2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
10 /* rx_user.c contains routines specific to the user space UNIX implementation of rx */
12 /* This controls the size of an fd_set; it must be defined early before
13 * the system headers define that type and the macros that operate on it.
14 * Its value should be as large as the maximum file descriptor limit we
15 * are likely to run into on any platform. Right now, that is 65536
16 * which is the default hard fd limit on Solaris 9 */
18 #define FD_SETSIZE 65536
21 #include <afsconfig.h>
22 #include <afs/param.h>
29 #include "rx_atomic.h"
30 #include "rx_globals.h"
33 #define MAXTHREADNAMELENGTH 64
35 int debugSelectFailure; /* # of times select failed */
38 * Sleep on the unique wait channel provided.
43 LWP_WaitProcess(addr);
47 * Wakeup any threads on the channel provided.
48 * They may be woken up spuriously, and must check any conditions.
51 rxi_Wakeup(void *addr)
53 LWP_NoYieldSignal(addr);
56 PROCESS rx_listenerPid = 0; /* LWP process id of socket listener process */
57 static void* rx_ListenerProc(void *dummy);
60 * Delay the current thread the specified number of seconds.
68 static int quitListening = 0;
70 /* This routine will kill the listener thread, if it exists. */
72 rxi_StopListener(void)
75 rxi_ReScheduleEvents();
78 /* This routine will get called by the event package whenever a new,
79 earlier than others, event is posted. If the Listener process
80 is blocked in selects, this will unblock it. It also can be called
81 to force a new trip through the rxi_Listener select loop when the set
82 of file descriptors it should be listening to changes... */
84 rxi_ReScheduleEvents(void)
87 IOMGR_Cancel(rx_listenerPid);
91 rxi_InitializeThreadSupport(void)
95 LWP_InitializeProcessSupport(LWP_NORMAL_PRIORITY, &junk);
97 FD_ZERO(&rx_selectMask);
101 rxi_StartServerProc(void *(*proc) (void *), int stacksize)
104 static int number = 0;
107 sprintf(name, "srv_%d", ++number);
108 LWP_CreateProcess(proc, stacksize, RX_PROCESS_PRIORITY, NULL,
109 "rx_ServerProc", &scratchPid);
111 (*registerProgram) (scratchPid, name);
115 rxi_StartListener(void)
117 /* Priority of listener should be high, so it can keep conns alive */
118 #define RX_LIST_STACK 24000
119 LWP_CreateProcess(rx_ListenerProc, RX_LIST_STACK, LWP_MAX_PRIORITY,
120 NULL, "rx_Listener", &rx_listenerPid);
122 (*registerProgram) (rx_listenerPid, "listener");
125 /* The main loop which listens to the net for datagrams, and handles timeouts
126 and retransmissions, etc. It also is responsible for scheduling the
127 execution of pending events (in conjunction with event.c).
129 Note interaction of nextPollTime and lastPollWorked. The idea is that if rx is not
130 keeping up with the incoming stream of packets (because there are threads that
131 are interfering with its running sufficiently often), rx does a polling select
132 before doing a real IOMGR_Select system call. Doing a real select means that
133 we don't have to let other processes run before processing more packets.
135 So, our algorithm is that if the last poll on the file descriptor found useful data, or
136 we're at the time nextPollTime (which is advanced so that it occurs every 3 or 4 seconds),
137 then we try the polling select before the IOMGR_Select. If we eventually catch up
138 (which we can tell by the polling select returning no input packets ready), then we
139 don't do a polling select again until several seconds later (via nextPollTime mechanism).
143 rxi_ListenerProc(fd_set * rfds, int *tnop, struct rx_call **newcallp)
147 struct rx_packet *p = (struct rx_packet *)0;
150 afs_int32 nextPollTime; /* time to next poll FD before sleeping */
151 int lastPollWorked, doingPoll; /* true iff last poll was useful */
152 struct timeval tv, *tvp;
158 char name[MAXTHREADNAMELENGTH] = "srv_0";
163 code = LWP_CurrentProcess(&pid);
165 osi_Panic("rxi_Listener: Can't get my pid.\n");
167 rx_listenerPid = pid;
169 (*swapNameProgram) (pid, "listener", &name[0]);
172 /* See if a check for additional packets was issued */
175 /* Grab a new packet only if necessary (otherwise re-use the old one) */
177 rxi_RestoreDataBufs(p);
179 if (!(p = rxi_AllocPacket(RX_PACKET_CLASS_RECEIVE)))
180 osi_Panic("rxi_ListenerProc: no packets!"); /* Shouldn't happen */
182 /* Wait for the next event time or a packet to arrive. */
183 /* event_RaiseEvents schedules any events whose time has come and
184 * then atomically computes the time to the next event, guaranteeing
185 * that this is positive. If there is no next event, it returns 0 */
187 if (!rxevent_RaiseEvents(&cv))
190 /* It's important to copy cv to tv, because the 4.3 documentation
191 * for select threatens that *tv may be updated after a select, in
192 * future editions of the system, to indicate how much of the time
193 * period has elapsed. So we shouldn't rely on tv not being altered. */
194 tv.tv_sec = cv.sec; /* Time to next event */
195 tv.tv_usec = cv.usec;
199 rx_atomic_inc(&rx_stats.selects);
201 *rfds = rx_selectMask;
203 if (lastPollWorked || nextPollTime < clock_Sec()) {
204 /* we're catching up, or haven't tried to for a few seconds */
206 nextPollTime = clock_Sec() + 4; /* try again in 4 seconds no matter what */
207 tv.tv_sec = tv.tv_usec = 0; /* make sure we poll */
209 code = select((int)(rx_maxSocketNumber + 1), rfds, 0, 0, tvp);
212 code = IOMGR_Select((int)(rx_maxSocketNumber + 1), rfds, 0, 0, tvp);
214 lastPollWorked = 0; /* default is that it didn't find anything */
218 LWP_DestroyProcess(pid);
224 * If it was a timer interrupt then we can assume that
225 * the time has advanced by roughly the value of the
226 * previous timeout, and that there is now at least
232 /* select or IOMGR_Select returned failure */
233 debugSelectFailure++; /* update debugging counter */
238 * IOMGR_Cancel is invoked whenever a new event is
239 * posted that is earlier than any existing events.
240 * So we re-evaluate the time, and then go back to
247 /* Packets have arrived, presumably:
248 * If it wasn't a timer interrupt, then no event should have
249 * timed out yet (well some event may have, but only just...), so
250 * we don't bother looking to see if any have timed out, but just
251 * go directly to reading the data packets
257 for (i = 0; p && i < rfds->fd_count; i++) {
258 socket = rfds->fd_array[i];
259 if (rxi_ReadPacket(socket, p, &host, &port)) {
261 p = rxi_ReceivePacket(p, socket, host, port, tnop,
263 if (newcallp && *newcallp) {
267 if (swapNameProgram) {
268 (*swapNameProgram) (rx_listenerPid, name, 0);
276 for (socket = rx_minSocketNumber;
277 p && socket <= rx_maxSocketNumber; socket++) {
278 if (!FD_ISSET(socket, rfds))
280 if (rxi_ReadPacket(socket, p, &host, &port)) {
281 p = rxi_ReceivePacket(p, socket, host, port, tnop,
283 if (newcallp && *newcallp) {
287 if (swapNameProgram) {
288 (*swapNameProgram) (rx_listenerPid, name, 0);
302 /* This is the listener process request loop. The listener process loop
303 * becomes a server thread when rxi_ListenerProc returns, and stays
304 * server thread until rxi_ServerProc returns. */
306 rx_ListenerProc(void *dummy)
310 struct rx_call *newcall;
313 if (!(rfds = IOMGR_AllocFDSet())) {
314 osi_Panic("rx_ListenerProc: no fd_sets!\n");
320 rxi_ListenerProc(rfds, &threadID, &newcall);
321 /* osi_Assert(threadID != -1); */
322 /* osi_Assert(newcall != NULL); */
323 sock = OSI_NULLSOCKET;
324 rxi_ServerProc(threadID, newcall, &sock);
325 /* osi_Assert(sock != OSI_NULLSOCKET); */
331 /* This is the server process request loop. The server process loop
332 * becomes a listener thread when rxi_ServerProc returns, and stays
333 * listener thread until rxi_ListenerProc returns. */
335 rx_ServerProc(void * unused)
339 struct rx_call *newcall = NULL;
342 if (!(rfds = IOMGR_AllocFDSet())) {
343 osi_Panic("rxi_ListenerProc: no fd_sets!\n");
346 rxi_MorePackets(rx_maxReceiveWindow + 2); /* alloc more packets */
347 rxi_dataQuota += rx_initSendWindow; /* Reserve some pkts for hard times */
348 /* threadID is used for making decisions in GetCall. Get it by bumping
349 * number of threads handling incoming calls */
350 threadID = rxi_availProcs++;
353 sock = OSI_NULLSOCKET;
354 rxi_ServerProc(threadID, newcall, &sock);
355 /* osi_Assert(sock != OSI_NULLSOCKET); */
357 rxi_ListenerProc(rfds, &threadID, &newcall);
358 /* osi_Assert(threadID != -1); */
359 /* osi_Assert(newcall != NULL); */
366 * Called from GetUDPSocket.
367 * Called from a single thread at startup.
368 * Returns 0 on success; -1 on failure.
371 rxi_Listen(osi_socket sock)
375 * Put the socket into non-blocking mode so that rx_Listener
376 * can do a polling read before entering select
378 if (fcntl(sock, F_SETFL, FNDELAY) == -1) {
380 (osi_Msg "rxi_Listen: unable to set non-blocking mode on socket\n");
384 if (sock > FD_SETSIZE - 1) {
385 (osi_Msg "rxi_Listen: socket descriptor > (FD_SETSIZE-1) = %d\n",
391 FD_SET(sock, &rx_selectMask);
392 if (sock > rx_maxSocketNumber)
393 rx_maxSocketNumber = sock;
394 if (sock < rx_minSocketNumber)
395 rx_minSocketNumber = sock;
403 rxi_Recvmsg(osi_socket socket, struct msghdr *msg_p, int flags)
405 #if defined(HAVE_LINUX_ERRQUEUE_H) && defined(ADAPT_PMTU)
406 while((rxi_HandleSocketError(socket)) > 0)
409 return recvmsg(socket, msg_p, flags);
413 * Simulate a blocking sendmsg on the non-blocking socket.
414 * It's non blocking because it was set that way for recvmsg.
417 rxi_Sendmsg(osi_socket socket, struct msghdr *msg_p, int flags)
419 fd_set *sfds = (fd_set *) 0;
420 while (sendmsg(socket, msg_p, flags) == -1) {
423 rx_atomic_inc(&rx_stats.sendSelects);
426 if (!(sfds = IOMGR_AllocFDSet())) {
427 (osi_Msg "rx failed to alloc fd_set: ");
428 perror("rx_sendmsg");
431 FD_SET(socket, sfds);
433 #if defined(HAVE_LINUX_ERRQUEUE_H) && defined(ADAPT_PMTU)
434 while((rxi_HandleSocketError(socket)) > 0)
438 if (WSAGetLastError())
439 #elif defined(AFS_LINUX22_ENV)
440 /* linux unfortunately returns ECONNREFUSED if the target port
441 * is no longer in use */
442 /* and EAGAIN if a UDP checksum is incorrect */
443 if (errno != EWOULDBLOCK && errno != ENOBUFS && errno != ECONNREFUSED
446 if (errno != EWOULDBLOCK && errno != ENOBUFS)
449 (osi_Msg "rx failed to send packet: ");
450 perror("rx_sendmsg");
455 if (WSAGetLastError() > 0)
456 return -WSAGetLastError();
460 while ((err = select(socket + 1, 0, sfds, 0, 0)) != 1) {
461 if (err >= 0 || errno != EINTR)
462 osi_Panic("rxi_sendmsg: select error %d.%d", err, errno);
464 FD_SET(socket, sfds);
468 IOMGR_FreeFDSet(sfds);