xref: /linux-6.15/kernel/fork.c (revision 6d20e6b2)
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 void mm_init_uprobes_state(struct mm_struct *mm)
789 {
790 #ifdef CONFIG_UPROBES
791 	mm->uprobes_state.xol_area = NULL;
792 #endif
793 }
794 
795 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
796 	struct user_namespace *user_ns)
797 {
798 	mm->mmap = NULL;
799 	mm->mm_rb = RB_ROOT;
800 	mm->vmacache_seqnum = 0;
801 	atomic_set(&mm->mm_users, 1);
802 	atomic_set(&mm->mm_count, 1);
803 	init_rwsem(&mm->mmap_sem);
804 	INIT_LIST_HEAD(&mm->mmlist);
805 	mm->core_state = NULL;
806 	atomic_long_set(&mm->nr_ptes, 0);
807 	mm_nr_pmds_init(mm);
808 	mm->map_count = 0;
809 	mm->locked_vm = 0;
810 	mm->pinned_vm = 0;
811 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
812 	spin_lock_init(&mm->page_table_lock);
813 	mm_init_cpumask(mm);
814 	mm_init_aio(mm);
815 	mm_init_owner(mm, p);
816 	RCU_INIT_POINTER(mm->exe_file, NULL);
817 	mmu_notifier_mm_init(mm);
818 	init_tlb_flush_pending(mm);
819 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
820 	mm->pmd_huge_pte = NULL;
821 #endif
822 	mm_init_uprobes_state(mm);
823 
824 	if (current->mm) {
825 		mm->flags = current->mm->flags & MMF_INIT_MASK;
826 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
827 	} else {
828 		mm->flags = default_dump_filter;
829 		mm->def_flags = 0;
830 	}
831 
832 	if (mm_alloc_pgd(mm))
833 		goto fail_nopgd;
834 
835 	if (init_new_context(p, mm))
836 		goto fail_nocontext;
837 
838 	mm->user_ns = get_user_ns(user_ns);
839 	return mm;
840 
841 fail_nocontext:
842 	mm_free_pgd(mm);
843 fail_nopgd:
844 	free_mm(mm);
845 	return NULL;
846 }
847 
848 static void check_mm(struct mm_struct *mm)
849 {
850 	int i;
851 
852 	for (i = 0; i < NR_MM_COUNTERS; i++) {
853 		long x = atomic_long_read(&mm->rss_stat.count[i]);
854 
855 		if (unlikely(x))
856 			printk(KERN_ALERT "BUG: Bad rss-counter state "
857 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
858 	}
859 
860 	if (atomic_long_read(&mm->nr_ptes))
861 		pr_alert("BUG: non-zero nr_ptes on freeing mm: %ld\n",
862 				atomic_long_read(&mm->nr_ptes));
863 	if (mm_nr_pmds(mm))
864 		pr_alert("BUG: non-zero nr_pmds on freeing mm: %ld\n",
865 				mm_nr_pmds(mm));
866 
867 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
868 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
869 #endif
870 }
871 
872 /*
873  * Allocate and initialize an mm_struct.
874  */
875 struct mm_struct *mm_alloc(void)
876 {
877 	struct mm_struct *mm;
878 
879 	mm = allocate_mm();
880 	if (!mm)
881 		return NULL;
882 
883 	memset(mm, 0, sizeof(*mm));
884 	return mm_init(mm, current, current_user_ns());
885 }
886 
887 /*
888  * Called when the last reference to the mm
889  * is dropped: either by a lazy thread or by
890  * mmput. Free the page directory and the mm.
891  */
892 void __mmdrop(struct mm_struct *mm)
893 {
894 	BUG_ON(mm == &init_mm);
895 	mm_free_pgd(mm);
896 	destroy_context(mm);
897 	mmu_notifier_mm_destroy(mm);
898 	check_mm(mm);
899 	put_user_ns(mm->user_ns);
900 	free_mm(mm);
901 }
902 EXPORT_SYMBOL_GPL(__mmdrop);
903 
904 static inline void __mmput(struct mm_struct *mm)
905 {
906 	VM_BUG_ON(atomic_read(&mm->mm_users));
907 
908 	uprobe_clear_state(mm);
909 	exit_aio(mm);
910 	ksm_exit(mm);
911 	khugepaged_exit(mm); /* must run before exit_mmap */
912 	exit_mmap(mm);
913 	mm_put_huge_zero_page(mm);
914 	set_mm_exe_file(mm, NULL);
915 	if (!list_empty(&mm->mmlist)) {
916 		spin_lock(&mmlist_lock);
917 		list_del(&mm->mmlist);
918 		spin_unlock(&mmlist_lock);
919 	}
920 	if (mm->binfmt)
921 		module_put(mm->binfmt->module);
922 	set_bit(MMF_OOM_SKIP, &mm->flags);
923 	mmdrop(mm);
924 }
925 
926 /*
927  * Decrement the use count and release all resources for an mm.
928  */
929 void mmput(struct mm_struct *mm)
930 {
931 	might_sleep();
932 
933 	if (atomic_dec_and_test(&mm->mm_users))
934 		__mmput(mm);
935 }
936 EXPORT_SYMBOL_GPL(mmput);
937 
938 #ifdef CONFIG_MMU
939 static void mmput_async_fn(struct work_struct *work)
940 {
941 	struct mm_struct *mm = container_of(work, struct mm_struct, async_put_work);
942 	__mmput(mm);
943 }
944 
945 void mmput_async(struct mm_struct *mm)
946 {
947 	if (atomic_dec_and_test(&mm->mm_users)) {
948 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
949 		schedule_work(&mm->async_put_work);
950 	}
951 }
952 #endif
953 
954 /**
955  * set_mm_exe_file - change a reference to the mm's executable file
956  *
957  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
958  *
959  * Main users are mmput() and sys_execve(). Callers prevent concurrent
960  * invocations: in mmput() nobody alive left, in execve task is single
961  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
962  * mm->exe_file, but does so without using set_mm_exe_file() in order
963  * to do avoid the need for any locks.
964  */
965 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
966 {
967 	struct file *old_exe_file;
968 
969 	/*
970 	 * It is safe to dereference the exe_file without RCU as
971 	 * this function is only called if nobody else can access
972 	 * this mm -- see comment above for justification.
973 	 */
974 	old_exe_file = rcu_dereference_raw(mm->exe_file);
975 
976 	if (new_exe_file)
977 		get_file(new_exe_file);
978 	rcu_assign_pointer(mm->exe_file, new_exe_file);
979 	if (old_exe_file)
980 		fput(old_exe_file);
981 }
982 
983 /**
984  * get_mm_exe_file - acquire a reference to the mm's executable file
985  *
986  * Returns %NULL if mm has no associated executable file.
987  * User must release file via fput().
988  */
989 struct file *get_mm_exe_file(struct mm_struct *mm)
990 {
991 	struct file *exe_file;
992 
993 	rcu_read_lock();
994 	exe_file = rcu_dereference(mm->exe_file);
995 	if (exe_file && !get_file_rcu(exe_file))
996 		exe_file = NULL;
997 	rcu_read_unlock();
998 	return exe_file;
999 }
1000 EXPORT_SYMBOL(get_mm_exe_file);
1001 
1002 /**
1003  * get_task_exe_file - acquire a reference to the task's executable file
1004  *
1005  * Returns %NULL if task's mm (if any) has no associated executable file or
1006  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1007  * User must release file via fput().
1008  */
1009 struct file *get_task_exe_file(struct task_struct *task)
1010 {
1011 	struct file *exe_file = NULL;
1012 	struct mm_struct *mm;
1013 
1014 	task_lock(task);
1015 	mm = task->mm;
1016 	if (mm) {
1017 		if (!(task->flags & PF_KTHREAD))
1018 			exe_file = get_mm_exe_file(mm);
1019 	}
1020 	task_unlock(task);
1021 	return exe_file;
1022 }
1023 EXPORT_SYMBOL(get_task_exe_file);
1024 
1025 /**
1026  * get_task_mm - acquire a reference to the task's mm
1027  *
1028  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1029  * this kernel workthread has transiently adopted a user mm with use_mm,
1030  * to do its AIO) is not set and if so returns a reference to it, after
1031  * bumping up the use count.  User must release the mm via mmput()
1032  * after use.  Typically used by /proc and ptrace.
1033  */
1034 struct mm_struct *get_task_mm(struct task_struct *task)
1035 {
1036 	struct mm_struct *mm;
1037 
1038 	task_lock(task);
1039 	mm = task->mm;
1040 	if (mm) {
1041 		if (task->flags & PF_KTHREAD)
1042 			mm = NULL;
1043 		else
1044 			mmget(mm);
1045 	}
1046 	task_unlock(task);
1047 	return mm;
1048 }
1049 EXPORT_SYMBOL_GPL(get_task_mm);
1050 
1051 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1052 {
1053 	struct mm_struct *mm;
1054 	int err;
1055 
1056 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1057 	if (err)
1058 		return ERR_PTR(err);
1059 
1060 	mm = get_task_mm(task);
1061 	if (mm && mm != current->mm &&
1062 			!ptrace_may_access(task, mode)) {
1063 		mmput(mm);
1064 		mm = ERR_PTR(-EACCES);
1065 	}
1066 	mutex_unlock(&task->signal->cred_guard_mutex);
1067 
1068 	return mm;
1069 }
1070 
1071 static void complete_vfork_done(struct task_struct *tsk)
1072 {
1073 	struct completion *vfork;
1074 
1075 	task_lock(tsk);
1076 	vfork = tsk->vfork_done;
1077 	if (likely(vfork)) {
1078 		tsk->vfork_done = NULL;
1079 		complete(vfork);
1080 	}
1081 	task_unlock(tsk);
1082 }
1083 
1084 static int wait_for_vfork_done(struct task_struct *child,
1085 				struct completion *vfork)
1086 {
1087 	int killed;
1088 
1089 	freezer_do_not_count();
1090 	killed = wait_for_completion_killable(vfork);
1091 	freezer_count();
1092 
1093 	if (killed) {
1094 		task_lock(child);
1095 		child->vfork_done = NULL;
1096 		task_unlock(child);
1097 	}
1098 
1099 	put_task_struct(child);
1100 	return killed;
1101 }
1102 
1103 /* Please note the differences between mmput and mm_release.
1104  * mmput is called whenever we stop holding onto a mm_struct,
1105  * error success whatever.
1106  *
1107  * mm_release is called after a mm_struct has been removed
1108  * from the current process.
1109  *
1110  * This difference is important for error handling, when we
1111  * only half set up a mm_struct for a new process and need to restore
1112  * the old one.  Because we mmput the new mm_struct before
1113  * restoring the old one. . .
1114  * Eric Biederman 10 January 1998
1115  */
1116 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1117 {
1118 	/* Get rid of any futexes when releasing the mm */
1119 #ifdef CONFIG_FUTEX
1120 	if (unlikely(tsk->robust_list)) {
1121 		exit_robust_list(tsk);
1122 		tsk->robust_list = NULL;
1123 	}
1124 #ifdef CONFIG_COMPAT
1125 	if (unlikely(tsk->compat_robust_list)) {
1126 		compat_exit_robust_list(tsk);
1127 		tsk->compat_robust_list = NULL;
1128 	}
1129 #endif
1130 	if (unlikely(!list_empty(&tsk->pi_state_list)))
1131 		exit_pi_state_list(tsk);
1132 #endif
1133 
1134 	uprobe_free_utask(tsk);
1135 
1136 	/* Get rid of any cached register state */
1137 	deactivate_mm(tsk, mm);
1138 
1139 	/*
1140 	 * Signal userspace if we're not exiting with a core dump
1141 	 * because we want to leave the value intact for debugging
1142 	 * purposes.
1143 	 */
1144 	if (tsk->clear_child_tid) {
1145 		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1146 		    atomic_read(&mm->mm_users) > 1) {
1147 			/*
1148 			 * We don't check the error code - if userspace has
1149 			 * not set up a proper pointer then tough luck.
1150 			 */
1151 			put_user(0, tsk->clear_child_tid);
1152 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1153 					1, NULL, NULL, 0);
1154 		}
1155 		tsk->clear_child_tid = NULL;
1156 	}
1157 
1158 	/*
1159 	 * All done, finally we can wake up parent and return this mm to him.
1160 	 * Also kthread_stop() uses this completion for synchronization.
1161 	 */
1162 	if (tsk->vfork_done)
1163 		complete_vfork_done(tsk);
1164 }
1165 
1166 /*
1167  * Allocate a new mm structure and copy contents from the
1168  * mm structure of the passed in task structure.
1169  */
1170 static struct mm_struct *dup_mm(struct task_struct *tsk)
1171 {
1172 	struct mm_struct *mm, *oldmm = current->mm;
1173 	int err;
1174 
1175 	mm = allocate_mm();
1176 	if (!mm)
1177 		goto fail_nomem;
1178 
1179 	memcpy(mm, oldmm, sizeof(*mm));
1180 
1181 	if (!mm_init(mm, tsk, mm->user_ns))
1182 		goto fail_nomem;
1183 
1184 	err = dup_mmap(mm, oldmm);
1185 	if (err)
1186 		goto free_pt;
1187 
1188 	mm->hiwater_rss = get_mm_rss(mm);
1189 	mm->hiwater_vm = mm->total_vm;
1190 
1191 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1192 		goto free_pt;
1193 
1194 	return mm;
1195 
1196 free_pt:
1197 	/* don't put binfmt in mmput, we haven't got module yet */
1198 	mm->binfmt = NULL;
1199 	mmput(mm);
1200 
1201 fail_nomem:
1202 	return NULL;
1203 }
1204 
1205 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1206 {
1207 	struct mm_struct *mm, *oldmm;
1208 	int retval;
1209 
1210 	tsk->min_flt = tsk->maj_flt = 0;
1211 	tsk->nvcsw = tsk->nivcsw = 0;
1212 #ifdef CONFIG_DETECT_HUNG_TASK
1213 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1214 #endif
1215 
1216 	tsk->mm = NULL;
1217 	tsk->active_mm = NULL;
1218 
1219 	/*
1220 	 * Are we cloning a kernel thread?
1221 	 *
1222 	 * We need to steal a active VM for that..
1223 	 */
1224 	oldmm = current->mm;
1225 	if (!oldmm)
1226 		return 0;
1227 
1228 	/* initialize the new vmacache entries */
1229 	vmacache_flush(tsk);
1230 
1231 	if (clone_flags & CLONE_VM) {
1232 		mmget(oldmm);
1233 		mm = oldmm;
1234 		goto good_mm;
1235 	}
1236 
1237 	retval = -ENOMEM;
1238 	mm = dup_mm(tsk);
1239 	if (!mm)
1240 		goto fail_nomem;
1241 
1242 good_mm:
1243 	tsk->mm = mm;
1244 	tsk->active_mm = mm;
1245 	return 0;
1246 
1247 fail_nomem:
1248 	return retval;
1249 }
1250 
1251 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1252 {
1253 	struct fs_struct *fs = current->fs;
1254 	if (clone_flags & CLONE_FS) {
1255 		/* tsk->fs is already what we want */
1256 		spin_lock(&fs->lock);
1257 		if (fs->in_exec) {
1258 			spin_unlock(&fs->lock);
1259 			return -EAGAIN;
1260 		}
1261 		fs->users++;
1262 		spin_unlock(&fs->lock);
1263 		return 0;
1264 	}
1265 	tsk->fs = copy_fs_struct(fs);
1266 	if (!tsk->fs)
1267 		return -ENOMEM;
1268 	return 0;
1269 }
1270 
1271 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1272 {
1273 	struct files_struct *oldf, *newf;
1274 	int error = 0;
1275 
1276 	/*
1277 	 * A background process may not have any files ...
1278 	 */
1279 	oldf = current->files;
1280 	if (!oldf)
1281 		goto out;
1282 
1283 	if (clone_flags & CLONE_FILES) {
1284 		atomic_inc(&oldf->count);
1285 		goto out;
1286 	}
1287 
1288 	newf = dup_fd(oldf, &error);
1289 	if (!newf)
1290 		goto out;
1291 
1292 	tsk->files = newf;
1293 	error = 0;
1294 out:
1295 	return error;
1296 }
1297 
1298 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1299 {
1300 #ifdef CONFIG_BLOCK
1301 	struct io_context *ioc = current->io_context;
1302 	struct io_context *new_ioc;
1303 
1304 	if (!ioc)
1305 		return 0;
1306 	/*
1307 	 * Share io context with parent, if CLONE_IO is set
1308 	 */
1309 	if (clone_flags & CLONE_IO) {
1310 		ioc_task_link(ioc);
1311 		tsk->io_context = ioc;
1312 	} else if (ioprio_valid(ioc->ioprio)) {
1313 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1314 		if (unlikely(!new_ioc))
1315 			return -ENOMEM;
1316 
1317 		new_ioc->ioprio = ioc->ioprio;
1318 		put_io_context(new_ioc);
1319 	}
1320 #endif
1321 	return 0;
1322 }
1323 
1324 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1325 {
1326 	struct sighand_struct *sig;
1327 
1328 	if (clone_flags & CLONE_SIGHAND) {
1329 		atomic_inc(&current->sighand->count);
1330 		return 0;
1331 	}
1332 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1333 	rcu_assign_pointer(tsk->sighand, sig);
1334 	if (!sig)
1335 		return -ENOMEM;
1336 
1337 	atomic_set(&sig->count, 1);
1338 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1339 	return 0;
1340 }
1341 
1342 void __cleanup_sighand(struct sighand_struct *sighand)
1343 {
1344 	if (atomic_dec_and_test(&sighand->count)) {
1345 		signalfd_cleanup(sighand);
1346 		/*
1347 		 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1348 		 * without an RCU grace period, see __lock_task_sighand().
1349 		 */
1350 		kmem_cache_free(sighand_cachep, sighand);
1351 	}
1352 }
1353 
1354 #ifdef CONFIG_POSIX_TIMERS
1355 /*
1356  * Initialize POSIX timer handling for a thread group.
1357  */
1358 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1359 {
1360 	unsigned long cpu_limit;
1361 
1362 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1363 	if (cpu_limit != RLIM_INFINITY) {
1364 		sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1365 		sig->cputimer.running = true;
1366 	}
1367 
1368 	/* The timer lists. */
1369 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1370 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1371 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1372 }
1373 #else
1374 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1375 #endif
1376 
1377 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1378 {
1379 	struct signal_struct *sig;
1380 
1381 	if (clone_flags & CLONE_THREAD)
1382 		return 0;
1383 
1384 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1385 	tsk->signal = sig;
1386 	if (!sig)
1387 		return -ENOMEM;
1388 
1389 	sig->nr_threads = 1;
1390 	atomic_set(&sig->live, 1);
1391 	atomic_set(&sig->sigcnt, 1);
1392 
1393 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1394 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1395 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1396 
1397 	init_waitqueue_head(&sig->wait_chldexit);
1398 	sig->curr_target = tsk;
1399 	init_sigpending(&sig->shared_pending);
1400 	seqlock_init(&sig->stats_lock);
1401 	prev_cputime_init(&sig->prev_cputime);
1402 
1403 #ifdef CONFIG_POSIX_TIMERS
1404 	INIT_LIST_HEAD(&sig->posix_timers);
1405 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1406 	sig->real_timer.function = it_real_fn;
1407 #endif
1408 
1409 	task_lock(current->group_leader);
1410 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1411 	task_unlock(current->group_leader);
1412 
1413 	posix_cpu_timers_init_group(sig);
1414 
1415 	tty_audit_fork(sig);
1416 	sched_autogroup_fork(sig);
1417 
1418 	sig->oom_score_adj = current->signal->oom_score_adj;
1419 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1420 
1421 	mutex_init(&sig->cred_guard_mutex);
1422 
1423 	return 0;
1424 }
1425 
1426 static void copy_seccomp(struct task_struct *p)
1427 {
1428 #ifdef CONFIG_SECCOMP
1429 	/*
1430 	 * Must be called with sighand->lock held, which is common to
1431 	 * all threads in the group. Holding cred_guard_mutex is not
1432 	 * needed because this new task is not yet running and cannot
1433 	 * be racing exec.
1434 	 */
1435 	assert_spin_locked(&current->sighand->siglock);
1436 
1437 	/* Ref-count the new filter user, and assign it. */
1438 	get_seccomp_filter(current);
1439 	p->seccomp = current->seccomp;
1440 
1441 	/*
1442 	 * Explicitly enable no_new_privs here in case it got set
1443 	 * between the task_struct being duplicated and holding the
1444 	 * sighand lock. The seccomp state and nnp must be in sync.
1445 	 */
1446 	if (task_no_new_privs(current))
1447 		task_set_no_new_privs(p);
1448 
1449 	/*
1450 	 * If the parent gained a seccomp mode after copying thread
1451 	 * flags and between before we held the sighand lock, we have
1452 	 * to manually enable the seccomp thread flag here.
1453 	 */
1454 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1455 		set_tsk_thread_flag(p, TIF_SECCOMP);
1456 #endif
1457 }
1458 
1459 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1460 {
1461 	current->clear_child_tid = tidptr;
1462 
1463 	return task_pid_vnr(current);
1464 }
1465 
1466 static void rt_mutex_init_task(struct task_struct *p)
1467 {
1468 	raw_spin_lock_init(&p->pi_lock);
1469 #ifdef CONFIG_RT_MUTEXES
1470 	p->pi_waiters = RB_ROOT;
1471 	p->pi_waiters_leftmost = NULL;
1472 	p->pi_top_task = NULL;
1473 	p->pi_blocked_on = NULL;
1474 #endif
1475 }
1476 
1477 #ifdef CONFIG_POSIX_TIMERS
1478 /*
1479  * Initialize POSIX timer handling for a single task.
1480  */
1481 static void posix_cpu_timers_init(struct task_struct *tsk)
1482 {
1483 	tsk->cputime_expires.prof_exp = 0;
1484 	tsk->cputime_expires.virt_exp = 0;
1485 	tsk->cputime_expires.sched_exp = 0;
1486 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1487 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1488 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1489 }
1490 #else
1491 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1492 #endif
1493 
1494 static inline void
1495 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1496 {
1497 	 task->pids[type].pid = pid;
1498 }
1499 
1500 static inline void rcu_copy_process(struct task_struct *p)
1501 {
1502 #ifdef CONFIG_PREEMPT_RCU
1503 	p->rcu_read_lock_nesting = 0;
1504 	p->rcu_read_unlock_special.s = 0;
1505 	p->rcu_blocked_node = NULL;
1506 	INIT_LIST_HEAD(&p->rcu_node_entry);
1507 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1508 #ifdef CONFIG_TASKS_RCU
1509 	p->rcu_tasks_holdout = false;
1510 	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1511 	p->rcu_tasks_idle_cpu = -1;
1512 #endif /* #ifdef CONFIG_TASKS_RCU */
1513 }
1514 
1515 /*
1516  * This creates a new process as a copy of the old one,
1517  * but does not actually start it yet.
1518  *
1519  * It copies the registers, and all the appropriate
1520  * parts of the process environment (as per the clone
1521  * flags). The actual kick-off is left to the caller.
1522  */
1523 static __latent_entropy struct task_struct *copy_process(
1524 					unsigned long clone_flags,
1525 					unsigned long stack_start,
1526 					unsigned long stack_size,
1527 					int __user *child_tidptr,
1528 					struct pid *pid,
1529 					int trace,
1530 					unsigned long tls,
1531 					int node)
1532 {
1533 	int retval;
1534 	struct task_struct *p;
1535 
1536 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1537 		return ERR_PTR(-EINVAL);
1538 
1539 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1540 		return ERR_PTR(-EINVAL);
1541 
1542 	/*
1543 	 * Thread groups must share signals as well, and detached threads
1544 	 * can only be started up within the thread group.
1545 	 */
1546 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1547 		return ERR_PTR(-EINVAL);
1548 
1549 	/*
1550 	 * Shared signal handlers imply shared VM. By way of the above,
1551 	 * thread groups also imply shared VM. Blocking this case allows
1552 	 * for various simplifications in other code.
1553 	 */
1554 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1555 		return ERR_PTR(-EINVAL);
1556 
1557 	/*
1558 	 * Siblings of global init remain as zombies on exit since they are
1559 	 * not reaped by their parent (swapper). To solve this and to avoid
1560 	 * multi-rooted process trees, prevent global and container-inits
1561 	 * from creating siblings.
1562 	 */
1563 	if ((clone_flags & CLONE_PARENT) &&
1564 				current->signal->flags & SIGNAL_UNKILLABLE)
1565 		return ERR_PTR(-EINVAL);
1566 
1567 	/*
1568 	 * If the new process will be in a different pid or user namespace
1569 	 * do not allow it to share a thread group with the forking task.
1570 	 */
1571 	if (clone_flags & CLONE_THREAD) {
1572 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1573 		    (task_active_pid_ns(current) !=
1574 				current->nsproxy->pid_ns_for_children))
1575 			return ERR_PTR(-EINVAL);
1576 	}
1577 
1578 	retval = security_task_create(clone_flags);
1579 	if (retval)
1580 		goto fork_out;
1581 
1582 	retval = -ENOMEM;
1583 	p = dup_task_struct(current, node);
1584 	if (!p)
1585 		goto fork_out;
1586 
1587 	/*
1588 	 * This _must_ happen before we call free_task(), i.e. before we jump
1589 	 * to any of the bad_fork_* labels. This is to avoid freeing
1590 	 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1591 	 * kernel threads (PF_KTHREAD).
1592 	 */
1593 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1594 	/*
1595 	 * Clear TID on mm_release()?
1596 	 */
1597 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1598 
1599 	ftrace_graph_init_task(p);
1600 
1601 	rt_mutex_init_task(p);
1602 
1603 #ifdef CONFIG_PROVE_LOCKING
1604 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1605 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1606 #endif
1607 	retval = -EAGAIN;
1608 	if (atomic_read(&p->real_cred->user->processes) >=
1609 			task_rlimit(p, RLIMIT_NPROC)) {
1610 		if (p->real_cred->user != INIT_USER &&
1611 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1612 			goto bad_fork_free;
1613 	}
1614 	current->flags &= ~PF_NPROC_EXCEEDED;
1615 
1616 	retval = copy_creds(p, clone_flags);
1617 	if (retval < 0)
1618 		goto bad_fork_free;
1619 
1620 	/*
1621 	 * If multiple threads are within copy_process(), then this check
1622 	 * triggers too late. This doesn't hurt, the check is only there
1623 	 * to stop root fork bombs.
1624 	 */
1625 	retval = -EAGAIN;
1626 	if (nr_threads >= max_threads)
1627 		goto bad_fork_cleanup_count;
1628 
1629 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1630 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1631 	p->flags |= PF_FORKNOEXEC;
1632 	INIT_LIST_HEAD(&p->children);
1633 	INIT_LIST_HEAD(&p->sibling);
1634 	rcu_copy_process(p);
1635 	p->vfork_done = NULL;
1636 	spin_lock_init(&p->alloc_lock);
1637 
1638 	init_sigpending(&p->pending);
1639 
1640 	p->utime = p->stime = p->gtime = 0;
1641 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1642 	p->utimescaled = p->stimescaled = 0;
1643 #endif
1644 	prev_cputime_init(&p->prev_cputime);
1645 
1646 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1647 	seqcount_init(&p->vtime.seqcount);
1648 	p->vtime.starttime = 0;
1649 	p->vtime.state = VTIME_INACTIVE;
1650 #endif
1651 
1652 #if defined(SPLIT_RSS_COUNTING)
1653 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1654 #endif
1655 
1656 	p->default_timer_slack_ns = current->timer_slack_ns;
1657 
1658 	task_io_accounting_init(&p->ioac);
1659 	acct_clear_integrals(p);
1660 
1661 	posix_cpu_timers_init(p);
1662 
1663 	p->start_time = ktime_get_ns();
1664 	p->real_start_time = ktime_get_boot_ns();
1665 	p->io_context = NULL;
1666 	p->audit_context = NULL;
1667 	cgroup_fork(p);
1668 #ifdef CONFIG_NUMA
1669 	p->mempolicy = mpol_dup(p->mempolicy);
1670 	if (IS_ERR(p->mempolicy)) {
1671 		retval = PTR_ERR(p->mempolicy);
1672 		p->mempolicy = NULL;
1673 		goto bad_fork_cleanup_threadgroup_lock;
1674 	}
1675 #endif
1676 #ifdef CONFIG_CPUSETS
1677 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1678 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1679 	seqcount_init(&p->mems_allowed_seq);
1680 #endif
1681 #ifdef CONFIG_TRACE_IRQFLAGS
1682 	p->irq_events = 0;
1683 	p->hardirqs_enabled = 0;
1684 	p->hardirq_enable_ip = 0;
1685 	p->hardirq_enable_event = 0;
1686 	p->hardirq_disable_ip = _THIS_IP_;
1687 	p->hardirq_disable_event = 0;
1688 	p->softirqs_enabled = 1;
1689 	p->softirq_enable_ip = _THIS_IP_;
1690 	p->softirq_enable_event = 0;
1691 	p->softirq_disable_ip = 0;
1692 	p->softirq_disable_event = 0;
1693 	p->hardirq_context = 0;
1694 	p->softirq_context = 0;
1695 #endif
1696 
1697 	p->pagefault_disabled = 0;
1698 
1699 #ifdef CONFIG_LOCKDEP
1700 	p->lockdep_depth = 0; /* no locks held yet */
1701 	p->curr_chain_key = 0;
1702 	p->lockdep_recursion = 0;
1703 #endif
1704 
1705 #ifdef CONFIG_DEBUG_MUTEXES
1706 	p->blocked_on = NULL; /* not blocked yet */
1707 #endif
1708 #ifdef CONFIG_BCACHE
1709 	p->sequential_io	= 0;
1710 	p->sequential_io_avg	= 0;
1711 #endif
1712 
1713 	/* Perform scheduler related setup. Assign this task to a CPU. */
1714 	retval = sched_fork(clone_flags, p);
1715 	if (retval)
1716 		goto bad_fork_cleanup_policy;
1717 
1718 	retval = perf_event_init_task(p);
1719 	if (retval)
1720 		goto bad_fork_cleanup_policy;
1721 	retval = audit_alloc(p);
1722 	if (retval)
1723 		goto bad_fork_cleanup_perf;
1724 	/* copy all the process information */
1725 	shm_init_task(p);
1726 	retval = security_task_alloc(p, clone_flags);
1727 	if (retval)
1728 		goto bad_fork_cleanup_audit;
1729 	retval = copy_semundo(clone_flags, p);
1730 	if (retval)
1731 		goto bad_fork_cleanup_security;
1732 	retval = copy_files(clone_flags, p);
1733 	if (retval)
1734 		goto bad_fork_cleanup_semundo;
1735 	retval = copy_fs(clone_flags, p);
1736 	if (retval)
1737 		goto bad_fork_cleanup_files;
1738 	retval = copy_sighand(clone_flags, p);
1739 	if (retval)
1740 		goto bad_fork_cleanup_fs;
1741 	retval = copy_signal(clone_flags, p);
1742 	if (retval)
1743 		goto bad_fork_cleanup_sighand;
1744 	retval = copy_mm(clone_flags, p);
1745 	if (retval)
1746 		goto bad_fork_cleanup_signal;
1747 	retval = copy_namespaces(clone_flags, p);
1748 	if (retval)
1749 		goto bad_fork_cleanup_mm;
1750 	retval = copy_io(clone_flags, p);
1751 	if (retval)
1752 		goto bad_fork_cleanup_namespaces;
1753 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1754 	if (retval)
1755 		goto bad_fork_cleanup_io;
1756 
1757 	if (pid != &init_struct_pid) {
1758 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1759 		if (IS_ERR(pid)) {
1760 			retval = PTR_ERR(pid);
1761 			goto bad_fork_cleanup_thread;
1762 		}
1763 	}
1764 
1765 #ifdef CONFIG_BLOCK
1766 	p->plug = NULL;
1767 #endif
1768 #ifdef CONFIG_FUTEX
1769 	p->robust_list = NULL;
1770 #ifdef CONFIG_COMPAT
1771 	p->compat_robust_list = NULL;
1772 #endif
1773 	INIT_LIST_HEAD(&p->pi_state_list);
1774 	p->pi_state_cache = NULL;
1775 #endif
1776 	/*
1777 	 * sigaltstack should be cleared when sharing the same VM
1778 	 */
1779 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1780 		sas_ss_reset(p);
1781 
1782 	/*
1783 	 * Syscall tracing and stepping should be turned off in the
1784 	 * child regardless of CLONE_PTRACE.
1785 	 */
1786 	user_disable_single_step(p);
1787 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1788 #ifdef TIF_SYSCALL_EMU
1789 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1790 #endif
1791 	clear_all_latency_tracing(p);
1792 
1793 	/* ok, now we should be set up.. */
1794 	p->pid = pid_nr(pid);
1795 	if (clone_flags & CLONE_THREAD) {
1796 		p->exit_signal = -1;
1797 		p->group_leader = current->group_leader;
1798 		p->tgid = current->tgid;
1799 	} else {
1800 		if (clone_flags & CLONE_PARENT)
1801 			p->exit_signal = current->group_leader->exit_signal;
1802 		else
1803 			p->exit_signal = (clone_flags & CSIGNAL);
1804 		p->group_leader = p;
1805 		p->tgid = p->pid;
1806 	}
1807 
1808 	p->nr_dirtied = 0;
1809 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1810 	p->dirty_paused_when = 0;
1811 
1812 	p->pdeath_signal = 0;
1813 	INIT_LIST_HEAD(&p->thread_group);
1814 	p->task_works = NULL;
1815 
1816 	cgroup_threadgroup_change_begin(current);
1817 	/*
1818 	 * Ensure that the cgroup subsystem policies allow the new process to be
1819 	 * forked. It should be noted the the new process's css_set can be changed
1820 	 * between here and cgroup_post_fork() if an organisation operation is in
1821 	 * progress.
1822 	 */
1823 	retval = cgroup_can_fork(p);
1824 	if (retval)
1825 		goto bad_fork_free_pid;
1826 
1827 	/*
1828 	 * Make it visible to the rest of the system, but dont wake it up yet.
1829 	 * Need tasklist lock for parent etc handling!
1830 	 */
1831 	write_lock_irq(&tasklist_lock);
1832 
1833 	/* CLONE_PARENT re-uses the old parent */
1834 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1835 		p->real_parent = current->real_parent;
1836 		p->parent_exec_id = current->parent_exec_id;
1837 	} else {
1838 		p->real_parent = current;
1839 		p->parent_exec_id = current->self_exec_id;
1840 	}
1841 
1842 	klp_copy_process(p);
1843 
1844 	spin_lock(&current->sighand->siglock);
1845 
1846 	/*
1847 	 * Copy seccomp details explicitly here, in case they were changed
1848 	 * before holding sighand lock.
1849 	 */
1850 	copy_seccomp(p);
1851 
1852 	/*
1853 	 * Process group and session signals need to be delivered to just the
1854 	 * parent before the fork or both the parent and the child after the
1855 	 * fork. Restart if a signal comes in before we add the new process to
1856 	 * it's process group.
1857 	 * A fatal signal pending means that current will exit, so the new
1858 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1859 	*/
1860 	recalc_sigpending();
1861 	if (signal_pending(current)) {
1862 		retval = -ERESTARTNOINTR;
1863 		goto bad_fork_cancel_cgroup;
1864 	}
1865 	if (unlikely(!(ns_of_pid(pid)->nr_hashed & PIDNS_HASH_ADDING))) {
1866 		retval = -ENOMEM;
1867 		goto bad_fork_cancel_cgroup;
1868 	}
1869 
1870 	if (likely(p->pid)) {
1871 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1872 
1873 		init_task_pid(p, PIDTYPE_PID, pid);
1874 		if (thread_group_leader(p)) {
1875 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1876 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1877 
1878 			if (is_child_reaper(pid)) {
1879 				ns_of_pid(pid)->child_reaper = p;
1880 				p->signal->flags |= SIGNAL_UNKILLABLE;
1881 			}
1882 
1883 			p->signal->leader_pid = pid;
1884 			p->signal->tty = tty_kref_get(current->signal->tty);
1885 			/*
1886 			 * Inherit has_child_subreaper flag under the same
1887 			 * tasklist_lock with adding child to the process tree
1888 			 * for propagate_has_child_subreaper optimization.
1889 			 */
1890 			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1891 							 p->real_parent->signal->is_child_subreaper;
1892 			list_add_tail(&p->sibling, &p->real_parent->children);
1893 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1894 			attach_pid(p, PIDTYPE_PGID);
1895 			attach_pid(p, PIDTYPE_SID);
1896 			__this_cpu_inc(process_counts);
1897 		} else {
1898 			current->signal->nr_threads++;
1899 			atomic_inc(&current->signal->live);
1900 			atomic_inc(&current->signal->sigcnt);
1901 			list_add_tail_rcu(&p->thread_group,
1902 					  &p->group_leader->thread_group);
1903 			list_add_tail_rcu(&p->thread_node,
1904 					  &p->signal->thread_head);
1905 		}
1906 		attach_pid(p, PIDTYPE_PID);
1907 		nr_threads++;
1908 	}
1909 
1910 	total_forks++;
1911 	spin_unlock(&current->sighand->siglock);
1912 	syscall_tracepoint_update(p);
1913 	write_unlock_irq(&tasklist_lock);
1914 
1915 	proc_fork_connector(p);
1916 	cgroup_post_fork(p);
1917 	cgroup_threadgroup_change_end(current);
1918 	perf_event_fork(p);
1919 
1920 	trace_task_newtask(p, clone_flags);
1921 	uprobe_copy_process(p, clone_flags);
1922 
1923 	return p;
1924 
1925 bad_fork_cancel_cgroup:
1926 	spin_unlock(&current->sighand->siglock);
1927 	write_unlock_irq(&tasklist_lock);
1928 	cgroup_cancel_fork(p);
1929 bad_fork_free_pid:
1930 	cgroup_threadgroup_change_end(current);
1931 	if (pid != &init_struct_pid)
1932 		free_pid(pid);
1933 bad_fork_cleanup_thread:
1934 	exit_thread(p);
1935 bad_fork_cleanup_io:
1936 	if (p->io_context)
1937 		exit_io_context(p);
1938 bad_fork_cleanup_namespaces:
1939 	exit_task_namespaces(p);
1940 bad_fork_cleanup_mm:
1941 	if (p->mm)
1942 		mmput(p->mm);
1943 bad_fork_cleanup_signal:
1944 	if (!(clone_flags & CLONE_THREAD))
1945 		free_signal_struct(p->signal);
1946 bad_fork_cleanup_sighand:
1947 	__cleanup_sighand(p->sighand);
1948 bad_fork_cleanup_fs:
1949 	exit_fs(p); /* blocking */
1950 bad_fork_cleanup_files:
1951 	exit_files(p); /* blocking */
1952 bad_fork_cleanup_semundo:
1953 	exit_sem(p);
1954 bad_fork_cleanup_security:
1955 	security_task_free(p);
1956 bad_fork_cleanup_audit:
1957 	audit_free(p);
1958 bad_fork_cleanup_perf:
1959 	perf_event_free_task(p);
1960 bad_fork_cleanup_policy:
1961 #ifdef CONFIG_NUMA
1962 	mpol_put(p->mempolicy);
1963 bad_fork_cleanup_threadgroup_lock:
1964 #endif
1965 	delayacct_tsk_free(p);
1966 bad_fork_cleanup_count:
1967 	atomic_dec(&p->cred->user->processes);
1968 	exit_creds(p);
1969 bad_fork_free:
1970 	p->state = TASK_DEAD;
1971 	put_task_stack(p);
1972 	free_task(p);
1973 fork_out:
1974 	return ERR_PTR(retval);
1975 }
1976 
1977 static inline void init_idle_pids(struct pid_link *links)
1978 {
1979 	enum pid_type type;
1980 
1981 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
1982 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
1983 		links[type].pid = &init_struct_pid;
1984 	}
1985 }
1986 
1987 struct task_struct *fork_idle(int cpu)
1988 {
1989 	struct task_struct *task;
1990 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
1991 			    cpu_to_node(cpu));
1992 	if (!IS_ERR(task)) {
1993 		init_idle_pids(task->pids);
1994 		init_idle(task, cpu);
1995 	}
1996 
1997 	return task;
1998 }
1999 
2000 /*
2001  *  Ok, this is the main fork-routine.
2002  *
2003  * It copies the process, and if successful kick-starts
2004  * it and waits for it to finish using the VM if required.
2005  */
2006 long _do_fork(unsigned long clone_flags,
2007 	      unsigned long stack_start,
2008 	      unsigned long stack_size,
2009 	      int __user *parent_tidptr,
2010 	      int __user *child_tidptr,
2011 	      unsigned long tls)
2012 {
2013 	struct task_struct *p;
2014 	int trace = 0;
2015 	long nr;
2016 
2017 	/*
2018 	 * Determine whether and which event to report to ptracer.  When
2019 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
2020 	 * requested, no event is reported; otherwise, report if the event
2021 	 * for the type of forking is enabled.
2022 	 */
2023 	if (!(clone_flags & CLONE_UNTRACED)) {
2024 		if (clone_flags & CLONE_VFORK)
2025 			trace = PTRACE_EVENT_VFORK;
2026 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
2027 			trace = PTRACE_EVENT_CLONE;
2028 		else
2029 			trace = PTRACE_EVENT_FORK;
2030 
2031 		if (likely(!ptrace_event_enabled(current, trace)))
2032 			trace = 0;
2033 	}
2034 
2035 	p = copy_process(clone_flags, stack_start, stack_size,
2036 			 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2037 	add_latent_entropy();
2038 	/*
2039 	 * Do this prior waking up the new thread - the thread pointer
2040 	 * might get invalid after that point, if the thread exits quickly.
2041 	 */
2042 	if (!IS_ERR(p)) {
2043 		struct completion vfork;
2044 		struct pid *pid;
2045 
2046 		trace_sched_process_fork(current, p);
2047 
2048 		pid = get_task_pid(p, PIDTYPE_PID);
2049 		nr = pid_vnr(pid);
2050 
2051 		if (clone_flags & CLONE_PARENT_SETTID)
2052 			put_user(nr, parent_tidptr);
2053 
2054 		if (clone_flags & CLONE_VFORK) {
2055 			p->vfork_done = &vfork;
2056 			init_completion(&vfork);
2057 			get_task_struct(p);
2058 		}
2059 
2060 		wake_up_new_task(p);
2061 
2062 		/* forking complete and child started to run, tell ptracer */
2063 		if (unlikely(trace))
2064 			ptrace_event_pid(trace, pid);
2065 
2066 		if (clone_flags & CLONE_VFORK) {
2067 			if (!wait_for_vfork_done(p, &vfork))
2068 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2069 		}
2070 
2071 		put_pid(pid);
2072 	} else {
2073 		nr = PTR_ERR(p);
2074 	}
2075 	return nr;
2076 }
2077 
2078 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2079 /* For compatibility with architectures that call do_fork directly rather than
2080  * using the syscall entry points below. */
2081 long do_fork(unsigned long clone_flags,
2082 	      unsigned long stack_start,
2083 	      unsigned long stack_size,
2084 	      int __user *parent_tidptr,
2085 	      int __user *child_tidptr)
2086 {
2087 	return _do_fork(clone_flags, stack_start, stack_size,
2088 			parent_tidptr, child_tidptr, 0);
2089 }
2090 #endif
2091 
2092 /*
2093  * Create a kernel thread.
2094  */
2095 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2096 {
2097 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2098 		(unsigned long)arg, NULL, NULL, 0);
2099 }
2100 
2101 #ifdef __ARCH_WANT_SYS_FORK
2102 SYSCALL_DEFINE0(fork)
2103 {
2104 #ifdef CONFIG_MMU
2105 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2106 #else
2107 	/* can not support in nommu mode */
2108 	return -EINVAL;
2109 #endif
2110 }
2111 #endif
2112 
2113 #ifdef __ARCH_WANT_SYS_VFORK
2114 SYSCALL_DEFINE0(vfork)
2115 {
2116 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2117 			0, NULL, NULL, 0);
2118 }
2119 #endif
2120 
2121 #ifdef __ARCH_WANT_SYS_CLONE
2122 #ifdef CONFIG_CLONE_BACKWARDS
2123 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2124 		 int __user *, parent_tidptr,
2125 		 unsigned long, tls,
2126 		 int __user *, child_tidptr)
2127 #elif defined(CONFIG_CLONE_BACKWARDS2)
2128 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2129 		 int __user *, parent_tidptr,
2130 		 int __user *, child_tidptr,
2131 		 unsigned long, tls)
2132 #elif defined(CONFIG_CLONE_BACKWARDS3)
2133 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2134 		int, stack_size,
2135 		int __user *, parent_tidptr,
2136 		int __user *, child_tidptr,
2137 		unsigned long, tls)
2138 #else
2139 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2140 		 int __user *, parent_tidptr,
2141 		 int __user *, child_tidptr,
2142 		 unsigned long, tls)
2143 #endif
2144 {
2145 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2146 }
2147 #endif
2148 
2149 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2150 {
2151 	struct task_struct *leader, *parent, *child;
2152 	int res;
2153 
2154 	read_lock(&tasklist_lock);
2155 	leader = top = top->group_leader;
2156 down:
2157 	for_each_thread(leader, parent) {
2158 		list_for_each_entry(child, &parent->children, sibling) {
2159 			res = visitor(child, data);
2160 			if (res) {
2161 				if (res < 0)
2162 					goto out;
2163 				leader = child;
2164 				goto down;
2165 			}
2166 up:
2167 			;
2168 		}
2169 	}
2170 
2171 	if (leader != top) {
2172 		child = leader;
2173 		parent = child->real_parent;
2174 		leader = parent->group_leader;
2175 		goto up;
2176 	}
2177 out:
2178 	read_unlock(&tasklist_lock);
2179 }
2180 
2181 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2182 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2183 #endif
2184 
2185 static void sighand_ctor(void *data)
2186 {
2187 	struct sighand_struct *sighand = data;
2188 
2189 	spin_lock_init(&sighand->siglock);
2190 	init_waitqueue_head(&sighand->signalfd_wqh);
2191 }
2192 
2193 void __init proc_caches_init(void)
2194 {
2195 	sighand_cachep = kmem_cache_create("sighand_cache",
2196 			sizeof(struct sighand_struct), 0,
2197 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2198 			SLAB_NOTRACK|SLAB_ACCOUNT, sighand_ctor);
2199 	signal_cachep = kmem_cache_create("signal_cache",
2200 			sizeof(struct signal_struct), 0,
2201 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2202 			NULL);
2203 	files_cachep = kmem_cache_create("files_cache",
2204 			sizeof(struct files_struct), 0,
2205 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2206 			NULL);
2207 	fs_cachep = kmem_cache_create("fs_cache",
2208 			sizeof(struct fs_struct), 0,
2209 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2210 			NULL);
2211 	/*
2212 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2213 	 * whole struct cpumask for the OFFSTACK case. We could change
2214 	 * this to *only* allocate as much of it as required by the
2215 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
2216 	 * is at the end of the structure, exactly for that reason.
2217 	 */
2218 	mm_cachep = kmem_cache_create("mm_struct",
2219 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2220 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK|SLAB_ACCOUNT,
2221 			NULL);
2222 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2223 	mmap_init();
2224 	nsproxy_cache_init();
2225 }
2226 
2227 /*
2228  * Check constraints on flags passed to the unshare system call.
2229  */
2230 static int check_unshare_flags(unsigned long unshare_flags)
2231 {
2232 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2233 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2234 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2235 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2236 		return -EINVAL;
2237 	/*
2238 	 * Not implemented, but pretend it works if there is nothing
2239 	 * to unshare.  Note that unsharing the address space or the
2240 	 * signal handlers also need to unshare the signal queues (aka
2241 	 * CLONE_THREAD).
2242 	 */
2243 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2244 		if (!thread_group_empty(current))
2245 			return -EINVAL;
2246 	}
2247 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2248 		if (atomic_read(&current->sighand->count) > 1)
2249 			return -EINVAL;
2250 	}
2251 	if (unshare_flags & CLONE_VM) {
2252 		if (!current_is_single_threaded())
2253 			return -EINVAL;
2254 	}
2255 
2256 	return 0;
2257 }
2258 
2259 /*
2260  * Unshare the filesystem structure if it is being shared
2261  */
2262 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2263 {
2264 	struct fs_struct *fs = current->fs;
2265 
2266 	if (!(unshare_flags & CLONE_FS) || !fs)
2267 		return 0;
2268 
2269 	/* don't need lock here; in the worst case we'll do useless copy */
2270 	if (fs->users == 1)
2271 		return 0;
2272 
2273 	*new_fsp = copy_fs_struct(fs);
2274 	if (!*new_fsp)
2275 		return -ENOMEM;
2276 
2277 	return 0;
2278 }
2279 
2280 /*
2281  * Unshare file descriptor table if it is being shared
2282  */
2283 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2284 {
2285 	struct files_struct *fd = current->files;
2286 	int error = 0;
2287 
2288 	if ((unshare_flags & CLONE_FILES) &&
2289 	    (fd && atomic_read(&fd->count) > 1)) {
2290 		*new_fdp = dup_fd(fd, &error);
2291 		if (!*new_fdp)
2292 			return error;
2293 	}
2294 
2295 	return 0;
2296 }
2297 
2298 /*
2299  * unshare allows a process to 'unshare' part of the process
2300  * context which was originally shared using clone.  copy_*
2301  * functions used by do_fork() cannot be used here directly
2302  * because they modify an inactive task_struct that is being
2303  * constructed. Here we are modifying the current, active,
2304  * task_struct.
2305  */
2306 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2307 {
2308 	struct fs_struct *fs, *new_fs = NULL;
2309 	struct files_struct *fd, *new_fd = NULL;
2310 	struct cred *new_cred = NULL;
2311 	struct nsproxy *new_nsproxy = NULL;
2312 	int do_sysvsem = 0;
2313 	int err;
2314 
2315 	/*
2316 	 * If unsharing a user namespace must also unshare the thread group
2317 	 * and unshare the filesystem root and working directories.
2318 	 */
2319 	if (unshare_flags & CLONE_NEWUSER)
2320 		unshare_flags |= CLONE_THREAD | CLONE_FS;
2321 	/*
2322 	 * If unsharing vm, must also unshare signal handlers.
2323 	 */
2324 	if (unshare_flags & CLONE_VM)
2325 		unshare_flags |= CLONE_SIGHAND;
2326 	/*
2327 	 * If unsharing a signal handlers, must also unshare the signal queues.
2328 	 */
2329 	if (unshare_flags & CLONE_SIGHAND)
2330 		unshare_flags |= CLONE_THREAD;
2331 	/*
2332 	 * If unsharing namespace, must also unshare filesystem information.
2333 	 */
2334 	if (unshare_flags & CLONE_NEWNS)
2335 		unshare_flags |= CLONE_FS;
2336 
2337 	err = check_unshare_flags(unshare_flags);
2338 	if (err)
2339 		goto bad_unshare_out;
2340 	/*
2341 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2342 	 * to a new ipc namespace, the semaphore arrays from the old
2343 	 * namespace are unreachable.
2344 	 */
2345 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2346 		do_sysvsem = 1;
2347 	err = unshare_fs(unshare_flags, &new_fs);
2348 	if (err)
2349 		goto bad_unshare_out;
2350 	err = unshare_fd(unshare_flags, &new_fd);
2351 	if (err)
2352 		goto bad_unshare_cleanup_fs;
2353 	err = unshare_userns(unshare_flags, &new_cred);
2354 	if (err)
2355 		goto bad_unshare_cleanup_fd;
2356 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2357 					 new_cred, new_fs);
2358 	if (err)
2359 		goto bad_unshare_cleanup_cred;
2360 
2361 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2362 		if (do_sysvsem) {
2363 			/*
2364 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2365 			 */
2366 			exit_sem(current);
2367 		}
2368 		if (unshare_flags & CLONE_NEWIPC) {
2369 			/* Orphan segments in old ns (see sem above). */
2370 			exit_shm(current);
2371 			shm_init_task(current);
2372 		}
2373 
2374 		if (new_nsproxy)
2375 			switch_task_namespaces(current, new_nsproxy);
2376 
2377 		task_lock(current);
2378 
2379 		if (new_fs) {
2380 			fs = current->fs;
2381 			spin_lock(&fs->lock);
2382 			current->fs = new_fs;
2383 			if (--fs->users)
2384 				new_fs = NULL;
2385 			else
2386 				new_fs = fs;
2387 			spin_unlock(&fs->lock);
2388 		}
2389 
2390 		if (new_fd) {
2391 			fd = current->files;
2392 			current->files = new_fd;
2393 			new_fd = fd;
2394 		}
2395 
2396 		task_unlock(current);
2397 
2398 		if (new_cred) {
2399 			/* Install the new user namespace */
2400 			commit_creds(new_cred);
2401 			new_cred = NULL;
2402 		}
2403 	}
2404 
2405 	perf_event_namespaces(current);
2406 
2407 bad_unshare_cleanup_cred:
2408 	if (new_cred)
2409 		put_cred(new_cred);
2410 bad_unshare_cleanup_fd:
2411 	if (new_fd)
2412 		put_files_struct(new_fd);
2413 
2414 bad_unshare_cleanup_fs:
2415 	if (new_fs)
2416 		free_fs_struct(new_fs);
2417 
2418 bad_unshare_out:
2419 	return err;
2420 }
2421 
2422 /*
2423  *	Helper to unshare the files of the current task.
2424  *	We don't want to expose copy_files internals to
2425  *	the exec layer of the kernel.
2426  */
2427 
2428 int unshare_files(struct files_struct **displaced)
2429 {
2430 	struct task_struct *task = current;
2431 	struct files_struct *copy = NULL;
2432 	int error;
2433 
2434 	error = unshare_fd(CLONE_FILES, &copy);
2435 	if (error || !copy) {
2436 		*displaced = NULL;
2437 		return error;
2438 	}
2439 	*displaced = task->files;
2440 	task_lock(task);
2441 	task->files = copy;
2442 	task_unlock(task);
2443 	return 0;
2444 }
2445 
2446 int sysctl_max_threads(struct ctl_table *table, int write,
2447 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2448 {
2449 	struct ctl_table t;
2450 	int ret;
2451 	int threads = max_threads;
2452 	int min = MIN_THREADS;
2453 	int max = MAX_THREADS;
2454 
2455 	t = *table;
2456 	t.data = &threads;
2457 	t.extra1 = &min;
2458 	t.extra2 = &max;
2459 
2460 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2461 	if (ret || !write)
2462 		return ret;
2463 
2464 	set_max_threads(threads);
2465 
2466 	return 0;
2467 }
2468