- Fix the stupid mess we created when making this from the cmake branch
[reactos.git] / lib / drivers / lwip / doc / rawapi.txt
1 Raw TCP/IP interface for lwIP
3 Authors: Adam Dunkels, Leon Woestenberg, Christiaan Simons
5 lwIP provides three Application Program's Interfaces (APIs) for programs
6 to use for communication with the TCP/IP code:
7 * low-level "core" / "callback" or "raw" API.
8 * higher-level "sequential" API.
9 * BSD-style socket API.
11 The sequential API provides a way for ordinary, sequential, programs
12 to use the lwIP stack. It is quite similar to the BSD socket API. The
13 model of execution is based on the blocking open-read-write-close
14 paradigm. Since the TCP/IP stack is event based by nature, the TCP/IP
15 code and the application program must reside in different execution
16 contexts (threads).
18 The socket API is a compatibility API for existing applications,
19 currently it is built on top of the sequential API. It is meant to
20 provide all functions needed to run socket API applications running
21 on other platforms (e.g. unix / windows etc.). However, due to limitations
22 in the specification of this API, there might be incompatibilities
23 that require small modifications of existing programs.
25 ** Threading
27 lwIP started targeting single-threaded environments. When adding multi-
28 threading support, instead of making the core thread-safe, another
29 approach was chosen: there is one main thread running the lwIP core
30 (also known as the "tcpip_thread"). The raw API may only be used from
31 this thread! Application threads using the sequential- or socket API
32 communicate with this main thread through message passing.
34 As such, the list of functions that may be called from
35 other threads or an ISR is very limited! Only functions
36 from these API header files are thread-safe:
37 - api.h
38 - netbuf.h
39 - netdb.h
40 - netifapi.h
41 - sockets.h
42 - sys.h
44 Additionaly, memory (de-)allocation functions may be
45 called from multiple threads (not ISR!) with NO_SYS=0
46 since they are protected by SYS_LIGHTWEIGHT_PROT and/or
47 semaphores.
49 Only since 1.3.0, if SYS_LIGHTWEIGHT_PROT is set to 1
51 pbuf_free() may also be called from another thread or
52 an ISR (since only then, mem_free - for PBUF_RAM - may
53 be called from an ISR: otherwise, the HEAP is only
54 protected by semaphores).
57 ** The remainder of this document discusses the "raw" API. **
59 The raw TCP/IP interface allows the application program to integrate
60 better with the TCP/IP code. Program execution is event based by
61 having callback functions being called from within the TCP/IP
62 code. The TCP/IP code and the application program both run in the same
63 thread. The sequential API has a much higher overhead and is not very
64 well suited for small systems since it forces a multithreaded paradigm
65 on the application.
67 The raw TCP/IP interface is not only faster in terms of code execution
68 time but is also less memory intensive. The drawback is that program
69 development is somewhat harder and application programs written for
70 the raw TCP/IP interface are more difficult to understand. Still, this
71 is the preferred way of writing applications that should be small in
72 code size and memory usage.
74 Both APIs can be used simultaneously by different application
75 programs. In fact, the sequential API is implemented as an application
76 program using the raw TCP/IP interface.
78 --- Callbacks
80 Program execution is driven by callbacks. Each callback is an ordinary
81 C function that is called from within the TCP/IP code. Every callback
82 function is passed the current TCP or UDP connection state as an
83 argument. Also, in order to be able to keep program specific state,
84 the callback functions are called with a program specified argument
85 that is independent of the TCP/IP state.
87 The function for setting the application connection state is:
89 - void tcp_arg(struct tcp_pcb *pcb, void *arg)
91 Specifies the program specific state that should be passed to all
92 other callback functions. The "pcb" argument is the current TCP
93 connection control block, and the "arg" argument is the argument
94 that will be passed to the callbacks.
97 --- TCP connection setup
99 The functions used for setting up connections is similar to that of
100 the sequential API and of the BSD socket API. A new TCP connection
101 identifier (i.e., a protocol control block - PCB) is created with the
102 tcp_new() function. This PCB can then be either set to listen for new
103 incoming connections or be explicitly connected to another host.
105 - struct tcp_pcb *tcp_new(void)
107 Creates a new connection identifier (PCB). If memory is not
108 available for creating the new pcb, NULL is returned.
110 - err_t tcp_bind(struct tcp_pcb *pcb, struct ip_addr *ipaddr,
111 u16_t port)
113 Binds the pcb to a local IP address and port number. The IP address
114 can be specified as IP_ADDR_ANY in order to bind the connection to
115 all local IP addresses.
117 If another connection is bound to the same port, the function will
118 return ERR_USE, otherwise ERR_OK is returned.
120 - struct tcp_pcb *tcp_listen(struct tcp_pcb *pcb)
122 Commands a pcb to start listening for incoming connections. When an
123 incoming connection is accepted, the function specified with the
124 tcp_accept() function will be called. The pcb will have to be bound
125 to a local port with the tcp_bind() function.
127 The tcp_listen() function returns a new connection identifier, and
128 the one passed as an argument to the function will be
129 deallocated. The reason for this behavior is that less memory is
130 needed for a connection that is listening, so tcp_listen() will
131 reclaim the memory needed for the original connection and allocate a
132 new smaller memory block for the listening connection.
134 tcp_listen() may return NULL if no memory was available for the
135 listening connection. If so, the memory associated with the pcb
136 passed as an argument to tcp_listen() will not be deallocated.
138 - struct tcp_pcb *tcp_listen_with_backlog(struct tcp_pcb *pcb, u8_t backlog)
140 Same as tcp_listen, but limits the number of outstanding connections
141 in the listen queue to the value specified by the backlog argument.
142 To use it, your need to set TCP_LISTEN_BACKLOG=1 in your lwipopts.h.
144 - void tcp_accepted(struct tcp_pcb *pcb)
146 Inform lwIP that an incoming connection has been accepted. This would
147 usually be called from the accept callback. This allows lwIP to perform
148 housekeeping tasks, such as allowing further incoming connections to be
149 queued in the listen backlog.
151 - void tcp_accept(struct tcp_pcb *pcb,
152 err_t (* accept)(void *arg, struct tcp_pcb *newpcb,
153 err_t err))
155 Specified the callback function that should be called when a new
156 connection arrives on a listening connection.
158 - err_t tcp_connect(struct tcp_pcb *pcb, struct ip_addr *ipaddr,
159 u16_t port, err_t (* connected)(void *arg,
160 struct tcp_pcb *tpcb,
161 err_t err));
163 Sets up the pcb to connect to the remote host and sends the
164 initial SYN segment which opens the connection.
166 The tcp_connect() function returns immediately; it does not wait for
167 the connection to be properly setup. Instead, it will call the
168 function specified as the fourth argument (the "connected" argument)
169 when the connection is established. If the connection could not be
170 properly established, either because the other host refused the
171 connection or because the other host didn't answer, the "err"
172 callback function of this pcb (registered with tcp_err, see below)
173 will be called.
175 The tcp_connect() function can return ERR_MEM if no memory is
176 available for enqueueing the SYN segment. If the SYN indeed was
177 enqueued successfully, the tcp_connect() function returns ERR_OK.
180 --- Sending TCP data
182 TCP data is sent by enqueueing the data with a call to
183 tcp_write(). When the data is successfully transmitted to the remote
184 host, the application will be notified with a call to a specified
185 callback function.
187 - err_t tcp_write(struct tcp_pcb *pcb, void *dataptr, u16_t len,
188 u8_t copy)
190 Enqueues the data pointed to by the argument dataptr. The length of
191 the data is passed as the len parameter. The copy argument is either
192 0 or 1 and indicates whether the new memory should be allocated for
193 the data to be copied into. If the argument is 0, no new memory
194 should be allocated and the data should only be referenced by
195 pointer.
197 The tcp_write() function will fail and return ERR_MEM if the length
198 of the data exceeds the current send buffer size or if the length of
199 the queue of outgoing segment is larger than the upper limit defined
200 in lwipopts.h. The number of bytes available in the output queue can
201 be retrieved with the tcp_sndbuf() function.
203 The proper way to use this function is to call the function with at
204 most tcp_sndbuf() bytes of data. If the function returns ERR_MEM,
205 the application should wait until some of the currently enqueued
206 data has been successfully received by the other host and try again.
208 - void tcp_sent(struct tcp_pcb *pcb,
209 err_t (* sent)(void *arg, struct tcp_pcb *tpcb,
210 u16_t len))
212 Specifies the callback function that should be called when data has
213 successfully been received (i.e., acknowledged) by the remote
214 host. The len argument passed to the callback function gives the
215 amount bytes that was acknowledged by the last acknowledgment.
218 --- Receiving TCP data
220 TCP data reception is callback based - an application specified
221 callback function is called when new data arrives. When the
222 application has taken the data, it has to call the tcp_recved()
223 function to indicate that TCP can advertise increase the receive
224 window.
226 - void tcp_recv(struct tcp_pcb *pcb,
227 err_t (* recv)(void *arg, struct tcp_pcb *tpcb,
228 struct pbuf *p, err_t err))
230 Sets the callback function that will be called when new data
231 arrives. The callback function will be passed a NULL pbuf to
232 indicate that the remote host has closed the connection. If
233 there are no errors and the callback function is to return
234 ERR_OK, then it must free the pbuf. Otherwise, it must not
235 free the pbuf so that lwIP core code can store it.
237 - void tcp_recved(struct tcp_pcb *pcb, u16_t len)
239 Must be called when the application has received the data. The len
240 argument indicates the length of the received data.
243 --- Application polling
245 When a connection is idle (i.e., no data is either transmitted or
246 received), lwIP will repeatedly poll the application by calling a
247 specified callback function. This can be used either as a watchdog
248 timer for killing connections that have stayed idle for too long, or
249 as a method of waiting for memory to become available. For instance,
250 if a call to tcp_write() has failed because memory wasn't available,
251 the application may use the polling functionality to call tcp_write()
252 again when the connection has been idle for a while.
254 - void tcp_poll(struct tcp_pcb *pcb, u8_t interval,
255 err_t (* poll)(void *arg, struct tcp_pcb *tpcb))
257 Specifies the polling interval and the callback function that should
258 be called to poll the application. The interval is specified in
259 number of TCP coarse grained timer shots, which typically occurs
260 twice a second. An interval of 10 means that the application would
261 be polled every 5 seconds.
264 --- Closing and aborting connections
266 - err_t tcp_close(struct tcp_pcb *pcb)
268 Closes the connection. The function may return ERR_MEM if no memory
269 was available for closing the connection. If so, the application
270 should wait and try again either by using the acknowledgment
271 callback or the polling functionality. If the close succeeds, the
272 function returns ERR_OK.
274 The pcb is deallocated by the TCP code after a call to tcp_close().
276 - void tcp_abort(struct tcp_pcb *pcb)
278 Aborts the connection by sending a RST (reset) segment to the remote
279 host. The pcb is deallocated. This function never fails.
281 ATTENTION: When calling this from one of the TCP callbacks, make
282 sure you always return ERR_ABRT (and never return ERR_ABRT otherwise
283 or you will risk accessing deallocated memory or memory leaks!
286 If a connection is aborted because of an error, the application is
287 alerted of this event by the err callback. Errors that might abort a
288 connection are when there is a shortage of memory. The callback
289 function to be called is set using the tcp_err() function.
291 - void tcp_err(struct tcp_pcb *pcb, void (* err)(void *arg,
292 err_t err))
294 The error callback function does not get the pcb passed to it as a
295 parameter since the pcb may already have been deallocated.
298 --- Lower layer TCP interface
300 TCP provides a simple interface to the lower layers of the
301 system. During system initialization, the function tcp_init() has
302 to be called before any other TCP function is called. When the system
303 is running, the two timer functions tcp_fasttmr() and tcp_slowtmr()
304 must be called with regular intervals. The tcp_fasttmr() should be
305 called every TCP_FAST_INTERVAL milliseconds (defined in tcp.h) and
306 tcp_slowtmr() should be called every TCP_SLOW_INTERVAL milliseconds.
309 --- UDP interface
311 The UDP interface is similar to that of TCP, but due to the lower
312 level of complexity of UDP, the interface is significantly simpler.
314 - struct udp_pcb *udp_new(void)
316 Creates a new UDP pcb which can be used for UDP communication. The
317 pcb is not active until it has either been bound to a local address
318 or connected to a remote address.
320 - void udp_remove(struct udp_pcb *pcb)
322 Removes and deallocates the pcb.
324 - err_t udp_bind(struct udp_pcb *pcb, struct ip_addr *ipaddr,
325 u16_t port)
327 Binds the pcb to a local address. The IP-address argument "ipaddr"
328 can be IP_ADDR_ANY to indicate that it should listen to any local IP
329 address. The function currently always return ERR_OK.
331 - err_t udp_connect(struct udp_pcb *pcb, struct ip_addr *ipaddr,
332 u16_t port)
334 Sets the remote end of the pcb. This function does not generate any
335 network traffic, but only set the remote address of the pcb.
337 - err_t udp_disconnect(struct udp_pcb *pcb)
339 Remove the remote end of the pcb. This function does not generate
340 any network traffic, but only removes the remote address of the pcb.
342 - err_t udp_send(struct udp_pcb *pcb, struct pbuf *p)
344 Sends the pbuf p. The pbuf is not deallocated.
346 - void udp_recv(struct udp_pcb *pcb,
347 void (* recv)(void *arg, struct udp_pcb *upcb,
348 struct pbuf *p,
349 struct ip_addr *addr,
350 u16_t port),
351 void *recv_arg)
353 Specifies a callback function that should be called when a UDP
354 datagram is received.
357 --- System initalization
359 A truly complete and generic sequence for initializing the lwip stack
360 cannot be given because it depends on the build configuration (lwipopts.h)
361 and additional initializations for your runtime environment (e.g. timers).
363 We can give you some idea on how to proceed when using the raw API.
364 We assume a configuration using a single Ethernet netif and the
365 UDP and TCP transport layers, IPv4 and the DHCP client.
367 Call these functions in the order of appearance:
369 - stats_init()
371 Clears the structure where runtime statistics are gathered.
373 - sys_init()
375 Not of much use since we set the NO_SYS 1 option in lwipopts.h,
376 to be called for easy configuration changes.
378 - mem_init()
380 Initializes the dynamic memory heap defined by MEM_SIZE.
382 - memp_init()
384 Initializes the memory pools defined by MEMP_NUM_x.
386 - pbuf_init()
388 Initializes the pbuf memory pool defined by PBUF_POOL_SIZE.
390 - etharp_init()
392 Initializes the ARP table and queue.
393 Note: you must call etharp_tmr at a ARP_TMR_INTERVAL (5 seconds) regular interval
394 after this initialization.
396 - ip_init()
398 Doesn't do much, it should be called to handle future changes.
400 - udp_init()
402 Clears the UDP PCB list.
404 - tcp_init()
406 Clears the TCP PCB list and clears some internal TCP timers.
407 Note: you must call tcp_fasttmr() and tcp_slowtmr() at the
408 predefined regular intervals after this initialization.
410 - netif_add(struct netif *netif, struct ip_addr *ipaddr,
411 struct ip_addr *netmask, struct ip_addr *gw,
412 void *state, err_t (* init)(struct netif *netif),
413 err_t (* input)(struct pbuf *p, struct netif *netif))
415 Adds your network interface to the netif_list. Allocate a struct
416 netif and pass a pointer to this structure as the first argument.
417 Give pointers to cleared ip_addr structures when using DHCP,
418 or fill them with sane numbers otherwise. The state pointer may be NULL.
420 The init function pointer must point to a initialization function for
421 your ethernet netif interface. The following code illustrates it's use.
423 err_t netif_if_init(struct netif *netif)
424 {
425 u8_t i;
427 for(i = 0; i < ETHARP_HWADDR_LEN; i++) netif->hwaddr[i] = some_eth_addr[i];
428 init_my_eth_device();
429 return ERR_OK;
430 }
432 For ethernet drivers, the input function pointer must point to the lwip
433 function ethernet_input() declared in "netif/etharp.h". Other drivers
434 must use ip_input() declared in "lwip/ip.h".
436 - netif_set_default(struct netif *netif)
438 Registers the default network interface.
440 - netif_set_up(struct netif *netif)
442 When the netif is fully configured this function must be called.
444 - dhcp_start(struct netif *netif)
446 Creates a new DHCP client for this interface on the first call.
447 Note: you must call dhcp_fine_tmr() and dhcp_coarse_tmr() at
448 the predefined regular intervals after starting the client.
450 You can peek in the netif->dhcp struct for the actual DHCP status.
453 --- Optimalization hints
455 The first thing you want to optimize is the lwip_standard_checksum()
456 routine from src/core/inet.c. You can override this standard
457 function with the #define LWIP_CHKSUM <your_checksum_routine>.
459 There are C examples given in inet.c or you might want to
460 craft an assembly function for this. RFC1071 is a good
461 introduction to this subject.
463 Other significant improvements can be made by supplying
464 assembly or inline replacements for htons() and htonl()
465 if you're using a little-endian architecture.
467 #define LWIP_PLATFORM_HTONS(x) <your_htons>
468 #define LWIP_PLATFORM_HTONL(x) <your_htonl>
470 Check your network interface driver if it reads at
471 a higher speed than the maximum wire-speed. If the
472 hardware isn't serviced frequently and fast enough
473 buffer overflows are likely to occur.
475 E.g. when using the cs8900 driver, call cs8900if_service(ethif)
476 as frequently as possible. When using an RTOS let the cs8900 interrupt
477 wake a high priority task that services your driver using a binary
478 semaphore or event flag. Some drivers might allow additional tuning
479 to match your application and network.
481 For a production release it is recommended to set LWIP_STATS to 0.
482 Note that speed performance isn't influenced much by simply setting
483 high values to the memory options.
485 For more optimization hints take a look at the lwIP wiki.
487 --- Zero-copy MACs
489 To achieve zero-copy on transmit, the data passed to the raw API must
490 remain unchanged until sent. Because the send- (or write-)functions return
491 when the packets have been enqueued for sending, data must be kept stable
492 after that, too.
494 This implies that PBUF_RAM/PBUF_POOL pbufs passed to raw-API send functions
495 must *not* be reused by the application unless their ref-count is 1.
497 For no-copy pbufs (PBUF_ROM/PBUF_REF), data must be kept unchanged, too,
498 but the stack/driver will/must copy PBUF_REF'ed data when enqueueing, while
499 PBUF_ROM-pbufs are just enqueued (as ROM-data is expected to never change).
501 Also, data passed to tcp_write without the copy-flag must not be changed!
503 Therefore, be careful which type of PBUF you use and if you copy TCP data
504 or not!