[Tarantool-patches] [PATCH] evio: workaround for wsl1 so_linger assertion

[Tarantool-patches] [PATCH] evio: workaround for wsl1 so_linger assertion

[Timur Safin]

SO_LINGER makes no much sense for unix-sockets, and Microsoft WSL
is returning EINVAL if setsockopts called for SO_LINGER over unix
sockets:

  [004] 2020-03-11 18:42:29.592 [29182] main/102/app sio.c:169 !> SystemError setsockopt(SO_LINGER), called on fd 16, aka
  [004] 2020-03-11 18:42:29.592 [29182] main/102/app F> can't initialize storage: setsockopt(SO_LINGER), called on fd 16,
  [004] 2020-03-11 18:42:29.592 [29182] main/102/app F> can't initialize storage: setsockopt(SO_LINGER), called on fd 16,

And it's sort of correct here, but the problem is Linux is simply
silently ignoring it, which passes tests.

After much debates we decided to work-around this case via CMAKE
define.

NB! In a future (April/May 2020), when WSL2 with full Linux kernel
would be released we should disable this check.
---

This patch replaces prior one tsafin/gh-4659-wsl-no-linger-assert because original
approach considered unsafe, and we rather want to workaround it via CMake instead.


Branch: https://github.com/tarantool/tarantool/tree/tsafin/wsl1-no-linger-assert

 cmake/os.cmake      | 13 +++++++++++++
 src/lib/core/evio.c |  2 ++
 2 files changed, 15 insertions(+)

diff --git a/cmake/os.cmake b/cmake/os.cmake
index 0ed554b9b..ee8870d21 100644
--- a/cmake/os.cmake
+++ b/cmake/os.cmake
@@ -11,6 +11,15 @@ if (${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
 # (see man page for feature_test_macros).
     add_definitions("-D_FILE_OFFSET_BITS=64")
     find_package_message(PLATFORM "Building for Linux" "${CMAKE_SYSTEM_NAME}")
+
+    # There are some subtle differences in Linux kernel calls
+    # implementation under WSL1 (which should go away with WSL2 kernel)
+    # so for a moment we introduce a way to distinguish Linux and
+    # Microsoft/WSL1
+    if (${CMAKE_SYSTEM} MATCHES "Linux-.*-Microsoft")
+        add_definitions("-DTARANTOOL_WSL_WORKAROUND_ENABLED=1")
+    endif()
+
 elseif (${CMAKE_SYSTEM_NAME} STREQUAL "kFreeBSD")
     set(TARGET_OS_FREEBSD 1)
     set(TARGET_OS_DEBIAN_FREEBSD 1)
@@ -19,6 +28,7 @@ elseif (${CMAKE_SYSTEM_NAME} STREQUAL "kFreeBSD")
     add_definitions("-D_FILE_OFFSET_BITS=64")
     find_package_message(PLATFORM "Building for Debian/kFreeBSD"
         "${CMAKE_SYSTEM_NAME}")
+
 elseif (${CMAKE_SYSTEM_NAME} STREQUAL "FreeBSD")
     set(TARGET_OS_FREEBSD 1)
     find_package_message(PLATFORM "Building for FreeBSD" "${CMAKE_SYSTEM_NAME}")
@@ -57,9 +67,11 @@ elseif (${CMAKE_SYSTEM_NAME} STREQUAL "FreeBSD")
                         "system libraries is not supported")
     endif()
     unset(REAL_OPENSSL_ROOT_DIR)
+
 elseif (${CMAKE_SYSTEM_NAME} STREQUAL "NetBSD")
     set(TARGET_OS_NETBSD 1)
     find_package_message(PLATFORM "Building for NetBSD" "${CMAKE_SYSTEM_NAME}")
+
 elseif (${CMAKE_SYSTEM_NAME} STREQUAL "Darwin")
     set(TARGET_OS_DARWIN 1)
 
@@ -154,6 +166,7 @@ elseif (${CMAKE_SYSTEM_NAME} STREQUAL "Darwin")
             endif()
         endif()
     endif()
+
 else()
     message (FATAL_ERROR "Unsupported platform -- ${CMAKE_SYSTEM_NAME}")
 endif()
diff --git a/src/lib/core/evio.c b/src/lib/core/evio.c
index 2152c15e6..6793236da 100644
--- a/src/lib/core/evio.c
+++ b/src/lib/core/evio.c
@@ -140,6 +140,7 @@ evio_setsockopt_server(int fd, int family, int type)
 		       &on, sizeof(on)))
 		return -1;
 
+#ifndef TARANTOOL_WSL_WORKAROUND_ENABLED
 	/* Send all buffered messages on socket before take
 	 * control out from close(2) or shutdown(2). */
 	struct linger linger = { 0, 0 };
@@ -147,6 +148,7 @@ evio_setsockopt_server(int fd, int family, int type)
 	if (sio_setsockopt(fd, SOL_SOCKET, SO_LINGER,
 		       &linger, sizeof(linger)))
 		return -1;
+#endif
 	if (type == SOCK_STREAM && family != AF_UNIX &&
 	    evio_setsockopt_keepalive(fd) != 0)
 		return -1;
-- 
2.20.1

Re: [Tarantool-patches] [PATCH] evio: workaround for wsl1 so_linger assertion

[Cyrill Gorcunov]

On Mon, Mar 16, 2020 at 06:25:22PM +0300, Timur Safin wrote:
> SO_LINGER makes no much sense for unix-sockets, and Microsoft WSL
> is returning EINVAL if setsockopts called for SO_LINGER over unix
> sockets:
> 
>   [004] 2020-03-11 18:42:29.592 [29182] main/102/app sio.c:169 !> SystemError setsockopt(SO_LINGER), called on fd 16, aka
>   [004] 2020-03-11 18:42:29.592 [29182] main/102/app F> can't initialize storage: setsockopt(SO_LINGER), called on fd 16,
>   [004] 2020-03-11 18:42:29.592 [29182] main/102/app F> can't initialize storage: setsockopt(SO_LINGER), called on fd 16,
> 
> And it's sort of correct here, but the problem is Linux is simply
> silently ignoring it, which passes tests.
> 
> After much debates we decided to work-around this case via CMAKE
> define.
> 
> NB! In a future (April/May 2020), when WSL2 with full Linux kernel
> would be released we should disable this check.
> ---
> 
> This patch replaces prior one tsafin/gh-4659-wsl-no-linger-assert because original
> approach considered unsafe, and we rather want to workaround it via CMake instead.
> 
> 
> Branch: https://github.com/tarantool/tarantool/tree/tsafin/wsl1-no-linger-assert
> 
>  cmake/os.cmake      | 13 +++++++++++++
>  src/lib/core/evio.c |  2 ++
>  2 files changed, 15 insertions(+)
> 
> diff --git a/cmake/os.cmake b/cmake/os.cmake
> index 0ed554b9b..ee8870d21 100644
> --- a/cmake/os.cmake
> +++ b/cmake/os.cmake
> @@ -11,6 +11,15 @@ if (${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
>  # (see man page for feature_test_macros).
>      add_definitions("-D_FILE_OFFSET_BITS=64")
>      find_package_message(PLATFORM "Building for Linux" "${CMAKE_SYSTEM_NAME}")
> +
> +    # There are some subtle differences in Linux kernel calls
> +    # implementation under WSL1 (which should go away with WSL2 kernel)
> +    # so for a moment we introduce a way to distinguish Linux and
> +    # Microsoft/WSL1
> +    if (${CMAKE_SYSTEM} MATCHES "Linux-.*-Microsoft")
> +        add_definitions("-DTARANTOOL_WSL_WORKAROUND_ENABLED=1")
> +    endif()
> +

Great! Thanks a huge, Timur! Since you mentioned WSL2 maybe we should
name it -DTARANTOOL_WSL_1 ? I don't insist though.

Acked-by: Cyrill Gorcunov <gorcunov@gmail.com>

Re: [Tarantool-patches] [PATCH] evio: workaround for wsl1 so_linger assertion

[Timur Safin]

: -----Original Message-----
: From: Cyrill Gorcunov <gorcunov@gmail.com>
: Sent: Monday, March 16, 2020 6:32 PM
: To: Timur Safin <tsafin@tarantool.org>
: Cc: Vladislav Shpilevoy <v.shpilevoy@tarantool.org>; Cyrill Gorcunov
: <gorcunov@tarantool.org>; tarantool-patches@dev.tarantool.org
: Subject: Re: [Tarantool-patches] [PATCH] evio: workaround for wsl1
: so_linger assertion
: 


: > +
: > +    # There are some subtle differences in Linux kernel calls
: > +    # implementation under WSL1 (which should go away with WSL2 kernel)
: > +    # so for a moment we introduce a way to distinguish Linux and
: > +    # Microsoft/WSL1
: > +    if (${CMAKE_SYSTEM} MATCHES "Linux-.*-Microsoft")
: > +        add_definitions("-DTARANTOOL_WSL_WORKAROUND_ENABLED=1")
: > +    endif()
: > +
: 
: Great! Thanks a huge, Timur! Since you mentioned WSL2 maybe we should
: name it -DTARANTOOL_WSL_1 ? I don't insist though.
: 
: Acked-by: Cyrill Gorcunov <gorcunov@gmail.com>

Agreed, will rename it shortly to *WSL1_WORKAROUND*

Timur

Re: [Tarantool-patches] [PATCH] evio: workaround for wsl1 so_linger assertion

[Cyrill Gorcunov]

On Mon, Mar 16, 2020 at 07:04:52PM +0300, Timur Safin wrote:
> : 
> : Acked-by: Cyrill Gorcunov <gorcunov@gmail.com>
> 
> Agreed, will rename it shortly to *WSL1_WORKAROUND*

Thanks. You can preserve my Ack tnen.

[Tarantool-patches] Отзыв: [PATCH] evio: workaround for wsl1 so_linger assertion

[Тимур Сафин]

[-- Attachment #1: Type: text/plain, Size: 113 bytes --]

оНКЭГНБЮРЕКЭ рХЛСП яЮТХМ УНРЕК АШ НРНГБЮРЭ ЯННАЫЕМХЕ, "[PATCH] evio:
workaround for wsl1 so_linger assertion".

[-- Attachment #2: winmail.dat --]
[-- Type: application/ms-tnef, Size: 1697 bytes --]