#endif /* KERNEL */
#include <opr/queue.h>
+#include <hcrypto/rand.h>
#include "rx.h"
#include "rx_clock.h"
struct rx_call **newcallp);
static struct rx_packet
*rxi_ReceiveAckPacket(struct rx_call *call, struct rx_packet *np,
- int istack);
+ int istack, int *a_invalid);
static struct rx_packet
*rxi_ReceiveResponsePacket(struct rx_connection *conn,
struct rx_packet *np, int istack);
static void rxi_ScheduleGrowMTUEvent(struct rx_call *call, int secs);
static void rxi_KeepAliveOn(struct rx_call *call);
static void rxi_GrowMTUOn(struct rx_call *call);
-static void rxi_ChallengeOn(struct rx_connection *conn);
+static int rxi_ChallengeOn(struct rx_connection *conn);
static int rxi_CheckCall(struct rx_call *call, int haveCTLock);
static void rxi_AckAllInTransmitQueue(struct rx_call *call);
static void rxi_CancelKeepAliveEvent(struct rx_call *call);
static void rxi_CancelDelayedAbortEvent(struct rx_call *call);
static void rxi_CancelGrowMTUEvent(struct rx_call *call);
+static void update_nextCid(void);
+
+#ifndef KERNEL
+static void rxi_Finalize_locked(void);
+#elif defined(UKERNEL)
+# define rxi_Finalize_locked() do { } while (0)
+#endif
#ifdef RX_ENABLE_LOCKS
struct rx_tq_debug {
static unsigned int rxi_rpc_process_stat_cnt;
-/*
- * rxi_busyChannelError is a boolean. It indicates whether or not RX_CALL_BUSY
- * errors should be reported to the application when a call channel appears busy
- * (inferred from the receipt of RX_PACKET_TYPE_BUSY packets on the channel),
- * and there are other call channels in the connection that are not busy.
- * If 0, we do not return errors upon receiving busy packets; we just keep
- * trying on the same call channel until we hit a timeout.
- */
-static afs_int32 rxi_busyChannelError = 0;
-
rx_atomic_t rx_nWaiting = RX_ATOMIC_INIT(0);
rx_atomic_t rx_nWaited = RX_ATOMIC_INIT(0);
* calls to process */
struct opr_queue rx_idleServerQueue;
+/* List of free rx_serverQueueEntry structs */
+struct opr_queue rx_freeServerQueue;
+
#if !defined(offsetof)
#include <stddef.h> /* for definition of offsetof() */
#endif
#ifdef RX_ENABLE_LOCKS
afs_kmutex_t rx_atomic_mutex;
+static afs_kmutex_t freeSQEList_lock;
#endif
/* Forward prototypes */
extern afs_kmutex_t rx_refcnt_mutex;
extern afs_kmutex_t des_init_mutex;
extern afs_kmutex_t des_random_mutex;
+#ifndef KERNEL
extern afs_kmutex_t rx_clock_mutex;
extern afs_kmutex_t rxi_connCacheMutex;
extern afs_kmutex_t event_handler_mutex;
extern afs_kcondvar_t rx_event_handler_cond;
extern afs_kcondvar_t rx_listener_cond;
+#endif /* !KERNEL */
static afs_kmutex_t epoch_mutex;
static afs_kmutex_t rx_init_mutex;
static void
rxi_InitPthread(void)
{
- MUTEX_INIT(&rx_clock_mutex, "clock", MUTEX_DEFAULT, 0);
- MUTEX_INIT(&rx_stats_mutex, "stats", MUTEX_DEFAULT, 0);
- MUTEX_INIT(&rx_atomic_mutex, "atomic", MUTEX_DEFAULT, 0);
MUTEX_INIT(&rx_quota_mutex, "quota", MUTEX_DEFAULT, 0);
MUTEX_INIT(&rx_pthread_mutex, "pthread", MUTEX_DEFAULT, 0);
MUTEX_INIT(&rx_packets_mutex, "packets", MUTEX_DEFAULT, 0);
MUTEX_INIT(&rx_refcnt_mutex, "refcnts", MUTEX_DEFAULT, 0);
- MUTEX_INIT(&epoch_mutex, "epoch", MUTEX_DEFAULT, 0);
- MUTEX_INIT(&rx_init_mutex, "init", MUTEX_DEFAULT, 0);
- MUTEX_INIT(&event_handler_mutex, "event handler", MUTEX_DEFAULT, 0);
+#ifndef KERNEL
+ MUTEX_INIT(&rx_clock_mutex, "clock", MUTEX_DEFAULT, 0);
MUTEX_INIT(&rxi_connCacheMutex, "conn cache", MUTEX_DEFAULT, 0);
+ MUTEX_INIT(&event_handler_mutex, "event handler", MUTEX_DEFAULT, 0);
MUTEX_INIT(&listener_mutex, "listener", MUTEX_DEFAULT, 0);
MUTEX_INIT(&rx_if_init_mutex, "if init", MUTEX_DEFAULT, 0);
MUTEX_INIT(&rx_if_mutex, "if", MUTEX_DEFAULT, 0);
+#endif
+ MUTEX_INIT(&rx_stats_mutex, "stats", MUTEX_DEFAULT, 0);
+ MUTEX_INIT(&rx_atomic_mutex, "atomic", MUTEX_DEFAULT, 0);
+ MUTEX_INIT(&epoch_mutex, "epoch", MUTEX_DEFAULT, 0);
+ MUTEX_INIT(&rx_init_mutex, "init", MUTEX_DEFAULT, 0);
MUTEX_INIT(&rx_debug_mutex, "debug", MUTEX_DEFAULT, 0);
+#ifndef KERNEL
CV_INIT(&rx_event_handler_cond, "evhand", CV_DEFAULT, 0);
CV_INIT(&rx_listener_cond, "rxlisten", CV_DEFAULT, 0);
+#endif
osi_Assert(pthread_key_create(&rx_thread_id_key, NULL) == 0);
osi_Assert(pthread_key_create(&rx_ts_info_key, NULL) == 0);
MUTEX_INIT(&rx_rpc_stats, "rx_rpc_stats", MUTEX_DEFAULT, 0);
MUTEX_INIT(&rx_freePktQ_lock, "rx_freePktQ_lock", MUTEX_DEFAULT, 0);
+ MUTEX_INIT(&rx_mallocedPktQ_lock, "rx_mallocedPktQ_lock", MUTEX_DEFAULT,
+ 0);
+
#ifdef RX_ENABLE_LOCKS
#ifdef RX_LOCKS_DB
rxdb_init();
MUTEX_INIT(&rx_connHashTable_lock, "rx_connHashTable_lock", MUTEX_DEFAULT,
0);
MUTEX_INIT(&rx_serverPool_lock, "rx_serverPool_lock", MUTEX_DEFAULT, 0);
+#ifndef KERNEL
MUTEX_INIT(&rxi_keyCreate_lock, "rxi_keyCreate_lock", MUTEX_DEFAULT, 0);
+#endif
#endif /* RX_ENABLE_LOCKS */
}
* tiers:
*
* rx_connHashTable_lock - synchronizes conn creation, rx_connHashTable access
+ * also protects updates to rx_nextCid
* conn_call_lock - used to synchonize rx_EndCall and rx_NewCall
* call->lock - locks call data fields.
* These are independent of each other:
#define CLEAR_CALL_QUEUE_LOCK(C)
#endif /* RX_ENABLE_LOCKS */
struct rx_serverQueueEntry *rx_waitForPacket = 0;
-struct rx_serverQueueEntry *rx_waitingForPacket = 0;
-/* ------------Exported Interfaces------------- */
-
-/* This function allows rxkad to set the epoch to a suitably random number
- * which rx_NewConnection will use in the future. The principle purpose is to
- * get rxnull connections to use the same epoch as the rxkad connections do, at
- * least once the first rxkad connection is established. This is important now
- * that the host/port addresses aren't used in FindConnection: the uniqueness
- * of epoch/cid matters and the start time won't do. */
-
-#ifdef AFS_PTHREAD_ENV
/*
- * This mutex protects the following global variables:
- * rx_epoch
+ * This mutex serializes calls to our initialization and shutdown routines
+ * (rx_InitHost, rx_Finalize and shutdown_rx). Only one thread can be running
+ * these at any time; all other threads must wait for it to finish running, and
+ * then examine the value of rxi_running afterwards.
*/
-
-#define LOCK_EPOCH MUTEX_ENTER(&epoch_mutex)
-#define UNLOCK_EPOCH MUTEX_EXIT(&epoch_mutex)
+#ifdef AFS_PTHREAD_ENV
+# define LOCK_RX_INIT MUTEX_ENTER(&rx_init_mutex)
+# define UNLOCK_RX_INIT MUTEX_EXIT(&rx_init_mutex)
#else
-#define LOCK_EPOCH
-#define UNLOCK_EPOCH
-#endif /* AFS_PTHREAD_ENV */
+# define LOCK_RX_INIT
+# define UNLOCK_RX_INIT
+#endif
-void
-rx_SetEpoch(afs_uint32 epoch)
+/* ------------Exported Interfaces------------- */
+
+static rx_atomic_t rxi_running = RX_ATOMIC_INIT(0);
+int
+rxi_IsRunning(void)
{
- LOCK_EPOCH;
- rx_epoch = epoch;
- UNLOCK_EPOCH;
+ return rx_atomic_read(&rxi_running);
}
/* Initialize rx. A port number may be mentioned, in which case this
* by the kernel. Whether this will ever overlap anything in
* /etc/services is anybody's guess... Returns 0 on success, -1 on
* error. */
-#ifndef AFS_NT40_ENV
-static
-#endif
-rx_atomic_t rxinit_status = RX_ATOMIC_INIT(1);
-
int
rx_InitHost(u_int host, u_int port)
{
#ifdef KERNEL
- osi_timeval_t tv;
+ osi_timeval32_t tv;
#else /* KERNEL */
struct timeval tv;
#endif /* KERNEL */
SPLVAR;
INIT_PTHREAD_LOCKS;
- if (!rx_atomic_test_and_clear_bit(&rxinit_status, 0))
+ LOCK_RX_INIT;
+ if (rxi_IsRunning()) {
+ UNLOCK_RX_INIT;
return 0; /* already started */
-
+ }
#ifdef RXDEBUG
rxi_DebugInit();
#endif
#ifdef AFS_NT40_ENV
if (afs_winsockInit() < 0)
- return -1;
+ goto error;
#endif
#ifndef KERNEL
rx_socket = rxi_GetHostUDPSocket(host, (u_short) port);
if (rx_socket == OSI_NULLSOCKET) {
- return RX_ADDRINUSE;
+ goto addrinuse;
}
#if defined(RX_ENABLE_LOCKS) && defined(KERNEL)
#ifdef RX_LOCKS_DB
MUTEX_INIT(&rx_connHashTable_lock, "rx_connHashTable_lock", MUTEX_DEFAULT,
0);
MUTEX_INIT(&rx_serverPool_lock, "rx_serverPool_lock", MUTEX_DEFAULT, 0);
+ MUTEX_INIT(&rx_mallocedPktQ_lock, "rx_mallocedPktQ_lock", MUTEX_DEFAULT,
+ 0);
+
#if defined(AFS_HPUX110_ENV)
if (!uniprocessor)
rx_sleepLock = alloc_spinlock(LAST_HELD_ORDER - 10, "rx_sleepLock");
opr_queue_Init(&rx_freePacketQueue);
rxi_NeedMorePackets = FALSE;
rx_nPackets = 0; /* rx_nPackets is managed by rxi_MorePackets* */
+ opr_queue_Init(&rx_mallocedPacketQueue);
/* enforce a minimum number of allocated packets */
if (rx_extraPackets < rxi_nSendFrags * rx_maxSendWindow)
socklen_t addrlen = sizeof(addr);
#endif
if (getsockname((intptr_t)rx_socket, (struct sockaddr *)&addr, &addrlen)) {
- rx_Finalize();
+ rxi_Finalize_locked();
osi_Free(htable, rx_hashTableSize * sizeof(struct rx_connection *));
- return -1;
+ goto error;
}
rx_port = addr.sin_port;
#endif
}
rx_stats.minRtt.sec = 9999999;
-#ifdef KERNEL
- rx_SetEpoch(tv.tv_sec | 0x80000000);
-#else
- rx_SetEpoch(tv.tv_sec); /* Start time of this package, rxkad
- * will provide a randomer value. */
-#endif
+ if (RAND_bytes(&rx_epoch, sizeof(rx_epoch)) != 1)
+ goto error;
+ rx_epoch = (rx_epoch & ~0x40000000) | 0x80000000;
+ if (RAND_bytes(&rx_nextCid, sizeof(rx_nextCid)) != 1)
+ goto error;
+ rx_nextCid &= RX_CIDMASK;
MUTEX_ENTER(&rx_quota_mutex);
rxi_dataQuota += rx_extraQuota; /* + extra pkts caller asked to rsrv */
MUTEX_EXIT(&rx_quota_mutex);
/* Initialize various global queues */
opr_queue_Init(&rx_idleServerQueue);
+ opr_queue_Init(&rx_freeServerQueue);
opr_queue_Init(&rx_incomingCallQueue);
opr_queue_Init(&rx_freeCallQueue);
rxi_StartListener();
USERPRI;
- rx_atomic_clear_bit(&rxinit_status, 0);
+
+ rx_atomic_set(&rxi_running, 1);
+ UNLOCK_RX_INIT;
+
return 0;
+
+ addrinuse:
+ UNLOCK_RX_INIT;
+ return RX_ADDRINUSE;
+
+ error:
+ UNLOCK_RX_INIT;
+ return -1;
}
int
{
struct clock now, retryTime;
+ MUTEX_ASSERT(&call->lock);
clock_GetTime(&now);
retryTime = now;
static_inline void
rxi_rto_cancel(struct rx_call *call)
{
- rxevent_Cancel(&call->resendEvent);
- CALL_RELE(call, RX_CALL_REFCOUNT_RESEND);
+ MUTEX_ASSERT(&call->lock);
+ if (rxevent_Cancel(&call->resendEvent))
+ CALL_RELE(call, RX_CALL_REFCOUNT_RESEND);
}
/*!
}
/**
- * Enables or disables the busy call channel error (RX_CALL_BUSY).
- *
- * @param[in] onoff Non-zero to enable busy call channel errors.
- *
- * @pre Neither rx_Init nor rx_InitHost have been called yet
- */
-void
-rx_SetBusyChannelError(afs_int32 onoff)
-{
- osi_Assert(rx_atomic_test_bit(&rxinit_status, 0));
- rxi_busyChannelError = onoff ? 1 : 0;
-}
-
-/**
* Set a delayed ack event on the specified call for the given time
*
* @param[in] call - the call on which to set the event
{
struct clock now, when;
+ MUTEX_ASSERT(&call->lock);
clock_GetTime(&now);
when = now;
clock_Add(&when, offset);
- if (call->delayedAckEvent && clock_Gt(&call->delayedAckTime, &when)) {
- /* The event we're cancelling already has a reference, so we don't
- * need a new one */
- rxevent_Cancel(&call->delayedAckEvent);
+ if (clock_Gt(&call->delayedAckTime, &when) &&
+ rxevent_Cancel(&call->delayedAckEvent)) {
+ /* We successfully cancelled an event too far in the future to install
+ * our new one; we can reuse the reference on the call. */
call->delayedAckEvent = rxevent_Post(&when, &now, rxi_SendDelayedAck,
call, NULL, 0);
call->delayedAckTime = when;
- } else if (!call->delayedAckEvent) {
+ } else if (call->delayedAckEvent == NULL) {
CALL_HOLD(call, RX_CALL_REFCOUNT_DELAY);
call->delayedAckEvent = rxevent_Post(&when, &now,
rxi_SendDelayedAck,
void
rxi_CancelDelayedAckEvent(struct rx_call *call)
{
- if (call->delayedAckEvent) {
- rxevent_Cancel(&call->delayedAckEvent);
+ MUTEX_ASSERT(&call->lock);
+ /* Only drop the ref if we cancelled it before it could run. */
+ if (rxevent_Cancel(&call->delayedAckEvent))
CALL_RELE(call, RX_CALL_REFCOUNT_DELAY);
- }
}
/* called with unincremented nRequestsRunning to see if it is OK to start
int serviceSecurityIndex)
{
int hashindex, i;
- afs_int32 cid;
struct rx_connection *conn;
+ int code;
SPLVAR;
#endif
NETPRI;
MUTEX_ENTER(&rx_connHashTable_lock);
- cid = (rx_nextCid += RX_MAXCALLS);
conn->type = RX_CLIENT_CONNECTION;
- conn->cid = cid;
conn->epoch = rx_epoch;
+ conn->cid = rx_nextCid;
+ update_nextCid();
conn->peer = rxi_FindPeer(shost, sport, 1);
conn->serviceId = sservice;
conn->securityObject = securityObject;
conn->lastBusy[i] = 0;
}
- RXS_NewConnection(securityObject, conn);
+ code = RXS_NewConnection(securityObject, conn);
hashindex =
CONN_HASH(shost, sport, conn->cid, conn->epoch, RX_CLIENT_CONNECTION);
rx_atomic_inc(&rx_stats.nClientConns);
MUTEX_EXIT(&rx_connHashTable_lock);
USERPRI;
+ if (code) {
+ rxi_ConnectionError(conn, code);
+ }
return conn;
}
rx_SetConnIdleDeadTime(struct rx_connection *conn, int seconds)
{
conn->idleDeadTime = seconds;
- conn->idleDeadDetection = (seconds ? 1 : 0);
rxi_CheckConnTimeouts(conn);
}
{
struct rx_connection **conn_ptr;
int havecalls = 0;
- struct rx_packet *packet;
int i;
SPLVAR;
if (conn->refCount > 0)
conn->refCount--;
else {
+#ifdef RX_REFCOUNT_CHECK
+ osi_Assert(conn->refCount == 0);
+#endif
if (rx_stats_active) {
MUTEX_ENTER(&rx_stats_mutex);
rxi_lowConnRefCount++;
if ((conn->type == RX_CLIENT_CONNECTION)
&& (conn->flags & (RX_CONN_MAKECALL_WAITING|RX_CONN_MAKECALL_ACTIVE))) {
conn->flags |= RX_CONN_DESTROY_ME;
+ MUTEX_EXIT(&rx_refcnt_mutex);
MUTEX_EXIT(&conn->conn_data_lock);
USERPRI;
return;
return;
}
- if (conn->natKeepAliveEvent) {
- rxi_NatKeepAliveOff(conn);
- }
-
- if (conn->delayedAbortEvent) {
- rxevent_Cancel(&conn->delayedAbortEvent);
- packet = rxi_AllocPacket(RX_PACKET_CLASS_SPECIAL);
- if (packet) {
- MUTEX_ENTER(&conn->conn_data_lock);
- rxi_SendConnectionAbort(conn, packet, 0, 1);
- MUTEX_EXIT(&conn->conn_data_lock);
- rxi_FreePacket(packet);
- }
- }
-
/* Remove from connection hash table before proceeding */
conn_ptr =
&rx_connHashTable[CONN_HASH
rxLastConn = 0;
/* Make sure the connection is completely reset before deleting it. */
- /* get rid of pending events that could zap us later */
- rxevent_Cancel(&conn->challengeEvent);
- rxevent_Cancel(&conn->checkReachEvent);
- rxevent_Cancel(&conn->natKeepAliveEvent);
+ /*
+ * Pending events hold a refcount, so we can't get here if they are
+ * non-NULL. */
+ osi_Assert(conn->challengeEvent == NULL);
+ osi_Assert(conn->delayedAbortEvent == NULL);
+ osi_Assert(conn->natKeepAliveEvent == NULL);
+ osi_Assert(conn->checkReachEvent == NULL);
/* Add the connection to the list of destroyed connections that
* need to be cleaned up. This is necessary to avoid deadlocks
rxi_WakeUpTransmitQueue(struct rx_call *call)
{
if (call->tqWaiters || (call->flags & RX_CALL_TQ_WAIT)) {
- dpf(("call %"AFS_PTR_FMT" has %d waiters and flags %d\n",
+ dpf(("call %p has %d waiters and flags %d\n",
call, call->tqWaiters, call->flags));
#ifdef RX_ENABLE_LOCKS
MUTEX_ASSERT(&call->lock);
SPLVAR;
clock_NewTime();
- dpf(("rx_NewCall(conn %"AFS_PTR_FMT")\n", conn));
+ dpf(("rx_NewCall(conn %p)\n", conn));
NETPRI;
clock_GetTime(&queueTime);
}
if (i < RX_MAXCALLS) {
conn->lastBusy[i] = 0;
- call->flags &= ~RX_CALL_PEER_BUSY;
break;
}
if (!wait)
MUTEX_EXIT(&call->lock);
USERPRI;
- dpf(("rx_NewCall(call %"AFS_PTR_FMT")\n", call));
+ dpf(("rx_NewCall(call %p)\n", call));
return call;
}
service->minProcs = 0;
service->maxProcs = 1;
service->idleDeadTime = 60;
- service->idleDeadErr = 0;
service->connDeadTime = rx_connDeadTime;
service->executeRequestProc = serviceProc;
service->checkReach = 0;
void *value)
{
int i;
+ int code;
for (i = 0; i<service->nSecurityObjects; i++) {
if (service->securityObjects[i]) {
- RXS_SetConfiguration(service->securityObjects[i], NULL, type,
- value, NULL);
+ code = RXS_SetConfiguration(service->securityObjects[i], NULL, type,
+ value, NULL);
+ if (code) {
+ return code;
+ }
}
}
return 0;
void
rx_WakeupServerProcs(void)
{
- struct rx_serverQueueEntry *np, *tqp;
+ struct rx_serverQueueEntry *np;
struct opr_queue *cursor;
SPLVAR;
osi_rxWakeup(rx_waitForPacket);
#endif /* RX_ENABLE_LOCKS */
MUTEX_ENTER(&freeSQEList_lock);
- for (np = rx_FreeSQEList; np; np = tqp) {
- tqp = *(struct rx_serverQueueEntry **)np;
+ for (opr_queue_Scan(&rx_freeServerQueue, cursor)) {
+ np = opr_queue_Entry(cursor, struct rx_serverQueueEntry, entry);
#ifdef RX_ENABLE_LOCKS
CV_BROADCAST(&np->cv);
#else /* RX_ENABLE_LOCKS */
MUTEX_ENTER(&freeSQEList_lock);
- if ((sq = rx_FreeSQEList)) {
- rx_FreeSQEList = *(struct rx_serverQueueEntry **)sq;
+ if (!opr_queue_IsEmpty(&rx_freeServerQueue)) {
+ sq = opr_queue_First(&rx_freeServerQueue, struct rx_serverQueueEntry,
+ entry);
+ opr_queue_Remove(&sq->entry);
MUTEX_EXIT(&freeSQEList_lock);
} else { /* otherwise allocate a new one and return that */
MUTEX_EXIT(&freeSQEList_lock);
}
MUTEX_ENTER(&rx_pthread_mutex);
if (tno == rxi_fcfs_thread_num
- || opr_queue_IsEnd(&rx_incomingCallQueue, cursor)) {
+ || opr_queue_IsLast(&rx_incomingCallQueue, cursor)) {
MUTEX_EXIT(&rx_pthread_mutex);
/* If we're the fcfs thread , then we'll just use
* this call. If we haven't been able to find an optimal
opr_queue_Remove(&call->entry);
MUTEX_EXIT(&rx_serverPool_lock);
MUTEX_ENTER(&call->lock);
+ CLEAR_CALL_QUEUE_LOCK(call);
if (call->flags & RX_CALL_WAIT_PROC) {
call->flags &= ~RX_CALL_WAIT_PROC;
|| opr_queue_First(&call->rq, struct rx_packet, entry)->header.seq != 1)
rxi_SendAck(call, 0, 0, RX_ACK_DELAY, 0);
- CLEAR_CALL_QUEUE_LOCK(call);
break;
} else {
/* If there are no eligible incoming calls, add this process
opr_queue_Append(&rx_idleServerQueue, &sq->entry);
#ifndef AFS_AIX41_ENV
rx_waitForPacket = sq;
-#else
- rx_waitingForPacket = sq;
#endif /* AFS_AIX41_ENV */
do {
CV_WAIT(&sq->cv, &rx_serverPool_lock);
#ifdef KERNEL
if (afs_termState == AFSOP_STOP_RXCALLBACK) {
- MUTEX_EXIT(&rx_serverPool_lock);
- return (struct rx_call *)0;
+ break;
}
#endif
} while (!(call = sq->newcall)
&& !(socketp && *socketp != OSI_NULLSOCKET));
+ if (opr_queue_IsOnQueue(&sq->entry)) {
+ opr_queue_Remove(&sq->entry);
+ }
MUTEX_EXIT(&rx_serverPool_lock);
if (call) {
MUTEX_ENTER(&call->lock);
}
MUTEX_ENTER(&freeSQEList_lock);
- *(struct rx_serverQueueEntry **)sq = rx_FreeSQEList;
- rx_FreeSQEList = sq;
+ opr_queue_Prepend(&rx_freeServerQueue, &sq->entry);
MUTEX_EXIT(&freeSQEList_lock);
if (call) {
#endif
rxi_calltrace(RX_CALL_START, call);
- dpf(("rx_GetCall(port=%d, service=%d) ==> call %"AFS_PTR_FMT"\n",
+ dpf(("rx_GetCall(port=%d, service=%d) ==> call %p\n",
call->conn->service->servicePort, call->conn->service->serviceId,
call));
NETPRI;
MUTEX_ENTER(&freeSQEList_lock);
- if ((sq = rx_FreeSQEList)) {
- rx_FreeSQEList = *(struct rx_serverQueueEntry **)sq;
+ if (!opr_queue_IsEmpty(&rx_freeServerQueue)) {
+ sq = opr_queue_First(&rx_freeServerQueue, struct rx_serverQueueEntry,
+ entry);
+ opr_queue_Remove(&sq->entry);
MUTEX_EXIT(&freeSQEList_lock);
} else { /* otherwise allocate a new one and return that */
MUTEX_EXIT(&freeSQEList_lock);
service = tcall->conn->service;
if (QuotaOK(service)) {
MUTEX_ENTER(&rx_pthread_mutex);
- /* XXX - If tcall->entry.next is NULL, then we're no longer
- * on a queue at all. This shouldn't happen. */
- if (tno == rxi_fcfs_thread_num || !tcall->entry.next) {
+ if (tno == rxi_fcfs_thread_num
+ || opr_queue_IsLast(&rx_incomingCallQueue, cursor)) {
MUTEX_EXIT(&rx_pthread_mutex);
/* If we're the fcfs thread, then we'll just use
* this call. If we haven't been able to find an optimal
if (call) {
opr_queue_Remove(&call->entry);
+ CLEAR_CALL_QUEUE_LOCK(call);
/* we can't schedule a call if there's no data!!! */
/* send an ack if there's no data, if we're missing the
* first packet, or we're missing something between first
MUTEX_EXIT(&sq->lock);
MUTEX_ENTER(&freeSQEList_lock);
- *(struct rx_serverQueueEntry **)sq = rx_FreeSQEList;
- rx_FreeSQEList = sq;
+ opr_queue_Prepend(&rx_freeServerQueue, &sq->entry);
MUTEX_EXIT(&freeSQEList_lock);
if (call) {
afs_int32 error;
SPLVAR;
- dpf(("rx_EndCall(call %"AFS_PTR_FMT" rc %d error %d abortCode %d)\n",
+ dpf(("rx_EndCall(call %p rc %d error %d abortCode %d)\n",
call, rc, call->error, call->abortCode));
NETPRI;
call->abortCount = 0;
}
- call->arrivalProc = (void (*)())0;
+ call->arrivalProc = NULL;
if (rc && call->error == 0) {
rxi_CallError(call, rc);
call->app.mode = RX_MODE_ERROR;
MUTEX_ENTER(&call->lock);
}
if (call->app.mode == RX_MODE_SENDING) {
- MUTEX_EXIT(&call->lock);
- rxi_FlushWrite(call);
- MUTEX_ENTER(&call->lock);
+ rxi_FlushWriteLocked(call);
}
rxi_calltrace(RX_CALL_END, call);
/* Call goes to hold state until reply packets are acknowledged */
MUTEX_ENTER(&conn->conn_call_lock);
MUTEX_ENTER(&call->lock);
- if (!(call->flags & RX_CALL_PEER_BUSY)) {
+ if (!call->error) {
+ /* While there are some circumstances where a call with an error is
+ * obviously not on a "busy" channel, be conservative (clearing
+ * lastBusy is just best-effort to possibly speed up rx_NewCall).
+ * The call channel is definitely not busy if we just successfully
+ * completed a call on it. */
conn->lastBusy[call->channel] = 0;
+
+ } else if (call->error == RX_CALL_TIMEOUT) {
+ /* The call is still probably running on the server side, so try to
+ * avoid this call channel in the future. */
+ conn->lastBusy[call->channel] = clock_Sec();
}
MUTEX_ENTER(&conn->conn_data_lock);
* Map errors to the local host's errno.h format.
*/
error = ntoh_syserr_conv(error);
+
+ /* If the caller said the call failed with some error, we had better
+ * return an error code. */
+ osi_Assert(!rc || error);
return error;
}
void
rx_Finalize(void)
{
- struct rx_connection **conn_ptr, **conn_end;
-
INIT_PTHREAD_LOCKS;
- if (rx_atomic_test_and_set_bit(&rxinit_status, 0))
+ LOCK_RX_INIT;
+ if (!rxi_IsRunning()) {
+ UNLOCK_RX_INIT;
return; /* Already shutdown. */
+ }
+ rxi_Finalize_locked();
+ UNLOCK_RX_INIT;
+}
+static void
+rxi_Finalize_locked(void)
+{
+ struct rx_connection **conn_ptr, **conn_end;
+ rx_atomic_set(&rxi_running, 0);
rxi_DeleteCachedConnections();
if (rx_connHashTable) {
MUTEX_ENTER(&rx_connHashTable_lock);
for (conn = *conn_ptr; conn; conn = next) {
next = conn->next;
if (conn->type == RX_CLIENT_CONNECTION) {
- MUTEX_ENTER(&rx_refcnt_mutex);
- conn->refCount++;
- MUTEX_EXIT(&rx_refcnt_mutex);
+ rx_GetConnection(conn);
#ifdef RX_ENABLE_LOCKS
rxi_DestroyConnectionNoLock(conn);
#else /* RX_ENABLE_LOCKS */
#ifdef AFS_NT40_ENV
afs_winsockCleanup();
#endif
-
}
#endif
struct opr_queue *cursor;
#endif
- dpf(("rxi_NewCall(conn %"AFS_PTR_FMT", channel %d)\n", conn, channel));
+ dpf(("rxi_NewCall(conn %p, channel %d)\n", conn, channel));
/* Grab an existing call structure, or allocate a new one.
* Existing call structures are assumed to have been left reset by
*/
MUTEX_ENTER(&conn->conn_data_lock);
if (conn->flags & RX_CONN_DESTROY_ME && !(conn->flags & RX_CONN_MAKECALL_WAITING)) {
- MUTEX_ENTER(&rx_refcnt_mutex);
- conn->refCount++;
- MUTEX_EXIT(&rx_refcnt_mutex);
+ rx_GetConnection(conn);
MUTEX_EXIT(&conn->conn_data_lock);
#ifdef RX_ENABLE_LOCKS
if (haveCTLock)
}
p = (char *)
-#if defined(KERNEL) && !defined(UKERNEL) && defined(AFS_FBSD80_ENV)
+#if defined(KERNEL) && !defined(UKERNEL) && defined(AFS_FBSD_ENV)
afs_osi_Alloc_NoSleep(size);
#else
osi_Alloc(size);
void
rxi_Free(void *addr, size_t size)
{
+ if (!addr) {
+ return;
+ }
if (rx_stats_active) {
rx_atomic_sub(&rxi_Allocsize, (int) size);
rx_atomic_dec(&rxi_Alloccnt);
if (peer->ifMTU < OLD_MAX_PACKET_SIZE)
peer->maxDgramPackets = 1;
/* We no longer have valid peer packet information */
- if (peer->maxPacketSize-RX_IPUDP_SIZE > peer->ifMTU)
+ if (peer->maxPacketSize + RX_HEADER_SIZE > peer->ifMTU)
peer->maxPacketSize = 0;
MUTEX_EXIT(&peer->peer_lock);
int *unknownService)
{
int hashindex, flag, i;
+ int code = 0;
struct rx_connection *conn;
*unknownService = 0;
hashindex = CONN_HASH(host, port, cid, epoch, type);
conn->nSpecific = 0;
conn->specific = NULL;
rx_SetConnDeadTime(conn, service->connDeadTime);
- conn->idleDeadTime = service->idleDeadTime;
- conn->idleDeadDetection = service->idleDeadErr ? 1 : 0;
+ rx_SetConnIdleDeadTime(conn, service->idleDeadTime);
for (i = 0; i < RX_MAXCALLS; i++) {
conn->twind[i] = rx_initSendWindow;
conn->rwind[i] = rx_initReceiveWindow;
}
/* Notify security object of the new connection */
- RXS_NewConnection(conn->securityObject, conn);
+ code = RXS_NewConnection(conn->securityObject, conn);
/* XXXX Connection timeout? */
if (service->newConnProc)
(*service->newConnProc) (conn);
rx_atomic_inc(&rx_stats.nServerConns);
}
- MUTEX_ENTER(&rx_refcnt_mutex);
- conn->refCount++;
- MUTEX_EXIT(&rx_refcnt_mutex);
+ rx_GetConnection(conn);
rxLastConn = conn; /* store this connection as the last conn used */
MUTEX_EXIT(&rx_connHashTable_lock);
- return conn;
-}
-
-/**
- * Timeout a call on a busy call channel if appropriate.
- *
- * @param[in] call The busy call.
- *
- * @pre 'call' is marked as busy (namely,
- * call->conn->lastBusy[call->channel] != 0)
- *
- * @pre call->lock is held
- * @pre rxi_busyChannelError is nonzero
- *
- * @note call->lock is dropped and reacquired
- */
-static void
-rxi_CheckBusy(struct rx_call *call)
-{
- struct rx_connection *conn = call->conn;
- int channel = call->channel;
- int freechannel = 0;
- int i;
-
- MUTEX_EXIT(&call->lock);
-
- MUTEX_ENTER(&conn->conn_call_lock);
-
- /* Are there any other call slots on this conn that we should try? Look for
- * slots that are empty and are either non-busy, or were marked as busy
- * longer than conn->secondsUntilDead seconds before this call started. */
-
- for (i = 0; i < RX_MAXCALLS && !freechannel; i++) {
- if (i == channel) {
- /* only look at channels that aren't us */
- continue;
- }
-
- if (conn->lastBusy[i]) {
- /* if this channel looked busy too recently, don't look at it */
- if (conn->lastBusy[i] >= call->startTime.sec) {
- continue;
- }
- if (call->startTime.sec - conn->lastBusy[i] < conn->secondsUntilDead) {
- continue;
- }
- }
-
- if (conn->call[i]) {
- struct rx_call *tcall = conn->call[i];
- MUTEX_ENTER(&tcall->lock);
- if (tcall->state == RX_STATE_DALLY) {
- freechannel = 1;
- }
- MUTEX_EXIT(&tcall->lock);
- } else {
- freechannel = 1;
- }
+ if (code) {
+ rxi_ConnectionError(conn, code);
}
-
- MUTEX_ENTER(&call->lock);
-
- /* Since the call->lock has been released it is possible that the call may
- * no longer be busy (the call channel cannot have been reallocated as we
- * haven't dropped the conn_call_lock) Therefore, we must confirm
- * that the call state has not changed when deciding whether or not to
- * force this application thread to retry by forcing a Timeout error. */
-
- if (freechannel && (call->flags & RX_CALL_PEER_BUSY)) {
- /* Since 'freechannel' is set, there exists another channel in this
- * rx_conn that the application thread might be able to use. We know
- * that we have the correct call since callNumber is unchanged, and we
- * know that the call is still busy. So, set the call error state to
- * rxi_busyChannelError so the application can retry the request,
- * presumably on a less-busy call channel. */
-
- rxi_CallError(call, RX_CALL_BUSY);
- }
- MUTEX_EXIT(&conn->conn_call_lock);
+ return conn;
}
/*!
rxi_AbortIfServerBusy(osi_socket socket, struct rx_connection *conn,
struct rx_packet *np)
{
+ afs_uint32 serial;
+
if ((rx_BusyThreshold > 0) &&
(rx_atomic_read(&rx_nWaiting) > rx_BusyThreshold)) {
+ MUTEX_ENTER(&conn->conn_data_lock);
+ serial = ++conn->serial;
+ MUTEX_EXIT(&conn->conn_data_lock);
rxi_SendRawAbort(socket, conn->peer->host, conn->peer->port,
- rx_BusyError, np, 0);
+ serial, rx_BusyError, np, 0);
if (rx_stats_active)
rx_atomic_inc(&rx_stats.nBusies);
return 1;
channel = np->header.cid & RX_CHANNELMASK;
MUTEX_ENTER(&conn->conn_call_lock);
call = conn->call[channel];
+ if (np->header.type == RX_PACKET_TYPE_BUSY) {
+ conn->lastBusy[channel] = clock_Sec();
+ }
if (!call || conn->callNumber[channel] != np->header.callNumber) {
MUTEX_EXIT(&conn->conn_call_lock);
if (rx_stats_active)
call = conn->call[channel];
if (!call) {
+ if (np->header.type != RX_PACKET_TYPE_DATA) {
+ /*
+ * Clients must send DATA packets at some point to create a new
+ * call. If the first packet we saw for this call channel is
+ * something else, then either the DATA packets got lost/delayed,
+ * or we were restarted and this is an existing call from before we
+ * were restarted. In the latter case, some clients get confused if
+ * we respond to such requests, so just drop the packet to make
+ * things easier for them.
+ */
+ MUTEX_EXIT(&conn->conn_call_lock);
+ if (rx_stats_active)
+ rx_atomic_inc(&rx_stats.spuriousPacketsRead);
+ return NULL;
+ }
+
+ if (np->header.seq > rx_maxReceiveWindow) {
+ /*
+ * This is a DATA packet for further along in the call than is
+ * possible for a new call. This is probably from an existing call
+ * that was in the middle of running when we were restarted; ignore
+ * it to avoid confusing clients. (See above comment about non-DATA
+ * packets.)
+ */
+ MUTEX_EXIT(&conn->conn_call_lock);
+ if (rx_stats_active)
+ rx_atomic_inc(&rx_stats.spuriousPacketsRead);
+ return NULL;
+ }
+
if (rxi_AbortIfServerBusy(socket, conn, np)) {
MUTEX_EXIT(&conn->conn_call_lock);
return NULL;
struct rx_connection *conn;
int type;
int unknownService = 0;
+ int invalid = 0;
#ifdef RXDEBUG
char *packetType;
#endif
* this is the first time the packet has been seen */
packetType = (np->header.type > 0 && np->header.type < RX_N_PACKET_TYPES)
? rx_packetTypes[np->header.type - 1] : "*UNKNOWN*";
- dpf(("R %d %s: %x.%d.%d.%d.%d.%d.%d flags %d, packet %"AFS_PTR_FMT"\n",
+ dpf(("R %d %s: %x.%d.%d.%d.%d.%d.%d flags %d, packet %p\n",
np->header.serial, packetType, ntohl(host), ntohs(port), np->header.serviceId,
np->header.epoch, np->header.cid, np->header.callNumber,
np->header.seq, np->header.flags, np));
addr.sin_family = AF_INET;
addr.sin_port = port;
addr.sin_addr.s_addr = host;
+ memset(&addr.sin_zero, 0, sizeof(addr.sin_zero));
#ifdef STRUCT_SOCKADDR_HAS_SA_LEN
addr.sin_len = sizeof(addr);
-#endif /* AFS_OSF_ENV */
+#endif
drop = (*rx_justReceived) (np, &addr);
/* drop packet if return value is non-zero */
if (drop)
don't abort an abort. */
if (!conn) {
if (unknownService && (np->header.type != RX_PACKET_TYPE_ABORT))
- rxi_SendRawAbort(socket, host, port, RX_INVALID_OPERATION,
+ rxi_SendRawAbort(socket, host, port, 0, RX_INVALID_OPERATION,
np, 0);
return np;
}
(void)rxi_SendAck(call, 0, np->header.serial,
RX_ACK_PING_RESPONSE, 1);
}
- np = rxi_ReceiveAckPacket(call, np, 1);
+ np = rxi_ReceiveAckPacket(call, np, 1, &invalid);
break;
case RX_PACKET_TYPE_ABORT: {
/* An abort packet: reset the call, passing the error up to the user. */
putConnection(conn);
return np; /* xmitting; drop packet */
}
- case RX_PACKET_TYPE_BUSY: {
- struct clock busyTime;
- clock_NewTime();
- clock_GetTime(&busyTime);
-
- MUTEX_EXIT(&call->lock);
-
- MUTEX_ENTER(&conn->conn_call_lock);
- MUTEX_ENTER(&call->lock);
- conn->lastBusy[call->channel] = busyTime.sec;
- call->flags |= RX_CALL_PEER_BUSY;
- MUTEX_EXIT(&call->lock);
- MUTEX_EXIT(&conn->conn_call_lock);
-
- putConnection(conn);
- return np;
- }
+ case RX_PACKET_TYPE_BUSY:
+ /* Mostly ignore BUSY packets. We will update lastReceiveTime below,
+ * so we don't think the endpoint is completely dead, but otherwise
+ * just act as if we never saw anything. If all we get are BUSY packets
+ * back, then we will eventually error out with RX_CALL_TIMEOUT if the
+ * connection is configured with idle/hard timeouts. */
+ break;
case RX_PACKET_TYPE_ACKALL:
/* All packets acknowledged, so we can drop all packets previously
np = rxi_SendCallAbort(call, np, 1, 0);
break;
};
- /* Note when this last legitimate packet was received, for keep-alive
- * processing. Note, we delay getting the time until now in the hope that
- * the packet will be delivered to the user before any get time is required
- * (if not, then the time won't actually be re-evaluated here). */
- call->lastReceiveTime = clock_Sec();
- /* we've received a legit packet, so the channel is not busy */
- call->flags &= ~RX_CALL_PEER_BUSY;
+ if (invalid) {
+ if (rx_stats_active)
+ rx_atomic_inc(&rx_stats.spuriousPacketsRead);
+ } else {
+ /*
+ * Note when this last legitimate packet was received, for keep-alive
+ * processing.
+ */
+ call->lastReceiveTime = clock_Sec();
+ }
MUTEX_EXIT(&call->lock);
putConnection(conn);
return np;
}
}
+/*
+ * Event handler function for connection-specific events for checking
+ * reachability. Also called directly from main code with |event| == NULL
+ * in order to trigger the initial reachability check.
+ *
+ * When |event| == NULL, must be called with the connection data lock held,
+ * but returns with the lock unlocked.
+ */
static void
rxi_CheckReachEvent(struct rxevent *event, void *arg1, void *arg2, int dummy)
{
struct clock when, now;
int i, waiting;
- MUTEX_ENTER(&conn->conn_data_lock);
+ if (event != NULL)
+ MUTEX_ENTER(&conn->conn_data_lock);
+ else
+ MUTEX_ASSERT(&conn->conn_data_lock);
- if (event)
+ if (event != NULL && event == conn->checkReachEvent)
rxevent_Put(&conn->checkReachEvent);
-
waiting = conn->flags & RX_CONN_ATTACHWAIT;
- if (event) {
- putConnection(conn);
- }
MUTEX_EXIT(&conn->conn_data_lock);
if (waiting) {
when.sec += RX_CHECKREACH_TIMEOUT;
MUTEX_ENTER(&conn->conn_data_lock);
if (!conn->checkReachEvent) {
- MUTEX_ENTER(&rx_refcnt_mutex);
- conn->refCount++;
- MUTEX_EXIT(&rx_refcnt_mutex);
+ rx_GetConnection(conn);
conn->checkReachEvent = rxevent_Post(&when, &now,
rxi_CheckReachEvent, conn,
NULL, 0);
MUTEX_EXIT(&conn->conn_data_lock);
}
}
+ /* If fired as an event handler, drop our refcount on the connection. */
+ if (event != NULL)
+ putConnection(conn);
}
static int
return 1;
}
conn->flags |= RX_CONN_ATTACHWAIT;
- MUTEX_EXIT(&conn->conn_data_lock);
- if (!conn->checkReachEvent)
+ if (conn->checkReachEvent == NULL) {
+ /* rxi_CheckReachEvent(NULL, ...) will drop the lock. */
rxi_CheckReachEvent(NULL, conn, call, 0);
+ } else {
+ MUTEX_EXIT(&conn->conn_data_lock);
+ }
return 1;
}
static void
TryAttach(struct rx_call *acall, osi_socket socket,
int *tnop, struct rx_call **newcallp,
- int reachOverride)
+ int reachOverride, int istack)
{
struct rx_connection *conn = acall->conn;
* may not any proc available
*/
} else {
- rxi_ChallengeOn(acall->conn);
+ int code;
+ code = rxi_ChallengeOn(acall->conn);
+ if (code) {
+ /*
+ * Ideally we would rxi_ConnectionError here, but doing that is
+ * difficult, because some callers may have locked 'call',
+ * _and_ another call on the same conn. So we cannot
+ * rxi_ConnectionError, since that needs to lock every call on
+ * the conn. But we can at least abort the call we have.
+ */
+ rxi_CallError(acall, code);
+ rxi_SendCallAbort(acall, NULL, istack, 0);
+ }
}
}
}
if (rx_stats_active)
rx_atomic_inc(&rx_stats.noPacketBuffersOnRead);
rxi_calltrace(RX_TRACE_DROP, call);
- dpf(("packet %"AFS_PTR_FMT" dropped on receipt - quota problems\n", np));
+ dpf(("packet %p dropped on receipt - quota problems\n", np));
/* We used to clear the receive queue here, in an attempt to free
* packets. However this is unsafe if the queue has received a
* soft ACK for the final packet */
&& opr_queue_First(&call->rq, struct rx_packet, entry)->header.seq == seq) {
if (rx_stats_active)
rx_atomic_inc(&rx_stats.dupPacketsRead);
- dpf(("packet %"AFS_PTR_FMT" dropped on receipt - duplicate\n", np));
+ dpf(("packet %p dropped on receipt - duplicate\n", np));
rxi_CancelDelayedAckEvent(call);
np = rxi_SendAck(call, np, serial, RX_ACK_DUPLICATE, istack);
ackNeeded = 0;
if (call->arrivalProc) {
(*call->arrivalProc) (call, call->arrivalProcHandle,
call->arrivalProcArg);
- call->arrivalProc = (void (*)())0;
+ call->arrivalProc = NULL;
}
/* Update last packet received */
* server thread is available, this thread becomes a server
* thread and the server thread becomes a listener thread. */
if (isFirst) {
- TryAttach(call, socket, tnop, newcallp, 0);
+ TryAttach(call, socket, tnop, newcallp, 0, istack);
}
}
/* This is not the expected next packet. */
}
static void
-rxi_UpdatePeerReach(struct rx_connection *conn, struct rx_call *acall)
+rxi_UpdatePeerReach(struct rx_connection *conn, struct rx_call *acall,
+ int istack)
{
struct rx_peer *peer = conn->peer;
if (call != acall)
MUTEX_ENTER(&call->lock);
/* tnop can be null if newcallp is null */
- TryAttach(call, (osi_socket) - 1, NULL, NULL, 1);
+ TryAttach(call, (osi_socket) - 1, NULL, NULL, 1, istack);
if (call != acall)
MUTEX_EXIT(&call->lock);
}
}
#endif
+static_inline int
+ack_is_valid(struct rx_call *call, afs_uint32 first, afs_uint32 prev)
+{
+ if (first < call->tfirst) {
+ /*
+ * The peer indicated that the window went backwards. That's not
+ * allowed; the window can only move forwards.
+ */
+ return 0;
+ }
+
+ if (first == call->tfirst && prev < call->tprev) {
+ /*
+ * The peer said the last DATA packet it received was seq X, but it
+ * already told us before that it had received data after X. This is
+ * probably just an out-of-order ACK, and so we can ignore it.
+ */
+ if (prev >= call->tfirst + call->twind) {
+ /*
+ * Some peers (OpenAFS libafs before 1.6.23) mistakenly set the
+ * previousPacket field to a serial number, not a sequence number.
+ * The sequence number the peer told us about is further than our
+ * transmit window, so it cannot possibly be correct; it's probably
+ * actually a serial number. Don't ignore packets based on this;
+ * the previousPacket information is not accurate.
+ */
+ return 1;
+ }
+
+ return 0;
+ }
+
+ /* Otherwise, the ack looks valid. */
+ return 1;
+}
/* The real smarts of the whole thing. */
static struct rx_packet *
rxi_ReceiveAckPacket(struct rx_call *call, struct rx_packet *np,
- int istack)
+ int istack, int *a_invalid)
{
struct rx_ackPacket *ap;
int nAcks;
int pktsize = 0; /* Set if we need to update the peer mtu */
int conn_data_locked = 0;
+ *a_invalid = 1;
+
if (rx_stats_active)
rx_atomic_inc(&rx_stats.ackPacketsRead);
ap = (struct rx_ackPacket *)rx_DataOf(np);
prev = ntohl(ap->previousPacket);
serial = ntohl(ap->serial);
- /*
- * Ignore ack packets received out of order while protecting
- * against peers that set the previousPacket field to a packet
- * serial number instead of a sequence number.
- */
- if (first < call->tfirst ||
- (first == call->tfirst && prev < call->tprev && prev < call->tfirst
- + call->twind)) {
+ if (!ack_is_valid(call, first, prev)) {
return np;
}
call->tprev = prev;
+ *a_invalid = 0;
+
if (np->header.flags & RX_SLOW_START_OK) {
call->flags |= RX_CALL_SLOW_START_OK;
}
if (ap->reason == RX_ACK_PING_RESPONSE)
- rxi_UpdatePeerReach(conn, call);
+ rxi_UpdatePeerReach(conn, call, istack);
if (conn->lastPacketSizeSeq) {
MUTEX_ENTER(&conn->conn_data_lock);
* but we are clearly receiving.
*/
if (!peer->maxPacketSize)
- peer->maxPacketSize = RX_MIN_PACKET_SIZE+RX_IPUDP_SIZE;
+ peer->maxPacketSize = RX_MIN_PACKET_SIZE - RX_HEADER_SIZE;
if (pktsize > peer->maxPacketSize) {
peer->maxPacketSize = pktsize;
- if ((pktsize-RX_IPUDP_SIZE > peer->ifMTU)) {
- peer->ifMTU=pktsize-RX_IPUDP_SIZE;
+ if ((pktsize + RX_HEADER_SIZE > peer->ifMTU)) {
+ peer->ifMTU = pktsize + RX_HEADER_SIZE;
peer->natMTU = rxi_AdjustIfMTU(peer->ifMTU);
rxi_ScheduleGrowMTUEvent(call, 1);
}
#ifdef RX_ENABLE_LOCKS
/* XXX Hack. Because we have to release the global call lock when sending
- * packets (osi_NetSend) we drop all acks while we're traversing the tq
+ * packets (rxi_NetSend) we drop all acks while we're traversing the tq
* in rxi_Start sending packets out because packets may move to the
* freePacketQueue as result of being here! So we drop these packets until
* we're safely out of the traversing. Really ugly!
* queue - they're not addressed by the contents of this ACK packet.
*/
- /* If the window has been extended by this acknowledge packet,
- * then wakeup a sender waiting in alloc for window space, or try
- * sending packets now, if he's been sitting on packets due to
- * lack of window space */
- if (call->tnext < (call->tfirst + call->twind)) {
-#ifdef RX_ENABLE_LOCKS
- CV_SIGNAL(&call->cv_twind);
-#else
- if (call->flags & RX_CALL_WAIT_WINDOW_ALLOC) {
- call->flags &= ~RX_CALL_WAIT_WINDOW_ALLOC;
- osi_rxWakeup(&call->twind);
- }
-#endif
- if (call->flags & RX_CALL_WAIT_WINDOW_SEND) {
- call->flags &= ~RX_CALL_WAIT_WINDOW_SEND;
- }
- }
-
/* if the ack packet has a receivelen field hanging off it,
* update our state */
if (np->length >= rx_AckDataSize(ap->nAcks) + 2 * sizeof(afs_int32)) {
rx_packetread(np, rx_AckDataSize(ap->nAcks) + (int)sizeof(afs_int32),
(int)sizeof(afs_int32), &tSize);
tSize = (afs_uint32) ntohl(tSize);
+ if (tSize > RX_MAX_PACKET_SIZE)
+ tSize = RX_MAX_PACKET_SIZE;
+ if (tSize < RX_MIN_PACKET_SIZE)
+ tSize = RX_MIN_PACKET_SIZE;
peer->natMTU = rxi_AdjustIfMTU(MIN(tSize, peer->ifMTU));
/* Get the maximum packet size to send to this peer */
rx_packetread(np, rx_AckDataSize(ap->nAcks), (int)sizeof(afs_int32),
&tSize);
tSize = (afs_uint32) ntohl(tSize);
+ if (tSize > RX_MAX_PACKET_SIZE)
+ tSize = RX_MAX_PACKET_SIZE;
+ if (tSize < RX_MIN_PACKET_SIZE)
+ tSize = RX_MIN_PACKET_SIZE;
tSize = (afs_uint32) MIN(tSize, rx_MyMaxSendSize);
tSize = rxi_AdjustMaxMTU(peer->natMTU, tSize);
rx_AckDataSize(ap->nAcks) + 2 * (int)sizeof(afs_int32),
(int)sizeof(afs_int32), &tSize);
tSize = (afs_uint32) ntohl(tSize); /* peer's receive window, if it's */
+ if (tSize == 0)
+ tSize = 1;
+ if (tSize >= rx_maxSendWindow)
+ tSize = rx_maxSendWindow;
if (tSize < call->twind) { /* smaller than our send */
call->twind = tSize; /* window, we must send less... */
call->ssthresh = MIN(call->twind, call->ssthresh);
rx_AckDataSize(ap->nAcks) + 2 * (int)sizeof(afs_int32),
sizeof(afs_int32), &tSize);
tSize = (afs_uint32) ntohl(tSize);
+ if (tSize == 0)
+ tSize = 1;
+ if (tSize >= rx_maxSendWindow)
+ tSize = rx_maxSendWindow;
/*
* As of AFS 3.5 we set the send window to match the receive window.
*/
call->MTU = OLD_MAX_PACKET_SIZE;
}
+ /* If the window has been extended by this acknowledge packet,
+ * then wakeup a sender waiting in alloc for window space, or try
+ * sending packets now, if he's been sitting on packets due to
+ * lack of window space */
+ if (call->tnext < (call->tfirst + call->twind)) {
+#ifdef RX_ENABLE_LOCKS
+ CV_SIGNAL(&call->cv_twind);
+#else
+ if (call->flags & RX_CALL_WAIT_WINDOW_ALLOC) {
+ call->flags &= ~RX_CALL_WAIT_WINDOW_ALLOC;
+ osi_rxWakeup(&call->twind);
+ }
+#endif
+ if (call->flags & RX_CALL_WAIT_WINDOW_SEND) {
+ call->flags &= ~RX_CALL_WAIT_WINDOW_SEND;
+ }
+ }
+
if (nNacked) {
/*
* Calculate how many datagrams were successfully received after
rxi_SendConnectionAbortLater(struct rx_connection *conn, int msec)
{
struct clock when, now;
+
+ MUTEX_ASSERT(&conn->conn_data_lock);
if (!conn->error) {
return;
}
clock_GetTime(&now);
when = now;
clock_Addmsec(&when, msec);
+ rx_GetConnection(conn);
conn->delayedAbortEvent =
rxevent_Post(&when, &now, rxi_SendDelayedConnAbort, conn, NULL, 0);
}
if (RXS_CheckAuthentication(conn->securityObject, conn) == 0)
return np;
+ if (!conn->securityChallengeSent) {
+ /* We've never sent out a challenge for this connection, so this
+ * response cannot possibly be correct; ignore it. This can happen
+ * if we sent a challenge to the client, then we were restarted, and
+ * then the client sent us a response. If we ignore the response, the
+ * client will eventually resend a data packet, causing us to send a
+ * new challenge and the client to send a new response. */
+ return np;
+ }
+
/* Otherwise, have the security object evaluate the response packet */
error = RXS_CheckResponse(conn->securityObject, conn, np);
if (error) {
* some calls went into attach-wait while we were waiting
* for authentication..
*/
- rxi_UpdatePeerReach(conn, NULL);
+ rxi_UpdatePeerReach(conn, NULL, istack);
}
return np;
}
if (opr_queue_IsOnQueue(&call->entry)) {
opr_queue_Remove(&call->entry);
}
+ CLEAR_CALL_QUEUE_LOCK(call);
}
call->state = RX_STATE_ACTIVE;
call->app.mode = RX_MODE_RECEIVING;
call->flags |= RX_CALL_ACKALL_SENT;
}
+/*
+ * Event handler for per-call delayed acks.
+ * Also called synchronously, with |event| == NULL, to send a "delayed" ack
+ * immediately.
+ */
static void
rxi_SendDelayedAck(struct rxevent *event, void *arg1, void *unused1,
int unused2)
MUTEX_ENTER(&call->lock);
if (event == call->delayedAckEvent)
rxevent_Put(&call->delayedAckEvent);
- CALL_RELE(call, RX_CALL_REFCOUNT_DELAY);
}
(void)rxi_SendAck(call, 0, 0, RX_ACK_DELAY, 0);
if (event)
rxevent_Put(&call->delayedAckEvent);
(void)rxi_SendAck(call, 0, 0, RX_ACK_DELAY, 0);
#endif /* RX_ENABLE_LOCKS */
+ /* Release the call reference for the event that fired. */
+ if (event)
+ CALL_RELE(call, RX_CALL_REFCOUNT_DELAY);
}
#ifdef RX_ENABLE_LOCKS
#ifdef RXDEBUG_PACKET
call->rqc -= count;
if ( call->rqc != 0 )
- dpf(("rxi_ClearReceiveQueue call %"AFS_PTR_FMT" rqc %u != 0\n", call, call->rqc));
+ dpf(("rxi_ClearReceiveQueue call %p rqc %u != 0\n", call, call->rqc));
#endif
call->flags &= ~(RX_CALL_RECEIVE_DONE | RX_CALL_HAVE_LAST);
}
rxi_SendCallAbort(struct rx_call *call, struct rx_packet *packet,
int istack, int force)
{
- afs_int32 error, cerror;
+ afs_int32 error;
struct clock when, now;
if (!call->error)
return packet;
- switch (call->error) {
- case RX_CALL_IDLE:
- case RX_CALL_BUSY:
- cerror = RX_CALL_TIMEOUT;
- break;
- default:
- cerror = call->error;
- }
-
/* Clients should never delay abort messages */
if (rx_IsClientConn(call->conn))
force = 1;
- if (call->abortCode != cerror) {
- call->abortCode = cerror;
+ /*
+ * An opcode that has been deprecated or has yet to be implemented is not
+ * a misbehavior of the client. Do not punish the client by introducing
+ * delays.
+ */
+ if (call->error == RXGEN_OPCODE) {
+ force = 1;
+ } else if (call->abortCode != call->error) {
+ call->abortCode = call->error;
call->abortCount = 0;
}
if (force || rxi_callAbortThreshhold == 0
|| call->abortCount < rxi_callAbortThreshhold) {
rxi_CancelDelayedAbortEvent(call);
- error = htonl(cerror);
- call->abortCount++;
+ error = htonl(call->error);
+ if (!force)
+ call->abortCount++;
packet =
rxi_SendSpecial(call, call->conn, packet, RX_PACKET_TYPE_ABORT,
(char *)&error, sizeof(error), istack);
static void
rxi_CancelDelayedAbortEvent(struct rx_call *call)
{
- if (call->delayedAbortEvent) {
- rxevent_Cancel(&call->delayedAbortEvent);
+ MUTEX_ASSERT(&call->lock);
+ if (rxevent_Cancel(&call->delayedAbortEvent))
CALL_RELE(call, RX_CALL_REFCOUNT_ABORT);
- }
}
/* Send an abort packet for the specified connection. Packet is an
if (force || rxi_connAbortThreshhold == 0
|| conn->abortCount < rxi_connAbortThreshhold) {
- rxevent_Cancel(&conn->delayedAbortEvent);
+ if (rxevent_Cancel(&conn->delayedAbortEvent))
+ putConnection(conn);
error = htonl(conn->error);
conn->abortCount++;
MUTEX_EXIT(&conn->conn_data_lock);
if (error) {
int i;
- dpf(("rxi_ConnectionError conn %"AFS_PTR_FMT" error %d\n", conn, error));
+ dpf(("rxi_ConnectionError conn %p error %d\n", conn, error));
MUTEX_ENTER(&conn->conn_data_lock);
- rxevent_Cancel(&conn->challengeEvent);
- rxevent_Cancel(&conn->natKeepAliveEvent);
- if (conn->checkReachEvent) {
- rxevent_Cancel(&conn->checkReachEvent);
+ if (rxevent_Cancel(&conn->challengeEvent))
+ putConnection(conn);
+ if (rxevent_Cancel(&conn->natKeepAliveEvent))
+ putConnection(conn);
+ if (rxevent_Cancel(&conn->checkReachEvent)) {
conn->flags &= ~(RX_CONN_ATTACHWAIT|RX_CONN_NAT_PING);
putConnection(conn);
}
rxi_CallError(struct rx_call *call, afs_int32 error)
{
MUTEX_ASSERT(&call->lock);
- dpf(("rxi_CallError call %"AFS_PTR_FMT" error %d call->error %d\n", call, error, call->error));
+ dpf(("rxi_CallError call %p error %d call->error %d\n", call, error, call->error));
if (call->error)
error = call->error;
struct rx_packet *packet;
MUTEX_ASSERT(&call->lock);
- dpf(("rxi_ResetCall(call %"AFS_PTR_FMT", newcall %d)\n", call, newcall));
+ dpf(("rxi_ResetCall(call %p, newcall %d)\n", call, newcall));
/* Notify anyone who is waiting for asynchronous packet arrival */
if (call->arrivalProc) {
(*call->arrivalProc) (call, call->arrivalProcHandle,
call->arrivalProcArg);
- call->arrivalProc = (void (*)())0;
+ call->arrivalProc = NULL;
}
rxi_ClearTransmitQueue(call, 1);
if (call->tqWaiters || (flags & RX_CALL_TQ_WAIT)) {
- dpf(("rcall %"AFS_PTR_FMT" has %d waiters and flags %d\n", call, call->tqWaiters, call->flags));
+ dpf(("rcall %p has %d waiters and flags %d\n", call, call->tqWaiters, call->flags));
}
call->flags = 0;
- if (!newcall && (flags & RX_CALL_PEER_BUSY)) {
- /* The call channel is still busy; resetting the call doesn't change
- * that. However, if 'newcall' is set, we are processing a call
- * structure that has either been recycled from the free list, or has
- * been newly allocated. So, RX_CALL_PEER_BUSY is not relevant if
- * 'newcall' is set, since it describes a completely different call
- * channel which we do not care about. */
- call->flags |= RX_CALL_PEER_BUSY;
- }
-
rxi_ClearReceiveQueue(call);
/* why init the queue if you just emptied it? queue_Init(&call->rq); */
int reason; Reason an acknowledge was prompted
*/
+#define RX_ZEROS 1024
+static char rx_zeros[RX_ZEROS];
+
struct rx_packet *
rxi_SendAck(struct rx_call *call,
struct rx_packet *optionalPacket, int serial, int reason,
*/
if (call->conn->peer->maxPacketSize &&
(call->conn->peer->maxPacketSize < OLD_MAX_PACKET_SIZE
- +RX_IPUDP_SIZE))
+ - RX_HEADER_SIZE))
padbytes = call->conn->peer->maxPacketSize+16;
else
padbytes = call->conn->peer->maxMTU + 128;
ap->nAcks = offset;
p->length = rx_AckDataSize(offset) + 4 * sizeof(afs_int32);
+ /* Must zero the 3 octets that rx_AckDataSize skips at the end of the
+ * ACK list.
+ */
+ rx_packetwrite(p, rx_AckDataSize(offset) - 3, 3, rx_zeros);
+
/* these are new for AFS 3.3 */
templ = rxi_AdjustMaxMTU(call->conn->peer->ifMTU, rx_maxReceiveSize);
templ = htonl(templ);
rx_packetwrite(p, rx_AckDataSize(offset) + 3 * sizeof(afs_int32),
sizeof(afs_int32), &templ);
+ p->length = rx_AckDataSize(offset) + 4 * sizeof(afs_int32);
+
p->header.serviceId = call->conn->serviceId;
p->header.cid = (call->conn->cid | call->channel);
p->header.callNumber = *call->callNumber;
p->header.epoch = call->conn->epoch;
p->header.type = RX_PACKET_TYPE_ACK;
p->header.flags = RX_SLOW_START_OK;
- if (reason == RX_ACK_PING) {
+ if (reason == RX_ACK_PING)
p->header.flags |= RX_REQUEST_ACK;
- if (padbytes) {
- p->length = padbytes +
- rx_AckDataSize(call->rwind) + 4 * sizeof(afs_int32);
- while (padbytes--)
- /* not fast but we can potentially use this if truncated
- * fragments are delivered to figure out the mtu.
- */
- rx_packetwrite(p, rx_AckDataSize(offset) + 4 *
- sizeof(afs_int32), sizeof(afs_int32),
- &padbytes);
+ while (padbytes > 0) {
+ if (padbytes > RX_ZEROS) {
+ rx_packetwrite(p, p->length, RX_ZEROS, rx_zeros);
+ p->length += RX_ZEROS;
+ padbytes -= RX_ZEROS;
+ } else {
+ rx_packetwrite(p, p->length, padbytes, rx_zeros);
+ p->length += padbytes;
+ padbytes = 0;
}
}
+
if (call->conn->type == RX_CLIENT_CONNECTION)
p->header.flags |= RX_CLIENT_INITIATED;
* processing), and for the connection (so that we can discover
* idle connections) */
conn->lastSendTime = call->lastSendTime = clock_Sec();
- /* Let a set of retransmits trigger an idle timeout */
- if (!xmit->resending)
- call->lastSendData = call->lastSendTime;
}
/* When sending packets we need to follow these rules:
/* Make sure that the event pointer is removed from the call
* structure, since there is no longer a per-call retransmission
* event pending. */
- if (event == call->resendEvent) {
- CALL_RELE(call, RX_CALL_REFCOUNT_RESEND);
+ if (event == call->resendEvent)
rxevent_Put(&call->resendEvent);
- }
rxi_CheckPeerDead(call);
- if (rxi_busyChannelError && (call->flags & RX_CALL_PEER_BUSY)) {
- rxi_CheckBusy(call);
- }
-
if (opr_queue_IsEmpty(&call->tq)) {
/* Nothing to do. This means that we've been raced, and that an
* ACK has come in between when we were triggered, and when we
rxi_Start(call, istack);
out:
+ CALL_RELE(call, RX_CALL_REFCOUNT_RESEND);
MUTEX_EXIT(&call->lock);
}
nXmitPackets, istack);
goto restart;
}
- dpf(("call %d xmit packet %"AFS_PTR_FMT"\n",
+ dpf(("call %d xmit packet %p\n",
*(call->callNumber), p));
call->xmitList[nXmitPackets++] = p;
}
rxi_Send(struct rx_call *call, struct rx_packet *p,
int istack)
{
+ int code;
struct rx_connection *conn = call->conn;
/* Stamp each packet with the user supplied status */
/* Allow the security object controlling this call's security to
* make any last-minute changes to the packet */
- RXS_SendPacket(conn->securityObject, call, p);
+ code = RXS_SendPacket(conn->securityObject, call, p);
+ if (code) {
+ MUTEX_EXIT(&call->lock);
+ CALL_HOLD(call, RX_CALL_REFCOUNT_SEND);
+ rxi_ConnectionError(conn, code);
+ CALL_RELE(call, RX_CALL_REFCOUNT_SEND);
+ MUTEX_ENTER(&call->lock);
+ return;
+ }
/* Since we're about to send SOME sort of packet to the peer, it's
* safe to nuke any scheduled end-of-packets ack */
(p->length <= (rx_AckDataSize(call->rwind) + 4 * sizeof(afs_int32))))
{
conn->lastSendTime = call->lastSendTime = clock_Sec();
- /* Don't count keepalive ping/acks here, so idleness can be tracked. */
- if ((p->header.type != RX_PACKET_TYPE_ACK) ||
- ((((struct rx_ackPacket *)rx_DataOf(p))->reason != RX_ACK_PING) &&
- (((struct rx_ackPacket *)rx_DataOf(p))->reason !=
- RX_ACK_PING_RESPONSE)))
- call->lastSendData = call->lastSendTime;
}
}
* number of seconds. */
if (now > (call->lastReceiveTime + deadTime)) {
if (call->state == RX_STATE_ACTIVE) {
-#ifdef AFS_ADAPT_PMTU
-# if defined(KERNEL) && defined(AFS_SUN5_ENV)
- ire_t *ire;
-# if defined(AFS_SUN510_ENV) && defined(GLOBAL_NETSTACKID)
- netstack_t *ns = netstack_find_by_stackid(GLOBAL_NETSTACKID);
- ip_stack_t *ipst = ns->netstack_ip;
-# endif
- ire = ire_cache_lookup(conn->peer->host
-# if defined(AFS_SUN510_ENV) && defined(ALL_ZONES)
- , ALL_ZONES
-# if defined(ICL_3_ARG) || defined(GLOBAL_NETSTACKID)
- , NULL
-# if defined(GLOBAL_NETSTACKID)
- , ipst
-# endif
-# endif
-# endif
- );
-
- if (ire && ire->ire_max_frag > 0)
- rxi_SetPeerMtu(NULL, conn->peer->host, 0,
- ire->ire_max_frag);
-# if defined(GLOBAL_NETSTACKID)
- netstack_rele(ns);
-# endif
-# endif
-#endif /* AFS_ADAPT_PMTU */
cerror = RX_CALL_DEAD;
goto mtuout;
} else {
* attached process can die reasonably gracefully. */
}
- if (conn->idleDeadDetection) {
- if (conn->idleDeadTime) {
- idleDeadTime = conn->idleDeadTime + fudgeFactor;
- }
-
- if (idleDeadTime) {
- /* see if we have a non-activity timeout */
- if (call->startWait && ((call->startWait + idleDeadTime) < now) &&
- (call->flags & RX_CALL_READER_WAIT)) {
- if (call->state == RX_STATE_ACTIVE) {
- cerror = RX_CALL_TIMEOUT;
- goto mtuout;
- }
- }
+ if (conn->idleDeadTime) {
+ idleDeadTime = conn->idleDeadTime + fudgeFactor;
+ }
- if (call->lastSendData && ((call->lastSendData + idleDeadTime) < now)) {
- if (call->state == RX_STATE_ACTIVE) {
- cerror = conn->service ? conn->service->idleDeadErr : RX_CALL_IDLE;
- idle_timeout = 1;
- goto mtuout;
- }
- }
- }
+ if (idleDeadTime) {
+ /* see if we have a non-activity timeout */
+ if (call->startWait && ((call->startWait + idleDeadTime) < now)) {
+ if (call->state == RX_STATE_ACTIVE) {
+ cerror = RX_CALL_TIMEOUT;
+ goto mtuout;
+ }
+ }
}
if (conn->hardDeadTime) {
call->lastReceiveTime) {
int oldMTU = conn->peer->ifMTU;
- /* if we thought we could send more, perhaps things got worse */
- if (conn->peer->maxPacketSize > conn->lastPacketSize)
- /* maxpacketsize will be cleared in rxi_SetPeerMtu */
- newmtu = MAX(conn->peer->maxPacketSize-RX_IPUDP_SIZE,
- conn->lastPacketSize-(128+RX_IPUDP_SIZE));
+ /* If we thought we could send more, perhaps things got worse.
+ * Shrink by 128 bytes and try again. */
+ if (conn->peer->maxPacketSize < conn->lastPacketSize)
+ /* maxPacketSize will be cleared in rxi_SetPeerMtu */
+ newmtu = MAX(conn->peer->maxPacketSize + RX_HEADER_SIZE,
+ conn->lastPacketSize - 128 + RX_HEADER_SIZE);
else
- newmtu = conn->lastPacketSize-(128+RX_IPUDP_SIZE);
+ newmtu = conn->lastPacketSize - 128 + RX_HEADER_SIZE;
/* minimum capped in SetPeerMtu */
rxi_SetPeerMtu(conn->peer, 0, 0, newmtu);
/* clean up */
- conn->lastPacketSize = 0;
+ conn->lastPacketSize = conn->lastPacketSizeSeq = 0;
/* needed so ResetCall doesn't clobber us. */
call->MTU = conn->peer->ifMTU;
struct sockaddr_in taddr;
char *tp;
char a[1] = { 0 };
+ int resched = 0;
struct iovec tmpiov[2];
osi_socket socket =
(conn->type ==
taddr.sin_family = AF_INET;
taddr.sin_port = rx_PortOf(rx_PeerOf(conn));
taddr.sin_addr.s_addr = rx_HostOf(rx_PeerOf(conn));
+ memset(&taddr.sin_zero, 0, sizeof(taddr.sin_zero));
#ifdef STRUCT_SOCKADDR_HAS_SA_LEN
taddr.sin_len = sizeof(struct sockaddr_in);
#endif
tmpiov[0].iov_base = tbuffer;
tmpiov[0].iov_len = 1 + sizeof(struct rx_header);
- osi_NetSend(socket, &taddr, tmpiov, 1, 1 + sizeof(struct rx_header), 1);
+ rxi_NetSend(socket, &taddr, tmpiov, 1, 1 + sizeof(struct rx_header), 1);
MUTEX_ENTER(&conn->conn_data_lock);
+ /* We ran, so the handle is no longer needed to try to cancel ourselves. */
+ if (event == conn->natKeepAliveEvent)
+ rxevent_Put(&conn->natKeepAliveEvent);
MUTEX_ENTER(&rx_refcnt_mutex);
/* Only reschedule ourselves if the connection would not be destroyed */
- if (conn->refCount <= 1) {
- rxevent_Put(&conn->natKeepAliveEvent);
- MUTEX_EXIT(&rx_refcnt_mutex);
- MUTEX_EXIT(&conn->conn_data_lock);
- rx_DestroyConnection(conn); /* drop the reference for this */
- } else {
- conn->refCount--; /* drop the reference for this */
- MUTEX_EXIT(&rx_refcnt_mutex);
- rxevent_Put(&conn->natKeepAliveEvent);
- rxi_ScheduleNatKeepAliveEvent(conn);
- MUTEX_EXIT(&conn->conn_data_lock);
+ if (conn->refCount > 1)
+ resched = 1;
+ if (conn->refCount <= 0) {
+#ifdef RX_REFCOUNT_CHECK
+ osi_Assert(conn->refCount == 0);
+#endif
+ if (rx_stats_active) {
+ MUTEX_ENTER(&rx_stats_mutex);
+ rxi_lowConnRefCount++;
+ MUTEX_EXIT(&rx_stats_mutex);
+ }
}
+ MUTEX_EXIT(&rx_refcnt_mutex);
+ if (resched)
+ rxi_ScheduleNatKeepAliveEvent(conn);
+ MUTEX_EXIT(&conn->conn_data_lock);
+ putConnection(conn);
}
static void
rxi_ScheduleNatKeepAliveEvent(struct rx_connection *conn)
{
+ MUTEX_ASSERT(&conn->conn_data_lock);
if (!conn->natKeepAliveEvent && conn->secondsUntilNatPing) {
struct clock when, now;
clock_GetTime(&now);
when = now;
when.sec += conn->secondsUntilNatPing;
- MUTEX_ENTER(&rx_refcnt_mutex);
- conn->refCount++; /* hold a reference for this */
- MUTEX_EXIT(&rx_refcnt_mutex);
+ rx_GetConnection(conn);
conn->natKeepAliveEvent =
rxevent_Post(&when, &now, rxi_NatKeepAliveEvent, conn, NULL, 0);
}
struct rx_connection *conn;
afs_uint32 now;
- CALL_RELE(call, RX_CALL_REFCOUNT_ALIVE);
MUTEX_ENTER(&call->lock);
if (event == call->keepAliveEvent)
if (rxi_CheckCall(call, 0)) {
MUTEX_EXIT(&call->lock);
+ CALL_RELE(call, RX_CALL_REFCOUNT_ALIVE);
return;
}
/* Don't try to keep alive dallying calls */
if (call->state == RX_STATE_DALLY) {
MUTEX_EXIT(&call->lock);
+ CALL_RELE(call, RX_CALL_REFCOUNT_ALIVE);
return;
}
}
rxi_ScheduleKeepAliveEvent(call);
MUTEX_EXIT(&call->lock);
+ CALL_RELE(call, RX_CALL_REFCOUNT_ALIVE);
}
/* Does what's on the nameplate. */
struct rx_call *call = arg1;
struct rx_connection *conn;
- CALL_RELE(call, RX_CALL_REFCOUNT_MTU);
MUTEX_ENTER(&call->lock);
if (event == call->growMTUEvent)
rxevent_Put(&call->growMTUEvent);
- if (rxi_CheckCall(call, 0)) {
- MUTEX_EXIT(&call->lock);
- return;
- }
+ if (rxi_CheckCall(call, 0))
+ goto out;
/* Don't bother with dallying calls */
- if (call->state == RX_STATE_DALLY) {
- MUTEX_EXIT(&call->lock);
- return;
- }
+ if (call->state == RX_STATE_DALLY)
+ goto out;
conn = call->conn;
*/
if ((conn->peer->maxPacketSize != 0) &&
(conn->peer->natMTU < RX_MAX_PACKET_SIZE) &&
- conn->idleDeadDetection)
+ conn->idleDeadTime)
(void)rxi_SendAck(call, NULL, 0, RX_ACK_MTU, 0);
rxi_ScheduleGrowMTUEvent(call, 0);
+out:
MUTEX_EXIT(&call->lock);
+ CALL_RELE(call, RX_CALL_REFCOUNT_MTU);
}
static void
rxi_ScheduleKeepAliveEvent(struct rx_call *call)
{
+ MUTEX_ASSERT(&call->lock);
if (!call->keepAliveEvent) {
struct clock when, now;
clock_GetTime(&now);
static void
rxi_CancelKeepAliveEvent(struct rx_call *call) {
- if (call->keepAliveEvent) {
- rxevent_Cancel(&call->keepAliveEvent);
+ MUTEX_ASSERT(&call->lock);
+ if (rxevent_Cancel(&call->keepAliveEvent))
CALL_RELE(call, RX_CALL_REFCOUNT_ALIVE);
- }
}
static void
rxi_ScheduleGrowMTUEvent(struct rx_call *call, int secs)
{
+ MUTEX_ASSERT(&call->lock);
if (!call->growMTUEvent) {
struct clock when, now;
static void
rxi_CancelGrowMTUEvent(struct rx_call *call)
{
- if (call->growMTUEvent) {
- rxevent_Cancel(&call->growMTUEvent);
+ MUTEX_ASSERT(&call->lock);
+ if (rxevent_Cancel(&call->growMTUEvent))
CALL_RELE(call, RX_CALL_REFCOUNT_MTU);
- }
+}
+
+/*
+ * Increment the counter for the next connection ID, handling overflow.
+ */
+static void
+update_nextCid(void)
+{
+ /* Overflow is technically undefined behavior; avoid it. */
+ if (rx_nextCid > MAX_AFS_INT32 - (1 << RX_CIDSHIFT))
+ rx_nextCid = -1 * ((MAX_AFS_INT32 / RX_CIDSHIFT) * RX_CIDSHIFT);
+ else
+ rx_nextCid += 1 << RX_CIDSHIFT;
}
static void
rxi_ScheduleKeepAliveEvent(call);
}
-void
-rx_KeepAliveOff(struct rx_call *call)
-{
- MUTEX_ENTER(&call->lock);
- rxi_CancelKeepAliveEvent(call);
- MUTEX_EXIT(&call->lock);
-}
-
-void
-rx_KeepAliveOn(struct rx_call *call)
-{
- MUTEX_ENTER(&call->lock);
- rxi_KeepAliveOn(call);
- MUTEX_EXIT(&call->lock);
-}
-
static void
rxi_GrowMTUOn(struct rx_call *call)
{
struct rx_packet *packet;
MUTEX_ENTER(&conn->conn_data_lock);
- rxevent_Put(&conn->delayedAbortEvent);
+ if (event == conn->delayedAbortEvent)
+ rxevent_Put(&conn->delayedAbortEvent);
error = htonl(conn->error);
conn->abortCount++;
MUTEX_EXIT(&conn->conn_data_lock);
sizeof(error), 0);
rxi_FreePacket(packet);
}
+ putConnection(conn);
}
/* This routine is called to send call abort messages
struct rx_packet *packet;
MUTEX_ENTER(&call->lock);
- rxevent_Put(&call->delayedAbortEvent);
+ if (event == call->delayedAbortEvent)
+ rxevent_Put(&call->delayedAbortEvent);
error = htonl(call->error);
call->abortCount++;
packet = rxi_AllocPacket(RX_PACKET_CLASS_SPECIAL);
CALL_RELE(call, RX_CALL_REFCOUNT_ABORT);
}
-/* This routine is called periodically (every RX_AUTH_REQUEST_TIMEOUT
+/*
+ * This routine is called periodically (every RX_AUTH_REQUEST_TIMEOUT
* seconds) to ask the client to authenticate itself. The routine
* issues a challenge to the client, which is obtained from the
- * security object associated with the connection */
+ * security object associated with the connection
+ *
+ * This routine is both an event handler and a function called directly;
+ * when called directly the passed |event| is NULL and the
+ * conn->conn->data>lock must must not be held. Also, when called as an
+ * an event handler, we must putConnection before we exit; but when called
+ * directly (the first challenge), we must NOT putConnection.
+ */
static void
rxi_ChallengeEvent(struct rxevent *event,
void *arg0, void *arg1, int tries)
{
struct rx_connection *conn = arg0;
+ int event_raised = 0; /* assume we were called directly */
- if (event)
+ MUTEX_ENTER(&conn->conn_data_lock);
+ if (event != NULL && event == conn->challengeEvent) {
+ event_raised = 1; /* called as an event */
rxevent_Put(&conn->challengeEvent);
+ }
+ MUTEX_EXIT(&conn->conn_data_lock);
/* If there are no active calls it is not worth re-issuing the
* challenge. If the client issues another call on this connection
* the challenge can be requested at that time.
*/
if (!rxi_HasActiveCalls(conn))
- return;
+ goto done;
if (RXS_CheckAuthentication(conn->securityObject, conn) != 0) {
struct rx_packet *packet;
}
}
MUTEX_EXIT(&conn->conn_call_lock);
- return;
+ goto done;
}
packet = rxi_AllocPacket(RX_PACKET_CLASS_SPECIAL);
if (packet) {
- /* If there's no packet available, do this later. */
- RXS_GetChallenge(conn->securityObject, conn, packet);
- rxi_SendSpecial((struct rx_call *)0, conn, packet,
- RX_PACKET_TYPE_CHALLENGE, NULL, -1, 0);
+ int code;
+ code = RXS_GetChallenge(conn->securityObject, conn, packet);
+ if (code && event_raised) {
+ /*
+ * We can only rxi_ConnectionError the connection if we are
+ * running as an event. Otherwise, the caller may have our call
+ * locked, and so we cannot call rxi_ConnectionError (since it
+ * tries to lock each call in the conn).
+ */
+ rxi_FreePacket(packet);
+ rxi_ConnectionError(conn, code);
+ goto done;
+ }
+ if (code == 0) {
+ /* Only send a challenge packet if we were able to allocate a
+ * packet, and the security layer successfully populated the
+ * challenge. */
+ rxi_SendSpecial((struct rx_call *)0, conn, packet,
+ RX_PACKET_TYPE_CHALLENGE, NULL, -1, 0);
+ conn->securityChallengeSent = 1;
+ }
rxi_FreePacket(packet);
}
clock_GetTime(&now);
when = now;
when.sec += RX_CHALLENGE_TIMEOUT;
- conn->challengeEvent =
- rxevent_Post(&when, &now, rxi_ChallengeEvent, conn, 0,
- (tries - 1));
+ MUTEX_ENTER(&conn->conn_data_lock);
+ /* Only reschedule ourselves if not already pending. */
+ if (conn->challengeEvent == NULL) {
+ rx_GetConnection(conn);
+ conn->challengeEvent =
+ rxevent_Post(&when, &now, rxi_ChallengeEvent, conn, 0,
+ (tries - 1));
+ }
+ MUTEX_EXIT(&conn->conn_data_lock);
}
+ done:
+ if (event_raised)
+ putConnection(conn);
}
/* Call this routine to start requesting the client to authenticate
* itself. This will continue until authentication is established,
* the call times out, or an invalid response is returned. The
* security object associated with the connection is asked to create
- * the challenge at this time. N.B. rxi_ChallengeOff is a macro,
- * defined earlier. */
-static void
+ * the challenge at this time. */
+static int
rxi_ChallengeOn(struct rx_connection *conn)
{
- if (!conn->challengeEvent) {
- RXS_CreateChallenge(conn->securityObject, conn);
+ int start = 0;
+ MUTEX_ENTER(&conn->conn_data_lock);
+ if (!conn->challengeEvent)
+ start = 1;
+ MUTEX_EXIT(&conn->conn_data_lock);
+ if (start) {
+ int code;
+ code = RXS_CreateChallenge(conn->securityObject, conn);
+ if (code) {
+ return code;
+ }
rxi_ChallengeEvent(NULL, conn, 0, RX_CHALLENGE_MAXTRIES);
- };
+ }
+ return 0;
}
return; /* somebody set the clock back, don't count this time. */
clock_Sub(&thisRtt, sentp);
- dpf(("rxi_ComputeRoundTripTime(call=%d packet=%"AFS_PTR_FMT" rttp=%d.%06d sec)\n",
+ dpf(("rxi_ComputeRoundTripTime(call=%d packet=%p rttp=%d.%06d sec)\n",
p->header.callNumber, p, thisRtt.sec, thisRtt.usec));
if (clock_IsZero(&thisRtt)) {
peer->rtt_dev = call->rtt_dev;
peer->rtt = call->rtt;
- dpf(("rxi_ComputeRoundTripTime(call=%d packet=%"AFS_PTR_FMT" rtt=%d ms, srtt=%d ms, rtt_dev=%d ms, timeout=%d.%06d sec)\n",
- p->header.callNumber, p, MSEC(&thisRtt), call->rtt >> 3, call->rtt_dev >> 2, (call->rto.sec), (call->rto.usec)));
+ dpf(("rxi_ComputeRoundTripTime(call=%d packet=%p rtt=%d ms, srtt=%d ms, "
+ "rtt_dev=%d ms, timeout=%d.%06d sec)\n",
+ p->header.callNumber, p, MSEC(&thisRtt), call->rtt >> 3,
+ call->rtt_dev >> 2, (call->rto.sec), (call->rto.usec)));
}
taddr.sin_family = AF_INET;
taddr.sin_port = remotePort;
taddr.sin_addr.s_addr = remoteAddr;
+ memset(&taddr.sin_zero, 0, sizeof(taddr.sin_zero));
#ifdef STRUCT_SOCKADDR_HAS_SA_LEN
taddr.sin_len = sizeof(struct sockaddr_in);
#endif
struct rx_serverQueueEntry *sq;
#endif /* KERNEL */
- if (rx_atomic_test_and_set_bit(&rxinit_status, 0))
+ LOCK_RX_INIT;
+ if (!rxi_IsRunning()) {
+ UNLOCK_RX_INIT;
return; /* Already shutdown. */
-
+ }
+ rx_atomic_set(&rxi_running, 0);
#ifndef KERNEL
rx_port = 0;
#ifndef AFS_PTHREAD_ENV
rxi_StopListener();
#endif /* AFS_PTHREAD_ENV */
shutdown_rxevent();
- rx_SetEpoch(0);
+ rx_epoch = 0;
#ifndef AFS_PTHREAD_ENV
#ifndef AFS_USE_GETTIMEOFDAY
clock_UnInit();
MUTEX_ENTER(&freeSQEList_lock);
- while ((np = rx_FreeSQEList)) {
- rx_FreeSQEList = *(struct rx_serverQueueEntry **)np;
+ while (!opr_queue_IsEmpty(&rx_freeServerQueue)) {
+ np = opr_queue_First(&rx_freeServerQueue, struct rx_serverQueueEntry,
+ entry);
+ opr_queue_Remove(&np->entry);
MUTEX_DESTROY(&np->lock);
rxi_Free(np, sizeof(*np));
}
rx_hashTableSize * sizeof(struct rx_connection *));
UNPIN(rx_peerHashTable, rx_hashTableSize * sizeof(struct rx_peer *));
- rxi_FreeAllPackets();
-
MUTEX_ENTER(&rx_quota_mutex);
rxi_dataQuota = RX_MAX_QUOTA;
rxi_availProcs = rxi_totalMin = rxi_minDeficit = 0;
MUTEX_EXIT(&rx_quota_mutex);
+ UNLOCK_RX_INIT;
}
#ifndef KERNEL
"rqc=%u,%u, tqc=%u,%u, iovqc=%u,%u, "
"lstatus=%u, rstatus=%u, error=%d, timeout=%u, "
"resendEvent=%d, keepAliveEvt=%d, delayedAckEvt=%d, delayedAbortEvt=%d, abortCode=%d, abortCount=%d, "
- "lastSendTime=%u, lastRecvTime=%u, lastSendData=%u"
+ "lastSendTime=%u, lastRecvTime=%u"
#ifdef RX_ENABLE_LOCKS
", refCount=%u"
#endif
(afs_uint32)c->rqc, (afs_uint32)rqc, (afs_uint32)c->tqc, (afs_uint32)tqc, (afs_uint32)c->iovqc, (afs_uint32)iovqc,
(afs_uint32)c->localStatus, (afs_uint32)c->remoteStatus, c->error, c->timeout,
c->resendEvent?1:0, c->keepAliveEvent?1:0, c->delayedAckEvent?1:0, c->delayedAbortEvent?1:0,
- c->abortCode, c->abortCount, c->lastSendTime, c->lastReceiveTime, c->lastSendData
+ c->abortCode, c->abortCount, c->lastSendTime, c->lastReceiveTime
#ifdef RX_ENABLE_LOCKS
, (afs_uint32)c->refCount
#endif
return 0;
}
#endif
+
+#ifdef AFS_RXERRQ_ENV
+void
+rxi_HandleSocketErrors(osi_socket sock)
+{
+ size_t cmsgbuf_len = 256;
+ void *cmsgbuf;
+# ifndef KERNEL
+ int errno_save = errno;
+# endif
+
+ cmsgbuf = rxi_Alloc(cmsgbuf_len);
+ if (cmsgbuf == NULL) {
+ goto done;
+ }
+
+ while (osi_HandleSocketError(sock, cmsgbuf, cmsgbuf_len))
+ ;
+
+ rxi_Free(cmsgbuf, cmsgbuf_len);
+
+ done:
+# ifndef KERNEL
+ errno = errno_save;
+# endif
+ return;
+}
+
+static int
+NetSend_retry(osi_socket sock, void *addr, struct iovec *dvec, int nvecs,
+ int length, int istack)
+{
+ int code;
+ int safety;
+ /*
+ * If an ICMP error comes in for any peer, sendmsg() can return -1 with an
+ * errno of EHOSTUNREACH, ENETUNREACH, etc. There may be no problem with
+ * sending this packet (an error is returned just to indicate we need to
+ * read in pending errors), but the packet wasn't actually sent.
+ *
+ * It's difficult to determine in general whether sendmsg() is returning an
+ * error due to a received ICMP error, or we're getting an actual error for
+ * this specific sendmsg() call, since there may be other threads running
+ * sendmsg/recvmsg/rxi_HandleSocketErrors at the same time. So, just retry
+ * the sendmsg a few times; make sure not to retry forever, in case we are
+ * getting an actual error from this sendmsg() call.
+ *
+ * Also note that if we accidentally drop a packet here that we didn't need
+ * to, it's not the end of the world. Packets get dropped, and we should be
+ * able to recover.
+ */
+ for (safety = 0; safety < RXI_SENDMSG_RETRY; safety++) {
+ code = osi_NetSend(sock, addr, dvec, nvecs, length, istack);
+ if (code == 0) {
+ return 0;
+ }
+ rxi_HandleSocketErrors(sock);
+ }
+ return code;
+
+}
+#endif
+
+int
+rxi_NetSend(osi_socket socket, void *addr, struct iovec *dvec,
+ int nvecs, int length, int istack)
+{
+ if (rxi_IsRunning()) {
+#ifdef AFS_RXERRQ_ENV
+ return NetSend_retry(socket, addr, dvec, nvecs, length, istack);
+#else
+ return osi_NetSend(socket, addr, dvec, nvecs, length, istack);
+#endif
+ }
+#ifdef AFS_NT40_ENV
+ return WSAESHUTDOWN;
+#else
+ return ESHUTDOWN;
+#endif
+}
git://git.openafs.org / openafs.git / blobdiff