Fossa is a multi-protocol networking library written in C. It provides easy to use event-driven interface that allows to implement network protocols or scalable network applications with little effort. Fossa releives developers from the burden of network programming complexity and let them concentrate on the logic, saving time and money.
Fossa has built-in support for several protocols, like HTTP and Websocket, and is ideal for embedded environments. Fossa has been designed as an open source platform for connecting devices and bringing them online.
-
Cross-platform: works on Linux/UNIX, QNX, eCos, Windows, Android, iPhone, etc
-
Single-threaded, asynchronous, non-blocking core with simple event-based API
-
Builtin protocols:
-
plain TCP, plain UDP, SSL/TLS (over TCP, one-way or two-way)
-
HTTP client, HTTP server
-
Websocket client, Websocket server
-
JSON-RPC client, JSON-RPC server
-
-
Tiny static and run-time footprint
-
Source code is both ISO C and ISO C++ compliant
-
Extensively tested and production-ready, trusted by many blue chip businesses
Fossa is a non-blocking, asyncronous event manager described by
struct ns_mgr
structure. That structure holds active connections.
Connections could be either listening, client or accepted.
Client connections are created by
ns_connect()
call. Listening connections are created by ns_bind()
call.
Accepted connections are those that incoming on a listening connection.
Each connection is described by struct ns_connection
structure, which has
a number of fields like socket, event handler function, send/receive buffer,
flags, et cetera.
ns_mgr_poll()
should be called in an infinite event loop.
ns_mgr_poll()
iterates over all sockets, accepts new connections,
sends and receives data, closes connections, and calls an event handler
function for each of those events.
Each connection has send and receive buffer, struct ns_connection::send_iobuf
and struct ns_connection::recv_iobuf
respectively. When data arrives,
Fossa appends received data to the recv_iobuf
and
triggers NS_RECV
event. User may send data back (ns_send()
or
ns_printf()
), which appends data to the send_iobuf
. When Fossa
successfully writes data to the socket, it discards it from send_iobuf
and
sends NS_SEND
event. When connection is closed, NS_CLOSE
event is sent.
-
Define an event handler function.
-
Initialize mgr by calling
ns_mgr_init()
. -
Create listening connections with
ns_bind()
and/or client connections withns_connect()
. Note that many connections can be created within a single manager. Connections can be created at any time, including within an event handler function. -
Call
ns_mgr_poll()
in a loop.
#include "fossa.h"
// This event handler implements TCP echo server
static void ev_handler(struct ns_connection *nc, int ev, void *ev_data) { // 1
struct iobuf *io = &nc->recv_iobuf;
switch (ev) {
case NS_RECV:
ns_send(nc, io->buf, io->len); // Echo received data back
iobuf_remove(io, io->len); // Discard data from recv buffer
break;
default:
break;
}
}
int main(void) {
struct ns_mgr mgr;
ns_mgr_init(&mgr, NULL); // 2
ns_bind(&mgr, "1234", ev_handler, NULL); // 3
// 4 - an event loop
for (;;) {
ns_mgr_poll(&mgr, 1000);
}
ns_mgr_free(&mgr);
return 0;
}
Fossa accepts incoming connections, reads and writes data, and
calls specified event handler for each connection when appropriate. An
event handler should examine received data, set connection flags if needed,
and send data back to the client by ns_send()
or ns_printf()
. Here is a
typical event flow for the accepted connection:
NS_ACCEPT
→ NS_RECV
→ …. → NS_CLOSE
. Below is a complete list
of events triggered by Fossa:
- NS_ACCEPT
-
sent when new server connection is accepted by a listening connection.
void *ev_data
isunion socket_address
of the remote peer. - NS_CONNECT
-
sent when a new client connection created by
ns_connect()
either failed or succeeded.void *ev_data
isint *success
. Ifsuccess
is 0 then connection has been established, otherwise it contains error code. Example code to check connection status:
static void ev_handler(struct ns_connection *nc, int ev, void *ev_data) {
int connect_status;
switch (ev) {
case NS_CONNECT:
connect_status = * (int *) ev_data;
if (connect_status == 0) {
/* Success */
} else {
/* Error */
printf("connect() error: %s\n", strerror(connect_status));
}
break;
...
- NS_RECV
-
New data is received and appended to the end of
recv_iobuf
.void *ev_data
isint *num_received_bytes
.
Warning
|
Fossa uses realloc() to expand receive buffer.
It is user’s responsibility to discard processed
data from the beginning of receive buffer, note the iobuf_remove()
call in the example above.
|
- NS_SEND
-
Fossa has written data to the remote peer and discarded written data from the
send_iobuf
.void *ev_data
isint *num_sent_bytes
- NS_POLL
-
Sent to all connections on each invocation of
ns_server_poll()
An event handler can set struct ns_connection::flags
attribute to control
the behavior of the connection. Below is a list of connection flags:
-
NSF_FINISHED_SENDING_DATA
tells Fossa that all data has been appended to thesend_iobuf
. As soon as Fossa sends it to the socket, the connection will be closed. -
NSF_BUFFER_BUT_DONT_SEND
tells Fossa to append data to thesend_iobuf
but hold on sending it, because the data will be modified later and then will be sent by clearingNSF_BUFFER_BUT_DONT_SEND
flag. -
NSF_SSL_HANDSHAKE_DONE
SSL only, set when SSL handshake is done -
NSF_CONNECTING
set when connection is in connecting state afterns_connect()
call but connect did not finish yet -
NSF_CLOSE_IMMEDIATELY
tells Fossa to close the connection immediately, usually after some error -
NSF_LISTENING
set for all listening connections -
NSF_UDP
set if connection is UDP -
NSF_IS_WEBSOCKET
set by Fossa if connection is a Websocket connection -
NSF_WEBSOCKET_NO_DEFRAG
should be set by a user if user wants to switch off automatic frame defragmentation -
NSF_USER_1
,NSF_USER_2
,NSF_USER_3
,NSF_USER_4
could be used by a developer to store application-specific state
Caution
|
Fossa manager instance is single threaded. It does not protect
it’s data structures by mutexes, therefore all functions that are dealing
with particular event manager should be called from the same thread,
with exception of mg_broadcast() function. It is fine to have different
event managers handled by different threads.
|
-
struct ns_connection
Describes a connection between two peers -
struct ns_mgr
Container for a bunch of connections -
struct iobuf
Describes piece of data
- void ns_mgr_init(struct ns_mgr *, void *user_data)
-
Initializes Fossa manager.
- void ns_mgr_free(struct ns_mgr *)
-
De-initializes skeleton manager, closes and deallocates all active connections.
- time_t ns_mgr_poll(struct ns_mgr *, int milliseconds)
-
This function performs the actual IO, and must be called in a loop (an event loop). Returns number current timestamp.
- void ns_broadcast(struct ns_mgr *, ns_event_handler_t cb, void *msg, size_t len)
-
Must be called from a different thread. Passes a message of a given length to all connections. Skeleton manager has a socketpair,
struct ns_mgr::ctl
, wherens_broadcast()
pushes the message.ns_mgr_poll()
wakes up, reads a message from the socket pair, and calls specified callback for each connection. Thus the callback function executes in event manager thread. Note thatns_broadcast()
is the only function that can be, and must be, called from a different thread. - void ns_next(struct ns_mgr *, struct ns_connection *)
-
Iterates over all active connections. Returns next connection from the list of active connections, or
NULL
if there is no more connections. Below is the iteration idiom:
for (c = ns_next(srv, NULL); c != NULL; c = ns_next(srv, c)) {
// Do something with connection `c`
}
- struct ns_connection *ns_add_sock(struct ns_mgr *, sock_t sock, ns_event_handler_t ev_handler)
-
Create a connection, associate it with the given socket and event handler, and add to the manager.
- struct ns_connection *ns_connect(struct ns_mgr *server, const char *addr, ns_event_handler_t ev_handler)
-
Connect to a remote host. If successful,
NS_CONNECT
event will be delivered to the new connection.addr
format is the same as for thens_bind()
call, just an IP address becomes mandatory:[PROTO://]HOST:PORT
PROTO
could betcp://
orudp://
. IfHOST
is not an IP address, Fossa will resolve it - beware that standard blocking resolver will be used. It is a good practice to pre-resolve hosts beforehands and use only IP addresses to avoid blockin an IO thread. Returns: new client connection, orNULL
on error. - struct ns_connection *ns_bind(struct ns_mgr *, const char *addr, ns_event_handler_t ev_handler)
-
Start listening on the given port.
addr
could be a port number, e.g."3128"
, or IP address with a port number, e.g."127.0.0.1:3128"
. Also, a protocol prefix could be specified, valid prefixes aretcp://
orudp://
.
Note that for UDP listening connections, only NS_RECV
and NS_CLOSE
are triggered.
If IP address is specified, Fossa binds to a specific interface only.
Also, port could be "0"
, in which case a random non-occupied port number
will be chosen. Return value: a listening connection on success, or
NULL
on error.
- const char *ns_set_ssl(struct ns_connection *nc, const char *cert, const char *ca_cert)
-
Enable SSL for a given connection. Connection must be TCP. For listening connection,
cert
is a path to a server certificate, and is mandatory.ca_cert
if non-NULL, specifies CA certificate for client authentication, enables two-way SSL. For client connections, bothcert
andca_cert
are optional and can be set to NULL. All certificates must be in PEM format. PEM file for server certificate should contain both certificate and the private key concatenated together. Returns: NULL if there is no error, or error string if there was error.
Snippet below shows how to generate self-signed SSL certificate using OpenSSL:
openssl req -x509 -nodes -newkey rsa:2048 -keyout key.pem -out cert.pem -days 365
cat cert.pem key.pem > my_ssl_cert.pem
- int ns_send(struct ns_connection *, const void *buf, int len)
- int ns_printf(struct ns_connection *, const char *fmt, …)
- int ns_vprintf(struct ns_connection *, const char *fmt, va_list ap)
-
These functions are for sending un-formatted and formatted data to the connection. Number of written bytes is returned. Note that these sending functions do not actually push data to the sockets, they just append data to the output buffer. The exception is UDP connections. For UDP, data is sent immediately, and returned value indicates an actual number of bytes sent to the socket.
- void *ns_start_thread(void *(*thread_function)(void *), void *param)
-
Starts a new thread
- int ns_socketpair2(sock_t [2], int proto)
-
Create a socket pair.
proto
can be eitherSOCK_STREAM
orSOCK_DGRAM
. Return 0 on failure, 1 on success. - void ns_set_close_on_exec(sock_t)
-
Set close-on-exec bit for a given socket.
- void ns_sock_to_str(sock_t sock, char *buf, size_t len, int flags)
-
Converts socket’s local or remote address into string.
flags
parameter is a bit mask that controls the behavior. If bit 2 is set (flags & 4
) then the remote address is stringified, otherwise local address is stringified. If bit 0 is set, then IP address is printed. If bit 1 is set, then port number is printed. If both port number and IP address are printed, they are separated by:
. - int ns_hexdump(const void *buf, int len, char *dst, int dst_len)
-
Takes a memory buffer
buf
of lengthlen
and creates a hex dump of that buffer indst
. - int ns_resolve(const char *domain_name, char *ip_addr_buf, size_t buf_len)
-
Converts domain name into IP address. This is a blocking call. Returns 1 on success, 0 on failure.
- int ns_stat(const char *path, ns_stat_t *st)
-
Perform a 64-bit
stat()
call against given file.path
should be UTF8 encoded. Return value is the same as forstat()
syscall. - FILE *ns_fopen(const char *path, const char *mode)
-
Open given file and return a file stream.
path
andmode
should be UTF8 encoded. Return value is the same as forfopen()
call. - int ns_open(const char *path, int flag, int mode)
-
Open given file and return file descriptor.
path
should be UTF8 encoded. Return value is the same as foropen()
syscall.
- void ns_set_protocol_http_websocket(struct ns_connection *)
-
Attach built-in HTTP event handler to the given connection. User-defined event handler will receive following extra events:
-
NS_HTTP_REQUEST: HTTP request has arrived. Parsed HTTP request is passed as
struct http_message
through the handler’svoid *ev_data
pointer. -
NS_HTTP_REPLY: HTTP reply has arrived. Parsed HTTP reply is passed as
struct http_message
through the handler’svoid *ev_data
pointer. -
NS_WEBSOCKET_HANDSHAKE_REQUEST: server has received websocket handshake request.
ev_data
contains parsed HTTP request. -
NS_WEBSOCKET_HANDSHAKE_DONE: server has completed Websocket handshake.
ev_data
isNULL
. -
NS_WEBSOCKET_FRAME: new websocket frame has arrived.
ev_data
isstruct websocket_message *
-
- void ns_send_websocket_handshake(struct ns_connection *nc, const char *uri, const char *extra_headers)
-
Sends websocket handshake to the server.
nc
must be a valid connection, connected to a server,uri
is an URI on the server,extra_headers
is extra HTTP headers to send orNULL
. This function is to be used by websocket client. - void ns_send_websocket_frame(struct ns_connection *nc, int op, const void *data, size_t data_len)
-
Send websocket frame to the remote end.
op
specifies frame’s type , one of:-
WEBSOCKET_OP_CONTINUE
-
WEBSOCKET_OP_TEXT
-
WEBSOCKET_OP_BINARY
-
WEBSOCKET_OP_CLOSE
-
WEBSOCKET_OP_PING
-
WEBSOCKET_OP_PONG
data
anddata_len
contain frame data.
-
void ns_send_websocket_framev(struct ns_connection *nc, int op, const struct ns_str *frames, int num_frames);
Send multiple websocket frames. Like ns_send_websocket_frame()
, but sends
multiple frames at once.
- void ns_printf_websocket_frame(struct ns_connection *nc, int op, const char *fmt, …)
-
Send websocket frame to the remote end. Like
ns_send_websocket_frame()
, but allows to create formatted message withprintf()
-like semantics. - struct ns_str *ns_get_http_header(struct http_message *, const char *)
-
Returns HTTP header if it is present in the HTTP message, or
NULL
. - int ns_parse_http(const char *s, int n, struct http_message *req)
-
Parses HTTP message. Return number of bytes parsed. If HTTP message is incomplete,
0
is returned. On parse error, negative number is returned. - int ns_get_http_var(const struct ns_str *buf, const char *name, char *dst, size_t dst_len)
-
Fetch an HTTP form variable
name
from abuf
into a buffer specified bydst
,dst_len
. Destination is always zero-terminated. Return length of a fetched variable. If not found, 0 is returned.buf
must be valid url-encoded buffer. If destination is too small,-1
is returned. - void ns_serve_http(struct ns_connection *nc, struct http_message *request, struct ns_serve_http_opts options)
-
Serve given HTTP request according to the
options
. Example code snippet:
static void ev_handler(struct ns_connection *nc, int ev, void *ev_data) {
struct http_message *hm = (struct http_message *) ev_data;
struct ns_serve_http_opts opts = { .document_root = "/var/www" }; // C99 syntax
switch (ev) {
case NS_HTTP_REQUEST:
ns_serve_http(nc, hm, opts);
break;
default:
break;
}
}
JSON-RPC module is implemented using Frozen JSON parser/generator. So for JSON-related functionality refer to Frozen documentation.
- int ns_rpc_parse_reply(const char *buf, int len, struct json_token *toks, int max_toks, struct ns_rpc_reply *reply, struct ns_rpc_error *error)
-
Parse JSON-RPC reply contained in
buf
,len
into JSON tokens arraytoks
,max_toks
. If buffer contains valid reply,reply
structure is populated. The result of RPC call is located inreply.result
. On error,error
structure is populated. Returns: the result of callingparse_json(buf, len, toks, max_toks)
. - int ns_rpc_create_request(char *buf, int len, const char *method, const char *id, const char *params_fmt, …)
-
Create JSON-RPC request in a given buffer. Return length of the request, which can be larger then
len
that indicates an overflow. - int ns_rpc_create_reply(char *buf, int len, const struct ns_rpc_request *req, const char *result_fmt, …)
-
Create JSON-RPC reply in a given buffer. Return length of the reply, which can be larger then
len
that indicates an overflow. - int ns_rpc_create_error(char *, int, struct ns_rpc_request *req, int, const char *, const char *, …)
-
Create JSON-RPC error in a given buffer. Return length of the error, which can be larger then
len
that indicates an overflow. - int ns_rpc_create_std_error(char *, int, struct ns_rpc_request *, int code)
-
Create JSON-RPC error in a given buffer. Return length of the error, which can be larger then
len
that indicates an overflow.code
could be one of:JSON_RPC_PARSE_ERROR
,JSON_RPC_INVALID_REQUEST_ERROR
,JSON_RPC_METHOD_NOT_FOUND_ERROR
,JSON_RPC_INVALID_PARAMS_ERROR
,JSON_RPC_INTERNAL_ERROR
,JSON_RPC_SERVER_ERROR
. - int ns_rpc_dispatch(const char *buf, int, char *dst, int dst_len, const char **methods, ns_rpc_handler_t *handlers)
-
Parses JSON-RPC request contained in
buf
,len
. Then, dispatches the request to the correct handler method. Valid method names should be specified in NULL terminated arraymethods
, and corresponding handlers inhandlers
. Result is put indst
,dst_len
. Return: length of the result, which can be larger thendst_len
that indicates an overflow.
-
examples/echo_server: a simple TCP echo server. It accepts incoming connections and echoes back any data that it receives
-
examples/publish_subscribe: implements pubsub pattern for TCP communication
-
examples/netcat: an implementation of Netcat utility with traffic hexdump and SSL support
Fossa is released under GNU GPL v.2. Businesses have an option to get non-restrictive, royalty-free commercial license and professional support from Cesanta Software.