From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from smtpng3.m.smailru.net (smtpng3.m.smailru.net [94.100.177.149]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by dev.tarantool.org (Postfix) with ESMTPS id F062046970E for ; Sat, 21 Dec 2019 20:50:52 +0300 (MSK) Content-Type: text/plain; charset=us-ascii Mime-Version: 1.0 (Mac OS X Mail 13.0 \(3594.4.19\)) From: Roman Khabibov In-Reply-To: <20191218145852.zvqd7kyuodyzv7cm@tkn_work_nb> Date: Sat, 21 Dec 2019 20:50:51 +0300 Content-Transfer-Encoding: quoted-printable Message-Id: <77F0F11F-1910-4078-B606-78336F63ACDB@tarantool.org> References: <20190723123954.dgtfsetqlmjhued3@tkn_work_nb> <704142F3-802F-4C1C-8FA2-6155BFF9A951@tarantool.org> <20190829004547.k24fktai33tlbl73@tkn_work_nb> <868EAF2C-A491-46C9-AD37-7512D6CAB213@tarantool.org> <20190906134509.egjpa2vxfdolkeme@tkn_work_nb> <20191101143929.7gv53nqgscwpbayv@tkn_work_nb> <0D96FEEE-3814-4C90-980F-723A837F8157@tarantool.org> <20191208194800.b6oiyzy3zx23mv4h@tkn_work_nb> <35E941DB-8252-478B-80C2-592FFAA6011E@tarantool.org> <20191218145852.zvqd7kyuodyzv7cm@tkn_work_nb> Subject: Re: [Tarantool-patches] [tarantool-patches] [server-dev] Re: Re: [PATCH v2 1/2] lua: return getaddrinfo() errors List-Id: Tarantool development patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Alexander Turenko Cc: tarantool-patches@dev.tarantool.org Hi! Thanks for the review. > On Dec 18, 2019, at 17:58, Alexander Turenko = wrote: >=20 >>>> @@ -1360,9 +1363,12 @@ local function lsocket_tcp_connect(self, = host, port) >>>> -- This function is broken by design >>>> local ga_opts =3D { family =3D 'AF_INET', type =3D 'SOCK_STREAM' = } >>>> local timeout =3D deadline - fiber.clock() >>>> - local dns =3D getaddrinfo(host, port, timeout, ga_opts) >>>> - if dns =3D=3D nil or #dns =3D=3D 0 then >>>> - self._errno =3D boxerrno.EINVAL >>>> + local dns, err =3D getaddrinfo(host, port, timeout, ga_opts) >>>> + if dns =3D=3D nil then >>>> + return nil, err >>>> + end >>>> + if #dns =3D=3D 0 then >>>> + boxerrno(self._errno) >>>> return nil, socket_error(self) >>>> end >>>> for _, remote in ipairs(dns) do >>>=20 >>> EINVAL here is changed to self._errno, which is always `nil` at = least >>> for typical usage of lua socket: `s =3D socket.tcp()`, then = `s:connect()`. >>>=20 >>> Let's keep the old behaviour (EINVAL) when getaddrinfo() returns = zero >>> amount of addresses. >>>=20 >>> RFC 2553 ('Basic Socket Interface Extensions for IPv6') states: >>>=20 >>> | Upon successful return a pointer to a linked list of one or more >>> | addrinfo structures is returned through the final argument. >>>=20 >>> RFC 3493 ('Basic Socket Interface Extensions for IPv6') states the = same, >>> but more explicitly: >>>=20 >>> | Upon successful return of getaddrinfo(), the location to which res >>> | points shall refer to a linked list of addrinfo structures, each = of >>> | which shall specify a socket address and information for use in >>> | creating a socket with which to use that socket address. The list >>> | shall include at least one addrinfo structure. >>>=20 >>> So this situation should not occur. But anyway we should not miss to = set >>> a certain errno in the case at least for consistency with = tcp_connect(). >>>=20 >>> Also I think that we should set errno to EINVAL when getaddrinfo() = fails >>> to keep the code backward compatible: say, a user lean on this value >>> when tcp_connect() / :connect() returns `nil`. >>=20 >> I understand so: >>=20 >> diff --git a/src/lua/socket.lua b/src/lua/socket.lua >> index a334ad45b..6c829a11c 100644 >> --- a/src/lua/socket.lua >> +++ b/src/lua/socket.lua >> @@ -995,7 +995,13 @@ local function getaddrinfo(host, port, timeout, = opts) >> end >>=20 >> end >> - return internal.getaddrinfo(host, port, timeout, ga_opts) >> + >> + local ret, err =3D internal.getaddrinfo(host, port, timeout, = ga_opts) >> + if ret =3D=3D nil then >> + boxerrno(boxerrno.EINVAL) >> + return nil, err >> + end >> + return ret >> end >>=20 >> -- tcp connector >> @@ -1041,9 +1047,12 @@ local function tcp_connect(host, port, = timeout) >> end >> local timeout =3D timeout or TIMEOUT_INFINITY >> local stop =3D fiber.clock() + timeout >> - local dns =3D getaddrinfo(host, port, timeout, { type =3D = 'SOCK_STREAM', >> + local dns, err =3D getaddrinfo(host, port, timeout, { type =3D = 'SOCK_STREAM', >> protocol =3D 'tcp' }) >> - if dns =3D=3D nil or #dns =3D=3D 0 then >> + if dns =3D=3D nil then >> + return nil, err >> + end >> + if #dns =3D=3D 0 then >> boxerrno(boxerrno.EINVAL) >> return nil >> end >> @@ -1360,8 +1369,12 @@ local function lsocket_tcp_connect(self, host, = port) >> -- This function is broken by design >> local ga_opts =3D { family =3D 'AF_INET', type =3D 'SOCK_STREAM' = } >> local timeout =3D deadline - fiber.clock() >> - local dns =3D getaddrinfo(host, port, timeout, ga_opts) >> - if dns =3D=3D nil or #dns =3D=3D 0 then >> + local dns, err =3D getaddrinfo(host, port, timeout, ga_opts) >> + if dns =3D=3D nil then >> + self._errno =3D boxerrno() >> + return nil, err >> + end >> + if #dns =3D=3D 0 then >> self._errno =3D boxerrno.EINVAL >> return nil, socket_error(self) >> end >> @@ -1547,9 +1560,12 @@ local function lsocket_connect(host, port) >> if host =3D=3D nil or port =3D=3D nil then >> error("Usage: luasocket.connect(host, port)") >> end >> - local s =3D tcp_connect(host, port) >> + local s, err =3D tcp_connect(host, port) >> if not s then >> - return nil, boxerrno.strerror() >> + if not err then >> + return nil, boxerrno.strerror() >> + end >> + return nil, err >> end >> setmetatable(s, lsocket_tcp_client_mt) >> return s >=20 > I meant that `self._errno =3D boxerrno.EINVAL` was changed to > `boxerrno(self._errno)` in `#dns =3D=3D 0` case and it does not more = affect > the second return value (an error message). I would return it back for > `#dns =3D=3D 0` case (only for it). >=20 > The motivation is just to don't change things that we don't asked to > change. >=20 > I didn't mean changes in getaddrinfo() function itself. @@ -1360,8 +1363,12 @@ local function lsocket_tcp_connect(self, host, = port) -- This function is broken by design local ga_opts =3D { family =3D 'AF_INET', type =3D 'SOCK_STREAM' } local timeout =3D deadline - fiber.clock() - local dns =3D getaddrinfo(host, port, timeout, ga_opts) - if dns =3D=3D nil or #dns =3D=3D 0 then + local dns, err =3D getaddrinfo(host, port, timeout, ga_opts) + if dns =3D=3D nil then + self._errno =3D boxerrno() + return nil, err + end + if #dns =3D=3D 0 then self._errno =3D boxerrno.EINVAL return nil, socket_error(self) end >>> Also I see that we don't document Lua Socket part of the API: >>> https://www.tarantool.io/en/doc/1.10/reference/reference_lua/socket/ >>=20 >> Is this the part of this patchset? Maybe, it would be better to do = doc >> request separately. >=20 > It was in context of your docbot request: there is no need to ask to > update Lua Socket API docs if we does not provide it at all. >=20 > Lua Socket has its own API documentation, so it is maybe okay to don't > document it again on our website. commit 7a85a217cb97faf03c8c817fc2748749abd91e91 Author: Roman Khabibov Date: Thu Nov 21 14:37:54 2019 +0300 lua: return getaddrinfo() errors =20 Add getaddrinfo() errors into the several fuctions of socket. Now getaddrinfo() can return a pair of values (nil and error message) in case of error. =20 Closes #4138 =20 @TarantoolBot document Title: socket API changes =20 * socket.getaddrinfo() =20 Can return error message as second return value in case of internal getaddrinfo() error. =20 Example: tarantool> socket.getaddrinfo('non_exists_hostname', 3301) --- - null - 'getaddrinfo: nodename nor servname provided, or not known' ... =20 * socket.tcp_connect() =20 Can return socket.getaddrinfo() error message as second return value. =20 Example: tarantool> socket.tcp_connect('non_exists_hostname', 3301) --- - null - 'getaddrinfo: nodename nor servname provided, or not known' ... diff --git a/src/box/lua/net_box.lua b/src/box/lua/net_box.lua index c2e1bb9c4..2fb3e6df6 100644 --- a/src/box/lua/net_box.lua +++ b/src/box/lua/net_box.lua @@ -150,9 +150,12 @@ local function next_id(id) return band(id + 1, = 0x7FFFFFFF) end local function establish_connection(host, port, timeout) local timeout =3D timeout or DEFAULT_CONNECT_TIMEOUT local begin =3D fiber.clock() - local s =3D socket.tcp_connect(host, port, timeout) + local s, err =3D socket.tcp_connect(host, port, timeout) if not s then - return nil, errno.strerror(errno()) + if not err then + return nil, errno.strerror(errno()) + end + return nil, err end local msg =3D s:read({chunk =3D IPROTO_GREETING_SIZE}, timeout - (fiber.clock() - begin)) diff --git a/src/lua/socket.c b/src/lua/socket.c index 130378caf..dbade2d27 100644 --- a/src/lua/socket.c +++ b/src/lua/socket.c @@ -774,6 +774,18 @@ lbox_getaddrinfo_result_wrapper(struct lua_State = *L) return 1; } =20 +/** + * Wrap coio_getaddrinfo() and call it. Push returned values onto + * @a L Lua stack. + * + * @param L Lua stack. + * + * @retval 1 Number of returned values by Lua function if + * coio_getaddrinfo() success. + * @retval 2 Number of returned values by Lua function if + * coio_getaddrinfo() is failed (first is nil, second is error + * message). + */ static int lbox_socket_getaddrinfo(struct lua_State *L) { @@ -816,7 +828,9 @@ lbox_socket_getaddrinfo(struct lua_State *L) =20 if (dns_res !=3D 0) { lua_pushnil(L); - return 1; + struct error *err =3D diag_get()->last; + lua_pushstring(L, err->errmsg); + return 2; } =20 /* no results */ diff --git a/src/lua/socket.lua b/src/lua/socket.lua index a334ad45b..f9256ecdc 100644 --- a/src/lua/socket.lua +++ b/src/lua/socket.lua @@ -1041,9 +1041,12 @@ local function tcp_connect(host, port, timeout) end local timeout =3D timeout or TIMEOUT_INFINITY local stop =3D fiber.clock() + timeout - local dns =3D getaddrinfo(host, port, timeout, { type =3D = 'SOCK_STREAM', + local dns, err =3D getaddrinfo(host, port, timeout, { type =3D = 'SOCK_STREAM', protocol =3D 'tcp' }) - if dns =3D=3D nil or #dns =3D=3D 0 then + if dns =3D=3D nil then + return nil, err + end + if #dns =3D=3D 0 then boxerrno(boxerrno.EINVAL) return nil end @@ -1360,8 +1363,12 @@ local function lsocket_tcp_connect(self, host, = port) -- This function is broken by design local ga_opts =3D { family =3D 'AF_INET', type =3D 'SOCK_STREAM' } local timeout =3D deadline - fiber.clock() - local dns =3D getaddrinfo(host, port, timeout, ga_opts) - if dns =3D=3D nil or #dns =3D=3D 0 then + local dns, err =3D getaddrinfo(host, port, timeout, ga_opts) + if dns =3D=3D nil then + self._errno =3D boxerrno() + return nil, err + end + if #dns =3D=3D 0 then self._errno =3D boxerrno.EINVAL return nil, socket_error(self) end @@ -1547,9 +1554,12 @@ local function lsocket_connect(host, port) if host =3D=3D nil or port =3D=3D nil then error("Usage: luasocket.connect(host, port)") end - local s =3D tcp_connect(host, port) + local s, err =3D tcp_connect(host, port) if not s then - return nil, boxerrno.strerror() + if not err then + return nil, boxerrno.strerror() + end + return nil, err end setmetatable(s, lsocket_tcp_client_mt) return s diff --git a/test/app/socket.result b/test/app/socket.result index fd299424c..6e5bd7a35 100644 --- a/test/app/socket.result +++ b/test/app/socket.result @@ -941,10 +941,20 @@ sc:close() - true ... -- tcp_connect --- test timeout -socket.tcp_connect('127.0.0.1', 80, 0.00000000001) +-- Test timeout. In this test, tcp_connect can return the second +-- output value from internal.getaddrinfo (usually on Mac OS, but +-- theoretically it can happen on Linux too). Sometimes +-- getaddrinfo() is timed out, sometimes connect. On Linux however +-- getaddrinfo is fast enough to never give timeout error in +-- the case. So, there are two sources of timeout errors that are +-- reported differently. This difference has appeared after +-- gh-4138 patch. +s, err =3D socket.tcp_connect('127.0.0.1', 80, 0.00000000001) --- -- null +... +s =3D=3D nil +--- +- true ... -- AF_INET s =3D socket('AF_INET', 'SOCK_STREAM', 'tcp') @@ -1606,7 +1616,7 @@ fio.stat(path) =3D=3D nil { socket.tcp_connect('abrakadabra#123') =3D=3D nil, errno.strerror() } --- - - true - - Invalid argument + - Input/output error ... -- wrong options for getaddrinfo socket.getaddrinfo('host', 'port', { type =3D 'WRONG' }) =3D=3D nil and = errno() =3D=3D errno.EINVAL @@ -2816,6 +2826,48 @@ test_run:cmd("clear filter") --- - true ... +-- gh-4138 Check getaddrinfo() error from socket:connect() only. +-- Error code and error message returned by getaddrinfo() depends +-- on system's gai_strerror(). +test_run:cmd("setopt delimiter ';'") +--- +- true +... +function check_err(err) + if err =3D=3D 'getaddrinfo: nodename nor servname provided, or not = known' or + err =3D=3D 'getaddrinfo: Servname not supported for ai_socktype' = or + err =3D=3D 'getaddrinfo: Name or service not known' then + return true + end + return false +end; +--- +... +test_run:cmd("setopt delimiter ''"); +--- +- true +... +s, err =3D socket.getaddrinfo('non_exists_hostname', 3301) +--- +... +check_err(err) +--- +- true +... +s, err =3D socket.connect('non_exists_hostname', 3301) +--- +... +check_err(err) +--- +- true +... +s, err =3D socket.tcp_connect('non_exists_hostname', 3301) +--- +... +check_err(err) +--- +- true +... -- case: sicket receive inconsistent behavior chan =3D fiber.channel() --- diff --git a/test/app/socket.test.lua b/test/app/socket.test.lua index c72d41763..7086e496b 100644 --- a/test/app/socket.test.lua +++ b/test/app/socket.test.lua @@ -300,8 +300,16 @@ sc:close() =20 -- tcp_connect =20 --- test timeout -socket.tcp_connect('127.0.0.1', 80, 0.00000000001) +-- Test timeout. In this test, tcp_connect can return the second +-- output value from internal.getaddrinfo (usually on Mac OS, but +-- theoretically it can happen on Linux too). Sometimes +-- getaddrinfo() is timed out, sometimes connect. On Linux however +-- getaddrinfo is fast enough to never give timeout error in +-- the case. So, there are two sources of timeout errors that are +-- reported differently. This difference has appeared after +-- gh-4138 patch. +s, err =3D socket.tcp_connect('127.0.0.1', 80, 0.00000000001) +s =3D=3D nil =20 -- AF_INET s =3D socket('AF_INET', 'SOCK_STREAM', 'tcp') @@ -960,6 +968,27 @@ server:close() =20 test_run:cmd("clear filter") =20 +-- gh-4138 Check getaddrinfo() error from socket:connect() only. +-- Error code and error message returned by getaddrinfo() depends +-- on system's gai_strerror(). +test_run:cmd("setopt delimiter ';'") +function check_err(err) + if err =3D=3D 'getaddrinfo: nodename nor servname provided, or not = known' or + err =3D=3D 'getaddrinfo: Servname not supported for ai_socktype' = or + err =3D=3D 'getaddrinfo: Name or service not known' then + return true + end + return false +end; +test_run:cmd("setopt delimiter ''"); + +s, err =3D socket.getaddrinfo('non_exists_hostname', 3301) +check_err(err) +s, err =3D socket.connect('non_exists_hostname', 3301) +check_err(err) +s, err =3D socket.tcp_connect('non_exists_hostname', 3301) +check_err(err) + -- case: sicket receive inconsistent behavior chan =3D fiber.channel() counter =3D 0 diff --git a/test/box/net.box.result b/test/box/net.box.result index e3dabf7d9..0635ca9e7 100644 --- a/test/box/net.box.result +++ b/test/box/net.box.result @@ -3927,6 +3927,34 @@ test_run:grep_log('default', '00000040:.*') --- - null ... +-- gh-4138 Check getaddrinfo() error from connect() only. Error +-- code and error message returned by getaddrinfo() depends on +-- system's gai_strerror(). +test_run:cmd("setopt delimiter ';'") +--- +- true +... +function check_err(err) + if err =3D=3D 'getaddrinfo: nodename nor servname provided, or not = known' or + err =3D=3D 'getaddrinfo: Servname not supported for ai_socktype' = or + err =3D=3D 'getaddrinfo: Name or service not known' then + return true + end + return false +end; +--- +... +test_run:cmd("setopt delimiter ''"); +--- +- true +... +s =3D remote.connect('non_exists_hostname:3301') +--- +... +check_err(s['error']) +--- +- true +... box.cfg{log_level=3Dlog_level} --- ... diff --git a/test/box/net.box.test.lua b/test/box/net.box.test.lua index 8e65ff470..9c9aee1ad 100644 --- a/test/box/net.box.test.lua +++ b/test/box/net.box.test.lua @@ -1582,4 +1582,21 @@ test_run:wait_log('default', '00000030:.*', nil, = 10) -- we expect nothing below, so don't wait test_run:grep_log('default', '00000040:.*') =20 +-- gh-4138 Check getaddrinfo() error from connect() only. Error +-- code and error message returned by getaddrinfo() depends on +-- system's gai_strerror(). +test_run:cmd("setopt delimiter ';'") +function check_err(err) + if err =3D=3D 'getaddrinfo: nodename nor servname provided, or not = known' or + err =3D=3D 'getaddrinfo: Servname not supported for ai_socktype' = or + err =3D=3D 'getaddrinfo: Name or service not known' then + return true + end + return false +end; +test_run:cmd("setopt delimiter ''"); + +s =3D remote.connect('non_exists_hostname:3301') +check_err(s['error']) + box.cfg{log_level=3Dlog_level}