xref: /linux-6.15/kernel/fork.c (revision a2b0fe74)
1 /*
2  *  linux/kernel/fork.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 /*
8  *  'fork.c' contains the help-routines for the 'fork' system call
9  * (see also entry.S and others).
10  * Fork is rather simple, once you get the hang of it, but the memory
11  * management can be a bitch. See 'mm/memory.c': 'copy_page_range()'
12  */
13 
14 #include <linux/slab.h>
15 #include <linux/sched/autogroup.h>
16 #include <linux/sched/mm.h>
17 #include <linux/sched/coredump.h>
18 #include <linux/sched/user.h>
19 #include <linux/sched/numa_balancing.h>
20 #include <linux/sched/stat.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/task_stack.h>
23 #include <linux/sched/cputime.h>
24 #include <linux/rtmutex.h>
25 #include <linux/init.h>
26 #include <linux/unistd.h>
27 #include <linux/module.h>
28 #include <linux/vmalloc.h>
29 #include <linux/completion.h>
30 #include <linux/personality.h>
31 #include <linux/mempolicy.h>
32 #include <linux/sem.h>
33 #include <linux/file.h>
34 #include <linux/fdtable.h>
35 #include <linux/iocontext.h>
36 #include <linux/key.h>
37 #include <linux/binfmts.h>
38 #include <linux/mman.h>
39 #include <linux/mmu_notifier.h>
40 #include <linux/hmm.h>
41 #include <linux/fs.h>
42 #include <linux/mm.h>
43 #include <linux/vmacache.h>
44 #include <linux/nsproxy.h>
45 #include <linux/capability.h>
46 #include <linux/cpu.h>
47 #include <linux/cgroup.h>
48 #include <linux/security.h>
49 #include <linux/hugetlb.h>
50 #include <linux/seccomp.h>
51 #include <linux/swap.h>
52 #include <linux/syscalls.h>
53 #include <linux/jiffies.h>
54 #include <linux/futex.h>
55 #include <linux/compat.h>
56 #include <linux/kthread.h>
57 #include <linux/task_io_accounting_ops.h>
58 #include <linux/rcupdate.h>
59 #include <linux/ptrace.h>
60 #include <linux/mount.h>
61 #include <linux/audit.h>
62 #include <linux/memcontrol.h>
63 #include <linux/ftrace.h>
64 #include <linux/proc_fs.h>
65 #include <linux/profile.h>
66 #include <linux/rmap.h>
67 #include <linux/ksm.h>
68 #include <linux/acct.h>
69 #include <linux/userfaultfd_k.h>
70 #include <linux/tsacct_kern.h>
71 #include <linux/cn_proc.h>
72 #include <linux/freezer.h>
73 #include <linux/delayacct.h>
74 #include <linux/taskstats_kern.h>
75 #include <linux/random.h>
76 #include <linux/tty.h>
77 #include <linux/blkdev.h>
78 #include <linux/fs_struct.h>
79 #include <linux/magic.h>
80 #include <linux/sched/mm.h>
81 #include <linux/perf_event.h>
82 #include <linux/posix-timers.h>
83 #include <linux/user-return-notifier.h>
84 #include <linux/oom.h>
85 #include <linux/khugepaged.h>
86 #include <linux/signalfd.h>
87 #include <linux/uprobes.h>
88 #include <linux/aio.h>
89 #include <linux/compiler.h>
90 #include <linux/sysctl.h>
91 #include <linux/kcov.h>
92 #include <linux/livepatch.h>
93 #include <linux/thread_info.h>
94 
95 #include <asm/pgtable.h>
96 #include <asm/pgalloc.h>
97 #include <linux/uaccess.h>
98 #include <asm/mmu_context.h>
99 #include <asm/cacheflush.h>
100 #include <asm/tlbflush.h>
101 
102 #include <trace/events/sched.h>
103 
104 #define CREATE_TRACE_POINTS
105 #include <trace/events/task.h>
106 
107 /*
108  * Minimum number of threads to boot the kernel
109  */
110 #define MIN_THREADS 20
111 
112 /*
113  * Maximum number of threads
114  */
115 #define MAX_THREADS FUTEX_TID_MASK
116 
117 /*
118  * Protected counters by write_lock_irq(&tasklist_lock)
119  */
120 unsigned long total_forks;	/* Handle normal Linux uptimes. */
121 int nr_threads;			/* The idle threads do not count.. */
122 
123 int max_threads;		/* tunable limit on nr_threads */
124 
125 DEFINE_PER_CPU(unsigned long, process_counts) = 0;
126 
127 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
128 
129 #ifdef CONFIG_PROVE_RCU
130 int lockdep_tasklist_lock_is_held(void)
131 {
132 	return lockdep_is_held(&tasklist_lock);
133 }
134 EXPORT_SYMBOL_GPL(lockdep_tasklist_lock_is_held);
135 #endif /* #ifdef CONFIG_PROVE_RCU */
136 
137 int nr_processes(void)
138 {
139 	int cpu;
140 	int total = 0;
141 
142 	for_each_possible_cpu(cpu)
143 		total += per_cpu(process_counts, cpu);
144 
145 	return total;
146 }
147 
148 void __weak arch_release_task_struct(struct task_struct *tsk)
149 {
150 }
151 
152 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
153 static struct kmem_cache *task_struct_cachep;
154 
155 static inline struct task_struct *alloc_task_struct_node(int node)
156 {
157 	return kmem_cache_alloc_node(task_struct_cachep, GFP_KERNEL, node);
158 }
159 
160 static inline void free_task_struct(struct task_struct *tsk)
161 {
162 	kmem_cache_free(task_struct_cachep, tsk);
163 }
164 #endif
165 
166 void __weak arch_release_thread_stack(unsigned long *stack)
167 {
168 }
169 
170 #ifndef CONFIG_ARCH_THREAD_STACK_ALLOCATOR
171 
172 /*
173  * Allocate pages if THREAD_SIZE is >= PAGE_SIZE, otherwise use a
174  * kmemcache based allocator.
175  */
176 # if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
177 
178 #ifdef CONFIG_VMAP_STACK
179 /*
180  * vmalloc() is a bit slow, and calling vfree() enough times will force a TLB
181  * flush.  Try to minimize the number of calls by caching stacks.
182  */
183 #define NR_CACHED_STACKS 2
184 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
185 
186 static int free_vm_stack_cache(unsigned int cpu)
187 {
188 	struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
189 	int i;
190 
191 	for (i = 0; i < NR_CACHED_STACKS; i++) {
192 		struct vm_struct *vm_stack = cached_vm_stacks[i];
193 
194 		if (!vm_stack)
195 			continue;
196 
197 		vfree(vm_stack->addr);
198 		cached_vm_stacks[i] = NULL;
199 	}
200 
201 	return 0;
202 }
203 #endif
204 
205 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
206 {
207 #ifdef CONFIG_VMAP_STACK
208 	void *stack;
209 	int i;
210 
211 	for (i = 0; i < NR_CACHED_STACKS; i++) {
212 		struct vm_struct *s;
213 
214 		s = this_cpu_xchg(cached_stacks[i], NULL);
215 
216 		if (!s)
217 			continue;
218 
219 #ifdef CONFIG_DEBUG_KMEMLEAK
220 		/* Clear stale pointers from reused stack. */
221 		memset(s->addr, 0, THREAD_SIZE);
222 #endif
223 		tsk->stack_vm_area = s;
224 		return s->addr;
225 	}
226 
227 	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
228 				     VMALLOC_START, VMALLOC_END,
229 				     THREADINFO_GFP,
230 				     PAGE_KERNEL,
231 				     0, node, __builtin_return_address(0));
232 
233 	/*
234 	 * We can't call find_vm_area() in interrupt context, and
235 	 * free_thread_stack() can be called in interrupt context,
236 	 * so cache the vm_struct.
237 	 */
238 	if (stack)
239 		tsk->stack_vm_area = find_vm_area(stack);
240 	return stack;
241 #else
242 	struct page *page = alloc_pages_node(node, THREADINFO_GFP,
243 					     THREAD_SIZE_ORDER);
244 
245 	return page ? page_address(page) : NULL;
246 #endif
247 }
248 
249 static inline void free_thread_stack(struct task_struct *tsk)
250 {
251 #ifdef CONFIG_VMAP_STACK
252 	if (task_stack_vm_area(tsk)) {
253 		int i;
254 
255 		for (i = 0; i < NR_CACHED_STACKS; i++) {
256 			if (this_cpu_cmpxchg(cached_stacks[i],
257 					NULL, tsk->stack_vm_area) != NULL)
258 				continue;
259 
260 			return;
261 		}
262 
263 		vfree_atomic(tsk->stack);
264 		return;
265 	}
266 #endif
267 
268 	__free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
269 }
270 # else
271 static struct kmem_cache *thread_stack_cache;
272 
273 static unsigned long *alloc_thread_stack_node(struct task_struct *tsk,
274 						  int node)
275 {
276 	return kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
277 }
278 
279 static void free_thread_stack(struct task_struct *tsk)
280 {
281 	kmem_cache_free(thread_stack_cache, tsk->stack);
282 }
283 
284 void thread_stack_cache_init(void)
285 {
286 	thread_stack_cache = kmem_cache_create("thread_stack", THREAD_SIZE,
287 					      THREAD_SIZE, 0, NULL);
288 	BUG_ON(thread_stack_cache == NULL);
289 }
290 # endif
291 #endif
292 
293 /* SLAB cache for signal_struct structures (tsk->signal) */
294 static struct kmem_cache *signal_cachep;
295 
296 /* SLAB cache for sighand_struct structures (tsk->sighand) */
297 struct kmem_cache *sighand_cachep;
298 
299 /* SLAB cache for files_struct structures (tsk->files) */
300 struct kmem_cache *files_cachep;
301 
302 /* SLAB cache for fs_struct structures (tsk->fs) */
303 struct kmem_cache *fs_cachep;
304 
305 /* SLAB cache for vm_area_struct structures */
306 struct kmem_cache *vm_area_cachep;
307 
308 /* SLAB cache for mm_struct structures (tsk->mm) */
309 static struct kmem_cache *mm_cachep;
310 
311 static void account_kernel_stack(struct task_struct *tsk, int account)
312 {
313 	void *stack = task_stack_page(tsk);
314 	struct vm_struct *vm = task_stack_vm_area(tsk);
315 
316 	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
317 
318 	if (vm) {
319 		int i;
320 
321 		BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
322 
323 		for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
324 			mod_zone_page_state(page_zone(vm->pages[i]),
325 					    NR_KERNEL_STACK_KB,
326 					    PAGE_SIZE / 1024 * account);
327 		}
328 
329 		/* All stack pages belong to the same memcg. */
330 		mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
331 				     account * (THREAD_SIZE / 1024));
332 	} else {
333 		/*
334 		 * All stack pages are in the same zone and belong to the
335 		 * same memcg.
336 		 */
337 		struct page *first_page = virt_to_page(stack);
338 
339 		mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
340 				    THREAD_SIZE / 1024 * account);
341 
342 		mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
343 				     account * (THREAD_SIZE / 1024));
344 	}
345 }
346 
347 static void release_task_stack(struct task_struct *tsk)
348 {
349 	if (WARN_ON(tsk->state != TASK_DEAD))
350 		return;  /* Better to leak the stack than to free prematurely */
351 
352 	account_kernel_stack(tsk, -1);
353 	arch_release_thread_stack(tsk->stack);
354 	free_thread_stack(tsk);
355 	tsk->stack = NULL;
356 #ifdef CONFIG_VMAP_STACK
357 	tsk->stack_vm_area = NULL;
358 #endif
359 }
360 
361 #ifdef CONFIG_THREAD_INFO_IN_TASK
362 void put_task_stack(struct task_struct *tsk)
363 {
364 	if (atomic_dec_and_test(&tsk->stack_refcount))
365 		release_task_stack(tsk);
366 }
367 #endif
368 
369 void free_task(struct task_struct *tsk)
370 {
371 #ifndef CONFIG_THREAD_INFO_IN_TASK
372 	/*
373 	 * The task is finally done with both the stack and thread_info,
374 	 * so free both.
375 	 */
376 	release_task_stack(tsk);
377 #else
378 	/*
379 	 * If the task had a separate stack allocation, it should be gone
380 	 * by now.
381 	 */
382 	WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
383 #endif
384 	rt_mutex_debug_task_free(tsk);
385 	ftrace_graph_exit_task(tsk);
386 	put_seccomp_filter(tsk);
387 	arch_release_task_struct(tsk);
388 	if (tsk->flags & PF_KTHREAD)
389 		free_kthread_struct(tsk);
390 	free_task_struct(tsk);
391 }
392 EXPORT_SYMBOL(free_task);
393 
394 #ifdef CONFIG_MMU
395 static __latent_entropy int dup_mmap(struct mm_struct *mm,
396 					struct mm_struct *oldmm)
397 {
398 	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
399 	struct rb_node **rb_link, *rb_parent;
400 	int retval;
401 	unsigned long charge;
402 	LIST_HEAD(uf);
403 
404 	uprobe_start_dup_mmap();
405 	if (down_write_killable(&oldmm->mmap_sem)) {
406 		retval = -EINTR;
407 		goto fail_uprobe_end;
408 	}
409 	flush_cache_dup_mm(oldmm);
410 	uprobe_dup_mmap(oldmm, mm);
411 	/*
412 	 * Not linked in yet - no deadlock potential:
413 	 */
414 	down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
415 
416 	/* No ordering required: file already has been exposed. */
417 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
418 
419 	mm->total_vm = oldmm->total_vm;
420 	mm->data_vm = oldmm->data_vm;
421 	mm->exec_vm = oldmm->exec_vm;
422 	mm->stack_vm = oldmm->stack_vm;
423 
424 	rb_link = &mm->mm_rb.rb_node;
425 	rb_parent = NULL;
426 	pprev = &mm->mmap;
427 	retval = ksm_fork(mm, oldmm);
428 	if (retval)
429 		goto out;
430 	retval = khugepaged_fork(mm, oldmm);
431 	if (retval)
432 		goto out;
433 
434 	prev = NULL;
435 	for (mpnt = oldmm->mmap; mpnt; mpnt = mpnt->vm_next) {
436 		struct file *file;
437 
438 		if (mpnt->vm_flags & VM_DONTCOPY) {
439 			vm_stat_account(mm, mpnt->vm_flags, -vma_pages(mpnt));
440 			continue;
441 		}
442 		charge = 0;
443 		if (mpnt->vm_flags & VM_ACCOUNT) {
444 			unsigned long len = vma_pages(mpnt);
445 
446 			if (security_vm_enough_memory_mm(oldmm, len)) /* sic */
447 				goto fail_nomem;
448 			charge = len;
449 		}
450 		tmp = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
451 		if (!tmp)
452 			goto fail_nomem;
453 		*tmp = *mpnt;
454 		INIT_LIST_HEAD(&tmp->anon_vma_chain);
455 		retval = vma_dup_policy(mpnt, tmp);
456 		if (retval)
457 			goto fail_nomem_policy;
458 		tmp->vm_mm = mm;
459 		retval = dup_userfaultfd(tmp, &uf);
460 		if (retval)
461 			goto fail_nomem_anon_vma_fork;
462 		if (tmp->vm_flags & VM_WIPEONFORK) {
463 			/* VM_WIPEONFORK gets a clean slate in the child. */
464 			tmp->anon_vma = NULL;
465 			if (anon_vma_prepare(tmp))
466 				goto fail_nomem_anon_vma_fork;
467 		} else if (anon_vma_fork(tmp, mpnt))
468 			goto fail_nomem_anon_vma_fork;
469 		tmp->vm_flags &= ~(VM_LOCKED | VM_LOCKONFAULT);
470 		tmp->vm_next = tmp->vm_prev = NULL;
471 		file = tmp->vm_file;
472 		if (file) {
473 			struct inode *inode = file_inode(file);
474 			struct address_space *mapping = file->f_mapping;
475 
476 			get_file(file);
477 			if (tmp->vm_flags & VM_DENYWRITE)
478 				atomic_dec(&inode->i_writecount);
479 			i_mmap_lock_write(mapping);
480 			if (tmp->vm_flags & VM_SHARED)
481 				atomic_inc(&mapping->i_mmap_writable);
482 			flush_dcache_mmap_lock(mapping);
483 			/* insert tmp into the share list, just after mpnt */
484 			vma_interval_tree_insert_after(tmp, mpnt,
485 					&mapping->i_mmap);
486 			flush_dcache_mmap_unlock(mapping);
487 			i_mmap_unlock_write(mapping);
488 		}
489 
490 		/*
491 		 * Clear hugetlb-related page reserves for children. This only
492 		 * affects MAP_PRIVATE mappings. Faults generated by the child
493 		 * are not guaranteed to succeed, even if read-only
494 		 */
495 		if (is_vm_hugetlb_page(tmp))
496 			reset_vma_resv_huge_pages(tmp);
497 
498 		/*
499 		 * Link in the new vma and copy the page table entries.
500 		 */
501 		*pprev = tmp;
502 		pprev = &tmp->vm_next;
503 		tmp->vm_prev = prev;
504 		prev = tmp;
505 
506 		__vma_link_rb(mm, tmp, rb_link, rb_parent);
507 		rb_link = &tmp->vm_rb.rb_right;
508 		rb_parent = &tmp->vm_rb;
509 
510 		mm->map_count++;
511 		if (!(tmp->vm_flags & VM_WIPEONFORK))
512 			retval = copy_page_range(mm, oldmm, mpnt);
513 
514 		if (tmp->vm_ops && tmp->vm_ops->open)
515 			tmp->vm_ops->open(tmp);
516 
517 		if (retval)
518 			goto out;
519 	}
520 	/* a new mm has just been created */
521 	arch_dup_mmap(oldmm, mm);
522 	retval = 0;
523 out:
524 	up_write(&mm->mmap_sem);
525 	flush_tlb_mm(oldmm);
526 	up_write(&oldmm->mmap_sem);
527 	dup_userfaultfd_complete(&uf);
528 fail_uprobe_end:
529 	uprobe_end_dup_mmap();
530 	return retval;
531 fail_nomem_anon_vma_fork:
532 	mpol_put(vma_policy(tmp));
533 fail_nomem_policy:
534 	kmem_cache_free(vm_area_cachep, tmp);
535 fail_nomem:
536 	retval = -ENOMEM;
537 	vm_unacct_memory(charge);
538 	goto out;
539 }
540 
541 static inline int mm_alloc_pgd(struct mm_struct *mm)
542 {
543 	mm->pgd = pgd_alloc(mm);
544 	if (unlikely(!mm->pgd))
545 		return -ENOMEM;
546 	return 0;
547 }
548 
549 static inline void mm_free_pgd(struct mm_struct *mm)
550 {
551 	pgd_free(mm, mm->pgd);
552 }
553 #else
554 static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
555 {
556 	down_write(&oldmm->mmap_sem);
557 	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
558 	up_write(&oldmm->mmap_sem);
559 	return 0;
560 }
561 #define mm_alloc_pgd(mm)	(0)
562 #define mm_free_pgd(mm)
563 #endif /* CONFIG_MMU */
564 
565 static void check_mm(struct mm_struct *mm)
566 {
567 	int i;
568 
569 	for (i = 0; i < NR_MM_COUNTERS; i++) {
570 		long x = atomic_long_read(&mm->rss_stat.count[i]);
571 
572 		if (unlikely(x))
573 			printk(KERN_ALERT "BUG: Bad rss-counter state "
574 					  "mm:%p idx:%d val:%ld\n", mm, i, x);
575 	}
576 
577 	if (mm_pgtables_bytes(mm))
578 		pr_alert("BUG: non-zero pgtables_bytes on freeing mm: %ld\n",
579 				mm_pgtables_bytes(mm));
580 
581 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
582 	VM_BUG_ON_MM(mm->pmd_huge_pte, mm);
583 #endif
584 }
585 
586 #define allocate_mm()	(kmem_cache_alloc(mm_cachep, GFP_KERNEL))
587 #define free_mm(mm)	(kmem_cache_free(mm_cachep, (mm)))
588 
589 /*
590  * Called when the last reference to the mm
591  * is dropped: either by a lazy thread or by
592  * mmput. Free the page directory and the mm.
593  */
594 static void __mmdrop(struct mm_struct *mm)
595 {
596 	BUG_ON(mm == &init_mm);
597 	mm_free_pgd(mm);
598 	destroy_context(mm);
599 	hmm_mm_destroy(mm);
600 	mmu_notifier_mm_destroy(mm);
601 	check_mm(mm);
602 	put_user_ns(mm->user_ns);
603 	free_mm(mm);
604 }
605 
606 void mmdrop(struct mm_struct *mm)
607 {
608 	if (unlikely(atomic_dec_and_test(&mm->mm_count)))
609 		__mmdrop(mm);
610 }
611 EXPORT_SYMBOL_GPL(mmdrop);
612 
613 static void mmdrop_async_fn(struct work_struct *work)
614 {
615 	struct mm_struct *mm;
616 
617 	mm = container_of(work, struct mm_struct, async_put_work);
618 	__mmdrop(mm);
619 }
620 
621 static void mmdrop_async(struct mm_struct *mm)
622 {
623 	if (unlikely(atomic_dec_and_test(&mm->mm_count))) {
624 		INIT_WORK(&mm->async_put_work, mmdrop_async_fn);
625 		schedule_work(&mm->async_put_work);
626 	}
627 }
628 
629 static inline void free_signal_struct(struct signal_struct *sig)
630 {
631 	taskstats_tgid_free(sig);
632 	sched_autogroup_exit(sig);
633 	/*
634 	 * __mmdrop is not safe to call from softirq context on x86 due to
635 	 * pgd_dtor so postpone it to the async context
636 	 */
637 	if (sig->oom_mm)
638 		mmdrop_async(sig->oom_mm);
639 	kmem_cache_free(signal_cachep, sig);
640 }
641 
642 static inline void put_signal_struct(struct signal_struct *sig)
643 {
644 	if (atomic_dec_and_test(&sig->sigcnt))
645 		free_signal_struct(sig);
646 }
647 
648 void __put_task_struct(struct task_struct *tsk)
649 {
650 	WARN_ON(!tsk->exit_state);
651 	WARN_ON(atomic_read(&tsk->usage));
652 	WARN_ON(tsk == current);
653 
654 	cgroup_free(tsk);
655 	task_numa_free(tsk);
656 	security_task_free(tsk);
657 	exit_creds(tsk);
658 	delayacct_tsk_free(tsk);
659 	put_signal_struct(tsk->signal);
660 
661 	if (!profile_handoff_task(tsk))
662 		free_task(tsk);
663 }
664 EXPORT_SYMBOL_GPL(__put_task_struct);
665 
666 void __init __weak arch_task_cache_init(void) { }
667 
668 /*
669  * set_max_threads
670  */
671 static void set_max_threads(unsigned int max_threads_suggested)
672 {
673 	u64 threads;
674 
675 	/*
676 	 * The number of threads shall be limited such that the thread
677 	 * structures may only consume a small part of the available memory.
678 	 */
679 	if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
680 		threads = MAX_THREADS;
681 	else
682 		threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
683 				    (u64) THREAD_SIZE * 8UL);
684 
685 	if (threads > max_threads_suggested)
686 		threads = max_threads_suggested;
687 
688 	max_threads = clamp_t(u64, threads, MIN_THREADS, MAX_THREADS);
689 }
690 
691 #ifdef CONFIG_ARCH_WANTS_DYNAMIC_TASK_STRUCT
692 /* Initialized by the architecture: */
693 int arch_task_struct_size __read_mostly;
694 #endif
695 
696 void __init fork_init(void)
697 {
698 	int i;
699 #ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
700 #ifndef ARCH_MIN_TASKALIGN
701 #define ARCH_MIN_TASKALIGN	0
702 #endif
703 	int align = max_t(int, L1_CACHE_BYTES, ARCH_MIN_TASKALIGN);
704 
705 	/* create a slab on which task_structs can be allocated */
706 	task_struct_cachep = kmem_cache_create("task_struct",
707 			arch_task_struct_size, align,
708 			SLAB_PANIC|SLAB_ACCOUNT, NULL);
709 #endif
710 
711 	/* do the arch specific task caches init */
712 	arch_task_cache_init();
713 
714 	set_max_threads(MAX_THREADS);
715 
716 	init_task.signal->rlim[RLIMIT_NPROC].rlim_cur = max_threads/2;
717 	init_task.signal->rlim[RLIMIT_NPROC].rlim_max = max_threads/2;
718 	init_task.signal->rlim[RLIMIT_SIGPENDING] =
719 		init_task.signal->rlim[RLIMIT_NPROC];
720 
721 	for (i = 0; i < UCOUNT_COUNTS; i++) {
722 		init_user_ns.ucount_max[i] = max_threads/2;
723 	}
724 
725 #ifdef CONFIG_VMAP_STACK
726 	cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "fork:vm_stack_cache",
727 			  NULL, free_vm_stack_cache);
728 #endif
729 
730 	lockdep_init_task(&init_task);
731 }
732 
733 int __weak arch_dup_task_struct(struct task_struct *dst,
734 					       struct task_struct *src)
735 {
736 	*dst = *src;
737 	return 0;
738 }
739 
740 void set_task_stack_end_magic(struct task_struct *tsk)
741 {
742 	unsigned long *stackend;
743 
744 	stackend = end_of_stack(tsk);
745 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
746 }
747 
748 static struct task_struct *dup_task_struct(struct task_struct *orig, int node)
749 {
750 	struct task_struct *tsk;
751 	unsigned long *stack;
752 	struct vm_struct *stack_vm_area;
753 	int err;
754 
755 	if (node == NUMA_NO_NODE)
756 		node = tsk_fork_get_node(orig);
757 	tsk = alloc_task_struct_node(node);
758 	if (!tsk)
759 		return NULL;
760 
761 	stack = alloc_thread_stack_node(tsk, node);
762 	if (!stack)
763 		goto free_tsk;
764 
765 	stack_vm_area = task_stack_vm_area(tsk);
766 
767 	err = arch_dup_task_struct(tsk, orig);
768 
769 	/*
770 	 * arch_dup_task_struct() clobbers the stack-related fields.  Make
771 	 * sure they're properly initialized before using any stack-related
772 	 * functions again.
773 	 */
774 	tsk->stack = stack;
775 #ifdef CONFIG_VMAP_STACK
776 	tsk->stack_vm_area = stack_vm_area;
777 #endif
778 #ifdef CONFIG_THREAD_INFO_IN_TASK
779 	atomic_set(&tsk->stack_refcount, 1);
780 #endif
781 
782 	if (err)
783 		goto free_stack;
784 
785 #ifdef CONFIG_SECCOMP
786 	/*
787 	 * We must handle setting up seccomp filters once we're under
788 	 * the sighand lock in case orig has changed between now and
789 	 * then. Until then, filter must be NULL to avoid messing up
790 	 * the usage counts on the error path calling free_task.
791 	 */
792 	tsk->seccomp.filter = NULL;
793 #endif
794 
795 	setup_thread_stack(tsk, orig);
796 	clear_user_return_notifier(tsk);
797 	clear_tsk_need_resched(tsk);
798 	set_task_stack_end_magic(tsk);
799 
800 #ifdef CONFIG_CC_STACKPROTECTOR
801 	tsk->stack_canary = get_random_canary();
802 #endif
803 
804 	/*
805 	 * One for us, one for whoever does the "release_task()" (usually
806 	 * parent)
807 	 */
808 	atomic_set(&tsk->usage, 2);
809 #ifdef CONFIG_BLK_DEV_IO_TRACE
810 	tsk->btrace_seq = 0;
811 #endif
812 	tsk->splice_pipe = NULL;
813 	tsk->task_frag.page = NULL;
814 	tsk->wake_q.next = NULL;
815 
816 	account_kernel_stack(tsk, 1);
817 
818 	kcov_task_init(tsk);
819 
820 #ifdef CONFIG_FAULT_INJECTION
821 	tsk->fail_nth = 0;
822 #endif
823 
824 	return tsk;
825 
826 free_stack:
827 	free_thread_stack(tsk);
828 free_tsk:
829 	free_task_struct(tsk);
830 	return NULL;
831 }
832 
833 __cacheline_aligned_in_smp DEFINE_SPINLOCK(mmlist_lock);
834 
835 static unsigned long default_dump_filter = MMF_DUMP_FILTER_DEFAULT;
836 
837 static int __init coredump_filter_setup(char *s)
838 {
839 	default_dump_filter =
840 		(simple_strtoul(s, NULL, 0) << MMF_DUMP_FILTER_SHIFT) &
841 		MMF_DUMP_FILTER_MASK;
842 	return 1;
843 }
844 
845 __setup("coredump_filter=", coredump_filter_setup);
846 
847 #include <linux/init_task.h>
848 
849 static void mm_init_aio(struct mm_struct *mm)
850 {
851 #ifdef CONFIG_AIO
852 	spin_lock_init(&mm->ioctx_lock);
853 	mm->ioctx_table = NULL;
854 #endif
855 }
856 
857 static void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
858 {
859 #ifdef CONFIG_MEMCG
860 	mm->owner = p;
861 #endif
862 }
863 
864 static void mm_init_uprobes_state(struct mm_struct *mm)
865 {
866 #ifdef CONFIG_UPROBES
867 	mm->uprobes_state.xol_area = NULL;
868 #endif
869 }
870 
871 static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
872 	struct user_namespace *user_ns)
873 {
874 	mm->mmap = NULL;
875 	mm->mm_rb = RB_ROOT;
876 	mm->vmacache_seqnum = 0;
877 	atomic_set(&mm->mm_users, 1);
878 	atomic_set(&mm->mm_count, 1);
879 	init_rwsem(&mm->mmap_sem);
880 	INIT_LIST_HEAD(&mm->mmlist);
881 	mm->core_state = NULL;
882 	mm_pgtables_bytes_init(mm);
883 	mm->map_count = 0;
884 	mm->locked_vm = 0;
885 	mm->pinned_vm = 0;
886 	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
887 	spin_lock_init(&mm->page_table_lock);
888 	mm_init_cpumask(mm);
889 	mm_init_aio(mm);
890 	mm_init_owner(mm, p);
891 	RCU_INIT_POINTER(mm->exe_file, NULL);
892 	mmu_notifier_mm_init(mm);
893 	hmm_mm_init(mm);
894 	init_tlb_flush_pending(mm);
895 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
896 	mm->pmd_huge_pte = NULL;
897 #endif
898 	mm_init_uprobes_state(mm);
899 
900 	if (current->mm) {
901 		mm->flags = current->mm->flags & MMF_INIT_MASK;
902 		mm->def_flags = current->mm->def_flags & VM_INIT_DEF_MASK;
903 	} else {
904 		mm->flags = default_dump_filter;
905 		mm->def_flags = 0;
906 	}
907 
908 	if (mm_alloc_pgd(mm))
909 		goto fail_nopgd;
910 
911 	if (init_new_context(p, mm))
912 		goto fail_nocontext;
913 
914 	mm->user_ns = get_user_ns(user_ns);
915 	return mm;
916 
917 fail_nocontext:
918 	mm_free_pgd(mm);
919 fail_nopgd:
920 	free_mm(mm);
921 	return NULL;
922 }
923 
924 /*
925  * Allocate and initialize an mm_struct.
926  */
927 struct mm_struct *mm_alloc(void)
928 {
929 	struct mm_struct *mm;
930 
931 	mm = allocate_mm();
932 	if (!mm)
933 		return NULL;
934 
935 	memset(mm, 0, sizeof(*mm));
936 	return mm_init(mm, current, current_user_ns());
937 }
938 
939 static inline void __mmput(struct mm_struct *mm)
940 {
941 	VM_BUG_ON(atomic_read(&mm->mm_users));
942 
943 	uprobe_clear_state(mm);
944 	exit_aio(mm);
945 	ksm_exit(mm);
946 	khugepaged_exit(mm); /* must run before exit_mmap */
947 	exit_mmap(mm);
948 	mm_put_huge_zero_page(mm);
949 	set_mm_exe_file(mm, NULL);
950 	if (!list_empty(&mm->mmlist)) {
951 		spin_lock(&mmlist_lock);
952 		list_del(&mm->mmlist);
953 		spin_unlock(&mmlist_lock);
954 	}
955 	if (mm->binfmt)
956 		module_put(mm->binfmt->module);
957 	mmdrop(mm);
958 }
959 
960 /*
961  * Decrement the use count and release all resources for an mm.
962  */
963 void mmput(struct mm_struct *mm)
964 {
965 	might_sleep();
966 
967 	if (atomic_dec_and_test(&mm->mm_users))
968 		__mmput(mm);
969 }
970 EXPORT_SYMBOL_GPL(mmput);
971 
972 #ifdef CONFIG_MMU
973 static void mmput_async_fn(struct work_struct *work)
974 {
975 	struct mm_struct *mm = container_of(work, struct mm_struct,
976 					    async_put_work);
977 
978 	__mmput(mm);
979 }
980 
981 void mmput_async(struct mm_struct *mm)
982 {
983 	if (atomic_dec_and_test(&mm->mm_users)) {
984 		INIT_WORK(&mm->async_put_work, mmput_async_fn);
985 		schedule_work(&mm->async_put_work);
986 	}
987 }
988 #endif
989 
990 /**
991  * set_mm_exe_file - change a reference to the mm's executable file
992  *
993  * This changes mm's executable file (shown as symlink /proc/[pid]/exe).
994  *
995  * Main users are mmput() and sys_execve(). Callers prevent concurrent
996  * invocations: in mmput() nobody alive left, in execve task is single
997  * threaded. sys_prctl(PR_SET_MM_MAP/EXE_FILE) also needs to set the
998  * mm->exe_file, but does so without using set_mm_exe_file() in order
999  * to do avoid the need for any locks.
1000  */
1001 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1002 {
1003 	struct file *old_exe_file;
1004 
1005 	/*
1006 	 * It is safe to dereference the exe_file without RCU as
1007 	 * this function is only called if nobody else can access
1008 	 * this mm -- see comment above for justification.
1009 	 */
1010 	old_exe_file = rcu_dereference_raw(mm->exe_file);
1011 
1012 	if (new_exe_file)
1013 		get_file(new_exe_file);
1014 	rcu_assign_pointer(mm->exe_file, new_exe_file);
1015 	if (old_exe_file)
1016 		fput(old_exe_file);
1017 }
1018 
1019 /**
1020  * get_mm_exe_file - acquire a reference to the mm's executable file
1021  *
1022  * Returns %NULL if mm has no associated executable file.
1023  * User must release file via fput().
1024  */
1025 struct file *get_mm_exe_file(struct mm_struct *mm)
1026 {
1027 	struct file *exe_file;
1028 
1029 	rcu_read_lock();
1030 	exe_file = rcu_dereference(mm->exe_file);
1031 	if (exe_file && !get_file_rcu(exe_file))
1032 		exe_file = NULL;
1033 	rcu_read_unlock();
1034 	return exe_file;
1035 }
1036 EXPORT_SYMBOL(get_mm_exe_file);
1037 
1038 /**
1039  * get_task_exe_file - acquire a reference to the task's executable file
1040  *
1041  * Returns %NULL if task's mm (if any) has no associated executable file or
1042  * this is a kernel thread with borrowed mm (see the comment above get_task_mm).
1043  * User must release file via fput().
1044  */
1045 struct file *get_task_exe_file(struct task_struct *task)
1046 {
1047 	struct file *exe_file = NULL;
1048 	struct mm_struct *mm;
1049 
1050 	task_lock(task);
1051 	mm = task->mm;
1052 	if (mm) {
1053 		if (!(task->flags & PF_KTHREAD))
1054 			exe_file = get_mm_exe_file(mm);
1055 	}
1056 	task_unlock(task);
1057 	return exe_file;
1058 }
1059 EXPORT_SYMBOL(get_task_exe_file);
1060 
1061 /**
1062  * get_task_mm - acquire a reference to the task's mm
1063  *
1064  * Returns %NULL if the task has no mm.  Checks PF_KTHREAD (meaning
1065  * this kernel workthread has transiently adopted a user mm with use_mm,
1066  * to do its AIO) is not set and if so returns a reference to it, after
1067  * bumping up the use count.  User must release the mm via mmput()
1068  * after use.  Typically used by /proc and ptrace.
1069  */
1070 struct mm_struct *get_task_mm(struct task_struct *task)
1071 {
1072 	struct mm_struct *mm;
1073 
1074 	task_lock(task);
1075 	mm = task->mm;
1076 	if (mm) {
1077 		if (task->flags & PF_KTHREAD)
1078 			mm = NULL;
1079 		else
1080 			mmget(mm);
1081 	}
1082 	task_unlock(task);
1083 	return mm;
1084 }
1085 EXPORT_SYMBOL_GPL(get_task_mm);
1086 
1087 struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
1088 {
1089 	struct mm_struct *mm;
1090 	int err;
1091 
1092 	err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
1093 	if (err)
1094 		return ERR_PTR(err);
1095 
1096 	mm = get_task_mm(task);
1097 	if (mm && mm != current->mm &&
1098 			!ptrace_may_access(task, mode)) {
1099 		mmput(mm);
1100 		mm = ERR_PTR(-EACCES);
1101 	}
1102 	mutex_unlock(&task->signal->cred_guard_mutex);
1103 
1104 	return mm;
1105 }
1106 
1107 static void complete_vfork_done(struct task_struct *tsk)
1108 {
1109 	struct completion *vfork;
1110 
1111 	task_lock(tsk);
1112 	vfork = tsk->vfork_done;
1113 	if (likely(vfork)) {
1114 		tsk->vfork_done = NULL;
1115 		complete(vfork);
1116 	}
1117 	task_unlock(tsk);
1118 }
1119 
1120 static int wait_for_vfork_done(struct task_struct *child,
1121 				struct completion *vfork)
1122 {
1123 	int killed;
1124 
1125 	freezer_do_not_count();
1126 	killed = wait_for_completion_killable(vfork);
1127 	freezer_count();
1128 
1129 	if (killed) {
1130 		task_lock(child);
1131 		child->vfork_done = NULL;
1132 		task_unlock(child);
1133 	}
1134 
1135 	put_task_struct(child);
1136 	return killed;
1137 }
1138 
1139 /* Please note the differences between mmput and mm_release.
1140  * mmput is called whenever we stop holding onto a mm_struct,
1141  * error success whatever.
1142  *
1143  * mm_release is called after a mm_struct has been removed
1144  * from the current process.
1145  *
1146  * This difference is important for error handling, when we
1147  * only half set up a mm_struct for a new process and need to restore
1148  * the old one.  Because we mmput the new mm_struct before
1149  * restoring the old one. . .
1150  * Eric Biederman 10 January 1998
1151  */
1152 void mm_release(struct task_struct *tsk, struct mm_struct *mm)
1153 {
1154 	/* Get rid of any futexes when releasing the mm */
1155 #ifdef CONFIG_FUTEX
1156 	if (unlikely(tsk->robust_list)) {
1157 		exit_robust_list(tsk);
1158 		tsk->robust_list = NULL;
1159 	}
1160 #ifdef CONFIG_COMPAT
1161 	if (unlikely(tsk->compat_robust_list)) {
1162 		compat_exit_robust_list(tsk);
1163 		tsk->compat_robust_list = NULL;
1164 	}
1165 #endif
1166 	if (unlikely(!list_empty(&tsk->pi_state_list)))
1167 		exit_pi_state_list(tsk);
1168 #endif
1169 
1170 	uprobe_free_utask(tsk);
1171 
1172 	/* Get rid of any cached register state */
1173 	deactivate_mm(tsk, mm);
1174 
1175 	/*
1176 	 * Signal userspace if we're not exiting with a core dump
1177 	 * because we want to leave the value intact for debugging
1178 	 * purposes.
1179 	 */
1180 	if (tsk->clear_child_tid) {
1181 		if (!(tsk->signal->flags & SIGNAL_GROUP_COREDUMP) &&
1182 		    atomic_read(&mm->mm_users) > 1) {
1183 			/*
1184 			 * We don't check the error code - if userspace has
1185 			 * not set up a proper pointer then tough luck.
1186 			 */
1187 			put_user(0, tsk->clear_child_tid);
1188 			sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
1189 					1, NULL, NULL, 0);
1190 		}
1191 		tsk->clear_child_tid = NULL;
1192 	}
1193 
1194 	/*
1195 	 * All done, finally we can wake up parent and return this mm to him.
1196 	 * Also kthread_stop() uses this completion for synchronization.
1197 	 */
1198 	if (tsk->vfork_done)
1199 		complete_vfork_done(tsk);
1200 }
1201 
1202 /*
1203  * Allocate a new mm structure and copy contents from the
1204  * mm structure of the passed in task structure.
1205  */
1206 static struct mm_struct *dup_mm(struct task_struct *tsk)
1207 {
1208 	struct mm_struct *mm, *oldmm = current->mm;
1209 	int err;
1210 
1211 	mm = allocate_mm();
1212 	if (!mm)
1213 		goto fail_nomem;
1214 
1215 	memcpy(mm, oldmm, sizeof(*mm));
1216 
1217 	if (!mm_init(mm, tsk, mm->user_ns))
1218 		goto fail_nomem;
1219 
1220 	err = dup_mmap(mm, oldmm);
1221 	if (err)
1222 		goto free_pt;
1223 
1224 	mm->hiwater_rss = get_mm_rss(mm);
1225 	mm->hiwater_vm = mm->total_vm;
1226 
1227 	if (mm->binfmt && !try_module_get(mm->binfmt->module))
1228 		goto free_pt;
1229 
1230 	return mm;
1231 
1232 free_pt:
1233 	/* don't put binfmt in mmput, we haven't got module yet */
1234 	mm->binfmt = NULL;
1235 	mmput(mm);
1236 
1237 fail_nomem:
1238 	return NULL;
1239 }
1240 
1241 static int copy_mm(unsigned long clone_flags, struct task_struct *tsk)
1242 {
1243 	struct mm_struct *mm, *oldmm;
1244 	int retval;
1245 
1246 	tsk->min_flt = tsk->maj_flt = 0;
1247 	tsk->nvcsw = tsk->nivcsw = 0;
1248 #ifdef CONFIG_DETECT_HUNG_TASK
1249 	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
1250 #endif
1251 
1252 	tsk->mm = NULL;
1253 	tsk->active_mm = NULL;
1254 
1255 	/*
1256 	 * Are we cloning a kernel thread?
1257 	 *
1258 	 * We need to steal a active VM for that..
1259 	 */
1260 	oldmm = current->mm;
1261 	if (!oldmm)
1262 		return 0;
1263 
1264 	/* initialize the new vmacache entries */
1265 	vmacache_flush(tsk);
1266 
1267 	if (clone_flags & CLONE_VM) {
1268 		mmget(oldmm);
1269 		mm = oldmm;
1270 		goto good_mm;
1271 	}
1272 
1273 	retval = -ENOMEM;
1274 	mm = dup_mm(tsk);
1275 	if (!mm)
1276 		goto fail_nomem;
1277 
1278 good_mm:
1279 	tsk->mm = mm;
1280 	tsk->active_mm = mm;
1281 	return 0;
1282 
1283 fail_nomem:
1284 	return retval;
1285 }
1286 
1287 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
1288 {
1289 	struct fs_struct *fs = current->fs;
1290 	if (clone_flags & CLONE_FS) {
1291 		/* tsk->fs is already what we want */
1292 		spin_lock(&fs->lock);
1293 		if (fs->in_exec) {
1294 			spin_unlock(&fs->lock);
1295 			return -EAGAIN;
1296 		}
1297 		fs->users++;
1298 		spin_unlock(&fs->lock);
1299 		return 0;
1300 	}
1301 	tsk->fs = copy_fs_struct(fs);
1302 	if (!tsk->fs)
1303 		return -ENOMEM;
1304 	return 0;
1305 }
1306 
1307 static int copy_files(unsigned long clone_flags, struct task_struct *tsk)
1308 {
1309 	struct files_struct *oldf, *newf;
1310 	int error = 0;
1311 
1312 	/*
1313 	 * A background process may not have any files ...
1314 	 */
1315 	oldf = current->files;
1316 	if (!oldf)
1317 		goto out;
1318 
1319 	if (clone_flags & CLONE_FILES) {
1320 		atomic_inc(&oldf->count);
1321 		goto out;
1322 	}
1323 
1324 	newf = dup_fd(oldf, &error);
1325 	if (!newf)
1326 		goto out;
1327 
1328 	tsk->files = newf;
1329 	error = 0;
1330 out:
1331 	return error;
1332 }
1333 
1334 static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
1335 {
1336 #ifdef CONFIG_BLOCK
1337 	struct io_context *ioc = current->io_context;
1338 	struct io_context *new_ioc;
1339 
1340 	if (!ioc)
1341 		return 0;
1342 	/*
1343 	 * Share io context with parent, if CLONE_IO is set
1344 	 */
1345 	if (clone_flags & CLONE_IO) {
1346 		ioc_task_link(ioc);
1347 		tsk->io_context = ioc;
1348 	} else if (ioprio_valid(ioc->ioprio)) {
1349 		new_ioc = get_task_io_context(tsk, GFP_KERNEL, NUMA_NO_NODE);
1350 		if (unlikely(!new_ioc))
1351 			return -ENOMEM;
1352 
1353 		new_ioc->ioprio = ioc->ioprio;
1354 		put_io_context(new_ioc);
1355 	}
1356 #endif
1357 	return 0;
1358 }
1359 
1360 static int copy_sighand(unsigned long clone_flags, struct task_struct *tsk)
1361 {
1362 	struct sighand_struct *sig;
1363 
1364 	if (clone_flags & CLONE_SIGHAND) {
1365 		atomic_inc(&current->sighand->count);
1366 		return 0;
1367 	}
1368 	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1369 	rcu_assign_pointer(tsk->sighand, sig);
1370 	if (!sig)
1371 		return -ENOMEM;
1372 
1373 	atomic_set(&sig->count, 1);
1374 	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1375 	return 0;
1376 }
1377 
1378 void __cleanup_sighand(struct sighand_struct *sighand)
1379 {
1380 	if (atomic_dec_and_test(&sighand->count)) {
1381 		signalfd_cleanup(sighand);
1382 		/*
1383 		 * sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
1384 		 * without an RCU grace period, see __lock_task_sighand().
1385 		 */
1386 		kmem_cache_free(sighand_cachep, sighand);
1387 	}
1388 }
1389 
1390 #ifdef CONFIG_POSIX_TIMERS
1391 /*
1392  * Initialize POSIX timer handling for a thread group.
1393  */
1394 static void posix_cpu_timers_init_group(struct signal_struct *sig)
1395 {
1396 	unsigned long cpu_limit;
1397 
1398 	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1399 	if (cpu_limit != RLIM_INFINITY) {
1400 		sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1401 		sig->cputimer.running = true;
1402 	}
1403 
1404 	/* The timer lists. */
1405 	INIT_LIST_HEAD(&sig->cpu_timers[0]);
1406 	INIT_LIST_HEAD(&sig->cpu_timers[1]);
1407 	INIT_LIST_HEAD(&sig->cpu_timers[2]);
1408 }
1409 #else
1410 static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1411 #endif
1412 
1413 static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1414 {
1415 	struct signal_struct *sig;
1416 
1417 	if (clone_flags & CLONE_THREAD)
1418 		return 0;
1419 
1420 	sig = kmem_cache_zalloc(signal_cachep, GFP_KERNEL);
1421 	tsk->signal = sig;
1422 	if (!sig)
1423 		return -ENOMEM;
1424 
1425 	sig->nr_threads = 1;
1426 	atomic_set(&sig->live, 1);
1427 	atomic_set(&sig->sigcnt, 1);
1428 
1429 	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1430 	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
1431 	tsk->thread_node = (struct list_head)LIST_HEAD_INIT(sig->thread_head);
1432 
1433 	init_waitqueue_head(&sig->wait_chldexit);
1434 	sig->curr_target = tsk;
1435 	init_sigpending(&sig->shared_pending);
1436 	seqlock_init(&sig->stats_lock);
1437 	prev_cputime_init(&sig->prev_cputime);
1438 
1439 #ifdef CONFIG_POSIX_TIMERS
1440 	INIT_LIST_HEAD(&sig->posix_timers);
1441 	hrtimer_init(&sig->real_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
1442 	sig->real_timer.function = it_real_fn;
1443 #endif
1444 
1445 	task_lock(current->group_leader);
1446 	memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
1447 	task_unlock(current->group_leader);
1448 
1449 	posix_cpu_timers_init_group(sig);
1450 
1451 	tty_audit_fork(sig);
1452 	sched_autogroup_fork(sig);
1453 
1454 	sig->oom_score_adj = current->signal->oom_score_adj;
1455 	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1456 
1457 	mutex_init(&sig->cred_guard_mutex);
1458 
1459 	return 0;
1460 }
1461 
1462 static void copy_seccomp(struct task_struct *p)
1463 {
1464 #ifdef CONFIG_SECCOMP
1465 	/*
1466 	 * Must be called with sighand->lock held, which is common to
1467 	 * all threads in the group. Holding cred_guard_mutex is not
1468 	 * needed because this new task is not yet running and cannot
1469 	 * be racing exec.
1470 	 */
1471 	assert_spin_locked(&current->sighand->siglock);
1472 
1473 	/* Ref-count the new filter user, and assign it. */
1474 	get_seccomp_filter(current);
1475 	p->seccomp = current->seccomp;
1476 
1477 	/*
1478 	 * Explicitly enable no_new_privs here in case it got set
1479 	 * between the task_struct being duplicated and holding the
1480 	 * sighand lock. The seccomp state and nnp must be in sync.
1481 	 */
1482 	if (task_no_new_privs(current))
1483 		task_set_no_new_privs(p);
1484 
1485 	/*
1486 	 * If the parent gained a seccomp mode after copying thread
1487 	 * flags and between before we held the sighand lock, we have
1488 	 * to manually enable the seccomp thread flag here.
1489 	 */
1490 	if (p->seccomp.mode != SECCOMP_MODE_DISABLED)
1491 		set_tsk_thread_flag(p, TIF_SECCOMP);
1492 #endif
1493 }
1494 
1495 SYSCALL_DEFINE1(set_tid_address, int __user *, tidptr)
1496 {
1497 	current->clear_child_tid = tidptr;
1498 
1499 	return task_pid_vnr(current);
1500 }
1501 
1502 static void rt_mutex_init_task(struct task_struct *p)
1503 {
1504 	raw_spin_lock_init(&p->pi_lock);
1505 #ifdef CONFIG_RT_MUTEXES
1506 	p->pi_waiters = RB_ROOT_CACHED;
1507 	p->pi_top_task = NULL;
1508 	p->pi_blocked_on = NULL;
1509 #endif
1510 }
1511 
1512 #ifdef CONFIG_POSIX_TIMERS
1513 /*
1514  * Initialize POSIX timer handling for a single task.
1515  */
1516 static void posix_cpu_timers_init(struct task_struct *tsk)
1517 {
1518 	tsk->cputime_expires.prof_exp = 0;
1519 	tsk->cputime_expires.virt_exp = 0;
1520 	tsk->cputime_expires.sched_exp = 0;
1521 	INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1522 	INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1523 	INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1524 }
1525 #else
1526 static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1527 #endif
1528 
1529 static inline void
1530 init_task_pid(struct task_struct *task, enum pid_type type, struct pid *pid)
1531 {
1532 	 task->pids[type].pid = pid;
1533 }
1534 
1535 static inline void rcu_copy_process(struct task_struct *p)
1536 {
1537 #ifdef CONFIG_PREEMPT_RCU
1538 	p->rcu_read_lock_nesting = 0;
1539 	p->rcu_read_unlock_special.s = 0;
1540 	p->rcu_blocked_node = NULL;
1541 	INIT_LIST_HEAD(&p->rcu_node_entry);
1542 #endif /* #ifdef CONFIG_PREEMPT_RCU */
1543 #ifdef CONFIG_TASKS_RCU
1544 	p->rcu_tasks_holdout = false;
1545 	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1546 	p->rcu_tasks_idle_cpu = -1;
1547 #endif /* #ifdef CONFIG_TASKS_RCU */
1548 }
1549 
1550 /*
1551  * This creates a new process as a copy of the old one,
1552  * but does not actually start it yet.
1553  *
1554  * It copies the registers, and all the appropriate
1555  * parts of the process environment (as per the clone
1556  * flags). The actual kick-off is left to the caller.
1557  */
1558 static __latent_entropy struct task_struct *copy_process(
1559 					unsigned long clone_flags,
1560 					unsigned long stack_start,
1561 					unsigned long stack_size,
1562 					int __user *child_tidptr,
1563 					struct pid *pid,
1564 					int trace,
1565 					unsigned long tls,
1566 					int node)
1567 {
1568 	int retval;
1569 	struct task_struct *p;
1570 
1571 	if ((clone_flags & (CLONE_NEWNS|CLONE_FS)) == (CLONE_NEWNS|CLONE_FS))
1572 		return ERR_PTR(-EINVAL);
1573 
1574 	if ((clone_flags & (CLONE_NEWUSER|CLONE_FS)) == (CLONE_NEWUSER|CLONE_FS))
1575 		return ERR_PTR(-EINVAL);
1576 
1577 	/*
1578 	 * Thread groups must share signals as well, and detached threads
1579 	 * can only be started up within the thread group.
1580 	 */
1581 	if ((clone_flags & CLONE_THREAD) && !(clone_flags & CLONE_SIGHAND))
1582 		return ERR_PTR(-EINVAL);
1583 
1584 	/*
1585 	 * Shared signal handlers imply shared VM. By way of the above,
1586 	 * thread groups also imply shared VM. Blocking this case allows
1587 	 * for various simplifications in other code.
1588 	 */
1589 	if ((clone_flags & CLONE_SIGHAND) && !(clone_flags & CLONE_VM))
1590 		return ERR_PTR(-EINVAL);
1591 
1592 	/*
1593 	 * Siblings of global init remain as zombies on exit since they are
1594 	 * not reaped by their parent (swapper). To solve this and to avoid
1595 	 * multi-rooted process trees, prevent global and container-inits
1596 	 * from creating siblings.
1597 	 */
1598 	if ((clone_flags & CLONE_PARENT) &&
1599 				current->signal->flags & SIGNAL_UNKILLABLE)
1600 		return ERR_PTR(-EINVAL);
1601 
1602 	/*
1603 	 * If the new process will be in a different pid or user namespace
1604 	 * do not allow it to share a thread group with the forking task.
1605 	 */
1606 	if (clone_flags & CLONE_THREAD) {
1607 		if ((clone_flags & (CLONE_NEWUSER | CLONE_NEWPID)) ||
1608 		    (task_active_pid_ns(current) !=
1609 				current->nsproxy->pid_ns_for_children))
1610 			return ERR_PTR(-EINVAL);
1611 	}
1612 
1613 	retval = -ENOMEM;
1614 	p = dup_task_struct(current, node);
1615 	if (!p)
1616 		goto fork_out;
1617 
1618 	/*
1619 	 * This _must_ happen before we call free_task(), i.e. before we jump
1620 	 * to any of the bad_fork_* labels. This is to avoid freeing
1621 	 * p->set_child_tid which is (ab)used as a kthread's data pointer for
1622 	 * kernel threads (PF_KTHREAD).
1623 	 */
1624 	p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
1625 	/*
1626 	 * Clear TID on mm_release()?
1627 	 */
1628 	p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
1629 
1630 	ftrace_graph_init_task(p);
1631 
1632 	rt_mutex_init_task(p);
1633 
1634 #ifdef CONFIG_PROVE_LOCKING
1635 	DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1636 	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1637 #endif
1638 	retval = -EAGAIN;
1639 	if (atomic_read(&p->real_cred->user->processes) >=
1640 			task_rlimit(p, RLIMIT_NPROC)) {
1641 		if (p->real_cred->user != INIT_USER &&
1642 		    !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN))
1643 			goto bad_fork_free;
1644 	}
1645 	current->flags &= ~PF_NPROC_EXCEEDED;
1646 
1647 	retval = copy_creds(p, clone_flags);
1648 	if (retval < 0)
1649 		goto bad_fork_free;
1650 
1651 	/*
1652 	 * If multiple threads are within copy_process(), then this check
1653 	 * triggers too late. This doesn't hurt, the check is only there
1654 	 * to stop root fork bombs.
1655 	 */
1656 	retval = -EAGAIN;
1657 	if (nr_threads >= max_threads)
1658 		goto bad_fork_cleanup_count;
1659 
1660 	delayacct_tsk_init(p);	/* Must remain after dup_task_struct() */
1661 	p->flags &= ~(PF_SUPERPRIV | PF_WQ_WORKER | PF_IDLE);
1662 	p->flags |= PF_FORKNOEXEC;
1663 	INIT_LIST_HEAD(&p->children);
1664 	INIT_LIST_HEAD(&p->sibling);
1665 	rcu_copy_process(p);
1666 	p->vfork_done = NULL;
1667 	spin_lock_init(&p->alloc_lock);
1668 
1669 	init_sigpending(&p->pending);
1670 
1671 	p->utime = p->stime = p->gtime = 0;
1672 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
1673 	p->utimescaled = p->stimescaled = 0;
1674 #endif
1675 	prev_cputime_init(&p->prev_cputime);
1676 
1677 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
1678 	seqcount_init(&p->vtime.seqcount);
1679 	p->vtime.starttime = 0;
1680 	p->vtime.state = VTIME_INACTIVE;
1681 #endif
1682 
1683 #if defined(SPLIT_RSS_COUNTING)
1684 	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1685 #endif
1686 
1687 	p->default_timer_slack_ns = current->timer_slack_ns;
1688 
1689 	task_io_accounting_init(&p->ioac);
1690 	acct_clear_integrals(p);
1691 
1692 	posix_cpu_timers_init(p);
1693 
1694 	p->start_time = ktime_get_ns();
1695 	p->real_start_time = ktime_get_boot_ns();
1696 	p->io_context = NULL;
1697 	p->audit_context = NULL;
1698 	cgroup_fork(p);
1699 #ifdef CONFIG_NUMA
1700 	p->mempolicy = mpol_dup(p->mempolicy);
1701 	if (IS_ERR(p->mempolicy)) {
1702 		retval = PTR_ERR(p->mempolicy);
1703 		p->mempolicy = NULL;
1704 		goto bad_fork_cleanup_threadgroup_lock;
1705 	}
1706 #endif
1707 #ifdef CONFIG_CPUSETS
1708 	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1709 	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1710 	seqcount_init(&p->mems_allowed_seq);
1711 #endif
1712 #ifdef CONFIG_TRACE_IRQFLAGS
1713 	p->irq_events = 0;
1714 	p->hardirqs_enabled = 0;
1715 	p->hardirq_enable_ip = 0;
1716 	p->hardirq_enable_event = 0;
1717 	p->hardirq_disable_ip = _THIS_IP_;
1718 	p->hardirq_disable_event = 0;
1719 	p->softirqs_enabled = 1;
1720 	p->softirq_enable_ip = _THIS_IP_;
1721 	p->softirq_enable_event = 0;
1722 	p->softirq_disable_ip = 0;
1723 	p->softirq_disable_event = 0;
1724 	p->hardirq_context = 0;
1725 	p->softirq_context = 0;
1726 #endif
1727 
1728 	p->pagefault_disabled = 0;
1729 
1730 #ifdef CONFIG_LOCKDEP
1731 	p->lockdep_depth = 0; /* no locks held yet */
1732 	p->curr_chain_key = 0;
1733 	p->lockdep_recursion = 0;
1734 	lockdep_init_task(p);
1735 #endif
1736 
1737 #ifdef CONFIG_DEBUG_MUTEXES
1738 	p->blocked_on = NULL; /* not blocked yet */
1739 #endif
1740 #ifdef CONFIG_BCACHE
1741 	p->sequential_io	= 0;
1742 	p->sequential_io_avg	= 0;
1743 #endif
1744 
1745 	/* Perform scheduler related setup. Assign this task to a CPU. */
1746 	retval = sched_fork(clone_flags, p);
1747 	if (retval)
1748 		goto bad_fork_cleanup_policy;
1749 
1750 	retval = perf_event_init_task(p);
1751 	if (retval)
1752 		goto bad_fork_cleanup_policy;
1753 	retval = audit_alloc(p);
1754 	if (retval)
1755 		goto bad_fork_cleanup_perf;
1756 	/* copy all the process information */
1757 	shm_init_task(p);
1758 	retval = security_task_alloc(p, clone_flags);
1759 	if (retval)
1760 		goto bad_fork_cleanup_audit;
1761 	retval = copy_semundo(clone_flags, p);
1762 	if (retval)
1763 		goto bad_fork_cleanup_security;
1764 	retval = copy_files(clone_flags, p);
1765 	if (retval)
1766 		goto bad_fork_cleanup_semundo;
1767 	retval = copy_fs(clone_flags, p);
1768 	if (retval)
1769 		goto bad_fork_cleanup_files;
1770 	retval = copy_sighand(clone_flags, p);
1771 	if (retval)
1772 		goto bad_fork_cleanup_fs;
1773 	retval = copy_signal(clone_flags, p);
1774 	if (retval)
1775 		goto bad_fork_cleanup_sighand;
1776 	retval = copy_mm(clone_flags, p);
1777 	if (retval)
1778 		goto bad_fork_cleanup_signal;
1779 	retval = copy_namespaces(clone_flags, p);
1780 	if (retval)
1781 		goto bad_fork_cleanup_mm;
1782 	retval = copy_io(clone_flags, p);
1783 	if (retval)
1784 		goto bad_fork_cleanup_namespaces;
1785 	retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
1786 	if (retval)
1787 		goto bad_fork_cleanup_io;
1788 
1789 	if (pid != &init_struct_pid) {
1790 		pid = alloc_pid(p->nsproxy->pid_ns_for_children);
1791 		if (IS_ERR(pid)) {
1792 			retval = PTR_ERR(pid);
1793 			goto bad_fork_cleanup_thread;
1794 		}
1795 	}
1796 
1797 #ifdef CONFIG_BLOCK
1798 	p->plug = NULL;
1799 #endif
1800 #ifdef CONFIG_FUTEX
1801 	p->robust_list = NULL;
1802 #ifdef CONFIG_COMPAT
1803 	p->compat_robust_list = NULL;
1804 #endif
1805 	INIT_LIST_HEAD(&p->pi_state_list);
1806 	p->pi_state_cache = NULL;
1807 #endif
1808 	/*
1809 	 * sigaltstack should be cleared when sharing the same VM
1810 	 */
1811 	if ((clone_flags & (CLONE_VM|CLONE_VFORK)) == CLONE_VM)
1812 		sas_ss_reset(p);
1813 
1814 	/*
1815 	 * Syscall tracing and stepping should be turned off in the
1816 	 * child regardless of CLONE_PTRACE.
1817 	 */
1818 	user_disable_single_step(p);
1819 	clear_tsk_thread_flag(p, TIF_SYSCALL_TRACE);
1820 #ifdef TIF_SYSCALL_EMU
1821 	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
1822 #endif
1823 	clear_all_latency_tracing(p);
1824 
1825 	/* ok, now we should be set up.. */
1826 	p->pid = pid_nr(pid);
1827 	if (clone_flags & CLONE_THREAD) {
1828 		p->exit_signal = -1;
1829 		p->group_leader = current->group_leader;
1830 		p->tgid = current->tgid;
1831 	} else {
1832 		if (clone_flags & CLONE_PARENT)
1833 			p->exit_signal = current->group_leader->exit_signal;
1834 		else
1835 			p->exit_signal = (clone_flags & CSIGNAL);
1836 		p->group_leader = p;
1837 		p->tgid = p->pid;
1838 	}
1839 
1840 	p->nr_dirtied = 0;
1841 	p->nr_dirtied_pause = 128 >> (PAGE_SHIFT - 10);
1842 	p->dirty_paused_when = 0;
1843 
1844 	p->pdeath_signal = 0;
1845 	INIT_LIST_HEAD(&p->thread_group);
1846 	p->task_works = NULL;
1847 
1848 	cgroup_threadgroup_change_begin(current);
1849 	/*
1850 	 * Ensure that the cgroup subsystem policies allow the new process to be
1851 	 * forked. It should be noted the the new process's css_set can be changed
1852 	 * between here and cgroup_post_fork() if an organisation operation is in
1853 	 * progress.
1854 	 */
1855 	retval = cgroup_can_fork(p);
1856 	if (retval)
1857 		goto bad_fork_free_pid;
1858 
1859 	/*
1860 	 * Make it visible to the rest of the system, but dont wake it up yet.
1861 	 * Need tasklist lock for parent etc handling!
1862 	 */
1863 	write_lock_irq(&tasklist_lock);
1864 
1865 	/* CLONE_PARENT re-uses the old parent */
1866 	if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1867 		p->real_parent = current->real_parent;
1868 		p->parent_exec_id = current->parent_exec_id;
1869 	} else {
1870 		p->real_parent = current;
1871 		p->parent_exec_id = current->self_exec_id;
1872 	}
1873 
1874 	klp_copy_process(p);
1875 
1876 	spin_lock(&current->sighand->siglock);
1877 
1878 	/*
1879 	 * Copy seccomp details explicitly here, in case they were changed
1880 	 * before holding sighand lock.
1881 	 */
1882 	copy_seccomp(p);
1883 
1884 	/*
1885 	 * Process group and session signals need to be delivered to just the
1886 	 * parent before the fork or both the parent and the child after the
1887 	 * fork. Restart if a signal comes in before we add the new process to
1888 	 * it's process group.
1889 	 * A fatal signal pending means that current will exit, so the new
1890 	 * thread can't slip out of an OOM kill (or normal SIGKILL).
1891 	*/
1892 	recalc_sigpending();
1893 	if (signal_pending(current)) {
1894 		retval = -ERESTARTNOINTR;
1895 		goto bad_fork_cancel_cgroup;
1896 	}
1897 	if (unlikely(!(ns_of_pid(pid)->pid_allocated & PIDNS_ADDING))) {
1898 		retval = -ENOMEM;
1899 		goto bad_fork_cancel_cgroup;
1900 	}
1901 
1902 	if (likely(p->pid)) {
1903 		ptrace_init_task(p, (clone_flags & CLONE_PTRACE) || trace);
1904 
1905 		init_task_pid(p, PIDTYPE_PID, pid);
1906 		if (thread_group_leader(p)) {
1907 			init_task_pid(p, PIDTYPE_PGID, task_pgrp(current));
1908 			init_task_pid(p, PIDTYPE_SID, task_session(current));
1909 
1910 			if (is_child_reaper(pid)) {
1911 				ns_of_pid(pid)->child_reaper = p;
1912 				p->signal->flags |= SIGNAL_UNKILLABLE;
1913 			}
1914 
1915 			p->signal->leader_pid = pid;
1916 			p->signal->tty = tty_kref_get(current->signal->tty);
1917 			/*
1918 			 * Inherit has_child_subreaper flag under the same
1919 			 * tasklist_lock with adding child to the process tree
1920 			 * for propagate_has_child_subreaper optimization.
1921 			 */
1922 			p->signal->has_child_subreaper = p->real_parent->signal->has_child_subreaper ||
1923 							 p->real_parent->signal->is_child_subreaper;
1924 			list_add_tail(&p->sibling, &p->real_parent->children);
1925 			list_add_tail_rcu(&p->tasks, &init_task.tasks);
1926 			attach_pid(p, PIDTYPE_PGID);
1927 			attach_pid(p, PIDTYPE_SID);
1928 			__this_cpu_inc(process_counts);
1929 		} else {
1930 			current->signal->nr_threads++;
1931 			atomic_inc(&current->signal->live);
1932 			atomic_inc(&current->signal->sigcnt);
1933 			list_add_tail_rcu(&p->thread_group,
1934 					  &p->group_leader->thread_group);
1935 			list_add_tail_rcu(&p->thread_node,
1936 					  &p->signal->thread_head);
1937 		}
1938 		attach_pid(p, PIDTYPE_PID);
1939 		nr_threads++;
1940 	}
1941 
1942 	total_forks++;
1943 	spin_unlock(&current->sighand->siglock);
1944 	syscall_tracepoint_update(p);
1945 	write_unlock_irq(&tasklist_lock);
1946 
1947 	proc_fork_connector(p);
1948 	cgroup_post_fork(p);
1949 	cgroup_threadgroup_change_end(current);
1950 	perf_event_fork(p);
1951 
1952 	trace_task_newtask(p, clone_flags);
1953 	uprobe_copy_process(p, clone_flags);
1954 
1955 	return p;
1956 
1957 bad_fork_cancel_cgroup:
1958 	spin_unlock(&current->sighand->siglock);
1959 	write_unlock_irq(&tasklist_lock);
1960 	cgroup_cancel_fork(p);
1961 bad_fork_free_pid:
1962 	cgroup_threadgroup_change_end(current);
1963 	if (pid != &init_struct_pid)
1964 		free_pid(pid);
1965 bad_fork_cleanup_thread:
1966 	exit_thread(p);
1967 bad_fork_cleanup_io:
1968 	if (p->io_context)
1969 		exit_io_context(p);
1970 bad_fork_cleanup_namespaces:
1971 	exit_task_namespaces(p);
1972 bad_fork_cleanup_mm:
1973 	if (p->mm)
1974 		mmput(p->mm);
1975 bad_fork_cleanup_signal:
1976 	if (!(clone_flags & CLONE_THREAD))
1977 		free_signal_struct(p->signal);
1978 bad_fork_cleanup_sighand:
1979 	__cleanup_sighand(p->sighand);
1980 bad_fork_cleanup_fs:
1981 	exit_fs(p); /* blocking */
1982 bad_fork_cleanup_files:
1983 	exit_files(p); /* blocking */
1984 bad_fork_cleanup_semundo:
1985 	exit_sem(p);
1986 bad_fork_cleanup_security:
1987 	security_task_free(p);
1988 bad_fork_cleanup_audit:
1989 	audit_free(p);
1990 bad_fork_cleanup_perf:
1991 	perf_event_free_task(p);
1992 bad_fork_cleanup_policy:
1993 	lockdep_free_task(p);
1994 #ifdef CONFIG_NUMA
1995 	mpol_put(p->mempolicy);
1996 bad_fork_cleanup_threadgroup_lock:
1997 #endif
1998 	delayacct_tsk_free(p);
1999 bad_fork_cleanup_count:
2000 	atomic_dec(&p->cred->user->processes);
2001 	exit_creds(p);
2002 bad_fork_free:
2003 	p->state = TASK_DEAD;
2004 	put_task_stack(p);
2005 	free_task(p);
2006 fork_out:
2007 	return ERR_PTR(retval);
2008 }
2009 
2010 static inline void init_idle_pids(struct pid_link *links)
2011 {
2012 	enum pid_type type;
2013 
2014 	for (type = PIDTYPE_PID; type < PIDTYPE_MAX; ++type) {
2015 		INIT_HLIST_NODE(&links[type].node); /* not really needed */
2016 		links[type].pid = &init_struct_pid;
2017 	}
2018 }
2019 
2020 struct task_struct *fork_idle(int cpu)
2021 {
2022 	struct task_struct *task;
2023 	task = copy_process(CLONE_VM, 0, 0, NULL, &init_struct_pid, 0, 0,
2024 			    cpu_to_node(cpu));
2025 	if (!IS_ERR(task)) {
2026 		init_idle_pids(task->pids);
2027 		init_idle(task, cpu);
2028 	}
2029 
2030 	return task;
2031 }
2032 
2033 /*
2034  *  Ok, this is the main fork-routine.
2035  *
2036  * It copies the process, and if successful kick-starts
2037  * it and waits for it to finish using the VM if required.
2038  */
2039 long _do_fork(unsigned long clone_flags,
2040 	      unsigned long stack_start,
2041 	      unsigned long stack_size,
2042 	      int __user *parent_tidptr,
2043 	      int __user *child_tidptr,
2044 	      unsigned long tls)
2045 {
2046 	struct task_struct *p;
2047 	int trace = 0;
2048 	long nr;
2049 
2050 	/*
2051 	 * Determine whether and which event to report to ptracer.  When
2052 	 * called from kernel_thread or CLONE_UNTRACED is explicitly
2053 	 * requested, no event is reported; otherwise, report if the event
2054 	 * for the type of forking is enabled.
2055 	 */
2056 	if (!(clone_flags & CLONE_UNTRACED)) {
2057 		if (clone_flags & CLONE_VFORK)
2058 			trace = PTRACE_EVENT_VFORK;
2059 		else if ((clone_flags & CSIGNAL) != SIGCHLD)
2060 			trace = PTRACE_EVENT_CLONE;
2061 		else
2062 			trace = PTRACE_EVENT_FORK;
2063 
2064 		if (likely(!ptrace_event_enabled(current, trace)))
2065 			trace = 0;
2066 	}
2067 
2068 	p = copy_process(clone_flags, stack_start, stack_size,
2069 			 child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
2070 	add_latent_entropy();
2071 	/*
2072 	 * Do this prior waking up the new thread - the thread pointer
2073 	 * might get invalid after that point, if the thread exits quickly.
2074 	 */
2075 	if (!IS_ERR(p)) {
2076 		struct completion vfork;
2077 		struct pid *pid;
2078 
2079 		trace_sched_process_fork(current, p);
2080 
2081 		pid = get_task_pid(p, PIDTYPE_PID);
2082 		nr = pid_vnr(pid);
2083 
2084 		if (clone_flags & CLONE_PARENT_SETTID)
2085 			put_user(nr, parent_tidptr);
2086 
2087 		if (clone_flags & CLONE_VFORK) {
2088 			p->vfork_done = &vfork;
2089 			init_completion(&vfork);
2090 			get_task_struct(p);
2091 		}
2092 
2093 		wake_up_new_task(p);
2094 
2095 		/* forking complete and child started to run, tell ptracer */
2096 		if (unlikely(trace))
2097 			ptrace_event_pid(trace, pid);
2098 
2099 		if (clone_flags & CLONE_VFORK) {
2100 			if (!wait_for_vfork_done(p, &vfork))
2101 				ptrace_event_pid(PTRACE_EVENT_VFORK_DONE, pid);
2102 		}
2103 
2104 		put_pid(pid);
2105 	} else {
2106 		nr = PTR_ERR(p);
2107 	}
2108 	return nr;
2109 }
2110 
2111 #ifndef CONFIG_HAVE_COPY_THREAD_TLS
2112 /* For compatibility with architectures that call do_fork directly rather than
2113  * using the syscall entry points below. */
2114 long do_fork(unsigned long clone_flags,
2115 	      unsigned long stack_start,
2116 	      unsigned long stack_size,
2117 	      int __user *parent_tidptr,
2118 	      int __user *child_tidptr)
2119 {
2120 	return _do_fork(clone_flags, stack_start, stack_size,
2121 			parent_tidptr, child_tidptr, 0);
2122 }
2123 #endif
2124 
2125 /*
2126  * Create a kernel thread.
2127  */
2128 pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
2129 {
2130 	return _do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn,
2131 		(unsigned long)arg, NULL, NULL, 0);
2132 }
2133 
2134 #ifdef __ARCH_WANT_SYS_FORK
2135 SYSCALL_DEFINE0(fork)
2136 {
2137 #ifdef CONFIG_MMU
2138 	return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
2139 #else
2140 	/* can not support in nommu mode */
2141 	return -EINVAL;
2142 #endif
2143 }
2144 #endif
2145 
2146 #ifdef __ARCH_WANT_SYS_VFORK
2147 SYSCALL_DEFINE0(vfork)
2148 {
2149 	return _do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, 0,
2150 			0, NULL, NULL, 0);
2151 }
2152 #endif
2153 
2154 #ifdef __ARCH_WANT_SYS_CLONE
2155 #ifdef CONFIG_CLONE_BACKWARDS
2156 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2157 		 int __user *, parent_tidptr,
2158 		 unsigned long, tls,
2159 		 int __user *, child_tidptr)
2160 #elif defined(CONFIG_CLONE_BACKWARDS2)
2161 SYSCALL_DEFINE5(clone, unsigned long, newsp, unsigned long, clone_flags,
2162 		 int __user *, parent_tidptr,
2163 		 int __user *, child_tidptr,
2164 		 unsigned long, tls)
2165 #elif defined(CONFIG_CLONE_BACKWARDS3)
2166 SYSCALL_DEFINE6(clone, unsigned long, clone_flags, unsigned long, newsp,
2167 		int, stack_size,
2168 		int __user *, parent_tidptr,
2169 		int __user *, child_tidptr,
2170 		unsigned long, tls)
2171 #else
2172 SYSCALL_DEFINE5(clone, unsigned long, clone_flags, unsigned long, newsp,
2173 		 int __user *, parent_tidptr,
2174 		 int __user *, child_tidptr,
2175 		 unsigned long, tls)
2176 #endif
2177 {
2178 	return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
2179 }
2180 #endif
2181 
2182 void walk_process_tree(struct task_struct *top, proc_visitor visitor, void *data)
2183 {
2184 	struct task_struct *leader, *parent, *child;
2185 	int res;
2186 
2187 	read_lock(&tasklist_lock);
2188 	leader = top = top->group_leader;
2189 down:
2190 	for_each_thread(leader, parent) {
2191 		list_for_each_entry(child, &parent->children, sibling) {
2192 			res = visitor(child, data);
2193 			if (res) {
2194 				if (res < 0)
2195 					goto out;
2196 				leader = child;
2197 				goto down;
2198 			}
2199 up:
2200 			;
2201 		}
2202 	}
2203 
2204 	if (leader != top) {
2205 		child = leader;
2206 		parent = child->real_parent;
2207 		leader = parent->group_leader;
2208 		goto up;
2209 	}
2210 out:
2211 	read_unlock(&tasklist_lock);
2212 }
2213 
2214 #ifndef ARCH_MIN_MMSTRUCT_ALIGN
2215 #define ARCH_MIN_MMSTRUCT_ALIGN 0
2216 #endif
2217 
2218 static void sighand_ctor(void *data)
2219 {
2220 	struct sighand_struct *sighand = data;
2221 
2222 	spin_lock_init(&sighand->siglock);
2223 	init_waitqueue_head(&sighand->signalfd_wqh);
2224 }
2225 
2226 void __init proc_caches_init(void)
2227 {
2228 	sighand_cachep = kmem_cache_create("sighand_cache",
2229 			sizeof(struct sighand_struct), 0,
2230 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_TYPESAFE_BY_RCU|
2231 			SLAB_ACCOUNT, sighand_ctor);
2232 	signal_cachep = kmem_cache_create("signal_cache",
2233 			sizeof(struct signal_struct), 0,
2234 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2235 			NULL);
2236 	files_cachep = kmem_cache_create("files_cache",
2237 			sizeof(struct files_struct), 0,
2238 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2239 			NULL);
2240 	fs_cachep = kmem_cache_create("fs_cache",
2241 			sizeof(struct fs_struct), 0,
2242 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2243 			NULL);
2244 	/*
2245 	 * FIXME! The "sizeof(struct mm_struct)" currently includes the
2246 	 * whole struct cpumask for the OFFSTACK case. We could change
2247 	 * this to *only* allocate as much of it as required by the
2248 	 * maximum number of CPU's we can ever have.  The cpumask_allocation
2249 	 * is at the end of the structure, exactly for that reason.
2250 	 */
2251 	mm_cachep = kmem_cache_create("mm_struct",
2252 			sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
2253 			SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT,
2254 			NULL);
2255 	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC|SLAB_ACCOUNT);
2256 	mmap_init();
2257 	nsproxy_cache_init();
2258 }
2259 
2260 /*
2261  * Check constraints on flags passed to the unshare system call.
2262  */
2263 static int check_unshare_flags(unsigned long unshare_flags)
2264 {
2265 	if (unshare_flags & ~(CLONE_THREAD|CLONE_FS|CLONE_NEWNS|CLONE_SIGHAND|
2266 				CLONE_VM|CLONE_FILES|CLONE_SYSVSEM|
2267 				CLONE_NEWUTS|CLONE_NEWIPC|CLONE_NEWNET|
2268 				CLONE_NEWUSER|CLONE_NEWPID|CLONE_NEWCGROUP))
2269 		return -EINVAL;
2270 	/*
2271 	 * Not implemented, but pretend it works if there is nothing
2272 	 * to unshare.  Note that unsharing the address space or the
2273 	 * signal handlers also need to unshare the signal queues (aka
2274 	 * CLONE_THREAD).
2275 	 */
2276 	if (unshare_flags & (CLONE_THREAD | CLONE_SIGHAND | CLONE_VM)) {
2277 		if (!thread_group_empty(current))
2278 			return -EINVAL;
2279 	}
2280 	if (unshare_flags & (CLONE_SIGHAND | CLONE_VM)) {
2281 		if (atomic_read(&current->sighand->count) > 1)
2282 			return -EINVAL;
2283 	}
2284 	if (unshare_flags & CLONE_VM) {
2285 		if (!current_is_single_threaded())
2286 			return -EINVAL;
2287 	}
2288 
2289 	return 0;
2290 }
2291 
2292 /*
2293  * Unshare the filesystem structure if it is being shared
2294  */
2295 static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
2296 {
2297 	struct fs_struct *fs = current->fs;
2298 
2299 	if (!(unshare_flags & CLONE_FS) || !fs)
2300 		return 0;
2301 
2302 	/* don't need lock here; in the worst case we'll do useless copy */
2303 	if (fs->users == 1)
2304 		return 0;
2305 
2306 	*new_fsp = copy_fs_struct(fs);
2307 	if (!*new_fsp)
2308 		return -ENOMEM;
2309 
2310 	return 0;
2311 }
2312 
2313 /*
2314  * Unshare file descriptor table if it is being shared
2315  */
2316 static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
2317 {
2318 	struct files_struct *fd = current->files;
2319 	int error = 0;
2320 
2321 	if ((unshare_flags & CLONE_FILES) &&
2322 	    (fd && atomic_read(&fd->count) > 1)) {
2323 		*new_fdp = dup_fd(fd, &error);
2324 		if (!*new_fdp)
2325 			return error;
2326 	}
2327 
2328 	return 0;
2329 }
2330 
2331 /*
2332  * unshare allows a process to 'unshare' part of the process
2333  * context which was originally shared using clone.  copy_*
2334  * functions used by do_fork() cannot be used here directly
2335  * because they modify an inactive task_struct that is being
2336  * constructed. Here we are modifying the current, active,
2337  * task_struct.
2338  */
2339 SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
2340 {
2341 	struct fs_struct *fs, *new_fs = NULL;
2342 	struct files_struct *fd, *new_fd = NULL;
2343 	struct cred *new_cred = NULL;
2344 	struct nsproxy *new_nsproxy = NULL;
2345 	int do_sysvsem = 0;
2346 	int err;
2347 
2348 	/*
2349 	 * If unsharing a user namespace must also unshare the thread group
2350 	 * and unshare the filesystem root and working directories.
2351 	 */
2352 	if (unshare_flags & CLONE_NEWUSER)
2353 		unshare_flags |= CLONE_THREAD | CLONE_FS;
2354 	/*
2355 	 * If unsharing vm, must also unshare signal handlers.
2356 	 */
2357 	if (unshare_flags & CLONE_VM)
2358 		unshare_flags |= CLONE_SIGHAND;
2359 	/*
2360 	 * If unsharing a signal handlers, must also unshare the signal queues.
2361 	 */
2362 	if (unshare_flags & CLONE_SIGHAND)
2363 		unshare_flags |= CLONE_THREAD;
2364 	/*
2365 	 * If unsharing namespace, must also unshare filesystem information.
2366 	 */
2367 	if (unshare_flags & CLONE_NEWNS)
2368 		unshare_flags |= CLONE_FS;
2369 
2370 	err = check_unshare_flags(unshare_flags);
2371 	if (err)
2372 		goto bad_unshare_out;
2373 	/*
2374 	 * CLONE_NEWIPC must also detach from the undolist: after switching
2375 	 * to a new ipc namespace, the semaphore arrays from the old
2376 	 * namespace are unreachable.
2377 	 */
2378 	if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
2379 		do_sysvsem = 1;
2380 	err = unshare_fs(unshare_flags, &new_fs);
2381 	if (err)
2382 		goto bad_unshare_out;
2383 	err = unshare_fd(unshare_flags, &new_fd);
2384 	if (err)
2385 		goto bad_unshare_cleanup_fs;
2386 	err = unshare_userns(unshare_flags, &new_cred);
2387 	if (err)
2388 		goto bad_unshare_cleanup_fd;
2389 	err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
2390 					 new_cred, new_fs);
2391 	if (err)
2392 		goto bad_unshare_cleanup_cred;
2393 
2394 	if (new_fs || new_fd || do_sysvsem || new_cred || new_nsproxy) {
2395 		if (do_sysvsem) {
2396 			/*
2397 			 * CLONE_SYSVSEM is equivalent to sys_exit().
2398 			 */
2399 			exit_sem(current);
2400 		}
2401 		if (unshare_flags & CLONE_NEWIPC) {
2402 			/* Orphan segments in old ns (see sem above). */
2403 			exit_shm(current);
2404 			shm_init_task(current);
2405 		}
2406 
2407 		if (new_nsproxy)
2408 			switch_task_namespaces(current, new_nsproxy);
2409 
2410 		task_lock(current);
2411 
2412 		if (new_fs) {
2413 			fs = current->fs;
2414 			spin_lock(&fs->lock);
2415 			current->fs = new_fs;
2416 			if (--fs->users)
2417 				new_fs = NULL;
2418 			else
2419 				new_fs = fs;
2420 			spin_unlock(&fs->lock);
2421 		}
2422 
2423 		if (new_fd) {
2424 			fd = current->files;
2425 			current->files = new_fd;
2426 			new_fd = fd;
2427 		}
2428 
2429 		task_unlock(current);
2430 
2431 		if (new_cred) {
2432 			/* Install the new user namespace */
2433 			commit_creds(new_cred);
2434 			new_cred = NULL;
2435 		}
2436 	}
2437 
2438 	perf_event_namespaces(current);
2439 
2440 bad_unshare_cleanup_cred:
2441 	if (new_cred)
2442 		put_cred(new_cred);
2443 bad_unshare_cleanup_fd:
2444 	if (new_fd)
2445 		put_files_struct(new_fd);
2446 
2447 bad_unshare_cleanup_fs:
2448 	if (new_fs)
2449 		free_fs_struct(new_fs);
2450 
2451 bad_unshare_out:
2452 	return err;
2453 }
2454 
2455 /*
2456  *	Helper to unshare the files of the current task.
2457  *	We don't want to expose copy_files internals to
2458  *	the exec layer of the kernel.
2459  */
2460 
2461 int unshare_files(struct files_struct **displaced)
2462 {
2463 	struct task_struct *task = current;
2464 	struct files_struct *copy = NULL;
2465 	int error;
2466 
2467 	error = unshare_fd(CLONE_FILES, &copy);
2468 	if (error || !copy) {
2469 		*displaced = NULL;
2470 		return error;
2471 	}
2472 	*displaced = task->files;
2473 	task_lock(task);
2474 	task->files = copy;
2475 	task_unlock(task);
2476 	return 0;
2477 }
2478 
2479 int sysctl_max_threads(struct ctl_table *table, int write,
2480 		       void __user *buffer, size_t *lenp, loff_t *ppos)
2481 {
2482 	struct ctl_table t;
2483 	int ret;
2484 	int threads = max_threads;
2485 	int min = MIN_THREADS;
2486 	int max = MAX_THREADS;
2487 
2488 	t = *table;
2489 	t.data = &threads;
2490 	t.extra1 = &min;
2491 	t.extra2 = &max;
2492 
2493 	ret = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
2494 	if (ret || !write)
2495 		return ret;
2496 
2497 	set_max_threads(threads);
2498 
2499 	return 0;
2500 }
2501