[PATCH v6 0/4] fast poll multishot mode

[PATCH v6 0/4] fast poll multishot mode

[Hao Xu]

Let multishot support multishot mode, currently only add accept as its
first consumer.
theoretical analysis:
  1) when connections come in fast
    - singleshot:
              add accept sqe(userspace) --> accept inline
                              ^                 |
                              |-----------------|
    - multishot:
             add accept sqe(userspace) --> accept inline
                                              ^     |
                                              |--*--|

    we do accept repeatedly in * place until get EAGAIN

  2) when connections come in at a low pressure
    similar thing like 1), we reduce a lot of userspace-kernel context
    switch and useless vfs_poll()


tests:
Did some tests, which goes in this way:

  server    client(multiple)
  accept    connect
  read      write
  write     read
  close     close

Basically, raise up a number of clients(on same machine with server) to
connect to the server, and then write some data to it, the server will
write those data back to the client after it receives them, and then
close the connection after write return. Then the client will read the
data and then close the connection. Here I test 10000 clients connect
one server, data size 128 bytes. And each client has a go routine for
it, so they come to the server in short time.
test 20 times before/after this patchset, time spent:(unit cycle, which
is the return value of clock())
before:
  1930136+1940725+1907981+1947601+1923812+1928226+1911087+1905897+1941075
  +1934374+1906614+1912504+1949110+1908790+1909951+1941672+1969525+1934984
  +1934226+1914385)/20.0 = 1927633.75
after:
  1858905+1917104+1895455+1963963+1892706+1889208+1874175+1904753+1874112
  +1874985+1882706+1884642+1864694+1906508+1916150+1924250+1869060+1889506
  +1871324+1940803)/20.0 = 1894750.45

(1927633.75 - 1894750.45) / 1927633.75 = 1.65%

v1->v2:
 - re-implement it against the reworked poll code

v2->v3:
 - fold in code tweak and clean from Jens
 - use io_issue_sqe rather than io_queue_sqe, since the former one
   return the internal error back which makes more sense
 - remove io_poll_clean() and its friends since they are not needed

v3->v4:
 - move the accept multishot flag to the proper patch
 - typo correction
 - remove improperly added signed-off-by

v4->v5:
 - address some email account issue

v5->v6:
 - support multishot accept with fixed file

Hao Xu (4):
  io_uring: add IORING_ACCEPT_MULTISHOT for accept
  io_uring: add REQ_F_APOLL_MULTISHOT for requests
  io_uring: let fast poll support multishot
  io_uring: implement multishot mode for accept

 fs/io_uring.c                 | 106 +++++++++++++++++++++++++++-------
 include/uapi/linux/io_uring.h |   5 ++
 2 files changed, 90 insertions(+), 21 deletions(-)


base-commit: 1b1d7b4bf1d9948c8dba5ee550459ce7c65ac019
-- 
2.36.0

[PATCH 1/4] io_uring: add IORING_ACCEPT_MULTISHOT for accept

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

add an accept_flag IORING_ACCEPT_MULTISHOT for accept, which is to
support multishot.

Signed-off-by: Hao Xu <howeyxu@tencent.com>
---
 include/uapi/linux/io_uring.h | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 36ec43dc7bf9..15f821af9242 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -232,6 +232,11 @@ enum {
  */
 #define IORING_RECVSEND_POLL_FIRST	(1U << 0)
 
+/*
+ * accept flags stored in sqe->ioprio
+ */
+#define IORING_ACCEPT_MULTISHOT	(1U << 0)
+
 /*
  * IO completion data structure (Completion Queue Entry)
  */
-- 
2.36.0

[PATCH 2/4] io_uring: add REQ_F_APOLL_MULTISHOT for requests

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

Add a flag to indicate multishot mode for fast poll. currently only
accept use it, but there may be more operations leveraging it in the
future. Also add a mask IO_APOLL_MULTI_POLLED which stands for
REQ_F_APOLL_MULTI | REQ_F_POLLED, to make the code short and cleaner.

Signed-off-by: Hao Xu <howeyxu@tencent.com>
---
 fs/io_uring.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index c4bd0faf0b27..e1e550de5956 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -116,6 +116,8 @@
 #define IO_REQ_CLEAN_SLOW_FLAGS (REQ_F_REFCOUNT | REQ_F_LINK | REQ_F_HARDLINK |\
 				 IO_REQ_CLEAN_FLAGS)
 
+#define IO_APOLL_MULTI_POLLED (REQ_F_APOLL_MULTISHOT | REQ_F_POLLED)
+
 #define IO_TCTX_REFS_CACHE_NR	(1U << 10)
 
 struct io_uring {
@@ -812,6 +814,7 @@ enum {
 	REQ_F_SINGLE_POLL_BIT,
 	REQ_F_DOUBLE_POLL_BIT,
 	REQ_F_PARTIAL_IO_BIT,
+	REQ_F_APOLL_MULTISHOT_BIT,
 	/* keep async read/write and isreg together and in order */
 	REQ_F_SUPPORT_NOWAIT_BIT,
 	REQ_F_ISREG_BIT,
@@ -876,6 +879,8 @@ enum {
 	REQ_F_DOUBLE_POLL	= BIT(REQ_F_DOUBLE_POLL_BIT),
 	/* request has already done partial IO */
 	REQ_F_PARTIAL_IO	= BIT(REQ_F_PARTIAL_IO_BIT),
+	/* fast poll multishot mode */
+	REQ_F_APOLL_MULTISHOT	= BIT(REQ_F_APOLL_MULTISHOT_BIT),
 };
 
 struct async_poll {
-- 
2.36.0

[PATCH 3/4] io_uring: let fast poll support multishot

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

For operations like accept, multishot is a useful feature, since we can
reduce a number of accept sqe. Let's integrate it to fast poll, it may
be good for other operations in the future.

Signed-off-by: Hao Xu <howeyxu@tencent.com>
---
 fs/io_uring.c | 47 ++++++++++++++++++++++++++++++++---------------
 1 file changed, 32 insertions(+), 15 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index e1e550de5956..a83405f657e0 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -6011,6 +6011,7 @@ static void io_poll_remove_entries(struct io_kiocb *req)
 	rcu_read_unlock();
 }
 
+static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags);
 /*
  * All poll tw should go through this. Checks for poll events, manages
  * references, does rewait, etc.
@@ -6019,10 +6020,10 @@ static void io_poll_remove_entries(struct io_kiocb *req)
  * either spurious wakeup or multishot CQE is served. 0 when it's done with
  * the request, then the mask is stored in req->cqe.res.
  */
-static int io_poll_check_events(struct io_kiocb *req, bool locked)
+static int io_poll_check_events(struct io_kiocb *req, bool *locked)
 {
 	struct io_ring_ctx *ctx = req->ctx;
-	int v;
+	int v, ret;
 
 	/* req->task == current here, checking PF_EXITING is safe */
 	if (unlikely(req->task->flags & PF_EXITING))
@@ -6046,23 +6047,37 @@ static int io_poll_check_events(struct io_kiocb *req, bool locked)
 			req->cqe.res = vfs_poll(req->file, &pt) & req->apoll_events;
 		}
 
-		/* multishot, just fill an CQE and proceed */
-		if (req->cqe.res && !(req->apoll_events & EPOLLONESHOT)) {
-			__poll_t mask = mangle_poll(req->cqe.res & req->apoll_events);
+		if ((unlikely(!req->cqe.res)))
+			continue;
+		if (req->apoll_events & EPOLLONESHOT)
+			return 0;
+
+		/* multishot, just fill a CQE and proceed */
+		if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
+			__poll_t mask = mangle_poll(req->cqe.res &
+						    req->apoll_events);
 			bool filled;
 
 			spin_lock(&ctx->completion_lock);
-			filled = io_fill_cqe_aux(ctx, req->cqe.user_data, mask,
-						 IORING_CQE_F_MORE);
+			filled = io_fill_cqe_aux(ctx, req->cqe.user_data,
+						 mask, IORING_CQE_F_MORE);
 			io_commit_cqring(ctx);
 			spin_unlock(&ctx->completion_lock);
-			if (unlikely(!filled))
-				return -ECANCELED;
-			io_cqring_ev_posted(ctx);
-		} else if (req->cqe.res) {
-			return 0;
+			if (filled) {
+				io_cqring_ev_posted(ctx);
+				continue;
+			}
+			return -ECANCELED;
 		}
 
+		io_tw_lock(req->ctx, locked);
+		if (unlikely(req->task->flags & PF_EXITING))
+			return -EFAULT;
+		ret = io_issue_sqe(req,
+				   IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
+		if (ret)
+			return ret;
+
 		/*
 		 * Release all references, retry if someone tried to restart
 		 * task_work while we were executing it.
@@ -6077,7 +6092,7 @@ static void io_poll_task_func(struct io_kiocb *req, bool *locked)
 	struct io_ring_ctx *ctx = req->ctx;
 	int ret;
 
-	ret = io_poll_check_events(req, *locked);
+	ret = io_poll_check_events(req, locked);
 	if (ret > 0)
 		return;
 
@@ -6102,7 +6117,7 @@ static void io_apoll_task_func(struct io_kiocb *req, bool *locked)
 	struct io_ring_ctx *ctx = req->ctx;
 	int ret;
 
-	ret = io_poll_check_events(req, *locked);
+	ret = io_poll_check_events(req, locked);
 	if (ret > 0)
 		return;
 
@@ -6343,7 +6358,7 @@ static int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
 	struct io_ring_ctx *ctx = req->ctx;
 	struct async_poll *apoll;
 	struct io_poll_table ipt;
-	__poll_t mask = IO_ASYNC_POLL_COMMON | POLLERR;
+	__poll_t mask = POLLPRI | POLLERR;
 	int ret;
 
 	if (!def->pollin && !def->pollout)
@@ -6352,6 +6367,8 @@ static int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
 		return IO_APOLL_ABORTED;
 	if ((req->flags & (REQ_F_POLLED|REQ_F_PARTIAL_IO)) == REQ_F_POLLED)
 		return IO_APOLL_ABORTED;
+	if (!(req->flags & REQ_F_APOLL_MULTISHOT))
+		mask |= EPOLLONESHOT;
 
 	if (def->pollin) {
 		mask |= POLLIN | POLLRDNORM;
-- 
2.36.0

[PATCH 4/4] io_uring: implement multishot mode for accept

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

Refactor io_accept() to support multishot mode.

theoretical analysis:
  1) when connections come in fast
    - singleshot:
              add accept sqe(userspace) --> accept inline
                              ^                 |
                              |-----------------|
    - multishot:
             add accept sqe(userspace) --> accept inline
                                              ^     |
                                              |--*--|

    we do accept repeatedly in * place until get EAGAIN

  2) when connections come in at a low pressure
    similar thing like 1), we reduce a lot of userspace-kernel context
    switch and useless vfs_poll()

tests:
Did some tests, which goes in this way:

  server    client(multiple)
  accept    connect
  read      write
  write     read
  close     close

Basically, raise up a number of clients(on same machine with server) to
connect to the server, and then write some data to it, the server will
write those data back to the client after it receives them, and then
close the connection after write return. Then the client will read the
data and then close the connection. Here I test 10000 clients connect
one server, data size 128 bytes. And each client has a go routine for
it, so they come to the server in short time.
test 20 times before/after this patchset, time spent:(unit cycle, which
is the return value of clock())
before:
  1930136+1940725+1907981+1947601+1923812+1928226+1911087+1905897+1941075
  +1934374+1906614+1912504+1949110+1908790+1909951+1941672+1969525+1934984
  +1934226+1914385)/20.0 = 1927633.75
after:
  1858905+1917104+1895455+1963963+1892706+1889208+1874175+1904753+1874112
  +1874985+1882706+1884642+1864694+1906508+1916150+1924250+1869060+1889506
  +1871324+1940803)/20.0 = 1894750.45

(1927633.75 - 1894750.45) / 1927633.75 = 1.65%

Signed-off-by: Hao Xu <howeyxu@tencent.com>
---
 fs/io_uring.c | 54 +++++++++++++++++++++++++++++++++++++++++++++------
 1 file changed, 48 insertions(+), 6 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index a83405f657e0..d4752e192ef9 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -1148,6 +1148,7 @@ static const struct io_op_def io_op_defs[] = {
 		.unbound_nonreg_file	= 1,
 		.pollin			= 1,
 		.poll_exclusive		= 1,
+		.ioprio			= 1,	/* used for flags */
 	},
 	[IORING_OP_ASYNC_CANCEL] = {
 		.audit_skip		= 1,
@@ -5762,6 +5763,7 @@ static int io_recv(struct io_kiocb *req, unsigned int issue_flags)
 static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 {
 	struct io_accept *accept = &req->accept;
+	unsigned flags;
 
 	if (sqe->len || sqe->buf_index)
 		return -EINVAL;
@@ -5770,19 +5772,30 @@ static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
 	accept->flags = READ_ONCE(sqe->accept_flags);
 	accept->nofile = rlimit(RLIMIT_NOFILE);
+	flags = READ_ONCE(sqe->ioprio);
+	if (flags & ~IORING_ACCEPT_MULTISHOT)
+		return -EINVAL;
 
 	accept->file_slot = READ_ONCE(sqe->file_index);
-	if (accept->file_slot && (accept->flags & SOCK_CLOEXEC))
-		return -EINVAL;
+	if (accept->file_slot) {
+		if (accept->flags & SOCK_CLOEXEC)
+			return -EINVAL;
+		if (flags & IORING_ACCEPT_MULTISHOT &&
+		    accept->file_slot != IORING_FILE_INDEX_ALLOC)
+			return -EINVAL;
+	}
 	if (accept->flags & ~(SOCK_CLOEXEC | SOCK_NONBLOCK))
 		return -EINVAL;
 	if (SOCK_NONBLOCK != O_NONBLOCK && (accept->flags & SOCK_NONBLOCK))
 		accept->flags = (accept->flags & ~SOCK_NONBLOCK) | O_NONBLOCK;
+	if (flags & IORING_ACCEPT_MULTISHOT)
+		req->flags |= REQ_F_APOLL_MULTISHOT;
 	return 0;
 }
 
 static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
 {
+	struct io_ring_ctx *ctx = req->ctx;
 	struct io_accept *accept = &req->accept;
 	bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
 	unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
@@ -5790,6 +5803,7 @@ static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
 	struct file *file;
 	int ret, fd;
 
+retry:
 	if (!fixed) {
 		fd = __get_unused_fd_flags(accept->flags, accept->nofile);
 		if (unlikely(fd < 0))
@@ -5801,8 +5815,17 @@ static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
 		if (!fixed)
 			put_unused_fd(fd);
 		ret = PTR_ERR(file);
-		if (ret == -EAGAIN && force_nonblock)
-			return -EAGAIN;
+		if (ret == -EAGAIN && force_nonblock) {
+			/*
+			 * if it's multishot and polled, we don't need to
+			 * return EAGAIN to arm the poll infra since it
+			 * has already been done
+			 */
+			if ((req->flags & IO_APOLL_MULTI_POLLED) ==
+			    IO_APOLL_MULTI_POLLED)
+				ret = 0;
+			return ret;
+		}
 		if (ret == -ERESTARTSYS)
 			ret = -EINTR;
 		req_set_fail(req);
@@ -5813,8 +5836,27 @@ static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
 		ret = io_fixed_fd_install(req, issue_flags, file,
 						accept->file_slot);
 	}
-	__io_req_complete(req, issue_flags, ret, 0);
-	return 0;
+
+	if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
+		__io_req_complete(req, issue_flags, ret, 0);
+		return 0;
+	}
+	if (ret >= 0) {
+		bool filled;
+
+		spin_lock(&ctx->completion_lock);
+		filled = io_fill_cqe_aux(ctx, req->cqe.user_data, ret,
+					 IORING_CQE_F_MORE);
+		io_commit_cqring(ctx);
+		spin_unlock(&ctx->completion_lock);
+		if (filled) {
+			io_cqring_ev_posted(ctx);
+			goto retry;
+		}
+		ret = -ECANCELED;
+	}
+
+	return ret;
 }
 
 static int io_connect_prep_async(struct io_kiocb *req)
-- 
2.36.0

Re: [PATCH v6 0/4] fast poll multishot mode

[Jens Axboe]

On Sat, 14 May 2022 22:20:42 +0800, Hao Xu wrote:
> Let multishot support multishot mode, currently only add accept as its
> first consumer.
> theoretical analysis:
>   1) when connections come in fast
>     - singleshot:
>               add accept sqe(userspace) --> accept inline
>                               ^                 |
>                               |-----------------|
>     - multishot:
>              add accept sqe(userspace) --> accept inline
>                                               ^     |
>                                               |--*--|
> 
> [...]

Applied, thanks!

[1/4] io_uring: add IORING_ACCEPT_MULTISHOT for accept
      commit: 390ed29b5e425ba00da2b6113b74a14949f71b02
[2/4] io_uring: add REQ_F_APOLL_MULTISHOT for requests
      commit: 227685ebfaba0bc7e2ddc47cef4556050b6d7a8f
[3/4] io_uring: let fast poll support multishot
      commit: dbc2564cfe0faff439dc46adb8c009589054ea46
[4/4] io_uring: implement multishot mode for accept
      commit: 4e86a2c980137f7be1ea600af5f1f5c8342ecc09

Best regards,
-- 
Jens Axboe

[PATCH 3/4] io_uring: let fast poll support multishot

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

For operations like accept, multishot is a useful feature, since we can
reduce a number of accept sqe. Let's integrate it to fast poll, it may
be good for other operations in the future.

Signed-off-by: Hao Xu <howeyxu@tencent.com>
---
 fs/io_uring.c | 47 ++++++++++++++++++++++++++++++++---------------
 1 file changed, 32 insertions(+), 15 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index c2ee184ac693..e0d12af04cd1 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -5955,6 +5955,7 @@ static void io_poll_remove_entries(struct io_kiocb *req)
 	rcu_read_unlock();
 }
 
+static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags);
 /*
  * All poll tw should go through this. Checks for poll events, manages
  * references, does rewait, etc.
@@ -5963,10 +5964,10 @@ static void io_poll_remove_entries(struct io_kiocb *req)
  * either spurious wakeup or multishot CQE is served. 0 when it's done with
  * the request, then the mask is stored in req->cqe.res.
  */
-static int io_poll_check_events(struct io_kiocb *req, bool locked)
+static int io_poll_check_events(struct io_kiocb *req, bool *locked)
 {
 	struct io_ring_ctx *ctx = req->ctx;
-	int v;
+	int v, ret;
 
 	/* req->task == current here, checking PF_EXITING is safe */
 	if (unlikely(req->task->flags & PF_EXITING))
@@ -5990,23 +5991,37 @@ static int io_poll_check_events(struct io_kiocb *req, bool locked)
 			req->cqe.res = vfs_poll(req->file, &pt) & req->apoll_events;
 		}
 
-		/* multishot, just fill an CQE and proceed */
-		if (req->cqe.res && !(req->apoll_events & EPOLLONESHOT)) {
-			__poll_t mask = mangle_poll(req->cqe.res & req->apoll_events);
+		if ((unlikely(!req->cqe.res)))
+			continue;
+		if (req->apoll_events & EPOLLONESHOT)
+			return 0;
+
+		/* multishot, just fill a CQE and proceed */
+		if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
+			__poll_t mask = mangle_poll(req->cqe.res &
+						    req->apoll_events);
 			bool filled;
 
 			spin_lock(&ctx->completion_lock);
-			filled = io_fill_cqe_aux(ctx, req->cqe.user_data, mask,
-						 IORING_CQE_F_MORE);
+			filled = io_fill_cqe_aux(ctx, req->cqe.user_data,
+						 mask, IORING_CQE_F_MORE);
 			io_commit_cqring(ctx);
 			spin_unlock(&ctx->completion_lock);
-			if (unlikely(!filled))
-				return -ECANCELED;
-			io_cqring_ev_posted(ctx);
-		} else if (req->cqe.res) {
-			return 0;
+			if (filled) {
+				io_cqring_ev_posted(ctx);
+				continue;
+			}
+			return -ECANCELED;
 		}
 
+		io_tw_lock(req->ctx, locked);
+		if (unlikely(req->task->flags & PF_EXITING))
+			return -EFAULT;
+		ret = io_issue_sqe(req,
+				   IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
+		if (ret)
+			return ret;
+
 		/*
 		 * Release all references, retry if someone tried to restart
 		 * task_work while we were executing it.
@@ -6021,7 +6036,7 @@ static void io_poll_task_func(struct io_kiocb *req, bool *locked)
 	struct io_ring_ctx *ctx = req->ctx;
 	int ret;
 
-	ret = io_poll_check_events(req, *locked);
+	ret = io_poll_check_events(req, locked);
 	if (ret > 0)
 		return;
 
@@ -6046,7 +6061,7 @@ static void io_apoll_task_func(struct io_kiocb *req, bool *locked)
 	struct io_ring_ctx *ctx = req->ctx;
 	int ret;
 
-	ret = io_poll_check_events(req, *locked);
+	ret = io_poll_check_events(req, locked);
 	if (ret > 0)
 		return;
 
@@ -6286,7 +6301,7 @@ static int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
 	struct io_ring_ctx *ctx = req->ctx;
 	struct async_poll *apoll;
 	struct io_poll_table ipt;
-	__poll_t mask = EPOLLONESHOT | POLLERR | POLLPRI;
+	__poll_t mask = POLLERR | POLLPRI;
 	int ret;
 
 	if (!def->pollin && !def->pollout)
@@ -6295,6 +6310,8 @@ static int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
 		return IO_APOLL_ABORTED;
 	if ((req->flags & (REQ_F_POLLED|REQ_F_PARTIAL_IO)) == REQ_F_POLLED)
 		return IO_APOLL_ABORTED;
+	if (!(req->flags & REQ_F_APOLL_MULTISHOT))
+		mask |= EPOLLONESHOT;
 
 	if (def->pollin) {
 		mask |= POLLIN | POLLRDNORM;
-- 
2.25.1

[PATCH 3/4] io_uring: let fast poll support multishot

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

For operations like accept, multishot is a useful feature, since we can
reduce a number of accept sqe. Let's integrate it to fast poll, it may
be good for other operations in the future.

Signed-off-by: Hao Xu <howeyxu@tencent.com>
---
 fs/io_uring.c | 47 ++++++++++++++++++++++++++++++++---------------
 1 file changed, 32 insertions(+), 15 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index c2ee184ac693..e0d12af04cd1 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -5955,6 +5955,7 @@ static void io_poll_remove_entries(struct io_kiocb *req)
 	rcu_read_unlock();
 }
 
+static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags);
 /*
  * All poll tw should go through this. Checks for poll events, manages
  * references, does rewait, etc.
@@ -5963,10 +5964,10 @@ static void io_poll_remove_entries(struct io_kiocb *req)
  * either spurious wakeup or multishot CQE is served. 0 when it's done with
  * the request, then the mask is stored in req->cqe.res.
  */
-static int io_poll_check_events(struct io_kiocb *req, bool locked)
+static int io_poll_check_events(struct io_kiocb *req, bool *locked)
 {
 	struct io_ring_ctx *ctx = req->ctx;
-	int v;
+	int v, ret;
 
 	/* req->task == current here, checking PF_EXITING is safe */
 	if (unlikely(req->task->flags & PF_EXITING))
@@ -5990,23 +5991,37 @@ static int io_poll_check_events(struct io_kiocb *req, bool locked)
 			req->cqe.res = vfs_poll(req->file, &pt) & req->apoll_events;
 		}
 
-		/* multishot, just fill an CQE and proceed */
-		if (req->cqe.res && !(req->apoll_events & EPOLLONESHOT)) {
-			__poll_t mask = mangle_poll(req->cqe.res & req->apoll_events);
+		if ((unlikely(!req->cqe.res)))
+			continue;
+		if (req->apoll_events & EPOLLONESHOT)
+			return 0;
+
+		/* multishot, just fill a CQE and proceed */
+		if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
+			__poll_t mask = mangle_poll(req->cqe.res &
+						    req->apoll_events);
 			bool filled;
 
 			spin_lock(&ctx->completion_lock);
-			filled = io_fill_cqe_aux(ctx, req->cqe.user_data, mask,
-						 IORING_CQE_F_MORE);
+			filled = io_fill_cqe_aux(ctx, req->cqe.user_data,
+						 mask, IORING_CQE_F_MORE);
 			io_commit_cqring(ctx);
 			spin_unlock(&ctx->completion_lock);
-			if (unlikely(!filled))
-				return -ECANCELED;
-			io_cqring_ev_posted(ctx);
-		} else if (req->cqe.res) {
-			return 0;
+			if (filled) {
+				io_cqring_ev_posted(ctx);
+				continue;
+			}
+			return -ECANCELED;
 		}
 
+		io_tw_lock(req->ctx, locked);
+		if (unlikely(req->task->flags & PF_EXITING))
+			return -EFAULT;
+		ret = io_issue_sqe(req,
+				   IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
+		if (ret)
+			return ret;
+
 		/*
 		 * Release all references, retry if someone tried to restart
 		 * task_work while we were executing it.
@@ -6021,7 +6036,7 @@ static void io_poll_task_func(struct io_kiocb *req, bool *locked)
 	struct io_ring_ctx *ctx = req->ctx;
 	int ret;
 
-	ret = io_poll_check_events(req, *locked);
+	ret = io_poll_check_events(req, locked);
 	if (ret > 0)
 		return;
 
@@ -6046,7 +6061,7 @@ static void io_apoll_task_func(struct io_kiocb *req, bool *locked)
 	struct io_ring_ctx *ctx = req->ctx;
 	int ret;
 
-	ret = io_poll_check_events(req, *locked);
+	ret = io_poll_check_events(req, locked);
 	if (ret > 0)
 		return;
 
@@ -6286,7 +6301,7 @@ static int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
 	struct io_ring_ctx *ctx = req->ctx;
 	struct async_poll *apoll;
 	struct io_poll_table ipt;
-	__poll_t mask = EPOLLONESHOT | POLLERR | POLLPRI;
+	__poll_t mask = POLLERR | POLLPRI;
 	int ret;
 
 	if (!def->pollin && !def->pollout)
@@ -6295,6 +6310,8 @@ static int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
 		return IO_APOLL_ABORTED;
 	if ((req->flags & (REQ_F_POLLED|REQ_F_PARTIAL_IO)) == REQ_F_POLLED)
 		return IO_APOLL_ABORTED;
+	if (!(req->flags & REQ_F_APOLL_MULTISHOT))
+		mask |= EPOLLONESHOT;
 
 	if (def->pollin) {
 		mask |= POLLIN | POLLRDNORM;
-- 
2.25.1

Re: [PATCH 3/4] io_uring: let fast poll support multishot

[Jens Axboe]

On 5/8/22 9:32 AM, Hao Xu wrote:
> From: Hao Xu <howeyxu@tencent.com>
> 
> For operations like accept, multishot is a useful feature, since we can
> reduce a number of accept sqe. Let's integrate it to fast poll, it may
> be good for other operations in the future.

This, and 1-2 look good to me now.

-- 
Jens Axboe

[PATCH 3/4] io_uring: let fast poll support multishot

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

For operations like accept, multishot is a useful feature, since we can
reduce a number of accept sqe. Let's integrate it to fast poll, it may
be good for other operations in the future.

Signed-off-by: Hao Xu <howeyxu@tencent.com>
---
 fs/io_uring.c | 47 ++++++++++++++++++++++++++++++++---------------
 1 file changed, 32 insertions(+), 15 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index c2ee184ac693..e0d12af04cd1 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -5955,6 +5955,7 @@ static void io_poll_remove_entries(struct io_kiocb *req)
 	rcu_read_unlock();
 }
 
+static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags);
 /*
  * All poll tw should go through this. Checks for poll events, manages
  * references, does rewait, etc.
@@ -5963,10 +5964,10 @@ static void io_poll_remove_entries(struct io_kiocb *req)
  * either spurious wakeup or multishot CQE is served. 0 when it's done with
  * the request, then the mask is stored in req->cqe.res.
  */
-static int io_poll_check_events(struct io_kiocb *req, bool locked)
+static int io_poll_check_events(struct io_kiocb *req, bool *locked)
 {
 	struct io_ring_ctx *ctx = req->ctx;
-	int v;
+	int v, ret;
 
 	/* req->task == current here, checking PF_EXITING is safe */
 	if (unlikely(req->task->flags & PF_EXITING))
@@ -5990,23 +5991,37 @@ static int io_poll_check_events(struct io_kiocb *req, bool locked)
 			req->cqe.res = vfs_poll(req->file, &pt) & req->apoll_events;
 		}
 
-		/* multishot, just fill an CQE and proceed */
-		if (req->cqe.res && !(req->apoll_events & EPOLLONESHOT)) {
-			__poll_t mask = mangle_poll(req->cqe.res & req->apoll_events);
+		if ((unlikely(!req->cqe.res)))
+			continue;
+		if (req->apoll_events & EPOLLONESHOT)
+			return 0;
+
+		/* multishot, just fill a CQE and proceed */
+		if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
+			__poll_t mask = mangle_poll(req->cqe.res &
+						    req->apoll_events);
 			bool filled;
 
 			spin_lock(&ctx->completion_lock);
-			filled = io_fill_cqe_aux(ctx, req->cqe.user_data, mask,
-						 IORING_CQE_F_MORE);
+			filled = io_fill_cqe_aux(ctx, req->cqe.user_data,
+						 mask, IORING_CQE_F_MORE);
 			io_commit_cqring(ctx);
 			spin_unlock(&ctx->completion_lock);
-			if (unlikely(!filled))
-				return -ECANCELED;
-			io_cqring_ev_posted(ctx);
-		} else if (req->cqe.res) {
-			return 0;
+			if (filled) {
+				io_cqring_ev_posted(ctx);
+				continue;
+			}
+			return -ECANCELED;
 		}
 
+		io_tw_lock(req->ctx, locked);
+		if (unlikely(req->task->flags & PF_EXITING))
+			return -EFAULT;
+		ret = io_issue_sqe(req,
+				   IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
+		if (ret)
+			return ret;
+
 		/*
 		 * Release all references, retry if someone tried to restart
 		 * task_work while we were executing it.
@@ -6021,7 +6036,7 @@ static void io_poll_task_func(struct io_kiocb *req, bool *locked)
 	struct io_ring_ctx *ctx = req->ctx;
 	int ret;
 
-	ret = io_poll_check_events(req, *locked);
+	ret = io_poll_check_events(req, locked);
 	if (ret > 0)
 		return;
 
@@ -6046,7 +6061,7 @@ static void io_apoll_task_func(struct io_kiocb *req, bool *locked)
 	struct io_ring_ctx *ctx = req->ctx;
 	int ret;
 
-	ret = io_poll_check_events(req, *locked);
+	ret = io_poll_check_events(req, locked);
 	if (ret > 0)
 		return;
 
@@ -6286,7 +6301,7 @@ static int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
 	struct io_ring_ctx *ctx = req->ctx;
 	struct async_poll *apoll;
 	struct io_poll_table ipt;
-	__poll_t mask = EPOLLONESHOT | POLLERR | POLLPRI;
+	__poll_t mask = POLLERR | POLLPRI;
 	int ret;
 
 	if (!def->pollin && !def->pollout)
@@ -6295,6 +6310,8 @@ static int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
 		return IO_APOLL_ABORTED;
 	if ((req->flags & (REQ_F_POLLED|REQ_F_PARTIAL_IO)) == REQ_F_POLLED)
 		return IO_APOLL_ABORTED;
+	if (!(req->flags & REQ_F_APOLL_MULTISHOT))
+		mask |= EPOLLONESHOT;
 
 	if (def->pollin) {
 		mask |= POLLIN | POLLRDNORM;
-- 
2.25.1

[PATCH v3 0/4] fast poll multishot mode

[Hao Xu]

Let multishot support multishot mode, currently only add accept as its
first comsumer.
theoretical analysis:
  1) when connections come in fast
    - singleshot:
              add accept sqe(userpsace) --> accept inline
                              ^                 |
                              |-----------------|
    - multishot:
             add accept sqe(userspace) --> accept inline
                                              ^     |
                                              |--*--|

    we do accept repeatedly in * place until get EAGAIN

  2) when connections come in at a low pressure
    similar thing like 1), we reduce a lot of userspace-kernel context
    switch and useless vfs_poll()


tests:
Did some tests, which goes in this way:

  server    client(multiple)
  accept    connect
  read      write
  write     read
  close     close

Basically, raise up a number of clients(on same machine with server) to
connect to the server, and then write some data to it, the server will
write those data back to the client after it receives them, and then
close the connection after write return. Then the client will read the
data and then close the connection. Here I test 10000 clients connect
one server, data size 128 bytes. And each client has a go routine for
it, so they come to the server in short time.
test 20 times before/after this patchset, time spent:(unit cycle, which
is the return value of clock())
before:
  1930136+1940725+1907981+1947601+1923812+1928226+1911087+1905897+1941075
  +1934374+1906614+1912504+1949110+1908790+1909951+1941672+1969525+1934984
  +1934226+1914385)/20.0 = 1927633.75
after:
  1858905+1917104+1895455+1963963+1892706+1889208+1874175+1904753+1874112
  +1874985+1882706+1884642+1864694+1906508+1916150+1924250+1869060+1889506
  +1871324+1940803)/20.0 = 1894750.45

(1927633.75 - 1894750.45) / 1927633.75 = 1.65%


A liburing test is here:
https://github.com/HowHsu/liburing/blob/multishot_accept/test/accept.c

v1->v2:
 - re-implement it against the reworked poll code

v2->v3:
 - fold in code tweak and clean from Jens
 - use io_issue_sqe rather than io_queue_sqe, since the former one
   return the internal error back which makes more sense
 - remove io_poll_clean() and its friends since they are not needed


Hao Xu (4):
  io_uring: add IORING_ACCEPT_MULTISHOT for accept
  io_uring: add REQ_F_APOLL_MULTISHOT for requests
  io_uring: let fast poll support multishot
  io_uring: implement multishot mode for accept

 fs/io_uring.c                 | 94 +++++++++++++++++++++++++++--------
 include/uapi/linux/io_uring.h |  5 ++
 2 files changed, 79 insertions(+), 20 deletions(-)


base-commit: 0a194603ba7ee67b4e39ec0ee5cda70a356ea618
-- 
2.36.0

[PATCH 3/4] io_uring: let fast poll support multishot

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

For operations like accept, multishot is a useful feature, since we can
reduce a number of accept sqe. Let's integrate it to fast poll, it may
be good for other operations in the future.

Signed-off-by: Hao Xu <howeyxu@tencent.com>
[fold in cleaning up]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c | 47 ++++++++++++++++++++++++++++++++---------------
 1 file changed, 32 insertions(+), 15 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index c2ee184ac693..e0d12af04cd1 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -5955,6 +5955,7 @@ static void io_poll_remove_entries(struct io_kiocb *req)
 	rcu_read_unlock();
 }
 
+static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags);
 /*
  * All poll tw should go through this. Checks for poll events, manages
  * references, does rewait, etc.
@@ -5963,10 +5964,10 @@ static void io_poll_remove_entries(struct io_kiocb *req)
  * either spurious wakeup or multishot CQE is served. 0 when it's done with
  * the request, then the mask is stored in req->cqe.res.
  */
-static int io_poll_check_events(struct io_kiocb *req, bool locked)
+static int io_poll_check_events(struct io_kiocb *req, bool *locked)
 {
 	struct io_ring_ctx *ctx = req->ctx;
-	int v;
+	int v, ret;
 
 	/* req->task == current here, checking PF_EXITING is safe */
 	if (unlikely(req->task->flags & PF_EXITING))
@@ -5990,23 +5991,37 @@ static int io_poll_check_events(struct io_kiocb *req, bool locked)
 			req->cqe.res = vfs_poll(req->file, &pt) & req->apoll_events;
 		}
 
-		/* multishot, just fill an CQE and proceed */
-		if (req->cqe.res && !(req->apoll_events & EPOLLONESHOT)) {
-			__poll_t mask = mangle_poll(req->cqe.res & req->apoll_events);
+		if ((unlikely(!req->cqe.res)))
+			continue;
+		if (req->apoll_events & EPOLLONESHOT)
+			return 0;
+
+		/* multishot, just fill a CQE and proceed */
+		if (!(req->flags & REQ_F_APOLL_MULTISHOT)) {
+			__poll_t mask = mangle_poll(req->cqe.res &
+						    req->apoll_events);
 			bool filled;
 
 			spin_lock(&ctx->completion_lock);
-			filled = io_fill_cqe_aux(ctx, req->cqe.user_data, mask,
-						 IORING_CQE_F_MORE);
+			filled = io_fill_cqe_aux(ctx, req->cqe.user_data,
+						 mask, IORING_CQE_F_MORE);
 			io_commit_cqring(ctx);
 			spin_unlock(&ctx->completion_lock);
-			if (unlikely(!filled))
-				return -ECANCELED;
-			io_cqring_ev_posted(ctx);
-		} else if (req->cqe.res) {
-			return 0;
+			if (filled) {
+				io_cqring_ev_posted(ctx);
+				continue;
+			}
+			return -ECANCELED;
 		}
 
+		io_tw_lock(req->ctx, locked);
+		if (unlikely(req->task->flags & PF_EXITING))
+			return -EFAULT;
+		ret = io_issue_sqe(req,
+				   IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
+		if (ret)
+			return ret;
+
 		/*
 		 * Release all references, retry if someone tried to restart
 		 * task_work while we were executing it.
@@ -6021,7 +6036,7 @@ static void io_poll_task_func(struct io_kiocb *req, bool *locked)
 	struct io_ring_ctx *ctx = req->ctx;
 	int ret;
 
-	ret = io_poll_check_events(req, *locked);
+	ret = io_poll_check_events(req, locked);
 	if (ret > 0)
 		return;
 
@@ -6046,7 +6061,7 @@ static void io_apoll_task_func(struct io_kiocb *req, bool *locked)
 	struct io_ring_ctx *ctx = req->ctx;
 	int ret;
 
-	ret = io_poll_check_events(req, *locked);
+	ret = io_poll_check_events(req, locked);
 	if (ret > 0)
 		return;
 
@@ -6286,7 +6301,7 @@ static int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
 	struct io_ring_ctx *ctx = req->ctx;
 	struct async_poll *apoll;
 	struct io_poll_table ipt;
-	__poll_t mask = EPOLLONESHOT | POLLERR | POLLPRI;
+	__poll_t mask = POLLERR | POLLPRI;
 	int ret;
 
 	if (!def->pollin && !def->pollout)
@@ -6295,6 +6310,8 @@ static int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
 		return IO_APOLL_ABORTED;
 	if ((req->flags & (REQ_F_POLLED|REQ_F_PARTIAL_IO)) == REQ_F_POLLED)
 		return IO_APOLL_ABORTED;
+	if (!(req->flags & REQ_F_APOLL_MULTISHOT))
+		mask |= EPOLLONESHOT;
 
 	if (def->pollin) {
 		mask |= POLLIN | POLLRDNORM;
-- 
2.36.0