[tarantool-patches] Re: [server-dev] Re: Re: [PATCH v2 1/2] lua: return getaddrinfo() errors

[Roman Khabibov <roman.habibov@tarantool.org>]

>>>> +-- gh-4138 Check getaddrinfo() error from socket:connect() only.
>>>> +-- Error code and error message returned by getaddrinfo() depends
>>>> +-- on system's gai_strerror(). So that there is no checking for
>>>> +-- certain error message.
>>>> +test_run:cmd("setopt delimiter ';'")
>>>> +---
>>>> +- true
>>>> +...
>>>> +function check_err(err)
>>>> +    if err == 'getaddrinfo: nodename nor servname provided, or not known' or
>>>> +       err == 'getaddrinfo: Servname not supported for ai_socktype' or
>>>> +       err == 'getaddrinfo: Name or service not known' then
>>>> +            return true
>>>> +    end
>>>> +    return false
>>>> +end;
>>>> +---
>>>> +...
>>>> +test_run:cmd("setopt delimiter ''");
>>>> +---
>>>> +- true
>>>> +...
>>>> +s, err = socket:connect('non_exists_hostname:3301')
>>> 
>>> socket.connect() should be called, the colon here is the mistake.
> 
> Please, look again to the code. One call is socket.connect(...), another
> is socket:connect().
> 
> Just in case, I suggest to add a patch to your patchset to keep youself
> from such mistakes:
> 
> | diff --git a/src/lua/socket.lua b/src/lua/socket.lua
> | index e4815adbb..a1c13d4db 100644
> | --- a/src/lua/socket.lua
> | +++ b/src/lua/socket.lua
> | @@ -1554,7 +1554,8 @@ local function lsocket_tcp()
> |  end
> |  
> |  local function lsocket_connect(host, port)
> | -    if host == nil or port == nil then
> | +    if type(host) ~= 'string' or (type(port) ~= 'string' and
> | +            type(port) ~= 'number') then
> | 		 error("Usage: luasocket.connect(host, port)")
> | 	 end
> | 	 local s, err = tcp_connect(host, port)
> | @@ -1569,7 +1570,8 @@ local function lsocket_connect(host, port)
> |  end
> |  
> |  local function lsocket_bind(host, port, backlog)
> | -    if host == nil or port == nil then
> | +    if type(host) ~= 'string' or (type(port) ~= 'string' and
> | +            type(port) ~= 'number') then
> | 		 error("Usage: luasocket.bind(host, port [, backlog])")
> | 	 end
> | 	 local function prepare(s) return backlog end
+s, err = socket.getaddrinfo('non_exists_hostname', 3301)
+check_err(err)
+s, err = socket.connect('non_exists_hostname', 3301)
+check_err(err)
+

Also, I added suggested patch.
>>> I think we should test both native and Lua Socket API, because both were
>>> changed.
>>> 
>>> Should we ever change Lua Socket API? Please, verify it with a
>>> documentation and an actual behaviour of the external module. We emulate
>>> it, so we should be compatible.
> 
> Are you look into this?
http://w3.impa.br/~diego/software/luasocket/tcp.html#connect
master:connect(address, port)

Attempts to connect a master object to a remote host, transforming it into a client object. Client objects support methods send, receive, getsockname, getpeername, settimeout, and close.

Address can be an IP address or a host name. Port must be an integer number in the range [1..64K). 

In case of error, the method returns nil followed by a string describing the error. In case of success, the method returns 1.

I looked to this doc. It seems like our behaviour corresponds this API.

commit 9810ccd869183de068939cdce5a157ad35916d36
Author: Roman Khabibov <roman.habibov@tarantool.org>
Date:   Tue Jul 30 15:49:13 2019 +0300

    lua: return getaddrinfo() errors
    
    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.
    
    Closes #4138
    
    @TarantoolBot document
    Title: socket_object:connect
    
    Connect to a remote host using tcp_connect() within. If it is no
    error, return a socket object.

diff --git a/src/box/lua/net_box.lua b/src/box/lua/net_box.lua
index 31a8c16b7..beab3b0a7 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 = timeout or DEFAULT_CONNECT_TIMEOUT
     local begin = fiber.clock()
-    local s = socket.tcp_connect(host, port, timeout)
+    local s, err = 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 = s:read({chunk = 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;
 }
 
+/**
+ * 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)
 
 	if (dns_res != 0) {
 		lua_pushnil(L);
-		return 1;
+		struct error *err = diag_get()->last;
+		lua_pushstring(L, err->errmsg);
+		return 2;
 	}
 
 	/* no results */
diff --git a/src/lua/socket.lua b/src/lua/socket.lua
index a334ad45b..e4815adbb 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 = timeout or TIMEOUT_INFINITY
     local stop = fiber.clock() + timeout
-    local dns = getaddrinfo(host, port, timeout, { type = 'SOCK_STREAM',
+    local dns, err = getaddrinfo(host, port, timeout, { type = 'SOCK_STREAM',
         protocol = 'tcp' })
-    if dns == nil or #dns == 0 then
+    if dns == nil then
+        return nil, err
+    end
+    if #dns == 0 then
         boxerrno(boxerrno.EINVAL)
         return nil
     end
@@ -1360,9 +1363,16 @@ local function lsocket_tcp_connect(self, host, port)
     -- This function is broken by design
     local ga_opts = { family = 'AF_INET', type = 'SOCK_STREAM' }
     local timeout = deadline - fiber.clock()
-    local dns = getaddrinfo(host, port, timeout, ga_opts)
+    local dns, err = getaddrinfo(host, port, timeout, ga_opts)
     if dns == nil or #dns == 0 then
-        self._errno = boxerrno.EINVAL
+        return nil, err
+    end
+    if dns == nil then
+        boxerrno(self._errno)
+        return nil, err
+    end
+    if #dns == 0 then
+        boxerrno(self._errno)
         return nil, socket_error(self)
     end
     for _, remote in ipairs(dns) do
@@ -1547,9 +1557,12 @@ local function lsocket_connect(host, port)
     if host == nil or port == nil then
         error("Usage: luasocket.connect(host, port)")
     end
-    local s = tcp_connect(host, port)
+    local s, err = 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..87b1d0387 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 = socket.tcp_connect('127.0.0.1', 80, 0.00000000001)
 ---
-- null
+...
+s == nil
+---
+- true
 ...
 -- AF_INET
 s = socket('AF_INET', 'SOCK_STREAM', 'tcp')
@@ -1606,7 +1616,7 @@ fio.stat(path) == nil
 { socket.tcp_connect('abrakadabra#123') == nil, errno.strerror() }
 ---
 - - true
-  - Invalid argument
+  - Input/output error
 ...
 -- wrong options for getaddrinfo
 socket.getaddrinfo('host', 'port', { type = 'WRONG' }) == nil and errno() == errno.EINVAL
@@ -2812,6 +2822,41 @@ server:close()
 ---
 - 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 == 'getaddrinfo: nodename nor servname provided, or not known' or
+       err == 'getaddrinfo: Servname not supported for ai_socktype' or
+       err == 'getaddrinfo: Name or service not known' then
+            return true
+    end
+    return false
+end;
+---
+...
+test_run:cmd("setopt delimiter ''");
+---
+- true
+...
+s, err = socket.getaddrinfo('non_exists_hostname', 3301)
+---
+...
+check_err(err)
+---
+- true
+...
+s, err = socket.connect('non_exists_hostname', 3301)
+---
+...
+check_err(err)
+---
+- true
+...
 test_run:cmd("clear filter")
 ---
 - true
diff --git a/test/app/socket.test.lua b/test/app/socket.test.lua
index c72d41763..08cb7fc9e 100644
--- a/test/app/socket.test.lua
+++ b/test/app/socket.test.lua
@@ -300,8 +300,16 @@ sc:close()
 
 -- 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 = socket.tcp_connect('127.0.0.1', 80, 0.00000000001)
+s == nil
 
 -- AF_INET
 s = socket('AF_INET', 'SOCK_STREAM', 'tcp')
@@ -958,6 +966,25 @@ fiber.cancel(echo_fiber)
 client:read(1, 5) == ''
 server:close()
 
+-- 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 == 'getaddrinfo: nodename nor servname provided, or not known' or
+       err == 'getaddrinfo: Servname not supported for ai_socktype' or
+       err == 'getaddrinfo: Name or service not known' then
+            return true
+    end
+    return false
+end;
+test_run:cmd("setopt delimiter ''");
+
+s, err = socket.getaddrinfo('non_exists_hostname', 3301)
+check_err(err)
+s, err = socket.connect('non_exists_hostname', 3301)
+check_err(err)
+
 test_run:cmd("clear filter")
 
 -- case: sicket receive inconsistent behavior
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 == 'getaddrinfo: nodename nor servname provided, or not known' or
+       err == 'getaddrinfo: Servname not supported for ai_socktype' or
+       err == 'getaddrinfo: Name or service not known' then
+            return true
+    end
+    return false
+end;
+---
+...
+test_run:cmd("setopt delimiter ''");
+---
+- true
+...
+s = remote.connect('non_exists_hostname:3301')
+---
+...
+check_err(s['error'])
+---
+- true
+...
 box.cfg{log_level=log_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:.*')
 
+-- 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 == 'getaddrinfo: nodename nor servname provided, or not known' or
+       err == 'getaddrinfo: Servname not supported for ai_socktype' or
+       err == 'getaddrinfo: Name or service not known' then
+            return true
+    end
+    return false
+end;
+test_run:cmd("setopt delimiter ''");
+
+s = remote.connect('non_exists_hostname:3301')
+check_err(s['error'])
+
 box.cfg{log_level=log_level}