~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	*
..	..	@@ -39,7 +40,6 @@
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	45	#include <linux/vmacache.h>
..	..	@@ -79,7 +79,6 @@
79	79	#include <linux/blkdev.h>
80	80	#include <linux/fs_struct.h>
81	81	#include <linux/magic.h>
82		-#include <linux/sched/mm.h>
83	82	#include <linux/perf_event.h>
84	83	#include <linux/posix-timers.h>
85	84	#include <linux/user-return-notifier.h>
..	..	@@ -93,10 +92,12 @@
93	92	#include <linux/kcov.h>
94	93	#include <linux/livepatch.h>
95	94	#include <linux/thread_info.h>
96		-#include <linux/cpufreq_times.h>
	95	+#include <linux/stackleak.h>
	96	+#include <linux/kasan.h>
97	97	#include <linux/scs.h>
	98	+#include <linux/io_uring.h>
	99	+#include <linux/cpufreq_times.h>
98	100
99		-#include <asm/pgtable.h>
100	101	#include <asm/pgalloc.h>
101	102	#include <linux/uaccess.h>
102	103	#include <asm/mmu_context.h>
..	..	@@ -108,6 +109,8 @@
108	109	#define CREATE_TRACE_POINTS
109	110	#include <trace/events/task.h>
110	111
	112	+#undef CREATE_TRACE_POINTS
	113	+#include <trace/hooks/sched.h>
111	114	/*
112	115	* Minimum number of threads to boot the kernel
113	116	*/
..	..	@@ -118,17 +121,29 @@
118	121	*/
119	122	#define MAX_THREADS FUTEX_TID_MASK
120	123
	124	+EXPORT_TRACEPOINT_SYMBOL_GPL(task_newtask);
	125	+
121	126	/*
122	127	* Protected counters by write_lock_irq(&tasklist_lock)
123	128	*/
124	129	unsigned long total_forks; /* Handle normal Linux uptimes. */
125	130	int nr_threads; /* The idle threads do not count.. */
126	131
127		-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	+};
128	142
129	143	DEFINE_PER_CPU(unsigned long, process_counts) = 0;
130	144
131	145	__cacheline_aligned DEFINE_RWLOCK(tasklist_lock); /* outer */
	146	+EXPORT_SYMBOL_GPL(tasklist_lock);
132	147
133	148	#ifdef CONFIG_PROVE_RCU
134	149	int lockdep_tasklist_lock_is_held(void)
..	..	@@ -216,6 +231,9 @@
216	231	if (!s)
217	232	continue;
218	233
	234	+ /* Mark stack accessible for KASAN. */
	235	+ kasan_unpoison_range(s->addr, THREAD_SIZE);
	236	+
219	237	/* Clear stale pointers from reused stack. */
220	238	memset(s->addr, 0, THREAD_SIZE);
221	239
..	..	@@ -224,9 +242,14 @@
224	242	return s->addr;
225	243	}
226	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	+ */
227	250	stack = __vmalloc_node_range(THREAD_SIZE, THREAD_ALIGN,
228	251	VMALLOC_START, VMALLOC_END,
229		- THREADINFO_GFP,
	252	+ THREADINFO_GFP & ~__GFP_ACCOUNT,
230	253	PAGE_KERNEL,
231	254	0, node, __builtin_return_address(0));
232	255
..	..	@@ -245,7 +268,7 @@
245	268	THREAD_SIZE_ORDER);
246	269
247	270	if (likely(page)) {
248		- tsk->stack = page_address(page);
	271	+ tsk->stack = kasan_reset_tag(page_address(page));
249	272	return tsk->stack;
250	273	}
251	274	return NULL;
..	..	@@ -255,8 +278,13 @@
255	278	static inline void free_thread_stack(struct task_struct *tsk)
256	279	{
257	280	#ifdef CONFIG_VMAP_STACK
258		- if (task_stack_vm_area(tsk)) {
	281	+ struct vm_struct *vm = task_stack_vm_area(tsk);
	282	+
	283	+ if (vm) {
259	284	int i;
	285	+
	286	+ for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++)
	287	+ memcg_kmem_uncharge_page(vm->pages[i], 0);
260	288
261	289	for (i = 0; i < NR_CACHED_STACKS; i++) {
262	290	if (this_cpu_cmpxchg(cached_stacks[i],
..	..	@@ -281,6 +309,7 @@
281	309	{
282	310	unsigned long *stack;
283	311	stack = kmem_cache_alloc_node(thread_stack_cache, THREADINFO_GFP, node);
	312	+ stack = kasan_reset_tag(stack);
284	313	tsk->stack = stack;
285	314	return stack;
286	315	}
..	..	@@ -333,8 +362,15 @@
333	362	struct vm_area_struct *new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL);
334	363
335	364	if (new) {
336		- new = orig;
337		- 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;
338	374	}
339	375	return new;
340	376	}
..	..	@@ -349,6 +385,22 @@
349	385	void *stack = task_stack_page(tsk);
350	386	struct vm_struct *vm = task_stack_vm_area(tsk);
351	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	+
352	404	BUILD_BUG_ON(IS_ENABLED(CONFIG_VMAP_STACK) && PAGE_SIZE % 1024 != 0);
353	405
354	406	if (vm) {
..	..	@@ -357,27 +409,19 @@
357	409	BUG_ON(vm->nr_pages != THREAD_SIZE / PAGE_SIZE);
358	410
359	411	for (i = 0; i < THREAD_SIZE / PAGE_SIZE; i++) {
360		- mod_zone_page_state(page_zone(vm->pages[i]),
361		- NR_KERNEL_STACK_KB,
362		- 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;
363	421	}
364		-
365		- /* All stack pages belong to the same memcg. */
366		- mod_memcg_page_state(vm->pages[0], MEMCG_KERNEL_STACK_KB,
367		- account * (THREAD_SIZE / 1024));
368		- } else {
369		- /*
370		- * All stack pages are in the same zone and belong to the
371		- * same memcg.
372		- */
373		- struct page *first_page = virt_to_page(stack);
374		-
375		- mod_zone_page_state(page_zone(first_page), NR_KERNEL_STACK_KB,
376		- THREAD_SIZE / 1024 * account);
377		-
378		- mod_memcg_page_state(first_page, MEMCG_KERNEL_STACK_KB,
379		- account * (THREAD_SIZE / 1024));
380	422	}
	423	+#endif
	424	+ return 0;
381	425	}
382	426
383	427	static void release_task_stack(struct task_struct *tsk)
..	..	@@ -396,16 +440,21 @@
396	440	#ifdef CONFIG_THREAD_INFO_IN_TASK
397	441	void put_task_stack(struct task_struct *tsk)
398	442	{
399		- if (atomic_dec_and_test(&tsk->stack_refcount))
	443	+ if (refcount_dec_and_test(&tsk->stack_refcount))
400	444	release_task_stack(tsk);
401	445	}
	446	+EXPORT_SYMBOL_GPL(put_task_stack);
402	447	#endif
403	448
404	449	void free_task(struct task_struct *tsk)
405	450	{
	451	+#ifdef CONFIG_SECCOMP
	452	+ WARN_ON_ONCE(tsk->seccomp.filter);
	453	+#endif
406	454	cpufreq_task_times_exit(tsk);
407	455	scs_release(tsk);
408	456
	457	+ trace_android_vh_free_task(tsk);
409	458	#ifndef CONFIG_THREAD_INFO_IN_TASK
410	459	/*
411	460	* The task is finally done with both the stack and thread_info,
..	..	@@ -417,11 +466,10 @@
417	466	* If the task had a separate stack allocation, it should be gone
418	467	* by now.
419	468	*/
420		- WARN_ON_ONCE(atomic_read(&tsk->stack_refcount) != 0);
	469	+ WARN_ON_ONCE(refcount_read(&tsk->stack_refcount) != 0);
421	470	#endif
422	471	rt_mutex_debug_task_free(tsk);
423	472	ftrace_graph_exit_task(tsk);
424		- put_seccomp_filter(tsk);
425	473	arch_release_task_struct(tsk);
426	474	if (tsk->flags & PF_KTHREAD)
427	475	free_kthread_struct(tsk);
..	..	@@ -433,14 +481,14 @@
433	481	static __latent_entropy int dup_mmap(struct mm_struct *mm,
434	482	struct mm_struct *oldmm)
435	483	{
436		- struct vm_area_struct mpnt, tmp, prev, *pprev;
	484	+ struct vm_area_struct mpnt, tmp, prev, pprev, last = NULL;
437	485	struct rb_node *rb_link, rb_parent;
438	486	int retval;
439	487	unsigned long charge;
440	488	LIST_HEAD(uf);
441	489
442	490	uprobe_start_dup_mmap();
443		- if (down_write_killable(&oldmm->mmap_sem)) {
	491	+ if (mmap_write_lock_killable(oldmm)) {
444	492	retval = -EINTR;
445	493	goto fail_uprobe_end;
446	494	}
..	..	@@ -449,7 +497,7 @@
449	497	/*
450	498	* Not linked in yet - no deadlock potential:
451	499	*/
452		- down_write_nested(&mm->mmap_sem, SINGLE_DEPTH_NESTING);
	500	+ mmap_write_lock_nested(mm, SINGLE_DEPTH_NESTING);
453	501
454	502	/* No ordering required: file already has been exposed. */
455	503	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
..	..	@@ -504,14 +552,15 @@
504	552	if (retval)
505	553	goto fail_nomem_anon_vma_fork;
506	554	if (tmp->vm_flags & VM_WIPEONFORK) {
507		- /* VM_WIPEONFORK gets a clean slate in the child. */
	555	+ /*
	556	+ * VM_WIPEONFORK gets a clean slate in the child.
	557	+ * Don't prepare anon_vma until fault since we don't
	558	+ * copy page for current vma.
	559	+ */
508	560	tmp->anon_vma = NULL;
509		- if (anon_vma_prepare(tmp))
510		- goto fail_nomem_anon_vma_fork;
511	561	} else if (anon_vma_fork(tmp, mpnt))
512	562	goto fail_nomem_anon_vma_fork;
513	563	tmp->vm_flags &= ~(VM_LOCKED \| VM_LOCKONFAULT);
514		- tmp->vm_next = tmp->vm_prev = NULL;
515	564	file = tmp->vm_file;
516	565	if (file) {
517	566	struct inode *inode = file_inode(file);
..	..	@@ -519,10 +568,10 @@
519	568
520	569	get_file(file);
521	570	if (tmp->vm_flags & VM_DENYWRITE)
522		- atomic_dec(&inode->i_writecount);
	571	+ put_write_access(inode);
523	572	i_mmap_lock_write(mapping);
524	573	if (tmp->vm_flags & VM_SHARED)
525		- atomic_inc(&mapping->i_mmap_writable);
	574	+ mapping_allow_writable(mapping);
526	575	flush_dcache_mmap_lock(mapping);
527	576	/* insert tmp into the share list, just after mpnt */
528	577	vma_interval_tree_insert_after(tmp, mpnt,
..	..	@@ -552,8 +601,18 @@
552	601	rb_parent = &tmp->vm_rb;
553	602
554	603	mm->map_count++;
555		- if (!(tmp->vm_flags & VM_WIPEONFORK))
556		- retval = copy_page_range(mm, oldmm, mpnt);
	604	+ if (!(tmp->vm_flags & VM_WIPEONFORK)) {
	605	+ if (IS_ENABLED(CONFIG_SPECULATIVE_PAGE_FAULT)) {
	606	+ /*
	607	+ * Mark this VMA as changing to prevent the
	608	+ * speculative page fault hanlder to process
	609	+ * it until the TLB are flushed below.
	610	+ */
	611	+ last = mpnt;
	612	+ vm_write_begin(mpnt);
	613	+ }
	614	+ retval = copy_page_range(tmp, mpnt);
	615	+ }
557	616
558	617	if (tmp->vm_ops && tmp->vm_ops->open)
559	618	tmp->vm_ops->open(tmp);
..	..	@@ -564,9 +623,25 @@
564	623	/* a new mm has just been created */
565	624	retval = arch_dup_mmap(oldmm, mm);
566	625	out:
567		- up_write(&mm->mmap_sem);
	626	+ mmap_write_unlock(mm);
568	627	flush_tlb_mm(oldmm);
569		- up_write(&oldmm->mmap_sem);
	628	+
	629	+ if (IS_ENABLED(CONFIG_SPECULATIVE_PAGE_FAULT)) {
	630	+ /*
	631	+ * Since the TLB has been flush, we can safely unmark the
	632	+ * copied VMAs and allows the speculative page fault handler to
	633	+ * process them again.
	634	+ * Walk back the VMA list from the last marked VMA.
	635	+ */
	636	+ for (; last; last = last->vm_prev) {
	637	+ if (last->vm_flags & VM_DONTCOPY)
	638	+ continue;
	639	+ if (!(last->vm_flags & VM_WIPEONFORK))
	640	+ vm_write_end(last);
	641	+ }
	642	+ }
	643	+
	644	+ mmap_write_unlock(oldmm);
570	645	dup_userfaultfd_complete(&uf);
571	646	fail_uprobe_end:
572	647	uprobe_end_dup_mmap();
..	..	@@ -596,9 +671,9 @@
596	671	#else
597	672	static int dup_mmap(struct mm_struct mm, struct mm_struct oldmm)
598	673	{
599		- down_write(&oldmm->mmap_sem);
	674	+ mmap_write_lock(oldmm);
600	675	RCU_INIT_POINTER(mm->exe_file, get_mm_exe_file(oldmm));
601		- up_write(&oldmm->mmap_sem);
	676	+ mmap_write_unlock(oldmm);
602	677	return 0;
603	678	}
604	679	#define mm_alloc_pgd(mm) (0)
..	..	@@ -609,12 +684,15 @@
609	684	{
610	685	int i;
611	686
	687	+ BUILD_BUG_ON_MSG(ARRAY_SIZE(resident_page_types) != NR_MM_COUNTERS,
	688	+ "Please make sure 'struct resident_page_types[]' is updated as well");
	689	+
612	690	for (i = 0; i < NR_MM_COUNTERS; i++) {
613	691	long x = atomic_long_read(&mm->rss_stat.count[i]);
614	692
615	693	if (unlikely(x))
616		- printk(KERN_ALERT "BUG: Bad rss-counter state "
617		- "mm:%p idx:%d val:%ld\n", mm, i, x);
	694	+ pr_alert("BUG: Bad rss-counter state mm:%p type:%s val:%ld\n",
	695	+ mm, resident_page_types[i], x);
618	696	}
619	697
620	698	if (mm_pgtables_bytes(mm))
..	..	@@ -641,8 +719,7 @@
641	719	WARN_ON_ONCE(mm == current->active_mm);
642	720	mm_free_pgd(mm);
643	721	destroy_context(mm);
644		- hmm_mm_destroy(mm);
645		- mmu_notifier_mm_destroy(mm);
	722	+ mmu_notifier_subscriptions_destroy(mm);
646	723	check_mm(mm);
647	724	put_user_ns(mm->user_ns);
648	725	free_mm(mm);
..	..	@@ -680,16 +757,17 @@
680	757
681	758	static inline void put_signal_struct(struct signal_struct *sig)
682	759	{
683		- if (atomic_dec_and_test(&sig->sigcnt))
	760	+ if (refcount_dec_and_test(&sig->sigcnt))
684	761	free_signal_struct(sig);
685	762	}
686	763
687	764	void __put_task_struct(struct task_struct *tsk)
688	765	{
689	766	WARN_ON(!tsk->exit_state);
690		- WARN_ON(atomic_read(&tsk->usage));
	767	+ WARN_ON(refcount_read(&tsk->usage));
691	768	WARN_ON(tsk == current);
692	769
	770	+ io_uring_free(tsk);
693	771	cgroup_free(tsk);
694	772	task_numa_free(tsk, true);
695	773	security_task_free(tsk);
..	..	@@ -702,6 +780,14 @@
702	780	}
703	781	EXPORT_SYMBOL_GPL(__put_task_struct);
704	782
	783	+void __put_task_struct_rcu_cb(struct rcu_head *rhp)
	784	+{
	785	+ struct task_struct *task = container_of(rhp, struct task_struct, rcu);
	786	+
	787	+ __put_task_struct(task);
	788	+}
	789	+EXPORT_SYMBOL_GPL(__put_task_struct_rcu_cb);
	790	+
705	791	void __init __weak arch_task_cache_init(void) { }
706	792
707	793	/*
..	..	@@ -710,15 +796,16 @@
710	796	static void set_max_threads(unsigned int max_threads_suggested)
711	797	{
712	798	u64 threads;
	799	+ unsigned long nr_pages = totalram_pages();
713	800
714	801	/*
715	802	* The number of threads shall be limited such that the thread
716	803	* structures may only consume a small part of the available memory.
717	804	*/
718		- if (fls64(totalram_pages) + fls64(PAGE_SIZE) > 64)
	805	+ if (fls64(nr_pages) + fls64(PAGE_SIZE) > 64)
719	806	threads = MAX_THREADS;
720	807	else
721		- threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
	808	+ threads = div64_u64((u64) nr_pages * (u64) PAGE_SIZE,
722	809	(u64) THREAD_SIZE * 8UL);
723	810
724	811	if (threads > max_threads_suggested)
..	..	@@ -732,6 +819,7 @@
732	819	int arch_task_struct_size __read_mostly;
733	820	#endif
734	821
	822	+#ifndef CONFIG_ARCH_TASK_STRUCT_ALLOCATOR
735	823	static void task_struct_whitelist(unsigned long offset, unsigned long size)
736	824	{
737	825	/* Fetch thread_struct whitelist for the architecture. */
..	..	@@ -746,6 +834,7 @@
746	834	else
747	835	*offset += offsetof(struct task_struct, thread);
748	836	}
	837	+#endif /* CONFIG_ARCH_TASK_STRUCT_ALLOCATOR */
749	838
750	839	void __init fork_init(void)
751	840	{
..	..	@@ -787,6 +876,7 @@
787	876	scs_init();
788	877
789	878	lockdep_init_task(&init_task);
	879	+ uprobes_init();
790	880	}
791	881
792	882	int __weak arch_dup_task_struct(struct task_struct *dst,
..	..	@@ -808,7 +898,7 @@
808	898	{
809	899	struct task_struct *tsk;
810	900	unsigned long *stack;
811		- struct vm_struct *stack_vm_area;
	901	+ struct vm_struct *stack_vm_area __maybe_unused;
812	902	int err;
813	903
814	904	if (node == NUMA_NO_NODE)
..	..	@@ -820,6 +910,9 @@
820	910	stack = alloc_thread_stack_node(tsk, node);
821	911	if (!stack)
822	912	goto free_tsk;
	913	+
	914	+ if (memcg_charge_kernel_stack(tsk))
	915	+ goto free_stack;
823	916
824	917	stack_vm_area = task_stack_vm_area(tsk);
825	918
..	..	@@ -835,7 +928,7 @@
835	928	tsk->stack_vm_area = stack_vm_area;
836	929	#endif
837	930	#ifdef CONFIG_THREAD_INFO_IN_TASK
838		- atomic_set(&tsk->stack_refcount, 1);
	931	+ refcount_set(&tsk->stack_refcount, 1);
839	932	#endif
840	933
841	934	if (err)
..	..	@@ -863,18 +956,23 @@
863	956	#ifdef CONFIG_STACKPROTECTOR
864	957	tsk->stack_canary = get_random_canary();
865	958	#endif
	959	+ if (orig->cpus_ptr == &orig->cpus_mask)
	960	+ tsk->cpus_ptr = &tsk->cpus_mask;
866	961
867	962	/*
868		- * One for us, one for whoever does the "release_task()" (usually
869		- * parent)
	963	+ * One for the user space visible state that goes away when reaped.
	964	+ * One for the scheduler.
870	965	*/
871		- atomic_set(&tsk->usage, 2);
	966	+ refcount_set(&tsk->rcu_users, 2);
	967	+ /* One for the rcu users */
	968	+ refcount_set(&tsk->usage, 1);
872	969	#ifdef CONFIG_BLK_DEV_IO_TRACE
873	970	tsk->btrace_seq = 0;
874	971	#endif
875	972	tsk->splice_pipe = NULL;
876	973	tsk->task_frag.page = NULL;
877	974	tsk->wake_q.next = NULL;
	975	+ tsk->pf_io_worker = NULL;
878	976
879	977	account_kernel_stack(tsk, 1);
880	978
..	..	@@ -892,6 +990,11 @@
892	990	#ifdef CONFIG_MEMCG
893	991	tsk->active_memcg = NULL;
894	992	#endif
	993	+
	994	+ android_init_vendor_data(tsk, 1);
	995	+ android_init_oem_data(tsk, 1);
	996	+
	997	+ trace_android_vh_dup_task_struct(tsk, orig);
895	998	return tsk;
896	999
897	1000	free_stack:
..	..	@@ -941,6 +1044,13 @@
941	1044	#endif
942	1045	}
943	1046
	1047	+static void mm_init_pasid(struct mm_struct *mm)
	1048	+{
	1049	+#ifdef CONFIG_IOMMU_SUPPORT
	1050	+ mm->pasid = INIT_PASID;
	1051	+#endif
	1052	+}
	1053	+
944	1054	static void mm_init_uprobes_state(struct mm_struct *mm)
945	1055	{
946	1056	#ifdef CONFIG_UPROBES
..	..	@@ -954,24 +1064,30 @@
954	1064	mm->mmap = NULL;
955	1065	mm->mm_rb = RB_ROOT;
956	1066	mm->vmacache_seqnum = 0;
	1067	+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
	1068	+ rwlock_init(&mm->mm_rb_lock);
	1069	+#endif
957	1070	atomic_set(&mm->mm_users, 1);
958	1071	atomic_set(&mm->mm_count, 1);
959		- init_rwsem(&mm->mmap_sem);
	1072	+ seqcount_init(&mm->write_protect_seq);
	1073	+ mmap_init_lock(mm);
960	1074	INIT_LIST_HEAD(&mm->mmlist);
961	1075	mm->core_state = NULL;
962	1076	mm_pgtables_bytes_init(mm);
963	1077	mm->map_count = 0;
964	1078	mm->locked_vm = 0;
965		- mm->pinned_vm = 0;
	1079	+ atomic_set(&mm->has_pinned, 0);
	1080	+ atomic64_set(&mm->pinned_vm, 0);
966	1081	memset(&mm->rss_stat, 0, sizeof(mm->rss_stat));
967	1082	spin_lock_init(&mm->page_table_lock);
968	1083	spin_lock_init(&mm->arg_lock);
969	1084	mm_init_cpumask(mm);
970	1085	mm_init_aio(mm);
971	1086	mm_init_owner(mm, p);
	1087	+ mm_init_pasid(mm);
972	1088	RCU_INIT_POINTER(mm->exe_file, NULL);
973		- mmu_notifier_mm_init(mm);
974		- hmm_mm_init(mm);
	1089	+ if (!mmu_notifier_subscriptions_init(mm))
	1090	+ goto fail_nopgd;
975	1091	init_tlb_flush_pending(mm);
976	1092	#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && !USE_SPLIT_PMD_PTLOCKS
977	1093	mm->pmd_huge_pte = NULL;
..	..	@@ -1046,8 +1162,10 @@
1046	1162	{
1047	1163	might_sleep();
1048	1164
1049		- if (atomic_dec_and_test(&mm->mm_users))
	1165	+ if (atomic_dec_and_test(&mm->mm_users)) {
	1166	+ trace_android_vh_mmput(NULL);
1050	1167	__mmput(mm);
	1168	+ }
1051	1169	}
1052	1170	EXPORT_SYMBOL_GPL(mmput);
1053	1171
..	..	@@ -1067,6 +1185,7 @@
1067	1185	schedule_work(&mm->async_put_work);
1068	1186	}
1069	1187	}
	1188	+EXPORT_SYMBOL_GPL(mmput_async);
1070	1189	#endif
1071	1190
1072	1191	/**
..	..	@@ -1171,7 +1290,7 @@
1171	1290	struct mm_struct *mm;
1172	1291	int err;
1173	1292
1174		- err = mutex_lock_killable(&task->signal->cred_guard_mutex);
	1293	+ err = down_read_killable(&task->signal->exec_update_lock);
1175	1294	if (err)
1176	1295	return ERR_PTR(err);
1177	1296
..	..	@@ -1181,7 +1300,7 @@
1181	1300	mmput(mm);
1182	1301	mm = ERR_PTR(-EACCES);
1183	1302	}
1184		- mutex_unlock(&task->signal->cred_guard_mutex);
	1303	+ up_read(&task->signal->exec_update_lock);
1185	1304
1186	1305	return mm;
1187	1306	}
..	..	@@ -1279,13 +1398,20 @@
1279	1398	mm_release(tsk, mm);
1280	1399	}
1281	1400
1282		-/*
1283		- * Allocate a new mm structure and copy contents from the
1284		- * mm structure of the passed in task structure.
	1401	+/**
	1402	+ * dup_mm() - duplicates an existing mm structure
	1403	+ * @tsk: the task_struct with which the new mm will be associated.
	1404	+ * @oldmm: the mm to duplicate.
	1405	+ *
	1406	+ * Allocates a new mm structure and duplicates the provided @oldmm structure
	1407	+ * content into it.
	1408	+ *
	1409	+ * Return: the duplicated mm or NULL on failure.
1285	1410	*/
1286		-static struct mm_struct dup_mm(struct task_struct tsk)
	1411	+static struct mm_struct dup_mm(struct task_struct tsk,
	1412	+ struct mm_struct *oldmm)
1287	1413	{
1288		- struct mm_struct mm, oldmm = current->mm;
	1414	+ struct mm_struct *mm;
1289	1415	int err;
1290	1416
1291	1417	mm = allocate_mm();
..	..	@@ -1353,7 +1479,7 @@
1353	1479	}
1354	1480
1355	1481	retval = -ENOMEM;
1356		- mm = dup_mm(tsk);
	1482	+ mm = dup_mm(tsk, current->mm);
1357	1483	if (!mm)
1358	1484	goto fail_nomem;
1359	1485
..	..	@@ -1403,7 +1529,7 @@
1403	1529	goto out;
1404	1530	}
1405	1531
1406		- newf = dup_fd(oldf, &error);
	1532	+ newf = dup_fd(oldf, NR_OPEN_MAX, &error);
1407	1533	if (!newf)
1408	1534	goto out;
1409	1535
..	..	@@ -1444,24 +1570,29 @@
1444	1570	struct sighand_struct *sig;
1445	1571
1446	1572	if (clone_flags & CLONE_SIGHAND) {
1447		- atomic_inc(&current->sighand->count);
	1573	+ refcount_inc(&current->sighand->count);
1448	1574	return 0;
1449	1575	}
1450	1576	sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1451		- rcu_assign_pointer(tsk->sighand, sig);
	1577	+ RCU_INIT_POINTER(tsk->sighand, sig);
1452	1578	if (!sig)
1453	1579	return -ENOMEM;
1454	1580
1455		- atomic_set(&sig->count, 1);
	1581	+ refcount_set(&sig->count, 1);
1456	1582	spin_lock_irq(&current->sighand->siglock);
1457	1583	memcpy(sig->action, current->sighand->action, sizeof(sig->action));
1458	1584	spin_unlock_irq(&current->sighand->siglock);
	1585	+
	1586	+ /* Reset all signal handler not set to SIG_IGN to SIG_DFL. */
	1587	+ if (clone_flags & CLONE_CLEAR_SIGHAND)
	1588	+ flush_signal_handlers(tsk, 0);
	1589	+
1459	1590	return 0;
1460	1591	}
1461	1592
1462	1593	void __cleanup_sighand(struct sighand_struct *sighand)
1463	1594	{
1464		- if (atomic_dec_and_test(&sighand->count)) {
	1595	+ if (refcount_dec_and_test(&sighand->count)) {
1465	1596	signalfd_cleanup(sighand);
1466	1597	/*
1467	1598	* sighand_cachep is SLAB_TYPESAFE_BY_RCU so we can free it
..	..	@@ -1471,28 +1602,17 @@
1471	1602	}
1472	1603	}
1473	1604
1474		-#ifdef CONFIG_POSIX_TIMERS
1475	1605	/*
1476	1606	* Initialize POSIX timer handling for a thread group.
1477	1607	*/
1478	1608	static void posix_cpu_timers_init_group(struct signal_struct *sig)
1479	1609	{
	1610	+ struct posix_cputimers *pct = &sig->posix_cputimers;
1480	1611	unsigned long cpu_limit;
1481	1612
1482	1613	cpu_limit = READ_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
1483		- if (cpu_limit != RLIM_INFINITY) {
1484		- sig->cputime_expires.prof_exp = cpu_limit * NSEC_PER_SEC;
1485		- sig->cputimer.running = true;
1486		- }
1487		-
1488		- /* The timer lists. */
1489		- INIT_LIST_HEAD(&sig->cpu_timers[0]);
1490		- INIT_LIST_HEAD(&sig->cpu_timers[1]);
1491		- INIT_LIST_HEAD(&sig->cpu_timers[2]);
	1614	+ posix_cputimers_group_init(pct, cpu_limit);
1492	1615	}
1493		-#else
1494		-static inline void posix_cpu_timers_init_group(struct signal_struct *sig) { }
1495		-#endif
1496	1616
1497	1617	static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1498	1618	{
..	..	@@ -1508,7 +1628,7 @@
1508	1628
1509	1629	sig->nr_threads = 1;
1510	1630	atomic_set(&sig->live, 1);
1511		- atomic_set(&sig->sigcnt, 1);
	1631	+ refcount_set(&sig->sigcnt, 1);
1512	1632
1513	1633	/* list_add(thread_node, thread_head) without INIT_LIST_HEAD() */
1514	1634	sig->thread_head = (struct list_head)LIST_HEAD_INIT(tsk->thread_node);
..	..	@@ -1540,6 +1660,7 @@
1540	1660	sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1541	1661
1542	1662	mutex_init(&sig->cred_guard_mutex);
	1663	+ init_rwsem(&sig->exec_update_lock);
1543	1664
1544	1665	return 0;
1545	1666	}
..	..	@@ -1594,23 +1715,6 @@
1594	1715	#endif
1595	1716	}
1596	1717
1597		-#ifdef CONFIG_POSIX_TIMERS
1598		-/*
1599		- * Initialize POSIX timer handling for a single task.
1600		- */
1601		-static void posix_cpu_timers_init(struct task_struct *tsk)
1602		-{
1603		- tsk->cputime_expires.prof_exp = 0;
1604		- tsk->cputime_expires.virt_exp = 0;
1605		- tsk->cputime_expires.sched_exp = 0;
1606		- INIT_LIST_HEAD(&tsk->cpu_timers[0]);
1607		- INIT_LIST_HEAD(&tsk->cpu_timers[1]);
1608		- INIT_LIST_HEAD(&tsk->cpu_timers[2]);
1609		-}
1610		-#else
1611		-static inline void posix_cpu_timers_init(struct task_struct *tsk) { }
1612		-#endif
1613		-
1614	1718	static inline void init_task_pid_links(struct task_struct *task)
1615	1719	{
1616	1720	enum pid_type type;
..	..	@@ -1642,7 +1746,125 @@
1642	1746	INIT_LIST_HEAD(&p->rcu_tasks_holdout_list);
1643	1747	p->rcu_tasks_idle_cpu = -1;
1644	1748	#endif /* #ifdef CONFIG_TASKS_RCU */
	1749	+#ifdef CONFIG_TASKS_TRACE_RCU
	1750	+ p->trc_reader_nesting = 0;
	1751	+ p->trc_reader_special.s = 0;
	1752	+ INIT_LIST_HEAD(&p->trc_holdout_list);
	1753	+#endif /* #ifdef CONFIG_TASKS_TRACE_RCU */
1645	1754	}
	1755	+
	1756	+struct pid pidfd_pid(const struct file file)
	1757	+{
	1758	+ if (file->f_op == &pidfd_fops)
	1759	+ return file->private_data;
	1760	+
	1761	+ return ERR_PTR(-EBADF);
	1762	+}
	1763	+
	1764	+static int pidfd_release(struct inode inode, struct file file)
	1765	+{
	1766	+ struct pid *pid = file->private_data;
	1767	+
	1768	+ file->private_data = NULL;
	1769	+ put_pid(pid);
	1770	+ return 0;
	1771	+}
	1772	+
	1773	+#ifdef CONFIG_PROC_FS
	1774	+/**
	1775	+ * pidfd_show_fdinfo - print information about a pidfd
	1776	+ * @m: proc fdinfo file
	1777	+ * @f: file referencing a pidfd
	1778	+ *
	1779	+ * Pid:
	1780	+ * This function will print the pid that a given pidfd refers to in the
	1781	+ * pid namespace of the procfs instance.
	1782	+ * If the pid namespace of the process is not a descendant of the pid
	1783	+ * namespace of the procfs instance 0 will be shown as its pid. This is
	1784	+ * similar to calling getppid() on a process whose parent is outside of
	1785	+ * its pid namespace.
	1786	+ *
	1787	+ * NSpid:
	1788	+ * If pid namespaces are supported then this function will also print
	1789	+ * the pid of a given pidfd refers to for all descendant pid namespaces
	1790	+ * starting from the current pid namespace of the instance, i.e. the
	1791	+ * Pid field and the first entry in the NSpid field will be identical.
	1792	+ * If the pid namespace of the process is not a descendant of the pid
	1793	+ * namespace of the procfs instance 0 will be shown as its first NSpid
	1794	+ * entry and no others will be shown.
	1795	+ * Note that this differs from the Pid and NSpid fields in
	1796	+ * /proc/<pid>/status where Pid and NSpid are always shown relative to
	1797	+ * the pid namespace of the procfs instance. The difference becomes
	1798	+ * obvious when sending around a pidfd between pid namespaces from a
	1799	+ * different branch of the tree, i.e. where no ancestoral relation is
	1800	+ * present between the pid namespaces:
	1801	+ * - create two new pid namespaces ns1 and ns2 in the initial pid
	1802	+ * namespace (also take care to create new mount namespaces in the
	1803	+ * new pid namespace and mount procfs)
	1804	+ * - create a process with a pidfd in ns1
	1805	+ * - send pidfd from ns1 to ns2
	1806	+ * - read /proc/self/fdinfo/<pidfd> and observe that both Pid and NSpid
	1807	+ * have exactly one entry, which is 0
	1808	+ */
	1809	+static void pidfd_show_fdinfo(struct seq_file m, struct file f)
	1810	+{
	1811	+ struct pid *pid = f->private_data;
	1812	+ struct pid_namespace *ns;
	1813	+ pid_t nr = -1;
	1814	+
	1815	+ if (likely(pid_has_task(pid, PIDTYPE_PID))) {
	1816	+ ns = proc_pid_ns(file_inode(m->file)->i_sb);
	1817	+ nr = pid_nr_ns(pid, ns);
	1818	+ }
	1819	+
	1820	+ seq_put_decimal_ll(m, "Pid:\t", nr);
	1821	+
	1822	+#ifdef CONFIG_PID_NS
	1823	+ seq_put_decimal_ll(m, "\nNSpid:\t", nr);
	1824	+ if (nr > 0) {
	1825	+ int i;
	1826	+
	1827	+ /* If nr is non-zero it means that 'pid' is valid and that
	1828	+ * ns, i.e. the pid namespace associated with the procfs
	1829	+ * instance, is in the pid namespace hierarchy of pid.
	1830	+ * Start at one below the already printed level.
	1831	+ */
	1832	+ for (i = ns->level + 1; i <= pid->level; i++)
	1833	+ seq_put_decimal_ll(m, "\t", pid->numbers[i].nr);
	1834	+ }
	1835	+#endif
	1836	+ seq_putc(m, '\n');
	1837	+}
	1838	+#endif
	1839	+
	1840	+/*
	1841	+ * Poll support for process exit notification.
	1842	+ */
	1843	+static __poll_t pidfd_poll(struct file file, struct poll_table_struct pts)
	1844	+{
	1845	+ struct pid *pid = file->private_data;
	1846	+ __poll_t poll_flags = 0;
	1847	+
	1848	+ poll_wait(file, &pid->wait_pidfd, pts);
	1849	+
	1850	+ /*
	1851	+ * Inform pollers only when the whole thread group exits.
	1852	+ * If the thread group leader exits before all other threads in the
	1853	+ * group, then poll(2) should block, similar to the wait(2) family.
	1854	+ */
	1855	+ if (thread_group_exited(pid))
	1856	+ poll_flags = EPOLLIN \| EPOLLRDNORM;
	1857	+
	1858	+ return poll_flags;
	1859	+}
	1860	+
	1861	+const struct file_operations pidfd_fops = {
	1862	+ .release = pidfd_release,
	1863	+ .poll = pidfd_poll,
	1864	+#ifdef CONFIG_PROC_FS
	1865	+ .show_fdinfo = pidfd_show_fdinfo,
	1866	+#endif
	1867	+};
1646	1868
1647	1869	static void __delayed_free_task(struct rcu_head *rhp)
1648	1870	{
..	..	@@ -1657,84 +1879,6 @@
1657	1879	call_rcu(&tsk->rcu, __delayed_free_task);
1658	1880	else
1659	1881	free_task(tsk);
1660		-}
1661		-
1662		-static int pidfd_release(struct inode inode, struct file file)
1663		-{
1664		- struct pid *pid = file->private_data;
1665		-
1666		- file->private_data = NULL;
1667		- put_pid(pid);
1668		- return 0;
1669		-}
1670		-
1671		-#ifdef CONFIG_PROC_FS
1672		-static void pidfd_show_fdinfo(struct seq_file m, struct file f)
1673		-{
1674		- struct pid_namespace *ns = proc_pid_ns(file_inode(m->file));
1675		- struct pid *pid = f->private_data;
1676		-
1677		- seq_put_decimal_ull(m, "Pid:\t", pid_nr_ns(pid, ns));
1678		- seq_putc(m, '\n');
1679		-}
1680		-#endif
1681		-
1682		-/*
1683		- * Poll support for process exit notification.
1684		- */
1685		-static __poll_t pidfd_poll(struct file file, struct poll_table_struct pts)
1686		-{
1687		- struct task_struct *task;
1688		- struct pid *pid = file->private_data;
1689		- __poll_t poll_flags = 0;
1690		-
1691		- poll_wait(file, &pid->wait_pidfd, pts);
1692		-
1693		- rcu_read_lock();
1694		- task = pid_task(pid, PIDTYPE_PID);
1695		- /*
1696		- * Inform pollers only when the whole thread group exits.
1697		- * If the thread group leader exits before all other threads in the
1698		- * group, then poll(2) should block, similar to the wait(2) family.
1699		- */
1700		- if (!task \|\| (task->exit_state && thread_group_empty(task)))
1701		- poll_flags = EPOLLIN \| EPOLLRDNORM;
1702		- rcu_read_unlock();
1703		-
1704		- return poll_flags;
1705		-}
1706		-
1707		-const struct file_operations pidfd_fops = {
1708		- .release = pidfd_release,
1709		- .poll = pidfd_poll,
1710		-#ifdef CONFIG_PROC_FS
1711		- .show_fdinfo = pidfd_show_fdinfo,
1712		-#endif
1713		-};
1714		-
1715		-/**
1716		- * pidfd_create() - Create a new pid file descriptor.
1717		- *
1718		- * @pid: struct pid that the pidfd will reference
1719		- *
1720		- * This creates a new pid file descriptor with the O_CLOEXEC flag set.
1721		- *
1722		- * Note, that this function can only be called after the fd table has
1723		- * been unshared to avoid leaking the pidfd to the new process.
1724		- *
1725		- * Return: On success, a cloexec pidfd is returned.
1726		- * On error, a negative errno number will be returned.
1727		- */
1728		-static int pidfd_create(struct pid *pid)
1729		-{
1730		- int fd;
1731		-
1732		- fd = anon_inode_getfd("[pidfd]", &pidfd_fops, get_pid(pid),
1733		- O_RDWR \| O_CLOEXEC);
1734		- if (fd < 0)
1735		- put_pid(pid);
1736		-
1737		- return fd;
1738	1882	}
1739	1883
1740	1884	static void copy_oom_score_adj(u64 clone_flags, struct task_struct *tsk)
..	..	@@ -1765,19 +1909,17 @@
1765	1909	* flags). The actual kick-off is left to the caller.
1766	1910	*/
1767	1911	static __latent_entropy struct task_struct *copy_process(
1768		- unsigned long clone_flags,
1769		- unsigned long stack_start,
1770		- unsigned long stack_size,
1771		- int __user *parent_tidptr,
1772		- int __user *child_tidptr,
1773	1912	struct pid *pid,
1774	1913	int trace,
1775		- unsigned long tls,
1776		- int node)
	1914	+ int node,
	1915	+ struct kernel_clone_args *args)
1777	1916	{
1778	1917	int pidfd = -1, retval;
1779	1918	struct task_struct *p;
1780	1919	struct multiprocess_signals delayed;
	1920	+ struct file *pidfile = NULL;
	1921	+ u64 clone_flags = args->flags;
	1922	+ struct nsproxy *nsp = current->nsproxy;
1781	1923
1782	1924	/*
1783	1925	* Don't allow sharing the root directory with processes in a different
..	..	@@ -1820,21 +1962,26 @@
1820	1962	*/
1821	1963	if (clone_flags & CLONE_THREAD) {
1822	1964	if ((clone_flags & (CLONE_NEWUSER \| CLONE_NEWPID)) \|\|
1823		- (task_active_pid_ns(current) !=
1824		- current->nsproxy->pid_ns_for_children))
	1965	+ (task_active_pid_ns(current) != nsp->pid_ns_for_children))
	1966	+ return ERR_PTR(-EINVAL);
	1967	+ }
	1968	+
	1969	+ /*
	1970	+ * If the new process will be in a different time namespace
	1971	+ * do not allow it to share VM or a thread group with the forking task.
	1972	+ */
	1973	+ if (clone_flags & (CLONE_THREAD \| CLONE_VM)) {
	1974	+ if (nsp->time_ns != nsp->time_ns_for_children)
1825	1975	return ERR_PTR(-EINVAL);
1826	1976	}
1827	1977
1828	1978	if (clone_flags & CLONE_PIDFD) {
1829	1979	/*
1830		- * - CLONE_PARENT_SETTID is useless for pidfds and also
1831		- * parent_tidptr is used to return pidfds.
1832	1980	* - CLONE_DETACHED is blocked so that we can potentially
1833	1981	* reuse it later for CLONE_PIDFD.
1834	1982	* - CLONE_THREAD is blocked until someone really needs it.
1835	1983	*/
1836		- if (clone_flags &
1837		- (CLONE_DETACHED \| CLONE_PARENT_SETTID \| CLONE_THREAD))
	1984	+ if (clone_flags & (CLONE_DETACHED \| CLONE_THREAD))
1838	1985	return ERR_PTR(-EINVAL);
1839	1986	}
1840	1987
..	..	@@ -1853,13 +2000,21 @@
1853	2000	recalc_sigpending();
1854	2001	spin_unlock_irq(&current->sighand->siglock);
1855	2002	retval = -ERESTARTNOINTR;
1856		- if (signal_pending(current))
	2003	+ if (task_sigpending(current))
1857	2004	goto fork_out;
1858	2005
1859	2006	retval = -ENOMEM;
1860	2007	p = dup_task_struct(current, node);
1861	2008	if (!p)
1862	2009	goto fork_out;
	2010	+ if (args->io_thread) {
	2011	+ /*
	2012	+ * Mark us an IO worker, and block any signal that isn't
	2013	+ * fatal or STOP
	2014	+ */
	2015	+ p->flags \|= PF_IO_WORKER;
	2016	+ siginitsetinv(&p->blocked, sigmask(SIGKILL)\|sigmask(SIGSTOP));
	2017	+ }
1863	2018
1864	2019	cpufreq_task_times_init(p);
1865	2020
..	..	@@ -1869,18 +2024,18 @@
1869	2024	* p->set_child_tid which is (ab)used as a kthread's data pointer for
1870	2025	* kernel threads (PF_KTHREAD).
1871	2026	*/
1872		- p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
	2027	+ p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? args->child_tid : NULL;
1873	2028	/*
1874	2029	* Clear TID on mm_release()?
1875	2030	*/
1876		- p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? child_tidptr : NULL;
	2031	+ p->clear_child_tid = (clone_flags & CLONE_CHILD_CLEARTID) ? args->child_tid : NULL;
1877	2032
1878	2033	ftrace_graph_init_task(p);
1879	2034
1880	2035	rt_mutex_init_task(p);
1881	2036
	2037	+ lockdep_assert_irqs_enabled();
1882	2038	#ifdef CONFIG_PROVE_LOCKING
1883		- DEBUG_LOCKS_WARN_ON(!p->hardirqs_enabled);
1884	2039	DEBUG_LOCKS_WARN_ON(!p->softirqs_enabled);
1885	2040	#endif
1886	2041	retval = -EAGAIN;
..	..	@@ -1902,7 +2057,7 @@
1902	2057	* to stop root fork bombs.
1903	2058	*/
1904	2059	retval = -EAGAIN;
1905		- if (nr_threads >= max_threads)
	2060	+ if (data_race(nr_threads >= max_threads))
1906	2061	goto bad_fork_cleanup_count;
1907	2062
1908	2063	delayacct_tsk_init(p); /* Must remain after dup_task_struct() */
..	..	@@ -1928,6 +2083,10 @@
1928	2083	p->vtime.state = VTIME_INACTIVE;
1929	2084	#endif
1930	2085
	2086	+#ifdef CONFIG_IO_URING
	2087	+ p->io_uring = NULL;
	2088	+#endif
	2089	+
1931	2090	#if defined(SPLIT_RSS_COUNTING)
1932	2091	memset(&p->rss_stat, 0, sizeof(p->rss_stat));
1933	2092	#endif
..	..	@@ -1941,7 +2100,7 @@
1941	2100	task_io_accounting_init(&p->ioac);
1942	2101	acct_clear_integrals(p);
1943	2102
1944		- posix_cpu_timers_init(p);
	2103	+ posix_cputimers_init(&p->posix_cputimers);
1945	2104
1946	2105	p->io_context = NULL;
1947	2106	audit_set_context(p, NULL);
..	..	@@ -1957,30 +2116,19 @@
1957	2116	#ifdef CONFIG_CPUSETS
1958	2117	p->cpuset_mem_spread_rotor = NUMA_NO_NODE;
1959	2118	p->cpuset_slab_spread_rotor = NUMA_NO_NODE;
1960		- seqcount_init(&p->mems_allowed_seq);
	2119	+ seqcount_spinlock_init(&p->mems_allowed_seq, &p->alloc_lock);
1961	2120	#endif
1962	2121	#ifdef CONFIG_TRACE_IRQFLAGS
1963		- p->irq_events = 0;
1964		- p->hardirqs_enabled = 0;
1965		- p->hardirq_enable_ip = 0;
1966		- p->hardirq_enable_event = 0;
1967		- p->hardirq_disable_ip = _THIS_IP_;
1968		- p->hardirq_disable_event = 0;
1969		- p->softirqs_enabled = 1;
1970		- p->softirq_enable_ip = _THIS_IP_;
1971		- p->softirq_enable_event = 0;
1972		- p->softirq_disable_ip = 0;
1973		- p->softirq_disable_event = 0;
1974		- p->hardirq_context = 0;
1975		- p->softirq_context = 0;
	2122	+ memset(&p->irqtrace, 0, sizeof(p->irqtrace));
	2123	+ p->irqtrace.hardirq_disable_ip = _THIS_IP_;
	2124	+ p->irqtrace.softirq_enable_ip = _THIS_IP_;
	2125	+ p->softirqs_enabled = 1;
	2126	+ p->softirq_context = 0;
1976	2127	#endif
1977	2128
1978	2129	p->pagefault_disabled = 0;
1979	2130
1980	2131	#ifdef CONFIG_LOCKDEP
1981		- p->lockdep_depth = 0; /* no locks held yet */
1982		- p->curr_chain_key = 0;
1983		- p->lockdep_recursion = 0;
1984	2132	lockdep_init_task(p);
1985	2133	#endif
1986	2134
..	..	@@ -2032,12 +2180,15 @@
2032	2180	retval = copy_io(clone_flags, p);
2033	2181	if (retval)
2034	2182	goto bad_fork_cleanup_namespaces;
2035		- retval = copy_thread_tls(clone_flags, stack_start, stack_size, p, tls);
	2183	+ retval = copy_thread(clone_flags, args->stack, args->stack_size, p, args->tls);
2036	2184	if (retval)
2037	2185	goto bad_fork_cleanup_io;
2038	2186
	2187	+ stackleak_task_init(p);
	2188	+
2039	2189	if (pid != &init_struct_pid) {
2040		- pid = alloc_pid(p->nsproxy->pid_ns_for_children);
	2190	+ pid = alloc_pid(p->nsproxy->pid_ns_for_children, args->set_tid,
	2191	+ args->set_tid_size);
2041	2192	if (IS_ERR(pid)) {
2042	2193	retval = PTR_ERR(pid);
2043	2194	goto bad_fork_cleanup_thread;
..	..	@@ -2050,12 +2201,22 @@
2050	2201	* if the fd table isn't shared).
2051	2202	*/
2052	2203	if (clone_flags & CLONE_PIDFD) {
2053		- retval = pidfd_create(pid);
	2204	+ retval = get_unused_fd_flags(O_RDWR \| O_CLOEXEC);
2054	2205	if (retval < 0)
2055	2206	goto bad_fork_free_pid;
2056	2207
2057	2208	pidfd = retval;
2058		- retval = put_user(pidfd, parent_tidptr);
	2209	+
	2210	+ pidfile = anon_inode_getfile("[pidfd]", &pidfd_fops, pid,
	2211	+ O_RDWR \| O_CLOEXEC);
	2212	+ if (IS_ERR(pidfile)) {
	2213	+ put_unused_fd(pidfd);
	2214	+ retval = PTR_ERR(pidfile);
	2215	+ goto bad_fork_free_pid;
	2216	+ }
	2217	+ get_pid(pid); /* held by pidfile now */
	2218	+
	2219	+ retval = put_user(pidfd, args->pidfd);
2059	2220	if (retval)
2060	2221	goto bad_fork_put_pidfd;
2061	2222	}
..	..	@@ -2080,7 +2241,7 @@
2080	2241	#ifdef TIF_SYSCALL_EMU
2081	2242	clear_tsk_thread_flag(p, TIF_SYSCALL_EMU);
2082	2243	#endif
2083		- clear_all_latency_tracing(p);
	2244	+ clear_tsk_latency_tracing(p);
2084	2245
2085	2246	/* ok, now we should be set up.. */
2086	2247	p->pid = pid_nr(pid);
..	..	@@ -2099,17 +2260,28 @@
2099	2260	p->pdeath_signal = 0;
2100	2261	INIT_LIST_HEAD(&p->thread_group);
2101	2262	p->task_works = NULL;
	2263	+ clear_posix_cputimers_work(p);
2102	2264
2103		- cgroup_threadgroup_change_begin(current);
2104	2265	/*
2105	2266	* Ensure that the cgroup subsystem policies allow the new process to be
2106		- * forked. It should be noted the the new process's css_set can be changed
	2267	+ * forked. It should be noted that the new process's css_set can be changed
2107	2268	* between here and cgroup_post_fork() if an organisation operation is in
2108	2269	* progress.
2109	2270	*/
2110		- retval = cgroup_can_fork(p);
	2271	+ retval = cgroup_can_fork(p, args);
2111	2272	if (retval)
2112		- goto bad_fork_cgroup_threadgroup_change_end;
	2273	+ goto bad_fork_put_pidfd;
	2274	+
	2275	+ /*
	2276	+ * Now that the cgroups are pinned, re-clone the parent cgroup and put
	2277	+ * the new task on the correct runqueue. All this before the task
	2278	+ * becomes visible.
	2279	+ *
	2280	+ * This isn't part of ->can_fork() because while the re-cloning is
	2281	+ * cgroup specific, it unconditionally needs to place the task on a
	2282	+ * runqueue.
	2283	+ */
	2284	+ sched_cgroup_fork(p, args);
2113	2285
2114	2286	/*
2115	2287	* From this point on we must avoid any synchronous user-space
..	..	@@ -2120,7 +2292,7 @@
2120	2292	*/
2121	2293
2122	2294	p->start_time = ktime_get_ns();
2123		- p->real_start_time = ktime_get_boot_ns();
	2295	+ p->start_boottime = ktime_get_boottime_ns();
2124	2296
2125	2297	/*
2126	2298	* Make it visible to the rest of the system, but dont wake it up yet.
..	..	@@ -2139,18 +2311,12 @@
2139	2311	} else {
2140	2312	p->real_parent = current;
2141	2313	p->parent_exec_id = current->self_exec_id;
2142		- p->exit_signal = (clone_flags & CSIGNAL);
	2314	+ p->exit_signal = args->exit_signal;
2143	2315	}
2144	2316
2145	2317	klp_copy_process(p);
2146	2318
2147	2319	spin_lock(&current->sighand->siglock);
2148		-
2149		- /*
2150		- * Copy seccomp details explicitly here, in case they were changed
2151		- * before holding sighand lock.
2152		- */
2153		- copy_seccomp(p);
2154	2320
2155	2321	rseq_fork(p, clone_flags);
2156	2322
..	..	@@ -2166,6 +2332,13 @@
2166	2332	goto bad_fork_cancel_cgroup;
2167	2333	}
2168	2334
	2335	+ /* No more failure paths after this point. */
	2336	+
	2337	+ /*
	2338	+ * Copy seccomp details explicitly here, in case they were changed
	2339	+ * before holding sighand lock.
	2340	+ */
	2341	+ copy_seccomp(p);
2169	2342
2170	2343	init_task_pid_links(p);
2171	2344	if (likely(p->pid)) {
..	..	@@ -2199,7 +2372,7 @@
2199	2372	} else {
2200	2373	current->signal->nr_threads++;
2201	2374	atomic_inc(&current->signal->live);
2202		- atomic_inc(&current->signal->sigcnt);
	2375	+ refcount_inc(&current->signal->sigcnt);
2203	2376	task_join_group_stop(p);
2204	2377	list_add_tail_rcu(&p->thread_group,
2205	2378	&p->group_leader->thread_group);
..	..	@@ -2215,9 +2388,12 @@
2215	2388	syscall_tracepoint_update(p);
2216	2389	write_unlock_irq(&tasklist_lock);
2217	2390
	2391	+ if (pidfile)
	2392	+ fd_install(pidfd, pidfile);
	2393	+
2218	2394	proc_fork_connector(p);
2219		- cgroup_post_fork(p);
2220		- cgroup_threadgroup_change_end(current);
	2395	+ sched_post_fork(p);
	2396	+ cgroup_post_fork(p, args);
2221	2397	perf_event_fork(p);
2222	2398
2223	2399	trace_task_newtask(p, clone_flags);
..	..	@@ -2230,12 +2406,12 @@
2230	2406	bad_fork_cancel_cgroup:
2231	2407	spin_unlock(&current->sighand->siglock);
2232	2408	write_unlock_irq(&tasklist_lock);
2233		- cgroup_cancel_fork(p);
2234		-bad_fork_cgroup_threadgroup_change_end:
2235		- cgroup_threadgroup_change_end(current);
	2409	+ cgroup_cancel_fork(p, args);
2236	2410	bad_fork_put_pidfd:
2237		- if (clone_flags & CLONE_PIDFD)
2238		- ksys_close(pidfd);
	2411	+ if (clone_flags & CLONE_PIDFD) {
	2412	+ fput(pidfile);
	2413	+ put_unused_fd(pidfd);
	2414	+ }
2239	2415	bad_fork_free_pid:
2240	2416	if (pid != &init_struct_pid)
2241	2417	free_pid(pid);
..	..	@@ -2299,11 +2475,14 @@
2299	2475	}
2300	2476	}
2301	2477
2302		-struct task_struct *fork_idle(int cpu)
	2478	+struct task_struct * __init fork_idle(int cpu)
2303	2479	{
2304	2480	struct task_struct *task;
2305		- task = copy_process(CLONE_VM, 0, 0, NULL, NULL, &init_struct_pid, 0, 0,
2306		- cpu_to_node(cpu));
	2481	+ struct kernel_clone_args args = {
	2482	+ .flags = CLONE_VM,
	2483	+ };
	2484	+
	2485	+ task = copy_process(&init_struct_pid, 0, cpu_to_node(cpu), &args);
2307	2486	if (!IS_ERR(task)) {
2308	2487	init_idle_pids(task);
2309	2488	init_idle(task, cpu);
..	..	@@ -2313,23 +2492,57 @@
2313	2492	}
2314	2493
2315	2494	/*
	2495	+ * This is like kernel_clone(), but shaved down and tailored to just
	2496	+ * creating io_uring workers. It returns a created task, or an error pointer.
	2497	+ * The returned task is inactive, and the caller must fire it up through
	2498	+ * wake_up_new_task(p). All signals are blocked in the created task.
	2499	+ */
	2500	+struct task_struct create_io_thread(int (fn)(void ), void arg, int node)
	2501	+{
	2502	+ unsigned long flags = CLONE_FS\|CLONE_FILES\|CLONE_SIGHAND\|CLONE_THREAD\|
	2503	+ CLONE_IO;
	2504	+ struct kernel_clone_args args = {
	2505	+ .flags = ((lower_32_bits(flags) \| CLONE_VM \|
	2506	+ CLONE_UNTRACED) & ~CSIGNAL),
	2507	+ .exit_signal = (lower_32_bits(flags) & CSIGNAL),
	2508	+ .stack = (unsigned long)fn,
	2509	+ .stack_size = (unsigned long)arg,
	2510	+ .io_thread = 1,
	2511	+ };
	2512	+
	2513	+ return copy_process(NULL, 0, node, &args);
	2514	+}
	2515	+
	2516	+/*
2316	2517	* Ok, this is the main fork-routine.
2317	2518	*
2318	2519	* It copies the process, and if successful kick-starts
2319	2520	* it and waits for it to finish using the VM if required.
	2521	+ *
	2522	+ * args->exit_signal is expected to be checked for sanity by the caller.
2320	2523	*/
2321		-long _do_fork(unsigned long clone_flags,
2322		- unsigned long stack_start,
2323		- unsigned long stack_size,
2324		- int __user *parent_tidptr,
2325		- int __user *child_tidptr,
2326		- unsigned long tls)
	2524	+pid_t kernel_clone(struct kernel_clone_args *args)
2327	2525	{
	2526	+ u64 clone_flags = args->flags;
2328	2527	struct completion vfork;
2329	2528	struct pid *pid;
2330	2529	struct task_struct *p;
2331	2530	int trace = 0;
2332		- long nr;
	2531	+ pid_t nr;
	2532	+
	2533	+ /*
	2534	+ * For legacy clone() calls, CLONE_PIDFD uses the parent_tid argument
	2535	+ * to return the pidfd. Hence, CLONE_PIDFD and CLONE_PARENT_SETTID are
	2536	+ * mutually exclusive. With clone3() CLONE_PIDFD has grown a separate
	2537	+ * field in struct clone_args and it still doesn't make sense to have
	2538	+ * them both point at the same memory location. Performing this check
	2539	+ * here has the advantage that we don't need to have a separate helper
	2540	+ * to check for legacy clone().
	2541	+ */
	2542	+ if ((args->flags & CLONE_PIDFD) &&
	2543	+ (args->flags & CLONE_PARENT_SETTID) &&
	2544	+ (args->pidfd == args->parent_tid))
	2545	+ return -EINVAL;
2333	2546
2334	2547	/*
2335	2548	* Determine whether and which event to report to ptracer. When
..	..	@@ -2340,7 +2553,7 @@
2340	2553	if (!(clone_flags & CLONE_UNTRACED)) {
2341	2554	if (clone_flags & CLONE_VFORK)
2342	2555	trace = PTRACE_EVENT_VFORK;
2343		- else if ((clone_flags & CSIGNAL) != SIGCHLD)
	2556	+ else if (args->exit_signal != SIGCHLD)
2344	2557	trace = PTRACE_EVENT_CLONE;
2345	2558	else
2346	2559	trace = PTRACE_EVENT_FORK;
..	..	@@ -2349,8 +2562,7 @@
2349	2562	trace = 0;
2350	2563	}
2351	2564
2352		- p = copy_process(clone_flags, stack_start, stack_size, parent_tidptr,
2353		- child_tidptr, NULL, trace, tls, NUMA_NO_NODE);
	2565	+ p = copy_process(NULL, trace, NUMA_NO_NODE, args);
2354	2566	add_latent_entropy();
2355	2567
2356	2568	if (IS_ERR(p))
..	..	@@ -2368,7 +2580,7 @@
2368	2580	nr = pid_vnr(pid);
2369	2581
2370	2582	if (clone_flags & CLONE_PARENT_SETTID)
2371		- put_user(nr, parent_tidptr);
	2583	+ put_user(nr, args->parent_tid);
2372	2584
2373	2585	if (clone_flags & CLONE_VFORK) {
2374	2586	p->vfork_done = &vfork;
..	..	@@ -2391,34 +2603,31 @@
2391	2603	return nr;
2392	2604	}
2393	2605
2394		-#ifndef CONFIG_HAVE_COPY_THREAD_TLS
2395		-/* For compatibility with architectures that call do_fork directly rather than
2396		- * using the syscall entry points below. */
2397		-long do_fork(unsigned long clone_flags,
2398		- unsigned long stack_start,
2399		- unsigned long stack_size,
2400		- int __user *parent_tidptr,
2401		- int __user *child_tidptr)
2402		-{
2403		- return _do_fork(clone_flags, stack_start, stack_size,
2404		- parent_tidptr, child_tidptr, 0);
2405		-}
2406		-#endif
2407		-
2408	2606	/*
2409	2607	* Create a kernel thread.
2410	2608	*/
2411	2609	pid_t kernel_thread(int (fn)(void ), void *arg, unsigned long flags)
2412	2610	{
2413		- return _do_fork(flags\|CLONE_VM\|CLONE_UNTRACED, (unsigned long)fn,
2414		- (unsigned long)arg, NULL, NULL, 0);
	2611	+ struct kernel_clone_args args = {
	2612	+ .flags = ((lower_32_bits(flags) \| CLONE_VM \|
	2613	+ CLONE_UNTRACED) & ~CSIGNAL),
	2614	+ .exit_signal = (lower_32_bits(flags) & CSIGNAL),
	2615	+ .stack = (unsigned long)fn,
	2616	+ .stack_size = (unsigned long)arg,
	2617	+ };
	2618	+
	2619	+ return kernel_clone(&args);
2415	2620	}
2416	2621
2417	2622	#ifdef __ARCH_WANT_SYS_FORK
2418	2623	SYSCALL_DEFINE0(fork)
2419	2624	{
2420	2625	#ifdef CONFIG_MMU
2421		- return _do_fork(SIGCHLD, 0, 0, NULL, NULL, 0);
	2626	+ struct kernel_clone_args args = {
	2627	+ .exit_signal = SIGCHLD,
	2628	+ };
	2629	+
	2630	+ return kernel_clone(&args);
2422	2631	#else
2423	2632	/* can not support in nommu mode */
2424	2633	return -EINVAL;
..	..	@@ -2429,8 +2638,12 @@
2429	2638	#ifdef __ARCH_WANT_SYS_VFORK
2430	2639	SYSCALL_DEFINE0(vfork)
2431	2640	{
2432		- return _do_fork(CLONE_VFORK \| CLONE_VM \| SIGCHLD, 0,
2433		- 0, NULL, NULL, 0);
	2641	+ struct kernel_clone_args args = {
	2642	+ .flags = CLONE_VFORK \| CLONE_VM,
	2643	+ .exit_signal = SIGCHLD,
	2644	+ };
	2645	+
	2646	+ return kernel_clone(&args);
2434	2647	}
2435	2648	#endif
2436	2649
..	..	@@ -2458,7 +2671,175 @@
2458	2671	unsigned long, tls)
2459	2672	#endif
2460	2673	{
2461		- return _do_fork(clone_flags, newsp, 0, parent_tidptr, child_tidptr, tls);
	2674	+ struct kernel_clone_args args = {
	2675	+ .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
	2676	+ .pidfd = parent_tidptr,
	2677	+ .child_tid = child_tidptr,
	2678	+ .parent_tid = parent_tidptr,
	2679	+ .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
	2680	+ .stack = newsp,
	2681	+ .tls = tls,
	2682	+ };
	2683	+
	2684	+ return kernel_clone(&args);
	2685	+}
	2686	+#endif
	2687	+
	2688	+#ifdef __ARCH_WANT_SYS_CLONE3
	2689	+
	2690	+noinline static int copy_clone_args_from_user(struct kernel_clone_args *kargs,
	2691	+ struct clone_args __user *uargs,
	2692	+ size_t usize)
	2693	+{
	2694	+ int err;
	2695	+ struct clone_args args;
	2696	+ pid_t *kset_tid = kargs->set_tid;
	2697	+
	2698	+ BUILD_BUG_ON(offsetofend(struct clone_args, tls) !=
	2699	+ CLONE_ARGS_SIZE_VER0);
	2700	+ BUILD_BUG_ON(offsetofend(struct clone_args, set_tid_size) !=
	2701	+ CLONE_ARGS_SIZE_VER1);
	2702	+ BUILD_BUG_ON(offsetofend(struct clone_args, cgroup) !=
	2703	+ CLONE_ARGS_SIZE_VER2);
	2704	+ BUILD_BUG_ON(sizeof(struct clone_args) != CLONE_ARGS_SIZE_VER2);
	2705	+
	2706	+ if (unlikely(usize > PAGE_SIZE))
	2707	+ return -E2BIG;
	2708	+ if (unlikely(usize < CLONE_ARGS_SIZE_VER0))
	2709	+ return -EINVAL;
	2710	+
	2711	+ err = copy_struct_from_user(&args, sizeof(args), uargs, usize);
	2712	+ if (err)
	2713	+ return err;
	2714	+
	2715	+ if (unlikely(args.set_tid_size > MAX_PID_NS_LEVEL))
	2716	+ return -EINVAL;
	2717	+
	2718	+ if (unlikely(!args.set_tid && args.set_tid_size > 0))
	2719	+ return -EINVAL;
	2720	+
	2721	+ if (unlikely(args.set_tid && args.set_tid_size == 0))
	2722	+ return -EINVAL;
	2723	+
	2724	+ /*
	2725	+ * Verify that higher 32bits of exit_signal are unset and that
	2726	+ * it is a valid signal
	2727	+ */
	2728	+ if (unlikely((args.exit_signal & ~((u64)CSIGNAL)) \|\|
	2729	+ !valid_signal(args.exit_signal)))
	2730	+ return -EINVAL;
	2731	+
	2732	+ if ((args.flags & CLONE_INTO_CGROUP) &&
	2733	+ (args.cgroup > INT_MAX \|\| usize < CLONE_ARGS_SIZE_VER2))
	2734	+ return -EINVAL;
	2735	+
	2736	+ *kargs = (struct kernel_clone_args){
	2737	+ .flags = args.flags,
	2738	+ .pidfd = u64_to_user_ptr(args.pidfd),
	2739	+ .child_tid = u64_to_user_ptr(args.child_tid),
	2740	+ .parent_tid = u64_to_user_ptr(args.parent_tid),
	2741	+ .exit_signal = args.exit_signal,
	2742	+ .stack = args.stack,
	2743	+ .stack_size = args.stack_size,
	2744	+ .tls = args.tls,
	2745	+ .set_tid_size = args.set_tid_size,
	2746	+ .cgroup = args.cgroup,
	2747	+ };
	2748	+
	2749	+ if (args.set_tid &&
	2750	+ copy_from_user(kset_tid, u64_to_user_ptr(args.set_tid),
	2751	+ (kargs->set_tid_size * sizeof(pid_t))))
	2752	+ return -EFAULT;
	2753	+
	2754	+ kargs->set_tid = kset_tid;
	2755	+
	2756	+ return 0;
	2757	+}
	2758	+
	2759	+/**
	2760	+ * clone3_stack_valid - check and prepare stack
	2761	+ * @kargs: kernel clone args
	2762	+ *
	2763	+ * Verify that the stack arguments userspace gave us are sane.
	2764	+ * In addition, set the stack direction for userspace since it's easy for us to
	2765	+ * determine.
	2766	+ */
	2767	+static inline bool clone3_stack_valid(struct kernel_clone_args *kargs)
	2768	+{
	2769	+ if (kargs->stack == 0) {
	2770	+ if (kargs->stack_size > 0)
	2771	+ return false;
	2772	+ } else {
	2773	+ if (kargs->stack_size == 0)
	2774	+ return false;
	2775	+
	2776	+ if (!access_ok((void __user *)kargs->stack, kargs->stack_size))
	2777	+ return false;
	2778	+
	2779	+#if !defined(CONFIG_STACK_GROWSUP) && !defined(CONFIG_IA64)
	2780	+ kargs->stack += kargs->stack_size;
	2781	+#endif
	2782	+ }
	2783	+
	2784	+ return true;
	2785	+}
	2786	+
	2787	+static bool clone3_args_valid(struct kernel_clone_args *kargs)
	2788	+{
	2789	+ /* Verify that no unknown flags are passed along. */
	2790	+ if (kargs->flags &
	2791	+ ~(CLONE_LEGACY_FLAGS \| CLONE_CLEAR_SIGHAND \| CLONE_INTO_CGROUP))
	2792	+ return false;
	2793	+
	2794	+ /*
	2795	+ * - make the CLONE_DETACHED bit reuseable for clone3
	2796	+ * - make the CSIGNAL bits reuseable for clone3
	2797	+ */
	2798	+ if (kargs->flags & (CLONE_DETACHED \| (CSIGNAL & (~CLONE_NEWTIME))))
	2799	+ return false;
	2800	+
	2801	+ if ((kargs->flags & (CLONE_SIGHAND \| CLONE_CLEAR_SIGHAND)) ==
	2802	+ (CLONE_SIGHAND \| CLONE_CLEAR_SIGHAND))
	2803	+ return false;
	2804	+
	2805	+ if ((kargs->flags & (CLONE_THREAD \| CLONE_PARENT)) &&
	2806	+ kargs->exit_signal)
	2807	+ return false;
	2808	+
	2809	+ if (!clone3_stack_valid(kargs))
	2810	+ return false;
	2811	+
	2812	+ return true;
	2813	+}
	2814	+
	2815	+/**
	2816	+ * clone3 - create a new process with specific properties
	2817	+ * @uargs: argument structure
	2818	+ * @size: size of @uargs
	2819	+ *
	2820	+ * clone3() is the extensible successor to clone()/clone2().
	2821	+ * It takes a struct as argument that is versioned by its size.
	2822	+ *
	2823	+ * Return: On success, a positive PID for the child process.
	2824	+ * On error, a negative errno number.
	2825	+ */
	2826	+SYSCALL_DEFINE2(clone3, struct clone_args __user *, uargs, size_t, size)
	2827	+{
	2828	+ int err;
	2829	+
	2830	+ struct kernel_clone_args kargs;
	2831	+ pid_t set_tid[MAX_PID_NS_LEVEL];
	2832	+
	2833	+ kargs.set_tid = set_tid;
	2834	+
	2835	+ err = copy_clone_args_from_user(&kargs, uargs, size);
	2836	+ if (err)
	2837	+ return err;
	2838	+
	2839	+ if (!clone3_args_valid(&kargs))
	2840	+ return -EINVAL;
	2841	+
	2842	+ return kernel_clone(&kargs);
2462	2843	}
2463	2844	#endif
2464	2845
..	..	@@ -2506,10 +2887,27 @@
2506	2887	init_waitqueue_head(&sighand->signalfd_wqh);
2507	2888	}
2508	2889
2509		-void __init proc_caches_init(void)
	2890	+void __init mm_cache_init(void)
2510	2891	{
2511	2892	unsigned int mm_size;
2512	2893
	2894	+ /*
	2895	+ * The mm_cpumask is located at the end of mm_struct, and is
	2896	+ * dynamically sized based on the maximum CPU number this system
	2897	+ * can have, taking hotplug into account (nr_cpu_ids).
	2898	+ */
	2899	+ mm_size = sizeof(struct mm_struct) + cpumask_size();
	2900	+
	2901	+ mm_cachep = kmem_cache_create_usercopy("mm_struct",
	2902	+ mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
	2903	+ SLAB_HWCACHE_ALIGN\|SLAB_PANIC\|SLAB_ACCOUNT,
	2904	+ offsetof(struct mm_struct, saved_auxv),
	2905	+ sizeof_field(struct mm_struct, saved_auxv),
	2906	+ NULL);
	2907	+}
	2908	+
	2909	+void __init proc_caches_init(void)
	2910	+{
2513	2911	sighand_cachep = kmem_cache_create("sighand_cache",
2514	2912	sizeof(struct sighand_struct), 0,
2515	2913	SLAB_HWCACHE_ALIGN\|SLAB_PANIC\|SLAB_TYPESAFE_BY_RCU\|
..	..	@@ -2527,19 +2925,6 @@
2527	2925	SLAB_HWCACHE_ALIGN\|SLAB_PANIC\|SLAB_ACCOUNT,
2528	2926	NULL);
2529	2927
2530		- /*
2531		- * The mm_cpumask is located at the end of mm_struct, and is
2532		- * dynamically sized based on the maximum CPU number this system
2533		- * can have, taking hotplug into account (nr_cpu_ids).
2534		- */
2535		- mm_size = sizeof(struct mm_struct) + cpumask_size();
2536		-
2537		- mm_cachep = kmem_cache_create_usercopy("mm_struct",
2538		- mm_size, ARCH_MIN_MMSTRUCT_ALIGN,
2539		- SLAB_HWCACHE_ALIGN\|SLAB_PANIC\|SLAB_ACCOUNT,
2540		- offsetof(struct mm_struct, saved_auxv),
2541		- sizeof_field(struct mm_struct, saved_auxv),
2542		- NULL);
2543	2928	vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC\|SLAB_ACCOUNT);
2544	2929	mmap_init();
2545	2930	nsproxy_cache_init();
..	..	@@ -2553,7 +2938,8 @@
2553	2938	if (unshare_flags & ~(CLONE_THREAD\|CLONE_FS\|CLONE_NEWNS\|CLONE_SIGHAND\|
2554	2939	CLONE_VM\|CLONE_FILES\|CLONE_SYSVSEM\|
2555	2940	CLONE_NEWUTS\|CLONE_NEWIPC\|CLONE_NEWNET\|
2556		- CLONE_NEWUSER\|CLONE_NEWPID\|CLONE_NEWCGROUP))
	2941	+ CLONE_NEWUSER\|CLONE_NEWPID\|CLONE_NEWCGROUP\|
	2942	+ CLONE_NEWTIME))
2557	2943	return -EINVAL;
2558	2944	/*
2559	2945	* Not implemented, but pretend it works if there is nothing
..	..	@@ -2566,7 +2952,7 @@
2566	2952	return -EINVAL;
2567	2953	}
2568	2954	if (unshare_flags & (CLONE_SIGHAND \| CLONE_VM)) {
2569		- if (atomic_read(&current->sighand->count) > 1)
	2955	+ if (refcount_read(&current->sighand->count) > 1)
2570	2956	return -EINVAL;
2571	2957	}
2572	2958	if (unshare_flags & CLONE_VM) {
..	..	@@ -2601,14 +2987,15 @@
2601	2987	/*
2602	2988	* Unshare file descriptor table if it is being shared
2603	2989	*/
2604		-static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp)
	2990	+int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
	2991	+ struct files_struct **new_fdp)
2605	2992	{
2606	2993	struct files_struct *fd = current->files;
2607	2994	int error = 0;
2608	2995
2609	2996	if ((unshare_flags & CLONE_FILES) &&
2610	2997	(fd && atomic_read(&fd->count) > 1)) {
2611		- *new_fdp = dup_fd(fd, &error);
	2998	+ *new_fdp = dup_fd(fd, max_fds, &error);
2612	2999	if (!*new_fdp)
2613	3000	return error;
2614	3001	}
..	..	@@ -2619,7 +3006,7 @@
2619	3006	/*
2620	3007	* unshare allows a process to 'unshare' part of the process
2621	3008	* context which was originally shared using clone. copy_*
2622		- * functions used by do_fork() cannot be used here directly
	3009	+ * functions used by kernel_clone() cannot be used here directly
2623	3010	* because they modify an inactive task_struct that is being
2624	3011	* constructed. Here we are modifying the current, active,
2625	3012	* task_struct.
..	..	@@ -2668,7 +3055,7 @@
2668	3055	err = unshare_fs(unshare_flags, &new_fs);
2669	3056	if (err)
2670	3057	goto bad_unshare_out;
2671		- err = unshare_fd(unshare_flags, &new_fd);
	3058	+ err = unshare_fd(unshare_flags, NR_OPEN_MAX, &new_fd);
2672	3059	if (err)
2673	3060	goto bad_unshare_cleanup_fs;
2674	3061	err = unshare_userns(unshare_flags, &new_cred);
..	..	@@ -2757,7 +3144,7 @@
2757	3144	struct files_struct *copy = NULL;
2758	3145	int error;
2759	3146
2760		- error = unshare_fd(CLONE_FILES, &copy);
	3147	+ error = unshare_fd(CLONE_FILES, NR_OPEN_MAX, &copy);
2761	3148	if (error \|\| !copy) {
2762	3149	*displaced = NULL;
2763	3150	return error;
..	..	@@ -2770,7 +3157,7 @@
2770	3157	}
2771	3158
2772	3159	int sysctl_max_threads(struct ctl_table *table, int write,
2773		- void __user buffer, size_t lenp, loff_t *ppos)
	3160	+ void buffer, size_t lenp, loff_t *ppos)
2774	3161	{
2775	3162	struct ctl_table t;
2776	3163	int ret;