int
comm_select(int msec)
{
static double last_timeout = 0.0;
int rc;
double start = current_dtime;
debug(5, 3) ("comm_select: timeout %d\n", msec);
if (msec > MAX_POLL_TIME)
msec = MAX_POLL_TIME;
#if DELAY_POOLS
/* We have delayed fds in queue? */
if (n_slow_fds)
msec = 0;
#endif
statCounter.select_loops++;
/* Check for disk io callbacks */
storeDirCallback();
/* Check timeouts once per second */
if (last_timeout + 0.999 < current_dtime) {
last_timeout = current_dtime;
checkTimeouts();
} else {
int max_timeout = (last_timeout + 1.0 - current_dtime) * 1000;
if (max_timeout < msec)
msec = max_timeout;
}
comm_select_handled = 0;
rc = do_comm_select(msec);
#if DELAY_POOLS
comm_call_slowfds();
#endif
getCurrentTime();
statCounter.select_time += (current_dtime - start);
if (rc == COMM_TIMEOUT)
debug(5, 8) ("comm_select: time out\n");
return rc;
}
static int
storeDiskdSend(int mtype, SwapDir * sd, int id, storeIOState * sio, int size, off_t offset, int shm_offset)
{
int x;
diomsg M;
static int send_errors = 0;
static int last_seq_no = 0;
static int seq_no = 0;
diskdinfo_t *diskdinfo = sd->fsdata;
struct timeval delay =
{0, 1};
M.mtype = mtype;
M.callback_data = sio;
M.size = size;
M.offset = offset;
M.status = -1;
M.shm_offset = (int) shm_offset;
M.id = id;
M.seq_no = ++seq_no;
if (M.callback_data)
cbdataLock(M.callback_data);
if (M.seq_no < last_seq_no)
debug(79, 1) ("WARNING: sequencing out of order\n");
/*
* We have to drain the queue here if necessary. If we don't,
* then we can have a lot of messages in the queue (probably
* up to 2*magic1) and we can run out of shared memory buffers.
*/
/*
* NOTE that it is important that we call storeDirCallback AFTER
* locking the callback data M.callback_data because we need
* to make sure the cbdata lock count doesn't go to zero (and
* get freed) before we have a chance to send the current message
* M!
*/
/*
* Note that we call storeDirCallback (for all SDs), rather
* than storeDiskdDirCallback for just this SD, so that while
* we're "blocking" on this SD we can also handle callbacks
* from other SDs that might be ready.
*/
while (diskdinfo->away > diskdinfo->magic2) {
select(0, NULL, NULL, NULL, &delay);
storeDirCallback();
if (delay.tv_usec < 1000000)
delay.tv_usec <<= 1;
}
#if 0
debug(79, 3) ("msgsnd %ld %d %d %p %d %" PRINTF_OFF_T " %d %d\n",
M.mtype,
M.id,
M.seq_no,
M.callback_data,
M.size,
M.offset,
M.status,
M.shm_offset);
#endif
x = msgsnd(diskdinfo->smsgid, &M, msg_snd_rcv_sz, IPC_NOWAIT);
last_seq_no = M.seq_no;
if (0 == x) {
diskd_stats.sent_count++;
diskdinfo->away++;
} else {
debug(79, 1) ("storeDiskdSend: msgsnd: %s\n", xstrerror());
if (M.callback_data)
cbdataUnlock(M.callback_data);
assert(++send_errors < 100);
}
return x;
}
/* Select on all sockets; call handlers for those that are ready. */
int
comm_select(int msec)
{
fd_set readfds;
fd_set pendingfds;
fd_set writefds;
#if DELAY_POOLS
BIT_ARRAY slowfds;
#endif
PF *hdl = NULL;
int fd;
int maxfd;
int num;
int pending;
int callicp = 0, callhttp = 0;
int calldns = 0;
int maxindex;
int k;
int j;
#if DEBUG_FDBITS
int i;
#endif
fd_mask *fdsp;
fd_mask *pfdsp;
fd_mask tmask;
static time_t last_timeout = 0;
struct timeval poll_time;
double timeout = current_dtime + (msec / 1000.0);
fde *F;
#if DELAY_POOLS
BA_INIT(slowfds,SQUID_NUMFD);
#endif
do {
#if !ALARM_UPDATES_TIME
double start;
getCurrentTime();
start = current_dtime;
#endif
#if DELAY_POOLS
BA_ZERO(slowfds);
#endif
/* Handle any fs callbacks that need doing */
storeDirCallback();
if (commCheckICPIncoming)
comm_select_icp_incoming();
if (commCheckDNSIncoming)
comm_select_dns_incoming();
if (commCheckHTTPIncoming)
comm_select_http_incoming();
callicp = calldns = callhttp = 0;
maxfd = Biggest_FD + 1;
xmemcpy(&readfds, &global_readfds,
howmany(maxfd, FD_MASK_BITS) * FD_MASK_BYTES);
xmemcpy(&writefds, &global_writefds,
howmany(maxfd, FD_MASK_BITS) * FD_MASK_BYTES);
/* remove stalled FDs, and deal with pending descriptors */
pending = 0;
FD_ZERO(&pendingfds);
maxindex = howmany(maxfd, FD_MASK_BITS);
fdsp = (fd_mask *) & readfds;
for (j = 0; j < maxindex; j++) {
if ((tmask = fdsp[j]) == 0)
continue; /* no bits here */
for (k = 0; k < FD_MASK_BITS; k++) {
if (!EBIT_TEST(tmask, k))
continue;
/* Found a set bit */
fd = (j * FD_MASK_BITS) + k;
switch (commDeferRead(fd)) {
case 0:
break;
case 1:
FD_CLR(fd, &readfds);
break;
#if DELAY_POOLS
case -1:
BA_SET(fd, slowfds);
break;
#endif
default:
fatalf("bad return value from commDeferRead(FD %d)\n", fd);
}
if (FD_ISSET(fd, &readfds) && fd_table[fd].flags.read_pending) {
FD_SET(fd, &pendingfds);
pending++;
}
}
}
#if DEBUG_FDBITS
for (i = 0; i < maxfd; i++) {
/* Check each open socket for a handler. */
#if DELAY_POOLS
if (fd_table[i].read_handler && commDeferRead(i) != 1) {
#else
if (fd_table[i].read_handler && !commDeferRead(i)) {
#endif
assert(FD_ISSET(i, &readfds));
}
if (fd_table[i].write_handler) {
assert(FD_ISSET(i, &writefds));
}
}
#endif
if (nreadfds + nwritefds == 0) {
assert(shutting_down);
#if DELAY_POOLS
BA_FREE(slowfds);
#endif
return COMM_SHUTDOWN;
}
if (msec > MAX_POLL_TIME)
msec = MAX_POLL_TIME;
if (pending)
msec = 0;
for (;;) {
poll_time.tv_sec = msec / 1000;
poll_time.tv_usec = (msec % 1000) * 1000;
statCounter.syscalls.selects++;
num = select(maxfd, &readfds, &writefds, NULL, &poll_time);
statCounter.select_loops++;
if (num >= 0 || pending > 0)
break;
if (ignoreErrno(errno))
break;
debug(50, 0) ("comm_select: select failure: %s\n",
xstrerror());
examine_select(&readfds, &writefds);
#if DELAY_POOLS
BA_FREE(slowfds);
#endif
return COMM_ERROR;
/* NOTREACHED */
}
if (num < 0 && !pending)
continue;
debug(5, num ? 5 : 8) ("comm_select: %d+%d FDs ready at %d\n",
num, pending, (int) squid_curtime);
statHistCount(&statCounter.select_fds_hist, num);
/* Check lifetime and timeout handlers ONCE each second.
* Replaces brain-dead check every time through the loop! */
if (squid_curtime > last_timeout) {
last_timeout = squid_curtime;
checkTimeouts();
}
if (num == 0 && pending == 0)
continue;
/* Scan return fd masks for ready descriptors */
fdsp = (fd_mask *) & readfds;
pfdsp = (fd_mask *) & pendingfds;
maxindex = howmany(maxfd, FD_MASK_BITS);
for (j = 0; j < maxindex; j++) {
if ((tmask = (fdsp[j] | pfdsp[j])) == 0)
continue; /* no bits here */
for (k = 0; k < FD_MASK_BITS; k++) {
if (tmask == 0)
break; /* no more bits left */
if (!EBIT_TEST(tmask, k))
continue;
/* Found a set bit */
fd = (j * FD_MASK_BITS) + k;
EBIT_CLR(tmask, k); /* this will be done */
#if DEBUG_FDBITS
debug(5, 9) ("FD %d bit set for reading\n", fd);
assert(FD_ISSET(fd, &readfds));
#endif
if (fdIsIcp(fd)) {
callicp = 1;
continue;
}
if (fdIsDns(fd)) {
calldns = 1;
continue;
}
if (fdIsHttp(fd)) {
callhttp = 1;
continue;
}
F = &fd_table[fd];
debug(5, 6) ("comm_select: FD %d ready for reading\n", fd);
if (NULL == (hdl = F->read_handler))
(void) 0;
#if DELAY_POOLS
else if (BA_ISSET(fd, slowfds))
commAddSlowFd(fd);
#endif
else {
F->read_handler = NULL;
commUpdateReadBits(fd, NULL);
hdl(fd, F->read_data);
statCounter.select_fds++;
if (commCheckICPIncoming)
comm_select_icp_incoming();
if (commCheckDNSIncoming)
comm_select_dns_incoming();
if (commCheckHTTPIncoming)
comm_select_http_incoming();
}
}
}
fdsp = (fd_mask *) & writefds;
for (j = 0; j < maxindex; j++) {
if ((tmask = fdsp[j]) == 0)
continue; /* no bits here */
for (k = 0; k < FD_MASK_BITS; k++) {
if (tmask == 0)
break; /* no more bits left */
if (!EBIT_TEST(tmask, k))
continue;
/* Found a set bit */
fd = (j * FD_MASK_BITS) + k;
EBIT_CLR(tmask, k); /* this will be done */
#if DEBUG_FDBITS
debug(5, 9) ("FD %d bit set for writing\n", fd);
assert(FD_ISSET(fd, &writefds));
#endif
if (fdIsIcp(fd)) {
callicp = 1;
continue;
}
if (fdIsDns(fd)) {
calldns = 1;
continue;
}
if (fdIsHttp(fd)) {
callhttp = 1;
continue;
}
F = &fd_table[fd];
debug(5, 5) ("comm_select: FD %d ready for writing\n", fd);
if ((hdl = F->write_handler)) {
F->write_handler = NULL;
commUpdateWriteBits(fd, NULL);
hdl(fd, F->write_data);
statCounter.select_fds++;
if (commCheckICPIncoming)
comm_select_icp_incoming();
if (commCheckDNSIncoming)
comm_select_dns_incoming();
if (commCheckHTTPIncoming)
comm_select_http_incoming();
}
}
}
if (callicp)
comm_select_icp_incoming();
if (calldns)
comm_select_dns_incoming();
if (callhttp)
comm_select_http_incoming();
#if DELAY_POOLS
while ((fd = commGetSlowFd()) != -1) {
F = &fd_table[fd];
debug(5, 6) ("comm_select: slow FD %d selected for reading\n", fd);
if ((hdl = F->read_handler)) {
F->read_handler = NULL;
commUpdateReadBits(fd, NULL);
hdl(fd, F->read_data);
statCounter.select_fds++;
if (commCheckICPIncoming)
comm_select_icp_incoming();
if (commCheckDNSIncoming)
comm_select_dns_incoming();
if (commCheckHTTPIncoming)
comm_select_http_incoming();
}
}
#endif
#if !ALARM_UPDATES_TIME
getCurrentTime();
statCounter.select_time += (current_dtime - start);
#endif
#if DELAY_POOLS
BA_FREE(slowfds);
#endif
return COMM_OK;
}
while (timeout > current_dtime);
debug(5, 8) ("comm_select: time out: %d\n", (int) squid_curtime);
#if DELAY_POOLS
BA_FREE(slowfds);
#endif
return COMM_TIMEOUT;
}
#endif
static void
#if HAVE_POLL
comm_poll_dns_incoming(void)
#else
comm_select_dns_incoming(void)
#endif
{
int nfds = 0;
int fds[2];
int nevents;
dns_io_events = 0;
if (DnsSocket < 0)
return;
fds[nfds++] = DnsSocket;
#if HAVE_POLL
nevents = comm_check_incoming_poll_handlers(nfds, fds);
#else
nevents = comm_check_incoming_select_handlers(nfds, fds);
#endif
if (nevents < 0)
return;
incoming_dns_interval += Config.comm_incoming.dns_average - nevents;
if (incoming_dns_interval < Config.comm_incoming.dns_min_poll)
incoming_dns_interval = Config.comm_incoming.dns_min_poll;
if (incoming_dns_interval > MAX_INCOMING_INTERVAL)
incoming_dns_interval = MAX_INCOMING_INTERVAL;
if (nevents > INCOMING_DNS_MAX)
nevents = INCOMING_DNS_MAX;
statHistCount(&statCounter.comm_dns_incoming, nevents);
}
/* poll all sockets; call handlers for those that are ready. */
int
comm_poll(int msec)
{
#if FD_CONFIG
struct pollfd *pfds = NULL;
#else
struct pollfd pfds[SQUID_NUMFD];
#endif
#if DELAY_POOLS
BIT_ARRAY slowfds;
#endif
PF *hdl = NULL;
int fd;
unsigned int i;
unsigned int maxfd;
unsigned int nfds;
unsigned int npending;
int num;
int callicp = 0, callhttp = 0;
int calldns = 0;
static time_t last_timeout = 0;
double timeout = current_dtime + (msec / 1000.0);
#if FD_CONFIG
pfds = xmalloc(sizeof(pfds[0])*SQUID_NUMFD);
#endif
#if DELAY_POOLS
BA_INIT(slowfds,SQUID_NUMFD);
#endif
do {
#if !ALARM_UPDATES_TIME
double start;
getCurrentTime();
start = current_dtime;
#endif
/* Handle any fs callbacks that need doing */
storeDirCallback();
#if DELAY_POOLS
BA_ZERO(slowfds);
#endif
if (commCheckICPIncoming)
comm_poll_icp_incoming();
if (commCheckDNSIncoming)
comm_poll_dns_incoming();
if (commCheckHTTPIncoming)
comm_poll_http_incoming();
callicp = calldns = callhttp = 0;
nfds = 0;
npending = 0;
maxfd = Biggest_FD + 1;
for (i = 0; i < maxfd; i++) {
int events;
events = 0;
/* Check each open socket for a handler. */
if (fd_table[i].read_handler) {
switch (commDeferRead(i)) {
case 0:
events |= POLLRDNORM;
break;
case 1:
break;
#if DELAY_POOLS
case -1:
events |= POLLRDNORM;
BA_SET(i, slowfds);
break;
#endif
default:
fatalf("bad return value from commDeferRead(FD %d)\n", i);
}
}
if (fd_table[i].write_handler)
events |= POLLWRNORM;
if (events) {
pfds[nfds].fd = i;
pfds[nfds].events = events;
pfds[nfds].revents = 0;
nfds++;
if ((events & POLLRDNORM) && fd_table[i].flags.read_pending)
npending++;
}
}
if (nfds == 0) {
assert(shutting_down);
#if DELAY_POOLS
BA_FREE(slowfds);
#endif
#if FD_CONFIG
xfree(pfds);
#endif
return COMM_SHUTDOWN;
}
if (npending)
msec = 0;
if (msec > MAX_POLL_TIME)
msec = MAX_POLL_TIME;
for (;;) {
statCounter.syscalls.polls++;
num = poll(pfds, nfds, msec);
statCounter.select_loops++;
if (num >= 0 || npending > 0)
break;
if (ignoreErrno(errno))
continue;
debug(5, 0) ("comm_poll: poll failure: %s\n", xstrerror());
assert(errno != EINVAL);
#if DELAY_POOLS
BA_FREE(slowfds);
#endif
#if FD_CONFIG
xfree(pfds);
#endif
return COMM_ERROR;
/* NOTREACHED */
}
debug(5, num ? 5 : 8) ("comm_poll: %d+%u FDs ready\n", num, npending);
statHistCount(&statCounter.select_fds_hist, num);
/* Check timeout handlers ONCE each second. */
if (squid_curtime > last_timeout) {
last_timeout = squid_curtime;
checkTimeouts();
}
if (num <= 0 && npending == 0)
continue;
/* scan each socket but the accept socket. Poll this
* more frequently to minimize losses due to the 5 connect
* limit in SunOS */
for (i = 0; i < nfds; i++) {
fde *F;
int revents = pfds[i].revents;
fd = pfds[i].fd;
if (fd == -1)
continue;
if (fd_table[fd].flags.read_pending)
revents |= POLLIN;
if (revents == 0)
continue;
if (fdIsIcp(fd)) {
callicp = 1;
continue;
}
if (fdIsDns(fd)) {
calldns = 1;
continue;
}
if (fdIsHttp(fd)) {
callhttp = 1;
continue;
}
F = &fd_table[fd];
if (revents & (POLLRDNORM | POLLIN | POLLHUP | POLLERR)) {
debug(5, 6) ("comm_poll: FD %d ready for reading\n", fd);
if (NULL == (hdl = F->read_handler))
(void) 0;
#if DELAY_POOLS
else if (BA_ISSET(fd, slowfds))
commAddSlowFd(fd);
#endif
else {
F->read_handler = NULL;
hdl(fd, F->read_data);
statCounter.select_fds++;
if (commCheckICPIncoming)
comm_poll_icp_incoming();
if (commCheckDNSIncoming)
comm_poll_dns_incoming();
if (commCheckHTTPIncoming)
comm_poll_http_incoming();
}
}
if (revents & (POLLWRNORM | POLLOUT | POLLHUP | POLLERR)) {
debug(5, 5) ("comm_poll: FD %d ready for writing\n", fd);
if ((hdl = F->write_handler)) {
F->write_handler = NULL;
hdl(fd, F->write_data);
statCounter.select_fds++;
if (commCheckICPIncoming)
comm_poll_icp_incoming();
if (commCheckDNSIncoming)
comm_poll_dns_incoming();
if (commCheckHTTPIncoming)
comm_poll_http_incoming();
}
}
if (revents & POLLNVAL) {
close_handler *ch;
debug(5, 0) ("WARNING: FD %d has handlers, but it's invalid.\n", fd);
debug(5, 0) ("FD %d is a %s\n", fd, fdTypeStr[F->type]);
debug(5, 0) ("--> %s\n", F->desc);
debug(5, 0) ("tmout:%p read:%p write:%p\n",
F->timeout_handler,
F->read_handler,
F->write_handler);
for (ch = F->close_handler; ch; ch = ch->next)
debug(5, 0) (" close handler: %p\n", ch->handler);
if (F->close_handler) {
commCallCloseHandlers(fd);
} else if (F->timeout_handler) {
debug(5, 0) ("comm_poll: Calling Timeout Handler\n");
F->timeout_handler(fd, F->timeout_data);
}
F->close_handler = NULL;
F->timeout_handler = NULL;
F->read_handler = NULL;
F->write_handler = NULL;
if (F->flags.open)
fd_close(fd);
}
}
if (callicp)
comm_poll_icp_incoming();
if (calldns)
comm_poll_dns_incoming();
if (callhttp)
comm_poll_http_incoming();
#if DELAY_POOLS
while ((fd = commGetSlowFd()) != -1) {
fde *F = &fd_table[fd];
debug(5, 6) ("comm_select: slow FD %d selected for reading\n", fd);
if ((hdl = F->read_handler)) {
F->read_handler = NULL;
hdl(fd, F->read_data);
statCounter.select_fds++;
if (commCheckICPIncoming)
comm_poll_icp_incoming();
if (commCheckDNSIncoming)
comm_poll_dns_incoming();
if (commCheckHTTPIncoming)
comm_poll_http_incoming();
}
}
#endif
#if !ALARM_UPDATES_TIME
getCurrentTime();
statCounter.select_time += (current_dtime - start);
#endif
#if DELAY_POOLS
BA_FREE(slowfds);
#endif
#if FD_CONFIG
xfree(pfds);
#endif
return COMM_OK;
}
while (timeout > current_dtime);
debug(5, 8) ("comm_poll: time out: %ld.\n", (long int) squid_curtime);
#if DELAY_POOLS
BA_FREE(slowfds);
#endif
#if FD_CONFIG
xfree(pfds);
#endif
return COMM_TIMEOUT;
}