
1748 	}
1749 
1750 	spin_lock(&current->sighand->siglock);
1751 
1752 	/*
1753 	 * Copy seccomp details explicitly here, in case they were changed
1754 	 * before holding sighand lock.
1755 	 */
1756 	copy_seccomp(p);
1757 
1758 	/*
1759 	 * Process group and session signals need to be delivered to just the
1760 	 * parent before the fork or both the parent and the child after the
1761 	 * fork. Restart if a signal comes in before we add the new process to
1762 	 * it's process group.
1763 	 * A fatal signal pending means that current will exit, so the new
1764 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1765 	*/
1766 	recalc_sigpending();
1767 	if (signal_pending(current)) {
1768 		spin_unlock(&current->sighand->siglock);
1769 		write_unlock_irq(&tasklist_lock);
1770 		retval = -ERESTARTNOINTR;
1771 		goto bad_fork_cancel_cgroup;
1772 	}
1773 
1774 	if (likely(p->pid)) {
1775 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1776 
1777 		init_task_pid(p, PIDTYPE_PID, pid);
1778 		if (thread_group_leader(p)) {
1779 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1780 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1781 
1782 			if (is_child_reaper(pid)) {
1783 				ns_of_pid(pid)->child_reaper = p;
1784 				p->signal->flags |= SIGNAL_UNKILLABLE;
1785 			}
1786 
1787 			p->signal->leader_pid = pid;
1788 			p->signal->tty = tty_kref_get(current->signal->tty);
1789 			list_add_tail(&p->sibling, &p->real_parent->children);
1790 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1791 			attach_pid(p, PIDTYPE_PGID);
1792 			attach_pid(p, PIDTYPE_SID);
1793 			__this_cpu_inc(process_counts);
1794 		} else {
1795 			current->signal->nr_threads++;
1796 			atomic_inc(&current->signal->live);
1797 			atomic_inc(&current->signal->sigcnt);
1798 			list_add_tail_rcu(&p->thread_group,
1799 					  &p->group_leader->thread_group);
1800 			list_add_tail_rcu(&p->thread_node,
1801 					  &p->signal->thread_head);
1802 		}
1803 		attach_pid(p, PIDTYPE_PID);
1804 		nr_threads++;
1805 	}
1806 
1807 	total_forks++;
1808 	spin_unlock(&current->sighand->siglock);
1809 	syscall_tracepoint_update(p);
1810 	write_unlock_irq(&tasklist_lock);
1811 
1812 	proc_fork_connector(p);
1813 	cgroup_post_fork(p);
1814 	threadgroup_change_end(current);
1815 	perf_event_fork(p);
1816 
1817 	trace_task_newtask(p, clone_flags);
1818 	uprobe_copy_process(p, clone_flags);
1819 
1820 	return p;
1821 
1822 bad_fork_cancel_cgroup:
1823 	cgroup_cancel_fork(p);
1824 bad_fork_free_pid:
1825 	threadgroup_change_end(current);
1826 	if (pid != &init_struct_pid)
1827 		free_pid(pid);
1828 bad_fork_cleanup_thread:
1829 	exit_thread(p);
1830 bad_fork_cleanup_io:
1831 	if (p->io_context)
1832 		exit_io_context(p);
1833 bad_fork_cleanup_namespaces:
1834 	exit_task_namespaces(p);
1835 bad_fork_cleanup_mm:
1836 	if (p->mm)
1837 		mmput(p->mm);
1838 bad_fork_cleanup_signal:
1839 	if (!(clone_flags & CLONE_THREAD))
1840 		free_signal_struct(p->signal);
1841 bad_fork_cleanup_sighand:
1842 	__cleanup_sighand(p->sighand);
1843 bad_fork_cleanup_fs:
1844 	exit_fs(p); /* blocking */
1845 bad_fork_cleanup_files:
1846 	exit_files(p); /* blocking */
1847 bad_fork_cleanup_semundo:
1848 	exit_sem(p);
1849 bad_fork_cleanup_audit:
1850 	audit_free(p);
1851 bad_fork_cleanup_perf:
1852 	perf_event_free_task(p);
1853 bad_fork_cleanup_policy:
1854 #ifdef CONFIG_NUMA
1855 	mpol_put(p->mempolicy);
1856 bad_fork_cleanup_threadgroup_lock:
1857 #endif
1858 	delayacct_tsk_free(p);
1859 bad_fork_cleanup_count:
1860 	atomic_dec(&p->cred->user->processes);
1861 	exit_creds(p);
1862 bad_fork_free:
1863 	put_task_stack(p);
1864 	free_task(p);
1865 fork_out:
1866 	return ERR_PTR(retval);
1867 }
1868 
1869 static inline void init_idle_pids(struct pid_link *links)
1870 {
1871 	enum pid_type type;
1872 
1873 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1874 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1875 		links[type].pid = &init_struct_pid;
1876 	}
1877 }
1878 
1879 struct task_struct *fork_idle(int cpu)
1880 {
1881 	struct task_struct *task;
1882 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1883 			    cpu_to_node(cpu));
1884 	if (!IS_ERR(task)) {
1885 		init_idle_pids(task->pids);
1886 		init_idle(task, cpu);
1887 	}
1888 
1889 	return task;
1890 }
1891 
1892 /*
1893  *  Ok, this is the main fork-routine.
1894  *
1895  * It copies the process, and if successful kick-starts
1896  * it and waits for it to finish using the VM if required.
1897  */
1898 long _do_fork(unsigned long clone_flags,
1899 	      unsigned long stack_start,
1900 	      unsigned long stack_size,
1901 	      int __user *parent_tidptr,
1902 	      int __user *child_tidptr,
1903 	      unsigned long tls)
1904 {
1905 	struct task_struct *p;
1906 	int trace = 0;
1907 	long nr;
1908 
1909 	/*
1910 	 * Determine whether and which event to report to ptracer.  When
1911 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
1912 	 * requested, no event is reported; otherwise, report if the event
1913 	 * for the type of forking is enabled.
1914 	 */
1915 	if (!(clone_flags & CLONE_UNTRACED)) {
1916 		if (clone_flags & CLONE_VFORK)
1917 			trace = PTRACE_EVENT_VFORK;
1918 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
1919 			trace = PTRACE_EVENT_CLONE;
1920 		else
1921 			trace = PTRACE_EVENT_FORK;
1922 
1923 		if (likely(!ptrace_event_enabled(current, trace)))
1924 			trace = 0;
1925 	}
1926 
1927 	p = copy_process(clone_flags, stack_start, stack_size,
1928 			 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
1929 	/*
1930 	 * Do this prior waking up the new thread - the thread pointer
1931 	 * might get invalid after that point, if the thread exits quickly.
1932 	 */
1933 	if (!IS_ERR(p)) {
1934 		struct completion vfork;
1935 		struct pid *pid;
1936 
1937 		trace_sched_process_fork(current, p);
1938 
1939 		pid = get_task_pid(p, PIDTYPE_PID);
1940 		nr = pid_vnr(pid);
1941 
1942 		if (clone_flags & CLONE_PARENT_SETTID)
1943 			put_user(nr, parent_tidptr);
1944 
1945 		if (clone_flags & CLONE_VFORK) {
1946 			p->vfork_done = &vfork;
1947 			init_completion(&vfork);
1948 			get_task_struct(p);
1949 		}
1950 
1951 		wake_up_new_task(p);
1952 
1953 		/* forking complete and child started to run, tell ptracer */
1954 		if (unlikely(trace))
1955 			ptrace_event_pid(trace, pid);
1956 
1957 		if (clone_flags & CLONE_VFORK) {
1958 			if (!wait_for_vfork_done(p, &vfork))
1959 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
1960 		}
1961 
1962 		put_pid(pid);
1963 	} else {
1964 		nr = PTR_ERR(p);
1965 	}
1966 	return nr;
1967 }
1968 
1969 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
1970 /* For compatibility with architectures that call do_fork directly rather than
1971  * using the syscall entry points below. */
1972 long do_fork(unsigned long clone_flags,
1973 	      unsigned long stack_start,
1974 	      unsigned long stack_size,
1975 	      int __user *parent_tidptr,
1976 	      int __user *child_tidptr)
1977 {
1978 	return _do_fork(clone_flags, stack_start, stack_size,
1979 			parent_tidptr, child_tidptr, 0);
1980 }
1981 #endif
1982 
1983 /*
1984  * Create a kernel thread.
1985  */
1986 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1987 {
1988 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
1989 		(unsigned long)arg, NULL, NULL, 0);
1990 }
1991 
1992 #ifdef __ARCH_WANT_SYS_FORK
1993 SYSCALL_DEFINE0(fork)
1994 {
1995 #ifdef CONFIG_MMU
1996 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
1997 #else
1998 	/* can not support in nommu mode */
1999 	return -EINVAL;
2000 #endif
2001 }
2002 #endif
2003 
2004 #ifdef __ARCH_WANT_SYS_VFORK
2005 SYSCALL_DEFINE0(vfork)
2006 {
2007 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2008 			0, NULL, NULL, 0);
2009 }
2010 #endif
2011 
2012 #ifdef __ARCH_WANT_SYS_CLONE
2013 #ifdef CONFIG_CLONE_BACKWARDS
2014 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2015 		 int __user *, parent_tidptr,
2016 		 unsigned long, tls,
2017 		 int __user *, child_tidptr)
2018 #elif defined(CONFIG_CLONE_BACKWARDS2)
2019 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2020 		 int __user *, parent_tidptr,
2021 		 int __user *, child_tidptr,
2022 		 unsigned long, tls)
2023 #elif defined(CONFIG_CLONE_BACKWARDS3)
2024 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2025 		int, stack_size,
2026 		int __user *, parent_tidptr,
2027 		int __user *, child_tidptr,
2028 		unsigned long, tls)
2029 #else
2030 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2031 		 int __user *, parent_tidptr,
2032 		 int __user *, child_tidptr,
2033 		 unsigned long, tls)
2034 #endif
2035 {
2036 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2037 }
2038 #endif
2039 
2040 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2041 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2042 #endif
2043 
2044 static void sighand_ctor(void *data)
2045 {
2046 	struct sighand_struct *sighand = data;
2047 
2048 	spin_lock_init(&sighand->siglock);
2049 	init_waitqueue_head(&sighand->signalfd_wqh);
2050 }
2051 
2052 void __init proc_caches_init(void)
2053 {
2054 	sighand_cachep = kmem_cache_create("sighand_cache",
2055 			sizeof(struct sighand_struct), 0,
2056 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
2057 			SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2058 	signal_cachep = kmem_cache_create("signal_cache",
2059 			sizeof(struct signal_struct), 0,
2060 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2061 			NULL);
2062 	files_cachep = kmem_cache_create("files_cache",
2063 			sizeof(struct files_struct), 0,
2064 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2065 			NULL);
2066 	fs_cachep = kmem_cache_create("fs_cache",
2067 			sizeof(struct fs_struct), 0,
2068 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2069 			NULL);
2070 	/*
2071 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2072 	 * whole struct cpumask for the OFFSTACK case. We could change
2073 	 * this to *only* allocate as much of it as required by the
2074 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
2075 	 * is at the end of the structure, exactly for that reason.
2076 	 */
2077 	mm_cachep = kmem_cache_create("mm_struct",
2078 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2079 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2080 			NULL);
2081 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2082 	mmap_init();
2083 	nsproxy_cache_init();
2084 }
2085 
2086 /*
2087  * Check constraints on flags passed to the unshare system call.
2088  */
2089 static int check_unshare_flags(unsigned long unshare_flags)
2090 {
2091 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2092 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2093 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2094 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2095 		return -EINVAL;
2096 	/*
2097 	 * Not implemented, but pretend it works if there is nothing
2098 	 * to unshare.  Note that unsharing the address space or the
2099 	 * signal handlers also need to unshare the signal queues (aka
2100 	 * CLONE_THREAD).
2101 	 */
2102 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2103 		if (!thread_group_empty(current))
2104 			return -EINVAL;
2105 	}
2106 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2107 		if (atomic_read(&current->sighand->count) > 1)
2108 			return -EINVAL;
2109 	}
2110 	if (unshare_flags & CLONE_VM) {
2111 		if (!current_is_single_threaded())
2112 			return -EINVAL;
2113 	}
2114 
2115 	return 0;
2116 }
2117 
2118 /*
2119  * Unshare the filesystem structure if it is being shared
2120  */
2121 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2122 {
2123 	struct fs_struct *fs = current->fs;
2124 
2125 	if (!(unshare_flags & CLONE_FS) || !fs)
2126 		return 0;
2127 
2128 	/* don't need lock here; in the worst case we'll do useless copy */
2129 	if (fs->users == 1)
2130 		return 0;
2131 
2132 	*new_fsp = copy_fs_struct(fs);
2133 	if (!*new_fsp)
2134 		return -ENOMEM;
2135 
2136 	return 0;
2137 }
2138 
2139 /*
2140  * Unshare file descriptor table if it is being shared
2141  */
2142 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2143 {
2144 	struct files_struct *fd = current->files;
2145 	int error = 0;
2146 
2147 	if ((unshare_flags & CLONE_FILES) &&
2148 	    (fd && atomic_read(&fd->count) > 1)) {
2149 		*new_fdp = dup_fd(fd, &error);
2150 		if (!*new_fdp)
2151 			return error;
2152 	}
2153 
2154 	return 0;
2155 }
2156 
2157 /*
2158  * unshare allows a process to 'unshare' part of the process
2159  * context which was originally shared using clone.  copy_*
2160  * functions used by do_fork() cannot be used here directly
2161  * because they modify an inactive task_struct that is being
2162  * constructed. Here we are modifying the current, active,
2163  * task_struct.
2164  */
2165 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2166 {
2167 	struct fs_struct *fs, *new_fs = NULL;
2168 	struct files_struct *fd, *new_fd = NULL;
2169 	struct cred *new_cred = NULL;
2170 	struct nsproxy *new_nsproxy = NULL;
2171 	int do_sysvsem = 0;
2172 	int err;
2173 
2174 	/*
2175 	 * If unsharing a user namespace must also unshare the thread group
2176 	 * and unshare the filesystem root and working directories.
2177 	 */
2178 	if (unshare_flags & CLONE_NEWUSER)
2179 		unshare_flags |= CLONE_THREAD | CLONE_FS;
2180 	/*
2181 	 * If unsharing vm, must also unshare signal handlers.
2182 	 */
2183 	if (unshare_flags & CLONE_VM)
2184 		unshare_flags |= CLONE_SIGHAND;
2185 	/*
2186 	 * If unsharing a signal handlers, must also unshare the signal queues.
2187 	 */
2188 	if (unshare_flags & CLONE_SIGHAND)
2189 		unshare_flags |= CLONE_THREAD;
2190 	/*
2191 	 * If unsharing namespace, must also unshare filesystem information.
2192 	 */
2193 	if (unshare_flags & CLONE_NEWNS)
2194 		unshare_flags |= CLONE_FS;
2195 
2196 	err = check_unshare_flags(unshare_flags);
2197 	if (err)
2198 		goto bad_unshare_out;
2199 	/*
2200 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2201 	 * to a new ipc namespace, the semaphore arrays from the old
2202 	 * namespace are unreachable.
2203 	 */
2204 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2205 		do_sysvsem = 1;
2206 	err = unshare_fs(unshare_flags, &new_fs);
2207 	if (err)
2208 		goto bad_unshare_out;
2209 	err = unshare_fd(unshare_flags, &new_fd);
2210 	if (err)
2211 		goto bad_unshare_cleanup_fs;
2212 	err = unshare_userns(unshare_flags, &new_cred);
2213 	if (err)
2214 		goto bad_unshare_cleanup_fd;
2215 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2216 					 new_cred, new_fs);
2217 	if (err)
2218 		goto bad_unshare_cleanup_cred;
2219 
2220 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2221 		if (do_sysvsem) {
2222 			/*
2223 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2224 			 */
2225 			exit_sem(current);
2226 		}
2227 		if (unshare_flags & CLONE_NEWIPC) {
2228 			/* Orphan segments in old ns (see sem above). */
2229 			exit_shm(current);
2230 			shm_init_task(current);
2231 		}
2232 
2233 		if (new_nsproxy)
2234 			switch_task_namespaces(current, new_nsproxy);
2235 
2236 		task_lock(current);
2237 
2238 		if (new_fs) {
2239 			fs = current->fs;
2240 			spin_lock(&fs->lock);
2241 			current->fs = new_fs;
2242 			if (--fs->users)
2243 				new_fs = NULL;
2244 			else
2245 				new_fs = fs;
2246 			spin_unlock(&fs->lock);
2247 		}
2248 
2249 		if (new_fd) {
2250 			fd = current->files;
2251 			current->files = new_fd;
2252 			new_fd = fd;
2253 		}
2254 
2255 		task_unlock(current);
2256 
2257 		if (new_cred) {
2258 			/* Install the new user namespace */
2259 			commit_creds(new_cred);
2260 			new_cred = NULL;
2261 		}
2262 	}
2263 
2264 bad_unshare_cleanup_cred:
2265 	if (new_cred)
2266 		put_cred(new_cred);
2267 bad_unshare_cleanup_fd:
2268 	if (new_fd)
2269 		put_files_struct(new_fd);
2270 
2271 bad_unshare_cleanup_fs:
2272 	if (new_fs)
2273 		free_fs_struct(new_fs);
2274 
2275 bad_unshare_out:
2276 	return err;
2277 }
2278 
2279 /*
2280  *	Helper to unshare the files of the current task.
2281  *	We don't want to expose copy_files internals to
2282  *	the exec layer of the kernel.
2283  */
2284 
2285 int unshare_files(struct files_struct **displaced)
2286 {
2287 	struct task_struct *task = current;
2288 	struct files_struct *copy = NULL;
2289 	int error;
2290 
2291 	error = unshare_fd(CLONE_FILES, &copy);
2292 	if (error || !copy) {
2293 		*displaced = NULL;
2294 		return error;
2295 	}
2296 	*displaced = task->files;
2297 	task_lock(task);
2298 	task->files = copy;
2299 	task_unlock(task);
2300 	return 0;
2301 }
2302 
2303 int sysctl_max_threads(struct ctl_table *table, int write,
2304 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2305 {
2306 	struct ctl_table t;
2307 	int ret;
2308 	int threads = max_threads;
2309 	int min = MIN_THREADS;
2310 	int max = MAX_THREADS;
2311 
2312 	t = *table;
2313 	t.data = &threads;
2314 	t.extra1 = &min;
2315 	t.extra2 = &max;
2316 
2317 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2318 	if (ret || !write)
2319 		return ret;
2320 
2321 	set_max_threads(threads);
2322 
2323 	return 0;
2324 }
2325