xref: /linux/kernel/fork.c (revision ef694e2d210111cc08c52fd25da9ab130170bfa3)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/kernel/fork.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 /*
9  *  'fork.c' contains the help-routines for the 'fork' system call
10  * (see also entry.S and others).
11  * Fork is rather simple, once you get the hang of it, but the memory
12  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
13  */
14 
15 #include <linux/anon_inodes.h>
16 #include <linux/slab.h>
17 #include <linux/sched/autogroup.h>
18 #include <linux/sched/mm.h>
19 #include <linux/sched/user.h>
20 #include <linux/sched/numa_balancing.h>
21 #include <linux/sched/stat.h>
22 #include <linux/sched/task.h>
23 #include <linux/sched/task_stack.h>
24 #include <linux/sched/cputime.h>
25 #include <linux/sched/ext.h>
26 #include <linux/sched/exec_state.h>
27 #include <linux/seq_file.h>
28 #include <linux/rtmutex.h>
29 #include <linux/init.h>
30 #include <linux/unistd.h>
31 #include <linux/module.h>
32 #include <linux/vmalloc.h>
33 #include <linux/completion.h>
34 #include <linux/personality.h>
35 #include <linux/mempolicy.h>
36 #include <linux/sem.h>
37 #include <linux/file.h>
38 #include <linux/fdtable.h>
39 #include <linux/iocontext.h>
40 #include <linux/key.h>
41 #include <linux/kmsan.h>
42 #include <linux/binfmts.h>
43 #include <linux/mman.h>
44 #include <linux/mmu_notifier.h>
45 #include <linux/fs.h>
46 #include <linux/mm.h>
47 #include <linux/mm_inline.h>
48 #include <linux/memblock.h>
49 #include <linux/nsproxy.h>
50 #include <linux/ns/ns_common_types.h>
51 #include <linux/capability.h>
52 #include <linux/cpu.h>
53 #include <linux/cgroup.h>
54 #include <linux/security.h>
55 #include <linux/hugetlb.h>
56 #include <linux/seccomp.h>
57 #include <linux/swap.h>
58 #include <linux/syscalls.h>
59 #include <linux/syscall_user_dispatch.h>
60 #include <linux/jiffies.h>
61 #include <linux/futex.h>
62 #include <linux/compat.h>
63 #include <linux/kthread.h>
64 #include <linux/task_io_accounting_ops.h>
65 #include <linux/rcupdate.h>
66 #include <linux/ptrace.h>
67 #include <linux/mount.h>
68 #include <linux/audit.h>
69 #include <linux/memcontrol.h>
70 #include <linux/ftrace.h>
71 #include <linux/proc_fs.h>
72 #include <linux/profile.h>
73 #include <linux/rmap.h>
74 #include <linux/ksm.h>
75 #include <linux/acct.h>
76 #include <linux/userfaultfd_k.h>
77 #include <linux/tsacct_kern.h>
78 #include <linux/cn_proc.h>
79 #include <linux/freezer.h>
80 #include <linux/delayacct.h>
81 #include <linux/taskstats_kern.h>
82 #include <linux/tty.h>
83 #include <linux/fs_struct.h>
84 #include <linux/magic.h>
85 #include <linux/perf_event.h>
86 #include <linux/posix-timers.h>
87 #include <linux/user-return-notifier.h>
88 #include <linux/oom.h>
89 #include <linux/khugepaged.h>
90 #include <linux/signalfd.h>
91 #include <linux/uprobes.h>
92 #include <linux/aio.h>
93 #include <linux/compiler.h>
94 #include <linux/sysctl.h>
95 #include <linux/kcov.h>
96 #include <linux/livepatch.h>
97 #include <linux/thread_info.h>
98 #include <linux/kstack_erase.h>
99 #include <linux/kasan.h>
100 #include <linux/randomize_kstack.h>
101 #include <linux/scs.h>
102 #include <linux/io_uring.h>
103 #include <linux/io_uring_types.h>
104 #include <linux/bpf.h>
105 #include <linux/stackprotector.h>
106 #include <linux/user_events.h>
107 #include <linux/iommu.h>
108 #include <linux/rseq.h>
109 #include <uapi/linux/pidfd.h>
110 #include <linux/pidfs.h>
111 #include <linux/tick.h>
112 #include <linux/unwind_deferred.h>
113 #include <linux/pgalloc.h>
114 #include <linux/uaccess.h>
115 
116 #include <asm/mmu_context.h>
117 #include <asm/cacheflush.h>
118 #include <asm/tlbflush.h>
119 
120 /* For dup_mmap(). */
121 #include "../mm/internal.h"
122 
123 #include <trace/events/sched.h>
124 
125 #define CREATE_TRACE_POINTS
126 #include <trace/events/task.h>
127 
128 #include <kunit/visibility.h>
129 
130 /*
131  * Minimum number of threads to boot the kernel
132  */
133 #define MIN_THREADS 20
134 
135 /*
136  * Maximum number of threads
137  */
138 #define MAX_THREADS FUTEX_TID_MASK
139 
140 /*
141  * Protected counters by write_lock_irq(&tasklist_lock)
142  */
143 unsigned long total_forks;	/* Handle normal Linux uptimes. */
144 int nr_threads;			/* The idle threads do not count.. */
145 
146 static int max_threads __read_mostly;		/* tunable limit on nr_threads */
147 
148 #define NAMED_ARRAY_INDEX(x)	[x] = __stringify(x)
149 
150 static const char * const resident_page_types[] = {
151 	NAMED_ARRAY_INDEX(MM_FILEPAGES),
152 	NAMED_ARRAY_INDEX(MM_ANONPAGES),
153 	NAMED_ARRAY_INDEX(MM_SWAPENTS),
154 	NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
155 };
156 
157 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
158 
159 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
160 
161 #ifdef CONFIG_PROVE_RCU
162 int lockdep_tasklist_lock_is_held(void)
163 {
164 	return lockdep_is_held(&tasklist_lock);
165 }
166 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
167 #endif /* #ifdef CONFIG_PROVE_RCU */
168 
169 int nr_processes(void)
170 {
171 	int cpu;
172 	int total = 0;
173 
174 	for_each_possible_cpu(cpu)
175 		total += per_cpu(process_counts, cpu);
176 
177 	return total;
178 }
179 
180 void __weak arch_release_task_struct(struct task_struct *tsk)
181 {
182 }
183 
184 static struct kmem_cache *task_struct_cachep;
185 
186 static inline struct task_struct *alloc_task_struct_node(int node)
187 {
188 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
189 }
190 
191 static inline void free_task_struct(struct task_struct *tsk)
192 {
193 	kmem_cache_free(task_struct_cachep, tsk);
194 }
195 
196 #ifdef CONFIG_VMAP_STACK
197 /*
198  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
199  * flush.  Try to minimize the number of calls by caching stacks.
200  */
201 #define NR_CACHED_STACKS 2
202 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
203 /*
204  * Allocated stacks are cached and later reused by new threads, so memcg
205  * accounting is performed by the code assigning/releasing stacks to tasks.
206  * We need a zeroed memory without __GFP_ACCOUNT.
207  */
208 #define GFP_VMAP_STACK (GFP_KERNEL | __GFP_ZERO | __GFP_SKIP_KASAN)
209 
210 struct vm_stack {
211 	struct rcu_head rcu;
212 	struct vm_struct *stack_vm_area;
213 };
214 
215 static struct vm_struct *alloc_thread_stack_node_from_cache(struct task_struct *tsk, int node)
216 {
217 	struct vm_struct *vm_area;
218 	unsigned int i;
219 
220 	/*
221 	 * If the node has memory, we are guaranteed the stacks are backed by local pages.
222 	 * Otherwise the pages are arbitrary.
223 	 *
224 	 * Note that depending on cpuset it is possible we will get migrated to a different
225 	 * node immediately after allocating here, so this does *not* guarantee locality for
226 	 * arbitrary callers.
227 	 */
228 	scoped_guard(preempt) {
229 		if (node != NUMA_NO_NODE && numa_node_id() != node)
230 			return NULL;
231 
232 		for (i = 0; i < NR_CACHED_STACKS; i++) {
233 			vm_area = this_cpu_xchg(cached_stacks[i], NULL);
234 			if (vm_area)
235 				return vm_area;
236 		}
237 	}
238 
239 	return NULL;
240 }
241 
242 static bool try_release_thread_stack_to_cache(struct vm_struct *vm_area)
243 {
244 	unsigned int i;
245 	int nid;
246 
247 	/*
248 	 * Don't cache stacks if any of the pages don't match the local domain, unless
249 	 * there is no local memory to begin with.
250 	 *
251 	 * Note that lack of local memory does not automatically mean it makes no difference
252 	 * performance-wise which other domain backs the stack. In this case we are merely
253 	 * trying to avoid constantly going to vmalloc.
254 	 */
255 	scoped_guard(preempt) {
256 		nid = numa_node_id();
257 		if (node_state(nid, N_MEMORY)) {
258 			for (i = 0; i < vm_area->nr_pages; i++) {
259 				struct page *page = vm_area->pages[i];
260 				if (page_to_nid(page) != nid)
261 					return false;
262 			}
263 		}
264 
265 		for (i = 0; i < NR_CACHED_STACKS; i++) {
266 			struct vm_struct *tmp = NULL;
267 
268 			if (this_cpu_try_cmpxchg(cached_stacks[i], &tmp, vm_area))
269 				return true;
270 		}
271 	}
272 	return false;
273 }
274 
275 static void thread_stack_free_rcu(struct rcu_head *rh)
276 {
277 	struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
278 	struct vm_struct *vm_area = vm_stack->stack_vm_area;
279 
280 	if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
281 		return;
282 
283 	vfree(vm_area->addr);
284 }
285 
286 static void thread_stack_delayed_free(struct task_struct *tsk)
287 {
288 	struct vm_stack *vm_stack = tsk->stack;
289 
290 	vm_stack->stack_vm_area = tsk->stack_vm_area;
291 	call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
292 }
293 
294 static int free_vm_stack_cache(unsigned int cpu)
295 {
296 	struct vm_struct **cached_vm_stack_areas = per_cpu_ptr(cached_stacks, cpu);
297 	int i;
298 
299 	for (i = 0; i < NR_CACHED_STACKS; i++) {
300 		struct vm_struct *vm_area = cached_vm_stack_areas[i];
301 
302 		if (!vm_area)
303 			continue;
304 
305 		vfree(vm_area->addr);
306 		cached_vm_stack_areas[i] = NULL;
307 	}
308 
309 	return 0;
310 }
311 
312 static int memcg_charge_kernel_stack(struct vm_struct *vm_area)
313 {
314 	int i;
315 	int ret;
316 	int nr_charged = 0;
317 
318 	BUG_ON(vm_area->nr_pages != THREAD_SIZE / PAGE_SIZE);
319 
320 	for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
321 		ret = memcg_kmem_charge_page(vm_area->pages[i], GFP_KERNEL, 0);
322 		if (ret)
323 			goto err;
324 		nr_charged++;
325 	}
326 	return 0;
327 err:
328 	for (i = 0; i < nr_charged; i++)
329 		memcg_kmem_uncharge_page(vm_area->pages[i], 0);
330 	return ret;
331 }
332 
333 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
334 {
335 	struct vm_struct *vm_area;
336 	void *stack;
337 
338 	vm_area = alloc_thread_stack_node_from_cache(tsk, node);
339 	if (vm_area) {
340 		if (memcg_charge_kernel_stack(vm_area)) {
341 			vfree(vm_area->addr);
342 			return -ENOMEM;
343 		}
344 
345 		/* Reset stack metadata. */
346 		if (!kasan_hw_tags_enabled())
347 			kasan_unpoison_range(vm_area->addr, THREAD_SIZE);
348 
349 		stack = kasan_reset_tag(vm_area->addr);
350 
351 		/* Clear stale pointers from reused stack. */
352 		clear_pages(vm_area->addr, vm_area->nr_pages);
353 
354 		tsk->stack_vm_area = vm_area;
355 		tsk->stack = stack;
356 		return 0;
357 	}
358 
359 	stack = __vmalloc_node(THREAD_SIZE, THREAD_ALIGN,
360 				     GFP_VMAP_STACK,
361 				     node, __builtin_return_address(0));
362 	if (!stack)
363 		return -ENOMEM;
364 
365 	vm_area = find_vm_area(stack);
366 	if (memcg_charge_kernel_stack(vm_area)) {
367 		vfree(stack);
368 		return -ENOMEM;
369 	}
370 	/*
371 	 * We can't call find_vm_area() in interrupt context, and
372 	 * free_thread_stack() can be called in interrupt context,
373 	 * so cache the vm_struct.
374 	 */
375 	tsk->stack_vm_area = vm_area;
376 	stack = kasan_reset_tag(stack);
377 	tsk->stack = stack;
378 	return 0;
379 }
380 
381 static void free_thread_stack(struct task_struct *tsk)
382 {
383 	if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
384 		thread_stack_delayed_free(tsk);
385 
386 	tsk->stack = NULL;
387 	tsk->stack_vm_area = NULL;
388 }
389 
390 #else /* !CONFIG_VMAP_STACK */
391 
392 /*
393  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
394  * kmemcache based allocator.
395  */
396 #if THREAD_SIZE >= PAGE_SIZE
397 
398 static void thread_stack_free_rcu(struct rcu_head *rh)
399 {
400 	__free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
401 }
402 
403 static void thread_stack_delayed_free(struct task_struct *tsk)
404 {
405 	struct rcu_head *rh = tsk->stack;
406 
407 	call_rcu(rh, thread_stack_free_rcu);
408 }
409 
410 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
411 {
412 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
413 					     THREAD_SIZE_ORDER);
414 
415 	if (likely(page)) {
416 		tsk->stack = kasan_reset_tag(page_address(page));
417 		return 0;
418 	}
419 	return -ENOMEM;
420 }
421 
422 static void free_thread_stack(struct task_struct *tsk)
423 {
424 	thread_stack_delayed_free(tsk);
425 	tsk->stack = NULL;
426 }
427 
428 #else /* !(THREAD_SIZE >= PAGE_SIZE) */
429 
430 static struct kmem_cache *thread_stack_cache;
431 
432 static void thread_stack_free_rcu(struct rcu_head *rh)
433 {
434 	kmem_cache_free(thread_stack_cache, rh);
435 }
436 
437 static void thread_stack_delayed_free(struct task_struct *tsk)
438 {
439 	struct rcu_head *rh = tsk->stack;
440 
441 	call_rcu(rh, thread_stack_free_rcu);
442 }
443 
444 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
445 {
446 	unsigned long *stack;
447 	stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
448 	stack = kasan_reset_tag(stack);
449 	tsk->stack = stack;
450 	return stack ? 0 : -ENOMEM;
451 }
452 
453 static void free_thread_stack(struct task_struct *tsk)
454 {
455 	thread_stack_delayed_free(tsk);
456 	tsk->stack = NULL;
457 }
458 
459 void thread_stack_cache_init(void)
460 {
461 	thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
462 					THREAD_SIZE, THREAD_SIZE, 0, 0,
463 					THREAD_SIZE, NULL);
464 	BUG_ON(thread_stack_cache == NULL);
465 }
466 
467 #endif /* THREAD_SIZE >= PAGE_SIZE */
468 #endif /* CONFIG_VMAP_STACK */
469 
470 /* SLAB cache for signal_struct structures (tsk->signal) */
471 static struct kmem_cache *signal_cachep;
472 
473 /* SLAB cache for sighand_struct structures (tsk->sighand) */
474 struct kmem_cache *sighand_cachep;
475 
476 /* SLAB cache for files_struct structures (tsk->files) */
477 struct kmem_cache *files_cachep;
478 
479 /* SLAB cache for fs_struct structures (tsk->fs) */
480 struct kmem_cache *fs_cachep;
481 
482 /* SLAB cache for mm_struct structures (tsk->mm) */
483 static struct kmem_cache *mm_cachep;
484 
485 static void account_kernel_stack(struct task_struct *tsk, int account)
486 {
487 	if (IS_ENABLED(CONFIG_VMAP_STACK)) {
488 		struct vm_struct *vm_area = task_stack_vm_area(tsk);
489 		int i;
490 
491 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
492 			mod_lruvec_page_state(vm_area->pages[i], NR_KERNEL_STACK_KB,
493 					      account * (PAGE_SIZE / 1024));
494 	} else {
495 		void *stack = task_stack_page(tsk);
496 
497 		/* All stack pages are in the same node. */
498 		mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
499 				      account * (THREAD_SIZE / 1024));
500 	}
501 }
502 
503 void exit_task_stack_account(struct task_struct *tsk)
504 {
505 	account_kernel_stack(tsk, -1);
506 
507 	if (IS_ENABLED(CONFIG_VMAP_STACK)) {
508 		struct vm_struct *vm_area;
509 		int i;
510 
511 		vm_area = task_stack_vm_area(tsk);
512 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
513 			memcg_kmem_uncharge_page(vm_area->pages[i], 0);
514 	}
515 }
516 
517 static void release_task_stack(struct task_struct *tsk)
518 {
519 	if (WARN_ON(READ_ONCE(tsk->__state) != TASK_DEAD))
520 		return;  /* Better to leak the stack than to free prematurely */
521 
522 	free_thread_stack(tsk);
523 }
524 
525 #ifdef CONFIG_THREAD_INFO_IN_TASK
526 void put_task_stack(struct task_struct *tsk)
527 {
528 	if (refcount_dec_and_test(&tsk->stack_refcount))
529 		release_task_stack(tsk);
530 }
531 #endif
532 
533 void free_task(struct task_struct *tsk)
534 {
535 #ifdef CONFIG_SECCOMP
536 	WARN_ON_ONCE(tsk->seccomp.filter);
537 #endif
538 	release_user_cpus_ptr(tsk);
539 	scs_release(tsk);
540 	smp_task_ipi_mask_free(tsk);
541 
542 #ifndef CONFIG_THREAD_INFO_IN_TASK
543 	/*
544 	 * The task is finally done with both the stack and thread_info,
545 	 * so free both.
546 	 */
547 	release_task_stack(tsk);
548 #else
549 	/*
550 	 * If the task had a separate stack allocation, it should be gone
551 	 * by now.
552 	 */
553 	WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
554 #endif
555 	rt_mutex_debug_task_free(tsk);
556 	ftrace_graph_exit_task(tsk);
557 	arch_release_task_struct(tsk);
558 	if (tsk->flags & PF_KTHREAD)
559 		free_kthread_struct(tsk);
560 	bpf_task_storage_free(tsk);
561 	put_task_exec_state(rcu_access_pointer(tsk->exec_state));
562 	free_task_struct(tsk);
563 }
564 EXPORT_SYMBOL(free_task);
565 
566 void dup_mm_exe_file(struct mm_struct *mm, struct mm_struct *oldmm)
567 {
568 	struct file *exe_file;
569 
570 	exe_file = get_mm_exe_file(oldmm);
571 	RCU_INIT_POINTER(mm->exe_file, exe_file);
572 	/*
573 	 * We depend on the oldmm having properly denied write access to the
574 	 * exe_file already.
575 	 */
576 	if (exe_file && exe_file_deny_write_access(exe_file))
577 		pr_warn_once("exe_file_deny_write_access() failed in %s\n", __func__);
578 }
579 
580 #ifdef CONFIG_MMU
581 static inline int mm_alloc_pgd(struct mm_struct *mm)
582 {
583 	mm->pgd = pgd_alloc(mm);
584 	if (unlikely(!mm->pgd))
585 		return -ENOMEM;
586 	return 0;
587 }
588 
589 static inline void mm_free_pgd(struct mm_struct *mm)
590 {
591 	pgd_free(mm, mm->pgd);
592 }
593 #else
594 #define mm_alloc_pgd(mm)	(0)
595 #define mm_free_pgd(mm)
596 #endif /* CONFIG_MMU */
597 
598 #ifdef CONFIG_MM_ID
599 static DEFINE_IDA(mm_ida);
600 
601 static inline int mm_alloc_id(struct mm_struct *mm)
602 {
603 	int ret;
604 
605 	ret = ida_alloc_range(&mm_ida, MM_ID_MIN, MM_ID_MAX, GFP_KERNEL);
606 	if (ret < 0)
607 		return ret;
608 	mm->mm_id = ret;
609 	return 0;
610 }
611 
612 static inline void mm_free_id(struct mm_struct *mm)
613 {
614 	const mm_id_t id = mm->mm_id;
615 
616 	mm->mm_id = MM_ID_DUMMY;
617 	if (id == MM_ID_DUMMY)
618 		return;
619 	if (WARN_ON_ONCE(id < MM_ID_MIN || id > MM_ID_MAX))
620 		return;
621 	ida_free(&mm_ida, id);
622 }
623 #else /* !CONFIG_MM_ID */
624 static inline int mm_alloc_id(struct mm_struct *mm) { return 0; }
625 static inline void mm_free_id(struct mm_struct *mm) {}
626 #endif /* CONFIG_MM_ID */
627 
628 static void check_mm(struct mm_struct *mm)
629 {
630 	int i;
631 
632 	BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
633 			 "Please make sure 'struct resident_page_types[]' is updated as well");
634 
635 	for (i = 0; i < NR_MM_COUNTERS; i++) {
636 		long x = percpu_counter_sum(&mm->rss_stat[i]);
637 
638 		if (unlikely(x)) {
639 			pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld Comm:%s Pid:%d\n",
640 				 mm, resident_page_types[i], x,
641 				 current->comm,
642 				 task_pid_nr(current));
643 		}
644 	}
645 
646 	if (mm_pgtables_bytes(mm))
647 		pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
648 				mm_pgtables_bytes(mm));
649 
650 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
651 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
652 #endif
653 }
654 
655 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
656 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
657 
658 static void do_check_lazy_tlb(void *arg)
659 {
660 	struct mm_struct *mm = arg;
661 
662 	WARN_ON_ONCE(current->active_mm == mm);
663 }
664 
665 static void do_shoot_lazy_tlb(void *arg)
666 {
667 	struct mm_struct *mm = arg;
668 
669 	if (current->active_mm == mm) {
670 		WARN_ON_ONCE(current->mm);
671 		current->active_mm = &init_mm;
672 		switch_mm(mm, &init_mm, current);
673 	}
674 }
675 
676 static void cleanup_lazy_tlbs(struct mm_struct *mm)
677 {
678 	if (!IS_ENABLED(CONFIG_MMU_LAZY_TLB_SHOOTDOWN)) {
679 		/*
680 		 * In this case, lazy tlb mms are refounted and would not reach
681 		 * __mmdrop until all CPUs have switched away and mmdrop()ed.
682 		 */
683 		return;
684 	}
685 
686 	/*
687 	 * Lazy mm shootdown does not refcount "lazy tlb mm" usage, rather it
688 	 * requires lazy mm users to switch to another mm when the refcount
689 	 * drops to zero, before the mm is freed. This requires IPIs here to
690 	 * switch kernel threads to init_mm.
691 	 *
692 	 * archs that use IPIs to flush TLBs can piggy-back that lazy tlb mm
693 	 * switch with the final userspace teardown TLB flush which leaves the
694 	 * mm lazy on this CPU but no others, reducing the need for additional
695 	 * IPIs here. There are cases where a final IPI is still required here,
696 	 * such as the final mmdrop being performed on a different CPU than the
697 	 * one exiting, or kernel threads using the mm when userspace exits.
698 	 *
699 	 * IPI overheads have not found to be expensive, but they could be
700 	 * reduced in a number of possible ways, for example (roughly
701 	 * increasing order of complexity):
702 	 * - The last lazy reference created by exit_mm() could instead switch
703 	 *   to init_mm, however it's probable this will run on the same CPU
704 	 *   immediately afterwards, so this may not reduce IPIs much.
705 	 * - A batch of mms requiring IPIs could be gathered and freed at once.
706 	 * - CPUs store active_mm where it can be remotely checked without a
707 	 *   lock, to filter out false-positives in the cpumask.
708 	 * - After mm_users or mm_count reaches zero, switching away from the
709 	 *   mm could clear mm_cpumask to reduce some IPIs, perhaps together
710 	 *   with some batching or delaying of the final IPIs.
711 	 * - A delayed freeing and RCU-like quiescing sequence based on mm
712 	 *   switching to avoid IPIs completely.
713 	 */
714 	on_each_cpu_mask(mm_cpumask(mm), do_shoot_lazy_tlb, (void *)mm, 1);
715 	if (IS_ENABLED(CONFIG_DEBUG_VM_SHOOT_LAZIES))
716 		on_each_cpu(do_check_lazy_tlb, (void *)mm, 1);
717 }
718 
719 /*
720  * Called when the last reference to the mm
721  * is dropped: either by a lazy thread or by
722  * mmput. Free the page directory and the mm.
723  */
724 void __mmdrop(struct mm_struct *mm)
725 {
726 	BUG_ON(mm == &init_mm);
727 	WARN_ON_ONCE(mm == current->mm);
728 
729 	/* Ensure no CPUs are using this as their lazy tlb mm */
730 	cleanup_lazy_tlbs(mm);
731 
732 	WARN_ON_ONCE(mm == current->active_mm);
733 	mm_destroy_sched(mm);
734 	mm_free_pgd(mm);
735 	mm_free_id(mm);
736 	destroy_context(mm);
737 	mmu_notifier_subscriptions_destroy(mm);
738 	check_mm(mm);
739 	mm_pasid_drop(mm);
740 	mm_destroy_cid(mm);
741 	percpu_counter_destroy_many(mm->rss_stat, NR_MM_COUNTERS);
742 
743 	free_mm(mm);
744 }
745 EXPORT_SYMBOL_GPL(__mmdrop);
746 
747 static void mmdrop_async_fn(struct work_struct *work)
748 {
749 	struct mm_struct *mm;
750 
751 	mm = container_of(work, struct mm_struct, async_put_work);
752 	__mmdrop(mm);
753 }
754 
755 static void mmdrop_async(struct mm_struct *mm)
756 {
757 	if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
758 		INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
759 		schedule_work(&mm->async_put_work);
760 	}
761 }
762 
763 static inline void free_signal_struct(struct signal_struct *sig)
764 {
765 	taskstats_tgid_free(sig);
766 	sched_autogroup_exit(sig);
767 	/*
768 	 * __mmdrop is not safe to call from softirq context on x86 due to
769 	 * pgd_dtor so postpone it to the async context
770 	 */
771 	if (sig->oom_mm)
772 		mmdrop_async(sig->oom_mm);
773 	kmem_cache_free(signal_cachep, sig);
774 }
775 
776 static inline void put_signal_struct(struct signal_struct *sig)
777 {
778 	if (refcount_dec_and_test(&sig->sigcnt))
779 		free_signal_struct(sig);
780 }
781 
782 void __put_task_struct(struct task_struct *tsk)
783 {
784 	WARN_ON(!tsk->exit_state);
785 	WARN_ON(refcount_read(&tsk->usage));
786 	WARN_ON(tsk == current);
787 
788 	unwind_task_free(tsk);
789 	io_uring_free(tsk);
790 	cgroup_task_free(tsk);
791 	task_numa_free(tsk, true);
792 	security_task_free(tsk);
793 	exit_creds(tsk);
794 	delayacct_tsk_free(tsk);
795 	put_signal_struct(tsk->signal);
796 	sched_core_free(tsk);
797 	free_task(tsk);
798 }
799 EXPORT_SYMBOL_GPL(__put_task_struct);
800 
801 void __put_task_struct_rcu_cb(struct rcu_head *rhp)
802 {
803 	struct task_struct *task = container_of(rhp, struct task_struct, rcu);
804 
805 	__put_task_struct(task);
806 }
807 EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
808 
809 void __init __weak arch_task_cache_init(void) { }
810 
811 /*
812  * set_max_threads
813  */
814 static void __init set_max_threads(unsigned int max_threads_suggested)
815 {
816 	u64 threads;
817 	unsigned long nr_pages = memblock_estimated_nr_free_pages();
818 
819 	/*
820 	 * The number of threads shall be limited such that the thread
821 	 * structures may only consume a small part of the available memory.
822 	 */
823 	if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
824 		threads = MAX_THREADS;
825 	else
826 		threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
827 				    (u64) THREAD_SIZE * 8UL);
828 
829 	if (threads > max_threads_suggested)
830 		threads = max_threads_suggested;
831 
832 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
833 }
834 
835 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
836 /* Initialized by the architecture: */
837 int arch_task_struct_size __read_mostly;
838 #endif
839 
840 static void __init task_struct_whitelist(unsigned long *offset, unsigned long *size)
841 {
842 	/* Fetch thread_struct whitelist for the architecture. */
843 	arch_thread_struct_whitelist(offset, size);
844 
845 	/*
846 	 * Handle zero-sized whitelist or empty thread_struct, otherwise
847 	 * adjust offset to position of thread_struct in task_struct.
848 	 */
849 	if (unlikely(*size == 0))
850 		*offset = 0;
851 	else
852 		*offset += offsetof(struct task_struct, thread);
853 }
854 
855 void __init fork_init(void)
856 {
857 	int i;
858 #ifndef ARCH_MIN_TASKALIGN
859 #define ARCH_MIN_TASKALIGN	0
860 #endif
861 	int align = max3(L1_CACHE_BYTES, ARCH_MIN_TASKALIGN,
862 			 __alignof__(struct task_struct));
863 	unsigned long useroffset, usersize;
864 
865 	/* create a slab on which task_structs can be allocated */
866 	task_struct_whitelist(&useroffset, &usersize);
867 	task_struct_cachep = kmem_cache_create_usercopy("task_struct",
868 			arch_task_struct_size, align,
869 			SLAB_PANIC|SLAB_ACCOUNT,
870 			useroffset, usersize, NULL);
871 
872 	/* do the arch specific task caches init */
873 	arch_task_cache_init();
874 
875 	set_max_threads(MAX_THREADS);
876 
877 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
878 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
879 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
880 		init_task.signal->rlim[RLIMIT_NPROC];
881 
882 	for (i = 0; i < UCOUNT_COUNTS; i++)
883 		init_user_ns.ucount_max[i] = max_threads/2;
884 
885 	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_NPROC,      RLIM_INFINITY);
886 	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MSGQUEUE,   RLIM_INFINITY);
887 	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_SIGPENDING, RLIM_INFINITY);
888 	set_userns_rlimit_max(&init_user_ns, UCOUNT_RLIMIT_MEMLOCK,    RLIM_INFINITY);
889 
890 #ifdef CONFIG_VMAP_STACK
891 	cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
892 			  NULL, free_vm_stack_cache);
893 #endif
894 
895 	scs_init();
896 
897 	lockdep_init_task(&init_task);
898 	uprobes_init();
899 }
900 
901 int __weak arch_dup_task_struct(struct task_struct *dst,
902 					       struct task_struct *src)
903 {
904 	*dst = *src;
905 	return 0;
906 }
907 
908 void set_task_stack_end_magic(struct task_struct *tsk)
909 {
910 	unsigned long *stackend;
911 
912 	stackend = end_of_stack(tsk);
913 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
914 }
915 
916 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
917 {
918 	struct task_struct *tsk;
919 	int err;
920 
921 	if (node == NUMA_NO_NODE)
922 		node = tsk_fork_get_node(orig);
923 	tsk = alloc_task_struct_node(node);
924 	if (!tsk)
925 		return NULL;
926 
927 	err = arch_dup_task_struct(tsk, orig);
928 	if (err)
929 		goto free_tsk;
930 
931 	err = alloc_thread_stack_node(tsk, node);
932 	if (err)
933 		goto free_tsk;
934 
935 #ifdef CONFIG_THREAD_INFO_IN_TASK
936 	refcount_set(&tsk->stack_refcount, 1);
937 #endif
938 	account_kernel_stack(tsk, 1);
939 
940 	err = smp_task_ipi_mask_alloc(tsk);
941 	if (err)
942 		goto free_stack;
943 
944 	err = scs_prepare(tsk, node);
945 	if (err)
946 		goto free_ipi_mask;
947 
948 #ifdef CONFIG_SECCOMP
949 	/*
950 	 * We must handle setting up seccomp filters once we're under
951 	 * the sighand lock in case orig has changed between now and
952 	 * then. Until then, filter must be NULL to avoid messing up
953 	 * the usage counts on the error path calling free_task.
954 	 */
955 	tsk->seccomp.filter = NULL;
956 #endif
957 
958 	RCU_INIT_POINTER(tsk->exec_state, NULL);
959 
960 	setup_thread_stack(tsk, orig);
961 	clear_user_return_notifier(tsk);
962 	clear_tsk_need_resched(tsk);
963 	set_task_stack_end_magic(tsk);
964 	clear_syscall_work_syscall_user_dispatch(tsk);
965 
966 #ifdef CONFIG_STACKPROTECTOR
967 	tsk->stack_canary = get_random_canary();
968 #endif
969 	if (orig->cpus_ptr == &orig->cpus_mask)
970 		tsk->cpus_ptr = &tsk->cpus_mask;
971 	dup_user_cpus_ptr(tsk, orig, node);
972 
973 	/*
974 	 * Nobody can keep the task while copy_process() may still fail, which
975 	 * frees it no matter what.
976 	 */
977 	refcount_set(&tsk->rcu_users, 0);
978 	/* One for the rcu users */
979 	refcount_set(&tsk->usage, 1);
980 #ifdef CONFIG_BLK_DEV_IO_TRACE
981 	tsk->btrace_seq = 0;
982 #endif
983 	tsk->splice_pipe = NULL;
984 	tsk->task_frag.page = NULL;
985 	tsk->wake_q.next = NULL;
986 	tsk->worker_private = NULL;
987 
988 	kcov_task_init(tsk);
989 	kmsan_task_create(tsk);
990 	kmap_local_fork(tsk);
991 
992 #ifdef CONFIG_FAULT_INJECTION
993 	tsk->fail_nth = 0;
994 #endif
995 
996 #ifdef CONFIG_BLK_CGROUP
997 	tsk->throttle_disk = NULL;
998 	tsk->use_memdelay = 0;
999 #endif
1000 
1001 #ifdef CONFIG_ARCH_HAS_CPU_PASID
1002 	tsk->pasid_activated = 0;
1003 #endif
1004 
1005 #ifdef CONFIG_MEMCG
1006 	tsk->active_memcg = NULL;
1007 #endif
1008 
1009 #ifdef CONFIG_X86_BUS_LOCK_DETECT
1010 	tsk->reported_split_lock = 0;
1011 #endif
1012 
1013 #ifdef CONFIG_SCHED_MM_CID
1014 	tsk->mm_cid.cid = MM_CID_UNSET;
1015 	tsk->mm_cid.active = 0;
1016 	INIT_HLIST_NODE(&tsk->mm_cid.node);
1017 #endif
1018 
1019 #ifdef CONFIG_BPF_SYSCALL
1020 	RCU_INIT_POINTER(tsk->bpf_storage, NULL);
1021 	tsk->bpf_ctx = NULL;
1022 #endif
1023 	return tsk;
1024 
1025 free_ipi_mask:
1026 	smp_task_ipi_mask_free(tsk);
1027 free_stack:
1028 	exit_task_stack_account(tsk);
1029 	free_thread_stack(tsk);
1030 free_tsk:
1031 	free_task_struct(tsk);
1032 	return NULL;
1033 }
1034 
1035 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
1036 
1037 static unsigned long coredump_filter = MMF_DUMP_FILTER_DEFAULT;
1038 
1039 static int __init coredump_filter_setup(char *s)
1040 {
1041 	if (kstrtoul(s, 0, &coredump_filter))
1042 		return 0;
1043 	coredump_filter <<= MMF_DUMP_FILTER_SHIFT;
1044 	coredump_filter &= MMF_DUMP_FILTER_MASK;
1045 	return 1;
1046 }
1047 
1048 __setup("coredump_filter=", coredump_filter_setup);
1049 
1050 #include <linux/init_task.h>
1051 
1052 static void mm_init_aio(struct mm_struct *mm)
1053 {
1054 #ifdef CONFIG_AIO
1055 	spin_lock_init(&mm->ioctx_lock);
1056 	mm->ioctx_table = NULL;
1057 #endif
1058 }
1059 
1060 static __always_inline void mm_clear_owner(struct mm_struct *mm,
1061 					   struct task_struct *p)
1062 {
1063 #ifdef CONFIG_MEMCG
1064 	if (mm->owner == p)
1065 		WRITE_ONCE(mm->owner, NULL);
1066 #endif
1067 }
1068 
1069 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
1070 {
1071 #ifdef CONFIG_MEMCG
1072 	mm->owner = p;
1073 #endif
1074 }
1075 
1076 static void mm_init_uprobes_state(struct mm_struct *mm)
1077 {
1078 #ifdef CONFIG_UPROBES
1079 	mm->uprobes_state.xol_area = NULL;
1080 #endif
1081 }
1082 
1083 static void mmap_init_lock(struct mm_struct *mm)
1084 {
1085 	init_rwsem(&mm->mmap_lock);
1086 	mm_lock_seqcount_init(mm);
1087 	rcuwait_init(&mm->vma_writer_wait);
1088 }
1089 
1090 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p)
1091 {
1092 	mt_init_flags(&mm->mm_mt, MM_MT_FLAGS);
1093 	mt_set_external_lock(&mm->mm_mt, &mm->mmap_lock);
1094 	atomic_set(&mm->mm_users, 1);
1095 	atomic_set(&mm->mm_count, 1);
1096 	seqcount_init(&mm->write_protect_seq);
1097 	mmap_init_lock(mm);
1098 	INIT_LIST_HEAD(&mm->mmlist);
1099 	mm_pgtables_bytes_init(mm);
1100 	mm->map_count = 0;
1101 	mm->locked_vm = 0;
1102 	atomic64_set(&mm->pinned_vm, 0);
1103 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1104 	spin_lock_init(&mm->page_table_lock);
1105 	spin_lock_init(&mm->arg_lock);
1106 	mm_init_cpumask(mm);
1107 	mm_init_aio(mm);
1108 	mm_init_owner(mm, p);
1109 	mm_pasid_init(mm);
1110 	RCU_INIT_POINTER(mm->exe_file, NULL);
1111 	mmu_notifier_subscriptions_init(mm);
1112 	init_tlb_flush_pending(mm);
1113 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !defined(CONFIG_SPLIT_PMD_PTLOCKS)
1114 	mm->pmd_huge_pte = NULL;
1115 #endif
1116 	mm_init_uprobes_state(mm);
1117 	hugetlb_count_init(mm);
1118 	futex_mm_init(mm);
1119 
1120 	mm_flags_clear_all(mm);
1121 	if (current->mm) {
1122 		unsigned long flags = __mm_flags_get_word(current->mm);
1123 
1124 		__mm_flags_overwrite_word(mm, mmf_init_legacy_flags(flags));
1125 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1126 	} else {
1127 		__mm_flags_overwrite_word(mm, coredump_filter);
1128 		mm->def_flags = 0;
1129 	}
1130 
1131 	if (mm_alloc_pgd(mm))
1132 		goto fail_mm_init;
1133 
1134 	if (mm_alloc_id(mm))
1135 		goto fail_noid;
1136 
1137 	if (init_new_context(p, mm))
1138 		goto fail_nocontext;
1139 
1140 	if (mm_alloc_cid(mm, p))
1141 		goto fail_cid;
1142 
1143 	if (mm_alloc_sched(mm))
1144 		goto fail_sched;
1145 
1146 	if (percpu_counter_init_many(mm->rss_stat, 0, GFP_KERNEL_ACCOUNT,
1147 				     NR_MM_COUNTERS))
1148 		goto fail_pcpu;
1149 
1150 	lru_gen_init_mm(mm);
1151 	return mm;
1152 
1153 fail_pcpu:
1154 	mm_destroy_sched(mm);
1155 fail_sched:
1156 	mm_destroy_cid(mm);
1157 fail_cid:
1158 	destroy_context(mm);
1159 fail_nocontext:
1160 	mm_free_id(mm);
1161 fail_noid:
1162 	mm_free_pgd(mm);
1163 fail_mm_init:
1164 	free_mm(mm);
1165 	return NULL;
1166 }
1167 
1168 /*
1169  * Allocate and initialize an mm_struct.
1170  */
1171 struct mm_struct *mm_alloc(void)
1172 {
1173 	struct mm_struct *mm;
1174 
1175 	mm = allocate_mm();
1176 	if (!mm)
1177 		return NULL;
1178 
1179 	memset(mm, 0, sizeof(*mm));
1180 	return mm_init(mm, current);
1181 }
1182 EXPORT_SYMBOL_IF_KUNIT(mm_alloc);
1183 
1184 static inline void __mmput(struct mm_struct *mm)
1185 {
1186 	VM_BUG_ON(atomic_read(&mm->mm_users));
1187 
1188 	uprobe_clear_state(mm);
1189 	exit_aio(mm);
1190 	ksm_exit(mm);
1191 	khugepaged_exit(mm); /* must run before exit_mmap */
1192 	exit_mmap(mm);
1193 	mm_put_huge_zero_folio(mm);
1194 	set_mm_exe_file(mm, NULL);
1195 	if (!list_empty(&mm->mmlist)) {
1196 		spin_lock(&mmlist_lock);
1197 		list_del(&mm->mmlist);
1198 		spin_unlock(&mmlist_lock);
1199 	}
1200 	if (mm->binfmt)
1201 		module_put(mm->binfmt->module);
1202 	lru_gen_del_mm(mm);
1203 	futex_hash_free(mm);
1204 	mmdrop(mm);
1205 }
1206 
1207 /*
1208  * Decrement the use count and release all resources for an mm.
1209  */
1210 void mmput(struct mm_struct *mm)
1211 {
1212 	might_sleep();
1213 
1214 	if (atomic_dec_and_test(&mm->mm_users))
1215 		__mmput(mm);
1216 }
1217 EXPORT_SYMBOL_GPL(mmput);
1218 
1219 #if defined(CONFIG_MMU) || defined(CONFIG_FUTEX_PRIVATE_HASH)
1220 static void mmput_async_fn(struct work_struct *work)
1221 {
1222 	struct mm_struct *mm = container_of(work, struct mm_struct,
1223 					    async_put_work);
1224 
1225 	__mmput(mm);
1226 }
1227 
1228 void mmput_async(struct mm_struct *mm)
1229 {
1230 	if (atomic_dec_and_test(&mm->mm_users)) {
1231 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
1232 		schedule_work(&mm->async_put_work);
1233 	}
1234 }
1235 EXPORT_SYMBOL_GPL(mmput_async);
1236 #endif
1237 
1238 /**
1239  * set_mm_exe_file - change a reference to the mm's executable file
1240  * @mm: The mm to change.
1241  * @new_exe_file: The new file to use.
1242  *
1243  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1244  *
1245  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1246  * invocations: in mmput() nobody alive left, in execve it happens before
1247  * the new mm is made visible to anyone.
1248  *
1249  * Can only fail if new_exe_file != NULL.
1250  */
1251 int set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1252 {
1253 	struct file *old_exe_file;
1254 
1255 	/*
1256 	 * It is safe to dereference the exe_file without RCU as
1257 	 * this function is only called if nobody else can access
1258 	 * this mm -- see comment above for justification.
1259 	 */
1260 	old_exe_file = rcu_dereference_raw(mm->exe_file);
1261 
1262 	if (new_exe_file) {
1263 		/*
1264 		 * We expect the caller (i.e., sys_execve) to already denied
1265 		 * write access, so this is unlikely to fail.
1266 		 */
1267 		if (unlikely(exe_file_deny_write_access(new_exe_file)))
1268 			return -EACCES;
1269 		get_file(new_exe_file);
1270 	}
1271 	rcu_assign_pointer(mm->exe_file, new_exe_file);
1272 	if (old_exe_file) {
1273 		exe_file_allow_write_access(old_exe_file);
1274 		fput(old_exe_file);
1275 	}
1276 	return 0;
1277 }
1278 
1279 /**
1280  * replace_mm_exe_file - replace a reference to the mm's executable file
1281  * @mm: The mm to change.
1282  * @new_exe_file: The new file to use.
1283  *
1284  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1285  *
1286  * Main user is sys_prctl(PR_SET_MM_MAP/EXE_FILE).
1287  */
1288 int replace_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1289 {
1290 	struct vm_area_struct *vma;
1291 	struct file *old_exe_file;
1292 	int ret = 0;
1293 
1294 	/* Forbid mm->exe_file change if old file still mapped. */
1295 	old_exe_file = get_mm_exe_file(mm);
1296 	if (old_exe_file) {
1297 		VMA_ITERATOR(vmi, mm, 0);
1298 		mmap_read_lock(mm);
1299 		for_each_vma(vmi, vma) {
1300 			if (!vma->vm_file)
1301 				continue;
1302 			if (path_equal(&vma->vm_file->f_path,
1303 				       &old_exe_file->f_path)) {
1304 				ret = -EBUSY;
1305 				break;
1306 			}
1307 		}
1308 		mmap_read_unlock(mm);
1309 		fput(old_exe_file);
1310 		if (ret)
1311 			return ret;
1312 	}
1313 
1314 	ret = exe_file_deny_write_access(new_exe_file);
1315 	if (ret)
1316 		return -EACCES;
1317 	get_file(new_exe_file);
1318 
1319 	/* set the new file */
1320 	mmap_write_lock(mm);
1321 	old_exe_file = rcu_dereference_raw(mm->exe_file);
1322 	rcu_assign_pointer(mm->exe_file, new_exe_file);
1323 	mmap_write_unlock(mm);
1324 
1325 	if (old_exe_file) {
1326 		exe_file_allow_write_access(old_exe_file);
1327 		fput(old_exe_file);
1328 	}
1329 	return 0;
1330 }
1331 
1332 /**
1333  * get_mm_exe_file - acquire a reference to the mm's executable file
1334  * @mm: The mm of interest.
1335  *
1336  * Returns %NULL if mm has no associated executable file.
1337  * User must release file via fput().
1338  */
1339 struct file *get_mm_exe_file(struct mm_struct *mm)
1340 {
1341 	struct file *exe_file;
1342 
1343 	rcu_read_lock();
1344 	exe_file = get_file_rcu(&mm->exe_file);
1345 	rcu_read_unlock();
1346 	return exe_file;
1347 }
1348 
1349 /**
1350  * get_task_exe_file - acquire a reference to the task's executable file
1351  * @task: The task.
1352  *
1353  * Returns %NULL if task's mm (if any) has no associated executable file or
1354  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1355  * User must release file via fput().
1356  */
1357 struct file *get_task_exe_file(struct task_struct *task)
1358 {
1359 	struct file *exe_file = NULL;
1360 	struct mm_struct *mm;
1361 
1362 	if (task->flags & PF_KTHREAD)
1363 		return NULL;
1364 
1365 	task_lock(task);
1366 	mm = task->mm;
1367 	if (mm)
1368 		exe_file = get_mm_exe_file(mm);
1369 	task_unlock(task);
1370 	return exe_file;
1371 }
1372 
1373 /**
1374  * get_task_mm - acquire a reference to the task's mm
1375  * @task: The task.
1376  *
1377  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1378  * this kernel workthread has transiently adopted a user mm with kthread_use_mm,
1379  * to do its AIO) is not set and if so returns a reference to it, after
1380  * bumping up the use count.  User must release the mm via mmput()
1381  * after use.  Typically used by /proc and ptrace.
1382  */
1383 struct mm_struct *get_task_mm(struct task_struct *task)
1384 {
1385 	struct mm_struct *mm;
1386 
1387 	if (task->flags & PF_KTHREAD)
1388 		return NULL;
1389 
1390 	task_lock(task);
1391 	mm = task->mm;
1392 	if (mm)
1393 		mmget(mm);
1394 	task_unlock(task);
1395 	return mm;
1396 }
1397 EXPORT_SYMBOL_GPL(get_task_mm);
1398 
1399 static bool may_access_mm(struct mm_struct *mm, struct task_struct *task, unsigned int mode)
1400 {
1401 	if (mm == current->mm)
1402 		return true;
1403 	if (ptrace_may_access(task, mode))
1404 		return true;
1405 	if ((mode & PTRACE_MODE_READ) && perfmon_capable())
1406 		return true;
1407 	return false;
1408 }
1409 
1410 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1411 {
1412 	struct mm_struct *mm;
1413 	int err;
1414 
1415 	err =  down_read_killable(&task->signal->exec_update_lock);
1416 	if (err)
1417 		return ERR_PTR(err);
1418 
1419 	mm = get_task_mm(task);
1420 	if (!mm) {
1421 		mm = ERR_PTR(-ESRCH);
1422 	} else if (!may_access_mm(mm, task, mode)) {
1423 		mmput(mm);
1424 		mm = ERR_PTR(-EACCES);
1425 	}
1426 	up_read(&task->signal->exec_update_lock);
1427 
1428 	return mm;
1429 }
1430 
1431 static void complete_vfork_done(struct task_struct *tsk)
1432 {
1433 	struct completion *vfork;
1434 
1435 	task_lock(tsk);
1436 	vfork = tsk->vfork_done;
1437 	if (likely(vfork)) {
1438 		tsk->vfork_done = NULL;
1439 		complete(vfork);
1440 	}
1441 	task_unlock(tsk);
1442 }
1443 
1444 static int wait_for_vfork_done(struct task_struct *child,
1445 				struct completion *vfork)
1446 {
1447 	unsigned int state = TASK_KILLABLE|TASK_FREEZABLE;
1448 	int killed;
1449 
1450 	cgroup_enter_frozen();
1451 	killed = wait_for_completion_state(vfork, state);
1452 	cgroup_leave_frozen(false);
1453 
1454 	if (killed) {
1455 		task_lock(child);
1456 		child->vfork_done = NULL;
1457 		task_unlock(child);
1458 	}
1459 
1460 	put_task_struct(child);
1461 	return killed;
1462 }
1463 
1464 /* Please note the differences between mmput and mm_release.
1465  * mmput is called whenever we stop holding onto a mm_struct,
1466  * error success whatever.
1467  *
1468  * mm_release is called after a mm_struct has been removed
1469  * from the current process.
1470  *
1471  * This difference is important for error handling, when we
1472  * only half set up a mm_struct for a new process and need to restore
1473  * the old one.  Because we mmput the new mm_struct before
1474  * restoring the old one. . .
1475  * Eric Biederman 10 January 1998
1476  */
1477 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1478 {
1479 	uprobe_free_utask(tsk);
1480 
1481 	/* Get rid of any cached register state */
1482 	deactivate_mm(tsk, mm);
1483 
1484 	/*
1485 	 * Signal userspace if we're not exiting with a core dump
1486 	 * because we want to leave the value intact for debugging
1487 	 * purposes.
1488 	 */
1489 	if (tsk->clear_child_tid) {
1490 		if (atomic_read(&mm->mm_users) > 1) {
1491 			/*
1492 			 * We don't check the error code - if userspace has
1493 			 * not set up a proper pointer then tough luck.
1494 			 */
1495 			put_user(0, tsk->clear_child_tid);
1496 			do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1497 					1, NULL, NULL, 0, 0);
1498 		}
1499 		tsk->clear_child_tid = NULL;
1500 	}
1501 
1502 	/*
1503 	 * All done, finally we can wake up parent and return this mm to him.
1504 	 * Also kthread_stop() uses this completion for synchronization.
1505 	 */
1506 	if (tsk->vfork_done)
1507 		complete_vfork_done(tsk);
1508 }
1509 
1510 void mm_exit_exec_release(struct task_struct *tsk, struct mm_struct *mm)
1511 {
1512 	futex_exit_exec_release(tsk);
1513 	mm_release(tsk, mm);
1514 }
1515 
1516 /**
1517  * dup_mm() - duplicates an existing mm structure
1518  * @tsk: the task_struct with which the new mm will be associated.
1519  * @oldmm: the mm to duplicate.
1520  *
1521  * Allocates a new mm structure and duplicates the provided @oldmm structure
1522  * content into it.
1523  *
1524  * Return: the duplicated mm or NULL on failure.
1525  */
1526 static struct mm_struct *dup_mm(struct task_struct *tsk,
1527 				struct mm_struct *oldmm)
1528 {
1529 	struct mm_struct *mm;
1530 	int err;
1531 
1532 	mm = allocate_mm();
1533 	if (!mm)
1534 		goto fail_nomem;
1535 
1536 	uprobe_start_dup_mmap();
1537 	if (mmap_write_lock_killable(oldmm)) {
1538 		free_mm(mm);
1539 		goto fail_uprobe;
1540 	}
1541 
1542 	memcpy(mm, oldmm, sizeof(*mm));
1543 
1544 	if (!mm_init(mm, tsk)) {
1545 		mmap_write_unlock(oldmm);
1546 		goto fail_uprobe;
1547 	}
1548 
1549 	err = dup_mmap(mm, oldmm);
1550 	if (err)
1551 		goto free_pt;
1552 	uprobe_end_dup_mmap();
1553 
1554 	mm->hiwater_rss = get_mm_rss(mm);
1555 	mm->hiwater_vm = mm->total_vm;
1556 
1557 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1558 		goto free_pt;
1559 
1560 	return mm;
1561 
1562 free_pt:
1563 	/* don't put binfmt in mmput, we haven't got module yet */
1564 	mm->binfmt = NULL;
1565 	mm_init_owner(mm, NULL);
1566 	mmput(mm);
1567 	if (!err)
1568 		return NULL;
1569 fail_uprobe:
1570 	uprobe_end_dup_mmap();
1571 fail_nomem:
1572 	return NULL;
1573 }
1574 
1575 static int copy_mm(u64 clone_flags, struct task_struct *tsk)
1576 {
1577 	struct mm_struct *mm, *oldmm;
1578 
1579 	tsk->min_flt = tsk->maj_flt = 0;
1580 	tsk->nvcsw = tsk->nivcsw = 0;
1581 #ifdef CONFIG_DETECT_HUNG_TASK
1582 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1583 	tsk->last_switch_time = 0;
1584 #endif
1585 
1586 	tsk->mm = NULL;
1587 	tsk->active_mm = NULL;
1588 
1589 	/*
1590 	 * Are we cloning a kernel thread?
1591 	 *
1592 	 * We need to steal a active VM for that..
1593 	 */
1594 	oldmm = current->mm;
1595 	if (!oldmm)
1596 		return 0;
1597 
1598 	if (clone_flags & CLONE_VM) {
1599 		mmget(oldmm);
1600 		mm = oldmm;
1601 	} else {
1602 		mm = dup_mm(tsk, current->mm);
1603 		if (!mm)
1604 			return -ENOMEM;
1605 	}
1606 
1607 	tsk->mm = mm;
1608 	tsk->active_mm = mm;
1609 	sched_cache_fork(tsk);
1610 	return 0;
1611 }
1612 
1613 static int copy_exec_state(u64 clone_flags, struct task_struct *tsk)
1614 {
1615 	struct task_exec_state *exec_state;
1616 
1617 	/* CLONE_VM siblings refcount-share the parent's exec_state. */
1618 	if (clone_flags & CLONE_VM) {
1619 		exec_state = rcu_dereference_protected(current->exec_state, true);
1620 		refcount_inc(&exec_state->count);
1621 		rcu_assign_pointer(tsk->exec_state, exec_state);
1622 		return 0;
1623 	}
1624 
1625 	/* Everyone else inherits a fresh copy. */
1626 	return task_exec_state_copy(tsk);
1627 }
1628 
1629 enum copy_fs_flags_t {
1630 	COPY_FS_UMH	= (1 << 0),
1631 	COPY_FS_IDLE	= (1 << 1),
1632 };
1633 
1634 static int copy_fs(u64 clone_flags, struct task_struct *tsk,
1635 		   enum copy_fs_flags_t flags)
1636 {
1637 	struct fs_struct *fs;
1638 
1639 	/*
1640 	 * Usermodehelper may copy userspace_init_fs filesystem state but
1641 	 * they don't get to create mount namespaces, share the
1642 	 * filesystem state, or be started from a non-initial mount
1643 	 * namespace.
1644 	 */
1645 	if (flags & COPY_FS_UMH) {
1646 		if (clone_flags & (CLONE_NEWNS | CLONE_FS))
1647 			return -EINVAL;
1648 		if (current->nsproxy->mnt_ns != &init_mnt_ns)
1649 			return -EINVAL;
1650 		fs = userspace_init_fs;
1651 	} else if (flags & COPY_FS_IDLE) {
1652 		fs = &init_fs;
1653 	} else {
1654 		fs = current->fs;
1655 		VFS_WARN_ON_ONCE(current->fs != current->real_fs);
1656 	}
1657 
1658 	if (clone_flags & CLONE_FS) {
1659 		read_seqlock_excl(&fs->seq);
1660 		/* "users" and "in_exec" locked for check_unsafe_exec() */
1661 		if (fs->in_exec) {
1662 			read_sequnlock_excl(&fs->seq);
1663 			return -EAGAIN;
1664 		}
1665 		fs->users++;
1666 		read_sequnlock_excl(&fs->seq);
1667 		tsk->real_fs = tsk->fs = fs;
1668 		return 0;
1669 	}
1670 	tsk->real_fs = tsk->fs = copy_fs_struct(fs);
1671 	if (!tsk->fs)
1672 		return -ENOMEM;
1673 	return 0;
1674 }
1675 
1676 static int copy_files(u64 clone_flags, struct task_struct *tsk,
1677 		      int no_files)
1678 {
1679 	struct files_struct *oldf, *newf;
1680 
1681 	/*
1682 	 * A background process may not have any files ...
1683 	 */
1684 	oldf = current->files;
1685 	if (!oldf)
1686 		return 0;
1687 
1688 	if (no_files) {
1689 		tsk->files = NULL;
1690 		return 0;
1691 	}
1692 
1693 	if (clone_flags & CLONE_FILES) {
1694 		atomic_inc(&oldf->count);
1695 		tsk->files = oldf;
1696 		return 0;
1697 	}
1698 
1699 	newf = dup_fd(oldf, NULL);
1700 	if (IS_ERR(newf))
1701 		return PTR_ERR(newf);
1702 
1703 	tsk->files = newf;
1704 	return 0;
1705 }
1706 
1707 static int copy_sighand(u64 clone_flags, struct task_struct *tsk)
1708 {
1709 	struct sighand_struct *sig;
1710 
1711 	if (clone_flags & CLONE_SIGHAND) {
1712 		refcount_inc(&current->sighand->count);
1713 		return 0;
1714 	}
1715 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1716 	RCU_INIT_POINTER(tsk->sighand, sig);
1717 	if (!sig)
1718 		return -ENOMEM;
1719 
1720 	refcount_set(&sig->count, 1);
1721 	spin_lock_irq(&current->sighand->siglock);
1722 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1723 	spin_unlock_irq(&current->sighand->siglock);
1724 
1725 	/* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1726 	if (clone_flags & CLONE_CLEAR_SIGHAND)
1727 		flush_signal_handlers(tsk, 0);
1728 
1729 	return 0;
1730 }
1731 
1732 void __cleanup_sighand(struct sighand_struct *sighand)
1733 {
1734 	if (refcount_dec_and_test(&sighand->count)) {
1735 		signalfd_cleanup(sighand);
1736 		/*
1737 		 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1738 		 * without an RCU grace period, see __lock_task_sighand().
1739 		 */
1740 		kmem_cache_free(sighand_cachep, sighand);
1741 	}
1742 }
1743 
1744 /*
1745  * Initialize POSIX timer handling for a thread group.
1746  */
1747 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1748 {
1749 	struct posix_cputimers *pct = &sig->posix_cputimers;
1750 	unsigned long cpu_limit;
1751 
1752 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1753 	posix_cputimers_group_init(pct, cpu_limit);
1754 }
1755 
1756 static int copy_signal(u64 clone_flags, struct task_struct *tsk)
1757 {
1758 	struct signal_struct *sig;
1759 
1760 	if (clone_flags & CLONE_THREAD)
1761 		return 0;
1762 
1763 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1764 	tsk->signal = sig;
1765 	if (!sig)
1766 		return -ENOMEM;
1767 
1768 	sig->nr_threads = 1;
1769 	sig->quick_threads = 1;
1770 	atomic_set(&sig->live, 1);
1771 	refcount_set(&sig->sigcnt, 1);
1772 
1773 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1774 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1775 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1776 
1777 	init_waitqueue_head(&sig->wait_chldexit);
1778 	sig->curr_target = tsk;
1779 	init_sigpending(&sig->shared_pending);
1780 	INIT_HLIST_HEAD(&sig->multiprocess);
1781 	seqlock_init(&sig->stats_lock);
1782 	prev_cputime_init(&sig->prev_cputime);
1783 
1784 #ifdef CONFIG_POSIX_TIMERS
1785 	INIT_HLIST_HEAD(&sig->posix_timers);
1786 	INIT_HLIST_HEAD(&sig->ignored_posix_timers);
1787 	hrtimer_setup(&sig->real_timer, it_real_fn, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1788 #endif
1789 
1790 	task_lock(current->group_leader);
1791 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1792 	task_unlock(current->group_leader);
1793 
1794 	posix_cpu_timers_init_group(sig);
1795 
1796 	tty_audit_fork(sig);
1797 	sched_autogroup_fork(sig);
1798 
1799 #ifdef CONFIG_CGROUPS
1800 	init_rwsem(&sig->cgroup_threadgroup_rwsem);
1801 #endif
1802 
1803 	sig->oom_score_adj = current->signal->oom_score_adj;
1804 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1805 
1806 	mutex_init(&sig->cred_guard_mutex);
1807 	init_rwsem(&sig->exec_update_lock);
1808 
1809 	return 0;
1810 }
1811 
1812 static void copy_seccomp(struct task_struct *p)
1813 {
1814 #ifdef CONFIG_SECCOMP
1815 	/*
1816 	 * Must be called with sighand->lock held, which is common to
1817 	 * all threads in the group. Holding cred_guard_mutex is not
1818 	 * needed because this new task is not yet running and cannot
1819 	 * be racing exec.
1820 	 */
1821 	assert_spin_locked(&current->sighand->siglock);
1822 
1823 	/* Ref-count the new filter user, and assign it. */
1824 	get_seccomp_filter(current);
1825 	p->seccomp = current->seccomp;
1826 
1827 	/*
1828 	 * Explicitly enable no_new_privs here in case it got set
1829 	 * between the task_struct being duplicated and holding the
1830 	 * sighand lock. The seccomp state and nnp must be in sync.
1831 	 */
1832 	if (task_no_new_privs(current))
1833 		task_set_no_new_privs(p);
1834 
1835 	/*
1836 	 * If the parent gained a seccomp mode after copying thread
1837 	 * flags and between before we held the sighand lock, we have
1838 	 * to manually enable the seccomp thread flag here.
1839 	 */
1840 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1841 		set_task_syscall_work(p, SECCOMP);
1842 #endif
1843 }
1844 
1845 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1846 {
1847 	current->clear_child_tid = tidptr;
1848 
1849 	return task_pid_vnr(current);
1850 }
1851 
1852 static void rt_mutex_init_task(struct task_struct *p)
1853 {
1854 	raw_spin_lock_init(&p->pi_lock);
1855 #ifdef CONFIG_RT_MUTEXES
1856 	p->pi_waiters = RB_ROOT_CACHED;
1857 	p->pi_top_task = NULL;
1858 	p->pi_blocked_on = NULL;
1859 #endif
1860 }
1861 
1862 static inline void init_task_pid_links(struct task_struct *task)
1863 {
1864 	enum pid_type type;
1865 
1866 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1867 		INIT_HLIST_NODE(&task->pid_links[type]);
1868 }
1869 
1870 static inline void
1871 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1872 {
1873 	if (type == PIDTYPE_PID)
1874 		task->thread_pid = pid;
1875 	else
1876 		task->signal->pids[type] = pid;
1877 }
1878 
1879 static inline void rcu_copy_process(struct task_struct *p)
1880 {
1881 #ifdef CONFIG_PREEMPT_RCU
1882 	p->rcu_read_lock_nesting = 0;
1883 	p->rcu_read_unlock_special.s = 0;
1884 	p->rcu_blocked_node = NULL;
1885 	INIT_LIST_HEAD(&p->rcu_node_entry);
1886 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1887 #ifdef CONFIG_TASKS_RCU
1888 	p->rcu_tasks_holdout = false;
1889 	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1890 	p->rcu_tasks_idle_cpu = -1;
1891 	INIT_LIST_HEAD(&p->rcu_tasks_exit_list);
1892 #endif /* #ifdef CONFIG_TASKS_RCU */
1893 #ifdef CONFIG_TASKS_TRACE_RCU
1894 	p->trc_reader_nesting = 0;
1895 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1896 }
1897 
1898 /**
1899  * pidfd_prepare - allocate a new pidfd_file and reserve a pidfd
1900  * @pid:   the struct pid for which to create a pidfd
1901  * @flags: flags of the new @pidfd
1902  * @ret_file: return the new pidfs file
1903  *
1904  * Allocate a new file that stashes @pid and reserve a new pidfd number in the
1905  * caller's file descriptor table. The pidfd is reserved but not installed yet.
1906  *
1907  * The helper verifies that @pid is still in use, without PIDFD_THREAD the
1908  * task identified by @pid must be a thread-group leader.
1909  *
1910  * If this function returns successfully the caller is responsible to either
1911  * call fd_install() passing the returned pidfd and pidfd file as arguments in
1912  * order to install the pidfd into its file descriptor table or they must use
1913  * put_unused_fd() and fput() on the returned pidfd and pidfd file
1914  * respectively.
1915  *
1916  * This function is useful when a pidfd must already be reserved but there
1917  * might still be points of failure afterwards and the caller wants to ensure
1918  * that no pidfd is leaked into its file descriptor table.
1919  *
1920  * Return: On success, a reserved pidfd is returned from the function and a new
1921  *         pidfd file is returned in the last argument to the function. On
1922  *         error, a negative error code is returned from the function and the
1923  *         last argument remains unchanged.
1924  */
1925 int pidfd_prepare(struct pid *pid, unsigned int flags, struct file **ret_file)
1926 {
1927 	struct file *pidfs_file;
1928 
1929 	/*
1930 	 * PIDFD_STALE is only allowed to be passed if the caller knows
1931 	 * that @pid is already registered in pidfs and thus
1932 	 * PIDFD_INFO_EXIT information is guaranteed to be available.
1933 	 */
1934 	if (!(flags & PIDFD_STALE)) {
1935 		/*
1936 		 * While holding the pidfd waitqueue lock removing the
1937 		 * task linkage for the thread-group leader pid
1938 		 * (PIDTYPE_TGID) isn't possible. Thus, if there's still
1939 		 * task linkage for PIDTYPE_PID not having thread-group
1940 		 * leader linkage for the pid means it wasn't a
1941 		 * thread-group leader in the first place.
1942 		 */
1943 		guard(spinlock_irq)(&pid->wait_pidfd.lock);
1944 
1945 		/* Task has already been reaped. */
1946 		if (!pid_has_task(pid, PIDTYPE_PID))
1947 			return -ESRCH;
1948 		/*
1949 		 * If this struct pid isn't used as a thread-group
1950 		 * leader but the caller requested to create a
1951 		 * thread-group leader pidfd then report ENOENT.
1952 		 */
1953 		if (!(flags & PIDFD_THREAD) && !pid_has_task(pid, PIDTYPE_TGID))
1954 			return -ENOENT;
1955 	}
1956 
1957 	CLASS(get_unused_fd, pidfd)(O_CLOEXEC);
1958 	if (pidfd < 0)
1959 		return pidfd;
1960 
1961 	pidfs_file = pidfs_alloc_file(pid, flags | O_RDWR);
1962 	if (IS_ERR(pidfs_file))
1963 		return PTR_ERR(pidfs_file);
1964 
1965 	*ret_file = pidfs_file;
1966 	return take_fd(pidfd);
1967 }
1968 
1969 static void __delayed_free_task(struct rcu_head *rhp)
1970 {
1971 	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1972 
1973 	free_task(tsk);
1974 }
1975 
1976 static __always_inline void delayed_free_task(struct task_struct *tsk)
1977 {
1978 	if (IS_ENABLED(CONFIG_MEMCG))
1979 		call_rcu(&tsk->rcu, __delayed_free_task);
1980 	else
1981 		free_task(tsk);
1982 }
1983 
1984 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1985 {
1986 	/* Skip if kernel thread */
1987 	if (!tsk->mm)
1988 		return;
1989 
1990 	/* Skip if spawning a thread or using vfork */
1991 	if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1992 		return;
1993 
1994 	/* We need to synchronize with __set_oom_adj */
1995 	mutex_lock(&oom_adj_mutex);
1996 	mm_flags_set(MMF_MULTIPROCESS, tsk->mm);
1997 	/* Update the values in case they were changed after copy_signal */
1998 	tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1999 	tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
2000 	mutex_unlock(&oom_adj_mutex);
2001 }
2002 
2003 #ifdef CONFIG_RV
2004 static void rv_task_fork(struct task_struct *p)
2005 {
2006 	memset(&p->rv, 0, sizeof(p->rv));
2007 }
2008 #else
2009 #define rv_task_fork(p) do {} while (0)
2010 #endif
2011 
2012 static bool need_futex_hash_allocate_default(u64 clone_flags)
2013 {
2014 	/*
2015 	 * Allocate a default futex hash for any sibling that will
2016 	 * share the parent's mm.
2017 	 */
2018 	return clone_flags & CLONE_VM;
2019 }
2020 
2021 /*
2022  * This creates a new process as a copy of the old one,
2023  * but does not actually start it yet.
2024  *
2025  * It copies the registers, and all the appropriate
2026  * parts of the process environment (as per the clone
2027  * flags). The actual kick-off is left to the caller.
2028  */
2029 __latent_entropy struct task_struct *copy_process(
2030 					struct pid *pid,
2031 					int trace,
2032 					int node,
2033 					struct kernel_clone_args *args)
2034 {
2035 	int pidfd = -1, retval;
2036 	struct task_struct *p;
2037 	struct multiprocess_signals delayed;
2038 	struct file *pidfile = NULL;
2039 	const u64 clone_flags = args->flags;
2040 	struct nsproxy *nsp = current->nsproxy;
2041 
2042 	/*
2043 	 * Don't allow sharing the root directory with processes in a different
2044 	 * namespace
2045 	 */
2046 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
2047 		return ERR_PTR(-EINVAL);
2048 
2049 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
2050 		return ERR_PTR(-EINVAL);
2051 
2052 	/*
2053 	 * Thread groups must share signals as well, and detached threads
2054 	 * can only be started up within the thread group.
2055 	 */
2056 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
2057 		return ERR_PTR(-EINVAL);
2058 
2059 	/*
2060 	 * Shared signal handlers imply shared VM. By way of the above,
2061 	 * thread groups also imply shared VM. Blocking this case allows
2062 	 * for various simplifications in other code.
2063 	 */
2064 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
2065 		return ERR_PTR(-EINVAL);
2066 
2067 	/*
2068 	 * Siblings of global init remain as zombies on exit since they are
2069 	 * not reaped by their parent (swapper). To solve this and to avoid
2070 	 * multi-rooted process trees, prevent global and container-inits
2071 	 * from creating siblings.
2072 	 */
2073 	if ((clone_flags & CLONE_PARENT) &&
2074 				current->signal->flags & SIGNAL_UNKILLABLE)
2075 		return ERR_PTR(-EINVAL);
2076 
2077 	/*
2078 	 * If the new process will be in a different pid or user namespace
2079 	 * do not allow it to share a thread group with the forking task.
2080 	 */
2081 	if (clone_flags & CLONE_THREAD) {
2082 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
2083 		    (task_active_pid_ns(current) != nsp->pid_ns_for_children))
2084 			return ERR_PTR(-EINVAL);
2085 	}
2086 
2087 	if (clone_flags & CLONE_PIDFD) {
2088 		/*
2089 		 * - CLONE_DETACHED is blocked so that we can potentially
2090 		 *   reuse it later for CLONE_PIDFD.
2091 		 */
2092 		if (clone_flags & CLONE_DETACHED)
2093 			return ERR_PTR(-EINVAL);
2094 	}
2095 
2096 	if (clone_flags & CLONE_AUTOREAP) {
2097 		if (clone_flags & CLONE_THREAD)
2098 			return ERR_PTR(-EINVAL);
2099 		if (clone_flags & CLONE_PARENT)
2100 			return ERR_PTR(-EINVAL);
2101 		if (args->exit_signal)
2102 			return ERR_PTR(-EINVAL);
2103 	}
2104 
2105 	if ((clone_flags & CLONE_PARENT) && current->signal->autoreap)
2106 		return ERR_PTR(-EINVAL);
2107 
2108 	if (clone_flags & CLONE_NNP) {
2109 		if (clone_flags & CLONE_THREAD)
2110 			return ERR_PTR(-EINVAL);
2111 	}
2112 
2113 	if (clone_flags & CLONE_PIDFD_AUTOKILL) {
2114 		if (!(clone_flags & CLONE_PIDFD))
2115 			return ERR_PTR(-EINVAL);
2116 		if (!(clone_flags & CLONE_AUTOREAP))
2117 			return ERR_PTR(-EINVAL);
2118 		if (clone_flags & CLONE_THREAD)
2119 			return ERR_PTR(-EINVAL);
2120 		/*
2121 		 * Without CLONE_NNP the child could escalate privileges
2122 		 * after being spawned, so require CAP_SYS_ADMIN.
2123 		 * With CLONE_NNP the child can't gain new privileges,
2124 		 * so allow unprivileged usage.
2125 		 */
2126 		if (!(clone_flags & CLONE_NNP) &&
2127 		    !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
2128 			return ERR_PTR(-EPERM);
2129 	}
2130 
2131 	/*
2132 	 * Force any signals received before this point to be delivered
2133 	 * before the fork happens.  Collect up signals sent to multiple
2134 	 * processes that happen during the fork and delay them so that
2135 	 * they appear to happen after the fork.
2136 	 */
2137 	sigemptyset(&delayed.signal);
2138 	INIT_HLIST_NODE(&delayed.node);
2139 
2140 	spin_lock_irq(&current->sighand->siglock);
2141 	if (!(clone_flags & CLONE_THREAD))
2142 		hlist_add_head(&delayed.node, &current->signal->multiprocess);
2143 	recalc_sigpending();
2144 	spin_unlock_irq(&current->sighand->siglock);
2145 	retval = -ERESTARTNOINTR;
2146 	if (task_sigpending(current))
2147 		goto fork_out;
2148 
2149 	retval = -ENOMEM;
2150 	p = dup_task_struct(current, node);
2151 	if (!p)
2152 		goto fork_out;
2153 	/*
2154 	 * Must run before the first fallible op, so error paths never
2155 	 * free the parent's ret_stack.
2156 	 */
2157 	ftrace_graph_init_task(p);
2158 	retval = copy_exec_state(clone_flags, p);
2159 	if (retval)
2160 		goto bad_fork_free;
2161 	p->flags &= ~PF_KTHREAD;
2162 	if (args->kthread)
2163 		p->flags |= PF_KTHREAD;
2164 	if (args->user_worker) {
2165 		/* A user worker takes only the signals nobody can block. */
2166 		p->flags |= PF_USER_WORKER;
2167 		siginitsetinv(&p->blocked, SIG_KERNEL_ONLY_MASK);
2168 	}
2169 	if (args->io_thread)
2170 		p->flags |= PF_IO_WORKER;
2171 
2172 	if (args->name)
2173 		strscpy_pad(p->comm, args->name, sizeof(p->comm));
2174 
2175 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
2176 	/*
2177 	 * TID is cleared in mm_release() when the task exits
2178 	 */
2179 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
2180 
2181 	rt_mutex_init_task(p);
2182 	raw_spin_lock_init(&p->blocked_lock);
2183 
2184 	lockdep_assert_irqs_enabled();
2185 #ifdef CONFIG_PROVE_LOCKING
2186 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
2187 #endif
2188 	retval = copy_creds(p, clone_flags);
2189 	if (retval < 0)
2190 		goto bad_fork_free;
2191 
2192 	retval = -EAGAIN;
2193 	if (is_rlimit_overlimit(task_ucounts(p), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
2194 		if (p->real_cred->user != INIT_USER &&
2195 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
2196 			goto bad_fork_cleanup_count;
2197 	}
2198 	current->flags &= ~PF_NPROC_EXCEEDED;
2199 
2200 	/*
2201 	 * If multiple threads are within copy_process(), then this check
2202 	 * triggers too late. This doesn't hurt, the check is only there
2203 	 * to stop root fork bombs.
2204 	 */
2205 	retval = -EAGAIN;
2206 	if (data_race(nr_threads >= max_threads))
2207 		goto bad_fork_cleanup_count;
2208 
2209 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
2210 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE | PF_NO_SETAFFINITY);
2211 	p->flags |= PF_FORKNOEXEC;
2212 	INIT_LIST_HEAD(&p->children);
2213 	INIT_LIST_HEAD(&p->sibling);
2214 	rcu_copy_process(p);
2215 	p->vfork_done = NULL;
2216 	/* Set by copy_files(), exit_files() on the error path skips NULL. */
2217 	p->files = NULL;
2218 	spin_lock_init(&p->alloc_lock);
2219 
2220 	init_sigpending(&p->pending);
2221 
2222 	p->utime = p->stime = p->gtime = 0;
2223 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2224 	p->utimescaled = p->stimescaled = 0;
2225 #endif
2226 	prev_cputime_init(&p->prev_cputime);
2227 
2228 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2229 	seqcount_init(&p->vtime.seqcount);
2230 	p->vtime.starttime = 0;
2231 	p->vtime.state = VTIME_INACTIVE;
2232 #endif
2233 
2234 #ifdef CONFIG_IO_URING
2235 	p->io_uring = NULL;
2236 	retval = io_uring_fork(p);
2237 	if (unlikely(retval))
2238 		goto bad_fork_cleanup_delayacct;
2239 	retval = -EAGAIN;
2240 #endif
2241 
2242 	p->default_timer_slack_ns = current->timer_slack_ns;
2243 
2244 #ifdef CONFIG_PSI
2245 	p->psi_flags = 0;
2246 #endif
2247 
2248 	task_io_accounting_init(&p->ioac);
2249 	acct_clear_integrals(p);
2250 
2251 	posix_cputimers_init(&p->posix_cputimers);
2252 	tick_dep_init_task(p);
2253 
2254 	p->io_context = NULL;
2255 	audit_set_context(p, NULL);
2256 	cgroup_fork(p);
2257 	if (args->kthread) {
2258 		if (!set_kthread_struct(p))
2259 			goto bad_fork_cleanup_delayacct;
2260 	}
2261 #ifdef CONFIG_NUMA
2262 	p->mempolicy = mpol_dup(p->mempolicy);
2263 	if (IS_ERR(p->mempolicy)) {
2264 		retval = PTR_ERR(p->mempolicy);
2265 		p->mempolicy = NULL;
2266 		goto bad_fork_cleanup_delayacct;
2267 	}
2268 #endif
2269 #ifdef CONFIG_CPUSETS
2270 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2271 	seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2272 #endif
2273 #ifdef CONFIG_TRACE_IRQFLAGS
2274 	memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2275 	p->irqtrace.hardirq_disable_ip	= _THIS_IP_;
2276 	p->irqtrace.softirq_enable_ip	= _THIS_IP_;
2277 	p->softirqs_enabled		= 1;
2278 	p->softirq_context		= 0;
2279 #endif
2280 
2281 	p->pagefault_disabled = 0;
2282 
2283 	lockdep_init_task(p);
2284 
2285 	p->blocked_on = NULL; /* not blocked yet */
2286 	p->blocked_donor = NULL; /* nobody is boosting p yet */
2287 
2288 #ifdef CONFIG_BCACHE
2289 	p->sequential_io	= 0;
2290 	p->sequential_io_avg	= 0;
2291 #endif
2292 
2293 	unwind_task_init(p);
2294 
2295 	/* Perform scheduler related setup. Assign this task to a CPU. */
2296 	retval = sched_fork(clone_flags, p);
2297 	if (retval)
2298 		goto bad_fork_cleanup_policy;
2299 
2300 	retval = perf_event_init_task(p, clone_flags);
2301 	if (retval)
2302 		goto bad_fork_sched_cancel_fork;
2303 	retval = audit_alloc(p);
2304 	if (retval)
2305 		goto bad_fork_cleanup_perf;
2306 	/* copy all the process information */
2307 	shm_init_task(p);
2308 	retval = security_task_alloc(p, clone_flags);
2309 	if (retval)
2310 		goto bad_fork_cleanup_audit;
2311 	retval = copy_semundo(clone_flags, p);
2312 	if (retval)
2313 		goto bad_fork_cleanup_security;
2314 	retval = copy_files(clone_flags, p, args->no_files);
2315 	if (retval)
2316 		goto bad_fork_cleanup_semundo;
2317 	retval = copy_fs(clone_flags, p,
2318 			 (args->umh ? COPY_FS_UMH : 0) |
2319 			 (args->idle ? COPY_FS_IDLE : 0));
2320 	if (retval)
2321 		goto bad_fork_cleanup_semundo;
2322 	retval = copy_sighand(clone_flags, p);
2323 	if (retval)
2324 		goto bad_fork_cleanup_fs;
2325 	retval = copy_signal(clone_flags, p);
2326 	if (retval)
2327 		goto bad_fork_cleanup_sighand;
2328 	retval = copy_mm(clone_flags, p);
2329 	if (retval)
2330 		goto bad_fork_cleanup_signal;
2331 	retval = copy_namespaces(clone_flags, p);
2332 	if (retval)
2333 		goto bad_fork_cleanup_mm;
2334 	retval = copy_io(clone_flags, p);
2335 	if (retval)
2336 		goto bad_fork_cleanup_namespaces;
2337 	retval = copy_thread(p, args);
2338 	if (retval)
2339 		goto bad_fork_cleanup_io;
2340 
2341 	stackleak_task_init(p);
2342 
2343 	if (pid != &init_struct_pid) {
2344 		pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2345 				args->set_tid_size);
2346 		if (IS_ERR(pid)) {
2347 			retval = PTR_ERR(pid);
2348 			goto bad_fork_cleanup_thread;
2349 		}
2350 	}
2351 
2352 	/*
2353 	 * This has to happen after we've potentially unshared the file
2354 	 * descriptor table (so that the pidfd doesn't leak into the child
2355 	 * if the fd table isn't shared).
2356 	 */
2357 	if (clone_flags & CLONE_PIDFD) {
2358 		unsigned flags = PIDFD_STALE;
2359 
2360 		if (clone_flags & CLONE_THREAD)
2361 			flags |= PIDFD_THREAD;
2362 		if (clone_flags & CLONE_PIDFD_AUTOKILL)
2363 			flags |= PIDFD_AUTOKILL;
2364 
2365 		/*
2366 		 * Note that no task has been attached to @pid yet indicate
2367 		 * that via CLONE_PIDFD.
2368 		 */
2369 		retval = pidfd_prepare(pid, flags, &pidfile);
2370 		if (retval < 0)
2371 			goto bad_fork_free_pid;
2372 		pidfd = retval;
2373 
2374 		retval = put_user(pidfd, args->pidfd);
2375 		if (retval)
2376 			goto bad_fork_put_pidfd;
2377 	}
2378 
2379 #ifdef CONFIG_BLOCK
2380 	p->plug = NULL;
2381 	p->flags &= ~PF_BLOCK_TS;
2382 #endif
2383 	futex_init_task(p);
2384 
2385 	/*
2386 	 * sigaltstack should be cleared when sharing the same VM
2387 	 */
2388 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2389 		sas_ss_reset(p);
2390 
2391 	/*
2392 	 * Syscall tracing and stepping should be turned off in the
2393 	 * child regardless of CLONE_PTRACE.
2394 	 */
2395 	user_disable_single_step(p);
2396 	clear_task_syscall_work(p, SYSCALL_TRACE);
2397 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2398 	clear_task_syscall_work(p, SYSCALL_EMU);
2399 #endif
2400 	clear_tsk_latency_tracing(p);
2401 
2402 	/* ok, now we should be set up.. */
2403 	p->pid = pid_nr(pid);
2404 	if (clone_flags & CLONE_THREAD) {
2405 		p->group_leader = current->group_leader;
2406 		p->tgid = current->tgid;
2407 	} else {
2408 		p->group_leader = p;
2409 		p->tgid = p->pid;
2410 	}
2411 
2412 	p->nr_dirtied = 0;
2413 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2414 	p->dirty_paused_when = 0;
2415 
2416 	p->pdeath_signal = 0;
2417 	p->task_works = NULL;
2418 	clear_posix_cputimers_work(p);
2419 
2420 #ifdef CONFIG_KRETPROBES
2421 	p->kretprobe_instances.first = NULL;
2422 #endif
2423 #ifdef CONFIG_RETHOOK
2424 	p->rethooks.first = NULL;
2425 #endif
2426 
2427 	/*
2428 	 * Ensure that the cgroup subsystem policies allow the new process to be
2429 	 * forked. It should be noted that the new process's css_set can be changed
2430 	 * between here and cgroup_post_fork() if an organisation operation is in
2431 	 * progress.
2432 	 */
2433 	retval = cgroup_can_fork(p, args);
2434 	if (retval)
2435 		goto bad_fork_put_pidfd;
2436 
2437 	/*
2438 	 * Now that the cgroups are pinned, re-clone the parent cgroup and put
2439 	 * the new task on the correct runqueue. All this *before* the task
2440 	 * becomes visible.
2441 	 *
2442 	 * This isn't part of ->can_fork() because while the re-cloning is
2443 	 * cgroup specific, it unconditionally needs to place the task on a
2444 	 * runqueue.
2445 	 */
2446 	retval = sched_cgroup_fork(p, args);
2447 	if (retval)
2448 		goto bad_fork_cancel_cgroup;
2449 
2450 	if (need_futex_hash_allocate_default(clone_flags)) {
2451 		retval = futex_hash_allocate_default();
2452 		if (retval)
2453 			goto bad_fork_cancel_cgroup;
2454 		/*
2455 		 * If we fail beyond this point we don't free the allocated
2456 		 * futex hash map. We assume that another thread will be created
2457 		 * and makes use of it. The hash map will be freed once the main
2458 		 * thread terminates.
2459 		 */
2460 	}
2461 	/*
2462 	 * From this point on we must avoid any synchronous user-space
2463 	 * communication until we take the tasklist-lock. In particular, we do
2464 	 * not want user-space to be able to predict the process start-time by
2465 	 * stalling fork(2) after we recorded the start_time but before it is
2466 	 * visible to the system.
2467 	 */
2468 
2469 	p->start_time = ktime_get_ns();
2470 	p->start_boottime = ktime_get_boottime_ns();
2471 
2472 	/*
2473 	 * Make it visible to the rest of the system, but dont wake it up yet.
2474 	 * Need tasklist lock for parent etc handling!
2475 	 */
2476 	write_lock_irq(&tasklist_lock);
2477 
2478 	/* CLONE_PARENT re-uses the old parent */
2479 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2480 		p->real_parent = current->real_parent;
2481 		p->parent_exec_id = current->parent_exec_id;
2482 		if (clone_flags & CLONE_THREAD)
2483 			p->exit_signal = -1;
2484 		else
2485 			p->exit_signal = current->group_leader->exit_signal;
2486 	} else {
2487 		p->real_parent = current;
2488 		p->parent_exec_id = current->self_exec_id;
2489 		p->exit_signal = args->exit_signal;
2490 	}
2491 
2492 	klp_copy_process(p);
2493 
2494 	sched_core_fork(p);
2495 
2496 	spin_lock(&current->sighand->siglock);
2497 
2498 	rv_task_fork(p);
2499 
2500 	rseq_fork(p, clone_flags);
2501 
2502 	/*
2503 	 * If zap_pid_ns_processes() was called after alloc_pid(), the new
2504 	 * child missed SIGKILL.  If current is not in the same namespace,
2505 	 * we can't rely on fatal_signal_pending() below.
2506 	 */
2507 	if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2508 		retval = -ENOMEM;
2509 		goto bad_fork_core_free;
2510 	}
2511 
2512 	/* Let kill terminate clone/fork in the middle */
2513 	if (fatal_signal_pending(current)) {
2514 		retval = -EINTR;
2515 		goto bad_fork_core_free;
2516 	}
2517 
2518 	/* No more failure paths after this point. */
2519 
2520 	/*
2521 	 * One for the user space visible state that goes away when reaped.
2522 	 * One for the scheduler.
2523 	 */
2524 	refcount_set(&p->rcu_users, 2);
2525 
2526 	/*
2527 	 * Copy seccomp details explicitly here, in case they were changed
2528 	 * before holding sighand lock.
2529 	 */
2530 	copy_seccomp(p);
2531 
2532 	if (clone_flags & CLONE_NNP)
2533 		task_set_no_new_privs(p);
2534 
2535 	init_task_pid_links(p);
2536 	if (likely(p->pid)) {
2537 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2538 
2539 		init_task_pid(p, PIDTYPE_PID, pid);
2540 		if (thread_group_leader(p)) {
2541 			init_task_pid(p, PIDTYPE_TGID, pid);
2542 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2543 			init_task_pid(p, PIDTYPE_SID, task_session(current));
2544 
2545 			if (is_child_reaper(pid)) {
2546 				struct pid_namespace *ns = ns_of_pid(pid);
2547 
2548 				ASSERT_EXCLUSIVE_WRITER(ns->child_reaper);
2549 				WRITE_ONCE(ns->child_reaper, p);
2550 				p->signal->flags |= SIGNAL_UNKILLABLE;
2551 			}
2552 			p->signal->shared_pending.signal = delayed.signal;
2553 			p->signal->tty = tty_kref_get(current->signal->tty);
2554 			/*
2555 			 * Inherit has_child_subreaper flag under the same
2556 			 * tasklist_lock with adding child to the process tree
2557 			 * for propagate_has_child_subreaper optimization.
2558 			 */
2559 			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2560 							 p->real_parent->signal->is_child_subreaper;
2561 			if (clone_flags & CLONE_AUTOREAP)
2562 				p->signal->autoreap = 1;
2563 			list_add_tail(&p->sibling, &p->real_parent->children);
2564 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
2565 			attach_pid(p, PIDTYPE_TGID);
2566 			attach_pid(p, PIDTYPE_PGID);
2567 			attach_pid(p, PIDTYPE_SID);
2568 			__this_cpu_inc(process_counts);
2569 		} else {
2570 			current->signal->nr_threads++;
2571 			current->signal->quick_threads++;
2572 			atomic_inc(&current->signal->live);
2573 			refcount_inc(&current->signal->sigcnt);
2574 			task_join_group_stop(p);
2575 			list_add_tail_rcu(&p->thread_node,
2576 					  &p->signal->thread_head);
2577 		}
2578 		attach_pid(p, PIDTYPE_PID);
2579 		nr_threads++;
2580 	}
2581 	total_forks++;
2582 	hlist_del_init(&delayed.node);
2583 	spin_unlock(&current->sighand->siglock);
2584 	syscall_tracepoint_update(p);
2585 	write_unlock_irq(&tasklist_lock);
2586 
2587 	if (pidfile)
2588 		fd_install(pidfd, pidfile);
2589 
2590 	proc_fork_connector(p);
2591 	/*
2592 	 * sched_ext needs @p to be associated with its cgroup in its post_fork
2593 	 * hook. cgroup_post_fork() should come before sched_post_fork().
2594 	 */
2595 	cgroup_post_fork(p, args);
2596 	sched_post_fork(p);
2597 	perf_event_fork(p);
2598 
2599 	trace_task_newtask(p, clone_flags);
2600 	uprobe_copy_process(p, clone_flags);
2601 	user_events_fork(p, clone_flags);
2602 
2603 	copy_oom_score_adj(clone_flags, p);
2604 
2605 	return p;
2606 
2607 bad_fork_core_free:
2608 	sched_core_free(p);
2609 	spin_unlock(&current->sighand->siglock);
2610 	write_unlock_irq(&tasklist_lock);
2611 bad_fork_cancel_cgroup:
2612 	cgroup_cancel_fork(p, args);
2613 bad_fork_put_pidfd:
2614 	if (clone_flags & CLONE_PIDFD) {
2615 		fput(pidfile);
2616 		put_unused_fd(pidfd);
2617 	}
2618 bad_fork_free_pid:
2619 	if (pid != &init_struct_pid)
2620 		free_pid(pid);
2621 bad_fork_cleanup_thread:
2622 	exit_thread(p);
2623 bad_fork_cleanup_io:
2624 	if (p->io_context)
2625 		exit_io_context(p);
2626 bad_fork_cleanup_namespaces:
2627 	exit_nsproxy_namespaces(p);
2628 bad_fork_cleanup_mm:
2629 	sched_cache_fork_cleanup(p);
2630 	if (p->mm) {
2631 		mm_clear_owner(p->mm, p);
2632 		mmput(p->mm);
2633 	}
2634 bad_fork_cleanup_signal:
2635 	if (!(clone_flags & CLONE_THREAD))
2636 		free_signal_struct(p->signal);
2637 bad_fork_cleanup_sighand:
2638 	__cleanup_sighand(p->sighand);
2639 bad_fork_cleanup_fs:
2640 	exit_fs(p); /* blocking */
2641 bad_fork_cleanup_semundo:
2642 	exit_sem(p);
2643 bad_fork_cleanup_security:
2644 	security_task_free(p);
2645 bad_fork_cleanup_audit:
2646 	audit_free(p);
2647 bad_fork_cleanup_perf:
2648 	perf_event_free_task(p);
2649 bad_fork_sched_cancel_fork:
2650 	sched_cancel_fork(p);
2651 	/* ->release() of a file may need scx_fork_rwsem for write. */
2652 	exit_files(p); /* blocking */
2653 bad_fork_cleanup_policy:
2654 	lockdep_free_task(p);
2655 #ifdef CONFIG_NUMA
2656 	mpol_put(p->mempolicy);
2657 #endif
2658 bad_fork_cleanup_delayacct:
2659 	io_uring_free(p);
2660 	delayacct_tsk_free(p);
2661 bad_fork_cleanup_count:
2662 	dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
2663 	exit_cred_namespaces(p);
2664 	exit_creds(p);
2665 bad_fork_free:
2666 	WRITE_ONCE(p->__state, TASK_DEAD);
2667 	exit_task_stack_account(p);
2668 	put_task_stack(p);
2669 	delayed_free_task(p);
2670 fork_out:
2671 	spin_lock_irq(&current->sighand->siglock);
2672 	hlist_del_init(&delayed.node);
2673 	spin_unlock_irq(&current->sighand->siglock);
2674 	return ERR_PTR(retval);
2675 }
2676 
2677 static inline void init_idle_pids(struct task_struct *idle)
2678 {
2679 	enum pid_type type;
2680 
2681 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2682 		INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2683 		init_task_pid(idle, type, &init_struct_pid);
2684 	}
2685 }
2686 
2687 static int idle_dummy(void *dummy)
2688 {
2689 	/* This function is never called */
2690 	return 0;
2691 }
2692 
2693 struct task_struct * __init fork_idle(int cpu)
2694 {
2695 	struct task_struct *task;
2696 	struct kernel_clone_args args = {
2697 		.flags		= CLONE_VM | CLONE_FS,
2698 		.fn		= &idle_dummy,
2699 		.fn_arg		= NULL,
2700 		.kthread	= 1,
2701 		.idle		= 1,
2702 	};
2703 
2704 	task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2705 	if (!IS_ERR(task)) {
2706 		init_idle_pids(task);
2707 		init_idle(task, cpu);
2708 	}
2709 
2710 	return task;
2711 }
2712 
2713 /*
2714  * This is like kernel_clone(), but shaved down and tailored to just
2715  * creating io_uring workers. It returns a created task, or an error pointer.
2716  * The returned task is inactive, and the caller must fire it up through
2717  * wake_up_new_task(p). All signals are blocked in the created task.
2718  */
2719 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2720 {
2721 	unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2722 			      CLONE_IO|CLONE_VM|CLONE_UNTRACED;
2723 	struct kernel_clone_args args = {
2724 		.flags		= flags,
2725 		.fn		= fn,
2726 		.fn_arg		= arg,
2727 		.io_thread	= 1,
2728 		.user_worker	= 1,
2729 	};
2730 
2731 	/* A creator past its fatal signal or its coredump point gets no thread. */
2732 	if (current->flags & (PF_SIGNALED | PF_POSTCOREDUMP))
2733 		return ERR_PTR(-EINTR);
2734 
2735 	return copy_process(NULL, 0, node, &args);
2736 }
2737 
2738 /*
2739  *  Ok, this is the main fork-routine.
2740  *
2741  * It copies the process, and if successful kick-starts
2742  * it and waits for it to finish using the VM if required.
2743  */
2744 pid_t kernel_clone(struct kernel_clone_args *args)
2745 {
2746 	u64 clone_flags = args->flags;
2747 	struct completion vfork;
2748 	struct pid *pid;
2749 	struct task_struct *p;
2750 	int trace = 0;
2751 	pid_t nr;
2752 
2753 	/*
2754 	 * Creating an empty mount namespace implies creating a new mount
2755 	 * namespace.  Set this before copy_process() so that the
2756 	 * CLONE_NEWNS|CLONE_FS mutual exclusion check works correctly.
2757 	 */
2758 	if (clone_flags & CLONE_EMPTY_MNTNS) {
2759 		clone_flags |= CLONE_NEWNS;
2760 		args->flags = clone_flags;
2761 	}
2762 
2763 	/*
2764 	 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2765 	 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2766 	 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2767 	 * field in struct clone_args and it still doesn't make sense to have
2768 	 * them both point at the same memory location. Performing this check
2769 	 * here has the advantage that we don't need to have a separate helper
2770 	 * to check for legacy clone().
2771 	 */
2772 	if ((clone_flags & CLONE_PIDFD) &&
2773 	    (clone_flags & CLONE_PARENT_SETTID) &&
2774 	    (args->pidfd == args->parent_tid))
2775 		return -EINVAL;
2776 
2777 	if (!valid_signal(args->exit_signal))
2778 		return -EINVAL;
2779 
2780 	/*
2781 	 * Determine whether and which event to report to ptracer.  When
2782 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
2783 	 * requested, no event is reported; otherwise, report if the event
2784 	 * for the type of forking is enabled.
2785 	 */
2786 	if (!(clone_flags & CLONE_UNTRACED)) {
2787 		if (clone_flags & CLONE_VFORK)
2788 			trace = PTRACE_EVENT_VFORK;
2789 		else if (args->exit_signal != SIGCHLD)
2790 			trace = PTRACE_EVENT_CLONE;
2791 		else
2792 			trace = PTRACE_EVENT_FORK;
2793 
2794 		if (likely(!ptrace_event_enabled(current, trace)))
2795 			trace = 0;
2796 	}
2797 
2798 	p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2799 	add_latent_entropy();
2800 
2801 	if (IS_ERR(p))
2802 		return PTR_ERR(p);
2803 
2804 	/*
2805 	 * Do this prior waking up the new thread - the thread pointer
2806 	 * might get invalid after that point, if the thread exits quickly.
2807 	 */
2808 	trace_sched_process_fork(current, p);
2809 
2810 	pid = get_task_pid(p, PIDTYPE_PID);
2811 	nr = pid_vnr(pid);
2812 
2813 	if (clone_flags & CLONE_PARENT_SETTID)
2814 		put_user(nr, args->parent_tid);
2815 
2816 	if (clone_flags & CLONE_VFORK) {
2817 		p->vfork_done = &vfork;
2818 		init_completion(&vfork);
2819 		get_task_struct(p);
2820 	}
2821 
2822 	if (IS_ENABLED(CONFIG_LRU_GEN_WALKS_MMU) && !(clone_flags & CLONE_VM)) {
2823 		/* lock the task to synchronize with memcg migration */
2824 		task_lock(p);
2825 		lru_gen_add_mm(p->mm);
2826 		task_unlock(p);
2827 	}
2828 
2829 	wake_up_new_task(p);
2830 
2831 	/* forking complete and child started to run, tell ptracer */
2832 	if (unlikely(trace))
2833 		ptrace_event_pid(trace, pid);
2834 
2835 	if (clone_flags & CLONE_VFORK) {
2836 		if (!wait_for_vfork_done(p, &vfork))
2837 			ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2838 	}
2839 
2840 	put_pid(pid);
2841 	return nr;
2842 }
2843 
2844 /*
2845  * Create a kernel thread.
2846  */
2847 pid_t kernel_thread(int (*fn)(void *), void *arg, const char *name,
2848 		    unsigned long flags)
2849 {
2850 	struct kernel_clone_args args = {
2851 		.flags		= ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2852 		.exit_signal	= (flags & CSIGNAL),
2853 		.fn		= fn,
2854 		.fn_arg		= arg,
2855 		.name		= name,
2856 		.kthread	= 1,
2857 	};
2858 
2859 	return kernel_clone(&args);
2860 }
2861 
2862 /*
2863  * Create a user mode thread.
2864  */
2865 pid_t user_mode_thread(int (*fn)(void *), void *arg, unsigned long flags)
2866 {
2867 	struct kernel_clone_args args = {
2868 		.flags		= ((flags | CLONE_VM | CLONE_UNTRACED) & ~CSIGNAL),
2869 		.exit_signal	= (flags & CSIGNAL),
2870 		.fn		= fn,
2871 		.fn_arg		= arg,
2872 		.umh		= 1,
2873 	};
2874 
2875 	return kernel_clone(&args);
2876 }
2877 
2878 #ifdef __ARCH_WANT_SYS_FORK
2879 SYSCALL_DEFINE0(fork)
2880 {
2881 #ifdef CONFIG_MMU
2882 	struct kernel_clone_args args = {
2883 		.exit_signal = SIGCHLD,
2884 	};
2885 
2886 	return kernel_clone(&args);
2887 #else
2888 	/* can not support in nommu mode */
2889 	return -EINVAL;
2890 #endif
2891 }
2892 #endif
2893 
2894 #ifdef __ARCH_WANT_SYS_VFORK
2895 SYSCALL_DEFINE0(vfork)
2896 {
2897 	struct kernel_clone_args args = {
2898 		.flags		= CLONE_VFORK | CLONE_VM,
2899 		.exit_signal	= SIGCHLD,
2900 	};
2901 
2902 	return kernel_clone(&args);
2903 }
2904 #endif
2905 
2906 #ifdef __ARCH_WANT_SYS_CLONE
2907 #ifdef CONFIG_CLONE_BACKWARDS
2908 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2909 		 int __user *, parent_tidptr,
2910 		 unsigned long, tls,
2911 		 int __user *, child_tidptr)
2912 #elif defined(CONFIG_CLONE_BACKWARDS2)
2913 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2914 		 int __user *, parent_tidptr,
2915 		 int __user *, child_tidptr,
2916 		 unsigned long, tls)
2917 #elif defined(CONFIG_CLONE_BACKWARDS3)
2918 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2919 		int, stack_size,
2920 		int __user *, parent_tidptr,
2921 		int __user *, child_tidptr,
2922 		unsigned long, tls)
2923 #else
2924 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2925 		 int __user *, parent_tidptr,
2926 		 int __user *, child_tidptr,
2927 		 unsigned long, tls)
2928 #endif
2929 {
2930 	struct kernel_clone_args args = {
2931 		.flags		= (lower_32_bits(clone_flags) & ~CSIGNAL),
2932 		.pidfd		= parent_tidptr,
2933 		.child_tid	= child_tidptr,
2934 		.parent_tid	= parent_tidptr,
2935 		.exit_signal	= (lower_32_bits(clone_flags) & CSIGNAL),
2936 		.stack		= newsp,
2937 		.tls		= tls,
2938 	};
2939 
2940 	return kernel_clone(&args);
2941 }
2942 #endif
2943 
2944 static noinline int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2945 					      struct clone_args __user *uargs,
2946 					      size_t usize)
2947 {
2948 	int err;
2949 	struct clone_args args;
2950 	pid_t *kset_tid = kargs->set_tid;
2951 
2952 	BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2953 		     CLONE_ARGS_SIZE_VER0);
2954 	BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2955 		     CLONE_ARGS_SIZE_VER1);
2956 	BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2957 		     CLONE_ARGS_SIZE_VER2);
2958 	BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2959 
2960 	if (unlikely(usize > PAGE_SIZE))
2961 		return -E2BIG;
2962 	if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2963 		return -EINVAL;
2964 
2965 	err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2966 	if (err)
2967 		return err;
2968 
2969 	if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2970 		return -EINVAL;
2971 
2972 	if (unlikely(!args.set_tid && args.set_tid_size > 0))
2973 		return -EINVAL;
2974 
2975 	if (unlikely(args.set_tid && args.set_tid_size == 0))
2976 		return -EINVAL;
2977 
2978 	/*
2979 	 * Verify that higher 32bits of exit_signal are unset
2980 	 */
2981 	if (unlikely(args.exit_signal & ~((u64)CSIGNAL)))
2982 		return -EINVAL;
2983 
2984 	if ((args.flags & CLONE_INTO_CGROUP) &&
2985 	    (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2986 		return -EINVAL;
2987 
2988 	*kargs = (struct kernel_clone_args){
2989 		.flags		= args.flags,
2990 		.pidfd		= u64_to_user_ptr(args.pidfd),
2991 		.child_tid	= u64_to_user_ptr(args.child_tid),
2992 		.parent_tid	= u64_to_user_ptr(args.parent_tid),
2993 		.exit_signal	= args.exit_signal,
2994 		.stack		= args.stack,
2995 		.stack_size	= args.stack_size,
2996 		.tls		= args.tls,
2997 		.set_tid_size	= args.set_tid_size,
2998 		.cgroup		= args.cgroup,
2999 	};
3000 
3001 	if (args.set_tid &&
3002 		copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
3003 			(kargs->set_tid_size * sizeof(pid_t))))
3004 		return -EFAULT;
3005 
3006 	kargs->set_tid = kset_tid;
3007 
3008 	return 0;
3009 }
3010 
3011 /**
3012  * clone3_stack_valid - check and prepare stack
3013  * @kargs: kernel clone args
3014  *
3015  * Verify that the stack arguments userspace gave us are sane.
3016  * In addition, set the stack direction for userspace since it's easy for us to
3017  * determine.
3018  */
3019 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
3020 {
3021 	if (kargs->stack == 0) {
3022 		if (kargs->stack_size > 0)
3023 			return false;
3024 	} else {
3025 		if (kargs->stack_size == 0)
3026 			return false;
3027 
3028 		if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
3029 			return false;
3030 
3031 #if !defined(CONFIG_STACK_GROWSUP)
3032 		kargs->stack += kargs->stack_size;
3033 #endif
3034 	}
3035 
3036 	return true;
3037 }
3038 
3039 static bool clone3_args_valid(struct kernel_clone_args *kargs)
3040 {
3041 	/* Verify that no unknown flags are passed along. */
3042 	if (kargs->flags &
3043 	    ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND |
3044 	      CLONE_INTO_CGROUP | CLONE_AUTOREAP | CLONE_NNP |
3045 	      CLONE_PIDFD_AUTOKILL | CLONE_EMPTY_MNTNS))
3046 		return false;
3047 
3048 	/*
3049 	 * - make the CLONE_DETACHED bit reusable for clone3
3050 	 * - make the CSIGNAL bits reusable for clone3
3051 	 */
3052 	if (kargs->flags & (CLONE_DETACHED | (CSIGNAL & (~CLONE_NEWTIME))))
3053 		return false;
3054 
3055 	if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
3056 	    (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
3057 		return false;
3058 
3059 	if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
3060 	    kargs->exit_signal)
3061 		return false;
3062 
3063 	if (!clone3_stack_valid(kargs))
3064 		return false;
3065 
3066 	return true;
3067 }
3068 
3069 /**
3070  * sys_clone3 - create a new process with specific properties
3071  * @uargs: argument structure
3072  * @size:  size of @uargs
3073  *
3074  * clone3() is the extensible successor to clone()/clone2().
3075  * It takes a struct as argument that is versioned by its size.
3076  *
3077  * Return: On success, a positive PID for the child process.
3078  *         On error, a negative errno number.
3079  */
3080 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
3081 {
3082 	int err;
3083 
3084 	struct kernel_clone_args kargs;
3085 	pid_t set_tid[MAX_PID_NS_LEVEL];
3086 
3087 #ifdef __ARCH_BROKEN_SYS_CLONE3
3088 #warning clone3() entry point is missing, please fix
3089 	return -ENOSYS;
3090 #endif
3091 
3092 	kargs.set_tid = set_tid;
3093 
3094 	err = copy_clone_args_from_user(&kargs, uargs, size);
3095 	if (err)
3096 		return err;
3097 
3098 	if (!clone3_args_valid(&kargs))
3099 		return -EINVAL;
3100 
3101 	return kernel_clone(&kargs);
3102 }
3103 
3104 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
3105 {
3106 	struct task_struct *leader, *parent, *child;
3107 	int res;
3108 
3109 	read_lock(&tasklist_lock);
3110 	leader = top = top->group_leader;
3111 down:
3112 	for_each_thread(leader, parent) {
3113 		list_for_each_entry(child, &parent->children, sibling) {
3114 			res = visitor(child, data);
3115 			if (res) {
3116 				if (res < 0)
3117 					goto out;
3118 				leader = child;
3119 				goto down;
3120 			}
3121 up:
3122 			;
3123 		}
3124 	}
3125 
3126 	if (leader != top) {
3127 		child = leader;
3128 		parent = child->real_parent;
3129 		leader = parent->group_leader;
3130 		goto up;
3131 	}
3132 out:
3133 	read_unlock(&tasklist_lock);
3134 }
3135 
3136 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
3137 #define ARCH_MIN_MMSTRUCT_ALIGN 0
3138 #endif
3139 
3140 static void sighand_ctor(void *data)
3141 {
3142 	struct sighand_struct *sighand = data;
3143 
3144 	spin_lock_init(&sighand->siglock);
3145 	init_waitqueue_head(&sighand->signalfd_wqh);
3146 }
3147 
3148 void __init mm_cache_init(void)
3149 {
3150 	unsigned int mm_size;
3151 
3152 	/*
3153 	 * The mm_cpumask is located at the end of mm_struct, and is
3154 	 * dynamically sized based on the maximum CPU number this system
3155 	 * can have, taking hotplug into account (nr_cpu_ids).
3156 	 */
3157 	mm_size = sizeof(struct mm_struct) + cpumask_size() + mm_cid_size();
3158 
3159 	mm_cachep = kmem_cache_create_usercopy("mm_struct",
3160 			mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
3161 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3162 			offsetof(struct mm_struct, saved_auxv),
3163 			sizeof_field(struct mm_struct, saved_auxv),
3164 			NULL);
3165 }
3166 
3167 void __init proc_caches_init(void)
3168 {
3169 	sighand_cachep = kmem_cache_create("sighand_cache",
3170 			sizeof(struct sighand_struct), 0,
3171 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
3172 			SLAB_ACCOUNT, sighand_ctor);
3173 	signal_cachep = kmem_cache_create("signal_cache",
3174 			sizeof(struct signal_struct), 0,
3175 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3176 			NULL);
3177 	exec_state_init();
3178 	files_cachep = kmem_cache_create("files_cache",
3179 			sizeof(struct files_struct), 0,
3180 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3181 			NULL);
3182 	fs_cachep = kmem_cache_create("fs_cache",
3183 			sizeof(struct fs_struct), 0,
3184 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
3185 			NULL);
3186 	mmap_init();
3187 	nsproxy_cache_init();
3188 }
3189 
3190 /*
3191  * Check constraints on flags passed to the unshare system call.
3192  */
3193 static int check_unshare_flags(unsigned long unshare_flags)
3194 {
3195 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_SIGHAND|
3196 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
3197 				CLONE_NS_ALL | UNSHARE_EMPTY_MNTNS))
3198 		return -EINVAL;
3199 	/*
3200 	 * Not implemented, but pretend it works if there is nothing
3201 	 * to unshare.  Note that unsharing the address space or the
3202 	 * signal handlers also need to unshare the signal queues (aka
3203 	 * CLONE_THREAD).
3204 	 */
3205 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
3206 		if (!thread_group_empty(current))
3207 			return -EINVAL;
3208 	}
3209 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
3210 		if (refcount_read(&current->sighand->count) > 1)
3211 			return -EINVAL;
3212 	}
3213 	if (unshare_flags & CLONE_VM) {
3214 		if (!current_is_single_threaded())
3215 			return -EINVAL;
3216 	}
3217 
3218 	return 0;
3219 }
3220 
3221 /*
3222  * Unshare the filesystem structure if it is being shared
3223  */
3224 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
3225 {
3226 	struct fs_struct *fs = current->fs;
3227 
3228 	if (!(unshare_flags & CLONE_FS) || !fs)
3229 		return 0;
3230 
3231 	/* don't need lock here; in the worst case we'll do useless copy */
3232 	if (!(unshare_flags & CLONE_NEWNS) && fs->users == 1)
3233 		return 0;
3234 
3235 	*new_fsp = copy_fs_struct(fs);
3236 	if (!*new_fsp)
3237 		return -ENOMEM;
3238 
3239 	return 0;
3240 }
3241 
3242 /*
3243  * unshare allows a process to 'unshare' part of the process
3244  * context which was originally shared using clone.  copy_*
3245  * functions used by kernel_clone() cannot be used here directly
3246  * because they modify an inactive task_struct that is being
3247  * constructed. Here we are modifying the current, active,
3248  * task_struct.
3249  */
3250 int ksys_unshare(unsigned long unshare_flags)
3251 {
3252 	struct fs_struct *new_fs = NULL;
3253 	struct files_struct *new_fd = NULL;
3254 	struct cred *new_cred = NULL;
3255 	struct nsproxy *new_nsproxy = NULL;
3256 	int do_sysvsem = 0;
3257 	int err;
3258 
3259 	/*
3260 	 * If unsharing a user namespace must also unshare the thread group
3261 	 * and unshare the filesystem root and working directories.
3262 	 */
3263 	if (unshare_flags & CLONE_NEWUSER)
3264 		unshare_flags |= CLONE_THREAD | CLONE_FS;
3265 	/*
3266 	 * If unsharing vm, must also unshare signal handlers.
3267 	 */
3268 	if (unshare_flags & CLONE_VM)
3269 		unshare_flags |= CLONE_SIGHAND;
3270 	/*
3271 	 * If unsharing a signal handlers, must also unshare the signal queues.
3272 	 */
3273 	if (unshare_flags & CLONE_SIGHAND)
3274 		unshare_flags |= CLONE_THREAD;
3275 	/*
3276 	 * If unsharing namespace, must also unshare filesystem information.
3277 	 */
3278 	if (unshare_flags & UNSHARE_EMPTY_MNTNS)
3279 		unshare_flags |= CLONE_NEWNS;
3280 	if (unshare_flags & CLONE_NEWNS)
3281 		unshare_flags |= CLONE_FS;
3282 
3283 	/* No unsharing with overriden fs state */
3284 	VFS_WARN_ON_ONCE(unshare_flags & (CLONE_NEWNS | CLONE_FS) &&
3285 			 current->fs != current->real_fs);
3286 
3287 	err = check_unshare_flags(unshare_flags);
3288 	if (err)
3289 		goto bad_unshare_out;
3290 	/*
3291 	 * CLONE_NEWIPC must also detach from the undolist: after switching
3292 	 * to a new ipc namespace, the semaphore arrays from the old
3293 	 * namespace are unreachable.
3294 	 */
3295 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
3296 		do_sysvsem = 1;
3297 	err = unshare_fs(unshare_flags, &new_fs);
3298 	if (err)
3299 		goto bad_unshare_out;
3300 	err = unshare_fd(unshare_flags, &new_fd);
3301 	if (err)
3302 		goto bad_unshare_cleanup_fs;
3303 	err = unshare_userns(unshare_flags, &new_cred);
3304 	if (err)
3305 		goto bad_unshare_cleanup_fd;
3306 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
3307 					 new_cred, new_fs);
3308 	if (err)
3309 		goto bad_unshare_cleanup_cred;
3310 	if (new_cred) {
3311 		err = set_cred_ucounts(new_cred);
3312 		if (err)
3313 			goto bad_unshare_cleanup_nsproxy;
3314 	}
3315 
3316 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
3317 		if (do_sysvsem) {
3318 			/*
3319 			 * CLONE_SYSVSEM is equivalent to sys_exit().
3320 			 */
3321 			exit_sem(current);
3322 		}
3323 		if (unshare_flags & CLONE_NEWIPC) {
3324 			/* Orphan segments in old ns (see sem above). */
3325 			exit_shm(current);
3326 			shm_init_task(current);
3327 		}
3328 
3329 		if (new_fs) {
3330 			new_fs = switch_fs_struct(new_fs);
3331 			if (new_fs)
3332 				free_fs_struct(no_free_ptr(new_fs));
3333 		}
3334 
3335 		if (new_nsproxy) {
3336 			switch_task_namespaces(current, new_nsproxy);
3337 			new_nsproxy = NULL;
3338 		}
3339 
3340 		if (new_fd)
3341 			switch_files_struct(current, no_free_ptr(new_fd));
3342 
3343 		if (new_cred) {
3344 			/* Install the new user namespace */
3345 			commit_creds(new_cred);
3346 			new_cred = NULL;
3347 		}
3348 	}
3349 
3350 	perf_event_namespaces(current);
3351 
3352 bad_unshare_cleanup_nsproxy:
3353 	/* never installed, so no active references to drop */
3354 	if (new_nsproxy)
3355 		nsproxy_free(new_nsproxy);
3356 bad_unshare_cleanup_cred:
3357 	if (new_cred)
3358 		put_cred(new_cred);
3359 bad_unshare_cleanup_fd:
3360 	if (new_fd)
3361 		put_files_struct(new_fd);
3362 bad_unshare_cleanup_fs:
3363 	if (new_fs)
3364 		free_fs_struct(new_fs);
3365 
3366 bad_unshare_out:
3367 	return err;
3368 }
3369 
3370 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3371 {
3372 	return ksys_unshare(unshare_flags);
3373 }
3374 
3375 static int sysctl_max_threads(const struct ctl_table *table, int write,
3376 		       void *buffer, size_t *lenp, loff_t *ppos)
3377 {
3378 	struct ctl_table t;
3379 	int ret;
3380 	int threads = max_threads;
3381 	int min = 1;
3382 	int max = MAX_THREADS;
3383 
3384 	t = *table;
3385 	t.data = &threads;
3386 	t.extra1 = &min;
3387 	t.extra2 = &max;
3388 
3389 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3390 	if (ret || !write)
3391 		return ret;
3392 
3393 	max_threads = threads;
3394 
3395 	return 0;
3396 }
3397 
3398 static const struct ctl_table fork_sysctl_table[] = {
3399 	{
3400 		.procname	= "threads-max",
3401 		.data		= NULL,
3402 		.maxlen		= sizeof(int),
3403 		.mode		= 0644,
3404 		.proc_handler	= sysctl_max_threads,
3405 	},
3406 };
3407 
3408 static int __init init_fork_sysctl(void)
3409 {
3410 	register_sysctl_init("kernel", fork_sysctl_table);
3411 	return 0;
3412 }
3413 
3414 subsys_initcall(init_fork_sysctl);
3415