xref: /openbmc/linux/kernel/fork.c (revision 7d4ad2e71d175bf7688e0a9f60e8df12591c9abd)
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13 
14 #include <linux/slab.h>
15 #include <linux/sched/autogroup.h>
16 #include <linux/sched/mm.h>
17 #include <linux/sched/coredump.h>
18 #include <linux/sched/user.h>
19 #include <linux/sched/numa_balancing.h>
20 #include <linux/sched/stat.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/task_stack.h>
23 #include <linux/sched/cputime.h>
24 #include <linux/rtmutex.h>
25 #include <linux/init.h>
26 #include <linux/unistd.h>
27 #include <linux/module.h>
28 #include <linux/vmalloc.h>
29 #include <linux/completion.h>
30 #include <linux/personality.h>
31 #include <linux/mempolicy.h>
32 #include <linux/sem.h>
33 #include <linux/file.h>
34 #include <linux/fdtable.h>
35 #include <linux/iocontext.h>
36 #include <linux/key.h>
37 #include <linux/binfmts.h>
38 #include <linux/mman.h>
39 #include <linux/mmu_notifier.h>
40 #include <linux/fs.h>
41 #include <linux/mm.h>
42 #include <linux/vmacache.h>
43 #include <linux/nsproxy.h>
44 #include <linux/capability.h>
45 #include <linux/cpu.h>
46 #include <linux/cgroup.h>
47 #include <linux/security.h>
48 #include <linux/hugetlb.h>
49 #include <linux/seccomp.h>
50 #include <linux/swap.h>
51 #include <linux/syscalls.h>
52 #include <linux/jiffies.h>
53 #include <linux/futex.h>
54 #include <linux/compat.h>
55 #include <linux/kthread.h>
56 #include <linux/task_io_accounting_ops.h>
57 #include <linux/rcupdate.h>
58 #include <linux/ptrace.h>
59 #include <linux/mount.h>
60 #include <linux/audit.h>
61 #include <linux/memcontrol.h>
62 #include <linux/ftrace.h>
63 #include <linux/proc_fs.h>
64 #include <linux/profile.h>
65 #include <linux/rmap.h>
66 #include <linux/ksm.h>
67 #include <linux/acct.h>
68 #include <linux/userfaultfd_k.h>
69 #include <linux/tsacct_kern.h>
70 #include <linux/cn_proc.h>
71 #include <linux/freezer.h>
72 #include <linux/delayacct.h>
73 #include <linux/taskstats_kern.h>
74 #include <linux/random.h>
75 #include <linux/tty.h>
76 #include <linux/blkdev.h>
77 #include <linux/fs_struct.h>
78 #include <linux/magic.h>
79 #include <linux/perf_event.h>
80 #include <linux/posix-timers.h>
81 #include <linux/user-return-notifier.h>
82 #include <linux/oom.h>
83 #include <linux/khugepaged.h>
84 #include <linux/signalfd.h>
85 #include <linux/uprobes.h>
86 #include <linux/aio.h>
87 #include <linux/compiler.h>
88 #include <linux/sysctl.h>
89 #include <linux/kcov.h>
90 #include <linux/livepatch.h>
91 
92 #include <asm/pgtable.h>
93 #include <asm/pgalloc.h>
94 #include <linux/uaccess.h>
95 #include <asm/mmu_context.h>
96 #include <asm/cacheflush.h>
97 #include <asm/tlbflush.h>
98 
99 #include <trace/events/sched.h>
100 
101 #define CREATE_TRACE_POINTS
102 #include <trace/events/task.h>
103 
104 /*
105  * Minimum number of threads to boot the kernel
106  */
107 #define MIN_THREADS 20
108 
109 /*
110  * Maximum number of threads
111  */
112 #define MAX_THREADS FUTEX_TID_MASK
113 
114 /*
115  * Protected counters by write_lock_irq(&tasklist_lock)
116  */
117 unsigned long total_forks;	/* Handle normal Linux uptimes. */
118 int nr_threads;			/* The idle threads do not count.. */
119 
120 int max_threads;		/* tunable limit on nr_threads */
121 
122 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
123 
124 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
125 
126 #ifdef CONFIG_PROVE_RCU
127 int lockdep_tasklist_lock_is_held(void)
128 {
129 	return lockdep_is_held(&tasklist_lock);
130 }
131 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
132 #endif /* #ifdef CONFIG_PROVE_RCU */
133 
134 int nr_processes(void)
135 {
136 	int cpu;
137 	int total = 0;
138 
139 	for_each_possible_cpu(cpu)
140 		total += per_cpu(process_counts, cpu);
141 
142 	return total;
143 }
144 
145 void __weak arch_release_task_struct(struct task_struct *tsk)
146 {
147 }
148 
149 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
150 static struct kmem_cache *task_struct_cachep;
151 
152 static inline struct task_struct *alloc_task_struct_node(int node)
153 {
154 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
155 }
156 
157 static inline void free_task_struct(struct task_struct *tsk)
158 {
159 	kmem_cache_free(task_struct_cachep, tsk);
160 }
161 #endif
162 
163 void __weak arch_release_thread_stack(unsigned long *stack)
164 {
165 }
166 
167 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
168 
169 /*
170  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
171  * kmemcache based allocator.
172  */
173 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
174 
175 #ifdef CONFIG_VMAP_STACK
176 /*
177  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
178  * flush.  Try to minimize the number of calls by caching stacks.
179  */
180 #define NR_CACHED_STACKS 2
181 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
182 
183 static int free_vm_stack_cache(unsigned int cpu)
184 {
185 	struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
186 	int i;
187 
188 	for (i = 0; i < NR_CACHED_STACKS; i++) {
189 		struct vm_struct *vm_stack = cached_vm_stacks[i];
190 
191 		if (!vm_stack)
192 			continue;
193 
194 		vfree(vm_stack->addr);
195 		cached_vm_stacks[i] = NULL;
196 	}
197 
198 	return 0;
199 }
200 #endif
201 
202 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
203 {
204 #ifdef CONFIG_VMAP_STACK
205 	void *stack;
206 	int i;
207 
208 	for (i = 0; i < NR_CACHED_STACKS; i++) {
209 		struct vm_struct *s;
210 
211 		s = this_cpu_xchg(cached_stacks[i], NULL);
212 
213 		if (!s)
214 			continue;
215 
216 		tsk->stack_vm_area = s;
217 		return s->addr;
218 	}
219 
220 	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_SIZE,
221 				     VMALLOC_START, VMALLOC_END,
222 				     THREADINFO_GFP,
223 				     PAGE_KERNEL,
224 				     0, node, __builtin_return_address(0));
225 
226 	/*
227 	 * We can't call find_vm_area() in interrupt context, and
228 	 * free_thread_stack() can be called in interrupt context,
229 	 * so cache the vm_struct.
230 	 */
231 	if (stack)
232 		tsk->stack_vm_area = find_vm_area(stack);
233 	return stack;
234 #else
235 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
236 					     THREAD_SIZE_ORDER);
237 
238 	return page ? page_address(page) : NULL;
239 #endif
240 }
241 
242 static inline void free_thread_stack(struct task_struct *tsk)
243 {
244 #ifdef CONFIG_VMAP_STACK
245 	if (task_stack_vm_area(tsk)) {
246 		int i;
247 
248 		for (i = 0; i < NR_CACHED_STACKS; i++) {
249 			if (this_cpu_cmpxchg(cached_stacks[i],
250 					NULL, tsk->stack_vm_area) != NULL)
251 				continue;
252 
253 			return;
254 		}
255 
256 		vfree_atomic(tsk->stack);
257 		return;
258 	}
259 #endif
260 
261 	__free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
262 }
263 # else
264 static struct kmem_cache *thread_stack_cache;
265 
266 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
267 						  int node)
268 {
269 	return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
270 }
271 
272 static void free_thread_stack(struct task_struct *tsk)
273 {
274 	kmem_cache_free(thread_stack_cache, tsk->stack);
275 }
276 
277 void thread_stack_cache_init(void)
278 {
279 	thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
280 					      THREAD_SIZE, 0, NULL);
281 	BUG_ON(thread_stack_cache == NULL);
282 }
283 # endif
284 #endif
285 
286 /* SLAB cache for signal_struct structures (tsk->signal) */
287 static struct kmem_cache *signal_cachep;
288 
289 /* SLAB cache for sighand_struct structures (tsk->sighand) */
290 struct kmem_cache *sighand_cachep;
291 
292 /* SLAB cache for files_struct structures (tsk->files) */
293 struct kmem_cache *files_cachep;
294 
295 /* SLAB cache for fs_struct structures (tsk->fs) */
296 struct kmem_cache *fs_cachep;
297 
298 /* SLAB cache for vm_area_struct structures */
299 struct kmem_cache *vm_area_cachep;
300 
301 /* SLAB cache for mm_struct structures (tsk->mm) */
302 static struct kmem_cache *mm_cachep;
303 
304 static void account_kernel_stack(struct task_struct *tsk, int account)
305 {
306 	void *stack = task_stack_page(tsk);
307 	struct vm_struct *vm = task_stack_vm_area(tsk);
308 
309 	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
310 
311 	if (vm) {
312 		int i;
313 
314 		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
315 
316 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
317 			mod_zone_page_state(page_zone(vm->pages[i]),
318 					    NR_KERNEL_STACK_KB,
319 					    PAGE_SIZE / 1024 * account);
320 		}
321 
322 		/* All stack pages belong to the same memcg. */
323 		mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
324 				     account * (THREAD_SIZE / 1024));
325 	} else {
326 		/*
327 		 * All stack pages are in the same zone and belong to the
328 		 * same memcg.
329 		 */
330 		struct page *first_page = virt_to_page(stack);
331 
332 		mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
333 				    THREAD_SIZE / 1024 * account);
334 
335 		mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
336 				     account * (THREAD_SIZE / 1024));
337 	}
338 }
339 
340 static void release_task_stack(struct task_struct *tsk)
341 {
342 	if (WARN_ON(tsk->state != TASK_DEAD))
343 		return;  /* Better to leak the stack than to free prematurely */
344 
345 	account_kernel_stack(tsk, -1);
346 	arch_release_thread_stack(tsk->stack);
347 	free_thread_stack(tsk);
348 	tsk->stack = NULL;
349 #ifdef CONFIG_VMAP_STACK
350 	tsk->stack_vm_area = NULL;
351 #endif
352 }
353 
354 #ifdef CONFIG_THREAD_INFO_IN_TASK
355 void put_task_stack(struct task_struct *tsk)
356 {
357 	if (atomic_dec_and_test(&tsk->stack_refcount))
358 		release_task_stack(tsk);
359 }
360 #endif
361 
362 void free_task(struct task_struct *tsk)
363 {
364 #ifndef CONFIG_THREAD_INFO_IN_TASK
365 	/*
366 	 * The task is finally done with both the stack and thread_info,
367 	 * so free both.
368 	 */
369 	release_task_stack(tsk);
370 #else
371 	/*
372 	 * If the task had a separate stack allocation, it should be gone
373 	 * by now.
374 	 */
375 	WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
376 #endif
377 	rt_mutex_debug_task_free(tsk);
378 	ftrace_graph_exit_task(tsk);
379 	put_seccomp_filter(tsk);
380 	arch_release_task_struct(tsk);
381 	if (tsk->flags & PF_KTHREAD)
382 		free_kthread_struct(tsk);
383 	free_task_struct(tsk);
384 }
385 EXPORT_SYMBOL(free_task);
386 
387 static inline void free_signal_struct(struct signal_struct *sig)
388 {
389 	taskstats_tgid_free(sig);
390 	sched_autogroup_exit(sig);
391 	/*
392 	 * __mmdrop is not safe to call from softirq context on x86 due to
393 	 * pgd_dtor so postpone it to the async context
394 	 */
395 	if (sig->oom_mm)
396 		mmdrop_async(sig->oom_mm);
397 	kmem_cache_free(signal_cachep, sig);
398 }
399 
400 static inline void put_signal_struct(struct signal_struct *sig)
401 {
402 	if (atomic_dec_and_test(&sig->sigcnt))
403 		free_signal_struct(sig);
404 }
405 
406 void __put_task_struct(struct task_struct *tsk)
407 {
408 	WARN_ON(!tsk->exit_state);
409 	WARN_ON(atomic_read(&tsk->usage));
410 	WARN_ON(tsk == current);
411 
412 	cgroup_free(tsk);
413 	task_numa_free(tsk);
414 	security_task_free(tsk);
415 	exit_creds(tsk);
416 	delayacct_tsk_free(tsk);
417 	put_signal_struct(tsk->signal);
418 
419 	if (!profile_handoff_task(tsk))
420 		free_task(tsk);
421 }
422 EXPORT_SYMBOL_GPL(__put_task_struct);
423 
424 void __init __weak arch_task_cache_init(void) { }
425 
426 /*
427  * set_max_threads
428  */
429 static void set_max_threads(unsigned int max_threads_suggested)
430 {
431 	u64 threads;
432 
433 	/*
434 	 * The number of threads shall be limited such that the thread
435 	 * structures may only consume a small part of the available memory.
436 	 */
437 	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
438 		threads = MAX_THREADS;
439 	else
440 		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
441 				    (u64) THREAD_SIZE * 8UL);
442 
443 	if (threads > max_threads_suggested)
444 		threads = max_threads_suggested;
445 
446 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
447 }
448 
449 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
450 /* Initialized by the architecture: */
451 int arch_task_struct_size __read_mostly;
452 #endif
453 
454 void __init fork_init(void)
455 {
456 	int i;
457 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
458 #ifndef ARCH_MIN_TASKALIGN
459 #define ARCH_MIN_TASKALIGN	0
460 #endif
461 	int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
462 
463 	/* create a slab on which task_structs can be allocated */
464 	task_struct_cachep = kmem_cache_create("task_struct",
465 			arch_task_struct_size, align,
466 			SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT, NULL);
467 #endif
468 
469 	/* do the arch specific task caches init */
470 	arch_task_cache_init();
471 
472 	set_max_threads(MAX_THREADS);
473 
474 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
475 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
476 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
477 		init_task.signal->rlim[RLIMIT_NPROC];
478 
479 	for (i = 0; i < UCOUNT_COUNTS; i++) {
480 		init_user_ns.ucount_max[i] = max_threads/2;
481 	}
482 
483 #ifdef CONFIG_VMAP_STACK
484 	cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
485 			  NULL, free_vm_stack_cache);
486 #endif
487 }
488 
489 int __weak arch_dup_task_struct(struct task_struct *dst,
490 					       struct task_struct *src)
491 {
492 	*dst = *src;
493 	return 0;
494 }
495 
496 void set_task_stack_end_magic(struct task_struct *tsk)
497 {
498 	unsigned long *stackend;
499 
500 	stackend = end_of_stack(tsk);
501 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
502 }
503 
504 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
505 {
506 	struct task_struct *tsk;
507 	unsigned long *stack;
508 	struct vm_struct *stack_vm_area;
509 	int err;
510 
511 	if (node == NUMA_NO_NODE)
512 		node = tsk_fork_get_node(orig);
513 	tsk = alloc_task_struct_node(node);
514 	if (!tsk)
515 		return NULL;
516 
517 	stack = alloc_thread_stack_node(tsk, node);
518 	if (!stack)
519 		goto free_tsk;
520 
521 	stack_vm_area = task_stack_vm_area(tsk);
522 
523 	err = arch_dup_task_struct(tsk, orig);
524 
525 	/*
526 	 * arch_dup_task_struct() clobbers the stack-related fields.  Make
527 	 * sure they're properly initialized before using any stack-related
528 	 * functions again.
529 	 */
530 	tsk->stack = stack;
531 #ifdef CONFIG_VMAP_STACK
532 	tsk->stack_vm_area = stack_vm_area;
533 #endif
534 #ifdef CONFIG_THREAD_INFO_IN_TASK
535 	atomic_set(&tsk->stack_refcount, 1);
536 #endif
537 
538 	if (err)
539 		goto free_stack;
540 
541 #ifdef CONFIG_SECCOMP
542 	/*
543 	 * We must handle setting up seccomp filters once we're under
544 	 * the sighand lock in case orig has changed between now and
545 	 * then. Until then, filter must be NULL to avoid messing up
546 	 * the usage counts on the error path calling free_task.
547 	 */
548 	tsk->seccomp.filter = NULL;
549 #endif
550 
551 	setup_thread_stack(tsk, orig);
552 	clear_user_return_notifier(tsk);
553 	clear_tsk_need_resched(tsk);
554 	set_task_stack_end_magic(tsk);
555 
556 #ifdef CONFIG_CC_STACKPROTECTOR
557 	tsk->stack_canary = get_random_canary();
558 #endif
559 
560 	/*
561 	 * One for us, one for whoever does the "release_task()" (usually
562 	 * parent)
563 	 */
564 	atomic_set(&tsk->usage, 2);
565 #ifdef CONFIG_BLK_DEV_IO_TRACE
566 	tsk->btrace_seq = 0;
567 #endif
568 	tsk->splice_pipe = NULL;
569 	tsk->task_frag.page = NULL;
570 	tsk->wake_q.next = NULL;
571 
572 	account_kernel_stack(tsk, 1);
573 
574 	kcov_task_init(tsk);
575 
576 #ifdef CONFIG_FAULT_INJECTION
577 	tsk->fail_nth = 0;
578 #endif
579 
580 	return tsk;
581 
582 free_stack:
583 	free_thread_stack(tsk);
584 free_tsk:
585 	free_task_struct(tsk);
586 	return NULL;
587 }
588 
589 #ifdef CONFIG_MMU
590 static __latent_entropy int dup_mmap(struct mm_struct *mm,
591 					struct mm_struct *oldmm)
592 {
593 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
594 	struct rb_node **rb_link, *rb_parent;
595 	int retval;
596 	unsigned long charge;
597 	LIST_HEAD(uf);
598 
599 	uprobe_start_dup_mmap();
600 	if (down_write_killable(&oldmm->mmap_sem)) {
601 		retval = -EINTR;
602 		goto fail_uprobe_end;
603 	}
604 	flush_cache_dup_mm(oldmm);
605 	uprobe_dup_mmap(oldmm, mm);
606 	/*
607 	 * Not linked in yet - no deadlock potential:
608 	 */
609 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
610 
611 	/* No ordering required: file already has been exposed. */
612 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
613 
614 	mm->total_vm = oldmm->total_vm;
615 	mm->data_vm = oldmm->data_vm;
616 	mm->exec_vm = oldmm->exec_vm;
617 	mm->stack_vm = oldmm->stack_vm;
618 
619 	rb_link = &mm->mm_rb.rb_node;
620 	rb_parent = NULL;
621 	pprev = &mm->mmap;
622 	retval = ksm_fork(mm, oldmm);
623 	if (retval)
624 		goto out;
625 	retval = khugepaged_fork(mm, oldmm);
626 	if (retval)
627 		goto out;
628 
629 	prev = NULL;
630 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
631 		struct file *file;
632 
633 		if (mpnt->vm_flags & VM_DONTCOPY) {
634 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
635 			continue;
636 		}
637 		charge = 0;
638 		if (mpnt->vm_flags & VM_ACCOUNT) {
639 			unsigned long len = vma_pages(mpnt);
640 
641 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
642 				goto fail_nomem;
643 			charge = len;
644 		}
645 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
646 		if (!tmp)
647 			goto fail_nomem;
648 		*tmp = *mpnt;
649 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
650 		retval = vma_dup_policy(mpnt, tmp);
651 		if (retval)
652 			goto fail_nomem_policy;
653 		tmp->vm_mm = mm;
654 		retval = dup_userfaultfd(tmp, &uf);
655 		if (retval)
656 			goto fail_nomem_anon_vma_fork;
657 		if (anon_vma_fork(tmp, mpnt))
658 			goto fail_nomem_anon_vma_fork;
659 		tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
660 		tmp->vm_next = tmp->vm_prev = NULL;
661 		file = tmp->vm_file;
662 		if (file) {
663 			struct inode *inode = file_inode(file);
664 			struct address_space *mapping = file->f_mapping;
665 
666 			get_file(file);
667 			if (tmp->vm_flags & VM_DENYWRITE)
668 				atomic_dec(&inode->i_writecount);
669 			i_mmap_lock_write(mapping);
670 			if (tmp->vm_flags & VM_SHARED)
671 				atomic_inc(&mapping->i_mmap_writable);
672 			flush_dcache_mmap_lock(mapping);
673 			/* insert tmp into the share list, just after mpnt */
674 			vma_interval_tree_insert_after(tmp, mpnt,
675 					&mapping->i_mmap);
676 			flush_dcache_mmap_unlock(mapping);
677 			i_mmap_unlock_write(mapping);
678 		}
679 
680 		/*
681 		 * Clear hugetlb-related page reserves for children. This only
682 		 * affects MAP_PRIVATE mappings. Faults generated by the child
683 		 * are not guaranteed to succeed, even if read-only
684 		 */
685 		if (is_vm_hugetlb_page(tmp))
686 			reset_vma_resv_huge_pages(tmp);
687 
688 		/*
689 		 * Link in the new vma and copy the page table entries.
690 		 */
691 		*pprev = tmp;
692 		pprev = &tmp->vm_next;
693 		tmp->vm_prev = prev;
694 		prev = tmp;
695 
696 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
697 		rb_link = &tmp->vm_rb.rb_right;
698 		rb_parent = &tmp->vm_rb;
699 
700 		mm->map_count++;
701 		retval = copy_page_range(mm, oldmm, mpnt);
702 
703 		if (tmp->vm_ops && tmp->vm_ops->open)
704 			tmp->vm_ops->open(tmp);
705 
706 		if (retval)
707 			goto out;
708 	}
709 	/* a new mm has just been created */
710 	arch_dup_mmap(oldmm, mm);
711 	retval = 0;
712 out:
713 	up_write(&mm->mmap_sem);
714 	flush_tlb_mm(oldmm);
715 	up_write(&oldmm->mmap_sem);
716 	dup_userfaultfd_complete(&uf);
717 fail_uprobe_end:
718 	uprobe_end_dup_mmap();
719 	return retval;
720 fail_nomem_anon_vma_fork:
721 	mpol_put(vma_policy(tmp));
722 fail_nomem_policy:
723 	kmem_cache_free(vm_area_cachep, tmp);
724 fail_nomem:
725 	retval = -ENOMEM;
726 	vm_unacct_memory(charge);
727 	goto out;
728 }
729 
730 static inline int mm_alloc_pgd(struct mm_struct *mm)
731 {
732 	mm->pgd = pgd_alloc(mm);
733 	if (unlikely(!mm->pgd))
734 		return -ENOMEM;
735 	return 0;
736 }
737 
738 static inline void mm_free_pgd(struct mm_struct *mm)
739 {
740 	pgd_free(mm, mm->pgd);
741 }
742 #else
743 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
744 {
745 	down_write(&oldmm->mmap_sem);
746 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
747 	up_write(&oldmm->mmap_sem);
748 	return 0;
749 }
750 #define mm_alloc_pgd(mm)	(0)
751 #define mm_free_pgd(mm)
752 #endif /* CONFIG_MMU */
753 
754 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
755 
756 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
757 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
758 
759 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
760 
761 static int __init coredump_filter_setup(char *s)
762 {
763 	default_dump_filter =
764 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
765 		MMF_DUMP_FILTER_MASK;
766 	return 1;
767 }
768 
769 __setup("coredump_filter=", coredump_filter_setup);
770 
771 #include <linux/init_task.h>
772 
773 static void mm_init_aio(struct mm_struct *mm)
774 {
775 #ifdef CONFIG_AIO
776 	spin_lock_init(&mm->ioctx_lock);
777 	mm->ioctx_table = NULL;
778 #endif
779 }
780 
781 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
782 {
783 #ifdef CONFIG_MEMCG
784 	mm->owner = p;
785 #endif
786 }
787 
788 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
789 	struct user_namespace *user_ns)
790 {
791 	mm->mmap = NULL;
792 	mm->mm_rb = RB_ROOT;
793 	mm->vmacache_seqnum = 0;
794 	atomic_set(&mm->mm_users, 1);
795 	atomic_set(&mm->mm_count, 1);
796 	init_rwsem(&mm->mmap_sem);
797 	INIT_LIST_HEAD(&mm->mmlist);
798 	mm->core_state = NULL;
799 	atomic_long_set(&mm->nr_ptes, 0);
800 	mm_nr_pmds_init(mm);
801 	mm->map_count = 0;
802 	mm->locked_vm = 0;
803 	mm->pinned_vm = 0;
804 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
805 	spin_lock_init(&mm->page_table_lock);
806 	mm_init_cpumask(mm);
807 	mm_init_aio(mm);
808 	mm_init_owner(mm, p);
809 	RCU_INIT_POINTER(mm->exe_file, NULL);
810 	mmu_notifier_mm_init(mm);
811 	init_tlb_flush_pending(mm);
812 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
813 	mm->pmd_huge_pte = NULL;
814 #endif
815 
816 	if (current->mm) {
817 		mm->flags = current->mm->flags & MMF_INIT_MASK;
818 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
819 	} else {
820 		mm->flags = default_dump_filter;
821 		mm->def_flags = 0;
822 	}
823 
824 	if (mm_alloc_pgd(mm))
825 		goto fail_nopgd;
826 
827 	if (init_new_context(p, mm))
828 		goto fail_nocontext;
829 
830 	mm->user_ns = get_user_ns(user_ns);
831 	return mm;
832 
833 fail_nocontext:
834 	mm_free_pgd(mm);
835 fail_nopgd:
836 	free_mm(mm);
837 	return NULL;
838 }
839 
840 static void check_mm(struct mm_struct *mm)
841 {
842 	int i;
843 
844 	for (i = 0; i < NR_MM_COUNTERS; i++) {
845 		long x = atomic_long_read(&mm->rss_stat.count[i]);
846 
847 		if (unlikely(x))
848 			printk(KERN_ALERT "BUG: Bad rss-counter state "
849 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
850 	}
851 
852 	if (atomic_long_read(&mm->nr_ptes))
853 		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
854 				atomic_long_read(&mm->nr_ptes));
855 	if (mm_nr_pmds(mm))
856 		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
857 				mm_nr_pmds(mm));
858 
859 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
860 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
861 #endif
862 }
863 
864 /*
865  * Allocate and initialize an mm_struct.
866  */
867 struct mm_struct *mm_alloc(void)
868 {
869 	struct mm_struct *mm;
870 
871 	mm = allocate_mm();
872 	if (!mm)
873 		return NULL;
874 
875 	memset(mm, 0, sizeof(*mm));
876 	return mm_init(mm, current, current_user_ns());
877 }
878 
879 /*
880  * Called when the last reference to the mm
881  * is dropped: either by a lazy thread or by
882  * mmput. Free the page directory and the mm.
883  */
884 void __mmdrop(struct mm_struct *mm)
885 {
886 	BUG_ON(mm == &init_mm);
887 	mm_free_pgd(mm);
888 	destroy_context(mm);
889 	mmu_notifier_mm_destroy(mm);
890 	check_mm(mm);
891 	put_user_ns(mm->user_ns);
892 	free_mm(mm);
893 }
894 EXPORT_SYMBOL_GPL(__mmdrop);
895 
896 static inline void __mmput(struct mm_struct *mm)
897 {
898 	VM_BUG_ON(atomic_read(&mm->mm_users));
899 
900 	uprobe_clear_state(mm);
901 	exit_aio(mm);
902 	ksm_exit(mm);
903 	khugepaged_exit(mm); /* must run before exit_mmap */
904 	exit_mmap(mm);
905 	mm_put_huge_zero_page(mm);
906 	set_mm_exe_file(mm, NULL);
907 	if (!list_empty(&mm->mmlist)) {
908 		spin_lock(&mmlist_lock);
909 		list_del(&mm->mmlist);
910 		spin_unlock(&mmlist_lock);
911 	}
912 	if (mm->binfmt)
913 		module_put(mm->binfmt->module);
914 	set_bit(MMF_OOM_SKIP, &mm->flags);
915 	mmdrop(mm);
916 }
917 
918 /*
919  * Decrement the use count and release all resources for an mm.
920  */
921 void mmput(struct mm_struct *mm)
922 {
923 	might_sleep();
924 
925 	if (atomic_dec_and_test(&mm->mm_users))
926 		__mmput(mm);
927 }
928 EXPORT_SYMBOL_GPL(mmput);
929 
930 #ifdef CONFIG_MMU
931 static void mmput_async_fn(struct work_struct *work)
932 {
933 	struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
934 	__mmput(mm);
935 }
936 
937 void mmput_async(struct mm_struct *mm)
938 {
939 	if (atomic_dec_and_test(&mm->mm_users)) {
940 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
941 		schedule_work(&mm->async_put_work);
942 	}
943 }
944 #endif
945 
946 /**
947  * set_mm_exe_file - change a reference to the mm's executable file
948  *
949  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
950  *
951  * Main users are mmput() and sys_execve(). Callers prevent concurrent
952  * invocations: in mmput() nobody alive left, in execve task is single
953  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
954  * mm->exe_file, but does so without using set_mm_exe_file() in order
955  * to do avoid the need for any locks.
956  */
957 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
958 {
959 	struct file *old_exe_file;
960 
961 	/*
962 	 * It is safe to dereference the exe_file without RCU as
963 	 * this function is only called if nobody else can access
964 	 * this mm -- see comment above for justification.
965 	 */
966 	old_exe_file = rcu_dereference_raw(mm->exe_file);
967 
968 	if (new_exe_file)
969 		get_file(new_exe_file);
970 	rcu_assign_pointer(mm->exe_file, new_exe_file);
971 	if (old_exe_file)
972 		fput(old_exe_file);
973 }
974 
975 /**
976  * get_mm_exe_file - acquire a reference to the mm's executable file
977  *
978  * Returns %NULL if mm has no associated executable file.
979  * User must release file via fput().
980  */
981 struct file *get_mm_exe_file(struct mm_struct *mm)
982 {
983 	struct file *exe_file;
984 
985 	rcu_read_lock();
986 	exe_file = rcu_dereference(mm->exe_file);
987 	if (exe_file && !get_file_rcu(exe_file))
988 		exe_file = NULL;
989 	rcu_read_unlock();
990 	return exe_file;
991 }
992 EXPORT_SYMBOL(get_mm_exe_file);
993 
994 /**
995  * get_task_exe_file - acquire a reference to the task's executable file
996  *
997  * Returns %NULL if task's mm (if any) has no associated executable file or
998  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
999  * User must release file via fput().
1000  */
1001 struct file *get_task_exe_file(struct task_struct *task)
1002 {
1003 	struct file *exe_file = NULL;
1004 	struct mm_struct *mm;
1005 
1006 	task_lock(task);
1007 	mm = task->mm;
1008 	if (mm) {
1009 		if (!(task->flags & PF_KTHREAD))
1010 			exe_file = get_mm_exe_file(mm);
1011 	}
1012 	task_unlock(task);
1013 	return exe_file;
1014 }
1015 EXPORT_SYMBOL(get_task_exe_file);
1016 
1017 /**
1018  * get_task_mm - acquire a reference to the task's mm
1019  *
1020  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1021  * this kernel workthread has transiently adopted a user mm with use_mm,
1022  * to do its AIO) is not set and if so returns a reference to it, after
1023  * bumping up the use count.  User must release the mm via mmput()
1024  * after use.  Typically used by /proc and ptrace.
1025  */
1026 struct mm_struct *get_task_mm(struct task_struct *task)
1027 {
1028 	struct mm_struct *mm;
1029 
1030 	task_lock(task);
1031 	mm = task->mm;
1032 	if (mm) {
1033 		if (task->flags & PF_KTHREAD)
1034 			mm = NULL;
1035 		else
1036 			mmget(mm);
1037 	}
1038 	task_unlock(task);
1039 	return mm;
1040 }
1041 EXPORT_SYMBOL_GPL(get_task_mm);
1042 
1043 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1044 {
1045 	struct mm_struct *mm;
1046 	int err;
1047 
1048 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1049 	if (err)
1050 		return ERR_PTR(err);
1051 
1052 	mm = get_task_mm(task);
1053 	if (mm && mm != current->mm &&
1054 			!ptrace_may_access(task, mode)) {
1055 		mmput(mm);
1056 		mm = ERR_PTR(-EACCES);
1057 	}
1058 	mutex_unlock(&task->signal->cred_guard_mutex);
1059 
1060 	return mm;
1061 }
1062 
1063 static void complete_vfork_done(struct task_struct *tsk)
1064 {
1065 	struct completion *vfork;
1066 
1067 	task_lock(tsk);
1068 	vfork = tsk->vfork_done;
1069 	if (likely(vfork)) {
1070 		tsk->vfork_done = NULL;
1071 		complete(vfork);
1072 	}
1073 	task_unlock(tsk);
1074 }
1075 
1076 static int wait_for_vfork_done(struct task_struct *child,
1077 				struct completion *vfork)
1078 {
1079 	int killed;
1080 
1081 	freezer_do_not_count();
1082 	killed = wait_for_completion_killable(vfork);
1083 	freezer_count();
1084 
1085 	if (killed) {
1086 		task_lock(child);
1087 		child->vfork_done = NULL;
1088 		task_unlock(child);
1089 	}
1090 
1091 	put_task_struct(child);
1092 	return killed;
1093 }
1094 
1095 /* Please note the differences between mmput and mm_release.
1096  * mmput is called whenever we stop holding onto a mm_struct,
1097  * error success whatever.
1098  *
1099  * mm_release is called after a mm_struct has been removed
1100  * from the current process.
1101  *
1102  * This difference is important for error handling, when we
1103  * only half set up a mm_struct for a new process and need to restore
1104  * the old one.  Because we mmput the new mm_struct before
1105  * restoring the old one. . .
1106  * Eric Biederman 10 January 1998
1107  */
1108 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1109 {
1110 	/* Get rid of any futexes when releasing the mm */
1111 #ifdef CONFIG_FUTEX
1112 	if (unlikely(tsk->robust_list)) {
1113 		exit_robust_list(tsk);
1114 		tsk->robust_list = NULL;
1115 	}
1116 #ifdef CONFIG_COMPAT
1117 	if (unlikely(tsk->compat_robust_list)) {
1118 		compat_exit_robust_list(tsk);
1119 		tsk->compat_robust_list = NULL;
1120 	}
1121 #endif
1122 	if (unlikely(!list_empty(&tsk->pi_state_list)))
1123 		exit_pi_state_list(tsk);
1124 #endif
1125 
1126 	uprobe_free_utask(tsk);
1127 
1128 	/* Get rid of any cached register state */
1129 	deactivate_mm(tsk, mm);
1130 
1131 	/*
1132 	 * Signal userspace if we're not exiting with a core dump
1133 	 * because we want to leave the value intact for debugging
1134 	 * purposes.
1135 	 */
1136 	if (tsk->clear_child_tid) {
1137 		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1138 		    atomic_read(&mm->mm_users) > 1) {
1139 			/*
1140 			 * We don't check the error code - if userspace has
1141 			 * not set up a proper pointer then tough luck.
1142 			 */
1143 			put_user(0, tsk->clear_child_tid);
1144 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1145 					1, NULL, NULL, 0);
1146 		}
1147 		tsk->clear_child_tid = NULL;
1148 	}
1149 
1150 	/*
1151 	 * All done, finally we can wake up parent and return this mm to him.
1152 	 * Also kthread_stop() uses this completion for synchronization.
1153 	 */
1154 	if (tsk->vfork_done)
1155 		complete_vfork_done(tsk);
1156 }
1157 
1158 /*
1159  * Allocate a new mm structure and copy contents from the
1160  * mm structure of the passed in task structure.
1161  */
1162 static struct mm_struct *dup_mm(struct task_struct *tsk)
1163 {
1164 	struct mm_struct *mm, *oldmm = current->mm;
1165 	int err;
1166 
1167 	mm = allocate_mm();
1168 	if (!mm)
1169 		goto fail_nomem;
1170 
1171 	memcpy(mm, oldmm, sizeof(*mm));
1172 
1173 	if (!mm_init(mm, tsk, mm->user_ns))
1174 		goto fail_nomem;
1175 
1176 	err = dup_mmap(mm, oldmm);
1177 	if (err)
1178 		goto free_pt;
1179 
1180 	mm->hiwater_rss = get_mm_rss(mm);
1181 	mm->hiwater_vm = mm->total_vm;
1182 
1183 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1184 		goto free_pt;
1185 
1186 	return mm;
1187 
1188 free_pt:
1189 	/* don't put binfmt in mmput, we haven't got module yet */
1190 	mm->binfmt = NULL;
1191 	mmput(mm);
1192 
1193 fail_nomem:
1194 	return NULL;
1195 }
1196 
1197 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1198 {
1199 	struct mm_struct *mm, *oldmm;
1200 	int retval;
1201 
1202 	tsk->min_flt = tsk->maj_flt = 0;
1203 	tsk->nvcsw = tsk->nivcsw = 0;
1204 #ifdef CONFIG_DETECT_HUNG_TASK
1205 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1206 #endif
1207 
1208 	tsk->mm = NULL;
1209 	tsk->active_mm = NULL;
1210 
1211 	/*
1212 	 * Are we cloning a kernel thread?
1213 	 *
1214 	 * We need to steal a active VM for that..
1215 	 */
1216 	oldmm = current->mm;
1217 	if (!oldmm)
1218 		return 0;
1219 
1220 	/* initialize the new vmacache entries */
1221 	vmacache_flush(tsk);
1222 
1223 	if (clone_flags & CLONE_VM) {
1224 		mmget(oldmm);
1225 		mm = oldmm;
1226 		goto good_mm;
1227 	}
1228 
1229 	retval = -ENOMEM;
1230 	mm = dup_mm(tsk);
1231 	if (!mm)
1232 		goto fail_nomem;
1233 
1234 good_mm:
1235 	tsk->mm = mm;
1236 	tsk->active_mm = mm;
1237 	return 0;
1238 
1239 fail_nomem:
1240 	return retval;
1241 }
1242 
1243 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1244 {
1245 	struct fs_struct *fs = current->fs;
1246 	if (clone_flags & CLONE_FS) {
1247 		/* tsk->fs is already what we want */
1248 		spin_lock(&fs->lock);
1249 		if (fs->in_exec) {
1250 			spin_unlock(&fs->lock);
1251 			return -EAGAIN;
1252 		}
1253 		fs->users++;
1254 		spin_unlock(&fs->lock);
1255 		return 0;
1256 	}
1257 	tsk->fs = copy_fs_struct(fs);
1258 	if (!tsk->fs)
1259 		return -ENOMEM;
1260 	return 0;
1261 }
1262 
1263 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1264 {
1265 	struct files_struct *oldf, *newf;
1266 	int error = 0;
1267 
1268 	/*
1269 	 * A background process may not have any files ...
1270 	 */
1271 	oldf = current->files;
1272 	if (!oldf)
1273 		goto out;
1274 
1275 	if (clone_flags & CLONE_FILES) {
1276 		atomic_inc(&oldf->count);
1277 		goto out;
1278 	}
1279 
1280 	newf = dup_fd(oldf, &error);
1281 	if (!newf)
1282 		goto out;
1283 
1284 	tsk->files = newf;
1285 	error = 0;
1286 out:
1287 	return error;
1288 }
1289 
1290 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1291 {
1292 #ifdef CONFIG_BLOCK
1293 	struct io_context *ioc = current->io_context;
1294 	struct io_context *new_ioc;
1295 
1296 	if (!ioc)
1297 		return 0;
1298 	/*
1299 	 * Share io context with parent, if CLONE_IO is set
1300 	 */
1301 	if (clone_flags & CLONE_IO) {
1302 		ioc_task_link(ioc);
1303 		tsk->io_context = ioc;
1304 	} else if (ioprio_valid(ioc->ioprio)) {
1305 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1306 		if (unlikely(!new_ioc))
1307 			return -ENOMEM;
1308 
1309 		new_ioc->ioprio = ioc->ioprio;
1310 		put_io_context(new_ioc);
1311 	}
1312 #endif
1313 	return 0;
1314 }
1315 
1316 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1317 {
1318 	struct sighand_struct *sig;
1319 
1320 	if (clone_flags & CLONE_SIGHAND) {
1321 		atomic_inc(&current->sighand->count);
1322 		return 0;
1323 	}
1324 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1325 	rcu_assign_pointer(tsk->sighand, sig);
1326 	if (!sig)
1327 		return -ENOMEM;
1328 
1329 	atomic_set(&sig->count, 1);
1330 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1331 	return 0;
1332 }
1333 
1334 void __cleanup_sighand(struct sighand_struct *sighand)
1335 {
1336 	if (atomic_dec_and_test(&sighand->count)) {
1337 		signalfd_cleanup(sighand);
1338 		/*
1339 		 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1340 		 * without an RCU grace period, see __lock_task_sighand().
1341 		 */
1342 		kmem_cache_free(sighand_cachep, sighand);
1343 	}
1344 }
1345 
1346 #ifdef CONFIG_POSIX_TIMERS
1347 /*
1348  * Initialize POSIX timer handling for a thread group.
1349  */
1350 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1351 {
1352 	unsigned long cpu_limit;
1353 
1354 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1355 	if (cpu_limit != RLIM_INFINITY) {
1356 		sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1357 		sig->cputimer.running = true;
1358 	}
1359 
1360 	/* The timer lists. */
1361 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1362 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1363 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1364 }
1365 #else
1366 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1367 #endif
1368 
1369 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1370 {
1371 	struct signal_struct *sig;
1372 
1373 	if (clone_flags & CLONE_THREAD)
1374 		return 0;
1375 
1376 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1377 	tsk->signal = sig;
1378 	if (!sig)
1379 		return -ENOMEM;
1380 
1381 	sig->nr_threads = 1;
1382 	atomic_set(&sig->live, 1);
1383 	atomic_set(&sig->sigcnt, 1);
1384 
1385 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1386 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1387 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1388 
1389 	init_waitqueue_head(&sig->wait_chldexit);
1390 	sig->curr_target = tsk;
1391 	init_sigpending(&sig->shared_pending);
1392 	seqlock_init(&sig->stats_lock);
1393 	prev_cputime_init(&sig->prev_cputime);
1394 
1395 #ifdef CONFIG_POSIX_TIMERS
1396 	INIT_LIST_HEAD(&sig->posix_timers);
1397 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1398 	sig->real_timer.function = it_real_fn;
1399 #endif
1400 
1401 	task_lock(current->group_leader);
1402 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1403 	task_unlock(current->group_leader);
1404 
1405 	posix_cpu_timers_init_group(sig);
1406 
1407 	tty_audit_fork(sig);
1408 	sched_autogroup_fork(sig);
1409 
1410 	sig->oom_score_adj = current->signal->oom_score_adj;
1411 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1412 
1413 	mutex_init(&sig->cred_guard_mutex);
1414 
1415 	return 0;
1416 }
1417 
1418 static void copy_seccomp(struct task_struct *p)
1419 {
1420 #ifdef CONFIG_SECCOMP
1421 	/*
1422 	 * Must be called with sighand->lock held, which is common to
1423 	 * all threads in the group. Holding cred_guard_mutex is not
1424 	 * needed because this new task is not yet running and cannot
1425 	 * be racing exec.
1426 	 */
1427 	assert_spin_locked(&current->sighand->siglock);
1428 
1429 	/* Ref-count the new filter user, and assign it. */
1430 	get_seccomp_filter(current);
1431 	p->seccomp = current->seccomp;
1432 
1433 	/*
1434 	 * Explicitly enable no_new_privs here in case it got set
1435 	 * between the task_struct being duplicated and holding the
1436 	 * sighand lock. The seccomp state and nnp must be in sync.
1437 	 */
1438 	if (task_no_new_privs(current))
1439 		task_set_no_new_privs(p);
1440 
1441 	/*
1442 	 * If the parent gained a seccomp mode after copying thread
1443 	 * flags and between before we held the sighand lock, we have
1444 	 * to manually enable the seccomp thread flag here.
1445 	 */
1446 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1447 		set_tsk_thread_flag(p, TIF_SECCOMP);
1448 #endif
1449 }
1450 
1451 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1452 {
1453 	current->clear_child_tid = tidptr;
1454 
1455 	return task_pid_vnr(current);
1456 }
1457 
1458 static void rt_mutex_init_task(struct task_struct *p)
1459 {
1460 	raw_spin_lock_init(&p->pi_lock);
1461 #ifdef CONFIG_RT_MUTEXES
1462 	p->pi_waiters = RB_ROOT;
1463 	p->pi_waiters_leftmost = NULL;
1464 	p->pi_top_task = NULL;
1465 	p->pi_blocked_on = NULL;
1466 #endif
1467 }
1468 
1469 #ifdef CONFIG_POSIX_TIMERS
1470 /*
1471  * Initialize POSIX timer handling for a single task.
1472  */
1473 static void posix_cpu_timers_init(struct task_struct *tsk)
1474 {
1475 	tsk->cputime_expires.prof_exp = 0;
1476 	tsk->cputime_expires.virt_exp = 0;
1477 	tsk->cputime_expires.sched_exp = 0;
1478 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1479 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1480 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1481 }
1482 #else
1483 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1484 #endif
1485 
1486 static inline void
1487 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1488 {
1489 	 task->pids[type].pid = pid;
1490 }
1491 
1492 static inline void rcu_copy_process(struct task_struct *p)
1493 {
1494 #ifdef CONFIG_PREEMPT_RCU
1495 	p->rcu_read_lock_nesting = 0;
1496 	p->rcu_read_unlock_special.s = 0;
1497 	p->rcu_blocked_node = NULL;
1498 	INIT_LIST_HEAD(&p->rcu_node_entry);
1499 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1500 #ifdef CONFIG_TASKS_RCU
1501 	p->rcu_tasks_holdout = false;
1502 	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1503 	p->rcu_tasks_idle_cpu = -1;
1504 #endif /* #ifdef CONFIG_TASKS_RCU */
1505 }
1506 
1507 /*
1508  * This creates a new process as a copy of the old one,
1509  * but does not actually start it yet.
1510  *
1511  * It copies the registers, and all the appropriate
1512  * parts of the process environment (as per the clone
1513  * flags). The actual kick-off is left to the caller.
1514  */
1515 static __latent_entropy struct task_struct *copy_process(
1516 					unsigned long clone_flags,
1517 					unsigned long stack_start,
1518 					unsigned long stack_size,
1519 					int __user *child_tidptr,
1520 					struct pid *pid,
1521 					int trace,
1522 					unsigned long tls,
1523 					int node)
1524 {
1525 	int retval;
1526 	struct task_struct *p;
1527 
1528 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1529 		return ERR_PTR(-EINVAL);
1530 
1531 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1532 		return ERR_PTR(-EINVAL);
1533 
1534 	/*
1535 	 * Thread groups must share signals as well, and detached threads
1536 	 * can only be started up within the thread group.
1537 	 */
1538 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1539 		return ERR_PTR(-EINVAL);
1540 
1541 	/*
1542 	 * Shared signal handlers imply shared VM. By way of the above,
1543 	 * thread groups also imply shared VM. Blocking this case allows
1544 	 * for various simplifications in other code.
1545 	 */
1546 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1547 		return ERR_PTR(-EINVAL);
1548 
1549 	/*
1550 	 * Siblings of global init remain as zombies on exit since they are
1551 	 * not reaped by their parent (swapper). To solve this and to avoid
1552 	 * multi-rooted process trees, prevent global and container-inits
1553 	 * from creating siblings.
1554 	 */
1555 	if ((clone_flags & CLONE_PARENT) &&
1556 				current->signal->flags & SIGNAL_UNKILLABLE)
1557 		return ERR_PTR(-EINVAL);
1558 
1559 	/*
1560 	 * If the new process will be in a different pid or user namespace
1561 	 * do not allow it to share a thread group with the forking task.
1562 	 */
1563 	if (clone_flags & CLONE_THREAD) {
1564 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1565 		    (task_active_pid_ns(current) !=
1566 				current->nsproxy->pid_ns_for_children))
1567 			return ERR_PTR(-EINVAL);
1568 	}
1569 
1570 	retval = security_task_create(clone_flags);
1571 	if (retval)
1572 		goto fork_out;
1573 
1574 	retval = -ENOMEM;
1575 	p = dup_task_struct(current, node);
1576 	if (!p)
1577 		goto fork_out;
1578 
1579 	/*
1580 	 * This _must_ happen before we call free_task(), i.e. before we jump
1581 	 * to any of the bad_fork_* labels. This is to avoid freeing
1582 	 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1583 	 * kernel threads (PF_KTHREAD).
1584 	 */
1585 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1586 	/*
1587 	 * Clear TID on mm_release()?
1588 	 */
1589 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1590 
1591 	ftrace_graph_init_task(p);
1592 
1593 	rt_mutex_init_task(p);
1594 
1595 #ifdef CONFIG_PROVE_LOCKING
1596 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1597 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1598 #endif
1599 	retval = -EAGAIN;
1600 	if (atomic_read(&p->real_cred->user->processes) >=
1601 			task_rlimit(p, RLIMIT_NPROC)) {
1602 		if (p->real_cred->user != INIT_USER &&
1603 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1604 			goto bad_fork_free;
1605 	}
1606 	current->flags &= ~PF_NPROC_EXCEEDED;
1607 
1608 	retval = copy_creds(p, clone_flags);
1609 	if (retval < 0)
1610 		goto bad_fork_free;
1611 
1612 	/*
1613 	 * If multiple threads are within copy_process(), then this check
1614 	 * triggers too late. This doesn't hurt, the check is only there
1615 	 * to stop root fork bombs.
1616 	 */
1617 	retval = -EAGAIN;
1618 	if (nr_threads >= max_threads)
1619 		goto bad_fork_cleanup_count;
1620 
1621 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1622 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1623 	p->flags |= PF_FORKNOEXEC;
1624 	INIT_LIST_HEAD(&p->children);
1625 	INIT_LIST_HEAD(&p->sibling);
1626 	rcu_copy_process(p);
1627 	p->vfork_done = NULL;
1628 	spin_lock_init(&p->alloc_lock);
1629 
1630 	init_sigpending(&p->pending);
1631 
1632 	p->utime = p->stime = p->gtime = 0;
1633 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1634 	p->utimescaled = p->stimescaled = 0;
1635 #endif
1636 	prev_cputime_init(&p->prev_cputime);
1637 
1638 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1639 	seqcount_init(&p->vtime.seqcount);
1640 	p->vtime.starttime = 0;
1641 	p->vtime.state = VTIME_INACTIVE;
1642 #endif
1643 
1644 #if defined(SPLIT_RSS_COUNTING)
1645 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1646 #endif
1647 
1648 	p->default_timer_slack_ns = current->timer_slack_ns;
1649 
1650 	task_io_accounting_init(&p->ioac);
1651 	acct_clear_integrals(p);
1652 
1653 	posix_cpu_timers_init(p);
1654 
1655 	p->start_time = ktime_get_ns();
1656 	p->real_start_time = ktime_get_boot_ns();
1657 	p->io_context = NULL;
1658 	p->audit_context = NULL;
1659 	cgroup_fork(p);
1660 #ifdef CONFIG_NUMA
1661 	p->mempolicy = mpol_dup(p->mempolicy);
1662 	if (IS_ERR(p->mempolicy)) {
1663 		retval = PTR_ERR(p->mempolicy);
1664 		p->mempolicy = NULL;
1665 		goto bad_fork_cleanup_threadgroup_lock;
1666 	}
1667 #endif
1668 #ifdef CONFIG_CPUSETS
1669 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1670 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1671 	seqcount_init(&p->mems_allowed_seq);
1672 #endif
1673 #ifdef CONFIG_TRACE_IRQFLAGS
1674 	p->irq_events = 0;
1675 	p->hardirqs_enabled = 0;
1676 	p->hardirq_enable_ip = 0;
1677 	p->hardirq_enable_event = 0;
1678 	p->hardirq_disable_ip = _THIS_IP_;
1679 	p->hardirq_disable_event = 0;
1680 	p->softirqs_enabled = 1;
1681 	p->softirq_enable_ip = _THIS_IP_;
1682 	p->softirq_enable_event = 0;
1683 	p->softirq_disable_ip = 0;
1684 	p->softirq_disable_event = 0;
1685 	p->hardirq_context = 0;
1686 	p->softirq_context = 0;
1687 #endif
1688 
1689 	p->pagefault_disabled = 0;
1690 
1691 #ifdef CONFIG_LOCKDEP
1692 	p->lockdep_depth = 0; /* no locks held yet */
1693 	p->curr_chain_key = 0;
1694 	p->lockdep_recursion = 0;
1695 #endif
1696 
1697 #ifdef CONFIG_DEBUG_MUTEXES
1698 	p->blocked_on = NULL; /* not blocked yet */
1699 #endif
1700 #ifdef CONFIG_BCACHE
1701 	p->sequential_io	= 0;
1702 	p->sequential_io_avg	= 0;
1703 #endif
1704 
1705 	/* Perform scheduler related setup. Assign this task to a CPU. */
1706 	retval = sched_fork(clone_flags, p);
1707 	if (retval)
1708 		goto bad_fork_cleanup_policy;
1709 
1710 	retval = perf_event_init_task(p);
1711 	if (retval)
1712 		goto bad_fork_cleanup_policy;
1713 	retval = audit_alloc(p);
1714 	if (retval)
1715 		goto bad_fork_cleanup_perf;
1716 	/* copy all the process information */
1717 	shm_init_task(p);
1718 	retval = security_task_alloc(p, clone_flags);
1719 	if (retval)
1720 		goto bad_fork_cleanup_audit;
1721 	retval = copy_semundo(clone_flags, p);
1722 	if (retval)
1723 		goto bad_fork_cleanup_security;
1724 	retval = copy_files(clone_flags, p);
1725 	if (retval)
1726 		goto bad_fork_cleanup_semundo;
1727 	retval = copy_fs(clone_flags, p);
1728 	if (retval)
1729 		goto bad_fork_cleanup_files;
1730 	retval = copy_sighand(clone_flags, p);
1731 	if (retval)
1732 		goto bad_fork_cleanup_fs;
1733 	retval = copy_signal(clone_flags, p);
1734 	if (retval)
1735 		goto bad_fork_cleanup_sighand;
1736 	retval = copy_mm(clone_flags, p);
1737 	if (retval)
1738 		goto bad_fork_cleanup_signal;
1739 	retval = copy_namespaces(clone_flags, p);
1740 	if (retval)
1741 		goto bad_fork_cleanup_mm;
1742 	retval = copy_io(clone_flags, p);
1743 	if (retval)
1744 		goto bad_fork_cleanup_namespaces;
1745 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1746 	if (retval)
1747 		goto bad_fork_cleanup_io;
1748 
1749 	if (pid != &init_struct_pid) {
1750 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1751 		if (IS_ERR(pid)) {
1752 			retval = PTR_ERR(pid);
1753 			goto bad_fork_cleanup_thread;
1754 		}
1755 	}
1756 
1757 #ifdef CONFIG_BLOCK
1758 	p->plug = NULL;
1759 #endif
1760 #ifdef CONFIG_FUTEX
1761 	p->robust_list = NULL;
1762 #ifdef CONFIG_COMPAT
1763 	p->compat_robust_list = NULL;
1764 #endif
1765 	INIT_LIST_HEAD(&p->pi_state_list);
1766 	p->pi_state_cache = NULL;
1767 #endif
1768 	/*
1769 	 * sigaltstack should be cleared when sharing the same VM
1770 	 */
1771 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1772 		sas_ss_reset(p);
1773 
1774 	/*
1775 	 * Syscall tracing and stepping should be turned off in the
1776 	 * child regardless of CLONE_PTRACE.
1777 	 */
1778 	user_disable_single_step(p);
1779 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1780 #ifdef TIF_SYSCALL_EMU
1781 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1782 #endif
1783 	clear_all_latency_tracing(p);
1784 
1785 	/* ok, now we should be set up.. */
1786 	p->pid = pid_nr(pid);
1787 	if (clone_flags & CLONE_THREAD) {
1788 		p->exit_signal = -1;
1789 		p->group_leader = current->group_leader;
1790 		p->tgid = current->tgid;
1791 	} else {
1792 		if (clone_flags & CLONE_PARENT)
1793 			p->exit_signal = current->group_leader->exit_signal;
1794 		else
1795 			p->exit_signal = (clone_flags & CSIGNAL);
1796 		p->group_leader = p;
1797 		p->tgid = p->pid;
1798 	}
1799 
1800 	p->nr_dirtied = 0;
1801 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1802 	p->dirty_paused_when = 0;
1803 
1804 	p->pdeath_signal = 0;
1805 	INIT_LIST_HEAD(&p->thread_group);
1806 	p->task_works = NULL;
1807 
1808 	cgroup_threadgroup_change_begin(current);
1809 	/*
1810 	 * Ensure that the cgroup subsystem policies allow the new process to be
1811 	 * forked. It should be noted the the new process's css_set can be changed
1812 	 * between here and cgroup_post_fork() if an organisation operation is in
1813 	 * progress.
1814 	 */
1815 	retval = cgroup_can_fork(p);
1816 	if (retval)
1817 		goto bad_fork_free_pid;
1818 
1819 	/*
1820 	 * Make it visible to the rest of the system, but dont wake it up yet.
1821 	 * Need tasklist lock for parent etc handling!
1822 	 */
1823 	write_lock_irq(&tasklist_lock);
1824 
1825 	/* CLONE_PARENT re-uses the old parent */
1826 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1827 		p->real_parent = current->real_parent;
1828 		p->parent_exec_id = current->parent_exec_id;
1829 	} else {
1830 		p->real_parent = current;
1831 		p->parent_exec_id = current->self_exec_id;
1832 	}
1833 
1834 	klp_copy_process(p);
1835 
1836 	spin_lock(&current->sighand->siglock);
1837 
1838 	/*
1839 	 * Copy seccomp details explicitly here, in case they were changed
1840 	 * before holding sighand lock.
1841 	 */
1842 	copy_seccomp(p);
1843 
1844 	/*
1845 	 * Process group and session signals need to be delivered to just the
1846 	 * parent before the fork or both the parent and the child after the
1847 	 * fork. Restart if a signal comes in before we add the new process to
1848 	 * it's process group.
1849 	 * A fatal signal pending means that current will exit, so the new
1850 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1851 	*/
1852 	recalc_sigpending();
1853 	if (signal_pending(current)) {
1854 		retval = -ERESTARTNOINTR;
1855 		goto bad_fork_cancel_cgroup;
1856 	}
1857 	if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1858 		retval = -ENOMEM;
1859 		goto bad_fork_cancel_cgroup;
1860 	}
1861 
1862 	if (likely(p->pid)) {
1863 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1864 
1865 		init_task_pid(p, PIDTYPE_PID, pid);
1866 		if (thread_group_leader(p)) {
1867 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1868 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1869 
1870 			if (is_child_reaper(pid)) {
1871 				ns_of_pid(pid)->child_reaper = p;
1872 				p->signal->flags |= SIGNAL_UNKILLABLE;
1873 			}
1874 
1875 			p->signal->leader_pid = pid;
1876 			p->signal->tty = tty_kref_get(current->signal->tty);
1877 			/*
1878 			 * Inherit has_child_subreaper flag under the same
1879 			 * tasklist_lock with adding child to the process tree
1880 			 * for propagate_has_child_subreaper optimization.
1881 			 */
1882 			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1883 							 p->real_parent->signal->is_child_subreaper;
1884 			list_add_tail(&p->sibling, &p->real_parent->children);
1885 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1886 			attach_pid(p, PIDTYPE_PGID);
1887 			attach_pid(p, PIDTYPE_SID);
1888 			__this_cpu_inc(process_counts);
1889 		} else {
1890 			current->signal->nr_threads++;
1891 			atomic_inc(&current->signal->live);
1892 			atomic_inc(&current->signal->sigcnt);
1893 			list_add_tail_rcu(&p->thread_group,
1894 					  &p->group_leader->thread_group);
1895 			list_add_tail_rcu(&p->thread_node,
1896 					  &p->signal->thread_head);
1897 		}
1898 		attach_pid(p, PIDTYPE_PID);
1899 		nr_threads++;
1900 	}
1901 
1902 	total_forks++;
1903 	spin_unlock(&current->sighand->siglock);
1904 	syscall_tracepoint_update(p);
1905 	write_unlock_irq(&tasklist_lock);
1906 
1907 	proc_fork_connector(p);
1908 	cgroup_post_fork(p);
1909 	cgroup_threadgroup_change_end(current);
1910 	perf_event_fork(p);
1911 
1912 	trace_task_newtask(p, clone_flags);
1913 	uprobe_copy_process(p, clone_flags);
1914 
1915 	return p;
1916 
1917 bad_fork_cancel_cgroup:
1918 	spin_unlock(&current->sighand->siglock);
1919 	write_unlock_irq(&tasklist_lock);
1920 	cgroup_cancel_fork(p);
1921 bad_fork_free_pid:
1922 	cgroup_threadgroup_change_end(current);
1923 	if (pid != &init_struct_pid)
1924 		free_pid(pid);
1925 bad_fork_cleanup_thread:
1926 	exit_thread(p);
1927 bad_fork_cleanup_io:
1928 	if (p->io_context)
1929 		exit_io_context(p);
1930 bad_fork_cleanup_namespaces:
1931 	exit_task_namespaces(p);
1932 bad_fork_cleanup_mm:
1933 	if (p->mm)
1934 		mmput(p->mm);
1935 bad_fork_cleanup_signal:
1936 	if (!(clone_flags & CLONE_THREAD))
1937 		free_signal_struct(p->signal);
1938 bad_fork_cleanup_sighand:
1939 	__cleanup_sighand(p->sighand);
1940 bad_fork_cleanup_fs:
1941 	exit_fs(p); /* blocking */
1942 bad_fork_cleanup_files:
1943 	exit_files(p); /* blocking */
1944 bad_fork_cleanup_semundo:
1945 	exit_sem(p);
1946 bad_fork_cleanup_security:
1947 	security_task_free(p);
1948 bad_fork_cleanup_audit:
1949 	audit_free(p);
1950 bad_fork_cleanup_perf:
1951 	perf_event_free_task(p);
1952 bad_fork_cleanup_policy:
1953 #ifdef CONFIG_NUMA
1954 	mpol_put(p->mempolicy);
1955 bad_fork_cleanup_threadgroup_lock:
1956 #endif
1957 	delayacct_tsk_free(p);
1958 bad_fork_cleanup_count:
1959 	atomic_dec(&p->cred->user->processes);
1960 	exit_creds(p);
1961 bad_fork_free:
1962 	p->state = TASK_DEAD;
1963 	put_task_stack(p);
1964 	free_task(p);
1965 fork_out:
1966 	return ERR_PTR(retval);
1967 }
1968 
1969 static inline void init_idle_pids(struct pid_link *links)
1970 {
1971 	enum pid_type type;
1972 
1973 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1974 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1975 		links[type].pid = &init_struct_pid;
1976 	}
1977 }
1978 
1979 struct task_struct *fork_idle(int cpu)
1980 {
1981 	struct task_struct *task;
1982 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1983 			    cpu_to_node(cpu));
1984 	if (!IS_ERR(task)) {
1985 		init_idle_pids(task->pids);
1986 		init_idle(task, cpu);
1987 	}
1988 
1989 	return task;
1990 }
1991 
1992 /*
1993  *  Ok, this is the main fork-routine.
1994  *
1995  * It copies the process, and if successful kick-starts
1996  * it and waits for it to finish using the VM if required.
1997  */
1998 long _do_fork(unsigned long clone_flags,
1999 	      unsigned long stack_start,
2000 	      unsigned long stack_size,
2001 	      int __user *parent_tidptr,
2002 	      int __user *child_tidptr,
2003 	      unsigned long tls)
2004 {
2005 	struct task_struct *p;
2006 	int trace = 0;
2007 	long nr;
2008 
2009 	/*
2010 	 * Determine whether and which event to report to ptracer.  When
2011 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
2012 	 * requested, no event is reported; otherwise, report if the event
2013 	 * for the type of forking is enabled.
2014 	 */
2015 	if (!(clone_flags & CLONE_UNTRACED)) {
2016 		if (clone_flags & CLONE_VFORK)
2017 			trace = PTRACE_EVENT_VFORK;
2018 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
2019 			trace = PTRACE_EVENT_CLONE;
2020 		else
2021 			trace = PTRACE_EVENT_FORK;
2022 
2023 		if (likely(!ptrace_event_enabled(current, trace)))
2024 			trace = 0;
2025 	}
2026 
2027 	p = copy_process(clone_flags, stack_start, stack_size,
2028 			 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2029 	add_latent_entropy();
2030 	/*
2031 	 * Do this prior waking up the new thread - the thread pointer
2032 	 * might get invalid after that point, if the thread exits quickly.
2033 	 */
2034 	if (!IS_ERR(p)) {
2035 		struct completion vfork;
2036 		struct pid *pid;
2037 
2038 		trace_sched_process_fork(current, p);
2039 
2040 		pid = get_task_pid(p, PIDTYPE_PID);
2041 		nr = pid_vnr(pid);
2042 
2043 		if (clone_flags & CLONE_PARENT_SETTID)
2044 			put_user(nr, parent_tidptr);
2045 
2046 		if (clone_flags & CLONE_VFORK) {
2047 			p->vfork_done = &vfork;
2048 			init_completion(&vfork);
2049 			get_task_struct(p);
2050 		}
2051 
2052 		wake_up_new_task(p);
2053 
2054 		/* forking complete and child started to run, tell ptracer */
2055 		if (unlikely(trace))
2056 			ptrace_event_pid(trace, pid);
2057 
2058 		if (clone_flags & CLONE_VFORK) {
2059 			if (!wait_for_vfork_done(p, &vfork))
2060 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2061 		}
2062 
2063 		put_pid(pid);
2064 	} else {
2065 		nr = PTR_ERR(p);
2066 	}
2067 	return nr;
2068 }
2069 
2070 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2071 /* For compatibility with architectures that call do_fork directly rather than
2072  * using the syscall entry points below. */
2073 long do_fork(unsigned long clone_flags,
2074 	      unsigned long stack_start,
2075 	      unsigned long stack_size,
2076 	      int __user *parent_tidptr,
2077 	      int __user *child_tidptr)
2078 {
2079 	return _do_fork(clone_flags, stack_start, stack_size,
2080 			parent_tidptr, child_tidptr, 0);
2081 }
2082 #endif
2083 
2084 /*
2085  * Create a kernel thread.
2086  */
2087 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2088 {
2089 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2090 		(unsigned long)arg, NULL, NULL, 0);
2091 }
2092 
2093 #ifdef __ARCH_WANT_SYS_FORK
2094 SYSCALL_DEFINE0(fork)
2095 {
2096 #ifdef CONFIG_MMU
2097 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2098 #else
2099 	/* can not support in nommu mode */
2100 	return -EINVAL;
2101 #endif
2102 }
2103 #endif
2104 
2105 #ifdef __ARCH_WANT_SYS_VFORK
2106 SYSCALL_DEFINE0(vfork)
2107 {
2108 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2109 			0, NULL, NULL, 0);
2110 }
2111 #endif
2112 
2113 #ifdef __ARCH_WANT_SYS_CLONE
2114 #ifdef CONFIG_CLONE_BACKWARDS
2115 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2116 		 int __user *, parent_tidptr,
2117 		 unsigned long, tls,
2118 		 int __user *, child_tidptr)
2119 #elif defined(CONFIG_CLONE_BACKWARDS2)
2120 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2121 		 int __user *, parent_tidptr,
2122 		 int __user *, child_tidptr,
2123 		 unsigned long, tls)
2124 #elif defined(CONFIG_CLONE_BACKWARDS3)
2125 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2126 		int, stack_size,
2127 		int __user *, parent_tidptr,
2128 		int __user *, child_tidptr,
2129 		unsigned long, tls)
2130 #else
2131 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2132 		 int __user *, parent_tidptr,
2133 		 int __user *, child_tidptr,
2134 		 unsigned long, tls)
2135 #endif
2136 {
2137 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2138 }
2139 #endif
2140 
2141 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2142 {
2143 	struct task_struct *leader, *parent, *child;
2144 	int res;
2145 
2146 	read_lock(&tasklist_lock);
2147 	leader = top = top->group_leader;
2148 down:
2149 	for_each_thread(leader, parent) {
2150 		list_for_each_entry(child, &parent->children, sibling) {
2151 			res = visitor(child, data);
2152 			if (res) {
2153 				if (res < 0)
2154 					goto out;
2155 				leader = child;
2156 				goto down;
2157 			}
2158 up:
2159 			;
2160 		}
2161 	}
2162 
2163 	if (leader != top) {
2164 		child = leader;
2165 		parent = child->real_parent;
2166 		leader = parent->group_leader;
2167 		goto up;
2168 	}
2169 out:
2170 	read_unlock(&tasklist_lock);
2171 }
2172 
2173 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2174 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2175 #endif
2176 
2177 static void sighand_ctor(void *data)
2178 {
2179 	struct sighand_struct *sighand = data;
2180 
2181 	spin_lock_init(&sighand->siglock);
2182 	init_waitqueue_head(&sighand->signalfd_wqh);
2183 }
2184 
2185 void __init proc_caches_init(void)
2186 {
2187 	sighand_cachep = kmem_cache_create("sighand_cache",
2188 			sizeof(struct sighand_struct), 0,
2189 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2190 			SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2191 	signal_cachep = kmem_cache_create("signal_cache",
2192 			sizeof(struct signal_struct), 0,
2193 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2194 			NULL);
2195 	files_cachep = kmem_cache_create("files_cache",
2196 			sizeof(struct files_struct), 0,
2197 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2198 			NULL);
2199 	fs_cachep = kmem_cache_create("fs_cache",
2200 			sizeof(struct fs_struct), 0,
2201 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2202 			NULL);
2203 	/*
2204 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2205 	 * whole struct cpumask for the OFFSTACK case. We could change
2206 	 * this to *only* allocate as much of it as required by the
2207 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
2208 	 * is at the end of the structure, exactly for that reason.
2209 	 */
2210 	mm_cachep = kmem_cache_create("mm_struct",
2211 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2212 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2213 			NULL);
2214 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2215 	mmap_init();
2216 	nsproxy_cache_init();
2217 }
2218 
2219 /*
2220  * Check constraints on flags passed to the unshare system call.
2221  */
2222 static int check_unshare_flags(unsigned long unshare_flags)
2223 {
2224 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2225 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2226 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2227 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2228 		return -EINVAL;
2229 	/*
2230 	 * Not implemented, but pretend it works if there is nothing
2231 	 * to unshare.  Note that unsharing the address space or the
2232 	 * signal handlers also need to unshare the signal queues (aka
2233 	 * CLONE_THREAD).
2234 	 */
2235 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2236 		if (!thread_group_empty(current))
2237 			return -EINVAL;
2238 	}
2239 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2240 		if (atomic_read(&current->sighand->count) > 1)
2241 			return -EINVAL;
2242 	}
2243 	if (unshare_flags & CLONE_VM) {
2244 		if (!current_is_single_threaded())
2245 			return -EINVAL;
2246 	}
2247 
2248 	return 0;
2249 }
2250 
2251 /*
2252  * Unshare the filesystem structure if it is being shared
2253  */
2254 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2255 {
2256 	struct fs_struct *fs = current->fs;
2257 
2258 	if (!(unshare_flags & CLONE_FS) || !fs)
2259 		return 0;
2260 
2261 	/* don't need lock here; in the worst case we'll do useless copy */
2262 	if (fs->users == 1)
2263 		return 0;
2264 
2265 	*new_fsp = copy_fs_struct(fs);
2266 	if (!*new_fsp)
2267 		return -ENOMEM;
2268 
2269 	return 0;
2270 }
2271 
2272 /*
2273  * Unshare file descriptor table if it is being shared
2274  */
2275 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2276 {
2277 	struct files_struct *fd = current->files;
2278 	int error = 0;
2279 
2280 	if ((unshare_flags & CLONE_FILES) &&
2281 	    (fd && atomic_read(&fd->count) > 1)) {
2282 		*new_fdp = dup_fd(fd, &error);
2283 		if (!*new_fdp)
2284 			return error;
2285 	}
2286 
2287 	return 0;
2288 }
2289 
2290 /*
2291  * unshare allows a process to 'unshare' part of the process
2292  * context which was originally shared using clone.  copy_*
2293  * functions used by do_fork() cannot be used here directly
2294  * because they modify an inactive task_struct that is being
2295  * constructed. Here we are modifying the current, active,
2296  * task_struct.
2297  */
2298 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2299 {
2300 	struct fs_struct *fs, *new_fs = NULL;
2301 	struct files_struct *fd, *new_fd = NULL;
2302 	struct cred *new_cred = NULL;
2303 	struct nsproxy *new_nsproxy = NULL;
2304 	int do_sysvsem = 0;
2305 	int err;
2306 
2307 	/*
2308 	 * If unsharing a user namespace must also unshare the thread group
2309 	 * and unshare the filesystem root and working directories.
2310 	 */
2311 	if (unshare_flags & CLONE_NEWUSER)
2312 		unshare_flags |= CLONE_THREAD | CLONE_FS;
2313 	/*
2314 	 * If unsharing vm, must also unshare signal handlers.
2315 	 */
2316 	if (unshare_flags & CLONE_VM)
2317 		unshare_flags |= CLONE_SIGHAND;
2318 	/*
2319 	 * If unsharing a signal handlers, must also unshare the signal queues.
2320 	 */
2321 	if (unshare_flags & CLONE_SIGHAND)
2322 		unshare_flags |= CLONE_THREAD;
2323 	/*
2324 	 * If unsharing namespace, must also unshare filesystem information.
2325 	 */
2326 	if (unshare_flags & CLONE_NEWNS)
2327 		unshare_flags |= CLONE_FS;
2328 
2329 	err = check_unshare_flags(unshare_flags);
2330 	if (err)
2331 		goto bad_unshare_out;
2332 	/*
2333 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2334 	 * to a new ipc namespace, the semaphore arrays from the old
2335 	 * namespace are unreachable.
2336 	 */
2337 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2338 		do_sysvsem = 1;
2339 	err = unshare_fs(unshare_flags, &new_fs);
2340 	if (err)
2341 		goto bad_unshare_out;
2342 	err = unshare_fd(unshare_flags, &new_fd);
2343 	if (err)
2344 		goto bad_unshare_cleanup_fs;
2345 	err = unshare_userns(unshare_flags, &new_cred);
2346 	if (err)
2347 		goto bad_unshare_cleanup_fd;
2348 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2349 					 new_cred, new_fs);
2350 	if (err)
2351 		goto bad_unshare_cleanup_cred;
2352 
2353 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2354 		if (do_sysvsem) {
2355 			/*
2356 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2357 			 */
2358 			exit_sem(current);
2359 		}
2360 		if (unshare_flags & CLONE_NEWIPC) {
2361 			/* Orphan segments in old ns (see sem above). */
2362 			exit_shm(current);
2363 			shm_init_task(current);
2364 		}
2365 
2366 		if (new_nsproxy)
2367 			switch_task_namespaces(current, new_nsproxy);
2368 
2369 		task_lock(current);
2370 
2371 		if (new_fs) {
2372 			fs = current->fs;
2373 			spin_lock(&fs->lock);
2374 			current->fs = new_fs;
2375 			if (--fs->users)
2376 				new_fs = NULL;
2377 			else
2378 				new_fs = fs;
2379 			spin_unlock(&fs->lock);
2380 		}
2381 
2382 		if (new_fd) {
2383 			fd = current->files;
2384 			current->files = new_fd;
2385 			new_fd = fd;
2386 		}
2387 
2388 		task_unlock(current);
2389 
2390 		if (new_cred) {
2391 			/* Install the new user namespace */
2392 			commit_creds(new_cred);
2393 			new_cred = NULL;
2394 		}
2395 	}
2396 
2397 	perf_event_namespaces(current);
2398 
2399 bad_unshare_cleanup_cred:
2400 	if (new_cred)
2401 		put_cred(new_cred);
2402 bad_unshare_cleanup_fd:
2403 	if (new_fd)
2404 		put_files_struct(new_fd);
2405 
2406 bad_unshare_cleanup_fs:
2407 	if (new_fs)
2408 		free_fs_struct(new_fs);
2409 
2410 bad_unshare_out:
2411 	return err;
2412 }
2413 
2414 /*
2415  *	Helper to unshare the files of the current task.
2416  *	We don't want to expose copy_files internals to
2417  *	the exec layer of the kernel.
2418  */
2419 
2420 int unshare_files(struct files_struct **displaced)
2421 {
2422 	struct task_struct *task = current;
2423 	struct files_struct *copy = NULL;
2424 	int error;
2425 
2426 	error = unshare_fd(CLONE_FILES, &copy);
2427 	if (error || !copy) {
2428 		*displaced = NULL;
2429 		return error;
2430 	}
2431 	*displaced = task->files;
2432 	task_lock(task);
2433 	task->files = copy;
2434 	task_unlock(task);
2435 	return 0;
2436 }
2437 
2438 int sysctl_max_threads(struct ctl_table *table, int write,
2439 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2440 {
2441 	struct ctl_table t;
2442 	int ret;
2443 	int threads = max_threads;
2444 	int min = MIN_THREADS;
2445 	int max = MAX_THREADS;
2446 
2447 	t = *table;
2448 	t.data = &threads;
2449 	t.extra1 = &min;
2450 	t.extra2 = &max;
2451 
2452 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2453 	if (ret || !write)
2454 		return ret;
2455 
2456 	set_max_threads(threads);
2457 
2458 	return 0;
2459 }
2460