- Chapter 1: A Simple UDP Server & Client
- 1.1.1. Introduction and Description of UDP
- 1.1.2. Creating a UDP/IP Server
- 1.1.2.1. Create the Socket
- 1.1.2.2. Identify/Name the Socket
- 1.1.2.3. Begin a Loop
- 1.1.2.4. Wait for a Message
- 1.1.2.5. Reply to Message
- 1.1.2.6. Close the Socket
- 1.1.2.7. End the Loop
- 1.2. Create Basic UDP Client
- Chapter 2: Layering DTLS onto Simple Server and Client
- 2.1.1. New Imports
- 2.1.2. Increase MSGLEN
- 2.1.3. Shifting Variables, Adding Functionality
- 2.1.3.1. Create Cleanup
- 2.1.3.2. Create Ctx
- 2.1.4. Tie Client Connection Together in Main
- 2.1.4.1. Add an Acknowledge
- 2.1.4.2. Remove Unnecessary Variables and Move all Declarations Together
- 2.1.4.3. Loop Shift
- 2.1.5. Update Main Function
- 2.1.5.1. If Defined, Turn on wolfSSL Debugging
- 2.1.5.2. Initialize wolfSSL, Load Certificates and Keys
- 2.1.5.3. Add Memory Freeing Functionality
- 2.1.6. Quick Recap
- 2.1.7. Adding DTLS to Server
- 2.1.7.1. Avoid Socket in Use Error
- 2.1.7.2. Await Datagram Arrival
- 2.1.7.3. Using wolfSSL to Open a Session with Client
- 2.1.7.4. Read the Message, Acknowledge that Message was Read
- 2.1.7.5. Free the Memory
- 2.2.1. Enable DTLS
- 2.2.2. wolfSSL Tutorial
- 2.2.2.1. Port Definition
- 2.2.2.2. Edit Arguments
- 2.2.2.3. Implement wolfSSL Read/Write Functions
- 2.2.3. Set Peer
- 2.2.4. Connect
- 2.2.5. Write/Read
- 2.2.6. Shutdown, Free, Cleanup
- 2.2.7. Adjust Makefile
- Chapter 3: Multithreading a DTLS Server with POSIX Threads
- 3.1.1. Elements Required for Thread Management
- 3.1.1.1. threadArgs Struct
- 3.1.1.2. Thread Creation and Deletion
- 3.1.1.3. ThreadHandler Method
- 3.1.1.3.1. pthread_detach
- 3.1.1.3.2. Instance Variables
- 3.1.1.3.3. memcpy
- 3.1.1.3.4. 'ssl' Object Creation
- 3.1.1.3.5. wolfSSL_accept, wolfSSL_read, wolfSSL_write, etc.
- 3.1.1.3.6. Thread Memory Cleanup
- 3.1.2. Review
- Chapter 4: Session Resumption with DTLS
- 4.1.1. Storage of the Previous Session Information
- 4.1.2. Memory Cleanup and Management
- 4.1.3. Reconnect with Old Session Data
- 4.1.4. Cleanup
- Chapter 5: Convert Server and Client to Nonblocking
- 5.1. A Nonblocking DTLS Client
- 5.1.1. A Note About Functions
- 5.1.2. Add New Headers to Top of File
- 5.1.3. Add Enum Variables for Testing Functions
- 5.1.4. Adding Nonblocking DTLS Functionality to Client
- 5.1.4.1. Variables
- 5.1.4.2. Adding a Loop 5.1.5. Datagram Sending
- 5.2. A Nonblocking DTLS Server
- 5.2.1. Add New Headers to Top of File
- 5.2.2. Add Enum Variables for Testing Functions
- 5.2.3. Handling Signals
- 5.2.3.1. Static variable
cleanup
- 5.2.3.2. Defining a Method to Handle Signals
- 5.2.3.3. Telling the Program to Use `sig_handler()
- 5.2.3.1. Static variable
- 5.2.4. Adding Nonblocking DTLS Functionality to Server
- 5.2.4.1. Variables
- 5.2.4.2. Adding a Loop
- 5.2.5. Final Note
- 5.1. A Nonblocking DTLS Client
- References
What is UDP? User Datagram Protocol (UDP) is a core member of Internet Protocol Suite (IPS). Messages sent using UDP are called datagrams. A client can send messages to other hosts (clients or servers that have host capabilities) on an Internet Protocol (IP) network without any prior communications to set up special transmissions channels or data paths. A formal definition can be found in Request for Comments (RFC) 768.
UDP uses a minimal amount of protocol mechanisms in that it requires no handshake. Rather UDP uses a checksum for data integrity and port numbers for addressing different functions at the source and destination of the datagram.
PROS: UDP is ideal for time conservation. The lack of having to re-transmit messages (if packets are lost) makes it suitable for real-time applications such as voice over IP or live streaming.
CONS: UDP provides no guarantees to the upper layer protocol for message delivery and the UDP protocol layer retains no backup of UDP messages once sent. Therefore messages are non-recoverable once sent.
In summary, these are the key points regarding UDP:
- No connection to create and maintain
- More control over when data is sent
- No error recovery
- No compensation for lost packets
- Packets may arrive out of order
- No congestion control
- Overall, UDP is lightweight but unreliable. Some applications where UDP is used include DNS, NFS, and SNMP.
UDP header consists of 4 fields, each of which is 16 bits.
- Field 1 contains information about the port from which the message originated.
- Field 2 contains information about the destination port.
- Field 3 defines the length in bytes of the UDP header and data. Allows for datagram of maximum size 65,535 bytes to be defined. If larger, set to 0. (IPv6 will allow larger).
- Field 4 contains a Checksum for security purposes. Contains some checksum of the header and data.
All 16 bit words are summed using “one's complement arithmetic”. The sum is then “one's complemented” and this is the value placed in the UDP checksum field.
Example: you have two 16 bit words as follows:
ONE's COMPLEMENT OF WORD1: 1 0 0 1 1 1 1 0 0 1 1 0 0 0 1 0
ONE's COMPLEMENT OF WORD2: 0 1 0 1 0 0 0 1 0 1 0 0 1 1 0 1
SUM: 1 1 1 0 1 1 1 1 1 0 1 0 1 1 1 1
-------------------------------
ONE's COMPLEMENT OF SUM: 0 0 0 1 0 0 0 0 0 1 0 1 0 0 0 0
The final value would be placed in the Checksum Field.
There are seven initial steps to creating a UDP/IP Server.
1. Create the socket
2. Identify the socket (IE give it a name)
3. <Begin a loop>
4. On the server wait for a message
5. Send a response to the client once message is received
6. Close the socket (in our case, just return to looking for packets arriving).
7. <End of loop>
A socket is created with the socket system call.
int sockfd = socket(domain, type, protocol);
Let's briefly discuss the parameters domain, type, and protocol.
- Domain The domain can also be referred to as the address family. It is the communication domain in which the socket should be created. Below you will see some of the examples domains or address families that we could work with. At the end will be a description of what we will choose specifically for a UDP server.
AF_INET: Internet Protocol (IP)
AF_INET6: IP version 6 (IPv6)
AF_UNIX: local channel, similar to pipes
AF_ISO: “In Search Of” (ISO) protocols
AF_NS: Xerox Network Systems protocols
- Type This is the type of service we will be providing with our UDP server. This is selected based on the requirements of the application, and will likely aid in determining which Domain (above) you will select ultimately.
SOCK_STREAM: a virtual circuit service
SOCK_DGRAM: a datagram service
SOCK_RAW: a direct IP service
- Protocol A protocol supports the sockets operation. This parameter is optional but is helpful in cases where the domain (family) supports multiple protocols. In these cases we can specify which protocol to use for said family. If the family supports only one type of protocol this value will be zero.
For this tutorial we want to select domain (domain = AF_NET)
and the datagram service (type = SOCK_DGRAM)
. There is only one form of datagram service therefore we do not need to specify a protocol for a UDP/IP server (protocol = 0)
.
#include <sys/socket.h>
…
int sockfd;
sockfd = socket(AF_INET, SOCK_DGRAM, 0);
if (sockfd < 0) {
perror("cannot create socket");
return 0;
}
printf("created socket: descriptor=%d\n", sockfd);
By naming the socket we are assigning a transport address to the socket (I.E. a port number in IP networking). This is more commonly referred to as “binding” an address. The bind system call is used to do this. In other words we are giving our server a unique address so that communication with our server can be established much like a person with a mailing address. They can receive communiques (letters) via their mailbox. Our server can do the same once it has a “bound” socket.
The transport address is defined in the socket address structure. Since sockets can receive and send data using a myriad of communication interfaces, the interface is very general. Rather than accepting a port number as a parameter, it will look for a “sockaddr” (short for socket address)structure whose format is based off the address family we chose.
# include <sys/socket.h>
contains the relevant “bind” call we will need. Since we already used it in 1.1.2.1 we do not need to include it a second time.
int
bind(int socket, const struct sockaddr *address, socklen_t address_len);
The first parameter “socket” is the socket we created in 1.1.2.1 (sockfd)
The second parameter (sockaddr)
will allow the operating system (OS) to read the first couple bytes that identify the address family. For UDP/IP networking we will use struct sockaddr_in
, which is defined in the netinet/in.h
system library.
struct sockaddr_in{
__uint8_t sin_len;
sa_family_t sin_family;
in_port_t sin_port;
struct in_addr sin_addr;
char sin_zero[8];
};
NOTE: this code will not be found in our example server. It is imported with the following call:
# include <netinet/in.h>
Before calling bind, we need to fill this struct. The three key parts we need to set are:
-
sin_family
The address family we used in 1.1.3.1(AF_INET)
. -
sin_port
The port number (transport address). This can either be explicitly declared, or you can allow the OS to assign one. Since we are creating a server, ideally we would want to explicitly declare a well known port so that clients know where to address their messages. However for this particular tutorial we will use the generic11111
(five ones). This will be defined directly beneath the include section of our code:# define SERV_PORT 11111
We can then call
SERV_PORT
where it is needed and if you, the client, are already using port11111
for any particular reason, you can then easily redefine it as needed for this tutorial. Additionally if you use# define SERV_PORT 0
, your system will use any available port. -
sin_addr
The address for our socket (your server machines IP address). With UDP/IP our server will have one IP address for each network interface. Since the address can vary based on transport methods and we are using a client computer to simulate a server, we will use theINADDR_ANY
.
memset(&servAddr, 0, sizeof(servAddr));
servAddr.sin_family = AF_INET;
servAddr.sin_port = htons(SERV_PORT);
inet_pton(AF_INET, argv[1], &servAddr.sin_addr);
- “htons” host to network - short : convert a number into a 16-bit network representation. This is commonly used to store a port number into a sockaddr structure.
- “htonl” host to network - long : convert a number into a 32-bit network representation. This is commonly used to store an IP address into a sockaddr structure.
- “ntohs” network to host - short : convert a 16-bit number from a network representation into the local processor`s format. This is commonly used to read a port number from a sockaddr structure.
- “ntohl” network to host - long : convert a 32-bit number from a network representation into the local processor`s format. This is commonly used to read an IP address from a sockaddr structure.
Using any of the above 4 macros will guarantee that your code remains portable regardless of the architecture you use in compilation.
Later when we layer on DTLS our server will set up a socket for listening via the “listen” system call. The server would then call “accept” upon hearing a request to communicate, and then wait for a connection to be established. UDP however in its base for is connectionless. So our server, as of right now, is capable of listening for a message purely due to the fact that it has a socket! We use recvfrom
system call to wait for an incoming datagram on a specific transport address (IP address, and port number).
The recvfrom
call is included with the # include <sys/socket.h>
therefore we do not need to include this library again since we already included it in 1.1.2.1.
int recvfrom(int socket, void* restrict buffer, size_t length,
int flags, struct sockaddr* restrict src_addr, socklen_t *src_len);
int socket
The first parametersocket
is the socket we created and bound in 1.1.2.1 & 1.1.2.2. The port number assigned to that socket via the “bind” tells us what portrecvfrom
will “watch” while awaiting incoming data transmissions.void* restrict buffer
The incoming data will be placed into memory at buffer.size_t length
No more thanlength
bytes will be transferred (that's the size of your buffer).int flags
For this tutorial we can ignore this last flags. However this parameter will allow us to “peek” at an incoming message without removing it from the queue or block until the request is fully satisfied. To ignore these flags, simply place a zero in as the parameter. See the man page forrecvfrom
to see an in-depth description of this parameter.struct sockaddr* restrict src_addr
Ifsrc_addr
is notNULL
, the IP address and port number of the sender of the data will be placed into memory atsrc_addr
.socklen_t *src_len
The size of the memory atsrc_addr
. Ifsrc_addr
isNULL
, thensrc_len
should also beNULL
. Example:struct sockaddr_in* cliaddr; socklen_t addrlen; addrlen = sizeof(struct sockaddr_in);
recvfrom
returns the number of bytes received, or -1
if an error occurred.
for (;;) {
printf("waiting for client message on port %d\n", SERV_PORT);
recvlen = recvfrom(sockfd, buf, MSGLEN, 0,
(struct sockaddr *)&cliaddr, &addrlen);
printf("heard %d bytes\n", recvlen);
if (recvlen > 0) {
buf[recvlen] = 0;
printf("I heard this: \"%s\"\n", buf);
}
else
printf("lost the connection to client\n");
}
Now we are able to receive messages from our clients but how do we let clients know their messages are being received? We have no connection to the server and we don’t know their IP address. Fortunately the recvfrom
call gave us the address, and it was placed in remaddr
:
recvlen = recvfrom(sockfd, buf, MSGLEN, 0, (struct sockaddr *)&cliaddr, &addrlen);
The server can use that address in sendto
and send a message back to the recipient’s address.
sendto(sockfd, buf, strlen(buf), 0, (struct sockaddr *)&cliaddr, addrlen)
This step is not necessary for our examples server, however can easily be accomplished with a call to close()
.
close(sockfd);
This concludes the simple UDP server portion of Chapter 1. Section 2 will now cover a Simple UDP Client.
Create a function to send and receive data.
This function will send a message to the server and then loop back. The function does not return anything and takes in 4 objects: the input variable name, a socket, a pointer to a socket address structure, and a length for the address.
Within this function, we will read in user input (fgets)
from the client and loop while this input is valid. This loop will send the input to the server using the sendto()
function. It will then read back the server’s echo with recvfrom()
and print this echo(fputs)
. Our function:
void DatagramClient (FILE* clientInput, WOLFSSL* ssl) {
int n = 0;
char sendLine[MAXLINE], recvLine[MAXLINE - 1];
fgets(sendLine, MAXLINE, clientInput);
if ( ( wolfSSL_write(ssl, sendLine, strlen(sendLine))) !=
strlen(sendLine))
err_sys("SSL_write failed");
n = wolfSSL_read(ssl, recvLine, sizeof(recvLine)-1);
if (n < 0) {
int readErr = wolfSSL_get_error(ssl, 0);
if(readErr != SSL_ERROR_WANT_READ)
err_sys("wolfSSL_read failed");
}
recvLine[n] = '\0';
fputs(recvLine, stdout);
}
This function can be accomplished within main without creating an additional function.
We will begin by adding the following libraries to pull from.
#include <wolfssl/ssl.h>
#include <errno.h>
#include <signal.h>
#include <unistd.h>
Next change the size of our MSGLEN
to 4096 to be more universal. This step is unnecessary if you’re testing against the client.c located in wolfssl/examples/client
as it will only send a message of length 14 or so. With this said, why not be able to handle a little user input if we want to test against a friend's client for example?
We will create a global static int
called cleanup
. This variable will be our signal to run wolfSSL_cleanup();
which will free the wolfSSL libraries and all allocated memory at the end of our program.
Now we declare a WOLFSSL_CTX
pointer and call it ctx
for simplicity.
Your global variable section should now look something like this:
# includes here...
WOLFSSL_CTX* ctx;
static int cleanup; /* To handle shutdown */
struct sockaddr_in servaddr; /* our server's address */
struct sockaddr_in cliaddr; /* the client's address */
Add the variable char ack
inside the main function. This will be a reply message that we send to our clients. Ack is short for “Acknowledge”. So our clients have some sort of feedback letting them know that we received their communication successfully. Section 7 of Chatper 2.1 sees the use of ack
.
We will no longer refer to the open socket as sockfd
, instead we will now call it listenfd
. This reminds us that the socket is just listening for packets to arrive and we are not actually handling those packets immediately like we did in Chapter 1. We will want to confirm our client is encrypting their data with an acceptable cypher suite prior to opening a communication channel with them.
With this change we will also rename addrlen
to clilen
to remind us that this socklen_t
is the length of the clients address and not to be confused with our socket’s address. We will no longer assign the length of clilen
upon declaration either. That will be handled later in our program. Remove msgnum
altogether. Take note: recvlen
is being declared here however is not used until we have verified our client is encrypting with DTLS version 1.2. Section 7 of Chapter 2.1 sees the use of recvlen
. Our variable section should now look something like this:
int on = 1;
int res = 1;
int connfd = 0;
int recvlen = 0; /* length of message */
int listenfd = 0; /* Initialize our socket */
WOLFSSL* ssl = NULL;
socklen_t clilen;
socklen_t len = sizeof(on);
unsigned char b[1500]; /* watch for incoming messages */
char buff[MSGLEN]; /* the incoming message */
char ack[] = "I hear you fashizzle!\n";
With the layering on of DTLS we will need to re-allocate our socket and re-bind our socket for each client connection. Since we will need to free up all memory allocated to handle these connections and additional security our loop will now change to a while
loop instead of a for
loop. We will loop on the condition that cleanup != 1
. If cleanup == 1
we will run wolfSSL_cleanup()
.
So while not 1 we will keep our socket open and continue listening for packets to arrive. Start a while
in your main function. Move the current code from main into the loop. At this point, you should have a loop similar to this inside your main function:
while (cleanup != 1) {
/* create a UDP/IP socket */
if ((sockfd = socket(AF_INET, SOCK_DGRAM, 0)) < 0) {
perror("cannot create socket");
return 0;
}
/* INADDR_ANY = IPaddr, socket = 11111, modify SERV_PORT to change */
memset((char *)&servaddr, 0, sizeof(servaddr));
/* host-to-network-long conversion (htonl) */
/* host-to-network-short conversion (htons) */
servaddr.sin_family = AF_INET;
servaddr.sin_addr.s_addr = htonl(INADDR_ANY);
servaddr.sin_port = htons(SERV_PORT);
if (bind(sockfd, (struct sockaddr *)&servaddr, sizeof(servaddr)) < 0) {
perror("bind failed");
return 0;
}
printf("waiting for client message on port %d\n", SERV_PORT);
recvlen = recvfrom(sockfd, buf, MSGLEN, 0,
(struct sockaddr *)&cliaddr, &addrlen);
printf("heard %d bytes\n", recvlen);
if (recvlen > 0) {
buf[recvlen] = 0;
printf("I heard this: \"%s\"\n", buf);
}
else {
printf("lost the connection to client\n");
}
sprintf(buf, "Message # %d received\n", msgnum++);
printf("reply sent \"%s\"\n", buf);
if (sendto(sockfd, buf, strlen(buf), 0,
(struct sockaddr *)&cliaddr, addrlen) < 0) {
perror("sendto");
}
/* continues to loop, use "Ctrl+C" to terminate listening */
}
This is pretty self-explanatory.
wolfSSL_Debugging_ON();
In order for these to load properly you will need to place a copy of the certs
file one directory above your current working directory. You can find a copy of the certs
file in wolfssl home directory. Simply copy and paste this file into the directory one up from your working directory, or change the file path in the code to search your wolfssl home directory for the certs file.
wolfSSL_Init(); /* Initialize wolfSSL */
/* Set ctx to DTLS 1.2 */
if ( (ctx = wolfSSL_CTX_new(wolfDTLSv1_2_server_method())) == NULL){
fprintf(stderr, "wolfSSL_CTX_new error.\n");
exit(EXIT_FAILURE);
}
/* Load CA certificates */
if (wolfSSL_CTX_load_verify_locations(ctx,"../certs/ca-cert.pem",0) !=
SSL_SUCCESS) {
fprintf(stderr, "Error loading ../certs/ca-cert.pem, "
"please check the file.\n");
exit(EXIT_FAILURE);
}
/* Load server certificates */
if (wolfSSL_CTX_use_certificate_file(ctx,"../certs/server-cert.pem",
SSL_FILETYPE_PEM) != SSL_SUCCESS) {
fprintf(stderr, "Error loading ../certs/server-cert.pem, "
"please check the file.\n");
exit(EXIT_FAILURE);
}
/* Load server Keys */
if (wolfSSL_CTX_use_PrivateKey_file(ctx,"../certs/server-key.pem",
SSL_FILETYPE_PEM) != SSL_SUCCESS) {
fprintf(stderr, "Error loading ../certs/server-key.pem, "
"please check the file.\n");
exit(EXIT_FAILURE);
}
Add a conditional statement that will free up any allocated memory at the termination of our program.
Last, add a return method for main()
.
if (cleanup == 1)
wolfSSL_CTX_free(ctx);
return 0;
So we’ve loaded all the certificates and keys we will need to encrypt any and all communications sent between our server and client. This encryption will be of type DTLS version 1.2 as seen in Figure 2.6, where you see wolfDTLSv1_2_server_method()
. In order for a client to now talk to our DTLS encrypted server they themselves will have to have certificates to verify our encryption, accept our key, and perform a DTLS handshake. See section 2 of this chapter for a tutorial on encrypting a client with DTLS version 1.2.
Our client handling is now running in a while
loop, so it will continue to listen for clients even after a client has communicated with us and the closed their port. Our program will re-allocate that socket, rebind that socket and continue to await the arrival of more datagrams from that same or different clients. Our first step to avoid potential errors with our program will be to avoid “socket already in use” errors. Initialize two dummy integers, res
and on
. set both equal to 1
. Then initialize a struct
of type socklen_t
(same as our clilen
for getting the length of the clients address) call it len
and set it equal to the size of on
. We will use these variables to set the socket options to avoid that error.
(socket)(level)(option_name)(option_value) (option_len)
int setsockopt(int socket, int level, int option_name, const void*option_value, socklen_t option_len);
res = setsockopt(listenfd, SOL_SOCKET, SO_REUSEADDR, &on, len);
The setsockopt()
function sets the option specified by the option_name
argument, at the protocol level specified by the level
argument, to the value pointed to by the option_value
argument for the socket associated with the file descriptor specified by the socket
argument.
The level
argument specifies the protocol level at which the option resides. To set options at the socket level, specify the level argument as SOL_SOCKET
. To set options at other levels, supply the appropriate protocol number for the protocol controlling the option. For example, to indicate that an option will be interpreted by the TCP (Transport Control Protocol), set level to the protocol number of TCP, as defined in the <netinet/in.h>
header.
We will also check to ensure that the method did not fail as if it did it we would think we were listening for clients when in reality our server would not be able to receive any datagrams on that socket and clients would be getting rejected without our knowledge.
int res = 1;
int on = 1;
socklen_t len = sizeof(on);
res = setsockopt(listenfd, SOL_SOCKET, SO_REUSEADDR, &on, len);
if (res < 0) {
printf("Setsockopt SO_REUSEADDR failed.\n");
cleanup = 1;
}
This error avoidance code will be inserted just after we finish our host to network short conversion of SERV_PORT
and just before we bind the socket.
NOTE: We have now changed our error handling from:
perror(“some message”); and return 0;
to:
printf(“some message”); and cleanup = 1;
All active error situations will now change to match this format and all future error situations will be of the same format. This will ensure that all memory is freed at the termination of our program.
Here is where we will now set clilen = sizeof(cliaddr);
as well. We will declare an unsigned char
that will behave as buff
does. Buff
will now be used by ssl
to read client messages and the new buff
(we will just call it b
) is for one purpose only: to “peek” at any incoming messages. We will not actually read those messages until later on. Right now we just want to see if there is a message waiting to be read or not. We must do this on a UDP server because no actual “connection” takes place… perseh. Packets arrive or they don’t - that’s how UDP works. Then we will perform some error handling on the fact that there are Datagrams waiting or there aren’t. If none are waiting we will want our while
loop to cycle and continue to await packet arrivals. If there are packets sitting there then we want to know about it.
/* set clilen to |cliaddr| */
clilen = sizeof(cliaddr); /* will be moved to the variable section later */
unsigned char b[1500]; /* will be moved to the variable section later */
int connfd = 0; /* will be moved to the variable section later */
connfd = (int)recvfrom(listenfd, (char *)&b, sizeof(b), MSG_PEEK,
(struct sockaddr*)&cliaddr, &clilen);
if (connfd < 0){
printf("No clients in que, enter idle state\n");
continue;
}
else if (connfd > 0) {
if (connect(listenfd, (const struct sockaddr *)&cliaddr,
sizeof(cliaddr)) != 0) {
printf("Udp connect failed.\n");
cleanup = 1;
}
}
else {
printf("Recvfrom failed.\n");
cleanup = 1;
}
printf("Connected!\n");
We say “Connected” for user friendly interpretation, but what we really mean is: “Datagrams have arrived, a client is attempting to communicate with us… let’s perform a handshake and see if they are using DTLS version 1.2… if so, we’ll read their Datagrams and see what they have to say”.
First we must declare an object that points to a WOLFSSL
structure. We will just call it ssl
. We will then assign it to use the correct cypher suite as previously defined by ctx
. We will again perform some error handling on this assignment and then set the file descriptor that will handle all incoming and outgoing messages for this session.
Once all that has been set up we are ready to check and see if our client is using an acceptable cypher suite. We accomplish this by making a call to wolfSSL_accept
on our ssl
object that is now pointing to the file descriptor that has an awaiting Datagram in it. (That's a lot - we know). We’ll use some fancy calls to error get methods so that if this part fails we will have a little bit of an idea as to why it failed and how to fix it.
/* initialize arg */
WOLFSSL* ssl;
/* Create the WOLFSSL Object */
if (( ssl = wolfSSL_new(ctx) ) == NULL) {
printf("wolfSSL_new error.\n");
cleanup = 1;
}
/* set the session ssl to client connection port */
wolfSSL_set_fd(ssl, listenfd);
if (wolfSSL_accept(ssl) != SSL_SUCCESS) {
int err = wolfSSL_get_error(ssl, 0);
char buffer[80];
printf("error = %d, %s\n", err, wolfSSL_ERR_error_string(err, buffer));
buffer[sizeof(buffer)-1]= 0;
printf("SSL_accept failed.\n");
cleanup = 1;
}
So our client is using DTLS version 1.2, we have a message waiting for us… so let’s read it! Then let’s send our client an acknowledgment that we have received their message. Don’t forget to handle any potential errors along the way!
if (( recvlen = wolfSSL_read(ssl, buff, sizeof(buff)-1)) > 0){
printf("heard %d bytes\n", recvlen);
buff[recvlen - 1] = 0;
printf("I heard this: \"%s\"\n", buff);
}
if (recvlen < 0) {
int readErr = wolfSSL_get_error(ssl, 0);
if(readErr != SSL_ERROR_WANT_READ) {
printf("SSL_read failed.\n");
cleanup = 1;
}
}
if (wolfSSL_write(ssl, ack, sizeof(ack)) < 0) {
printf("wolfSSL_write fail.\n");
cleanup = 1;
}
else
printf("lost the connection to client\n");
printf("reply sent \"%s\"\n", ack);
Finally we need to free up all memory that was allocated on our server for this particular session. If the client so chooses they can connect with us again using the same session key. But we do not want to bog down our server with unnecessary messages from previous sessions, once those messages have been handled. So we reset the file descriptor to 0 allocated bits, and shutdown our active ssl
object and free the memory. Lastly we print out a line letting us know that we have reached the end of our loop successfully, all memory is free, and we are returning to the top to listen for the next client that might want to talk to us.
wolfSSL_set_fd(ssl, 0);
wolfSSL_shutdown(ssl);
wolfSSL_free(ssl);
printf("Client left return to idle state\n");
This concludes Section 1 of Chapter 2 on “Layering DTLS onto a UDP Server”. Section 2 will now cover Layering DTLS onto a UDP Client.
As stated in chapter 4 of the wolfSSL manual, DTLS is enabled by using the --enable-dtls
build option when building wolfSSL. If you have not done so, this should be your first step.
Walk through chapter 11 in the wolfSSL tutorial. Follow the guides for TLS/SSL using the correct wolfDTLS client method. There are a few adjustments to be made for DTLS, detailed in the next few steps.
Make sure you have the correct port defined in your program. For example,
# define SERV_PORT 11111
for server with 11111
set as the port.
Edit the arguments in your send and receive function to just 2 arguments: a FILE*
object and a WOLFSSL*
object (previously this function had 4 arguments).
Change sendto
and recvfrom
functions to wolfSSL_write
and wolfSSL_read
.
Delete the last 3 arguments that were in sendto
and recvfrom
. In the wolfSSL_read()
call, change the first argument from a socket to the WOLFSSL*
object from the original function call.
Make a call to the wolfSSL_dtls_set_peer()
function. It will take in as arguments your WOLFSSL*
object, a pointer to the address carrying your sockaddr_in
structure, and size
of the structure. Example:
wolfSSL_dtls_set_peer(ssl, &servAddr, sizeof(servAddr));
This function will be called between where you built the sockaddr_in
structure and your socket creation.
Add a wolfSSL connect function below the call to wolfSSL_set_fd()
and pass the WOLFSSL*
object you created as the argument. Include error checking. Example:
if (wolfSSL_connect(ssl) != SSL_SUCCESS) {
int err1 = wolfSSL_get_error(ssl, 0);
char buffer[80];
printf("err = %d, %s\n", err1,yaSSL_ERR_error_string(err1, buffer));
err_sys("SSL_connect failed");
}
Call your wolfSSL_write
/wolfSSL_read
functions. Example:
wolfSSL_read(ssl, buff, sizeof(buff)-1);
wolfSSL_write(ssl, ack, sizeof(ack));
Make calls to wolfSSL_shutdown()
, wolfSSL_free()
, wolfSSL_CTX_free()
, and wolfSSL_Cleanup()
with correct parameters in each. This can be done in the read/write function or at the end of the main method.
Include -DWOLFSSL_DTLS
before -o
in your compilation line. This will include the DTLS method you chose.
To make a DTLS server multithreaded, the most effective way to do this would be to use POSIX threads.
From Wikipedia's POSIX Threads page:
POSIX Threads ... allows a program to control multiple different flows of work that overlap in time. Each flow of work is referred to as a thread, and creation and control over these flows is achieved by making calls to the POSIX Threads API.
In order to multithread a DTLS server, three things will need to be completed.
- Elements required for thread management should be added
- Creation of a
ThreadHandler
method - Cleanup
(a quick note: be sure to have gone over signal handling and have code for signal handling in your main method prior to attempting multithread your server. Information on signal handling can be found in section 11.12 of the SSL tutorial)
The main elements of thread management using pthread
are
- Data given to each thread created (usually in the form of a struct)
- Thread creation and deletion method calls (usually
pthread_create
,pthread_join
,pthread_detach
, etc.) - A thread ID
- A ThreadHandler method
The following sections go over each of the items listed above.
The threadArgs
struct is needed to so that we can simultaneously pass multiple pieces of information to our pthread
being created. It will allow us to pass in a file descriptor, message, size, and whatever else we can think of - but for now it will just contain those three things. Below is the definition of the threadArgs
struct.
typedef struct {
int activefd;
char b[MSGLEN];
int size;
} threadArgs;
The actual creation of each threadArgs*
object will be done at the beginning of each iteration of the while loop located within AwaitDGram
. Inside of that loop, we will create the object, and also dynamically allocate it with malloc
so that it is accessible to the thread and us from anywhere.
while (cleanup != 1) {
threadArgs* args;
args = (threadArgs *) malloc(sizeof(threadArgs));
.
.
.
}
After a client has connected to the server, we need to create a thread and "spin" it to handle receiving and sending messages (spinning a thread is just spinning it off of the current process and giving it a method to take care of itself). For thread creation, there are two things to do - create a thread ID, and call pthread_create
.
From the man page of pthread_create
,
NAME
pthread_create - create a new thread
SYNOPSIS
#include <pthread.h>
int pthread_create(pthread_t *thread, const pthread_attr_t *attr,
void *(*start_routine) (void *), void *arg);
When we call pthread_create
, it should be done like this:
if (cleanup != 1) {
pthread_t threadid;
pthread_create(&threadid, NULL, ThreadHandler, args);
}
note ThreadHandler
being passed as the 3rd argument - this is a method we will define
To control what the thread does when it is deleted/its resources are freed, we will have to define a ThreadHandler
method.
When writing a method to handle thread execution, imagine that every thread is its own server being executed, and that the thread handler method is the main
method of each of these servers. But it isn't entirely the same as a main method - it takes a void pointer as input, and requires some special features to keep itself in check. A common and simple outline of what a ThreadHandler
should include is listed below.
- A
pthread_detach( pthread_self() )
call - Instance variables (file descriptors, message length, reply,
WOLFSSL*
object) - A
memcpy
call ssl
object creationwolfSSL_accept
,wolfSSL_read
andwolfSSL_write
calls with error checking for all three- Memory cleanup (
wolfSSL_shutdown
,wolfSSL_free
, etc.)
As we go over items 1 through 6, we'll continue adding meat to our ThreadHandler
method.
Each commented section will be filled in as each of the following sections is completed.
Calling pthread_detach
should be the very first thing done within the method - it specifies that when the thread terminates, its resources will be released. The parameter passed to pthread_detach
should be the thread itself.
Our ThreadHandler
code now looks like this:
void* ThreadHandler(void *input)
{
/* pthread_detach call */
pthread_detach( pthread_self() );
/* Instance variables */
/* memcpy call */
/* miscellaneous standard server calls */
/* Thread memory cleanup */
}
In total, there are 7 local variables used inside of ThreadHandler
. They are all listed below with a brief description of their purposes.
threadArgs* args = (threadArgs*) input;
int recvLen = 0; /* Length of message received */
int activefd = args->activefd; /* active file descriptor */
int msgLen = args->size; /* Length of message */
unsigned char buff[msgLen]; /* Incoming message from client */
char back[] = "MESSAGE RECEIVED" /* Message sent to client */
WOLFSSL* ssl; /* SSL object */
Putting these variables into our code gives us a new ThreadHandler
method:
void* ThreadHandler(void *input)
{
/* pthread_detach call */
pthread_detach( pthread_self() );
/* Instance variables */
threadArgs* args = (threadArgs*) input;
int recvLen = 0; /* Length of message received */
int activefd = args->activefd; /* active file descriptor */
int msgLen = args->size; /* Length of message */
unsigned char buff[msgLen]; /* Incoming message from client */
char back[] = "MESSAGE RECEIVED" /* Message sent to client */
WOLFSSL* ssl; /* SSL object */
/* memcpy call */
/* miscellaneous standard server calls */
/* Thread memory cleanup */
}
Calling memcpy
allows the user to clear out the previous data located in args->b
. The method should be called like the example is below.
memcpy(buff, args->b, msgLen);
Putting this call into our code gives this ThreadHandler
method:
void* ThreadHandler(void *input)
{
/* pthread_detach call */
pthread_detach( pthread_self() );
/* Instance variables */
threadArgs* args = (threadArgs*) input;
int recvLen = 0; /* Length of message received */
int activefd = args->activefd; /* active file descriptor */
int msgLen = args->size; /* Length of message */
unsigned char buff[msgLen]; /* Incoming message from client */
char back[] = "MESSAGE RECEIVED" /* Message sent to client */
WOLFSSL* ssl; /* SSL object */
/* memcpy call */
memcpy(buff, args->b, msgLen);
/* miscellaneous standard server calls */
/* Thread memory cleanup */
}
There shouldn't be anything special about the creation of the ssl
object. It should be identical to a normal creation, but should be assigned inside of ThreadHandler
. The purpose of creating a WOLFSSL
object in ThreadHandler
is that each thread will maintain a DTLS connection to a client.
void* ThreadHandler(void *input)
{
/* pthread_detach call */
pthread_detach( pthread_self() );
/* Instance variables */
threadArgs* args = (threadArgs*) input;
int recvLen = 0; /* Length of message received */
int activefd = args->activefd; /* active file descriptor */
int msgLen = args->size; /* Length of message */
unsigned char buff[msgLen]; /* Incoming message from client */
char back[] = "MESSAGE RECEIVED" /* Message sent to client */
WOLFSSL* ssl; /* SSL object */
/* memcpy call */
memcpy(buff, args->b, msgLen);
/* miscellaneous standard server calls */
/* Create the WOLFSSL Object */
if ((ssl = wolfSSL_new(ctx)) == NULL) {
printf("wolfSSL_new error.\n");
cleanup = 1;
return NULL;
}
/* Thread memory cleanup */
}
For the same reason that each thread will create a WOLFSSL
object, each thread will also make calls to wolfSSL_accept
, wolfSSL_read
, and wolfSSL_write
. The calls should be almost identical to the regular calls inside of server-dtls.c
, but with minor changes.
Important: don't forget to set the file descriptor as well
continue
keywords should be changed toreturn NULL
, asThreadHandler
does not loop- If
wolfSSL_write
returns a negative number inThreadHandler
, there should be areturn NULL
instead ofreturn 1
(ThreadHandler
is invoked differently fromAwaitDGram
.
After making these changes and adding them to ThreadHandler
, it now looks like this:
void* ThreadHandler(void *input)
{
/* pthread_detach call */
pthread_detach( pthread_self() );
/* Instance variables */
threadArgs* args = (threadArgs*) input;
int recvLen = 0; /* Length of message received */
int activefd = args->activefd; /* active file descriptor */
int msgLen = args->size; /* Length of message */
unsigned char buff[msgLen]; /* Incoming message from client */
char back[] = "MESSAGE RECEIVED" /* Message sent to client */
WOLFSSL* ssl; /* SSL object */
/* memcpy call */
memcpy(buff, args->b, msgLen);
/* miscellaneous standard server calls */
/* Create the WOLFSSL Object */
if ((ssl = wolfSSL_new(ctx)) == NULL) {
printf("wolfSSL_new error.\n");
cleanup = 1;
return NULL;
}
wolfSSL_set_fd(ssl, activefd);
/* Attempt to accept connection */
if (wolfSSL_accept(ssl) != SSL_SUCCESS) {
int e = wolfSSL_get_error(ssl, 0);
printf("error = %d, %s\n", e, wolfSSL_ERR_reason_error_string(e));
printf("SSL_accept failed.\n");
return NULL;
}
/* Attempt to read a message from the client, check for errors twice */
if ((recvLen = wolfSSL_read(ssl, buff, msgLen-1)) > 0) {
printf("heard %d bytes\n", recvLen);
buff[recvLen] = 0;
printf("I heard this: \"%s\"\n", buff);
}
else if (recvLen < 0) {
int readErr = wolfSSL_get_error(ssl, 0);
if(readErr != SSL_ERROR_WANT_READ) {
printf("SSL_read failed.\n");
cleanup = 1;
return NULL;
}
}
/* Send reply to the client */
if (wolfSSL_write(ssl, ack, sizeof(ack)) < 0) {
printf("wolfSSL_write fail.\n");
cleanup = 1;
return NULL;
}
else {
printf("Sending reply.\n");
}
printf("reply sent \"%s\"\n", ack);
/* Thread memory cleanup */
}
To clean up the mess left behind when the client and server are finished with their business, we'll need to free/close the objects created at the beginning of the ThreadHandler
method, and call a few subtle thread memory management calls. Below is a list of items that need to be freed, and our finished ThreadHandler
code is immediately following.
ssl
activefd
input
- The current thread
void* ThreadHandler(void *input)
{
/* pthread_detach call */
pthread_detach( pthread_self() );
/* Instance variables */
threadArgs* args = (threadArgs*) input;
int recvLen = 0; /* Length of message received */
int activefd = args->activefd; /* active file descriptor */
int msgLen = args->size; /* Length of message */
unsigned char buff[msgLen]; /* Incoming message from client */
char back[] = "MESSAGE RECEIVED" /* Message sent to client */
WOLFSSL* ssl; /* SSL object */
/* memcpy call */
memcpy(buff, args->b, msgLen);
/* miscellaneous standard server calls */
/* Create the WOLFSSL Object */
if ((ssl = wolfSSL_new(ctx)) == NULL) {
printf("wolfSSL_new error.\n");
cleanup = 1;
return NULL;
}
wolfSSL_set_fd(ssl, activefd);
/* Attempt to accept connection */
if (wolfSSL_accept(ssl) != SSL_SUCCESS) {
int e = wolfSSL_get_error(ssl, 0);
printf("error = %d, %s\n", e, wolfSSL_ERR_reason_error_string(e));
printf("SSL_accept failed.\n");
return NULL;
}
/* Attempt to read a message from the client, check for errors twice */
if ((recvLen = wolfSSL_read(ssl, buff, msgLen-1)) > 0) {
printf("heard %d bytes\n", recvLen);
buff[recvLen] = 0;
printf("I heard this: \"%s\"\n", buff);
}
else if (recvLen < 0) {
int readErr = wolfSSL_get_error(ssl, 0);
if(readErr != SSL_ERROR_WANT_READ) {
printf("SSL_read failed.\n");
cleanup = 1;
return NULL;
}
}
/* Send reply to the client */
if (wolfSSL_write(ssl, ack, sizeof(ack)) < 0) {
printf("wolfSSL_write fail.\n");
cleanup = 1;
return NULL;
}
else {
printf("Sending reply.\n");
}
printf("reply sent \"%s\"\n", ack);
printf("Client left, return to idle state.\n"
"Cleaning up & exiting thread.\n");
/* Thread memory cleanup */
wolfSSL_shutdown(ssl); /* cleans up the ssl object */
wolfSSL_free(ssl);
close(activefd); /* closes activefd (who'd have thought?) */
free(input); /* cleans up the input object */
pthread_exit(input); /* terminates the thread that it is called in */
}
And that's it - that is all that is required to create a multithreaded DTLS server.
There were several things added to a DTLS server to enable threading. Below is a concise list of all that was added and done to make a threaded DTLS server.
- Defined
threadArgs
struct with data about the current client and session - Calling
pthread_create
and passing in client/session data with a definedThreadHandler
method. - A method (ours was called
ThreadHandler
) to tell the threads how to behaveThreadHandler
used its ownWOLFSSL
objects and terminated itself at the end of the method
In the event that a connection between a client and a server is lost, it can be useful to resume with the previous session data. To implement this, all the work can be done on the client's side, since the client will be the one with direct access to their session info before a disconnect.
To utilize session resumption, 4 things have to be completed.
- Storage of the previous session with a new WOLFSSL* object created with the same method call as the previous WOLFSSL* object
- Memory cleanup and management of the previously interrupted session
- Reconnect with the old session data
- Cleanup
(a quick note: to allow for infinite session resumptions, simply perform these actions inside of a loop.)
When keeping track of previous sessions for resumption, the simplest and easiest way to do this is to create new objects and assign the old session data to them. An example of this is shown below.
session = wolfSSL_get_session(ssl);
sslResume = wolfSSL_new(ctx);
The two variables above are the most important for session resumption.
session
is an object of typeWOLFSSL_SESSION*
sslResume
is an object of typeWOLFSSL*
Now it's time to clean up and clear out all the old objects. To clean up the memory, this can be done in standard fashion with calls to the functions wolfSSL_shutdown
and wolfSSL_free
- and don't forget to call close
on the socket file descriptor.
An example of this is shown below.
wolfSSL_shutdown(ssl);
wolfSSL_free(ssl);
close(sockfd);
In addition, it is important to clear the memory holding the server address and reassign it. An example of this is shown below.
memset(&servAddr, 0, sizeof(servAddr));
servAddr.sin_family = AF_INET;
servAddr.sin_port = htons(SERV_PORT);
if ( (inet_pton(AF_INET, host, &servAddr.sin_addr)) < 1) {
printf("Error and/or invalid IP address");
return 1;
}
The reconnection is nearly identical to the way that a client will connect to a server the first time. There are calls to wolfSSL_dtls_set_peer
, socket
, wolfSSL_set_fd
, and so on, but there are also calls to other methods.
All the method calls needed with new variables and the distinct new method calls are listed below in the order that they need to be called.
wolfSSL_dtls_set_peer(sslResume, &servAddr, sizeof(servAddr))
- This method call uses the old session data (
sslResume
) and with the same server address. It needs to be called after thememset
call listed in section 2 (Memory cleanup and management).
socket(AF_INET, SOCK_DGRAM, 0)
will need to be in an identical conditional as the prior call tosocket
.wolfSSL_set_fd(sslResume, sockfd)
- This method call uses the old session data (
sslResume
) and with the same socket file descriptor.
wolfSSL_set_session(sslResume, session)
- This is a new method call. On failure, the return value is (you guessed it) an
SSL_FAILURE
code. On success, it will return (you could try out for jeopardy!) anSSL_SUCCESS
code.
wolfSSL_connect(sslResume)
will need to be called in an identical conditional as the prior call towolfSSL_connect
.wolfSSL_session_reused(sslResume)
- This method call returns the
resuming
value from inside thesslResume
object. - This method call is used with a conditional to tell the user whether or not a session was reused. An example is shown below.
if (wolfSSL_session_reused(sslResume)) {
printf("Reused session ID\n");
} else {
printf("Did not reuse session ID\n");
}
There isn't anything special about cleanup with reused sessions. All that needs to be done is to call wolfSSL_shutdown
and wolfssl_free
with the sslResume
object.
Again, if wanted, this can all be isolated inside a loop that will break upon receiving a signal from the user. To find out more about signal handling, visit the wolfSSL SSL Tutorial, specifically section 11.12.
If you compare this tutorial and the functions we discuss implementing to the available completed example code, you will notice these functions are missing from the example code. The functions have been combined into main()
in order to allow the code to read linearly and be more understandable. For the purposes of this tutorial, you may still add the noted functions to achieve the same functionality. All of the functions are simply hidden within the main function. Most are commented with their original function names when they occur. This tutorial is in the process of being updated to match this more appropriately.
# include <errno.h> /* error checking */
# include <fcntl.h> /* set file status flags */
In both the server and client files, add the following before the first function:
enum {
TEST_SELECT_FAIL,
TEST_TIMEOUT,
TEST_RECV_READY,
TEST_ERROR_READY
};
From Quora,
Many IO related system calls, like read(2), will block, that is not return, until there is activity. A socket can be placed in "non-blocking mode" which means that these system calls will return immediately if there is nothing pending.
To make our client a nonblocking client, we need to add 2 things (they should come before we attempt to send a datagram to the server, as it would be much more difficult to send a datagram without having a connection!).
1. Variables
2. A loop that iterates until an SSL_SUCCESS
or error
- The loop will also have conditional statements inside that check the progress of the nonblocking code - they will be mainly used to determine the outcome of the loop and keep the user updated.
The only part of the functionality that is truly nonblocking is the selection of sockets. It will not wait until there is activity on the server side to return - it will return after a specified amount of time regardless of if it can connect or not.
Before adding anything else, there are 2 method calls to be added. In the main()
method, add the following chunk immediately after the call to wolfSSL_set_fd(ssl, sockfd)
:
wolfSSL_dtls_set_using_nonblock(ssl, 1);
fcntl(sockfd, F_SETFL, O_NONBLOCK);
Almost all the variables that need to be added are int
types. There are also fd_set
and timeval
variables. Instantiation at declaration is not necessary, as most of the variables will be reset at every iteration of the loop in mentioned above in 2. In addition, some variables are also dependent upon other variables and their data in the main()
method. Figure 5...? shows the variable declarations. They should be declared along with all other variables inside of the main()
method.
Figure 5.1
/* Variables used for nonblocking DTLS connect */
int ret;
int error;
int nb_sockfd;
int select_ret;
int currTimeout;
/* Variables used for nonblocking DTLS select */
int nfds;
int result;
fd_set recvfds, errfds;
struct timeval timeout;
Three of the variables used for nonblocking DTLS connect will be assigned prior to the loop. Figure 5.2 shows where they are assigned.
Figure 5.2
fcntl(sockfd, F_SETFL, O_NONBLOCK);
/*****************************************************************************/
/* Connect using Nonblocking - DTLS version */
ret = wolfSSL_connect(ssl); /* Connection test (should fail) */
error = wolfSSL_get_error(ssl, 0); /* Find out why connect failed */
nb_sockfd = (int) wolfSSL_get_fd(ssl); /* To protect the original socket */
1 variable from nonblocking DTLS connect and 2 from nonblocking DTLS select will be assigned at the beginning of each iteration.
2 variables from nonblocking DTLS select will be assigned in the middle of the loop.
The variable assignments will be shown when the loop is discussed in Section 5.1.4.2 next.
The body of the nonblocking DTLS connection and functionality occur within a while loop that continues to loop while ret
has not succeeded in a connection, and while error
is at least an SSL_ERROR_WANT_READ
or an SSL_ERROR_WANT_WRITE
.
Figure 5.3
/*****************************************************************************/
/* Connect using Nonblocking - DTLS version */
ret = wolfSSL_connect(ssl); /* Connection test (should fail) */
error = wolfSSL_get_error(ssl, 0); /* Find out why connect failed */
nb_sockfd = (int) wolfSSL_get_fd(ssl); /* To protect the original socket */
while (ret != SSL_SUCCESS &&
(error = SSL_ERROR_WANT_READ || error == SSL_ERROR_WANT_WRITE)) {
/* Variables that reset every iteration */
currTimeout = wolfSSL_dtls_get_current_timeout(ssl);
nfds = nb_sockfd + 1;
timeout = (struct timeval) { (currTimeout > 0) ? currTimeout : 0, 0};
/* Keep the user updated */
if (error == SSL_ERROR_WANT_READ) {
printf("... client would read block.\n");
} else {
printf("... client would write block.\n");
}
/* nonblocking DTLS select variables that reset every iteration *
* (recvfds, nb_sockfd and errfds must be manually reset) */
FD_ZERO(&recvfds);
FD_SET(nb_sockfd, &recvfds);
FD_ZERO(&errfds);
FD_SET(nb_sockfd, &errfds);
result = select(nfds, &recvfds, NULL, &errfds, &timeout);
select_ret = TEST_SELECT_FAIL;
if (result == 0) {
select_ret = TEST_TIMEOUT;
}
else if (result > 0) {
if (FD_ISSET(nb_sockfd, &recvfds)) {
select_ret = TEST_RECV_READY;
}
else if (FD_ISSET(nb_sockfd, &errfds)) {
select_ret = TEST_ERROR_READY;
}
}
/* End nonblocking DTLS selection functionality */
/* Determine the outcome of nonblocking selection */
if ( select_ret == TEST_RECV_READY ||
select_ret == TEST_ERROR_READY ) {
ret = wolfSSL_connect(ssl);
error = wolfSSL_get_error(ssl, 0);
}
else if (select_ret == TEST_TIMEOUT && !wolfSSL_dtls(ssl)) {
error = 2;
}
else if (select_ret == TEST_TIMEOUT && wolfSSL_dtls(ssl) &&
wolfSSL_dtls_got_timeout(ssl) >= 0) {
error = 2;
}
else {
error = SSL_FATAL_ERROR;
}
}
/* Confirm that the loop exited properly */
if (ret != SSL_SUCCESS) {
printf("SSL_connect failed with %d\n", ret);
}
/* */
/*****************************************************************************/
Now the client needs to send a message to the server. If it didn't, then there would have been no point in connecting to the server.
The way that a datagram from a nonblocking DTLS client is sent is identical to the way that a datagram is sent from a regular DTLS client. Remember, what makes a DTLS client nonblocking is how it connects to a server, not how it sends its message.
The datagram sending code is included below for reference in Figure 5.4. It requires no special variable usage or extra conditionals/loops in a nonblocking client.
Figure 5.4
/* Code for sending a datagram to the server */
int n = 0;
char sendLine[MAXLINE], recvLine[MAXLINE - 1];
while (fgets(sendLine, MAXLINE, stdin) != NULL) {
while ( ( wolfSSL_write(ssl, sendLine, strlen(sendLine))) !=
strlen(sendLine)) {
printf("SSL_write failed");
}
while ( (n = wolfSSL_read(ssl, recvLine, sizeof(recvLine)-1)) <= 0) {
int readErr = wolfSSL_get_error(ssl, 0);
if(readErr != SSL_ERROR_WANT_READ) {
printf("wolfSSL_read failed");
}
}
recvLine[n] = '\0';
fputs(recvLine, stdout);
}
#include <errno.h> /* error checking */
#include <fcntl.h> /* set file status flags */
In both the server and client files, add the following before the first function:
enum {
TEST_SELECT_FAIL,
TEST_TIMEOUT,
TEST_RECV_READY,
TEST_ERROR_READY
};
In the server file, it can be useful to get a user's input that tells the server to shut down. This is done with handling signals. To do this, we will need to have a static variable cleanup
that tells the server when the signal has been passed, a method that will modify cleanup
, and code that states this method is what needs to be executed when the server is passed a signal.
This variable cleanup
is just a global static integer. It does not need to be initialized at declaration. Be sure to include in your while loop that you wish to loop until the original condition is true or cleanup
is equal to 1.
To let the program know in an simple way that a signal has passed, all that needs to be done is writing a method that will reassign cleanup
.
The method needs to take in a const int
type, and if desired, you can also add a print statement that tells the user what has happened. An example of a sig_handler
method is shown below in Figure 5.5.
Figure 5.5
/* For handling ^C interrupts or other signals */
void sig_handler(const int sig)
{
printf("\nSIGINT %d handled\n", sig);
cleanup = 1;
return;
}
If we just leave the server with the sig_handler
definition and it looping until cleanup is changed, nothing will actually change when a user passes a signal to the server. To fix this, add the code in Figure 5.6 immediately after creating the variables in main()
.
Figure 5.6
/*Code for handling signals */
struct sigaction act, oact;
act.sa_handler = sig_handler;
sigemptyset(&act.sa_mask);
act.sa_flags = 0;
sigaction(SIGINT, &act, &oact);
To make our server a nonblocking server, we need to add 2 things. They are the same as the two things we had to add for our client - variables and a while loop.
1. Variables
2. A while loop that loops while cleanup is not 1 and the original conditions are true
Don't forget that immediately after the call to wolfSSL_set_fd()
, we need to make a call wolfSSL_dtls_set_using_nonblock()
. The call to this method is shown below.
wolfSSL_dtls_set_using_nonblock(ssl, 1);
As with our client, there are several variables needed for our nonblocking DTLS server. Figure 5.7 lists them.
Figure 5.7
/* DTLS set nonblocking flag */
int flags = fcntl(*(&listenfd), F_GETFL, 0);
/* NonBlockingSSL_Accept variables */
int ret;
int select_ret;
int currTimeout;
int error;
int result;
int nfds;
fd_set recvfds, errfds;
struct timeval timeout;
/* udp-read-connect variables */
int bytesRecvd;
unsigned char b[MSGLEN];
struct sockaddr_in cliAddr;
socklen_t clilen;
As with the client, the while loop has variables that are assigned prior to, at the beginning of, and in the center of the loop.
For brevity, they will not all be pointed out and explained - the code will be displayed in Figure 5.8, and all that is new will have descriptive comments that explain the differences.
Figure 5.8
/*****************************************************************************/
/* AwaitDatagram code */
/* This code will loop until a ^C (SIGINT 2) is passed by the user */
cont = 0;
while (cleanup != 1) {
clilen = sizeof(cliAddr);
timeout.tv_sec = (currTimeout > 0) ? currTimeout : 0;
/* Create a UDP/IP socket */
if ((listenfd = socket(AF_INET, SOCK_DGRAM, 0)) < 0 ) {
printf("Cannot create socket.\n");
cont = 1;
}
printf("Socket allocated\n");
/* DTLS set nonblocking */
if (flags < 0) {
printf("fcntl get failed");
cleanup = 1;
}
flags = fcntl(*(&listenfd), F_SETFL, flags | O_NONBLOCK);
if (flags < 0) {
printf("fcntl set failed.\n");
cleanup = 1;
}
/* Clear servAddr each loop */
memset((char *)&servAddr, 0, sizeof(servAddr));
/* host-to-network-long conversion (htonl) */
/* host-to-network-short conversion (htons) */
servAddr.sin_family = AF_INET;
servAddr.sin_addr.s_addr = htonl(INADDR_ANY);
servAddr.sin_port = htons(SERV_PORT);
/* Eliminate socket already in use error */
res = setsockopt(listenfd, SOL_SOCKET, SO_REUSEADDR, &on, len);
if (res < 0) {
printf("Setsockopt SO_REUSEADDR failed.\n");
cont = 1;
}
/*Bind Socket*/
if (bind(listenfd,
(struct sockaddr *)&servAddr, sizeof(servAddr)) < 0) {
printf("Bind failed.\n");
cont = 1;
}
printf("Awaiting client connection on port %d\n", SERV_PORT);
/* UDP-read-connect */
do {
if (cleanup == 1) {
cont = 1;
break;
}
bytesRecvd = (int)recvfrom(listenfd, (char*)b, sizeof(b), MSG_PEEK,
(struct sockaddr*)&cliAddr, &clilen);
} while (bytesRecvd <= 0);
if (bytesRecvd > 0) {
if (connect(listenfd, (const struct sockaddr*)&cliAddr,
sizeof(cliAddr)) != 0) {
printf("udp connect failed.\n");
}
}
else {
printf("recvfrom failed.\n");
}
printf("Connected!\n");
/* ensure b is empty upon each call */
memset(&b, 0, sizeof(b));
clientfd = listenfd;
/* Create the WOLFSSL Object */
if (( ssl = wolfSSL_new(ctx)) == NULL) {
printf("wolfSSL_new error.\n");
cont = 1;
}
/* set clilen to |cliAddr| */
printf("Connected!\n");
/* set the/ session ssl to client connection port */
wolfSSL_set_fd(ssl, clientfd);
wolfSSL_dtls_set_using_nonblock(ssl, 1);
/*****************************************************************************/
/* NonBlockingDTLS_Connect code */
/* listenfd states where to listen ()
ret = wolfSSL_accept(ssl);
currTimeout = 1;
error = wolfSSL_get_error(ssl, 0);
listenfd = (int) wolfSSL_get_fd(ssl);
nfds = listenfd + 1;
/* Loop until there has been a successful connection *
* or until there has been a signal */
while (cleanup != 1 && (ret != SSL_SUCCESS &&
(error == SSL_ERROR_WANT_READ ||
error == SSL_ERROR_WANT_WRITE))) {
if (cleanup == 1) {
wolfSSL_free(ssl);
wolfSSL_shutdown(ssl);
break;
}
/* Keep the user updated */
if (error == SSL_ERROR_WANT_READ)
printf("... server would read block\n");
else
printf("... server would write block\n");
currTimeout = wolfSSL_dtls_get_current_timeout(ssl);
FD_ZERO(&recvfds);
FD_SET(listenfd, &recvfds);
FD_ZERO(&errfds);
FD_SET(listenfd, &errfds);
/* This is where the term 'nonblocking' comes into use */
result = select(nfds, &recvfds, NULL, &errfds, &timeout);
if (result == 0) {
select_ret = TEST_TIMEOUT;
}
else if (result > 0) {
if (FD_ISSET(listenfd, &recvfds)) {
select_ret = TEST_RECV_READY;
}
else if (FD_ISSET(listenfd, &errfds)) {
select_ret = TEST_ERROR_READY;
}
}
else {
select_ret = TEST_SELECT_FAIL;
}
if ((select_ret == TEST_RECV_READY) ||
(select_ret == TEST_ERROR_READY)) {
ret = wolfSSL_accept(ssl);
error = wolfSSL_get_error(ssl, 0);
}
else if (select_ret == TEST_TIMEOUT && !wolfSSL_dtls(ssl)) {
error = SSL_ERROR_WANT_READ;
}
else if (select_ret == TEST_TIMEOUT && wolfSSL_dtls(ssl) &&
wolfSSL_dtls_got_timeout(ssl) >= 0) {
error = SSL_ERROR_WANT_READ;
}
else {
error = SSL_FATAL_ERROR;
}
}
if (ret != SSL_SUCCESS) {
printf("SSL_accept failed with %d.\n", ret);
cont = 1;
}
else {
cont = 0;
}
if (cont != 0) {
printf("NonBlockingSSL_Accept failed.\n");
cont = 1;
}
/* end NonBlockingDTLS_Connect code */
/*****************************************************************************/
/* Begin: Reply to the client */
recvLen = wolfSSL_read(ssl, buff, sizeof(buff)-1);
/* Begin do-while read */
do {
if (cleanup == 1) {
memset(buff, 0, sizeof(buff));
break;
}
if (recvLen < 0) {
readWriteErr = wolfSSL_get_error(ssl, 0);
if (readWriteErr != SSL_ERROR_WANT_READ) {
printf("Read Error, error was: %d.\n", readWriteErr);
cleanup = 1;
}
else {
recvLen = wolfSSL_read(ssl, buff, sizeof(buff)-1);
}
}
} while (readWriteErr == SSL_ERROR_WANT_READ &&
recvLen < 0 &&
cleanup != 1);
/* End do-while read */
if (recvLen > 0) {
buff[recvLen] = 0;
printf("I heard this:\"%s\"\n", buff);
}
else {
printf("Connection Timed Out.\n");
}
/* Begin do-while write */
do {
if (cleanup == 1) {
memset(&buff, 0, sizeof(buff));
break;
}
readWriteErr = wolfSSL_get_error(ssl, 0);
if (wolfSSL_write(ssl, ack, sizeof(ack)) < 0) {
printf("Write error.\n");
cleanup = 1;
}
printf("Reply sent:\"%s\"\n", ack);
} while(readWriteErr == SSL_ERROR_WANT_WRITE && cleanup != 1);
/* End do-while write */
/* free allocated memory */
memset(buff, 0, sizeof(buff));
wolfSSL_free(ssl);
/* End: Reply to the Client */
}
/* End await datagram code */
/*****************************************************************************/
/* End of the main method */
if (cont == 1 || cleanup == 1) {
wolfSSL_CTX_free(ctx);
wolfSSL_Cleanup();
}
return 0;
The code above was taken directly from the DTLS server nonblocking file.
Be sure to keep in mind that the AwaitDatagram
code is essentially one large loop that will attempt to listen for a client (in a nonblocking fashion) at every iteration, and will close the loop upon a signal passed by the user.
And that's it! The server has been made into a nonblocking server, and the client has been made into a nonblocking client.
- Paul Krzyzanowski, “Programming with UDP sockets”, Copyright 2003-2014, PK.ORG
- The Open Group, “setsockopt - set the socket options”, Copyright © 1997, The Single UNIX ® Specification, Version 2
- https://en.wikipedia.org/wiki/POSIX_Threads
- https://www.quora.com/What-exactly-does-it-mean-for-a-web-server-to-be-blocking-versus-non-blocking