xref: /openbmc/linux/kernel/fork.c (revision 8ffdff6a)
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/coredump.h>
20 #include <linux/sched/user.h>
21 #include <linux/sched/numa_balancing.h>
22 #include <linux/sched/stat.h>
23 #include <linux/sched/task.h>
24 #include <linux/sched/task_stack.h>
25 #include <linux/sched/cputime.h>
26 #include <linux/seq_file.h>
27 #include <linux/rtmutex.h>
28 #include <linux/init.h>
29 #include <linux/unistd.h>
30 #include <linux/module.h>
31 #include <linux/vmalloc.h>
32 #include <linux/completion.h>
33 #include <linux/personality.h>
34 #include <linux/mempolicy.h>
35 #include <linux/sem.h>
36 #include <linux/file.h>
37 #include <linux/fdtable.h>
38 #include <linux/iocontext.h>
39 #include <linux/key.h>
40 #include <linux/binfmts.h>
41 #include <linux/mman.h>
42 #include <linux/mmu_notifier.h>
43 #include <linux/fs.h>
44 #include <linux/mm.h>
45 #include <linux/vmacache.h>
46 #include <linux/nsproxy.h>
47 #include <linux/capability.h>
48 #include <linux/cpu.h>
49 #include <linux/cgroup.h>
50 #include <linux/security.h>
51 #include <linux/hugetlb.h>
52 #include <linux/seccomp.h>
53 #include <linux/swap.h>
54 #include <linux/syscalls.h>
55 #include <linux/jiffies.h>
56 #include <linux/futex.h>
57 #include <linux/compat.h>
58 #include <linux/kthread.h>
59 #include <linux/task_io_accounting_ops.h>
60 #include <linux/rcupdate.h>
61 #include <linux/ptrace.h>
62 #include <linux/mount.h>
63 #include <linux/audit.h>
64 #include <linux/memcontrol.h>
65 #include <linux/ftrace.h>
66 #include <linux/proc_fs.h>
67 #include <linux/profile.h>
68 #include <linux/rmap.h>
69 #include <linux/ksm.h>
70 #include <linux/acct.h>
71 #include <linux/userfaultfd_k.h>
72 #include <linux/tsacct_kern.h>
73 #include <linux/cn_proc.h>
74 #include <linux/freezer.h>
75 #include <linux/delayacct.h>
76 #include <linux/taskstats_kern.h>
77 #include <linux/random.h>
78 #include <linux/tty.h>
79 #include <linux/blkdev.h>
80 #include <linux/fs_struct.h>
81 #include <linux/magic.h>
82 #include <linux/perf_event.h>
83 #include <linux/posix-timers.h>
84 #include <linux/user-return-notifier.h>
85 #include <linux/oom.h>
86 #include <linux/khugepaged.h>
87 #include <linux/signalfd.h>
88 #include <linux/uprobes.h>
89 #include <linux/aio.h>
90 #include <linux/compiler.h>
91 #include <linux/sysctl.h>
92 #include <linux/kcov.h>
93 #include <linux/livepatch.h>
94 #include <linux/thread_info.h>
95 #include <linux/stackleak.h>
96 #include <linux/kasan.h>
97 #include <linux/scs.h>
98 #include <linux/io_uring.h>
99 
100 #include <asm/pgalloc.h>
101 #include <linux/uaccess.h>
102 #include <asm/mmu_context.h>
103 #include <asm/cacheflush.h>
104 #include <asm/tlbflush.h>
105 
106 #include <trace/events/sched.h>
107 
108 #define CREATE_TRACE_POINTS
109 #include <trace/events/task.h>
110 
111 /*
112  * Minimum number of threads to boot the kernel
113  */
114 #define MIN_THREADS 20
115 
116 /*
117  * Maximum number of threads
118  */
119 #define MAX_THREADS FUTEX_TID_MASK
120 
121 /*
122  * Protected counters by write_lock_irq(&tasklist_lock)
123  */
124 unsigned long total_forks;	/* Handle normal Linux uptimes. */
125 int nr_threads;			/* The idle threads do not count.. */
126 
127 static int max_threads;		/* tunable limit on nr_threads */
128 
129 #define NAMED_ARRAY_INDEX(x)	[x] = __stringify(x)
130 
131 static const char * const resident_page_types[] = {
132 	NAMED_ARRAY_INDEX(MM_FILEPAGES),
133 	NAMED_ARRAY_INDEX(MM_ANONPAGES),
134 	NAMED_ARRAY_INDEX(MM_SWAPENTS),
135 	NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
136 };
137 
138 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
139 
140 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
141 
142 #ifdef CONFIG_PROVE_RCU
143 int lockdep_tasklist_lock_is_held(void)
144 {
145 	return lockdep_is_held(&tasklist_lock);
146 }
147 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
148 #endif /* #ifdef CONFIG_PROVE_RCU */
149 
150 int nr_processes(void)
151 {
152 	int cpu;
153 	int total = 0;
154 
155 	for_each_possible_cpu(cpu)
156 		total += per_cpu(process_counts, cpu);
157 
158 	return total;
159 }
160 
161 void __weak arch_release_task_struct(struct task_struct *tsk)
162 {
163 }
164 
165 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
166 static struct kmem_cache *task_struct_cachep;
167 
168 static inline struct task_struct *alloc_task_struct_node(int node)
169 {
170 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
171 }
172 
173 static inline void free_task_struct(struct task_struct *tsk)
174 {
175 	kmem_cache_free(task_struct_cachep, tsk);
176 }
177 #endif
178 
179 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
180 
181 /*
182  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
183  * kmemcache based allocator.
184  */
185 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
186 
187 #ifdef CONFIG_VMAP_STACK
188 /*
189  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
190  * flush.  Try to minimize the number of calls by caching stacks.
191  */
192 #define NR_CACHED_STACKS 2
193 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
194 
195 static int free_vm_stack_cache(unsigned int cpu)
196 {
197 	struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
198 	int i;
199 
200 	for (i = 0; i < NR_CACHED_STACKS; i++) {
201 		struct vm_struct *vm_stack = cached_vm_stacks[i];
202 
203 		if (!vm_stack)
204 			continue;
205 
206 		vfree(vm_stack->addr);
207 		cached_vm_stacks[i] = NULL;
208 	}
209 
210 	return 0;
211 }
212 #endif
213 
214 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
215 {
216 #ifdef CONFIG_VMAP_STACK
217 	void *stack;
218 	int i;
219 
220 	for (i = 0; i < NR_CACHED_STACKS; i++) {
221 		struct vm_struct *s;
222 
223 		s = this_cpu_xchg(cached_stacks[i], NULL);
224 
225 		if (!s)
226 			continue;
227 
228 		/* Mark stack accessible for KASAN. */
229 		kasan_unpoison_range(s->addr, THREAD_SIZE);
230 
231 		/* Clear stale pointers from reused stack. */
232 		memset(s->addr, 0, THREAD_SIZE);
233 
234 		tsk->stack_vm_area = s;
235 		tsk->stack = s->addr;
236 		return s->addr;
237 	}
238 
239 	/*
240 	 * Allocated stacks are cached and later reused by new threads,
241 	 * so memcg accounting is performed manually on assigning/releasing
242 	 * stacks to tasks. Drop __GFP_ACCOUNT.
243 	 */
244 	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
245 				     VMALLOC_START, VMALLOC_END,
246 				     THREADINFO_GFP & ~__GFP_ACCOUNT,
247 				     PAGE_KERNEL,
248 				     0, node, __builtin_return_address(0));
249 
250 	/*
251 	 * We can't call find_vm_area() in interrupt context, and
252 	 * free_thread_stack() can be called in interrupt context,
253 	 * so cache the vm_struct.
254 	 */
255 	if (stack) {
256 		tsk->stack_vm_area = find_vm_area(stack);
257 		tsk->stack = stack;
258 	}
259 	return stack;
260 #else
261 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
262 					     THREAD_SIZE_ORDER);
263 
264 	if (likely(page)) {
265 		tsk->stack = kasan_reset_tag(page_address(page));
266 		return tsk->stack;
267 	}
268 	return NULL;
269 #endif
270 }
271 
272 static inline void free_thread_stack(struct task_struct *tsk)
273 {
274 #ifdef CONFIG_VMAP_STACK
275 	struct vm_struct *vm = task_stack_vm_area(tsk);
276 
277 	if (vm) {
278 		int i;
279 
280 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
281 			memcg_kmem_uncharge_page(vm->pages[i], 0);
282 
283 		for (i = 0; i < NR_CACHED_STACKS; i++) {
284 			if (this_cpu_cmpxchg(cached_stacks[i],
285 					NULL, tsk->stack_vm_area) != NULL)
286 				continue;
287 
288 			return;
289 		}
290 
291 		vfree_atomic(tsk->stack);
292 		return;
293 	}
294 #endif
295 
296 	__free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
297 }
298 # else
299 static struct kmem_cache *thread_stack_cache;
300 
301 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
302 						  int node)
303 {
304 	unsigned long *stack;
305 	stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
306 	stack = kasan_reset_tag(stack);
307 	tsk->stack = stack;
308 	return stack;
309 }
310 
311 static void free_thread_stack(struct task_struct *tsk)
312 {
313 	kmem_cache_free(thread_stack_cache, tsk->stack);
314 }
315 
316 void thread_stack_cache_init(void)
317 {
318 	thread_stack_cache = kmem_cache_create_usercopy("thread_stack",
319 					THREAD_SIZE, THREAD_SIZE, 0, 0,
320 					THREAD_SIZE, NULL);
321 	BUG_ON(thread_stack_cache == NULL);
322 }
323 # endif
324 #endif
325 
326 /* SLAB cache for signal_struct structures (tsk->signal) */
327 static struct kmem_cache *signal_cachep;
328 
329 /* SLAB cache for sighand_struct structures (tsk->sighand) */
330 struct kmem_cache *sighand_cachep;
331 
332 /* SLAB cache for files_struct structures (tsk->files) */
333 struct kmem_cache *files_cachep;
334 
335 /* SLAB cache for fs_struct structures (tsk->fs) */
336 struct kmem_cache *fs_cachep;
337 
338 /* SLAB cache for vm_area_struct structures */
339 static struct kmem_cache *vm_area_cachep;
340 
341 /* SLAB cache for mm_struct structures (tsk->mm) */
342 static struct kmem_cache *mm_cachep;
343 
344 struct vm_area_struct *vm_area_alloc(struct mm_struct *mm)
345 {
346 	struct vm_area_struct *vma;
347 
348 	vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
349 	if (vma)
350 		vma_init(vma, mm);
351 	return vma;
352 }
353 
354 struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig)
355 {
356 	struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
357 
358 	if (new) {
359 		ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
360 		ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
361 		/*
362 		 * orig->shared.rb may be modified concurrently, but the clone
363 		 * will be reinitialized.
364 		 */
365 		*new = data_race(*orig);
366 		INIT_LIST_HEAD(&new->anon_vma_chain);
367 		new->vm_next = new->vm_prev = NULL;
368 	}
369 	return new;
370 }
371 
372 void vm_area_free(struct vm_area_struct *vma)
373 {
374 	kmem_cache_free(vm_area_cachep, vma);
375 }
376 
377 static void account_kernel_stack(struct task_struct *tsk, int account)
378 {
379 	void *stack = task_stack_page(tsk);
380 	struct vm_struct *vm = task_stack_vm_area(tsk);
381 
382 
383 	/* All stack pages are in the same node. */
384 	if (vm)
385 		mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
386 				      account * (THREAD_SIZE / 1024));
387 	else
388 		mod_lruvec_kmem_state(stack, NR_KERNEL_STACK_KB,
389 				      account * (THREAD_SIZE / 1024));
390 }
391 
392 static int memcg_charge_kernel_stack(struct task_struct *tsk)
393 {
394 #ifdef CONFIG_VMAP_STACK
395 	struct vm_struct *vm = task_stack_vm_area(tsk);
396 	int ret;
397 
398 	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
399 
400 	if (vm) {
401 		int i;
402 
403 		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
404 
405 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
406 			/*
407 			 * If memcg_kmem_charge_page() fails, page's
408 			 * memory cgroup pointer is NULL, and
409 			 * memcg_kmem_uncharge_page() in free_thread_stack()
410 			 * will ignore this page.
411 			 */
412 			ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
413 						     0);
414 			if (ret)
415 				return ret;
416 		}
417 	}
418 #endif
419 	return 0;
420 }
421 
422 static void release_task_stack(struct task_struct *tsk)
423 {
424 	if (WARN_ON(tsk->state != TASK_DEAD))
425 		return;  /* Better to leak the stack than to free prematurely */
426 
427 	account_kernel_stack(tsk, -1);
428 	free_thread_stack(tsk);
429 	tsk->stack = NULL;
430 #ifdef CONFIG_VMAP_STACK
431 	tsk->stack_vm_area = NULL;
432 #endif
433 }
434 
435 #ifdef CONFIG_THREAD_INFO_IN_TASK
436 void put_task_stack(struct task_struct *tsk)
437 {
438 	if (refcount_dec_and_test(&tsk->stack_refcount))
439 		release_task_stack(tsk);
440 }
441 #endif
442 
443 void free_task(struct task_struct *tsk)
444 {
445 	scs_release(tsk);
446 
447 #ifndef CONFIG_THREAD_INFO_IN_TASK
448 	/*
449 	 * The task is finally done with both the stack and thread_info,
450 	 * so free both.
451 	 */
452 	release_task_stack(tsk);
453 #else
454 	/*
455 	 * If the task had a separate stack allocation, it should be gone
456 	 * by now.
457 	 */
458 	WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
459 #endif
460 	rt_mutex_debug_task_free(tsk);
461 	ftrace_graph_exit_task(tsk);
462 	arch_release_task_struct(tsk);
463 	if (tsk->flags & PF_KTHREAD)
464 		free_kthread_struct(tsk);
465 	free_task_struct(tsk);
466 }
467 EXPORT_SYMBOL(free_task);
468 
469 #ifdef CONFIG_MMU
470 static __latent_entropy int dup_mmap(struct mm_struct *mm,
471 					struct mm_struct *oldmm)
472 {
473 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
474 	struct rb_node **rb_link, *rb_parent;
475 	int retval;
476 	unsigned long charge;
477 	LIST_HEAD(uf);
478 
479 	uprobe_start_dup_mmap();
480 	if (mmap_write_lock_killable(oldmm)) {
481 		retval = -EINTR;
482 		goto fail_uprobe_end;
483 	}
484 	flush_cache_dup_mm(oldmm);
485 	uprobe_dup_mmap(oldmm, mm);
486 	/*
487 	 * Not linked in yet - no deadlock potential:
488 	 */
489 	mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
490 
491 	/* No ordering required: file already has been exposed. */
492 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
493 
494 	mm->total_vm = oldmm->total_vm;
495 	mm->data_vm = oldmm->data_vm;
496 	mm->exec_vm = oldmm->exec_vm;
497 	mm->stack_vm = oldmm->stack_vm;
498 
499 	rb_link = &mm->mm_rb.rb_node;
500 	rb_parent = NULL;
501 	pprev = &mm->mmap;
502 	retval = ksm_fork(mm, oldmm);
503 	if (retval)
504 		goto out;
505 	retval = khugepaged_fork(mm, oldmm);
506 	if (retval)
507 		goto out;
508 
509 	prev = NULL;
510 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
511 		struct file *file;
512 
513 		if (mpnt->vm_flags & VM_DONTCOPY) {
514 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
515 			continue;
516 		}
517 		charge = 0;
518 		/*
519 		 * Don't duplicate many vmas if we've been oom-killed (for
520 		 * example)
521 		 */
522 		if (fatal_signal_pending(current)) {
523 			retval = -EINTR;
524 			goto out;
525 		}
526 		if (mpnt->vm_flags & VM_ACCOUNT) {
527 			unsigned long len = vma_pages(mpnt);
528 
529 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
530 				goto fail_nomem;
531 			charge = len;
532 		}
533 		tmp = vm_area_dup(mpnt);
534 		if (!tmp)
535 			goto fail_nomem;
536 		retval = vma_dup_policy(mpnt, tmp);
537 		if (retval)
538 			goto fail_nomem_policy;
539 		tmp->vm_mm = mm;
540 		retval = dup_userfaultfd(tmp, &uf);
541 		if (retval)
542 			goto fail_nomem_anon_vma_fork;
543 		if (tmp->vm_flags & VM_WIPEONFORK) {
544 			/*
545 			 * VM_WIPEONFORK gets a clean slate in the child.
546 			 * Don't prepare anon_vma until fault since we don't
547 			 * copy page for current vma.
548 			 */
549 			tmp->anon_vma = NULL;
550 		} else if (anon_vma_fork(tmp, mpnt))
551 			goto fail_nomem_anon_vma_fork;
552 		tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
553 		file = tmp->vm_file;
554 		if (file) {
555 			struct inode *inode = file_inode(file);
556 			struct address_space *mapping = file->f_mapping;
557 
558 			get_file(file);
559 			if (tmp->vm_flags & VM_DENYWRITE)
560 				put_write_access(inode);
561 			i_mmap_lock_write(mapping);
562 			if (tmp->vm_flags & VM_SHARED)
563 				mapping_allow_writable(mapping);
564 			flush_dcache_mmap_lock(mapping);
565 			/* insert tmp into the share list, just after mpnt */
566 			vma_interval_tree_insert_after(tmp, mpnt,
567 					&mapping->i_mmap);
568 			flush_dcache_mmap_unlock(mapping);
569 			i_mmap_unlock_write(mapping);
570 		}
571 
572 		/*
573 		 * Clear hugetlb-related page reserves for children. This only
574 		 * affects MAP_PRIVATE mappings. Faults generated by the child
575 		 * are not guaranteed to succeed, even if read-only
576 		 */
577 		if (is_vm_hugetlb_page(tmp))
578 			reset_vma_resv_huge_pages(tmp);
579 
580 		/*
581 		 * Link in the new vma and copy the page table entries.
582 		 */
583 		*pprev = tmp;
584 		pprev = &tmp->vm_next;
585 		tmp->vm_prev = prev;
586 		prev = tmp;
587 
588 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
589 		rb_link = &tmp->vm_rb.rb_right;
590 		rb_parent = &tmp->vm_rb;
591 
592 		mm->map_count++;
593 		if (!(tmp->vm_flags & VM_WIPEONFORK))
594 			retval = copy_page_range(tmp, mpnt);
595 
596 		if (tmp->vm_ops && tmp->vm_ops->open)
597 			tmp->vm_ops->open(tmp);
598 
599 		if (retval)
600 			goto out;
601 	}
602 	/* a new mm has just been created */
603 	retval = arch_dup_mmap(oldmm, mm);
604 out:
605 	mmap_write_unlock(mm);
606 	flush_tlb_mm(oldmm);
607 	mmap_write_unlock(oldmm);
608 	dup_userfaultfd_complete(&uf);
609 fail_uprobe_end:
610 	uprobe_end_dup_mmap();
611 	return retval;
612 fail_nomem_anon_vma_fork:
613 	mpol_put(vma_policy(tmp));
614 fail_nomem_policy:
615 	vm_area_free(tmp);
616 fail_nomem:
617 	retval = -ENOMEM;
618 	vm_unacct_memory(charge);
619 	goto out;
620 }
621 
622 static inline int mm_alloc_pgd(struct mm_struct *mm)
623 {
624 	mm->pgd = pgd_alloc(mm);
625 	if (unlikely(!mm->pgd))
626 		return -ENOMEM;
627 	return 0;
628 }
629 
630 static inline void mm_free_pgd(struct mm_struct *mm)
631 {
632 	pgd_free(mm, mm->pgd);
633 }
634 #else
635 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
636 {
637 	mmap_write_lock(oldmm);
638 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
639 	mmap_write_unlock(oldmm);
640 	return 0;
641 }
642 #define mm_alloc_pgd(mm)	(0)
643 #define mm_free_pgd(mm)
644 #endif /* CONFIG_MMU */
645 
646 static void check_mm(struct mm_struct *mm)
647 {
648 	int i;
649 
650 	BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
651 			 "Please make sure 'struct resident_page_types[]' is updated as well");
652 
653 	for (i = 0; i < NR_MM_COUNTERS; i++) {
654 		long x = atomic_long_read(&mm->rss_stat.count[i]);
655 
656 		if (unlikely(x))
657 			pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
658 				 mm, resident_page_types[i], x);
659 	}
660 
661 	if (mm_pgtables_bytes(mm))
662 		pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
663 				mm_pgtables_bytes(mm));
664 
665 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
666 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
667 #endif
668 }
669 
670 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
671 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
672 
673 /*
674  * Called when the last reference to the mm
675  * is dropped: either by a lazy thread or by
676  * mmput. Free the page directory and the mm.
677  */
678 void __mmdrop(struct mm_struct *mm)
679 {
680 	BUG_ON(mm == &init_mm);
681 	WARN_ON_ONCE(mm == current->mm);
682 	WARN_ON_ONCE(mm == current->active_mm);
683 	mm_free_pgd(mm);
684 	destroy_context(mm);
685 	mmu_notifier_subscriptions_destroy(mm);
686 	check_mm(mm);
687 	put_user_ns(mm->user_ns);
688 	free_mm(mm);
689 }
690 EXPORT_SYMBOL_GPL(__mmdrop);
691 
692 static void mmdrop_async_fn(struct work_struct *work)
693 {
694 	struct mm_struct *mm;
695 
696 	mm = container_of(work, struct mm_struct, async_put_work);
697 	__mmdrop(mm);
698 }
699 
700 static void mmdrop_async(struct mm_struct *mm)
701 {
702 	if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
703 		INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
704 		schedule_work(&mm->async_put_work);
705 	}
706 }
707 
708 static inline void free_signal_struct(struct signal_struct *sig)
709 {
710 	taskstats_tgid_free(sig);
711 	sched_autogroup_exit(sig);
712 	/*
713 	 * __mmdrop is not safe to call from softirq context on x86 due to
714 	 * pgd_dtor so postpone it to the async context
715 	 */
716 	if (sig->oom_mm)
717 		mmdrop_async(sig->oom_mm);
718 	kmem_cache_free(signal_cachep, sig);
719 }
720 
721 static inline void put_signal_struct(struct signal_struct *sig)
722 {
723 	if (refcount_dec_and_test(&sig->sigcnt))
724 		free_signal_struct(sig);
725 }
726 
727 void __put_task_struct(struct task_struct *tsk)
728 {
729 	WARN_ON(!tsk->exit_state);
730 	WARN_ON(refcount_read(&tsk->usage));
731 	WARN_ON(tsk == current);
732 
733 	io_uring_free(tsk);
734 	cgroup_free(tsk);
735 	task_numa_free(tsk, true);
736 	security_task_free(tsk);
737 	exit_creds(tsk);
738 	delayacct_tsk_free(tsk);
739 	put_signal_struct(tsk->signal);
740 
741 	if (!profile_handoff_task(tsk))
742 		free_task(tsk);
743 }
744 EXPORT_SYMBOL_GPL(__put_task_struct);
745 
746 void __init __weak arch_task_cache_init(void) { }
747 
748 /*
749  * set_max_threads
750  */
751 static void set_max_threads(unsigned int max_threads_suggested)
752 {
753 	u64 threads;
754 	unsigned long nr_pages = totalram_pages();
755 
756 	/*
757 	 * The number of threads shall be limited such that the thread
758 	 * structures may only consume a small part of the available memory.
759 	 */
760 	if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
761 		threads = MAX_THREADS;
762 	else
763 		threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
764 				    (u64) THREAD_SIZE * 8UL);
765 
766 	if (threads > max_threads_suggested)
767 		threads = max_threads_suggested;
768 
769 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
770 }
771 
772 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
773 /* Initialized by the architecture: */
774 int arch_task_struct_size __read_mostly;
775 #endif
776 
777 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
778 static void task_struct_whitelist(unsigned long *offset, unsigned long *size)
779 {
780 	/* Fetch thread_struct whitelist for the architecture. */
781 	arch_thread_struct_whitelist(offset, size);
782 
783 	/*
784 	 * Handle zero-sized whitelist or empty thread_struct, otherwise
785 	 * adjust offset to position of thread_struct in task_struct.
786 	 */
787 	if (unlikely(*size == 0))
788 		*offset = 0;
789 	else
790 		*offset += offsetof(struct task_struct, thread);
791 }
792 #endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
793 
794 void __init fork_init(void)
795 {
796 	int i;
797 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
798 #ifndef ARCH_MIN_TASKALIGN
799 #define ARCH_MIN_TASKALIGN	0
800 #endif
801 	int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
802 	unsigned long useroffset, usersize;
803 
804 	/* create a slab on which task_structs can be allocated */
805 	task_struct_whitelist(&useroffset, &usersize);
806 	task_struct_cachep = kmem_cache_create_usercopy("task_struct",
807 			arch_task_struct_size, align,
808 			SLAB_PANIC|SLAB_ACCOUNT,
809 			useroffset, usersize, NULL);
810 #endif
811 
812 	/* do the arch specific task caches init */
813 	arch_task_cache_init();
814 
815 	set_max_threads(MAX_THREADS);
816 
817 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
818 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
819 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
820 		init_task.signal->rlim[RLIMIT_NPROC];
821 
822 	for (i = 0; i < UCOUNT_COUNTS; i++)
823 		init_user_ns.ucount_max[i] = max_threads/2;
824 
825 #ifdef CONFIG_VMAP_STACK
826 	cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
827 			  NULL, free_vm_stack_cache);
828 #endif
829 
830 	scs_init();
831 
832 	lockdep_init_task(&init_task);
833 	uprobes_init();
834 }
835 
836 int __weak arch_dup_task_struct(struct task_struct *dst,
837 					       struct task_struct *src)
838 {
839 	*dst = *src;
840 	return 0;
841 }
842 
843 void set_task_stack_end_magic(struct task_struct *tsk)
844 {
845 	unsigned long *stackend;
846 
847 	stackend = end_of_stack(tsk);
848 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
849 }
850 
851 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
852 {
853 	struct task_struct *tsk;
854 	unsigned long *stack;
855 	struct vm_struct *stack_vm_area __maybe_unused;
856 	int err;
857 
858 	if (node == NUMA_NO_NODE)
859 		node = tsk_fork_get_node(orig);
860 	tsk = alloc_task_struct_node(node);
861 	if (!tsk)
862 		return NULL;
863 
864 	stack = alloc_thread_stack_node(tsk, node);
865 	if (!stack)
866 		goto free_tsk;
867 
868 	if (memcg_charge_kernel_stack(tsk))
869 		goto free_stack;
870 
871 	stack_vm_area = task_stack_vm_area(tsk);
872 
873 	err = arch_dup_task_struct(tsk, orig);
874 
875 	/*
876 	 * arch_dup_task_struct() clobbers the stack-related fields.  Make
877 	 * sure they're properly initialized before using any stack-related
878 	 * functions again.
879 	 */
880 	tsk->stack = stack;
881 #ifdef CONFIG_VMAP_STACK
882 	tsk->stack_vm_area = stack_vm_area;
883 #endif
884 #ifdef CONFIG_THREAD_INFO_IN_TASK
885 	refcount_set(&tsk->stack_refcount, 1);
886 #endif
887 
888 	if (err)
889 		goto free_stack;
890 
891 	err = scs_prepare(tsk, node);
892 	if (err)
893 		goto free_stack;
894 
895 #ifdef CONFIG_SECCOMP
896 	/*
897 	 * We must handle setting up seccomp filters once we're under
898 	 * the sighand lock in case orig has changed between now and
899 	 * then. Until then, filter must be NULL to avoid messing up
900 	 * the usage counts on the error path calling free_task.
901 	 */
902 	tsk->seccomp.filter = NULL;
903 #endif
904 
905 	setup_thread_stack(tsk, orig);
906 	clear_user_return_notifier(tsk);
907 	clear_tsk_need_resched(tsk);
908 	set_task_stack_end_magic(tsk);
909 	clear_syscall_work_syscall_user_dispatch(tsk);
910 
911 #ifdef CONFIG_STACKPROTECTOR
912 	tsk->stack_canary = get_random_canary();
913 #endif
914 	if (orig->cpus_ptr == &orig->cpus_mask)
915 		tsk->cpus_ptr = &tsk->cpus_mask;
916 
917 	/*
918 	 * One for the user space visible state that goes away when reaped.
919 	 * One for the scheduler.
920 	 */
921 	refcount_set(&tsk->rcu_users, 2);
922 	/* One for the rcu users */
923 	refcount_set(&tsk->usage, 1);
924 #ifdef CONFIG_BLK_DEV_IO_TRACE
925 	tsk->btrace_seq = 0;
926 #endif
927 	tsk->splice_pipe = NULL;
928 	tsk->task_frag.page = NULL;
929 	tsk->wake_q.next = NULL;
930 
931 	account_kernel_stack(tsk, 1);
932 
933 	kcov_task_init(tsk);
934 	kmap_local_fork(tsk);
935 
936 #ifdef CONFIG_FAULT_INJECTION
937 	tsk->fail_nth = 0;
938 #endif
939 
940 #ifdef CONFIG_BLK_CGROUP
941 	tsk->throttle_queue = NULL;
942 	tsk->use_memdelay = 0;
943 #endif
944 
945 #ifdef CONFIG_MEMCG
946 	tsk->active_memcg = NULL;
947 #endif
948 	return tsk;
949 
950 free_stack:
951 	free_thread_stack(tsk);
952 free_tsk:
953 	free_task_struct(tsk);
954 	return NULL;
955 }
956 
957 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
958 
959 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
960 
961 static int __init coredump_filter_setup(char *s)
962 {
963 	default_dump_filter =
964 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
965 		MMF_DUMP_FILTER_MASK;
966 	return 1;
967 }
968 
969 __setup("coredump_filter=", coredump_filter_setup);
970 
971 #include <linux/init_task.h>
972 
973 static void mm_init_aio(struct mm_struct *mm)
974 {
975 #ifdef CONFIG_AIO
976 	spin_lock_init(&mm->ioctx_lock);
977 	mm->ioctx_table = NULL;
978 #endif
979 }
980 
981 static __always_inline void mm_clear_owner(struct mm_struct *mm,
982 					   struct task_struct *p)
983 {
984 #ifdef CONFIG_MEMCG
985 	if (mm->owner == p)
986 		WRITE_ONCE(mm->owner, NULL);
987 #endif
988 }
989 
990 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
991 {
992 #ifdef CONFIG_MEMCG
993 	mm->owner = p;
994 #endif
995 }
996 
997 static void mm_init_pasid(struct mm_struct *mm)
998 {
999 #ifdef CONFIG_IOMMU_SUPPORT
1000 	mm->pasid = INIT_PASID;
1001 #endif
1002 }
1003 
1004 static void mm_init_uprobes_state(struct mm_struct *mm)
1005 {
1006 #ifdef CONFIG_UPROBES
1007 	mm->uprobes_state.xol_area = NULL;
1008 #endif
1009 }
1010 
1011 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
1012 	struct user_namespace *user_ns)
1013 {
1014 	mm->mmap = NULL;
1015 	mm->mm_rb = RB_ROOT;
1016 	mm->vmacache_seqnum = 0;
1017 	atomic_set(&mm->mm_users, 1);
1018 	atomic_set(&mm->mm_count, 1);
1019 	seqcount_init(&mm->write_protect_seq);
1020 	mmap_init_lock(mm);
1021 	INIT_LIST_HEAD(&mm->mmlist);
1022 	mm->core_state = NULL;
1023 	mm_pgtables_bytes_init(mm);
1024 	mm->map_count = 0;
1025 	mm->locked_vm = 0;
1026 	atomic_set(&mm->has_pinned, 0);
1027 	atomic64_set(&mm->pinned_vm, 0);
1028 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1029 	spin_lock_init(&mm->page_table_lock);
1030 	spin_lock_init(&mm->arg_lock);
1031 	mm_init_cpumask(mm);
1032 	mm_init_aio(mm);
1033 	mm_init_owner(mm, p);
1034 	mm_init_pasid(mm);
1035 	RCU_INIT_POINTER(mm->exe_file, NULL);
1036 	mmu_notifier_subscriptions_init(mm);
1037 	init_tlb_flush_pending(mm);
1038 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1039 	mm->pmd_huge_pte = NULL;
1040 #endif
1041 	mm_init_uprobes_state(mm);
1042 
1043 	if (current->mm) {
1044 		mm->flags = current->mm->flags & MMF_INIT_MASK;
1045 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
1046 	} else {
1047 		mm->flags = default_dump_filter;
1048 		mm->def_flags = 0;
1049 	}
1050 
1051 	if (mm_alloc_pgd(mm))
1052 		goto fail_nopgd;
1053 
1054 	if (init_new_context(p, mm))
1055 		goto fail_nocontext;
1056 
1057 	mm->user_ns = get_user_ns(user_ns);
1058 	return mm;
1059 
1060 fail_nocontext:
1061 	mm_free_pgd(mm);
1062 fail_nopgd:
1063 	free_mm(mm);
1064 	return NULL;
1065 }
1066 
1067 /*
1068  * Allocate and initialize an mm_struct.
1069  */
1070 struct mm_struct *mm_alloc(void)
1071 {
1072 	struct mm_struct *mm;
1073 
1074 	mm = allocate_mm();
1075 	if (!mm)
1076 		return NULL;
1077 
1078 	memset(mm, 0, sizeof(*mm));
1079 	return mm_init(mm, current, current_user_ns());
1080 }
1081 
1082 static inline void __mmput(struct mm_struct *mm)
1083 {
1084 	VM_BUG_ON(atomic_read(&mm->mm_users));
1085 
1086 	uprobe_clear_state(mm);
1087 	exit_aio(mm);
1088 	ksm_exit(mm);
1089 	khugepaged_exit(mm); /* must run before exit_mmap */
1090 	exit_mmap(mm);
1091 	mm_put_huge_zero_page(mm);
1092 	set_mm_exe_file(mm, NULL);
1093 	if (!list_empty(&mm->mmlist)) {
1094 		spin_lock(&mmlist_lock);
1095 		list_del(&mm->mmlist);
1096 		spin_unlock(&mmlist_lock);
1097 	}
1098 	if (mm->binfmt)
1099 		module_put(mm->binfmt->module);
1100 	mmdrop(mm);
1101 }
1102 
1103 /*
1104  * Decrement the use count and release all resources for an mm.
1105  */
1106 void mmput(struct mm_struct *mm)
1107 {
1108 	might_sleep();
1109 
1110 	if (atomic_dec_and_test(&mm->mm_users))
1111 		__mmput(mm);
1112 }
1113 EXPORT_SYMBOL_GPL(mmput);
1114 
1115 #ifdef CONFIG_MMU
1116 static void mmput_async_fn(struct work_struct *work)
1117 {
1118 	struct mm_struct *mm = container_of(work, struct mm_struct,
1119 					    async_put_work);
1120 
1121 	__mmput(mm);
1122 }
1123 
1124 void mmput_async(struct mm_struct *mm)
1125 {
1126 	if (atomic_dec_and_test(&mm->mm_users)) {
1127 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
1128 		schedule_work(&mm->async_put_work);
1129 	}
1130 }
1131 #endif
1132 
1133 /**
1134  * set_mm_exe_file - change a reference to the mm's executable file
1135  *
1136  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
1137  *
1138  * Main users are mmput() and sys_execve(). Callers prevent concurrent
1139  * invocations: in mmput() nobody alive left, in execve task is single
1140  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
1141  * mm->exe_file, but does so without using set_mm_exe_file() in order
1142  * to do avoid the need for any locks.
1143  */
1144 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1145 {
1146 	struct file *old_exe_file;
1147 
1148 	/*
1149 	 * It is safe to dereference the exe_file without RCU as
1150 	 * this function is only called if nobody else can access
1151 	 * this mm -- see comment above for justification.
1152 	 */
1153 	old_exe_file = rcu_dereference_raw(mm->exe_file);
1154 
1155 	if (new_exe_file)
1156 		get_file(new_exe_file);
1157 	rcu_assign_pointer(mm->exe_file, new_exe_file);
1158 	if (old_exe_file)
1159 		fput(old_exe_file);
1160 }
1161 
1162 /**
1163  * get_mm_exe_file - acquire a reference to the mm's executable file
1164  *
1165  * Returns %NULL if mm has no associated executable file.
1166  * User must release file via fput().
1167  */
1168 struct file *get_mm_exe_file(struct mm_struct *mm)
1169 {
1170 	struct file *exe_file;
1171 
1172 	rcu_read_lock();
1173 	exe_file = rcu_dereference(mm->exe_file);
1174 	if (exe_file && !get_file_rcu(exe_file))
1175 		exe_file = NULL;
1176 	rcu_read_unlock();
1177 	return exe_file;
1178 }
1179 EXPORT_SYMBOL(get_mm_exe_file);
1180 
1181 /**
1182  * get_task_exe_file - acquire a reference to the task's executable file
1183  *
1184  * Returns %NULL if task's mm (if any) has no associated executable file or
1185  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1186  * User must release file via fput().
1187  */
1188 struct file *get_task_exe_file(struct task_struct *task)
1189 {
1190 	struct file *exe_file = NULL;
1191 	struct mm_struct *mm;
1192 
1193 	task_lock(task);
1194 	mm = task->mm;
1195 	if (mm) {
1196 		if (!(task->flags & PF_KTHREAD))
1197 			exe_file = get_mm_exe_file(mm);
1198 	}
1199 	task_unlock(task);
1200 	return exe_file;
1201 }
1202 EXPORT_SYMBOL(get_task_exe_file);
1203 
1204 /**
1205  * get_task_mm - acquire a reference to the task's mm
1206  *
1207  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1208  * this kernel workthread has transiently adopted a user mm with use_mm,
1209  * to do its AIO) is not set and if so returns a reference to it, after
1210  * bumping up the use count.  User must release the mm via mmput()
1211  * after use.  Typically used by /proc and ptrace.
1212  */
1213 struct mm_struct *get_task_mm(struct task_struct *task)
1214 {
1215 	struct mm_struct *mm;
1216 
1217 	task_lock(task);
1218 	mm = task->mm;
1219 	if (mm) {
1220 		if (task->flags & PF_KTHREAD)
1221 			mm = NULL;
1222 		else
1223 			mmget(mm);
1224 	}
1225 	task_unlock(task);
1226 	return mm;
1227 }
1228 EXPORT_SYMBOL_GPL(get_task_mm);
1229 
1230 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1231 {
1232 	struct mm_struct *mm;
1233 	int err;
1234 
1235 	err =  down_read_killable(&task->signal->exec_update_lock);
1236 	if (err)
1237 		return ERR_PTR(err);
1238 
1239 	mm = get_task_mm(task);
1240 	if (mm && mm != current->mm &&
1241 			!ptrace_may_access(task, mode)) {
1242 		mmput(mm);
1243 		mm = ERR_PTR(-EACCES);
1244 	}
1245 	up_read(&task->signal->exec_update_lock);
1246 
1247 	return mm;
1248 }
1249 
1250 static void complete_vfork_done(struct task_struct *tsk)
1251 {
1252 	struct completion *vfork;
1253 
1254 	task_lock(tsk);
1255 	vfork = tsk->vfork_done;
1256 	if (likely(vfork)) {
1257 		tsk->vfork_done = NULL;
1258 		complete(vfork);
1259 	}
1260 	task_unlock(tsk);
1261 }
1262 
1263 static int wait_for_vfork_done(struct task_struct *child,
1264 				struct completion *vfork)
1265 {
1266 	int killed;
1267 
1268 	freezer_do_not_count();
1269 	cgroup_enter_frozen();
1270 	killed = wait_for_completion_killable(vfork);
1271 	cgroup_leave_frozen(false);
1272 	freezer_count();
1273 
1274 	if (killed) {
1275 		task_lock(child);
1276 		child->vfork_done = NULL;
1277 		task_unlock(child);
1278 	}
1279 
1280 	put_task_struct(child);
1281 	return killed;
1282 }
1283 
1284 /* Please note the differences between mmput and mm_release.
1285  * mmput is called whenever we stop holding onto a mm_struct,
1286  * error success whatever.
1287  *
1288  * mm_release is called after a mm_struct has been removed
1289  * from the current process.
1290  *
1291  * This difference is important for error handling, when we
1292  * only half set up a mm_struct for a new process and need to restore
1293  * the old one.  Because we mmput the new mm_struct before
1294  * restoring the old one. . .
1295  * Eric Biederman 10 January 1998
1296  */
1297 static void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1298 {
1299 	uprobe_free_utask(tsk);
1300 
1301 	/* Get rid of any cached register state */
1302 	deactivate_mm(tsk, mm);
1303 
1304 	/*
1305 	 * Signal userspace if we're not exiting with a core dump
1306 	 * because we want to leave the value intact for debugging
1307 	 * purposes.
1308 	 */
1309 	if (tsk->clear_child_tid) {
1310 		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1311 		    atomic_read(&mm->mm_users) > 1) {
1312 			/*
1313 			 * We don't check the error code - if userspace has
1314 			 * not set up a proper pointer then tough luck.
1315 			 */
1316 			put_user(0, tsk->clear_child_tid);
1317 			do_futex(tsk->clear_child_tid, FUTEX_WAKE,
1318 					1, NULL, NULL, 0, 0);
1319 		}
1320 		tsk->clear_child_tid = NULL;
1321 	}
1322 
1323 	/*
1324 	 * All done, finally we can wake up parent and return this mm to him.
1325 	 * Also kthread_stop() uses this completion for synchronization.
1326 	 */
1327 	if (tsk->vfork_done)
1328 		complete_vfork_done(tsk);
1329 }
1330 
1331 void exit_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1332 {
1333 	futex_exit_release(tsk);
1334 	mm_release(tsk, mm);
1335 }
1336 
1337 void exec_mm_release(struct task_struct *tsk, struct mm_struct *mm)
1338 {
1339 	futex_exec_release(tsk);
1340 	mm_release(tsk, mm);
1341 }
1342 
1343 /**
1344  * dup_mm() - duplicates an existing mm structure
1345  * @tsk: the task_struct with which the new mm will be associated.
1346  * @oldmm: the mm to duplicate.
1347  *
1348  * Allocates a new mm structure and duplicates the provided @oldmm structure
1349  * content into it.
1350  *
1351  * Return: the duplicated mm or NULL on failure.
1352  */
1353 static struct mm_struct *dup_mm(struct task_struct *tsk,
1354 				struct mm_struct *oldmm)
1355 {
1356 	struct mm_struct *mm;
1357 	int err;
1358 
1359 	mm = allocate_mm();
1360 	if (!mm)
1361 		goto fail_nomem;
1362 
1363 	memcpy(mm, oldmm, sizeof(*mm));
1364 
1365 	if (!mm_init(mm, tsk, mm->user_ns))
1366 		goto fail_nomem;
1367 
1368 	err = dup_mmap(mm, oldmm);
1369 	if (err)
1370 		goto free_pt;
1371 
1372 	mm->hiwater_rss = get_mm_rss(mm);
1373 	mm->hiwater_vm = mm->total_vm;
1374 
1375 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1376 		goto free_pt;
1377 
1378 	return mm;
1379 
1380 free_pt:
1381 	/* don't put binfmt in mmput, we haven't got module yet */
1382 	mm->binfmt = NULL;
1383 	mm_init_owner(mm, NULL);
1384 	mmput(mm);
1385 
1386 fail_nomem:
1387 	return NULL;
1388 }
1389 
1390 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1391 {
1392 	struct mm_struct *mm, *oldmm;
1393 	int retval;
1394 
1395 	tsk->min_flt = tsk->maj_flt = 0;
1396 	tsk->nvcsw = tsk->nivcsw = 0;
1397 #ifdef CONFIG_DETECT_HUNG_TASK
1398 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1399 	tsk->last_switch_time = 0;
1400 #endif
1401 
1402 	tsk->mm = NULL;
1403 	tsk->active_mm = NULL;
1404 
1405 	/*
1406 	 * Are we cloning a kernel thread?
1407 	 *
1408 	 * We need to steal a active VM for that..
1409 	 */
1410 	oldmm = current->mm;
1411 	if (!oldmm)
1412 		return 0;
1413 
1414 	/* initialize the new vmacache entries */
1415 	vmacache_flush(tsk);
1416 
1417 	if (clone_flags & CLONE_VM) {
1418 		mmget(oldmm);
1419 		mm = oldmm;
1420 		goto good_mm;
1421 	}
1422 
1423 	retval = -ENOMEM;
1424 	mm = dup_mm(tsk, current->mm);
1425 	if (!mm)
1426 		goto fail_nomem;
1427 
1428 good_mm:
1429 	tsk->mm = mm;
1430 	tsk->active_mm = mm;
1431 	return 0;
1432 
1433 fail_nomem:
1434 	return retval;
1435 }
1436 
1437 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1438 {
1439 	struct fs_struct *fs = current->fs;
1440 	if (clone_flags & CLONE_FS) {
1441 		/* tsk->fs is already what we want */
1442 		spin_lock(&fs->lock);
1443 		if (fs->in_exec) {
1444 			spin_unlock(&fs->lock);
1445 			return -EAGAIN;
1446 		}
1447 		fs->users++;
1448 		spin_unlock(&fs->lock);
1449 		return 0;
1450 	}
1451 	tsk->fs = copy_fs_struct(fs);
1452 	if (!tsk->fs)
1453 		return -ENOMEM;
1454 	return 0;
1455 }
1456 
1457 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1458 {
1459 	struct files_struct *oldf, *newf;
1460 	int error = 0;
1461 
1462 	/*
1463 	 * A background process may not have any files ...
1464 	 */
1465 	oldf = current->files;
1466 	if (!oldf)
1467 		goto out;
1468 
1469 	if (clone_flags & CLONE_FILES) {
1470 		atomic_inc(&oldf->count);
1471 		goto out;
1472 	}
1473 
1474 	newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1475 	if (!newf)
1476 		goto out;
1477 
1478 	tsk->files = newf;
1479 	error = 0;
1480 out:
1481 	return error;
1482 }
1483 
1484 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1485 {
1486 #ifdef CONFIG_BLOCK
1487 	struct io_context *ioc = current->io_context;
1488 	struct io_context *new_ioc;
1489 
1490 	if (!ioc)
1491 		return 0;
1492 	/*
1493 	 * Share io context with parent, if CLONE_IO is set
1494 	 */
1495 	if (clone_flags & CLONE_IO) {
1496 		ioc_task_link(ioc);
1497 		tsk->io_context = ioc;
1498 	} else if (ioprio_valid(ioc->ioprio)) {
1499 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1500 		if (unlikely(!new_ioc))
1501 			return -ENOMEM;
1502 
1503 		new_ioc->ioprio = ioc->ioprio;
1504 		put_io_context(new_ioc);
1505 	}
1506 #endif
1507 	return 0;
1508 }
1509 
1510 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1511 {
1512 	struct sighand_struct *sig;
1513 
1514 	if (clone_flags & CLONE_SIGHAND) {
1515 		refcount_inc(&current->sighand->count);
1516 		return 0;
1517 	}
1518 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1519 	RCU_INIT_POINTER(tsk->sighand, sig);
1520 	if (!sig)
1521 		return -ENOMEM;
1522 
1523 	refcount_set(&sig->count, 1);
1524 	spin_lock_irq(&current->sighand->siglock);
1525 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1526 	spin_unlock_irq(&current->sighand->siglock);
1527 
1528 	/* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
1529 	if (clone_flags & CLONE_CLEAR_SIGHAND)
1530 		flush_signal_handlers(tsk, 0);
1531 
1532 	return 0;
1533 }
1534 
1535 void __cleanup_sighand(struct sighand_struct *sighand)
1536 {
1537 	if (refcount_dec_and_test(&sighand->count)) {
1538 		signalfd_cleanup(sighand);
1539 		/*
1540 		 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1541 		 * without an RCU grace period, see __lock_task_sighand().
1542 		 */
1543 		kmem_cache_free(sighand_cachep, sighand);
1544 	}
1545 }
1546 
1547 /*
1548  * Initialize POSIX timer handling for a thread group.
1549  */
1550 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1551 {
1552 	struct posix_cputimers *pct = &sig->posix_cputimers;
1553 	unsigned long cpu_limit;
1554 
1555 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1556 	posix_cputimers_group_init(pct, cpu_limit);
1557 }
1558 
1559 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1560 {
1561 	struct signal_struct *sig;
1562 
1563 	if (clone_flags & CLONE_THREAD)
1564 		return 0;
1565 
1566 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1567 	tsk->signal = sig;
1568 	if (!sig)
1569 		return -ENOMEM;
1570 
1571 	sig->nr_threads = 1;
1572 	atomic_set(&sig->live, 1);
1573 	refcount_set(&sig->sigcnt, 1);
1574 
1575 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1576 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1577 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1578 
1579 	init_waitqueue_head(&sig->wait_chldexit);
1580 	sig->curr_target = tsk;
1581 	init_sigpending(&sig->shared_pending);
1582 	INIT_HLIST_HEAD(&sig->multiprocess);
1583 	seqlock_init(&sig->stats_lock);
1584 	prev_cputime_init(&sig->prev_cputime);
1585 
1586 #ifdef CONFIG_POSIX_TIMERS
1587 	INIT_LIST_HEAD(&sig->posix_timers);
1588 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1589 	sig->real_timer.function = it_real_fn;
1590 #endif
1591 
1592 	task_lock(current->group_leader);
1593 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1594 	task_unlock(current->group_leader);
1595 
1596 	posix_cpu_timers_init_group(sig);
1597 
1598 	tty_audit_fork(sig);
1599 	sched_autogroup_fork(sig);
1600 
1601 	sig->oom_score_adj = current->signal->oom_score_adj;
1602 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1603 
1604 	mutex_init(&sig->cred_guard_mutex);
1605 	init_rwsem(&sig->exec_update_lock);
1606 
1607 	return 0;
1608 }
1609 
1610 static void copy_seccomp(struct task_struct *p)
1611 {
1612 #ifdef CONFIG_SECCOMP
1613 	/*
1614 	 * Must be called with sighand->lock held, which is common to
1615 	 * all threads in the group. Holding cred_guard_mutex is not
1616 	 * needed because this new task is not yet running and cannot
1617 	 * be racing exec.
1618 	 */
1619 	assert_spin_locked(&current->sighand->siglock);
1620 
1621 	/* Ref-count the new filter user, and assign it. */
1622 	get_seccomp_filter(current);
1623 	p->seccomp = current->seccomp;
1624 
1625 	/*
1626 	 * Explicitly enable no_new_privs here in case it got set
1627 	 * between the task_struct being duplicated and holding the
1628 	 * sighand lock. The seccomp state and nnp must be in sync.
1629 	 */
1630 	if (task_no_new_privs(current))
1631 		task_set_no_new_privs(p);
1632 
1633 	/*
1634 	 * If the parent gained a seccomp mode after copying thread
1635 	 * flags and between before we held the sighand lock, we have
1636 	 * to manually enable the seccomp thread flag here.
1637 	 */
1638 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1639 		set_task_syscall_work(p, SECCOMP);
1640 #endif
1641 }
1642 
1643 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1644 {
1645 	current->clear_child_tid = tidptr;
1646 
1647 	return task_pid_vnr(current);
1648 }
1649 
1650 static void rt_mutex_init_task(struct task_struct *p)
1651 {
1652 	raw_spin_lock_init(&p->pi_lock);
1653 #ifdef CONFIG_RT_MUTEXES
1654 	p->pi_waiters = RB_ROOT_CACHED;
1655 	p->pi_top_task = NULL;
1656 	p->pi_blocked_on = NULL;
1657 #endif
1658 }
1659 
1660 static inline void init_task_pid_links(struct task_struct *task)
1661 {
1662 	enum pid_type type;
1663 
1664 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type)
1665 		INIT_HLIST_NODE(&task->pid_links[type]);
1666 }
1667 
1668 static inline void
1669 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1670 {
1671 	if (type == PIDTYPE_PID)
1672 		task->thread_pid = pid;
1673 	else
1674 		task->signal->pids[type] = pid;
1675 }
1676 
1677 static inline void rcu_copy_process(struct task_struct *p)
1678 {
1679 #ifdef CONFIG_PREEMPT_RCU
1680 	p->rcu_read_lock_nesting = 0;
1681 	p->rcu_read_unlock_special.s = 0;
1682 	p->rcu_blocked_node = NULL;
1683 	INIT_LIST_HEAD(&p->rcu_node_entry);
1684 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1685 #ifdef CONFIG_TASKS_RCU
1686 	p->rcu_tasks_holdout = false;
1687 	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1688 	p->rcu_tasks_idle_cpu = -1;
1689 #endif /* #ifdef CONFIG_TASKS_RCU */
1690 #ifdef CONFIG_TASKS_TRACE_RCU
1691 	p->trc_reader_nesting = 0;
1692 	p->trc_reader_special.s = 0;
1693 	INIT_LIST_HEAD(&p->trc_holdout_list);
1694 #endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1695 }
1696 
1697 struct pid *pidfd_pid(const struct file *file)
1698 {
1699 	if (file->f_op == &pidfd_fops)
1700 		return file->private_data;
1701 
1702 	return ERR_PTR(-EBADF);
1703 }
1704 
1705 static int pidfd_release(struct inode *inode, struct file *file)
1706 {
1707 	struct pid *pid = file->private_data;
1708 
1709 	file->private_data = NULL;
1710 	put_pid(pid);
1711 	return 0;
1712 }
1713 
1714 #ifdef CONFIG_PROC_FS
1715 /**
1716  * pidfd_show_fdinfo - print information about a pidfd
1717  * @m: proc fdinfo file
1718  * @f: file referencing a pidfd
1719  *
1720  * Pid:
1721  * This function will print the pid that a given pidfd refers to in the
1722  * pid namespace of the procfs instance.
1723  * If the pid namespace of the process is not a descendant of the pid
1724  * namespace of the procfs instance 0 will be shown as its pid. This is
1725  * similar to calling getppid() on a process whose parent is outside of
1726  * its pid namespace.
1727  *
1728  * NSpid:
1729  * If pid namespaces are supported then this function will also print
1730  * the pid of a given pidfd refers to for all descendant pid namespaces
1731  * starting from the current pid namespace of the instance, i.e. the
1732  * Pid field and the first entry in the NSpid field will be identical.
1733  * If the pid namespace of the process is not a descendant of the pid
1734  * namespace of the procfs instance 0 will be shown as its first NSpid
1735  * entry and no others will be shown.
1736  * Note that this differs from the Pid and NSpid fields in
1737  * /proc/<pid>/status where Pid and NSpid are always shown relative to
1738  * the  pid namespace of the procfs instance. The difference becomes
1739  * obvious when sending around a pidfd between pid namespaces from a
1740  * different branch of the tree, i.e. where no ancestoral relation is
1741  * present between the pid namespaces:
1742  * - create two new pid namespaces ns1 and ns2 in the initial pid
1743  *   namespace (also take care to create new mount namespaces in the
1744  *   new pid namespace and mount procfs)
1745  * - create a process with a pidfd in ns1
1746  * - send pidfd from ns1 to ns2
1747  * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
1748  *   have exactly one entry, which is 0
1749  */
1750 static void pidfd_show_fdinfo(struct seq_file *m, struct file *f)
1751 {
1752 	struct pid *pid = f->private_data;
1753 	struct pid_namespace *ns;
1754 	pid_t nr = -1;
1755 
1756 	if (likely(pid_has_task(pid, PIDTYPE_PID))) {
1757 		ns = proc_pid_ns(file_inode(m->file)->i_sb);
1758 		nr = pid_nr_ns(pid, ns);
1759 	}
1760 
1761 	seq_put_decimal_ll(m, "Pid:\t", nr);
1762 
1763 #ifdef CONFIG_PID_NS
1764 	seq_put_decimal_ll(m, "\nNSpid:\t", nr);
1765 	if (nr > 0) {
1766 		int i;
1767 
1768 		/* If nr is non-zero it means that 'pid' is valid and that
1769 		 * ns, i.e. the pid namespace associated with the procfs
1770 		 * instance, is in the pid namespace hierarchy of pid.
1771 		 * Start at one below the already printed level.
1772 		 */
1773 		for (i = ns->level + 1; i <= pid->level; i++)
1774 			seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
1775 	}
1776 #endif
1777 	seq_putc(m, '\n');
1778 }
1779 #endif
1780 
1781 /*
1782  * Poll support for process exit notification.
1783  */
1784 static __poll_t pidfd_poll(struct file *file, struct poll_table_struct *pts)
1785 {
1786 	struct pid *pid = file->private_data;
1787 	__poll_t poll_flags = 0;
1788 
1789 	poll_wait(file, &pid->wait_pidfd, pts);
1790 
1791 	/*
1792 	 * Inform pollers only when the whole thread group exits.
1793 	 * If the thread group leader exits before all other threads in the
1794 	 * group, then poll(2) should block, similar to the wait(2) family.
1795 	 */
1796 	if (thread_group_exited(pid))
1797 		poll_flags = EPOLLIN | EPOLLRDNORM;
1798 
1799 	return poll_flags;
1800 }
1801 
1802 const struct file_operations pidfd_fops = {
1803 	.release = pidfd_release,
1804 	.poll = pidfd_poll,
1805 #ifdef CONFIG_PROC_FS
1806 	.show_fdinfo = pidfd_show_fdinfo,
1807 #endif
1808 };
1809 
1810 static void __delayed_free_task(struct rcu_head *rhp)
1811 {
1812 	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
1813 
1814 	free_task(tsk);
1815 }
1816 
1817 static __always_inline void delayed_free_task(struct task_struct *tsk)
1818 {
1819 	if (IS_ENABLED(CONFIG_MEMCG))
1820 		call_rcu(&tsk->rcu, __delayed_free_task);
1821 	else
1822 		free_task(tsk);
1823 }
1824 
1825 static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
1826 {
1827 	/* Skip if kernel thread */
1828 	if (!tsk->mm)
1829 		return;
1830 
1831 	/* Skip if spawning a thread or using vfork */
1832 	if ((clone_flags & (CLONE_VM | CLONE_THREAD | CLONE_VFORK)) != CLONE_VM)
1833 		return;
1834 
1835 	/* We need to synchronize with __set_oom_adj */
1836 	mutex_lock(&oom_adj_mutex);
1837 	set_bit(MMF_MULTIPROCESS, &tsk->mm->flags);
1838 	/* Update the values in case they were changed after copy_signal */
1839 	tsk->signal->oom_score_adj = current->signal->oom_score_adj;
1840 	tsk->signal->oom_score_adj_min = current->signal->oom_score_adj_min;
1841 	mutex_unlock(&oom_adj_mutex);
1842 }
1843 
1844 /*
1845  * This creates a new process as a copy of the old one,
1846  * but does not actually start it yet.
1847  *
1848  * It copies the registers, and all the appropriate
1849  * parts of the process environment (as per the clone
1850  * flags). The actual kick-off is left to the caller.
1851  */
1852 static __latent_entropy struct task_struct *copy_process(
1853 					struct pid *pid,
1854 					int trace,
1855 					int node,
1856 					struct kernel_clone_args *args)
1857 {
1858 	int pidfd = -1, retval;
1859 	struct task_struct *p;
1860 	struct multiprocess_signals delayed;
1861 	struct file *pidfile = NULL;
1862 	u64 clone_flags = args->flags;
1863 	struct nsproxy *nsp = current->nsproxy;
1864 
1865 	/*
1866 	 * Don't allow sharing the root directory with processes in a different
1867 	 * namespace
1868 	 */
1869 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1870 		return ERR_PTR(-EINVAL);
1871 
1872 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1873 		return ERR_PTR(-EINVAL);
1874 
1875 	/*
1876 	 * Thread groups must share signals as well, and detached threads
1877 	 * can only be started up within the thread group.
1878 	 */
1879 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1880 		return ERR_PTR(-EINVAL);
1881 
1882 	/*
1883 	 * Shared signal handlers imply shared VM. By way of the above,
1884 	 * thread groups also imply shared VM. Blocking this case allows
1885 	 * for various simplifications in other code.
1886 	 */
1887 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1888 		return ERR_PTR(-EINVAL);
1889 
1890 	/*
1891 	 * Siblings of global init remain as zombies on exit since they are
1892 	 * not reaped by their parent (swapper). To solve this and to avoid
1893 	 * multi-rooted process trees, prevent global and container-inits
1894 	 * from creating siblings.
1895 	 */
1896 	if ((clone_flags & CLONE_PARENT) &&
1897 				current->signal->flags & SIGNAL_UNKILLABLE)
1898 		return ERR_PTR(-EINVAL);
1899 
1900 	/*
1901 	 * If the new process will be in a different pid or user namespace
1902 	 * do not allow it to share a thread group with the forking task.
1903 	 */
1904 	if (clone_flags & CLONE_THREAD) {
1905 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1906 		    (task_active_pid_ns(current) != nsp->pid_ns_for_children))
1907 			return ERR_PTR(-EINVAL);
1908 	}
1909 
1910 	/*
1911 	 * If the new process will be in a different time namespace
1912 	 * do not allow it to share VM or a thread group with the forking task.
1913 	 */
1914 	if (clone_flags & (CLONE_THREAD | CLONE_VM)) {
1915 		if (nsp->time_ns != nsp->time_ns_for_children)
1916 			return ERR_PTR(-EINVAL);
1917 	}
1918 
1919 	if (clone_flags & CLONE_PIDFD) {
1920 		/*
1921 		 * - CLONE_DETACHED is blocked so that we can potentially
1922 		 *   reuse it later for CLONE_PIDFD.
1923 		 * - CLONE_THREAD is blocked until someone really needs it.
1924 		 */
1925 		if (clone_flags & (CLONE_DETACHED | CLONE_THREAD))
1926 			return ERR_PTR(-EINVAL);
1927 	}
1928 
1929 	/*
1930 	 * Force any signals received before this point to be delivered
1931 	 * before the fork happens.  Collect up signals sent to multiple
1932 	 * processes that happen during the fork and delay them so that
1933 	 * they appear to happen after the fork.
1934 	 */
1935 	sigemptyset(&delayed.signal);
1936 	INIT_HLIST_NODE(&delayed.node);
1937 
1938 	spin_lock_irq(&current->sighand->siglock);
1939 	if (!(clone_flags & CLONE_THREAD))
1940 		hlist_add_head(&delayed.node, &current->signal->multiprocess);
1941 	recalc_sigpending();
1942 	spin_unlock_irq(&current->sighand->siglock);
1943 	retval = -ERESTARTNOINTR;
1944 	if (signal_pending(current))
1945 		goto fork_out;
1946 
1947 	retval = -ENOMEM;
1948 	p = dup_task_struct(current, node);
1949 	if (!p)
1950 		goto fork_out;
1951 	if (args->io_thread) {
1952 		/*
1953 		 * Mark us an IO worker, and block any signal that isn't
1954 		 * fatal or STOP
1955 		 */
1956 		p->flags |= PF_IO_WORKER;
1957 		siginitsetinv(&p->blocked, sigmask(SIGKILL)|sigmask(SIGSTOP));
1958 	}
1959 
1960 	/*
1961 	 * This _must_ happen before we call free_task(), i.e. before we jump
1962 	 * to any of the bad_fork_* labels. This is to avoid freeing
1963 	 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1964 	 * kernel threads (PF_KTHREAD).
1965 	 */
1966 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1967 	/*
1968 	 * Clear TID on mm_release()?
1969 	 */
1970 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1971 
1972 	ftrace_graph_init_task(p);
1973 
1974 	rt_mutex_init_task(p);
1975 
1976 	lockdep_assert_irqs_enabled();
1977 #ifdef CONFIG_PROVE_LOCKING
1978 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1979 #endif
1980 	retval = -EAGAIN;
1981 	if (atomic_read(&p->real_cred->user->processes) >=
1982 			task_rlimit(p, RLIMIT_NPROC)) {
1983 		if (p->real_cred->user != INIT_USER &&
1984 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1985 			goto bad_fork_free;
1986 	}
1987 	current->flags &= ~PF_NPROC_EXCEEDED;
1988 
1989 	retval = copy_creds(p, clone_flags);
1990 	if (retval < 0)
1991 		goto bad_fork_free;
1992 
1993 	/*
1994 	 * If multiple threads are within copy_process(), then this check
1995 	 * triggers too late. This doesn't hurt, the check is only there
1996 	 * to stop root fork bombs.
1997 	 */
1998 	retval = -EAGAIN;
1999 	if (data_race(nr_threads >= max_threads))
2000 		goto bad_fork_cleanup_count;
2001 
2002 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
2003 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
2004 	p->flags |= PF_FORKNOEXEC;
2005 	INIT_LIST_HEAD(&p->children);
2006 	INIT_LIST_HEAD(&p->sibling);
2007 	rcu_copy_process(p);
2008 	p->vfork_done = NULL;
2009 	spin_lock_init(&p->alloc_lock);
2010 
2011 	init_sigpending(&p->pending);
2012 
2013 	p->utime = p->stime = p->gtime = 0;
2014 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
2015 	p->utimescaled = p->stimescaled = 0;
2016 #endif
2017 	prev_cputime_init(&p->prev_cputime);
2018 
2019 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
2020 	seqcount_init(&p->vtime.seqcount);
2021 	p->vtime.starttime = 0;
2022 	p->vtime.state = VTIME_INACTIVE;
2023 #endif
2024 
2025 #ifdef CONFIG_IO_URING
2026 	p->io_uring = NULL;
2027 #endif
2028 
2029 #if defined(SPLIT_RSS_COUNTING)
2030 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
2031 #endif
2032 
2033 	p->default_timer_slack_ns = current->timer_slack_ns;
2034 
2035 #ifdef CONFIG_PSI
2036 	p->psi_flags = 0;
2037 #endif
2038 
2039 	task_io_accounting_init(&p->ioac);
2040 	acct_clear_integrals(p);
2041 
2042 	posix_cputimers_init(&p->posix_cputimers);
2043 
2044 	p->io_context = NULL;
2045 	audit_set_context(p, NULL);
2046 	cgroup_fork(p);
2047 #ifdef CONFIG_NUMA
2048 	p->mempolicy = mpol_dup(p->mempolicy);
2049 	if (IS_ERR(p->mempolicy)) {
2050 		retval = PTR_ERR(p->mempolicy);
2051 		p->mempolicy = NULL;
2052 		goto bad_fork_cleanup_threadgroup_lock;
2053 	}
2054 #endif
2055 #ifdef CONFIG_CPUSETS
2056 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2057 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2058 	seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2059 #endif
2060 #ifdef CONFIG_TRACE_IRQFLAGS
2061 	memset(&p->irqtrace, 0, sizeof(p->irqtrace));
2062 	p->irqtrace.hardirq_disable_ip	= _THIS_IP_;
2063 	p->irqtrace.softirq_enable_ip	= _THIS_IP_;
2064 	p->softirqs_enabled		= 1;
2065 	p->softirq_context		= 0;
2066 #endif
2067 
2068 	p->pagefault_disabled = 0;
2069 
2070 #ifdef CONFIG_LOCKDEP
2071 	lockdep_init_task(p);
2072 #endif
2073 
2074 #ifdef CONFIG_DEBUG_MUTEXES
2075 	p->blocked_on = NULL; /* not blocked yet */
2076 #endif
2077 #ifdef CONFIG_BCACHE
2078 	p->sequential_io	= 0;
2079 	p->sequential_io_avg	= 0;
2080 #endif
2081 
2082 	/* Perform scheduler related setup. Assign this task to a CPU. */
2083 	retval = sched_fork(clone_flags, p);
2084 	if (retval)
2085 		goto bad_fork_cleanup_policy;
2086 
2087 	retval = perf_event_init_task(p);
2088 	if (retval)
2089 		goto bad_fork_cleanup_policy;
2090 	retval = audit_alloc(p);
2091 	if (retval)
2092 		goto bad_fork_cleanup_perf;
2093 	/* copy all the process information */
2094 	shm_init_task(p);
2095 	retval = security_task_alloc(p, clone_flags);
2096 	if (retval)
2097 		goto bad_fork_cleanup_audit;
2098 	retval = copy_semundo(clone_flags, p);
2099 	if (retval)
2100 		goto bad_fork_cleanup_security;
2101 	retval = copy_files(clone_flags, p);
2102 	if (retval)
2103 		goto bad_fork_cleanup_semundo;
2104 	retval = copy_fs(clone_flags, p);
2105 	if (retval)
2106 		goto bad_fork_cleanup_files;
2107 	retval = copy_sighand(clone_flags, p);
2108 	if (retval)
2109 		goto bad_fork_cleanup_fs;
2110 	retval = copy_signal(clone_flags, p);
2111 	if (retval)
2112 		goto bad_fork_cleanup_sighand;
2113 	retval = copy_mm(clone_flags, p);
2114 	if (retval)
2115 		goto bad_fork_cleanup_signal;
2116 	retval = copy_namespaces(clone_flags, p);
2117 	if (retval)
2118 		goto bad_fork_cleanup_mm;
2119 	retval = copy_io(clone_flags, p);
2120 	if (retval)
2121 		goto bad_fork_cleanup_namespaces;
2122 	retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2123 	if (retval)
2124 		goto bad_fork_cleanup_io;
2125 
2126 	stackleak_task_init(p);
2127 
2128 	if (pid != &init_struct_pid) {
2129 		pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
2130 				args->set_tid_size);
2131 		if (IS_ERR(pid)) {
2132 			retval = PTR_ERR(pid);
2133 			goto bad_fork_cleanup_thread;
2134 		}
2135 	}
2136 
2137 	/*
2138 	 * This has to happen after we've potentially unshared the file
2139 	 * descriptor table (so that the pidfd doesn't leak into the child
2140 	 * if the fd table isn't shared).
2141 	 */
2142 	if (clone_flags & CLONE_PIDFD) {
2143 		retval = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
2144 		if (retval < 0)
2145 			goto bad_fork_free_pid;
2146 
2147 		pidfd = retval;
2148 
2149 		pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
2150 					      O_RDWR | O_CLOEXEC);
2151 		if (IS_ERR(pidfile)) {
2152 			put_unused_fd(pidfd);
2153 			retval = PTR_ERR(pidfile);
2154 			goto bad_fork_free_pid;
2155 		}
2156 		get_pid(pid);	/* held by pidfile now */
2157 
2158 		retval = put_user(pidfd, args->pidfd);
2159 		if (retval)
2160 			goto bad_fork_put_pidfd;
2161 	}
2162 
2163 #ifdef CONFIG_BLOCK
2164 	p->plug = NULL;
2165 #endif
2166 	futex_init_task(p);
2167 
2168 	/*
2169 	 * sigaltstack should be cleared when sharing the same VM
2170 	 */
2171 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
2172 		sas_ss_reset(p);
2173 
2174 	/*
2175 	 * Syscall tracing and stepping should be turned off in the
2176 	 * child regardless of CLONE_PTRACE.
2177 	 */
2178 	user_disable_single_step(p);
2179 	clear_task_syscall_work(p, SYSCALL_TRACE);
2180 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
2181 	clear_task_syscall_work(p, SYSCALL_EMU);
2182 #endif
2183 	clear_tsk_latency_tracing(p);
2184 
2185 	/* ok, now we should be set up.. */
2186 	p->pid = pid_nr(pid);
2187 	if (clone_flags & CLONE_THREAD) {
2188 		p->group_leader = current->group_leader;
2189 		p->tgid = current->tgid;
2190 	} else {
2191 		p->group_leader = p;
2192 		p->tgid = p->pid;
2193 	}
2194 
2195 	p->nr_dirtied = 0;
2196 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
2197 	p->dirty_paused_when = 0;
2198 
2199 	p->pdeath_signal = 0;
2200 	INIT_LIST_HEAD(&p->thread_group);
2201 	p->task_works = NULL;
2202 
2203 #ifdef CONFIG_KRETPROBES
2204 	p->kretprobe_instances.first = NULL;
2205 #endif
2206 
2207 	/*
2208 	 * Ensure that the cgroup subsystem policies allow the new process to be
2209 	 * forked. It should be noted that the new process's css_set can be changed
2210 	 * between here and cgroup_post_fork() if an organisation operation is in
2211 	 * progress.
2212 	 */
2213 	retval = cgroup_can_fork(p, args);
2214 	if (retval)
2215 		goto bad_fork_put_pidfd;
2216 
2217 	/*
2218 	 * From this point on we must avoid any synchronous user-space
2219 	 * communication until we take the tasklist-lock. In particular, we do
2220 	 * not want user-space to be able to predict the process start-time by
2221 	 * stalling fork(2) after we recorded the start_time but before it is
2222 	 * visible to the system.
2223 	 */
2224 
2225 	p->start_time = ktime_get_ns();
2226 	p->start_boottime = ktime_get_boottime_ns();
2227 
2228 	/*
2229 	 * Make it visible to the rest of the system, but dont wake it up yet.
2230 	 * Need tasklist lock for parent etc handling!
2231 	 */
2232 	write_lock_irq(&tasklist_lock);
2233 
2234 	/* CLONE_PARENT re-uses the old parent */
2235 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
2236 		p->real_parent = current->real_parent;
2237 		p->parent_exec_id = current->parent_exec_id;
2238 		if (clone_flags & CLONE_THREAD)
2239 			p->exit_signal = -1;
2240 		else
2241 			p->exit_signal = current->group_leader->exit_signal;
2242 	} else {
2243 		p->real_parent = current;
2244 		p->parent_exec_id = current->self_exec_id;
2245 		p->exit_signal = args->exit_signal;
2246 	}
2247 
2248 	klp_copy_process(p);
2249 
2250 	spin_lock(&current->sighand->siglock);
2251 
2252 	/*
2253 	 * Copy seccomp details explicitly here, in case they were changed
2254 	 * before holding sighand lock.
2255 	 */
2256 	copy_seccomp(p);
2257 
2258 	rseq_fork(p, clone_flags);
2259 
2260 	/* Don't start children in a dying pid namespace */
2261 	if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
2262 		retval = -ENOMEM;
2263 		goto bad_fork_cancel_cgroup;
2264 	}
2265 
2266 	/* Let kill terminate clone/fork in the middle */
2267 	if (fatal_signal_pending(current)) {
2268 		retval = -EINTR;
2269 		goto bad_fork_cancel_cgroup;
2270 	}
2271 
2272 	/* past the last point of failure */
2273 	if (pidfile)
2274 		fd_install(pidfd, pidfile);
2275 
2276 	init_task_pid_links(p);
2277 	if (likely(p->pid)) {
2278 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
2279 
2280 		init_task_pid(p, PIDTYPE_PID, pid);
2281 		if (thread_group_leader(p)) {
2282 			init_task_pid(p, PIDTYPE_TGID, pid);
2283 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
2284 			init_task_pid(p, PIDTYPE_SID, task_session(current));
2285 
2286 			if (is_child_reaper(pid)) {
2287 				ns_of_pid(pid)->child_reaper = p;
2288 				p->signal->flags |= SIGNAL_UNKILLABLE;
2289 			}
2290 			p->signal->shared_pending.signal = delayed.signal;
2291 			p->signal->tty = tty_kref_get(current->signal->tty);
2292 			/*
2293 			 * Inherit has_child_subreaper flag under the same
2294 			 * tasklist_lock with adding child to the process tree
2295 			 * for propagate_has_child_subreaper optimization.
2296 			 */
2297 			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
2298 							 p->real_parent->signal->is_child_subreaper;
2299 			list_add_tail(&p->sibling, &p->real_parent->children);
2300 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
2301 			attach_pid(p, PIDTYPE_TGID);
2302 			attach_pid(p, PIDTYPE_PGID);
2303 			attach_pid(p, PIDTYPE_SID);
2304 			__this_cpu_inc(process_counts);
2305 		} else {
2306 			current->signal->nr_threads++;
2307 			atomic_inc(&current->signal->live);
2308 			refcount_inc(&current->signal->sigcnt);
2309 			task_join_group_stop(p);
2310 			list_add_tail_rcu(&p->thread_group,
2311 					  &p->group_leader->thread_group);
2312 			list_add_tail_rcu(&p->thread_node,
2313 					  &p->signal->thread_head);
2314 		}
2315 		attach_pid(p, PIDTYPE_PID);
2316 		nr_threads++;
2317 	}
2318 	total_forks++;
2319 	hlist_del_init(&delayed.node);
2320 	spin_unlock(&current->sighand->siglock);
2321 	syscall_tracepoint_update(p);
2322 	write_unlock_irq(&tasklist_lock);
2323 
2324 	proc_fork_connector(p);
2325 	sched_post_fork(p);
2326 	cgroup_post_fork(p, args);
2327 	perf_event_fork(p);
2328 
2329 	trace_task_newtask(p, clone_flags);
2330 	uprobe_copy_process(p, clone_flags);
2331 
2332 	copy_oom_score_adj(clone_flags, p);
2333 
2334 	return p;
2335 
2336 bad_fork_cancel_cgroup:
2337 	spin_unlock(&current->sighand->siglock);
2338 	write_unlock_irq(&tasklist_lock);
2339 	cgroup_cancel_fork(p, args);
2340 bad_fork_put_pidfd:
2341 	if (clone_flags & CLONE_PIDFD) {
2342 		fput(pidfile);
2343 		put_unused_fd(pidfd);
2344 	}
2345 bad_fork_free_pid:
2346 	if (pid != &init_struct_pid)
2347 		free_pid(pid);
2348 bad_fork_cleanup_thread:
2349 	exit_thread(p);
2350 bad_fork_cleanup_io:
2351 	if (p->io_context)
2352 		exit_io_context(p);
2353 bad_fork_cleanup_namespaces:
2354 	exit_task_namespaces(p);
2355 bad_fork_cleanup_mm:
2356 	if (p->mm) {
2357 		mm_clear_owner(p->mm, p);
2358 		mmput(p->mm);
2359 	}
2360 bad_fork_cleanup_signal:
2361 	if (!(clone_flags & CLONE_THREAD))
2362 		free_signal_struct(p->signal);
2363 bad_fork_cleanup_sighand:
2364 	__cleanup_sighand(p->sighand);
2365 bad_fork_cleanup_fs:
2366 	exit_fs(p); /* blocking */
2367 bad_fork_cleanup_files:
2368 	exit_files(p); /* blocking */
2369 bad_fork_cleanup_semundo:
2370 	exit_sem(p);
2371 bad_fork_cleanup_security:
2372 	security_task_free(p);
2373 bad_fork_cleanup_audit:
2374 	audit_free(p);
2375 bad_fork_cleanup_perf:
2376 	perf_event_free_task(p);
2377 bad_fork_cleanup_policy:
2378 	lockdep_free_task(p);
2379 #ifdef CONFIG_NUMA
2380 	mpol_put(p->mempolicy);
2381 bad_fork_cleanup_threadgroup_lock:
2382 #endif
2383 	delayacct_tsk_free(p);
2384 bad_fork_cleanup_count:
2385 	atomic_dec(&p->cred->user->processes);
2386 	exit_creds(p);
2387 bad_fork_free:
2388 	p->state = TASK_DEAD;
2389 	put_task_stack(p);
2390 	delayed_free_task(p);
2391 fork_out:
2392 	spin_lock_irq(&current->sighand->siglock);
2393 	hlist_del_init(&delayed.node);
2394 	spin_unlock_irq(&current->sighand->siglock);
2395 	return ERR_PTR(retval);
2396 }
2397 
2398 static inline void init_idle_pids(struct task_struct *idle)
2399 {
2400 	enum pid_type type;
2401 
2402 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2403 		INIT_HLIST_NODE(&idle->pid_links[type]); /* not really needed */
2404 		init_task_pid(idle, type, &init_struct_pid);
2405 	}
2406 }
2407 
2408 struct task_struct *fork_idle(int cpu)
2409 {
2410 	struct task_struct *task;
2411 	struct kernel_clone_args args = {
2412 		.flags = CLONE_VM,
2413 	};
2414 
2415 	task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2416 	if (!IS_ERR(task)) {
2417 		init_idle_pids(task);
2418 		init_idle(task, cpu);
2419 	}
2420 
2421 	return task;
2422 }
2423 
2424 struct mm_struct *copy_init_mm(void)
2425 {
2426 	return dup_mm(NULL, &init_mm);
2427 }
2428 
2429 /*
2430  * This is like kernel_clone(), but shaved down and tailored to just
2431  * creating io_uring workers. It returns a created task, or an error pointer.
2432  * The returned task is inactive, and the caller must fire it up through
2433  * wake_up_new_task(p). All signals are blocked in the created task.
2434  */
2435 struct task_struct *create_io_thread(int (*fn)(void *), void *arg, int node)
2436 {
2437 	unsigned long flags = CLONE_FS|CLONE_FILES|CLONE_SIGHAND|CLONE_THREAD|
2438 				CLONE_IO;
2439 	struct kernel_clone_args args = {
2440 		.flags		= ((lower_32_bits(flags) | CLONE_VM |
2441 				    CLONE_UNTRACED) & ~CSIGNAL),
2442 		.exit_signal	= (lower_32_bits(flags) & CSIGNAL),
2443 		.stack		= (unsigned long)fn,
2444 		.stack_size	= (unsigned long)arg,
2445 		.io_thread	= 1,
2446 	};
2447 
2448 	return copy_process(NULL, 0, node, &args);
2449 }
2450 
2451 /*
2452  *  Ok, this is the main fork-routine.
2453  *
2454  * It copies the process, and if successful kick-starts
2455  * it and waits for it to finish using the VM if required.
2456  *
2457  * args->exit_signal is expected to be checked for sanity by the caller.
2458  */
2459 pid_t kernel_clone(struct kernel_clone_args *args)
2460 {
2461 	u64 clone_flags = args->flags;
2462 	struct completion vfork;
2463 	struct pid *pid;
2464 	struct task_struct *p;
2465 	int trace = 0;
2466 	pid_t nr;
2467 
2468 	/*
2469 	 * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
2470 	 * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
2471 	 * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
2472 	 * field in struct clone_args and it still doesn't make sense to have
2473 	 * them both point at the same memory location. Performing this check
2474 	 * here has the advantage that we don't need to have a separate helper
2475 	 * to check for legacy clone().
2476 	 */
2477 	if ((args->flags & CLONE_PIDFD) &&
2478 	    (args->flags & CLONE_PARENT_SETTID) &&
2479 	    (args->pidfd == args->parent_tid))
2480 		return -EINVAL;
2481 
2482 	/*
2483 	 * Determine whether and which event to report to ptracer.  When
2484 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
2485 	 * requested, no event is reported; otherwise, report if the event
2486 	 * for the type of forking is enabled.
2487 	 */
2488 	if (!(clone_flags & CLONE_UNTRACED)) {
2489 		if (clone_flags & CLONE_VFORK)
2490 			trace = PTRACE_EVENT_VFORK;
2491 		else if (args->exit_signal != SIGCHLD)
2492 			trace = PTRACE_EVENT_CLONE;
2493 		else
2494 			trace = PTRACE_EVENT_FORK;
2495 
2496 		if (likely(!ptrace_event_enabled(current, trace)))
2497 			trace = 0;
2498 	}
2499 
2500 	p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2501 	add_latent_entropy();
2502 
2503 	if (IS_ERR(p))
2504 		return PTR_ERR(p);
2505 
2506 	/*
2507 	 * Do this prior waking up the new thread - the thread pointer
2508 	 * might get invalid after that point, if the thread exits quickly.
2509 	 */
2510 	trace_sched_process_fork(current, p);
2511 
2512 	pid = get_task_pid(p, PIDTYPE_PID);
2513 	nr = pid_vnr(pid);
2514 
2515 	if (clone_flags & CLONE_PARENT_SETTID)
2516 		put_user(nr, args->parent_tid);
2517 
2518 	if (clone_flags & CLONE_VFORK) {
2519 		p->vfork_done = &vfork;
2520 		init_completion(&vfork);
2521 		get_task_struct(p);
2522 	}
2523 
2524 	wake_up_new_task(p);
2525 
2526 	/* forking complete and child started to run, tell ptracer */
2527 	if (unlikely(trace))
2528 		ptrace_event_pid(trace, pid);
2529 
2530 	if (clone_flags & CLONE_VFORK) {
2531 		if (!wait_for_vfork_done(p, &vfork))
2532 			ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2533 	}
2534 
2535 	put_pid(pid);
2536 	return nr;
2537 }
2538 
2539 /*
2540  * Create a kernel thread.
2541  */
2542 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2543 {
2544 	struct kernel_clone_args args = {
2545 		.flags		= ((lower_32_bits(flags) | CLONE_VM |
2546 				    CLONE_UNTRACED) & ~CSIGNAL),
2547 		.exit_signal	= (lower_32_bits(flags) & CSIGNAL),
2548 		.stack		= (unsigned long)fn,
2549 		.stack_size	= (unsigned long)arg,
2550 	};
2551 
2552 	return kernel_clone(&args);
2553 }
2554 
2555 #ifdef __ARCH_WANT_SYS_FORK
2556 SYSCALL_DEFINE0(fork)
2557 {
2558 #ifdef CONFIG_MMU
2559 	struct kernel_clone_args args = {
2560 		.exit_signal = SIGCHLD,
2561 	};
2562 
2563 	return kernel_clone(&args);
2564 #else
2565 	/* can not support in nommu mode */
2566 	return -EINVAL;
2567 #endif
2568 }
2569 #endif
2570 
2571 #ifdef __ARCH_WANT_SYS_VFORK
2572 SYSCALL_DEFINE0(vfork)
2573 {
2574 	struct kernel_clone_args args = {
2575 		.flags		= CLONE_VFORK | CLONE_VM,
2576 		.exit_signal	= SIGCHLD,
2577 	};
2578 
2579 	return kernel_clone(&args);
2580 }
2581 #endif
2582 
2583 #ifdef __ARCH_WANT_SYS_CLONE
2584 #ifdef CONFIG_CLONE_BACKWARDS
2585 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2586 		 int __user *, parent_tidptr,
2587 		 unsigned long, tls,
2588 		 int __user *, child_tidptr)
2589 #elif defined(CONFIG_CLONE_BACKWARDS2)
2590 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2591 		 int __user *, parent_tidptr,
2592 		 int __user *, child_tidptr,
2593 		 unsigned long, tls)
2594 #elif defined(CONFIG_CLONE_BACKWARDS3)
2595 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2596 		int, stack_size,
2597 		int __user *, parent_tidptr,
2598 		int __user *, child_tidptr,
2599 		unsigned long, tls)
2600 #else
2601 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2602 		 int __user *, parent_tidptr,
2603 		 int __user *, child_tidptr,
2604 		 unsigned long, tls)
2605 #endif
2606 {
2607 	struct kernel_clone_args args = {
2608 		.flags		= (lower_32_bits(clone_flags) & ~CSIGNAL),
2609 		.pidfd		= parent_tidptr,
2610 		.child_tid	= child_tidptr,
2611 		.parent_tid	= parent_tidptr,
2612 		.exit_signal	= (lower_32_bits(clone_flags) & CSIGNAL),
2613 		.stack		= newsp,
2614 		.tls		= tls,
2615 	};
2616 
2617 	return kernel_clone(&args);
2618 }
2619 #endif
2620 
2621 #ifdef __ARCH_WANT_SYS_CLONE3
2622 
2623 noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
2624 					      struct clone_args __user *uargs,
2625 					      size_t usize)
2626 {
2627 	int err;
2628 	struct clone_args args;
2629 	pid_t *kset_tid = kargs->set_tid;
2630 
2631 	BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
2632 		     CLONE_ARGS_SIZE_VER0);
2633 	BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
2634 		     CLONE_ARGS_SIZE_VER1);
2635 	BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
2636 		     CLONE_ARGS_SIZE_VER2);
2637 	BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
2638 
2639 	if (unlikely(usize > PAGE_SIZE))
2640 		return -E2BIG;
2641 	if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
2642 		return -EINVAL;
2643 
2644 	err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
2645 	if (err)
2646 		return err;
2647 
2648 	if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
2649 		return -EINVAL;
2650 
2651 	if (unlikely(!args.set_tid && args.set_tid_size > 0))
2652 		return -EINVAL;
2653 
2654 	if (unlikely(args.set_tid && args.set_tid_size == 0))
2655 		return -EINVAL;
2656 
2657 	/*
2658 	 * Verify that higher 32bits of exit_signal are unset and that
2659 	 * it is a valid signal
2660 	 */
2661 	if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) ||
2662 		     !valid_signal(args.exit_signal)))
2663 		return -EINVAL;
2664 
2665 	if ((args.flags & CLONE_INTO_CGROUP) &&
2666 	    (args.cgroup > INT_MAX || usize < CLONE_ARGS_SIZE_VER2))
2667 		return -EINVAL;
2668 
2669 	*kargs = (struct kernel_clone_args){
2670 		.flags		= args.flags,
2671 		.pidfd		= u64_to_user_ptr(args.pidfd),
2672 		.child_tid	= u64_to_user_ptr(args.child_tid),
2673 		.parent_tid	= u64_to_user_ptr(args.parent_tid),
2674 		.exit_signal	= args.exit_signal,
2675 		.stack		= args.stack,
2676 		.stack_size	= args.stack_size,
2677 		.tls		= args.tls,
2678 		.set_tid_size	= args.set_tid_size,
2679 		.cgroup		= args.cgroup,
2680 	};
2681 
2682 	if (args.set_tid &&
2683 		copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
2684 			(kargs->set_tid_size * sizeof(pid_t))))
2685 		return -EFAULT;
2686 
2687 	kargs->set_tid = kset_tid;
2688 
2689 	return 0;
2690 }
2691 
2692 /**
2693  * clone3_stack_valid - check and prepare stack
2694  * @kargs: kernel clone args
2695  *
2696  * Verify that the stack arguments userspace gave us are sane.
2697  * In addition, set the stack direction for userspace since it's easy for us to
2698  * determine.
2699  */
2700 static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
2701 {
2702 	if (kargs->stack == 0) {
2703 		if (kargs->stack_size > 0)
2704 			return false;
2705 	} else {
2706 		if (kargs->stack_size == 0)
2707 			return false;
2708 
2709 		if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
2710 			return false;
2711 
2712 #if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
2713 		kargs->stack += kargs->stack_size;
2714 #endif
2715 	}
2716 
2717 	return true;
2718 }
2719 
2720 static bool clone3_args_valid(struct kernel_clone_args *kargs)
2721 {
2722 	/* Verify that no unknown flags are passed along. */
2723 	if (kargs->flags &
2724 	    ~(CLONE_LEGACY_FLAGS | CLONE_CLEAR_SIGHAND | CLONE_INTO_CGROUP))
2725 		return false;
2726 
2727 	/*
2728 	 * - make the CLONE_DETACHED bit reuseable for clone3
2729 	 * - make the CSIGNAL bits reuseable for clone3
2730 	 */
2731 	if (kargs->flags & (CLONE_DETACHED | CSIGNAL))
2732 		return false;
2733 
2734 	if ((kargs->flags & (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND)) ==
2735 	    (CLONE_SIGHAND | CLONE_CLEAR_SIGHAND))
2736 		return false;
2737 
2738 	if ((kargs->flags & (CLONE_THREAD | CLONE_PARENT)) &&
2739 	    kargs->exit_signal)
2740 		return false;
2741 
2742 	if (!clone3_stack_valid(kargs))
2743 		return false;
2744 
2745 	return true;
2746 }
2747 
2748 /**
2749  * clone3 - create a new process with specific properties
2750  * @uargs: argument structure
2751  * @size:  size of @uargs
2752  *
2753  * clone3() is the extensible successor to clone()/clone2().
2754  * It takes a struct as argument that is versioned by its size.
2755  *
2756  * Return: On success, a positive PID for the child process.
2757  *         On error, a negative errno number.
2758  */
2759 SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
2760 {
2761 	int err;
2762 
2763 	struct kernel_clone_args kargs;
2764 	pid_t set_tid[MAX_PID_NS_LEVEL];
2765 
2766 	kargs.set_tid = set_tid;
2767 
2768 	err = copy_clone_args_from_user(&kargs, uargs, size);
2769 	if (err)
2770 		return err;
2771 
2772 	if (!clone3_args_valid(&kargs))
2773 		return -EINVAL;
2774 
2775 	return kernel_clone(&kargs);
2776 }
2777 #endif
2778 
2779 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2780 {
2781 	struct task_struct *leader, *parent, *child;
2782 	int res;
2783 
2784 	read_lock(&tasklist_lock);
2785 	leader = top = top->group_leader;
2786 down:
2787 	for_each_thread(leader, parent) {
2788 		list_for_each_entry(child, &parent->children, sibling) {
2789 			res = visitor(child, data);
2790 			if (res) {
2791 				if (res < 0)
2792 					goto out;
2793 				leader = child;
2794 				goto down;
2795 			}
2796 up:
2797 			;
2798 		}
2799 	}
2800 
2801 	if (leader != top) {
2802 		child = leader;
2803 		parent = child->real_parent;
2804 		leader = parent->group_leader;
2805 		goto up;
2806 	}
2807 out:
2808 	read_unlock(&tasklist_lock);
2809 }
2810 
2811 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2812 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2813 #endif
2814 
2815 static void sighand_ctor(void *data)
2816 {
2817 	struct sighand_struct *sighand = data;
2818 
2819 	spin_lock_init(&sighand->siglock);
2820 	init_waitqueue_head(&sighand->signalfd_wqh);
2821 }
2822 
2823 void __init proc_caches_init(void)
2824 {
2825 	unsigned int mm_size;
2826 
2827 	sighand_cachep = kmem_cache_create("sighand_cache",
2828 			sizeof(struct sighand_struct), 0,
2829 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2830 			SLAB_ACCOUNT, sighand_ctor);
2831 	signal_cachep = kmem_cache_create("signal_cache",
2832 			sizeof(struct signal_struct), 0,
2833 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2834 			NULL);
2835 	files_cachep = kmem_cache_create("files_cache",
2836 			sizeof(struct files_struct), 0,
2837 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2838 			NULL);
2839 	fs_cachep = kmem_cache_create("fs_cache",
2840 			sizeof(struct fs_struct), 0,
2841 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2842 			NULL);
2843 
2844 	/*
2845 	 * The mm_cpumask is located at the end of mm_struct, and is
2846 	 * dynamically sized based on the maximum CPU number this system
2847 	 * can have, taking hotplug into account (nr_cpu_ids).
2848 	 */
2849 	mm_size = sizeof(struct mm_struct) + cpumask_size();
2850 
2851 	mm_cachep = kmem_cache_create_usercopy("mm_struct",
2852 			mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2853 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2854 			offsetof(struct mm_struct, saved_auxv),
2855 			sizeof_field(struct mm_struct, saved_auxv),
2856 			NULL);
2857 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2858 	mmap_init();
2859 	nsproxy_cache_init();
2860 }
2861 
2862 /*
2863  * Check constraints on flags passed to the unshare system call.
2864  */
2865 static int check_unshare_flags(unsigned long unshare_flags)
2866 {
2867 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2868 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2869 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2870 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP|
2871 				CLONE_NEWTIME))
2872 		return -EINVAL;
2873 	/*
2874 	 * Not implemented, but pretend it works if there is nothing
2875 	 * to unshare.  Note that unsharing the address space or the
2876 	 * signal handlers also need to unshare the signal queues (aka
2877 	 * CLONE_THREAD).
2878 	 */
2879 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2880 		if (!thread_group_empty(current))
2881 			return -EINVAL;
2882 	}
2883 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2884 		if (refcount_read(&current->sighand->count) > 1)
2885 			return -EINVAL;
2886 	}
2887 	if (unshare_flags & CLONE_VM) {
2888 		if (!current_is_single_threaded())
2889 			return -EINVAL;
2890 	}
2891 
2892 	return 0;
2893 }
2894 
2895 /*
2896  * Unshare the filesystem structure if it is being shared
2897  */
2898 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2899 {
2900 	struct fs_struct *fs = current->fs;
2901 
2902 	if (!(unshare_flags & CLONE_FS) || !fs)
2903 		return 0;
2904 
2905 	/* don't need lock here; in the worst case we'll do useless copy */
2906 	if (fs->users == 1)
2907 		return 0;
2908 
2909 	*new_fsp = copy_fs_struct(fs);
2910 	if (!*new_fsp)
2911 		return -ENOMEM;
2912 
2913 	return 0;
2914 }
2915 
2916 /*
2917  * Unshare file descriptor table if it is being shared
2918  */
2919 int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
2920 	       struct files_struct **new_fdp)
2921 {
2922 	struct files_struct *fd = current->files;
2923 	int error = 0;
2924 
2925 	if ((unshare_flags & CLONE_FILES) &&
2926 	    (fd && atomic_read(&fd->count) > 1)) {
2927 		*new_fdp = dup_fd(fd, max_fds, &error);
2928 		if (!*new_fdp)
2929 			return error;
2930 	}
2931 
2932 	return 0;
2933 }
2934 
2935 /*
2936  * unshare allows a process to 'unshare' part of the process
2937  * context which was originally shared using clone.  copy_*
2938  * functions used by kernel_clone() cannot be used here directly
2939  * because they modify an inactive task_struct that is being
2940  * constructed. Here we are modifying the current, active,
2941  * task_struct.
2942  */
2943 int ksys_unshare(unsigned long unshare_flags)
2944 {
2945 	struct fs_struct *fs, *new_fs = NULL;
2946 	struct files_struct *fd, *new_fd = NULL;
2947 	struct cred *new_cred = NULL;
2948 	struct nsproxy *new_nsproxy = NULL;
2949 	int do_sysvsem = 0;
2950 	int err;
2951 
2952 	/*
2953 	 * If unsharing a user namespace must also unshare the thread group
2954 	 * and unshare the filesystem root and working directories.
2955 	 */
2956 	if (unshare_flags & CLONE_NEWUSER)
2957 		unshare_flags |= CLONE_THREAD | CLONE_FS;
2958 	/*
2959 	 * If unsharing vm, must also unshare signal handlers.
2960 	 */
2961 	if (unshare_flags & CLONE_VM)
2962 		unshare_flags |= CLONE_SIGHAND;
2963 	/*
2964 	 * If unsharing a signal handlers, must also unshare the signal queues.
2965 	 */
2966 	if (unshare_flags & CLONE_SIGHAND)
2967 		unshare_flags |= CLONE_THREAD;
2968 	/*
2969 	 * If unsharing namespace, must also unshare filesystem information.
2970 	 */
2971 	if (unshare_flags & CLONE_NEWNS)
2972 		unshare_flags |= CLONE_FS;
2973 
2974 	err = check_unshare_flags(unshare_flags);
2975 	if (err)
2976 		goto bad_unshare_out;
2977 	/*
2978 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2979 	 * to a new ipc namespace, the semaphore arrays from the old
2980 	 * namespace are unreachable.
2981 	 */
2982 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2983 		do_sysvsem = 1;
2984 	err = unshare_fs(unshare_flags, &new_fs);
2985 	if (err)
2986 		goto bad_unshare_out;
2987 	err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2988 	if (err)
2989 		goto bad_unshare_cleanup_fs;
2990 	err = unshare_userns(unshare_flags, &new_cred);
2991 	if (err)
2992 		goto bad_unshare_cleanup_fd;
2993 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2994 					 new_cred, new_fs);
2995 	if (err)
2996 		goto bad_unshare_cleanup_cred;
2997 
2998 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2999 		if (do_sysvsem) {
3000 			/*
3001 			 * CLONE_SYSVSEM is equivalent to sys_exit().
3002 			 */
3003 			exit_sem(current);
3004 		}
3005 		if (unshare_flags & CLONE_NEWIPC) {
3006 			/* Orphan segments in old ns (see sem above). */
3007 			exit_shm(current);
3008 			shm_init_task(current);
3009 		}
3010 
3011 		if (new_nsproxy)
3012 			switch_task_namespaces(current, new_nsproxy);
3013 
3014 		task_lock(current);
3015 
3016 		if (new_fs) {
3017 			fs = current->fs;
3018 			spin_lock(&fs->lock);
3019 			current->fs = new_fs;
3020 			if (--fs->users)
3021 				new_fs = NULL;
3022 			else
3023 				new_fs = fs;
3024 			spin_unlock(&fs->lock);
3025 		}
3026 
3027 		if (new_fd) {
3028 			fd = current->files;
3029 			current->files = new_fd;
3030 			new_fd = fd;
3031 		}
3032 
3033 		task_unlock(current);
3034 
3035 		if (new_cred) {
3036 			/* Install the new user namespace */
3037 			commit_creds(new_cred);
3038 			new_cred = NULL;
3039 		}
3040 	}
3041 
3042 	perf_event_namespaces(current);
3043 
3044 bad_unshare_cleanup_cred:
3045 	if (new_cred)
3046 		put_cred(new_cred);
3047 bad_unshare_cleanup_fd:
3048 	if (new_fd)
3049 		put_files_struct(new_fd);
3050 
3051 bad_unshare_cleanup_fs:
3052 	if (new_fs)
3053 		free_fs_struct(new_fs);
3054 
3055 bad_unshare_out:
3056 	return err;
3057 }
3058 
3059 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
3060 {
3061 	return ksys_unshare(unshare_flags);
3062 }
3063 
3064 /*
3065  *	Helper to unshare the files of the current task.
3066  *	We don't want to expose copy_files internals to
3067  *	the exec layer of the kernel.
3068  */
3069 
3070 int unshare_files(void)
3071 {
3072 	struct task_struct *task = current;
3073 	struct files_struct *old, *copy = NULL;
3074 	int error;
3075 
3076 	error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
3077 	if (error || !copy)
3078 		return error;
3079 
3080 	old = task->files;
3081 	task_lock(task);
3082 	task->files = copy;
3083 	task_unlock(task);
3084 	put_files_struct(old);
3085 	return 0;
3086 }
3087 
3088 int sysctl_max_threads(struct ctl_table *table, int write,
3089 		       void *buffer, size_t *lenp, loff_t *ppos)
3090 {
3091 	struct ctl_table t;
3092 	int ret;
3093 	int threads = max_threads;
3094 	int min = 1;
3095 	int max = MAX_THREADS;
3096 
3097 	t = *table;
3098 	t.data = &threads;
3099 	t.extra1 = &min;
3100 	t.extra2 = &max;
3101 
3102 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
3103 	if (ret || !write)
3104 		return ret;
3105 
3106 	max_threads = threads;
3107 
3108 	return 0;
3109 }
3110