~hc/RK356X_SDK_RELEASE.git

..	..	@@ -1,3 +1,4 @@
	1	+// SPDX-License-Identifier: GPL-2.0-only
1	2	/*
2	3	* linux/kernel/fork.c
3	4	*
..	..	@@ -39,10 +40,8 @@
39	40	#include <linux/binfmts.h>
40	41	#include <linux/mman.h>
41	42	#include <linux/mmu_notifier.h>
42		-#include <linux/hmm.h>
43	43	#include <linux/fs.h>
44	44	#include <linux/mm.h>
45		-#include <linux/kprobes.h>
46	45	#include <linux/vmacache.h>
47	46	#include <linux/nsproxy.h>
48	47	#include <linux/capability.h>
..	..	@@ -80,7 +79,6 @@
80	79	#include <linux/blkdev.h>
81	80	#include <linux/fs_struct.h>
82	81	#include <linux/magic.h>
83		-#include <linux/sched/mm.h>
84	82	#include <linux/perf_event.h>
85	83	#include <linux/posix-timers.h>
86	84	#include <linux/user-return-notifier.h>
..	..	@@ -94,10 +92,12 @@
94	92	#include <linux/kcov.h>
95	93	#include <linux/livepatch.h>
96	94	#include <linux/thread_info.h>
97		-#include <linux/cpufreq_times.h>
	95	+#include <linux/stackleak.h>
	96	+#include <linux/kasan.h>
98	97	#include <linux/scs.h>
	98	+#include <linux/io_uring.h>
	99	+#include <linux/cpufreq_times.h>
99	100
100		-#include <asm/pgtable.h>
101	101	#include <asm/pgalloc.h>
102	102	#include <linux/uaccess.h>
103	103	#include <asm/mmu_context.h>
..	..	@@ -109,6 +109,8 @@
109	109	#define CREATE_TRACE_POINTS
110	110	#include <trace/events/task.h>
111	111
	112	+#undef CREATE_TRACE_POINTS
	113	+#include <trace/hooks/sched.h>
112	114	/*
113	115	* Minimum number of threads to boot the kernel
114	116	*/
..	..	@@ -119,17 +121,29 @@
119	121	*/
120	122	#define MAX_THREADS FUTEX_TID_MASK
121	123
	124	+EXPORT_TRACEPOINT_SYMBOL_GPL(task_newtask);
	125	+
122	126	/*
123	127	* Protected counters by write_lock_irq(&tasklist_lock)
124	128	*/
125	129	unsigned long total_forks; /* Handle normal Linux uptimes. */
126	130	int nr_threads; /* The idle threads do not count.. */
127	131
128		-int max_threads; /* tunable limit on nr_threads */
	132	+static int max_threads; /* tunable limit on nr_threads */
	133	+
	134	+#define NAMED_ARRAY_INDEX(x) [x] = __stringify(x)
	135	+
	136	+static const char * const resident_page_types[] = {
	137	+ NAMED_ARRAY_INDEX(MM_FILEPAGES),
	138	+ NAMED_ARRAY_INDEX(MM_ANONPAGES),
	139	+ NAMED_ARRAY_INDEX(MM_SWAPENTS),
	140	+ NAMED_ARRAY_INDEX(MM_SHMEMPAGES),
	141	+};
129	142
130	143	DEFINE_PER_CPU(unsigned long, process_counts) = 0;
131	144
132	145	__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
	146	+EXPORT_SYMBOL_GPL(tasklist_lock);
133	147
134	148	#ifdef CONFIG_PROVE_RCU
135	149	int lockdep_tasklist_lock_is_held(void)
..	..	@@ -217,6 +231,9 @@
217	231	if (!s)
218	232	continue;
219	233
	234	+ /* Mark stack accessible for KASAN. */
	235	+ kasan_unpoison_range(s->addr, THREAD_SIZE);
	236	+
220	237	/* Clear stale pointers from reused stack. */
221	238	memset(s->addr, 0, THREAD_SIZE);
222	239
..	..	@@ -225,9 +242,14 @@
225	242	return s->addr;
226	243	}
227	244
	245	+ /*
	246	+ * Allocated stacks are cached and later reused by new threads,
	247	+ * so memcg accounting is performed manually on assigning/releasing
	248	+ * stacks to tasks. Drop __GFP_ACCOUNT.
	249	+ */
228	250	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
229	251	VMALLOC_START, VMALLOC_END,
230		- THREADINFO_GFP,
	252	+ THREADINFO_GFP & ~__GFP_ACCOUNT,
231	253	PAGE_KERNEL,
232	254	0, node, __builtin_return_address(0));
233	255
..	..	@@ -246,7 +268,7 @@
246	268	THREAD_SIZE_ORDER);
247	269
248	270	if (likely(page)) {
249		- tsk->stack = page_address(page);
	271	+ tsk->stack = kasan_reset_tag(page_address(page));
250	272	return tsk->stack;
251	273	}
252	274	return NULL;
..	..	@@ -256,8 +278,13 @@
256	278	static inline void free_thread_stack(struct task_struct *tsk)
257	279	{
258	280	#ifdef CONFIG_VMAP_STACK
259		- if (task_stack_vm_area(tsk)) {
	281	+ struct vm_struct *vm = task_stack_vm_area(tsk);
	282	+
	283	+ if (vm) {
260	284	int i;
	285	+
	286	+ for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
	287	+ memcg_kmem_uncharge_page(vm->pages[i], 0);
261	288
262	289	for (i = 0; i < NR_CACHED_STACKS; i++) {
263	290	if (this_cpu_cmpxchg(cached_stacks[i],
..	..	@@ -282,6 +309,7 @@
282	309	{
283	310	unsigned long *stack;
284	311	stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
	312	+ stack = kasan_reset_tag(stack);
285	313	tsk->stack = stack;
286	314	return stack;
287	315	}
..	..	@@ -334,8 +362,15 @@
334	362	struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
335	363
336	364	if (new) {
337		- new = orig;
338		- INIT_LIST_HEAD(&new->anon_vma_chain);
	365	+ ASSERT_EXCLUSIVE_WRITER(orig->vm_flags);
	366	+ ASSERT_EXCLUSIVE_WRITER(orig->vm_file);
	367	+ /*
	368	+ * orig->shared.rb may be modified concurrently, but the clone
	369	+ * will be reinitialized.
	370	+ */
	371	+ new = data_race(orig);
	372	+ INIT_VMA(new);
	373	+ new->vm_next = new->vm_prev = NULL;
339	374	}
340	375	return new;
341	376	}
..	..	@@ -350,6 +385,22 @@
350	385	void *stack = task_stack_page(tsk);
351	386	struct vm_struct *vm = task_stack_vm_area(tsk);
352	387
	388	+
	389	+ /* All stack pages are in the same node. */
	390	+ if (vm)
	391	+ mod_lruvec_page_state(vm->pages[0], NR_KERNEL_STACK_KB,
	392	+ account * (THREAD_SIZE / 1024));
	393	+ else
	394	+ mod_lruvec_slab_state(stack, NR_KERNEL_STACK_KB,
	395	+ account * (THREAD_SIZE / 1024));
	396	+}
	397	+
	398	+static int memcg_charge_kernel_stack(struct task_struct *tsk)
	399	+{
	400	+#ifdef CONFIG_VMAP_STACK
	401	+ struct vm_struct *vm = task_stack_vm_area(tsk);
	402	+ int ret;
	403	+
353	404	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
354	405
355	406	if (vm) {
..	..	@@ -358,27 +409,19 @@
358	409	BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
359	410
360	411	for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
361		- mod_zone_page_state(page_zone(vm->pages[i]),
362		- NR_KERNEL_STACK_KB,
363		- PAGE_SIZE / 1024 * account);
	412	+ /*
	413	+ * If memcg_kmem_charge_page() fails, page->mem_cgroup
	414	+ * pointer is NULL, and memcg_kmem_uncharge_page() in
	415	+ * free_thread_stack() will ignore this page.
	416	+ */
	417	+ ret = memcg_kmem_charge_page(vm->pages[i], GFP_KERNEL,
	418	+ 0);
	419	+ if (ret)
	420	+ return ret;
364	421	}
365		-
366		- /* All stack pages belong to the same memcg. */
367		- mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
368		- account * (THREAD_SIZE / 1024));
369		- } else {
370		- /*
371		- * All stack pages are in the same zone and belong to the
372		- * same memcg.
373		- */
374		- struct page *first_page = virt_to_page(stack);
375		-
376		- mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
377		- THREAD_SIZE / 1024 * account);
378		-
379		- mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
380		- account * (THREAD_SIZE / 1024));
381	422	}
	423	+#endif
	424	+ return 0;
382	425	}
383	426
384	427	static void release_task_stack(struct task_struct *tsk)
..	..	@@ -397,9 +440,10 @@
397	440	#ifdef CONFIG_THREAD_INFO_IN_TASK
398	441	void put_task_stack(struct task_struct *tsk)
399	442	{
400		- if (atomic_dec_and_test(&tsk->stack_refcount))
	443	+ if (refcount_dec_and_test(&tsk->stack_refcount))
401	444	release_task_stack(tsk);
402	445	}
	446	+EXPORT_SYMBOL_GPL(put_task_stack);
403	447	#endif
404	448
405	449	void free_task(struct task_struct *tsk)
..	..	@@ -407,6 +451,7 @@
407	451	cpufreq_task_times_exit(tsk);
408	452	scs_release(tsk);
409	453
	454	+ trace_android_vh_free_task(tsk);
410	455	#ifndef CONFIG_THREAD_INFO_IN_TASK
411	456	/*
412	457	* The task is finally done with both the stack and thread_info,
..	..	@@ -418,11 +463,10 @@
418	463	* If the task had a separate stack allocation, it should be gone
419	464	* by now.
420	465	*/
421		- WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
	466	+ WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
422	467	#endif
423	468	rt_mutex_debug_task_free(tsk);
424	469	ftrace_graph_exit_task(tsk);
425		- put_seccomp_filter(tsk);
426	470	arch_release_task_struct(tsk);
427	471	if (tsk->flags & PF_KTHREAD)
428	472	free_kthread_struct(tsk);
..	..	@@ -434,14 +478,14 @@
434	478	static __latent_entropy int dup_mmap(struct mm_struct *mm,
435	479	struct mm_struct *oldmm)
436	480	{
437		- struct vm_area_struct mpnt, tmp, prev, *pprev;
	481	+ struct vm_area_struct mpnt, tmp, prev, pprev, last = NULL;
438	482	struct rb_node *rb_link, rb_parent;
439	483	int retval;
440	484	unsigned long charge;
441	485	LIST_HEAD(uf);
442	486
443	487	uprobe_start_dup_mmap();
444		- if (down_write_killable(&oldmm->mmap_sem)) {
	488	+ if (mmap_write_lock_killable(oldmm)) {
445	489	retval = -EINTR;
446	490	goto fail_uprobe_end;
447	491	}
..	..	@@ -450,7 +494,7 @@
450	494	/*
451	495	* Not linked in yet - no deadlock potential:
452	496	*/
453		- down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
	497	+ mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
454	498
455	499	/* No ordering required: file already has been exposed. */
456	500	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
..	..	@@ -505,14 +549,15 @@
505	549	if (retval)
506	550	goto fail_nomem_anon_vma_fork;
507	551	if (tmp->vm_flags & VM_WIPEONFORK) {
508		- /* VM_WIPEONFORK gets a clean slate in the child. */
	552	+ /*
	553	+ * VM_WIPEONFORK gets a clean slate in the child.
	554	+ * Don't prepare anon_vma until fault since we don't
	555	+ * copy page for current vma.
	556	+ */
509	557	tmp->anon_vma = NULL;
510		- if (anon_vma_prepare(tmp))
511		- goto fail_nomem_anon_vma_fork;
512	558	} else if (anon_vma_fork(tmp, mpnt))
513	559	goto fail_nomem_anon_vma_fork;
514	560	tmp->vm_flags &= ~(VM_LOCKED \| VM_LOCKONFAULT);
515		- tmp->vm_next = tmp->vm_prev = NULL;
516	561	file = tmp->vm_file;
517	562	if (file) {
518	563	struct inode *inode = file_inode(file);
..	..	@@ -520,10 +565,10 @@
520	565
521	566	get_file(file);
522	567	if (tmp->vm_flags & VM_DENYWRITE)
523		- atomic_dec(&inode->i_writecount);
	568	+ put_write_access(inode);
524	569	i_mmap_lock_write(mapping);
525	570	if (tmp->vm_flags & VM_SHARED)
526		- atomic_inc(&mapping->i_mmap_writable);
	571	+ mapping_allow_writable(mapping);
527	572	flush_dcache_mmap_lock(mapping);
528	573	/* insert tmp into the share list, just after mpnt */
529	574	vma_interval_tree_insert_after(tmp, mpnt,
..	..	@@ -553,8 +598,18 @@
553	598	rb_parent = &tmp->vm_rb;
554	599
555	600	mm->map_count++;
556		- if (!(tmp->vm_flags & VM_WIPEONFORK))
557		- retval = copy_page_range(mm, oldmm, mpnt);
	601	+ if (!(tmp->vm_flags & VM_WIPEONFORK)) {
	602	+ if (IS_ENABLED(CONFIG_SPECULATIVE_PAGE_FAULT)) {
	603	+ /*
	604	+ * Mark this VMA as changing to prevent the
	605	+ * speculative page fault hanlder to process
	606	+ * it until the TLB are flushed below.
	607	+ */
	608	+ last = mpnt;
	609	+ vm_write_begin(mpnt);
	610	+ }
	611	+ retval = copy_page_range(tmp, mpnt);
	612	+ }
558	613
559	614	if (tmp->vm_ops && tmp->vm_ops->open)
560	615	tmp->vm_ops->open(tmp);
..	..	@@ -565,9 +620,25 @@
565	620	/* a new mm has just been created */
566	621	retval = arch_dup_mmap(oldmm, mm);
567	622	out:
568		- up_write(&mm->mmap_sem);
	623	+ mmap_write_unlock(mm);
569	624	flush_tlb_mm(oldmm);
570		- up_write(&oldmm->mmap_sem);
	625	+
	626	+ if (IS_ENABLED(CONFIG_SPECULATIVE_PAGE_FAULT)) {
	627	+ /*
	628	+ * Since the TLB has been flush, we can safely unmark the
	629	+ * copied VMAs and allows the speculative page fault handler to
	630	+ * process them again.
	631	+ * Walk back the VMA list from the last marked VMA.
	632	+ */
	633	+ for (; last; last = last->vm_prev) {
	634	+ if (last->vm_flags & VM_DONTCOPY)
	635	+ continue;
	636	+ if (!(last->vm_flags & VM_WIPEONFORK))
	637	+ vm_write_end(last);
	638	+ }
	639	+ }
	640	+
	641	+ mmap_write_unlock(oldmm);
571	642	dup_userfaultfd_complete(&uf);
572	643	fail_uprobe_end:
573	644	uprobe_end_dup_mmap();
..	..	@@ -597,9 +668,9 @@
597	668	#else
598	669	static int dup_mmap(struct mm_struct mm, struct mm_struct oldmm)
599	670	{
600		- down_write(&oldmm->mmap_sem);
	671	+ mmap_write_lock(oldmm);
601	672	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
602		- up_write(&oldmm->mmap_sem);
	673	+ mmap_write_unlock(oldmm);
603	674	return 0;
604	675	}
605	676	#define mm_alloc_pgd(mm) (0)
..	..	@@ -610,12 +681,15 @@
610	681	{
611	682	int i;
612	683
	684	+ BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
	685	+ "Please make sure 'struct resident_page_types[]' is updated as well");
	686	+
613	687	for (i = 0; i < NR_MM_COUNTERS; i++) {
614	688	long x = atomic_long_read(&mm->rss_stat.count[i]);
615	689
616	690	if (unlikely(x))
617		- printk(KERN_ALERT "BUG: Bad rss-counter state "
618		- "mm:%p idx:%d val:%ld\n", mm, i, x);
	691	+ pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
	692	+ mm, resident_page_types[i], x);
619	693	}
620	694
621	695	if (mm_pgtables_bytes(mm))
..	..	@@ -642,26 +716,12 @@
642	716	WARN_ON_ONCE(mm == current->active_mm);
643	717	mm_free_pgd(mm);
644	718	destroy_context(mm);
645		- hmm_mm_destroy(mm);
646		- mmu_notifier_mm_destroy(mm);
	719	+ mmu_notifier_subscriptions_destroy(mm);
647	720	check_mm(mm);
648	721	put_user_ns(mm->user_ns);
649	722	free_mm(mm);
650	723	}
651	724	EXPORT_SYMBOL_GPL(__mmdrop);
652		-
653		-#ifdef CONFIG_PREEMPT_RT_BASE
654		-/*
655		- * RCU callback for delayed mm drop. Not strictly rcu, but we don't
656		- * want another facility to make this work.
657		- */
658		-void __mmdrop_delayed(struct rcu_head *rhp)
659		-{
660		- struct mm_struct *mm = container_of(rhp, struct mm_struct, delayed_drop);
661		-
662		- __mmdrop(mm);
663		-}
664		-#endif
665	725
666	726	static void mmdrop_async_fn(struct work_struct *work)
667	727	{
..	..	@@ -694,27 +754,17 @@
694	754
695	755	static inline void put_signal_struct(struct signal_struct *sig)
696	756	{
697		- if (atomic_dec_and_test(&sig->sigcnt))
	757	+ if (refcount_dec_and_test(&sig->sigcnt))
698	758	free_signal_struct(sig);
699	759	}
700		-#ifdef CONFIG_PREEMPT_RT_BASE
701		-static
702		-#endif
	760	+
703	761	void __put_task_struct(struct task_struct *tsk)
704	762	{
705	763	WARN_ON(!tsk->exit_state);
706		- WARN_ON(atomic_read(&tsk->usage));
	764	+ WARN_ON(refcount_read(&tsk->usage));
707	765	WARN_ON(tsk == current);
708	766
709		- /*
710		- * Remove function-return probe instances associated with this
711		- * task and put them back on the free list.
712		- */
713		- kprobe_flush_task(tsk);
714		-
715		- /* Task is done with its stack. */
716		- put_task_stack(tsk);
717		-
	767	+ io_uring_free(tsk);
718	768	cgroup_free(tsk);
719	769	task_numa_free(tsk, true);
720	770	security_task_free(tsk);
..	..	@@ -725,18 +775,7 @@
725	775	if (!profile_handoff_task(tsk))
726	776	free_task(tsk);
727	777	}
728		-#ifndef CONFIG_PREEMPT_RT_BASE
729	778	EXPORT_SYMBOL_GPL(__put_task_struct);
730		-#else
731		-void __put_task_struct_cb(struct rcu_head *rhp)
732		-{
733		- struct task_struct *tsk = container_of(rhp, struct task_struct, put_rcu);
734		-
735		- __put_task_struct(tsk);
736		-
737		-}
738		-EXPORT_SYMBOL_GPL(__put_task_struct_cb);
739		-#endif
740	779
741	780	void __init __weak arch_task_cache_init(void) { }
742	781
..	..	@@ -746,15 +785,16 @@
746	785	static void set_max_threads(unsigned int max_threads_suggested)
747	786	{
748	787	u64 threads;
	788	+ unsigned long nr_pages = totalram_pages();
749	789
750	790	/*
751	791	* The number of threads shall be limited such that the thread
752	792	* structures may only consume a small part of the available memory.
753	793	*/
754		- if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
	794	+ if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
755	795	threads = MAX_THREADS;
756	796	else
757		- threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
	797	+ threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
758	798	(u64) THREAD_SIZE * 8UL);
759	799
760	800	if (threads > max_threads_suggested)
..	..	@@ -768,6 +808,7 @@
768	808	int arch_task_struct_size __read_mostly;
769	809	#endif
770	810
	811	+#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
771	812	static void task_struct_whitelist(unsigned long offset, unsigned long size)
772	813	{
773	814	/* Fetch thread_struct whitelist for the architecture. */
..	..	@@ -782,6 +823,7 @@
782	823	else
783	824	*offset += offsetof(struct task_struct, thread);
784	825	}
	826	+#endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
785	827
786	828	void __init fork_init(void)
787	829	{
..	..	@@ -823,6 +865,7 @@
823	865	scs_init();
824	866
825	867	lockdep_init_task(&init_task);
	868	+ uprobes_init();
826	869	}
827	870
828	871	int __weak arch_dup_task_struct(struct task_struct *dst,
..	..	@@ -844,7 +887,7 @@
844	887	{
845	888	struct task_struct *tsk;
846	889	unsigned long *stack;
847		- struct vm_struct *stack_vm_area;
	890	+ struct vm_struct *stack_vm_area __maybe_unused;
848	891	int err;
849	892
850	893	if (node == NUMA_NO_NODE)
..	..	@@ -856,6 +899,9 @@
856	899	stack = alloc_thread_stack_node(tsk, node);
857	900	if (!stack)
858	901	goto free_tsk;
	902	+
	903	+ if (memcg_charge_kernel_stack(tsk))
	904	+ goto free_stack;
859	905
860	906	stack_vm_area = task_stack_vm_area(tsk);
861	907
..	..	@@ -871,7 +917,7 @@
871	917	tsk->stack_vm_area = stack_vm_area;
872	918	#endif
873	919	#ifdef CONFIG_THREAD_INFO_IN_TASK
874		- atomic_set(&tsk->stack_refcount, 1);
	920	+ refcount_set(&tsk->stack_refcount, 1);
875	921	#endif
876	922
877	923	if (err)
..	..	@@ -903,17 +949,19 @@
903	949	tsk->cpus_ptr = &tsk->cpus_mask;
904	950
905	951	/*
906		- * One for us, one for whoever does the "release_task()" (usually
907		- * parent)
	952	+ * One for the user space visible state that goes away when reaped.
	953	+ * One for the scheduler.
908	954	*/
909		- atomic_set(&tsk->usage, 2);
	955	+ refcount_set(&tsk->rcu_users, 2);
	956	+ /* One for the rcu users */
	957	+ refcount_set(&tsk->usage, 1);
910	958	#ifdef CONFIG_BLK_DEV_IO_TRACE
911	959	tsk->btrace_seq = 0;
912	960	#endif
913	961	tsk->splice_pipe = NULL;
914	962	tsk->task_frag.page = NULL;
915	963	tsk->wake_q.next = NULL;
916		- tsk->wake_q_sleeper.next = NULL;
	964	+ tsk->pf_io_worker = NULL;
917	965
918	966	account_kernel_stack(tsk, 1);
919	967
..	..	@@ -931,6 +979,11 @@
931	979	#ifdef CONFIG_MEMCG
932	980	tsk->active_memcg = NULL;
933	981	#endif
	982	+
	983	+ android_init_vendor_data(tsk, 1);
	984	+ android_init_oem_data(tsk, 1);
	985	+
	986	+ trace_android_vh_dup_task_struct(tsk, orig);
934	987	return tsk;
935	988
936	989	free_stack:
..	..	@@ -980,6 +1033,13 @@
980	1033	#endif
981	1034	}
982	1035
	1036	+static void mm_init_pasid(struct mm_struct *mm)
	1037	+{
	1038	+#ifdef CONFIG_IOMMU_SUPPORT
	1039	+ mm->pasid = INIT_PASID;
	1040	+#endif
	1041	+}
	1042	+
983	1043	static void mm_init_uprobes_state(struct mm_struct *mm)
984	1044	{
985	1045	#ifdef CONFIG_UPROBES
..	..	@@ -993,24 +1053,30 @@
993	1053	mm->mmap = NULL;
994	1054	mm->mm_rb = RB_ROOT;
995	1055	mm->vmacache_seqnum = 0;
	1056	+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
	1057	+ rwlock_init(&mm->mm_rb_lock);
	1058	+#endif
996	1059	atomic_set(&mm->mm_users, 1);
997	1060	atomic_set(&mm->mm_count, 1);
998		- init_rwsem(&mm->mmap_sem);
	1061	+ seqcount_init(&mm->write_protect_seq);
	1062	+ mmap_init_lock(mm);
999	1063	INIT_LIST_HEAD(&mm->mmlist);
1000	1064	mm->core_state = NULL;
1001	1065	mm_pgtables_bytes_init(mm);
1002	1066	mm->map_count = 0;
1003	1067	mm->locked_vm = 0;
1004		- mm->pinned_vm = 0;
	1068	+ atomic_set(&mm->has_pinned, 0);
	1069	+ atomic64_set(&mm->pinned_vm, 0);
1005	1070	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
1006	1071	spin_lock_init(&mm->page_table_lock);
1007	1072	spin_lock_init(&mm->arg_lock);
1008	1073	mm_init_cpumask(mm);
1009	1074	mm_init_aio(mm);
1010	1075	mm_init_owner(mm, p);
	1076	+ mm_init_pasid(mm);
1011	1077	RCU_INIT_POINTER(mm->exe_file, NULL);
1012		- mmu_notifier_mm_init(mm);
1013		- hmm_mm_init(mm);
	1078	+ if (!mmu_notifier_subscriptions_init(mm))
	1079	+ goto fail_nopgd;
1014	1080	init_tlb_flush_pending(mm);
1015	1081	#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
1016	1082	mm->pmd_huge_pte = NULL;
..	..	@@ -1085,8 +1151,10 @@
1085	1151	{
1086	1152	might_sleep();
1087	1153
1088		- if (atomic_dec_and_test(&mm->mm_users))
	1154	+ if (atomic_dec_and_test(&mm->mm_users)) {
	1155	+ trace_android_vh_mmput(NULL);
1089	1156	__mmput(mm);
	1157	+ }
1090	1158	}
1091	1159	EXPORT_SYMBOL_GPL(mmput);
1092	1160
..	..	@@ -1106,6 +1174,7 @@
1106	1174	schedule_work(&mm->async_put_work);
1107	1175	}
1108	1176	}
	1177	+EXPORT_SYMBOL_GPL(mmput_async);
1109	1178	#endif
1110	1179
1111	1180	/**
..	..	@@ -1210,7 +1279,7 @@
1210	1279	struct mm_struct *mm;
1211	1280	int err;
1212	1281
1213		- err = mutex_lock_killable(&task->signal->cred_guard_mutex);
	1282	+ err = down_read_killable(&task->signal->exec_update_lock);
1214	1283	if (err)
1215	1284	return ERR_PTR(err);
1216	1285
..	..	@@ -1220,7 +1289,7 @@
1220	1289	mmput(mm);
1221	1290	mm = ERR_PTR(-EACCES);
1222	1291	}
1223		- mutex_unlock(&task->signal->cred_guard_mutex);
	1292	+ up_read(&task->signal->exec_update_lock);
1224	1293
1225	1294	return mm;
1226	1295	}
..	..	@@ -1318,13 +1387,20 @@
1318	1387	mm_release(tsk, mm);
1319	1388	}
1320	1389
1321		-/*
1322		- * Allocate a new mm structure and copy contents from the
1323		- * mm structure of the passed in task structure.
	1390	+/**
	1391	+ * dup_mm() - duplicates an existing mm structure
	1392	+ * @tsk: the task_struct with which the new mm will be associated.
	1393	+ * @oldmm: the mm to duplicate.
	1394	+ *
	1395	+ * Allocates a new mm structure and duplicates the provided @oldmm structure
	1396	+ * content into it.
	1397	+ *
	1398	+ * Return: the duplicated mm or NULL on failure.
1324	1399	*/
1325		-static struct mm_struct dup_mm(struct task_struct tsk)
	1400	+static struct mm_struct dup_mm(struct task_struct tsk,
	1401	+ struct mm_struct *oldmm)
1326	1402	{
1327		- struct mm_struct mm, oldmm = current->mm;
	1403	+ struct mm_struct *mm;
1328	1404	int err;
1329	1405
1330	1406	mm = allocate_mm();
..	..	@@ -1392,7 +1468,7 @@
1392	1468	}
1393	1469
1394	1470	retval = -ENOMEM;
1395		- mm = dup_mm(tsk);
	1471	+ mm = dup_mm(tsk, current->mm);
1396	1472	if (!mm)
1397	1473	goto fail_nomem;
1398	1474
..	..	@@ -1442,7 +1518,7 @@
1442	1518	goto out;
1443	1519	}
1444	1520
1445		- newf = dup_fd(oldf, &error);
	1521	+ newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1446	1522	if (!newf)
1447	1523	goto out;
1448	1524
..	..	@@ -1483,24 +1559,29 @@
1483	1559	struct sighand_struct *sig;
1484	1560
1485	1561	if (clone_flags & CLONE_SIGHAND) {
1486		- atomic_inc(&current->sighand->count);
	1562	+ refcount_inc(&current->sighand->count);
1487	1563	return 0;
1488	1564	}
1489	1565	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1490		- rcu_assign_pointer(tsk->sighand, sig);
	1566	+ RCU_INIT_POINTER(tsk->sighand, sig);
1491	1567	if (!sig)
1492	1568	return -ENOMEM;
1493	1569
1494		- atomic_set(&sig->count, 1);
	1570	+ refcount_set(&sig->count, 1);
1495	1571	spin_lock_irq(&current->sighand->siglock);
1496	1572	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1497	1573	spin_unlock_irq(&current->sighand->siglock);
	1574	+
	1575	+ /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
	1576	+ if (clone_flags & CLONE_CLEAR_SIGHAND)
	1577	+ flush_signal_handlers(tsk, 0);
	1578	+
1498	1579	return 0;
1499	1580	}
1500	1581
1501	1582	void __cleanup_sighand(struct sighand_struct *sighand)
1502	1583	{
1503		- if (atomic_dec_and_test(&sighand->count)) {
	1584	+ if (refcount_dec_and_test(&sighand->count)) {
1504	1585	signalfd_cleanup(sighand);
1505	1586	/*
1506	1587	* sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
..	..	@@ -1510,28 +1591,17 @@
1510	1591	}
1511	1592	}
1512	1593
1513		-#ifdef CONFIG_POSIX_TIMERS
1514	1594	/*
1515	1595	* Initialize POSIX timer handling for a thread group.
1516	1596	*/
1517	1597	static void posix_cpu_timers_init_group(struct signal_struct *sig)
1518	1598	{
	1599	+ struct posix_cputimers *pct = &sig->posix_cputimers;
1519	1600	unsigned long cpu_limit;
1520	1601
1521	1602	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1522		- if (cpu_limit != RLIM_INFINITY) {
1523		- sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1524		- sig->cputimer.running = true;
1525		- }
1526		-
1527		- /* The timer lists. */
1528		- INIT_LIST_HEAD(&sig->cpu_timers[0]);
1529		- INIT_LIST_HEAD(&sig->cpu_timers[1]);
1530		- INIT_LIST_HEAD(&sig->cpu_timers[2]);
	1603	+ posix_cputimers_group_init(pct, cpu_limit);
1531	1604	}
1532		-#else
1533		-static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1534		-#endif
1535	1605
1536	1606	static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1537	1607	{
..	..	@@ -1547,7 +1617,7 @@
1547	1617
1548	1618	sig->nr_threads = 1;
1549	1619	atomic_set(&sig->live, 1);
1550		- atomic_set(&sig->sigcnt, 1);
	1620	+ refcount_set(&sig->sigcnt, 1);
1551	1621
1552	1622	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1553	1623	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
..	..	@@ -1579,6 +1649,7 @@
1579	1649	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1580	1650
1581	1651	mutex_init(&sig->cred_guard_mutex);
	1652	+ init_rwsem(&sig->exec_update_lock);
1582	1653
1583	1654	return 0;
1584	1655	}
..	..	@@ -1633,26 +1704,6 @@
1633	1704	#endif
1634	1705	}
1635	1706
1636		-#ifdef CONFIG_POSIX_TIMERS
1637		-/*
1638		- * Initialize POSIX timer handling for a single task.
1639		- */
1640		-static void posix_cpu_timers_init(struct task_struct *tsk)
1641		-{
1642		-#ifdef CONFIG_PREEMPT_RT_BASE
1643		- tsk->posix_timer_list = NULL;
1644		-#endif
1645		- tsk->cputime_expires.prof_exp = 0;
1646		- tsk->cputime_expires.virt_exp = 0;
1647		- tsk->cputime_expires.sched_exp = 0;
1648		- INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1649		- INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1650		- INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1651		-}
1652		-#else
1653		-static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1654		-#endif
1655		-
1656	1707	static inline void init_task_pid_links(struct task_struct *task)
1657	1708	{
1658	1709	enum pid_type type;
..	..	@@ -1684,7 +1735,125 @@
1684	1735	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1685	1736	p->rcu_tasks_idle_cpu = -1;
1686	1737	#endif /* #ifdef CONFIG_TASKS_RCU */
	1738	+#ifdef CONFIG_TASKS_TRACE_RCU
	1739	+ p->trc_reader_nesting = 0;
	1740	+ p->trc_reader_special.s = 0;
	1741	+ INIT_LIST_HEAD(&p->trc_holdout_list);
	1742	+#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1687	1743	}
	1744	+
	1745	+struct pid pidfd_pid(const struct file file)
	1746	+{
	1747	+ if (file->f_op == &pidfd_fops)
	1748	+ return file->private_data;
	1749	+
	1750	+ return ERR_PTR(-EBADF);
	1751	+}
	1752	+
	1753	+static int pidfd_release(struct inode inode, struct file file)
	1754	+{
	1755	+ struct pid *pid = file->private_data;
	1756	+
	1757	+ file->private_data = NULL;
	1758	+ put_pid(pid);
	1759	+ return 0;
	1760	+}
	1761	+
	1762	+#ifdef CONFIG_PROC_FS
	1763	+/**
	1764	+ * pidfd_show_fdinfo - print information about a pidfd
	1765	+ * @m: proc fdinfo file
	1766	+ * @f: file referencing a pidfd
	1767	+ *
	1768	+ * Pid:
	1769	+ * This function will print the pid that a given pidfd refers to in the
	1770	+ * pid namespace of the procfs instance.
	1771	+ * If the pid namespace of the process is not a descendant of the pid
	1772	+ * namespace of the procfs instance 0 will be shown as its pid. This is
	1773	+ * similar to calling getppid() on a process whose parent is outside of
	1774	+ * its pid namespace.
	1775	+ *
	1776	+ * NSpid:
	1777	+ * If pid namespaces are supported then this function will also print
	1778	+ * the pid of a given pidfd refers to for all descendant pid namespaces
	1779	+ * starting from the current pid namespace of the instance, i.e. the
	1780	+ * Pid field and the first entry in the NSpid field will be identical.
	1781	+ * If the pid namespace of the process is not a descendant of the pid
	1782	+ * namespace of the procfs instance 0 will be shown as its first NSpid
	1783	+ * entry and no others will be shown.
	1784	+ * Note that this differs from the Pid and NSpid fields in
	1785	+ * /proc/<pid>/status where Pid and NSpid are always shown relative to
	1786	+ * the pid namespace of the procfs instance. The difference becomes
	1787	+ * obvious when sending around a pidfd between pid namespaces from a
	1788	+ * different branch of the tree, i.e. where no ancestoral relation is
	1789	+ * present between the pid namespaces:
	1790	+ * - create two new pid namespaces ns1 and ns2 in the initial pid
	1791	+ * namespace (also take care to create new mount namespaces in the
	1792	+ * new pid namespace and mount procfs)
	1793	+ * - create a process with a pidfd in ns1
	1794	+ * - send pidfd from ns1 to ns2
	1795	+ * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
	1796	+ * have exactly one entry, which is 0
	1797	+ */
	1798	+static void pidfd_show_fdinfo(struct seq_file m, struct file f)
	1799	+{
	1800	+ struct pid *pid = f->private_data;
	1801	+ struct pid_namespace *ns;
	1802	+ pid_t nr = -1;
	1803	+
	1804	+ if (likely(pid_has_task(pid, PIDTYPE_PID))) {
	1805	+ ns = proc_pid_ns(file_inode(m->file)->i_sb);
	1806	+ nr = pid_nr_ns(pid, ns);
	1807	+ }
	1808	+
	1809	+ seq_put_decimal_ll(m, "Pid:\t", nr);
	1810	+
	1811	+#ifdef CONFIG_PID_NS
	1812	+ seq_put_decimal_ll(m, "\nNSpid:\t", nr);
	1813	+ if (nr > 0) {
	1814	+ int i;
	1815	+
	1816	+ /* If nr is non-zero it means that 'pid' is valid and that
	1817	+ * ns, i.e. the pid namespace associated with the procfs
	1818	+ * instance, is in the pid namespace hierarchy of pid.
	1819	+ * Start at one below the already printed level.
	1820	+ */
	1821	+ for (i = ns->level + 1; i <= pid->level; i++)
	1822	+ seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
	1823	+ }
	1824	+#endif
	1825	+ seq_putc(m, '\n');
	1826	+}
	1827	+#endif
	1828	+
	1829	+/*
	1830	+ * Poll support for process exit notification.
	1831	+ */
	1832	+static __poll_t pidfd_poll(struct file file, struct poll_table_struct pts)
	1833	+{
	1834	+ struct pid *pid = file->private_data;
	1835	+ __poll_t poll_flags = 0;
	1836	+
	1837	+ poll_wait(file, &pid->wait_pidfd, pts);
	1838	+
	1839	+ /*
	1840	+ * Inform pollers only when the whole thread group exits.
	1841	+ * If the thread group leader exits before all other threads in the
	1842	+ * group, then poll(2) should block, similar to the wait(2) family.
	1843	+ */
	1844	+ if (thread_group_exited(pid))
	1845	+ poll_flags = EPOLLIN \| EPOLLRDNORM;
	1846	+
	1847	+ return poll_flags;
	1848	+}
	1849	+
	1850	+const struct file_operations pidfd_fops = {
	1851	+ .release = pidfd_release,
	1852	+ .poll = pidfd_poll,
	1853	+#ifdef CONFIG_PROC_FS
	1854	+ .show_fdinfo = pidfd_show_fdinfo,
	1855	+#endif
	1856	+};
1688	1857
1689	1858	static void __delayed_free_task(struct rcu_head *rhp)
1690	1859	{
..	..	@@ -1699,84 +1868,6 @@
1699	1868	call_rcu(&tsk->rcu, __delayed_free_task);
1700	1869	else
1701	1870	free_task(tsk);
1702		-}
1703		-
1704		-static int pidfd_release(struct inode inode, struct file file)
1705		-{
1706		- struct pid *pid = file->private_data;
1707		-
1708		- file->private_data = NULL;
1709		- put_pid(pid);
1710		- return 0;
1711		-}
1712		-
1713		-#ifdef CONFIG_PROC_FS
1714		-static void pidfd_show_fdinfo(struct seq_file m, struct file f)
1715		-{
1716		- struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1717		- struct pid *pid = f->private_data;
1718		-
1719		- seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1720		- seq_putc(m, '\n');
1721		-}
1722		-#endif
1723		-
1724		-/*
1725		- * Poll support for process exit notification.
1726		- */
1727		-static __poll_t pidfd_poll(struct file file, struct poll_table_struct pts)
1728		-{
1729		- struct task_struct *task;
1730		- struct pid *pid = file->private_data;
1731		- __poll_t poll_flags = 0;
1732		-
1733		- poll_wait(file, &pid->wait_pidfd, pts);
1734		-
1735		- rcu_read_lock();
1736		- task = pid_task(pid, PIDTYPE_PID);
1737		- /*
1738		- * Inform pollers only when the whole thread group exits.
1739		- * If the thread group leader exits before all other threads in the
1740		- * group, then poll(2) should block, similar to the wait(2) family.
1741		- */
1742		- if (!task \|\| (task->exit_state && thread_group_empty(task)))
1743		- poll_flags = EPOLLIN \| EPOLLRDNORM;
1744		- rcu_read_unlock();
1745		-
1746		- return poll_flags;
1747		-}
1748		-
1749		-const struct file_operations pidfd_fops = {
1750		- .release = pidfd_release,
1751		- .poll = pidfd_poll,
1752		-#ifdef CONFIG_PROC_FS
1753		- .show_fdinfo = pidfd_show_fdinfo,
1754		-#endif
1755		-};
1756		-
1757		-/**
1758		- * pidfd_create() - Create a new pid file descriptor.
1759		- *
1760		- * @pid: struct pid that the pidfd will reference
1761		- *
1762		- * This creates a new pid file descriptor with the O_CLOEXEC flag set.
1763		- *
1764		- * Note, that this function can only be called after the fd table has
1765		- * been unshared to avoid leaking the pidfd to the new process.
1766		- *
1767		- * Return: On success, a cloexec pidfd is returned.
1768		- * On error, a negative errno number will be returned.
1769		- */
1770		-static int pidfd_create(struct pid *pid)
1771		-{
1772		- int fd;
1773		-
1774		- fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid),
1775		- O_RDWR \| O_CLOEXEC);
1776		- if (fd < 0)
1777		- put_pid(pid);
1778		-
1779		- return fd;
1780	1871	}
1781	1872
1782	1873	static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
..	..	@@ -1807,19 +1898,17 @@
1807	1898	* flags). The actual kick-off is left to the caller.
1808	1899	*/
1809	1900	static __latent_entropy struct task_struct *copy_process(
1810		- unsigned long clone_flags,
1811		- unsigned long stack_start,
1812		- unsigned long stack_size,
1813		- int __user *parent_tidptr,
1814		- int __user *child_tidptr,
1815	1901	struct pid *pid,
1816	1902	int trace,
1817		- unsigned long tls,
1818		- int node)
	1903	+ int node,
	1904	+ struct kernel_clone_args *args)
1819	1905	{
1820	1906	int pidfd = -1, retval;
1821	1907	struct task_struct *p;
1822	1908	struct multiprocess_signals delayed;
	1909	+ struct file *pidfile = NULL;
	1910	+ u64 clone_flags = args->flags;
	1911	+ struct nsproxy *nsp = current->nsproxy;
1823	1912
1824	1913	/*
1825	1914	* Don't allow sharing the root directory with processes in a different
..	..	@@ -1862,21 +1951,26 @@
1862	1951	*/
1863	1952	if (clone_flags & CLONE_THREAD) {
1864	1953	if ((clone_flags & (CLONE_NEWUSER \| CLONE_NEWPID)) \|\|
1865		- (task_active_pid_ns(current) !=
1866		- current->nsproxy->pid_ns_for_children))
	1954	+ (task_active_pid_ns(current) != nsp->pid_ns_for_children))
	1955	+ return ERR_PTR(-EINVAL);
	1956	+ }
	1957	+
	1958	+ /*
	1959	+ * If the new process will be in a different time namespace
	1960	+ * do not allow it to share VM or a thread group with the forking task.
	1961	+ */
	1962	+ if (clone_flags & (CLONE_THREAD \| CLONE_VM)) {
	1963	+ if (nsp->time_ns != nsp->time_ns_for_children)
1867	1964	return ERR_PTR(-EINVAL);
1868	1965	}
1869	1966
1870	1967	if (clone_flags & CLONE_PIDFD) {
1871	1968	/*
1872		- * - CLONE_PARENT_SETTID is useless for pidfds and also
1873		- * parent_tidptr is used to return pidfds.
1874	1969	* - CLONE_DETACHED is blocked so that we can potentially
1875	1970	* reuse it later for CLONE_PIDFD.
1876	1971	* - CLONE_THREAD is blocked until someone really needs it.
1877	1972	*/
1878		- if (clone_flags &
1879		- (CLONE_DETACHED \| CLONE_PARENT_SETTID \| CLONE_THREAD))
	1973	+ if (clone_flags & (CLONE_DETACHED \| CLONE_THREAD))
1880	1974	return ERR_PTR(-EINVAL);
1881	1975	}
1882	1976
..	..	@@ -1895,13 +1989,21 @@
1895	1989	recalc_sigpending();
1896	1990	spin_unlock_irq(&current->sighand->siglock);
1897	1991	retval = -ERESTARTNOINTR;
1898		- if (signal_pending(current))
	1992	+ if (task_sigpending(current))
1899	1993	goto fork_out;
1900	1994
1901	1995	retval = -ENOMEM;
1902	1996	p = dup_task_struct(current, node);
1903	1997	if (!p)
1904	1998	goto fork_out;
	1999	+ if (args->io_thread) {
	2000	+ /*
	2001	+ * Mark us an IO worker, and block any signal that isn't
	2002	+ * fatal or STOP
	2003	+ */
	2004	+ p->flags \|= PF_IO_WORKER;
	2005	+ siginitsetinv(&p->blocked, sigmask(SIGKILL)\|sigmask(SIGSTOP));
	2006	+ }
1905	2007
1906	2008	cpufreq_task_times_init(p);
1907	2009
..	..	@@ -1911,18 +2013,18 @@
1911	2013	* p->set_child_tid which is (ab)used as a kthread's data pointer for
1912	2014	* kernel threads (PF_KTHREAD).
1913	2015	*/
1914		- p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
	2016	+ p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1915	2017	/*
1916	2018	* Clear TID on mm_release()?
1917	2019	*/
1918		- p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
	2020	+ p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1919	2021
1920	2022	ftrace_graph_init_task(p);
1921	2023
1922	2024	rt_mutex_init_task(p);
1923	2025
	2026	+ lockdep_assert_irqs_enabled();
1924	2027	#ifdef CONFIG_PROVE_LOCKING
1925		- DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1926	2028	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1927	2029	#endif
1928	2030	retval = -EAGAIN;
..	..	@@ -1944,7 +2046,7 @@
1944	2046	* to stop root fork bombs.
1945	2047	*/
1946	2048	retval = -EAGAIN;
1947		- if (nr_threads >= max_threads)
	2049	+ if (data_race(nr_threads >= max_threads))
1948	2050	goto bad_fork_cleanup_count;
1949	2051
1950	2052	delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
..	..	@@ -1957,7 +2059,6 @@
1957	2059	spin_lock_init(&p->alloc_lock);
1958	2060
1959	2061	init_sigpending(&p->pending);
1960		- p->sigqueue_cache = NULL;
1961	2062
1962	2063	p->utime = p->stime = p->gtime = 0;
1963	2064	#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
..	..	@@ -1969,6 +2070,10 @@
1969	2070	seqcount_init(&p->vtime.seqcount);
1970	2071	p->vtime.starttime = 0;
1971	2072	p->vtime.state = VTIME_INACTIVE;
	2073	+#endif
	2074	+
	2075	+#ifdef CONFIG_IO_URING
	2076	+ p->io_uring = NULL;
1972	2077	#endif
1973	2078
1974	2079	#if defined(SPLIT_RSS_COUNTING)
..	..	@@ -1984,7 +2089,7 @@
1984	2089	task_io_accounting_init(&p->ioac);
1985	2090	acct_clear_integrals(p);
1986	2091
1987		- posix_cpu_timers_init(p);
	2092	+ posix_cputimers_init(&p->posix_cputimers);
1988	2093
1989	2094	p->io_context = NULL;
1990	2095	audit_set_context(p, NULL);
..	..	@@ -2000,30 +2105,19 @@
2000	2105	#ifdef CONFIG_CPUSETS
2001	2106	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
2002	2107	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
2003		- seqcount_init(&p->mems_allowed_seq);
	2108	+ seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
2004	2109	#endif
2005	2110	#ifdef CONFIG_TRACE_IRQFLAGS
2006		- p->irq_events = 0;
2007		- p->hardirqs_enabled = 0;
2008		- p->hardirq_enable_ip = 0;
2009		- p->hardirq_enable_event = 0;
2010		- p->hardirq_disable_ip = _THIS_IP_;
2011		- p->hardirq_disable_event = 0;
2012		- p->softirqs_enabled = 1;
2013		- p->softirq_enable_ip = _THIS_IP_;
2014		- p->softirq_enable_event = 0;
2015		- p->softirq_disable_ip = 0;
2016		- p->softirq_disable_event = 0;
2017		- p->hardirq_context = 0;
2018		- p->softirq_context = 0;
	2111	+ memset(&p->irqtrace, 0, sizeof(p->irqtrace));
	2112	+ p->irqtrace.hardirq_disable_ip = _THIS_IP_;
	2113	+ p->irqtrace.softirq_enable_ip = _THIS_IP_;
	2114	+ p->softirqs_enabled = 1;
	2115	+ p->softirq_context = 0;
2019	2116	#endif
2020	2117
2021	2118	p->pagefault_disabled = 0;
2022	2119
2023	2120	#ifdef CONFIG_LOCKDEP
2024		- p->lockdep_depth = 0; /* no locks held yet */
2025		- p->curr_chain_key = 0;
2026		- p->lockdep_recursion = 0;
2027	2121	lockdep_init_task(p);
2028	2122	#endif
2029	2123
..	..	@@ -2075,12 +2169,15 @@
2075	2169	retval = copy_io(clone_flags, p);
2076	2170	if (retval)
2077	2171	goto bad_fork_cleanup_namespaces;
2078		- retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
	2172	+ retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2079	2173	if (retval)
2080	2174	goto bad_fork_cleanup_io;
2081	2175
	2176	+ stackleak_task_init(p);
	2177	+
2082	2178	if (pid != &init_struct_pid) {
2083		- pid = alloc_pid(p->nsproxy->pid_ns_for_children);
	2179	+ pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
	2180	+ args->set_tid_size);
2084	2181	if (IS_ERR(pid)) {
2085	2182	retval = PTR_ERR(pid);
2086	2183	goto bad_fork_cleanup_thread;
..	..	@@ -2093,12 +2190,22 @@
2093	2190	* if the fd table isn't shared).
2094	2191	*/
2095	2192	if (clone_flags & CLONE_PIDFD) {
2096		- retval = pidfd_create(pid);
	2193	+ retval = get_unused_fd_flags(O_RDWR \| O_CLOEXEC);
2097	2194	if (retval < 0)
2098	2195	goto bad_fork_free_pid;
2099	2196
2100	2197	pidfd = retval;
2101		- retval = put_user(pidfd, parent_tidptr);
	2198	+
	2199	+ pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
	2200	+ O_RDWR \| O_CLOEXEC);
	2201	+ if (IS_ERR(pidfile)) {
	2202	+ put_unused_fd(pidfd);
	2203	+ retval = PTR_ERR(pidfile);
	2204	+ goto bad_fork_free_pid;
	2205	+ }
	2206	+ get_pid(pid); /* held by pidfile now */
	2207	+
	2208	+ retval = put_user(pidfd, args->pidfd);
2102	2209	if (retval)
2103	2210	goto bad_fork_put_pidfd;
2104	2211	}
..	..	@@ -2123,7 +2230,7 @@
2123	2230	#ifdef TIF_SYSCALL_EMU
2124	2231	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2125	2232	#endif
2126		- clear_all_latency_tracing(p);
	2233	+ clear_tsk_latency_tracing(p);
2127	2234
2128	2235	/* ok, now we should be set up.. */
2129	2236	p->pid = pid_nr(pid);
..	..	@@ -2142,17 +2249,28 @@
2142	2249	p->pdeath_signal = 0;
2143	2250	INIT_LIST_HEAD(&p->thread_group);
2144	2251	p->task_works = NULL;
	2252	+ clear_posix_cputimers_work(p);
2145	2253
2146		- cgroup_threadgroup_change_begin(current);
2147	2254	/*
2148	2255	* Ensure that the cgroup subsystem policies allow the new process to be
2149		- * forked. It should be noted the the new process's css_set can be changed
	2256	+ * forked. It should be noted that the new process's css_set can be changed
2150	2257	* between here and cgroup_post_fork() if an organisation operation is in
2151	2258	* progress.
2152	2259	*/
2153		- retval = cgroup_can_fork(p);
	2260	+ retval = cgroup_can_fork(p, args);
2154	2261	if (retval)
2155		- goto bad_fork_cgroup_threadgroup_change_end;
	2262	+ goto bad_fork_put_pidfd;
	2263	+
	2264	+ /*
	2265	+ * Now that the cgroups are pinned, re-clone the parent cgroup and put
	2266	+ * the new task on the correct runqueue. All this before the task
	2267	+ * becomes visible.
	2268	+ *
	2269	+ * This isn't part of ->can_fork() because while the re-cloning is
	2270	+ * cgroup specific, it unconditionally needs to place the task on a
	2271	+ * runqueue.
	2272	+ */
	2273	+ sched_cgroup_fork(p, args);
2156	2274
2157	2275	/*
2158	2276	* From this point on we must avoid any synchronous user-space
..	..	@@ -2163,7 +2281,7 @@
2163	2281	*/
2164	2282
2165	2283	p->start_time = ktime_get_ns();
2166		- p->real_start_time = ktime_get_boot_ns();
	2284	+ p->start_boottime = ktime_get_boottime_ns();
2167	2285
2168	2286	/*
2169	2287	* Make it visible to the rest of the system, but dont wake it up yet.
..	..	@@ -2182,7 +2300,7 @@
2182	2300	} else {
2183	2301	p->real_parent = current;
2184	2302	p->parent_exec_id = current->self_exec_id;
2185		- p->exit_signal = (clone_flags & CSIGNAL);
	2303	+ p->exit_signal = args->exit_signal;
2186	2304	}
2187	2305
2188	2306	klp_copy_process(p);
..	..	@@ -2208,7 +2326,6 @@
2208	2326	retval = -EINTR;
2209	2327	goto bad_fork_cancel_cgroup;
2210	2328	}
2211		-
2212	2329
2213	2330	init_task_pid_links(p);
2214	2331	if (likely(p->pid)) {
..	..	@@ -2242,7 +2359,7 @@
2242	2359	} else {
2243	2360	current->signal->nr_threads++;
2244	2361	atomic_inc(&current->signal->live);
2245		- atomic_inc(&current->signal->sigcnt);
	2362	+ refcount_inc(&current->signal->sigcnt);
2246	2363	task_join_group_stop(p);
2247	2364	list_add_tail_rcu(&p->thread_group,
2248	2365	&p->group_leader->thread_group);
..	..	@@ -2258,9 +2375,12 @@
2258	2375	syscall_tracepoint_update(p);
2259	2376	write_unlock_irq(&tasklist_lock);
2260	2377
	2378	+ if (pidfile)
	2379	+ fd_install(pidfd, pidfile);
	2380	+
2261	2381	proc_fork_connector(p);
2262		- cgroup_post_fork(p);
2263		- cgroup_threadgroup_change_end(current);
	2382	+ sched_post_fork(p);
	2383	+ cgroup_post_fork(p, args);
2264	2384	perf_event_fork(p);
2265	2385
2266	2386	trace_task_newtask(p, clone_flags);
..	..	@@ -2273,12 +2393,12 @@
2273	2393	bad_fork_cancel_cgroup:
2274	2394	spin_unlock(&current->sighand->siglock);
2275	2395	write_unlock_irq(&tasklist_lock);
2276		- cgroup_cancel_fork(p);
2277		-bad_fork_cgroup_threadgroup_change_end:
2278		- cgroup_threadgroup_change_end(current);
	2396	+ cgroup_cancel_fork(p, args);
2279	2397	bad_fork_put_pidfd:
2280		- if (clone_flags & CLONE_PIDFD)
2281		- ksys_close(pidfd);
	2398	+ if (clone_flags & CLONE_PIDFD) {
	2399	+ fput(pidfile);
	2400	+ put_unused_fd(pidfd);
	2401	+ }
2282	2402	bad_fork_free_pid:
2283	2403	if (pid != &init_struct_pid)
2284	2404	free_pid(pid);
..	..	@@ -2342,11 +2462,14 @@
2342	2462	}
2343	2463	}
2344	2464
2345		-struct task_struct *fork_idle(int cpu)
	2465	+struct task_struct * __init fork_idle(int cpu)
2346	2466	{
2347	2467	struct task_struct *task;
2348		- task = copy_process(CLONE_VM, 0, 0, NULL, NULL, &init_struct_pid, 0, 0,
2349		- cpu_to_node(cpu));
	2468	+ struct kernel_clone_args args = {
	2469	+ .flags = CLONE_VM,
	2470	+ };
	2471	+
	2472	+ task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2350	2473	if (!IS_ERR(task)) {
2351	2474	init_idle_pids(task);
2352	2475	init_idle(task, cpu);
..	..	@@ -2355,24 +2478,63 @@
2355	2478	return task;
2356	2479	}
2357	2480
	2481	+struct mm_struct *copy_init_mm(void)
	2482	+{
	2483	+ return dup_mm(NULL, &init_mm);
	2484	+}
	2485	+
	2486	+/*
	2487	+ * This is like kernel_clone(), but shaved down and tailored to just
	2488	+ * creating io_uring workers. It returns a created task, or an error pointer.
	2489	+ * The returned task is inactive, and the caller must fire it up through
	2490	+ * wake_up_new_task(p). All signals are blocked in the created task.
	2491	+ */
	2492	+struct task_struct create_io_thread(int (fn)(void ), void arg, int node)
	2493	+{
	2494	+ unsigned long flags = CLONE_FS\|CLONE_FILES\|CLONE_SIGHAND\|CLONE_THREAD\|
	2495	+ CLONE_IO;
	2496	+ struct kernel_clone_args args = {
	2497	+ .flags = ((lower_32_bits(flags) \| CLONE_VM \|
	2498	+ CLONE_UNTRACED) & ~CSIGNAL),
	2499	+ .exit_signal = (lower_32_bits(flags) & CSIGNAL),
	2500	+ .stack = (unsigned long)fn,
	2501	+ .stack_size = (unsigned long)arg,
	2502	+ .io_thread = 1,
	2503	+ };
	2504	+
	2505	+ return copy_process(NULL, 0, node, &args);
	2506	+}
	2507	+
2358	2508	/*
2359	2509	* Ok, this is the main fork-routine.
2360	2510	*
2361	2511	* It copies the process, and if successful kick-starts
2362	2512	* it and waits for it to finish using the VM if required.
	2513	+ *
	2514	+ * args->exit_signal is expected to be checked for sanity by the caller.
2363	2515	*/
2364		-long _do_fork(unsigned long clone_flags,
2365		- unsigned long stack_start,
2366		- unsigned long stack_size,
2367		- int __user *parent_tidptr,
2368		- int __user *child_tidptr,
2369		- unsigned long tls)
	2516	+pid_t kernel_clone(struct kernel_clone_args *args)
2370	2517	{
	2518	+ u64 clone_flags = args->flags;
2371	2519	struct completion vfork;
2372	2520	struct pid *pid;
2373	2521	struct task_struct *p;
2374	2522	int trace = 0;
2375		- long nr;
	2523	+ pid_t nr;
	2524	+
	2525	+ /*
	2526	+ * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
	2527	+ * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
	2528	+ * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
	2529	+ * field in struct clone_args and it still doesn't make sense to have
	2530	+ * them both point at the same memory location. Performing this check
	2531	+ * here has the advantage that we don't need to have a separate helper
	2532	+ * to check for legacy clone().
	2533	+ */
	2534	+ if ((args->flags & CLONE_PIDFD) &&
	2535	+ (args->flags & CLONE_PARENT_SETTID) &&
	2536	+ (args->pidfd == args->parent_tid))
	2537	+ return -EINVAL;
2376	2538
2377	2539	/*
2378	2540	* Determine whether and which event to report to ptracer. When
..	..	@@ -2383,7 +2545,7 @@
2383	2545	if (!(clone_flags & CLONE_UNTRACED)) {
2384	2546	if (clone_flags & CLONE_VFORK)
2385	2547	trace = PTRACE_EVENT_VFORK;
2386		- else if ((clone_flags & CSIGNAL) != SIGCHLD)
	2548	+ else if (args->exit_signal != SIGCHLD)
2387	2549	trace = PTRACE_EVENT_CLONE;
2388	2550	else
2389	2551	trace = PTRACE_EVENT_FORK;
..	..	@@ -2392,8 +2554,7 @@
2392	2554	trace = 0;
2393	2555	}
2394	2556
2395		- p = copy_process(clone_flags, stack_start, stack_size, parent_tidptr,
2396		- child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
	2557	+ p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2397	2558	add_latent_entropy();
2398	2559
2399	2560	if (IS_ERR(p))
..	..	@@ -2411,7 +2572,7 @@
2411	2572	nr = pid_vnr(pid);
2412	2573
2413	2574	if (clone_flags & CLONE_PARENT_SETTID)
2414		- put_user(nr, parent_tidptr);
	2575	+ put_user(nr, args->parent_tid);
2415	2576
2416	2577	if (clone_flags & CLONE_VFORK) {
2417	2578	p->vfork_done = &vfork;
..	..	@@ -2434,34 +2595,31 @@
2434	2595	return nr;
2435	2596	}
2436	2597
2437		-#ifndef CONFIG_HAVE_COPY_THREAD_TLS
2438		-/* For compatibility with architectures that call do_fork directly rather than
2439		- * using the syscall entry points below. */
2440		-long do_fork(unsigned long clone_flags,
2441		- unsigned long stack_start,
2442		- unsigned long stack_size,
2443		- int __user *parent_tidptr,
2444		- int __user *child_tidptr)
2445		-{
2446		- return _do_fork(clone_flags, stack_start, stack_size,
2447		- parent_tidptr, child_tidptr, 0);
2448		-}
2449		-#endif
2450		-
2451	2598	/*
2452	2599	* Create a kernel thread.
2453	2600	*/
2454	2601	pid_t kernel_thread(int (fn)(void ), void *arg, unsigned long flags)
2455	2602	{
2456		- return _do_fork(flags\|CLONE_VM\|CLONE_UNTRACED, (unsigned long)fn,
2457		- (unsigned long)arg, NULL, NULL, 0);
	2603	+ struct kernel_clone_args args = {
	2604	+ .flags = ((lower_32_bits(flags) \| CLONE_VM \|
	2605	+ CLONE_UNTRACED) & ~CSIGNAL),
	2606	+ .exit_signal = (lower_32_bits(flags) & CSIGNAL),
	2607	+ .stack = (unsigned long)fn,
	2608	+ .stack_size = (unsigned long)arg,
	2609	+ };
	2610	+
	2611	+ return kernel_clone(&args);
2458	2612	}
2459	2613
2460	2614	#ifdef __ARCH_WANT_SYS_FORK
2461	2615	SYSCALL_DEFINE0(fork)
2462	2616	{
2463	2617	#ifdef CONFIG_MMU
2464		- return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
	2618	+ struct kernel_clone_args args = {
	2619	+ .exit_signal = SIGCHLD,
	2620	+ };
	2621	+
	2622	+ return kernel_clone(&args);
2465	2623	#else
2466	2624	/* can not support in nommu mode */
2467	2625	return -EINVAL;
..	..	@@ -2472,8 +2630,12 @@
2472	2630	#ifdef __ARCH_WANT_SYS_VFORK
2473	2631	SYSCALL_DEFINE0(vfork)
2474	2632	{
2475		- return _do_fork(CLONE_VFORK \| CLONE_VM \| SIGCHLD, 0,
2476		- 0, NULL, NULL, 0);
	2633	+ struct kernel_clone_args args = {
	2634	+ .flags = CLONE_VFORK \| CLONE_VM,
	2635	+ .exit_signal = SIGCHLD,
	2636	+ };
	2637	+
	2638	+ return kernel_clone(&args);
2477	2639	}
2478	2640	#endif
2479	2641
..	..	@@ -2501,7 +2663,175 @@
2501	2663	unsigned long, tls)
2502	2664	#endif
2503	2665	{
2504		- return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
	2666	+ struct kernel_clone_args args = {
	2667	+ .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
	2668	+ .pidfd = parent_tidptr,
	2669	+ .child_tid = child_tidptr,
	2670	+ .parent_tid = parent_tidptr,
	2671	+ .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
	2672	+ .stack = newsp,
	2673	+ .tls = tls,
	2674	+ };
	2675	+
	2676	+ return kernel_clone(&args);
	2677	+}
	2678	+#endif
	2679	+
	2680	+#ifdef __ARCH_WANT_SYS_CLONE3
	2681	+
	2682	+noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
	2683	+ struct clone_args __user *uargs,
	2684	+ size_t usize)
	2685	+{
	2686	+ int err;
	2687	+ struct clone_args args;
	2688	+ pid_t *kset_tid = kargs->set_tid;
	2689	+
	2690	+ BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
	2691	+ CLONE_ARGS_SIZE_VER0);
	2692	+ BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
	2693	+ CLONE_ARGS_SIZE_VER1);
	2694	+ BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
	2695	+ CLONE_ARGS_SIZE_VER2);
	2696	+ BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
	2697	+
	2698	+ if (unlikely(usize > PAGE_SIZE))
	2699	+ return -E2BIG;
	2700	+ if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
	2701	+ return -EINVAL;
	2702	+
	2703	+ err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
	2704	+ if (err)
	2705	+ return err;
	2706	+
	2707	+ if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
	2708	+ return -EINVAL;
	2709	+
	2710	+ if (unlikely(!args.set_tid && args.set_tid_size > 0))
	2711	+ return -EINVAL;
	2712	+
	2713	+ if (unlikely(args.set_tid && args.set_tid_size == 0))
	2714	+ return -EINVAL;
	2715	+
	2716	+ /*
	2717	+ * Verify that higher 32bits of exit_signal are unset and that
	2718	+ * it is a valid signal
	2719	+ */
	2720	+ if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) \|\|
	2721	+ !valid_signal(args.exit_signal)))
	2722	+ return -EINVAL;
	2723	+
	2724	+ if ((args.flags & CLONE_INTO_CGROUP) &&
	2725	+ (args.cgroup > INT_MAX \|\| usize < CLONE_ARGS_SIZE_VER2))
	2726	+ return -EINVAL;
	2727	+
	2728	+ *kargs = (struct kernel_clone_args){
	2729	+ .flags = args.flags,
	2730	+ .pidfd = u64_to_user_ptr(args.pidfd),
	2731	+ .child_tid = u64_to_user_ptr(args.child_tid),
	2732	+ .parent_tid = u64_to_user_ptr(args.parent_tid),
	2733	+ .exit_signal = args.exit_signal,
	2734	+ .stack = args.stack,
	2735	+ .stack_size = args.stack_size,
	2736	+ .tls = args.tls,
	2737	+ .set_tid_size = args.set_tid_size,
	2738	+ .cgroup = args.cgroup,
	2739	+ };
	2740	+
	2741	+ if (args.set_tid &&
	2742	+ copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
	2743	+ (kargs->set_tid_size * sizeof(pid_t))))
	2744	+ return -EFAULT;
	2745	+
	2746	+ kargs->set_tid = kset_tid;
	2747	+
	2748	+ return 0;
	2749	+}
	2750	+
	2751	+/**
	2752	+ * clone3_stack_valid - check and prepare stack
	2753	+ * @kargs: kernel clone args
	2754	+ *
	2755	+ * Verify that the stack arguments userspace gave us are sane.
	2756	+ * In addition, set the stack direction for userspace since it's easy for us to
	2757	+ * determine.
	2758	+ */
	2759	+static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
	2760	+{
	2761	+ if (kargs->stack == 0) {
	2762	+ if (kargs->stack_size > 0)
	2763	+ return false;
	2764	+ } else {
	2765	+ if (kargs->stack_size == 0)
	2766	+ return false;
	2767	+
	2768	+ if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
	2769	+ return false;
	2770	+
	2771	+#if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
	2772	+ kargs->stack += kargs->stack_size;
	2773	+#endif
	2774	+ }
	2775	+
	2776	+ return true;
	2777	+}
	2778	+
	2779	+static bool clone3_args_valid(struct kernel_clone_args *kargs)
	2780	+{
	2781	+ /* Verify that no unknown flags are passed along. */
	2782	+ if (kargs->flags &
	2783	+ ~(CLONE_LEGACY_FLAGS \| CLONE_CLEAR_SIGHAND \| CLONE_INTO_CGROUP))
	2784	+ return false;
	2785	+
	2786	+ /*
	2787	+ * - make the CLONE_DETACHED bit reuseable for clone3
	2788	+ * - make the CSIGNAL bits reuseable for clone3
	2789	+ */
	2790	+ if (kargs->flags & (CLONE_DETACHED \| CSIGNAL))
	2791	+ return false;
	2792	+
	2793	+ if ((kargs->flags & (CLONE_SIGHAND \| CLONE_CLEAR_SIGHAND)) ==
	2794	+ (CLONE_SIGHAND \| CLONE_CLEAR_SIGHAND))
	2795	+ return false;
	2796	+
	2797	+ if ((kargs->flags & (CLONE_THREAD \| CLONE_PARENT)) &&
	2798	+ kargs->exit_signal)
	2799	+ return false;
	2800	+
	2801	+ if (!clone3_stack_valid(kargs))
	2802	+ return false;
	2803	+
	2804	+ return true;
	2805	+}
	2806	+
	2807	+/**
	2808	+ * clone3 - create a new process with specific properties
	2809	+ * @uargs: argument structure
	2810	+ * @size: size of @uargs
	2811	+ *
	2812	+ * clone3() is the extensible successor to clone()/clone2().
	2813	+ * It takes a struct as argument that is versioned by its size.
	2814	+ *
	2815	+ * Return: On success, a positive PID for the child process.
	2816	+ * On error, a negative errno number.
	2817	+ */
	2818	+SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
	2819	+{
	2820	+ int err;
	2821	+
	2822	+ struct kernel_clone_args kargs;
	2823	+ pid_t set_tid[MAX_PID_NS_LEVEL];
	2824	+
	2825	+ kargs.set_tid = set_tid;
	2826	+
	2827	+ err = copy_clone_args_from_user(&kargs, uargs, size);
	2828	+ if (err)
	2829	+ return err;
	2830	+
	2831	+ if (!clone3_args_valid(&kargs))
	2832	+ return -EINVAL;
	2833	+
	2834	+ return kernel_clone(&kargs);
2505	2835	}
2506	2836	#endif
2507	2837
..	..	@@ -2596,7 +2926,8 @@
2596	2926	if (unshare_flags & ~(CLONE_THREAD\|CLONE_FS\|CLONE_NEWNS\|CLONE_SIGHAND\|
2597	2927	CLONE_VM\|CLONE_FILES\|CLONE_SYSVSEM\|
2598	2928	CLONE_NEWUTS\|CLONE_NEWIPC\|CLONE_NEWNET\|
2599		- CLONE_NEWUSER\|CLONE_NEWPID\|CLONE_NEWCGROUP))
	2929	+ CLONE_NEWUSER\|CLONE_NEWPID\|CLONE_NEWCGROUP\|
	2930	+ CLONE_NEWTIME))
2600	2931	return -EINVAL;
2601	2932	/*
2602	2933	* Not implemented, but pretend it works if there is nothing
..	..	@@ -2609,7 +2940,7 @@
2609	2940	return -EINVAL;
2610	2941	}
2611	2942	if (unshare_flags & (CLONE_SIGHAND \| CLONE_VM)) {
2612		- if (atomic_read(&current->sighand->count) > 1)
	2943	+ if (refcount_read(&current->sighand->count) > 1)
2613	2944	return -EINVAL;
2614	2945	}
2615	2946	if (unshare_flags & CLONE_VM) {
..	..	@@ -2644,14 +2975,15 @@
2644	2975	/*
2645	2976	* Unshare file descriptor table if it is being shared
2646	2977	*/
2647		-static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
	2978	+int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
	2979	+ struct files_struct **new_fdp)
2648	2980	{
2649	2981	struct files_struct *fd = current->files;
2650	2982	int error = 0;
2651	2983
2652	2984	if ((unshare_flags & CLONE_FILES) &&
2653	2985	(fd && atomic_read(&fd->count) > 1)) {
2654		- *new_fdp = dup_fd(fd, &error);
	2986	+ *new_fdp = dup_fd(fd, max_fds, &error);
2655	2987	if (!*new_fdp)
2656	2988	return error;
2657	2989	}
..	..	@@ -2662,7 +2994,7 @@
2662	2994	/*
2663	2995	* unshare allows a process to 'unshare' part of the process
2664	2996	* context which was originally shared using clone. copy_*
2665		- * functions used by do_fork() cannot be used here directly
	2997	+ * functions used by kernel_clone() cannot be used here directly
2666	2998	* because they modify an inactive task_struct that is being
2667	2999	* constructed. Here we are modifying the current, active,
2668	3000	* task_struct.
..	..	@@ -2711,7 +3043,7 @@
2711	3043	err = unshare_fs(unshare_flags, &new_fs);
2712	3044	if (err)
2713	3045	goto bad_unshare_out;
2714		- err = unshare_fd(unshare_flags, &new_fd);
	3046	+ err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2715	3047	if (err)
2716	3048	goto bad_unshare_cleanup_fs;
2717	3049	err = unshare_userns(unshare_flags, &new_cred);
..	..	@@ -2800,7 +3132,7 @@
2800	3132	struct files_struct *copy = NULL;
2801	3133	int error;
2802	3134
2803		- error = unshare_fd(CLONE_FILES, &copy);
	3135	+ error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
2804	3136	if (error \|\| !copy) {
2805	3137	*displaced = NULL;
2806	3138	return error;
..	..	@@ -2813,7 +3145,7 @@
2813	3145	}
2814	3146
2815	3147	int sysctl_max_threads(struct ctl_table *table, int write,
2816		- void __user buffer, size_t lenp, loff_t *ppos)
	3148	+ void buffer, size_t lenp, loff_t *ppos)
2817	3149	{
2818	3150	struct ctl_table t;
2819	3151	int ret;