The previous patch for supporting the TCP_CORK socket option broke the TCP_NODELAY option, which previously worked accidentally - because it is the only TCP socket option that had the same numeric value in Linux as in BSD. This patch returns support for it, and tests this support (the test is the same test for TCP_CORK, it's just that with TCP_NODELAY we don't need to "uncork" - each small write sends its data immediately so right after the second write, the reader can almost immediately read all the data).
Signed-off-by: Nadav Har'El <[email protected]> --- bsd/sys/compat/linux/linux_socket.cc | 2 ++ tests/tst-tcp-cork.cc | 38 ++++++++++++++++++++++-------------- 2 files changed, 25 insertions(+), 15 deletions(-) diff --git a/bsd/sys/compat/linux/linux_socket.cc b/bsd/sys/compat/linux/linux_socket.cc index 141da64..c1bbc49 100644 --- a/bsd/sys/compat/linux/linux_socket.cc +++ b/bsd/sys/compat/linux/linux_socket.cc @@ -1200,6 +1200,8 @@ int linux_to_bsd_tcp_sockopt(int name) // Not using the constants because we never know what will the compiler // will insert here. They are interface, so they shouldn't change. switch (name) { + case 1: // TCP_NODELAY + return 0x001; case 3: // TCP_CORK return 0x004; // TCP_NOPUSH case 4: // TCP_KEEPIDLE diff --git a/tests/tst-tcp-cork.cc b/tests/tst-tcp-cork.cc index 614af50..5f0dc5a 100644 --- a/tests/tst-tcp-cork.cc +++ b/tests/tst-tcp-cork.cc @@ -6,16 +6,17 @@ */ /* - * This is a test for the TCP_CORK socket option described in the tcp(7) - * manual page. The server thread opens a socket, accepts connection and then - * writes to in two separate writes of 100 bytes. If the first write is - * not acknowleged for 100ms (because of "delayed ack"), the second write - * is not sent (because of Naggle's algorithm). If the client measures the - * amount of time it takes to get from 1 to 200 bytes, he will see more than - * 100ms. + * This is a test for the TCP_CORK and TCP_NODELAY socket options described + * in the tcp(7) manual page. The server thread opens a socket, accepts + * connection and then writes to in two separate writes of 100 bytes. If the + * first write is not acknowleged for 100ms (because of "delayed ack"), the + * second write is not sent (because of Naggle's algorithm). If the client + * measures the amount of time it takes to get from 1 to 200 bytes, he will + * see more than 100ms. * But if TCP_CORK is used - and correctly supported - the two writes will * be flushed immediately as TCP_CORK is released, and this will take much - * less than 100ms. + * less than 100ms. Similarly, when TCP_NODELAY is used, each write will be + * set out immediately, and not be delayed by Naggle's algorithm. */ #include <stdio.h> @@ -82,7 +83,7 @@ static int server(void) after-before).count(); } -static int client(bool cork) +static int client(bool cork, bool nodelay) { int sock; struct sockaddr_in sa; @@ -105,6 +106,12 @@ static int client(bool cork) perror("setsockopt failed"); } } + if (nodelay) { + int optval = 1; + if (setsockopt(sock, IPPROTO_TCP, TCP_NODELAY, &optval, sizeof(int)) < 0) { + perror("setsockopt failed"); + } + } // Make two small writes, of 100 bytes each. Sleep for 5ms between them, // hoping to give the kernel enough time to decide to send the first @@ -134,15 +141,15 @@ static int client(bool cork) return 0; } -static bool test(bool cork) +static bool test(bool cork, bool nodelay) { int srv_result; std::thread t([&srv_result] { srv_result = server(); }); sleep(1); - int clnt_result = client(cork); + int clnt_result = client(cork, nodelay); t.join(); - printf("test %d time (ms): %d \n", cork, srv_result); - if (cork) { + printf("test %d %d time (ms): %d \n", cork, nodelay, srv_result); + if (cork || nodelay) { // If TCP_CORK is used, both batches of 100 bytes are sent as soon as // we release the TCP_CORK option, which takes about 5ms because of the // 5ms delay we have between the two batches. It can be a bit more than @@ -177,8 +184,9 @@ static void report(bool ok, const char* msg) int main(int argc, char **argv) { - report(test(false), "Without TCP_CORK, delayed ack comes into play"); - report(test(true), "With TCP_CORK, no nagle so no delayed ack"); + report(test(false, false), "Without TCP_CORK, delayed ack comes into play"); + report(test(true, false), "With TCP_CORK, no nagle so no delayed ack"); + report(test(false, true), "With TCP_NODELAY, no nagle so no delayed ack"); printf("SUMMARY: %d tests, %d failures\n", tests, fails); return fails; } -- 2.9.4 -- You received this message because you are subscribed to the Google Groups "OSv Development" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
