Merge tag 'for-6.3/io_uring-2023-02-16' of git://git.kernel.dk/linux

Pull io_uring updates from Jens Axboe:

 - Cleanup series making the async prep and handling of
   REQ_F_FORCE_ASYNC easier to follow and verify (Dylan)

 - Enable specifying specific flags for OP_MSG_RING (Breno)

 - Enable use of KASAN with the internal request cache (Breno)

 - Split the opcode definition structs into a hot and cold part (Breno)

 - OP_MSG_RING fixes (Pavel, me)

 - Fix an issue with IOPOLL cancelation and PREEMPT_NONE (me)

 - Handle TIF_NOTIFY_RESUME for the io-wq threads that never return to
   userspace (me)

 - Add support for using io_uring_register() with a registered ring fd
   (Josh)

 - Improve handling of poll on the ring fd (Pavel)

 - Series improving the task_work handling (Pavel)

 - Misc cleanups, fixes, improvements (Dmitrii, Quanfa, Richard, Pavel,
   me)

* tag 'for-6.3/io_uring-2023-02-16' of git://git.kernel.dk/linux: (51 commits)
  io_uring: Support calling io_uring_register with a registered ring fd
  io_uring,audit: don't log IORING_OP_MADVISE
  io_uring: mark task TASK_RUNNING before handling resume/task work
  io_uring: always go async for unsupported open flags
  io_uring: always go async for unsupported fadvise flags
  io_uring: for requests that require async, force it
  io_uring: if a linked request has REQ_F_FORCE_ASYNC then run it async
  io_uring: add reschedule point to handle_tw_list()
  io_uring: add a conditional reschedule to the IOPOLL cancelation loop
  io_uring: return normal tw run linking optimisation
  io_uring: refactor tctx_task_work
  io_uring: refactor io_put_task helpers
  io_uring: refactor req allocation
  io_uring: improve io_get_sqe
  io_uring: kill outdated comment about overflow flush
  io_uring: use user visible tail in io_uring_poll()
  io_uring: pass in io_issue_def to io_assign_file()
  io_uring: Enable KASAN for request cache
  io_uring: handle TIF_NOTIFY_RESUME when checking for task_work
  io_uring/msg-ring: ensure flags passing works for task_work completions
  ...

20 files changed, 699 insertions, 409 deletions

diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index 128a67a40065..0efe4d784358 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -195,21 +195,23 @@ struct io_alloc_cache {
 struct io_ring_ctx {
 	/* const or read-mostly hot data */
 	struct {
-		struct percpu_ref	refs;
-
-		struct io_rings		*rings;
 		unsigned int		flags;
-		enum task_work_notify_mode	notify_method;
-		unsigned int		compat: 1;
 		unsigned int		drain_next: 1;
 		unsigned int		restricted: 1;
 		unsigned int		off_timeout_used: 1;
 		unsigned int		drain_active: 1;
-		unsigned int		drain_disabled: 1;
 		unsigned int		has_evfd: 1;
-		unsigned int		syscall_iopoll: 1;
 		/* all CQEs should be posted only by the submitter task */
 		unsigned int		task_complete: 1;
+		unsigned int		syscall_iopoll: 1;
+		unsigned int		poll_activated: 1;
+		unsigned int		drain_disabled: 1;
+		unsigned int		compat: 1;
+
+		enum task_work_notify_mode	notify_method;
+		struct io_rings			*rings;
+		struct task_struct		*submitter_task;
+		struct percpu_ref		refs;
 	} ____cacheline_aligned_in_smp;
 
 	/* submission data */
@@ -293,6 +295,7 @@ struct io_ring_ctx {
 		spinlock_t		completion_lock;
 
 		bool			poll_multi_queue;
+		bool			cq_waiting;
 
 		/*
 		 * ->iopoll_list is protected by the ctx->uring_lock for
@@ -318,9 +321,8 @@ struct io_ring_ctx {
 	} ____cacheline_aligned_in_smp;
 
 	/* Keep this last, we don't need it for the fast path */
-
+	struct wait_queue_head		poll_wq;
 	struct io_restriction		restrictions;
-	struct task_struct		*submitter_task;
 
 	/* slow path rsrc auxilary data, used by update/register */
 	struct io_rsrc_node		*rsrc_backup_node;
@@ -357,6 +359,7 @@ struct io_ring_ctx {
 	u32				iowq_limits[2];
 	bool				iowq_limits_set;
 
+	struct callback_head		poll_wq_task_work;
 	struct list_head		defer_list;
 	unsigned			sq_thread_idle;
 	/* protected by ->completion_lock */
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 2780bce62faf..97661a60b28c 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -347,6 +347,8 @@ enum {
  *				applicable for IORING_MSG_DATA, obviously.
  */
 #define IORING_MSG_RING_CQE_SKIP	(1U << 0)
+/* Pass through the flags from sqe->file_index to cqe->flags */
+#define IORING_MSG_RING_FLAGS_PASS	(1U << 1)
 
 /*
  * IO completion data structure (Completion Queue Entry)
@@ -470,6 +472,7 @@ struct io_uring_params {
 #define IORING_FEAT_RSRC_TAGS		(1U << 10)
 #define IORING_FEAT_CQE_SKIP		(1U << 11)
 #define IORING_FEAT_LINKED_FILE		(1U << 12)
+#define IORING_FEAT_REG_REG_RING	(1U << 13)
 
 /*
  * io_uring_register(2) opcodes and arguments
@@ -517,7 +520,10 @@ enum {
 	IORING_REGISTER_FILE_ALLOC_RANGE	= 25,
 
 	/* this goes last */
-	IORING_REGISTER_LAST
+	IORING_REGISTER_LAST,
+
+	/* flag added to the opcode to use a registered ring fd */
+	IORING_REGISTER_USE_REGISTERED_RING	= 1U << 31
 };
 
 /* io-wq worker categories */
diff --git a/io_uring/advise.c b/io_uring/advise.c
index 449c6f14649f..7085804c513c 100644
--- a/io_uring/advise.c
+++ b/io_uring/advise.c
@@ -39,6 +39,7 @@ int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	ma->addr = READ_ONCE(sqe->addr);
 	ma->len = READ_ONCE(sqe->len);
 	ma->advice = READ_ONCE(sqe->fadvise_advice);
+	req->flags |= REQ_F_FORCE_ASYNC;
 	return 0;
 #else
 	return -EOPNOTSUPP;
@@ -51,8 +52,7 @@ int io_madvise(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_madvise *ma = io_kiocb_to_cmd(req, struct io_madvise);
 	int ret;
 
-	if (issue_flags & IO_URING_F_NONBLOCK)
-		return -EAGAIN;
+	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
 
 	ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice);
 	io_req_set_res(req, ret, 0);
@@ -62,6 +62,18 @@ int io_madvise(struct io_kiocb *req, unsigned int issue_flags)
 #endif
 }
 
+static bool io_fadvise_force_async(struct io_fadvise *fa)
+{
+	switch (fa->advice) {
+	case POSIX_FADV_NORMAL:
+	case POSIX_FADV_RANDOM:
+	case POSIX_FADV_SEQUENTIAL:
+		return false;
+	default:
+		return true;
+	}
+}
+
 int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 {
 	struct io_fadvise *fa = io_kiocb_to_cmd(req, struct io_fadvise);
@@ -72,6 +84,8 @@ int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	fa->offset = READ_ONCE(sqe->off);
 	fa->len = READ_ONCE(sqe->len);
 	fa->advice = READ_ONCE(sqe->fadvise_advice);
+	if (io_fadvise_force_async(fa))
+		req->flags |= REQ_F_FORCE_ASYNC;
 	return 0;
 }
 
@@ -80,16 +94,7 @@ int io_fadvise(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_fadvise *fa = io_kiocb_to_cmd(req, struct io_fadvise);
 	int ret;
 
-	if (issue_flags & IO_URING_F_NONBLOCK) {
-		switch (fa->advice) {
-		case POSIX_FADV_NORMAL:
-		case POSIX_FADV_RANDOM:
-		case POSIX_FADV_SEQUENTIAL:
-			break;
-		default:
-			return -EAGAIN;
-		}
-	}
+	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK && io_fadvise_force_async(fa));
 
 	ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
 	if (ret < 0)
diff --git a/io_uring/fs.c b/io_uring/fs.c
index 7100c293c13a..f6a69a549fd4 100644
--- a/io_uring/fs.c
+++ b/io_uring/fs.c
@@ -74,6 +74,7 @@ int io_renameat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	}
 
 	req->flags |= REQ_F_NEED_CLEANUP;
+	req->flags |= REQ_F_FORCE_ASYNC;
 	return 0;
 }
 
@@ -82,8 +83,7 @@ int io_renameat(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_rename *ren = io_kiocb_to_cmd(req, struct io_rename);
 	int ret;
 
-	if (issue_flags & IO_URING_F_NONBLOCK)
-		return -EAGAIN;
+	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
 
 	ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd,
 				ren->newpath, ren->flags);
@@ -123,6 +123,7 @@ int io_unlinkat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 		return PTR_ERR(un->filename);
 
 	req->flags |= REQ_F_NEED_CLEANUP;
+	req->flags |= REQ_F_FORCE_ASYNC;
 	return 0;
 }
 
@@ -131,8 +132,7 @@ int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_unlink *un = io_kiocb_to_cmd(req, struct io_unlink);
 	int ret;
 
-	if (issue_flags & IO_URING_F_NONBLOCK)
-		return -EAGAIN;
+	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
 
 	if (un->flags & AT_REMOVEDIR)
 		ret = do_rmdir(un->dfd, un->filename);
@@ -170,6 +170,7 @@ int io_mkdirat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 		return PTR_ERR(mkd->filename);
 
 	req->flags |= REQ_F_NEED_CLEANUP;
+	req->flags |= REQ_F_FORCE_ASYNC;
 	return 0;
 }
 
@@ -178,8 +179,7 @@ int io_mkdirat(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_mkdir *mkd = io_kiocb_to_cmd(req, struct io_mkdir);
 	int ret;
 
-	if (issue_flags & IO_URING_F_NONBLOCK)
-		return -EAGAIN;
+	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
 
 	ret = do_mkdirat(mkd->dfd, mkd->filename, mkd->mode);
 
@@ -220,6 +220,7 @@ int io_symlinkat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	}
 
 	req->flags |= REQ_F_NEED_CLEANUP;
+	req->flags |= REQ_F_FORCE_ASYNC;
 	return 0;
 }
 
@@ -228,8 +229,7 @@ int io_symlinkat(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_link *sl = io_kiocb_to_cmd(req, struct io_link);
 	int ret;
 
-	if (issue_flags & IO_URING_F_NONBLOCK)
-		return -EAGAIN;
+	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
 
 	ret = do_symlinkat(sl->oldpath, sl->new_dfd, sl->newpath);
 
@@ -265,6 +265,7 @@ int io_linkat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	}
 
 	req->flags |= REQ_F_NEED_CLEANUP;
+	req->flags |= REQ_F_FORCE_ASYNC;
 	return 0;
 }
 
@@ -273,8 +274,7 @@ int io_linkat(struct io_kiocb *req, unsigned int issue_flags)
 	struct io_link *lnk = io_kiocb_to_cmd(req, struct io_link);
 	int ret;
 
-	if (issue_flags & IO_URING_F_NONBLOCK)
-		return -EAGAIN;
+	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
 
 	ret = do_linkat(lnk->old_dfd, lnk->oldpath, lnk->new_dfd,
 				lnk->newpath, lnk->flags);
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index db623b3185c8..3b915deb4d08 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -151,7 +151,7 @@ static void io_move_task_work_from_local(struct io_ring_ctx *ctx);
 static void __io_submit_flush_completions(struct io_ring_ctx *ctx);
 static __cold void io_fallback_tw(struct io_uring_task *tctx);
 
-static struct kmem_cache *req_cachep;
+struct kmem_cache *req_cachep;
 
 struct sock *io_uring_get_socket(struct file *file)
 {
@@ -230,6 +230,7 @@ static inline void req_fail_link_node(struct io_kiocb *req, int res)
 static inline void io_req_add_to_cache(struct io_kiocb *req, struct io_ring_ctx *ctx)
 {
 	wq_stack_add_head(&req->comp_list, &ctx->submit_state.free_list);
+	kasan_poison_object_data(req_cachep, req);
 }
 
 static __cold void io_ring_ctx_ref_free(struct percpu_ref *ref)
@@ -245,17 +246,15 @@ static __cold void io_fallback_req_func(struct work_struct *work)
 						fallback_work.work);
 	struct llist_node *node = llist_del_all(&ctx->fallback_llist);
 	struct io_kiocb *req, *tmp;
-	bool locked = false;
+	bool locked = true;
 
-	percpu_ref_get(&ctx->refs);
+	mutex_lock(&ctx->uring_lock);
 	llist_for_each_entry_safe(req, tmp, node, io_task_work.node)
 		req->io_task_work.func(req, &locked);
-
-	if (locked) {
-		io_submit_flush_completions(ctx);
-		mutex_unlock(&ctx->uring_lock);
-	}
-	percpu_ref_put(&ctx->refs);
+	if (WARN_ON_ONCE(!locked))
+		return;
+	io_submit_flush_completions(ctx);
+	mutex_unlock(&ctx->uring_lock);
 }
 
 static int io_alloc_hash_table(struct io_hash_table *table, unsigned bits)
@@ -316,6 +315,7 @@ static __cold struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 	xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1);
 	mutex_init(&ctx->uring_lock);
 	init_waitqueue_head(&ctx->cq_wait);
+	init_waitqueue_head(&ctx->poll_wq);
 	spin_lock_init(&ctx->completion_lock);
 	spin_lock_init(&ctx->timeout_lock);
 	INIT_WQ_LIST(&ctx->iopoll_list);
@@ -407,7 +407,7 @@ static inline void io_arm_ltimeout(struct io_kiocb *req)
 
 static void io_prep_async_work(struct io_kiocb *req)
 {
-	const struct io_op_def *def = &io_op_defs[req->opcode];
+	const struct io_issue_def *def = &io_issue_defs[req->opcode];
 	struct io_ring_ctx *ctx = req->ctx;
 
 	if (!(req->flags & REQ_F_CREDS)) {
@@ -572,6 +572,8 @@ static void io_eventfd_flush_signal(struct io_ring_ctx *ctx)
 
 void __io_commit_cqring_flush(struct io_ring_ctx *ctx)
 {
+	if (ctx->poll_activated)
+		io_poll_wq_wake(ctx);
 	if (ctx->off_timeout_used)
 		io_flush_timeouts(ctx);
 	if (ctx->drain_active) {
@@ -618,6 +620,25 @@ static inline void __io_cq_unlock_post(struct io_ring_ctx *ctx)
 	io_cqring_wake(ctx);
 }
 
+static inline void __io_cq_unlock_post_flush(struct io_ring_ctx *ctx)
+	__releases(ctx->completion_lock)
+{
+	io_commit_cqring(ctx);
+	__io_cq_unlock(ctx);
+	io_commit_cqring_flush(ctx);
+
+	/*
+	 * As ->task_complete implies that the ring is single tasked, cq_wait
+	 * may only be waited on by the current in io_cqring_wait(), but since
+	 * it will re-check the wakeup conditions once we return we can safely
+	 * skip waking it up.
+	 */
+	if (!(ctx->flags & IORING_SETUP_DEFER_TASKRUN)) {
+		smp_mb();
+		__io_cqring_wake(ctx);
+	}
+}
+
 void io_cq_unlock_post(struct io_ring_ctx *ctx)
 	__releases(ctx->completion_lock)
 {
@@ -645,7 +666,6 @@ static void io_cqring_overflow_kill(struct io_ring_ctx *ctx)
 	}
 }
 
-/* Returns true if there are no backlogged entries after the flush */
 static void __io_cqring_overflow_flush(struct io_ring_ctx *ctx)
 {
 	size_t cqe_size = sizeof(struct io_uring_cqe);
@@ -693,7 +713,8 @@ static void io_cqring_overflow_flush(struct io_ring_ctx *ctx)
 		io_cqring_do_overflow_flush(ctx);
 }
 
-void __io_put_task(struct task_struct *task, int nr)
+/* can be called by any task */
+static void io_put_task_remote(struct task_struct *task, int nr)
 {
 	struct io_uring_task *tctx = task->io_uring;
 
@@ -703,6 +724,21 @@ void __io_put_task(struct task_struct *task, int nr)
 	put_task_struct_many(task, nr);
 }
 
+/* used by a task to put its own references */
+static void io_put_task_local(struct task_struct *task, int nr)
+{
+	task->io_uring->cached_refs += nr;
+}
+
+/* must to be called somewhat shortly after putting a request */
+static inline void io_put_task(struct task_struct *task, int nr)
+{
+	if (likely(task == current))
+		io_put_task_local(task, nr);
+	else
+		io_put_task_remote(task, nr);
+}
+
 void io_task_refs_refill(struct io_uring_task *tctx)
 {
 	unsigned int refill = -tctx->cached_refs + IO_TCTX_REFS_CACHE_NR;
@@ -945,15 +981,15 @@ static void __io_req_complete_post(struct io_kiocb *req)
 				req->link = NULL;
 			}
 		}
+		io_put_kbuf_comp(req);
+		io_dismantle_req(req);
 		io_req_put_rsrc(req);
 		/*
 		 * Selected buffer deallocation in io_clean_op() assumes that
 		 * we don't hold ->completion_lock. Clean them here to avoid
 		 * deadlocks.
 		 */
-		io_put_kbuf_comp(req);
-		io_dismantle_req(req);
-		io_put_task(req->task, 1);
+		io_put_task_remote(req->task, 1);
 		wq_list_add_head(&req->comp_list, &ctx->locked_free_list);
 		ctx->locked_free_nr++;
 	}
@@ -980,7 +1016,7 @@ void io_req_complete_post(struct io_kiocb *req, unsigned issue_flags)
 void io_req_defer_failed(struct io_kiocb *req, s32 res)
 	__must_hold(&ctx->uring_lock)
 {
-	const struct io_op_def *def = &io_op_defs[req->opcode];
+	const struct io_cold_def *def = &io_cold_defs[req->opcode];
 
 	lockdep_assert_held(&req->ctx->uring_lock);
 
@@ -1076,7 +1112,7 @@ __cold void io_free_req(struct io_kiocb *req)
 
 	io_req_put_rsrc(req);
 	io_dismantle_req(req);
-	io_put_task(req->task, 1);
+	io_put_task_remote(req->task, 1);
 
 	spin_lock(&ctx->completion_lock);
 	wq_list_add_head(&req->comp_list, &ctx->locked_free_list);
@@ -1130,7 +1166,7 @@ static unsigned int handle_tw_list(struct llist_node *node,
 {
 	unsigned int count = 0;
 
-	while (node != last) {
+	while (node && node != last) {
 		struct llist_node *next = node->next;
 		struct io_kiocb *req = container_of(node, struct io_kiocb,
 						    io_task_work.node);
@@ -1143,10 +1179,16 @@ static unsigned int handle_tw_list(struct llist_node *node,
 			/* if not contended, grab and improve batching */
 			*locked = mutex_trylock(&(*ctx)->uring_lock);
 			percpu_ref_get(&(*ctx)->refs);
-		}
+		} else if (!*locked)
+			*locked = mutex_trylock(&(*ctx)->uring_lock);
 		req->io_task_work.func(req, locked);
 		node = next;
 		count++;
+		if (unlikely(need_resched())) {
+			ctx_flush_and_put(*ctx, locked);
+			*ctx = NULL;
+			cond_resched();
+		}
 	}
 
 	return count;
@@ -1190,23 +1232,29 @@ void tctx_task_work(struct callback_head *cb)
 						  task_work);
 	struct llist_node fake = {};
 	struct llist_node *node;
-	unsigned int loops = 1;
-	unsigned int count;
+	unsigned int loops = 0;
+	unsigned int count = 0;
 
 	if (unlikely(current->flags & PF_EXITING)) {
 		io_fallback_tw(tctx);
 		return;
 	}
 
-	node = io_llist_xchg(&tctx->task_list, &fake);
-	count = handle_tw_list(node, &ctx, &uring_locked, NULL);
-	node = io_llist_cmpxchg(&tctx->task_list, &fake, NULL);
-	while (node != &fake) {
+	do {
 		loops++;
 		node = io_llist_xchg(&tctx->task_list, &fake);
 		count += handle_tw_list(node, &ctx, &uring_locked, &fake);
+
+		/* skip expensive cmpxchg if there are items in the list */
+		if (READ_ONCE(tctx->task_list.first) != &fake)
+			continue;
+		if (uring_locked && !wq_list_empty(&ctx->submit_state.compl_reqs)) {
+			io_submit_flush_completions(ctx);
+			if (READ_ONCE(tctx->task_list.first) != &fake)
+				continue;
+		}
 		node = io_llist_cmpxchg(&tctx->task_list, &fake, NULL);
-	}
+	} while (node != &fake);
 
 	ctx_flush_and_put(ctx, &uring_locked);
 
@@ -1241,7 +1289,7 @@ static void io_req_local_work_add(struct io_kiocb *req)
 		percpu_ref_put(&ctx->refs);
 		return;
 	}
-	/* need it for the following io_cqring_wake() */
+	/* needed for the following wake up */
 	smp_mb__after_atomic();
 
 	if (unlikely(atomic_read(&req->task->io_uring->in_idle))) {
@@ -1252,10 +1300,11 @@ static void io_req_local_work_add(struct io_kiocb *req)
 
 	if (ctx->flags & IORING_SETUP_TASKRUN_FLAG)
 		atomic_or(IORING_SQ_TASKRUN, &ctx->rings->sq_flags);
-
 	if (ctx->has_evfd)
 		io_eventfd_signal(ctx);
-	__io_cqring_wake(ctx);
+
+	if (READ_ONCE(ctx->cq_waiting))
+		wake_up_state(ctx->submitter_task, TASK_INTERRUPTIBLE);
 	percpu_ref_put(&ctx->refs);
 }
 
@@ -1296,21 +1345,19 @@ static void __cold io_move_task_work_from_local(struct io_ring_ctx *ctx)
 	}
 }
 
-int __io_run_local_work(struct io_ring_ctx *ctx, bool *locked)
+static int __io_run_local_work(struct io_ring_ctx *ctx, bool *locked)
 {
 	struct llist_node *node;
-	struct llist_node fake;
-	struct llist_node *current_final = NULL;
-	int ret;
-	unsigned int loops = 1;
+	unsigned int loops = 0;
+	int ret = 0;
 
-	if (unlikely(ctx->submitter_task != current))
+	if (WARN_ON_ONCE(ctx->submitter_task != current))
 		return -EEXIST;
-
-	node = io_llist_xchg(&ctx->work_llist, &fake);
-	ret = 0;
+	if (ctx->flags & IORING_SETUP_TASKRUN_FLAG)
+		atomic_andnot(IORING_SQ_TASKRUN, &ctx->rings->sq_flags);
 again:
-	while (node != current_final) {
+	node = io_llist_xchg(&ctx->work_llist, NULL);
+	while (node) {
 		struct llist_node *next = node->next;
 		struct io_kiocb *req = container_of(node, struct io_kiocb,
 						    io_task_work.node);
@@ -1319,26 +1366,20 @@ again:
 		ret++;
 		node = next;
 	}
+	loops++;
 
-	if (ctx->flags & IORING_SETUP_TASKRUN_FLAG)
-		atomic_andnot(IORING_SQ_TASKRUN, &ctx->rings->sq_flags);
-
-	node = io_llist_cmpxchg(&ctx->work_llist, &fake, NULL);
-	if (node != &fake) {
-		loops++;
-		current_final = &fake;
-		node = io_llist_xchg(&ctx->work_llist, &fake);
+	if (!llist_empty(&ctx->work_llist))
 		goto again;
-	}
-
-	if (*locked)
+	if (*locked) {
 		io_submit_flush_completions(ctx);
+		if (!llist_empty(&ctx->work_llist))
+			goto again;
+	}
 	trace_io_uring_local_work_run(ctx, ret, loops);
 	return ret;
-
 }
 
-int io_run_local_work(struct io_ring_ctx *ctx)
+static inline int io_run_local_work_locked(struct io_ring_ctx *ctx)
 {
 	bool locked;
 	int ret;
@@ -1346,8 +1387,19 @@ int io_run_local_work(struct io_ring_ctx *ctx)
 	if (llist_empty(&ctx->work_llist))
 		return 0;
 
-	__set_current_state(TASK_RUNNING);
-	locked = mutex_trylock(&ctx->uring_lock);
+	locked = true;
+	ret = __io_run_local_work(ctx, &locked);
+	/* shouldn't happen! */
+	if (WARN_ON_ONCE(!locked))
+		mutex_lock(&ctx->uring_lock);
+	return ret;
+}
+
+static int io_run_local_work(struct io_ring_ctx *ctx)
+{
+	bool locked = mutex_trylock(&ctx->uring_lock);
+	int ret;
+
 	ret = __io_run_local_work(ctx, &locked);
 	if (locked)
 		mutex_unlock(&ctx->uring_lock);
@@ -1365,10 +1417,12 @@ void io_req_task_submit(struct io_kiocb *req, bool *locked)
 {
 	io_tw_lock(req->ctx, locked);
 	/* req->task == current here, checking PF_EXITING is safe */
-	if (likely(!(req->task->flags & PF_EXITING)))
-		io_queue_sqe(req);
-	else
+	if (unlikely(req->task->flags & PF_EXITING))
 		io_req_defer_failed(req, -EFAULT);
+	else if (req->flags & REQ_F_FORCE_ASYNC)
+		io_queue_iowq(req, locked);
+	else
+		io_queue_sqe(req);
 }
 
 void io_req_task_queue_fail(struct io_kiocb *req, int ret)
@@ -1467,7 +1521,7 @@ static void __io_submit_flush_completions(struct io_ring_ctx *ctx)
 			}
 		}
 	}
-	__io_cq_unlock_post(ctx);
+	__io_cq_unlock_post_flush(ctx);
 
 	if (!wq_list_empty(&ctx->submit_state.compl_reqs)) {
 		io_free_batch_list(ctx, state->compl_reqs.first);
@@ -1708,8 +1762,8 @@ unsigned int io_file_get_flags(struct file *file)
 
 bool io_alloc_async_data(struct io_kiocb *req)
 {
-	WARN_ON_ONCE(!io_op_defs[req->opcode].async_size);
-	req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL);
+	WARN_ON_ONCE(!io_cold_defs[req->opcode].async_size);
+	req->async_data = kmalloc(io_cold_defs[req->opcode].async_size, GFP_KERNEL);
 	if (req->async_data) {
 		req->flags |= REQ_F_ASYNC_DATA;
 		return false;
@@ -1719,20 +1773,21 @@ bool io_alloc_async_data(struct io_kiocb *req)
 
 int io_req_prep_async(struct io_kiocb *req)
 {
-	const struct io_op_def *def = &io_op_defs[req->opcode];
+	const struct io_cold_def *cdef = &io_cold_defs[req->opcode];
+	const struct io_issue_def *def = &io_issue_defs[req->opcode];
 
 	/* assign early for deferred execution for non-fixed file */
 	if (def->needs_file && !(req->flags & REQ_F_FIXED_FILE))
 		req->file = io_file_get_normal(req, req->cqe.fd);
-	if (!def->prep_async)
+	if (!cdef->prep_async)
 		return 0;
 	if (WARN_ON_ONCE(req_has_async_data(req)))
 		return -EFAULT;
-	if (!io_op_defs[req->opcode].manual_alloc) {
+	if (!def->manual_alloc) {
 		if (io_alloc_async_data(req))
 			return -EAGAIN;
 	}
-	return def->prep_async(req);
+	return cdef->prep_async(req);
 }
 
 static u32 io_get_sequence(struct io_kiocb *req)
@@ -1796,7 +1851,7 @@ static void io_clean_op(struct io_kiocb *req)
 	}
 
 	if (req->flags & REQ_F_NEED_CLEANUP) {
-		const struct io_op_def *def = &io_op_defs[req->opcode];
+		const struct io_cold_def *def = &io_cold_defs[req->opcode];
 
 		if (def->cleanup)
 			def->cleanup(req);
@@ -1820,9 +1875,10 @@ static void io_clean_op(struct io_kiocb *req)
 	req->flags &= ~IO_REQ_CLEAN_FLAGS;
 }
 
-static bool io_assign_file(struct io_kiocb *req, unsigned int issue_flags)
+static bool io_assign_file(struct io_kiocb *req, const struct io_issue_def *def,
+			   unsigned int issue_flags)
 {
-	if (req->file || !io_op_defs[req->opcode].needs_file)
+	if (req->file || !def->needs_file)
 		return true;
 
 	if (req->flags & REQ_F_FIXED_FILE)
@@ -1835,11 +1891,11 @@ static bool io_assign_file(struct io_kiocb *req, unsigned int issue_flags)
 
 static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
 {
-	const struct io_op_def *def = &io_op_defs[req->opcode];
+	const struct io_issue_def *def = &io_issue_defs[req->opcode];
 	const struct cred *creds = NULL;
 	int ret;
 
-	if (unlikely(!io_assign_file(req, issue_flags)))
+	if (unlikely(!io_assign_file(req, def, issue_flags)))
 		return -EBADF;
 
 	if (unlikely((req->flags & REQ_F_CREDS) && req->creds != current_cred()))
@@ -1889,7 +1945,7 @@ struct io_wq_work *io_wq_free_work(struct io_wq_work *work)
 void io_wq_submit_work(struct io_wq_work *work)
 {
 	struct io_kiocb *req = container_of(work, struct io_kiocb, work);
-	const struct io_op_def *def = &io_op_defs[req->opcode];
+	const struct io_issue_def *def = &io_issue_defs[req->opcode];
 	unsigned int issue_flags = IO_URING_F_UNLOCKED | IO_URING_F_IOWQ;
 	bool needs_poll = false;
 	int ret = 0, err = -ECANCELED;
@@ -1908,7 +1964,7 @@ fail:
 		io_req_task_queue_fail(req, err);
 		return;
 	}
-	if (!io_assign_file(req, issue_flags)) {
+	if (!io_assign_file(req, def, issue_flags)) {
 		err = -EBADF;
 		work->flags |= IO_WQ_WORK_CANCEL;
 		goto fail;
@@ -2104,7 +2160,7 @@ static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
 		       const struct io_uring_sqe *sqe)
 	__must_hold(&ctx->uring_lock)
 {
-	const struct io_op_def *def;
+	const struct io_issue_def *def;
 	unsigned int sqe_flags;
 	int personality;
 	u8 opcode;
@@ -2122,7 +2178,7 @@ static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
 		req->opcode = 0;
 		return -EINVAL;
 	}
-	def = &io_op_defs[opcode];
+	def = &io_issue_defs[opcode];
 	if (unlikely(sqe_flags & ~SQE_COMMON_FLAGS)) {
 		/* enforce forwards compatibility on users */
 		if (sqe_flags & ~SQE_VALID_FLAGS)
@@ -2333,7 +2389,7 @@ static void io_commit_sqring(struct io_ring_ctx *ctx)
  * used, it's important that those reads are done through READ_ONCE() to
  * prevent a re-load down the line.
  */
-static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx)
+static bool io_get_sqe(struct io_ring_ctx *ctx, const struct io_uring_sqe **sqe)
 {
 	unsigned head, mask = ctx->sq_entries - 1;
 	unsigned sq_idx = ctx->cached_sq_head++ & mask;
@@ -2351,14 +2407,15 @@ static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx)
 		/* double index for 128-byte SQEs, twice as long */
 		if (ctx->flags & IORING_SETUP_SQE128)
 			head <<= 1;
-		return &ctx->sq_sqes[head];
+		*sqe = &ctx->sq_sqes[head];
+		return true;
 	}
 
 	/* drop invalid entries */
 	ctx->cq_extra--;
 	WRITE_ONCE(ctx->rings->sq_dropped,
 		   READ_ONCE(ctx->rings->sq_dropped) + 1);
-	return NULL;
+	return false;
 }
 
 int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
@@ -2379,11 +2436,9 @@ int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
 		const struct io_uring_sqe *sqe;
 		struct io_kiocb *req;
 
-		if (unlikely(!io_alloc_req_refill(ctx)))
+		if (unlikely(!io_alloc_req(ctx, &req)))
 			break;
-		req = io_alloc_req(ctx);
-		sqe = io_get_sqe(ctx);
-		if (unlikely(!sqe)) {
+		if (unlikely(!io_get_sqe(ctx, &sqe))) {
 			io_req_add_to_cache(req, ctx);
 			break;
 		}
@@ -2418,13 +2473,13 @@ struct io_wait_queue {
 	struct io_ring_ctx *ctx;
 	unsigned cq_tail;
 	unsigned nr_timeouts;
+	ktime_t timeout;
 };
 
 static inline bool io_has_work(struct io_ring_ctx *ctx)
 {
 	return test_bit(IO_CHECK_CQ_OVERFLOW_BIT, &ctx->check_cq) ||
-	       ((ctx->flags & IORING_SETUP_DEFER_TASKRUN) &&
-		!llist_empty(&ctx->work_llist));
+	       !llist_empty(&ctx->work_llist);
 }
 
 static inline bool io_should_wake(struct io_wait_queue *iowq)
@@ -2443,22 +2498,25 @@ static inline bool io_should_wake(struct io_wait_queue *iowq)
 static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
 			    int wake_flags, void *key)
 {
-	struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
-							wq);
-	struct io_ring_ctx *ctx = iowq->ctx;
+	struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue, wq);
 
 	/*
 	 * Cannot safely flush overflowed CQEs from here, ensure we wake up
 	 * the task, and the next invocation will do it.
 	 */
-	if (io_should_wake(iowq) || io_has_work(ctx))
+	if (io_should_wake(iowq) || io_has_work(iowq->ctx))
 		return autoremove_wake_function(curr, mode, wake_flags, key);
 	return -1;
 }
 
 int io_run_task_work_sig(struct io_ring_ctx *ctx)
 {
-	if (io_run_task_work_ctx(ctx) > 0)
+	if (!llist_empty(&ctx->work_llist)) {
+		__set_current_state(TASK_RUNNING);
+		if (io_run_local_work(ctx) > 0)
+			return 1;
+	}
+	if (io_run_task_work() > 0)
 		return 1;
 	if (task_sigpending(current))
 		return -EINTR;
@@ -2467,35 +2525,23 @@ int io_run_task_work_sig(struct io_ring_ctx *ctx)
 
 /* when returns >0, the caller should retry */
 static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
-					  struct io_wait_queue *iowq,
-					  ktime_t *timeout)
+					  struct io_wait_queue *iowq)
 {
-	int ret;
-	unsigned long check_cq;
-
-	/* make sure we run task_work before checking for signals */
-	ret = io_run_task_work_sig(ctx);
-	if (ret || io_should_wake(iowq))
-		return ret;
-
-	check_cq = READ_ONCE(ctx->check_cq);
-	if (unlikely(check_cq)) {
-		/* let the caller flush overflows, retry */
-		if (check_cq & BIT(IO_CHECK_CQ_OVERFLOW_BIT))
-			return 1;
-		if (check_cq & BIT(IO_CHECK_CQ_DROPPED_BIT))
-			return -EBADR;
-	}
-	if (!schedule_hrtimeout(timeout, HRTIMER_MODE_ABS))
+	if (unlikely(READ_ONCE(ctx->check_cq)))
+		return 1;
+	if (unlikely(!llist_empty(&ctx->work_llist)))
+		return 1;
+	if (unlikely(test_thread_flag(TIF_NOTIFY_SIGNAL)))
+		return 1;
+	if (unlikely(task_sigpending(current)))
+		return -EINTR;