[RFC,v1,12/12] test/vsock: MSG_ZEROCOPY support for vsock_perf

Message ID 03570f48-f56a-2af4-9579-15a685127aeb@sberdevices.ru
State New
Headers
Series vsock: MSG_ZEROCOPY flag support |

Commit Message

Arseniy Krasnov Feb. 6, 2023, 7:06 a.m. UTC
  To use this option pass '--zc' parameter:

./vsock_perf --zc --sender <cid> --port <port> --bytes <bytes to send>

With this option MSG_ZEROCOPY flag will be passed to the 'send()' call.

Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
---
 tools/testing/vsock/vsock_perf.c | 127 +++++++++++++++++++++++++++++--
 1 file changed, 120 insertions(+), 7 deletions(-)

-- 
2.25.1
  

Comments

Stefano Garzarella Feb. 16, 2023, 3:29 p.m. UTC | #1
On Mon, Feb 06, 2023 at 07:06:32AM +0000, Arseniy Krasnov wrote:
>To use this option pass '--zc' parameter:

--zerocopy or --zero-copy maybe better follow what we did with the other 
parameters :-)

>
>./vsock_perf --zc --sender <cid> --port <port> --bytes <bytes to send>
>
>With this option MSG_ZEROCOPY flag will be passed to the 'send()' call.
>
>Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
>---
> tools/testing/vsock/vsock_perf.c | 127 +++++++++++++++++++++++++++++--
> 1 file changed, 120 insertions(+), 7 deletions(-)
>
>diff --git a/tools/testing/vsock/vsock_perf.c b/tools/testing/vsock/vsock_perf.c
>index a72520338f84..1d435be9b48e 100644
>--- a/tools/testing/vsock/vsock_perf.c
>+++ b/tools/testing/vsock/vsock_perf.c
>@@ -18,6 +18,8 @@
> #include <poll.h>
> #include <sys/socket.h>
> #include <linux/vm_sockets.h>
>+#include <sys/mman.h>
>+#include <linux/errqueue.h>
>
> #define DEFAULT_BUF_SIZE_BYTES	(128 * 1024)
> #define DEFAULT_TO_SEND_BYTES	(64 * 1024)
>@@ -28,9 +30,14 @@
> #define BYTES_PER_GB		(1024 * 1024 * 1024ULL)
> #define NSEC_PER_SEC		(1000000000ULL)
>
>+#ifndef SOL_VSOCK
>+#define SOL_VSOCK 287
>+#endif

I thought we use the current kernel headers when we compile the tests,
do we need to fix something in the makefile?

>+
> static unsigned int port = DEFAULT_PORT;
> static unsigned long buf_size_bytes = DEFAULT_BUF_SIZE_BYTES;
> static unsigned long vsock_buf_bytes = DEFAULT_VSOCK_BUF_BYTES;
>+static bool zerocopy;
>
> static void error(const char *s)
> {
>@@ -247,15 +254,74 @@ static void run_receiver(unsigned long rcvlowat_bytes)
> 	close(fd);
> }
>
>+static void recv_completion(int fd)
>+{
>+	struct sock_extended_err *serr;
>+	char cmsg_data[128];
>+	struct cmsghdr *cm;
>+	struct msghdr msg;
>+	int ret;
>+
>+	msg.msg_control = cmsg_data;
>+	msg.msg_controllen = sizeof(cmsg_data);
>+
>+	ret = recvmsg(fd, &msg, MSG_ERRQUEUE);
>+	if (ret == -1)
>+		return;
>+
>+	cm = CMSG_FIRSTHDR(&msg);
>+	if (!cm) {
>+		fprintf(stderr, "cmsg: no cmsg\n");
>+		return;
>+	}
>+
>+	if (cm->cmsg_level != SOL_VSOCK) {
>+		fprintf(stderr, "cmsg: unexpected 'cmsg_level'\n");
>+		return;
>+	}
>+
>+	if (cm->cmsg_type) {
>+		fprintf(stderr, "cmsg: unexpected 'cmsg_type'\n");
>+		return;
>+	}
>+
>+	serr = (void *)CMSG_DATA(cm);
>+	if (serr->ee_origin != SO_EE_ORIGIN_ZEROCOPY) {
>+		fprintf(stderr, "serr: wrong origin\n");
>+		return;
>+	}
>+
>+	if (serr->ee_errno) {
>+		fprintf(stderr, "serr: wrong error code\n");
>+		return;
>+	}
>+
>+	if (zerocopy && (serr->ee_code & SO_EE_CODE_ZEROCOPY_COPIED))
>+		fprintf(stderr, "warning: copy instead of zerocopy\n");
>+}
>+
>+static void enable_so_zerocopy(int fd)
>+{
>+	int val = 1;
>+
>+	if (setsockopt(fd, SOL_SOCKET, SO_ZEROCOPY, &val, sizeof(val)))
>+		error("setsockopt(SO_ZEROCOPY)");
>+}
>+
> static void run_sender(int peer_cid, unsigned long to_send_bytes)
> {
> 	time_t tx_begin_ns;
> 	time_t tx_total_ns;
> 	size_t total_send;
>+	time_t time_in_send;
> 	void *data;
> 	int fd;
>
>-	printf("Run as sender\n");
>+	if (zerocopy)
>+		printf("Run as sender MSG_ZEROCOPY\n");
>+	else
>+		printf("Run as sender\n");
>+
> 	printf("Connect to %i:%u\n", peer_cid, port);
> 	printf("Send %lu bytes\n", to_send_bytes);
> 	printf("TX buffer %lu bytes\n", buf_size_bytes);
>@@ -265,25 +331,58 @@ static void run_sender(int peer_cid, unsigned long to_send_bytes)
> 	if (fd < 0)
> 		exit(EXIT_FAILURE);
>
>-	data = malloc(buf_size_bytes);
>+	if (zerocopy) {
>+		enable_so_zerocopy(fd);
>
>-	if (!data) {
>-		fprintf(stderr, "'malloc()' failed\n");
>-		exit(EXIT_FAILURE);
>+		data = mmap(NULL, buf_size_bytes, PROT_READ | PROT_WRITE,
>+			    MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
>+		if (data == MAP_FAILED) {
>+			perror("mmap");
>+			exit(EXIT_FAILURE);
>+		}
>+	} else {
>+		data = malloc(buf_size_bytes);
>+
>+		if (!data) {
>+			fprintf(stderr, "'malloc()' failed\n");
>+			exit(EXIT_FAILURE);
>+		}
> 	}

Eventually to simplify the code I think we can use the mmaped buffer in
both cases.

>
> 	memset(data, 0, buf_size_bytes);
> 	total_send = 0;
>+	time_in_send = 0;
> 	tx_begin_ns = current_nsec();
>
> 	while (total_send < to_send_bytes) {
> 		ssize_t sent;
>+		size_t rest_bytes;
>+		time_t before;
>+
>+		rest_bytes = to_send_bytes - total_send;
>
>-		sent = write(fd, data, buf_size_bytes);
>+		before = current_nsec();
>+		sent = send(fd, data, (rest_bytes > buf_size_bytes) ?
>+			    buf_size_bytes : rest_bytes,
>+			    zerocopy ? MSG_ZEROCOPY : 0);
>+		time_in_send += (current_nsec() - before);
>
> 		if (sent <= 0)
> 			error("write");
>
>+		if (zerocopy) {
>+			struct pollfd fds = { 0 };
>+
>+			fds.fd = fd;

Which event are we waiting for here?

>+
>+			if (poll(&fds, 1, -1) < 0) {
>+				perror("poll");
>+				exit(EXIT_FAILURE);
>+			}

We need this because we use only one buffer, but if we use more than
one, we could take full advantage of zerocopy, right?

Otherwise, I don't think it's a fair comparison with non-zerocopy.

Thanks,
Stefano

>+
>+			recv_completion(fd);
>+		}
>+
> 		total_send += sent;
> 	}
>
>@@ -294,9 +393,14 @@ static void run_sender(int peer_cid, unsigned long to_send_bytes)
> 	       get_gbps(total_send * 8, tx_total_ns));
> 	printf("total time in 'write()': %f sec\n",
> 	       (float)tx_total_ns / NSEC_PER_SEC);
>+	printf("time in send %f\n", (float)time_in_send / NSEC_PER_SEC);
>
> 	close(fd);
>-	free(data);
>+
>+	if (zerocopy)
>+		munmap(data, buf_size_bytes);
>+	else
>+		free(data);
> }
>
> static const char optstring[] = "";
>@@ -336,6 +440,11 @@ static const struct option longopts[] = {
> 		.has_arg = required_argument,
> 		.val = 'R',
> 	},
>+	{
>+		.name = "zc",
>+		.has_arg = no_argument,
>+		.val = 'Z',
>+	},
> 	{},
> };
>
>@@ -351,6 +460,7 @@ static void usage(void)
> 	       "  --help			This message\n"
> 	       "  --sender   <cid>		Sender mode (receiver default)\n"
> 	       "                                <cid> of the receiver to connect to\n"
>+	       "  --zc				Enable zerocopy\n"
> 	       "  --port     <port>		Port (default %d)\n"
> 	       "  --bytes    <bytes>KMG		Bytes to send (default %d)\n"
> 	       "  --buf-size <bytes>KMG		Data buffer size (default %d). In sender mode\n"
>@@ -413,6 +523,9 @@ int main(int argc, char **argv)
> 		case 'H': /* Help. */
> 			usage();
> 			break;
>+		case 'Z': /* Zerocopy. */
>+			zerocopy = true;
>+			break;
> 		default:
> 			usage();
> 		}
>-- 
>2.25.1
>
  
Arseniy Krasnov Feb. 20, 2023, 9:05 a.m. UTC | #2
On 16.02.2023 18:29, Stefano Garzarella wrote:
> On Mon, Feb 06, 2023 at 07:06:32AM +0000, Arseniy Krasnov wrote:
>> To use this option pass '--zc' parameter:
> 
> --zerocopy or --zero-copy maybe better follow what we did with the other parameters :-)
> 
>>
>> ./vsock_perf --zc --sender <cid> --port <port> --bytes <bytes to send>
>>
>> With this option MSG_ZEROCOPY flag will be passed to the 'send()' call.
>>
>> Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
>> ---
>> tools/testing/vsock/vsock_perf.c | 127 +++++++++++++++++++++++++++++--
>> 1 file changed, 120 insertions(+), 7 deletions(-)
>>
>> diff --git a/tools/testing/vsock/vsock_perf.c b/tools/testing/vsock/vsock_perf.c
>> index a72520338f84..1d435be9b48e 100644
>> --- a/tools/testing/vsock/vsock_perf.c
>> +++ b/tools/testing/vsock/vsock_perf.c
>> @@ -18,6 +18,8 @@
>> #include <poll.h>
>> #include <sys/socket.h>
>> #include <linux/vm_sockets.h>
>> +#include <sys/mman.h>
>> +#include <linux/errqueue.h>
>>
>> #define DEFAULT_BUF_SIZE_BYTES    (128 * 1024)
>> #define DEFAULT_TO_SEND_BYTES    (64 * 1024)
>> @@ -28,9 +30,14 @@
>> #define BYTES_PER_GB        (1024 * 1024 * 1024ULL)
>> #define NSEC_PER_SEC        (1000000000ULL)
>>
>> +#ifndef SOL_VSOCK
>> +#define SOL_VSOCK 287
>> +#endif
> 
> I thought we use the current kernel headers when we compile the tests,
> do we need to fix something in the makefile?
Not sure, of course we are using uapi. But i see, that defines like SOL_XXX is not
defined in uapi headers. For example SOL_IP is defined in include/linux/socket.h,
but userspace app uses SOL_IP from in.h (at least on my machine). E.g. SOL_XXX is
not exported to user.
> 
>> +
>> static unsigned int port = DEFAULT_PORT;
>> static unsigned long buf_size_bytes = DEFAULT_BUF_SIZE_BYTES;
>> static unsigned long vsock_buf_bytes = DEFAULT_VSOCK_BUF_BYTES;
>> +static bool zerocopy;
>>
>> static void error(const char *s)
>> {
>> @@ -247,15 +254,74 @@ static void run_receiver(unsigned long rcvlowat_bytes)
>>     close(fd);
>> }
>>
>> +static void recv_completion(int fd)
>> +{
>> +    struct sock_extended_err *serr;
>> +    char cmsg_data[128];
>> +    struct cmsghdr *cm;
>> +    struct msghdr msg;
>> +    int ret;
>> +
>> +    msg.msg_control = cmsg_data;
>> +    msg.msg_controllen = sizeof(cmsg_data);
>> +
>> +    ret = recvmsg(fd, &msg, MSG_ERRQUEUE);
>> +    if (ret == -1)
>> +        return;
>> +
>> +    cm = CMSG_FIRSTHDR(&msg);
>> +    if (!cm) {
>> +        fprintf(stderr, "cmsg: no cmsg\n");
>> +        return;
>> +    }
>> +
>> +    if (cm->cmsg_level != SOL_VSOCK) {
>> +        fprintf(stderr, "cmsg: unexpected 'cmsg_level'\n");
>> +        return;
>> +    }
>> +
>> +    if (cm->cmsg_type) {
>> +        fprintf(stderr, "cmsg: unexpected 'cmsg_type'\n");
>> +        return;
>> +    }
>> +
>> +    serr = (void *)CMSG_DATA(cm);
>> +    if (serr->ee_origin != SO_EE_ORIGIN_ZEROCOPY) {
>> +        fprintf(stderr, "serr: wrong origin\n");
>> +        return;
>> +    }
>> +
>> +    if (serr->ee_errno) {
>> +        fprintf(stderr, "serr: wrong error code\n");
>> +        return;
>> +    }
>> +
>> +    if (zerocopy && (serr->ee_code & SO_EE_CODE_ZEROCOPY_COPIED))
>> +        fprintf(stderr, "warning: copy instead of zerocopy\n");
>> +}
>> +
>> +static void enable_so_zerocopy(int fd)
>> +{
>> +    int val = 1;
>> +
>> +    if (setsockopt(fd, SOL_SOCKET, SO_ZEROCOPY, &val, sizeof(val)))
>> +        error("setsockopt(SO_ZEROCOPY)");
>> +}
>> +
>> static void run_sender(int peer_cid, unsigned long to_send_bytes)
>> {
>>     time_t tx_begin_ns;
>>     time_t tx_total_ns;
>>     size_t total_send;
>> +    time_t time_in_send;
>>     void *data;
>>     int fd;
>>
>> -    printf("Run as sender\n");
>> +    if (zerocopy)
>> +        printf("Run as sender MSG_ZEROCOPY\n");
>> +    else
>> +        printf("Run as sender\n");
>> +
>>     printf("Connect to %i:%u\n", peer_cid, port);
>>     printf("Send %lu bytes\n", to_send_bytes);
>>     printf("TX buffer %lu bytes\n", buf_size_bytes);
>> @@ -265,25 +331,58 @@ static void run_sender(int peer_cid, unsigned long to_send_bytes)
>>     if (fd < 0)
>>         exit(EXIT_FAILURE);
>>
>> -    data = malloc(buf_size_bytes);
>> +    if (zerocopy) {
>> +        enable_so_zerocopy(fd);
>>
>> -    if (!data) {
>> -        fprintf(stderr, "'malloc()' failed\n");
>> -        exit(EXIT_FAILURE);
>> +        data = mmap(NULL, buf_size_bytes, PROT_READ | PROT_WRITE,
>> +                MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
>> +        if (data == MAP_FAILED) {
>> +            perror("mmap");
>> +            exit(EXIT_FAILURE);
>> +        }
>> +    } else {
>> +        data = malloc(buf_size_bytes);
>> +
>> +        if (!data) {
>> +            fprintf(stderr, "'malloc()' failed\n");
>> +            exit(EXIT_FAILURE);
>> +        }
>>     }
> 
> Eventually to simplify the code I think we can use the mmaped buffer in
> both cases.
> 
>>
>>     memset(data, 0, buf_size_bytes);
>>     total_send = 0;
>> +    time_in_send = 0;
>>     tx_begin_ns = current_nsec();
>>
>>     while (total_send < to_send_bytes) {
>>         ssize_t sent;
>> +        size_t rest_bytes;
>> +        time_t before;
>> +
>> +        rest_bytes = to_send_bytes - total_send;
>>
>> -        sent = write(fd, data, buf_size_bytes);
>> +        before = current_nsec();
>> +        sent = send(fd, data, (rest_bytes > buf_size_bytes) ?
>> +                buf_size_bytes : rest_bytes,
>> +                zerocopy ? MSG_ZEROCOPY : 0);
>> +        time_in_send += (current_nsec() - before);
>>
>>         if (sent <= 0)
>>             error("write");
>>
>> +        if (zerocopy) {
>> +            struct pollfd fds = { 0 };
>> +
>> +            fds.fd = fd;
> 
> Which event are we waiting for here?
> 
>> +
>> +            if (poll(&fds, 1, -1) < 0) {
>> +                perror("poll");
>> +                exit(EXIT_FAILURE);
>> +            }
> 
> We need this because we use only one buffer, but if we use more than
> one, we could take full advantage of zerocopy, right?
> 
> Otherwise, I don't think it's a fair comparison with non-zerocopy.
Yes or course, in next versions i'll update this test for using
more long iovs.
> 
> Thanks,
> Stefano
> 
>> +
>> +            recv_completion(fd);
>> +        }
>> +
>>         total_send += sent;
>>     }
>>
>> @@ -294,9 +393,14 @@ static void run_sender(int peer_cid, unsigned long to_send_bytes)
>>            get_gbps(total_send * 8, tx_total_ns));
>>     printf("total time in 'write()': %f sec\n",
>>            (float)tx_total_ns / NSEC_PER_SEC);
>> +    printf("time in send %f\n", (float)time_in_send / NSEC_PER_SEC);
>>
>>     close(fd);
>> -    free(data);
>> +
>> +    if (zerocopy)
>> +        munmap(data, buf_size_bytes);
>> +    else
>> +        free(data);
>> }
>>
>> static const char optstring[] = "";
>> @@ -336,6 +440,11 @@ static const struct option longopts[] = {
>>         .has_arg = required_argument,
>>         .val = 'R',
>>     },
>> +    {
>> +        .name = "zc",
>> +        .has_arg = no_argument,
>> +        .val = 'Z',
>> +    },
>>     {},
>> };
>>
>> @@ -351,6 +460,7 @@ static void usage(void)
>>            "  --help            This message\n"
>>            "  --sender   <cid>        Sender mode (receiver default)\n"
>>            "                                <cid> of the receiver to connect to\n"
>> +           "  --zc                Enable zerocopy\n"
>>            "  --port     <port>        Port (default %d)\n"
>>            "  --bytes    <bytes>KMG        Bytes to send (default %d)\n"
>>            "  --buf-size <bytes>KMG        Data buffer size (default %d). In sender mode\n"
>> @@ -413,6 +523,9 @@ int main(int argc, char **argv)
>>         case 'H': /* Help. */
>>             usage();
>>             break;
>> +        case 'Z': /* Zerocopy. */
>> +            zerocopy = true;
>> +            break;
>>         default:
>>             usage();
>>         }
>> -- 
>> 2.25.1
>>
>
  
Stefano Garzarella Feb. 28, 2023, 10:32 a.m. UTC | #3
On Mon, Feb 20, 2023 at 09:05:12AM +0000, Krasnov Arseniy wrote:
>On 16.02.2023 18:29, Stefano Garzarella wrote:
>> On Mon, Feb 06, 2023 at 07:06:32AM +0000, Arseniy Krasnov wrote:
>>> To use this option pass '--zc' parameter:
>>
>> --zerocopy or --zero-copy maybe better follow what we did with the other parameters :-)
>>
>>>
>>> ./vsock_perf --zc --sender <cid> --port <port> --bytes <bytes to send>
>>>
>>> With this option MSG_ZEROCOPY flag will be passed to the 'send()' call.
>>>
>>> Signed-off-by: Arseniy Krasnov <AVKrasnov@sberdevices.ru>
>>> ---
>>> tools/testing/vsock/vsock_perf.c | 127 +++++++++++++++++++++++++++++--
>>> 1 file changed, 120 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/tools/testing/vsock/vsock_perf.c b/tools/testing/vsock/vsock_perf.c
>>> index a72520338f84..1d435be9b48e 100644
>>> --- a/tools/testing/vsock/vsock_perf.c
>>> +++ b/tools/testing/vsock/vsock_perf.c
>>> @@ -18,6 +18,8 @@
>>> #include <poll.h>
>>> #include <sys/socket.h>
>>> #include <linux/vm_sockets.h>
>>> +#include <sys/mman.h>
>>> +#include <linux/errqueue.h>
>>>
>>> #define DEFAULT_BUF_SIZE_BYTES    (128 * 1024)
>>> #define DEFAULT_TO_SEND_BYTES    (64 * 1024)
>>> @@ -28,9 +30,14 @@
>>> #define BYTES_PER_GB        (1024 * 1024 * 1024ULL)
>>> #define NSEC_PER_SEC        (1000000000ULL)
>>>
>>> +#ifndef SOL_VSOCK
>>> +#define SOL_VSOCK 287
>>> +#endif
>>
>> I thought we use the current kernel headers when we compile the tests,
>> do we need to fix something in the makefile?
>Not sure, of course we are using uapi. But i see, that defines like SOL_XXX is not
>defined in uapi headers. For example SOL_IP is defined in include/linux/socket.h,
>but userspace app uses SOL_IP from in.h (at least on my machine). E.g. SOL_XXX is
>not exported to user.

Right, I see only few SOL_* in the uapi, e.g. SOL_TIPC in 
uapi/linux/tipc.h

So it's fine for now, otherwise we can define it in 
uapi/linux/vm_sockets.h

Thanks,
Stefano
  

Patch

diff --git a/tools/testing/vsock/vsock_perf.c b/tools/testing/vsock/vsock_perf.c
index a72520338f84..1d435be9b48e 100644
--- a/tools/testing/vsock/vsock_perf.c
+++ b/tools/testing/vsock/vsock_perf.c
@@ -18,6 +18,8 @@ 
 #include <poll.h>
 #include <sys/socket.h>
 #include <linux/vm_sockets.h>
+#include <sys/mman.h>
+#include <linux/errqueue.h>
 
 #define DEFAULT_BUF_SIZE_BYTES	(128 * 1024)
 #define DEFAULT_TO_SEND_BYTES	(64 * 1024)
@@ -28,9 +30,14 @@ 
 #define BYTES_PER_GB		(1024 * 1024 * 1024ULL)
 #define NSEC_PER_SEC		(1000000000ULL)
 
+#ifndef SOL_VSOCK
+#define SOL_VSOCK 287
+#endif
+
 static unsigned int port = DEFAULT_PORT;
 static unsigned long buf_size_bytes = DEFAULT_BUF_SIZE_BYTES;
 static unsigned long vsock_buf_bytes = DEFAULT_VSOCK_BUF_BYTES;
+static bool zerocopy;
 
 static void error(const char *s)
 {
@@ -247,15 +254,74 @@  static void run_receiver(unsigned long rcvlowat_bytes)
 	close(fd);
 }
 
+static void recv_completion(int fd)
+{
+	struct sock_extended_err *serr;
+	char cmsg_data[128];
+	struct cmsghdr *cm;
+	struct msghdr msg;
+	int ret;
+
+	msg.msg_control = cmsg_data;
+	msg.msg_controllen = sizeof(cmsg_data);
+
+	ret = recvmsg(fd, &msg, MSG_ERRQUEUE);
+	if (ret == -1)
+		return;
+
+	cm = CMSG_FIRSTHDR(&msg);
+	if (!cm) {
+		fprintf(stderr, "cmsg: no cmsg\n");
+		return;
+	}
+
+	if (cm->cmsg_level != SOL_VSOCK) {
+		fprintf(stderr, "cmsg: unexpected 'cmsg_level'\n");
+		return;
+	}
+
+	if (cm->cmsg_type) {
+		fprintf(stderr, "cmsg: unexpected 'cmsg_type'\n");
+		return;
+	}
+
+	serr = (void *)CMSG_DATA(cm);
+	if (serr->ee_origin != SO_EE_ORIGIN_ZEROCOPY) {
+		fprintf(stderr, "serr: wrong origin\n");
+		return;
+	}
+
+	if (serr->ee_errno) {
+		fprintf(stderr, "serr: wrong error code\n");
+		return;
+	}
+
+	if (zerocopy && (serr->ee_code & SO_EE_CODE_ZEROCOPY_COPIED))
+		fprintf(stderr, "warning: copy instead of zerocopy\n");
+}
+
+static void enable_so_zerocopy(int fd)
+{
+	int val = 1;
+
+	if (setsockopt(fd, SOL_SOCKET, SO_ZEROCOPY, &val, sizeof(val)))
+		error("setsockopt(SO_ZEROCOPY)");
+}
+
 static void run_sender(int peer_cid, unsigned long to_send_bytes)
 {
 	time_t tx_begin_ns;
 	time_t tx_total_ns;
 	size_t total_send;
+	time_t time_in_send;
 	void *data;
 	int fd;
 
-	printf("Run as sender\n");
+	if (zerocopy)
+		printf("Run as sender MSG_ZEROCOPY\n");
+	else
+		printf("Run as sender\n");
+
 	printf("Connect to %i:%u\n", peer_cid, port);
 	printf("Send %lu bytes\n", to_send_bytes);
 	printf("TX buffer %lu bytes\n", buf_size_bytes);
@@ -265,25 +331,58 @@  static void run_sender(int peer_cid, unsigned long to_send_bytes)
 	if (fd < 0)
 		exit(EXIT_FAILURE);
 
-	data = malloc(buf_size_bytes);
+	if (zerocopy) {
+		enable_so_zerocopy(fd);
 
-	if (!data) {
-		fprintf(stderr, "'malloc()' failed\n");
-		exit(EXIT_FAILURE);
+		data = mmap(NULL, buf_size_bytes, PROT_READ | PROT_WRITE,
+			    MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+		if (data == MAP_FAILED) {
+			perror("mmap");
+			exit(EXIT_FAILURE);
+		}
+	} else {
+		data = malloc(buf_size_bytes);
+
+		if (!data) {
+			fprintf(stderr, "'malloc()' failed\n");
+			exit(EXIT_FAILURE);
+		}
 	}
 
 	memset(data, 0, buf_size_bytes);
 	total_send = 0;
+	time_in_send = 0;
 	tx_begin_ns = current_nsec();
 
 	while (total_send < to_send_bytes) {
 		ssize_t sent;
+		size_t rest_bytes;
+		time_t before;
+
+		rest_bytes = to_send_bytes - total_send;
 
-		sent = write(fd, data, buf_size_bytes);
+		before = current_nsec();
+		sent = send(fd, data, (rest_bytes > buf_size_bytes) ?
+			    buf_size_bytes : rest_bytes,
+			    zerocopy ? MSG_ZEROCOPY : 0);
+		time_in_send += (current_nsec() - before);
 
 		if (sent <= 0)
 			error("write");
 
+		if (zerocopy) {
+			struct pollfd fds = { 0 };
+
+			fds.fd = fd;
+
+			if (poll(&fds, 1, -1) < 0) {
+				perror("poll");
+				exit(EXIT_FAILURE);
+			}
+
+			recv_completion(fd);
+		}
+
 		total_send += sent;
 	}
 
@@ -294,9 +393,14 @@  static void run_sender(int peer_cid, unsigned long to_send_bytes)
 	       get_gbps(total_send * 8, tx_total_ns));
 	printf("total time in 'write()': %f sec\n",
 	       (float)tx_total_ns / NSEC_PER_SEC);
+	printf("time in send %f\n", (float)time_in_send / NSEC_PER_SEC);
 
 	close(fd);
-	free(data);
+
+	if (zerocopy)
+		munmap(data, buf_size_bytes);
+	else
+		free(data);
 }
 
 static const char optstring[] = "";
@@ -336,6 +440,11 @@  static const struct option longopts[] = {
 		.has_arg = required_argument,
 		.val = 'R',
 	},
+	{
+		.name = "zc",
+		.has_arg = no_argument,
+		.val = 'Z',
+	},
 	{},
 };
 
@@ -351,6 +460,7 @@  static void usage(void)
 	       "  --help			This message\n"
 	       "  --sender   <cid>		Sender mode (receiver default)\n"
 	       "                                <cid> of the receiver to connect to\n"
+	       "  --zc				Enable zerocopy\n"
 	       "  --port     <port>		Port (default %d)\n"
 	       "  --bytes    <bytes>KMG		Bytes to send (default %d)\n"
 	       "  --buf-size <bytes>KMG		Data buffer size (default %d). In sender mode\n"
@@ -413,6 +523,9 @@  int main(int argc, char **argv)
 		case 'H': /* Help. */
 			usage();
 			break;
+		case 'Z': /* Zerocopy. */
+			zerocopy = true;
+			break;
 		default:
 			usage();
 		}