static cache_t *
set_cache(cache_t **ccp, head_t *chp, int noreclaim)
{
/*
* Install a cache element:
*
* The caller passes the address of cache descriptor ["chp"] and the
* hash chain into which the new element is to be linked ["ccp"]. This
* routine allocates a new cache_t structure (or, if the maximum number
* of elements has already been allocated, reclaims the oldest element
* from the cache), links it into the indicated hash chain, and returns
* its address to the caller.
*/
cache_t *cap;
if ((chp->count < chp->maxblks) &&
(cap = (cache_t *)bkmem_alloc(chp->size))) {
/*
* We haven't reached the maximum cache size yet.
* Allocate a new "cache_t" struct to be added to the
* cache.
*/
chp->count += 1;
} else {
if (noreclaim)
return (NULL);
/*
* Cache is full. Use the "reclaim_cache" routine to
* remove the oldest element from the cache. This
* will become the cache_t struct associated with the
* new element.
*/
cap = reclaim_cache(chp, -1);
chp->purges += 1;
}
bzero((char *)cap, chp->size);
cap->chn = ccp;
cap->link[Prev] = (cache_t *)chp;
cap->link[Next] = chp->aged[Frst];
cap->link[Prev]->link[Next] = cap->link[Next]->link[Prev] = cap;
if ((cap->link[Hash+Next] = *ccp) != 0)
(*ccp)->link[Hash+Prev] = cap;
return (*ccp = cap);
}
int
set_rdcache(int dev, char *name, int pnum, int inum)
{
/*
* Reliably set the dcache
*
* This routine is the same as set_dcache except that it
* return 1 if the entry could not be entered into
* the cache without a purge.
*/
int len = strlen(name) + 1;
dc_t *dcp =
(dc_t *)set_cache(&dc_hash[DC_HASH(dev, name, len)],
&dc_head, 1);
if (dcp == NULL)
return (1);
if ((dcp->dc_hdr.data = (void *)bkmem_alloc(len)) == NULL) {
/*
* Not enough memory to make a copy of the name!
* There's probably not enough to do much else either!
*/
prom_panic("no memory for directory cache");
/* NOTREACHED */
}
/*
* Allocate a buffer for the pathname component, and
* make this the "data" portion of the generalize
* "cache_t" struct. Also fill in the cache-specific
* fields (pnum, inum).
*/
dcp->dc_pnum = pnum;
dcp->dc_inum = inum;
dcp->dc_hdr.dev = dev;
dcp->dc_hdr.size = len;
bcopy(name, (char *)dcp->dc_hdr.data, len);
return (0);
}
void
set_dcache(int dev, char *name, int pnum, int inum)
{
/*
* Build Directory Cache Entry:
*
* This routine creates directory cache entries to be retrieved later
* via "get_dcache". The cache key is composed of three parts: The
* device specifier, the file name ("name"), and the file number of
* the directory containing that name ("pnum"). The data portion of
* the entry consists of the file number ("inum").
*/
int len = strlen(name)+1;
dc_t *dcp =
(dc_t *)set_cache(&dc_hash[DC_HASH(dev, name, len)], &dc_head, 0);
if (dcp->dc_hdr.data = (void *)bkmem_alloc(len)) {
/*
* Allocate a buffer for the pathname component, and
* make this the "data" portion of the generalize
* "cache_t" struct. Also fill in the cache-specific
* fields (pnum, inum).
*/
dcp->dc_pnum = pnum;
dcp->dc_inum = inum;
dcp->dc_hdr.dev = dev;
dcp->dc_hdr.size = len;
bcopy(name, (char *)dcp->dc_hdr.data, len);
} else {
/*
* Not enough memory to make a copy of the name!
* There's probably not enough to do much else either!
*/
prom_panic("no memory for directory cache");
}
}
/*
* our version of brpc_call(). We cache in portnumber in to->sin_port for
* your convenience. to and from addresses are taken and received in network
* order.
*/
enum clnt_stat
brpc_call(
rpcprog_t prog, /* rpc program number to call. */
rpcvers_t vers, /* rpc program version */
rpcproc_t proc, /* rpc procedure to call */
xdrproc_t in_xdr, /* routine to serialize arguments */
caddr_t args, /* arg vector for remote call */
xdrproc_t out_xdr, /* routine to deserialize results */
caddr_t ret, /* addr of buf to place results in */
int rexmit, /* retransmission interval (secs) */
int wait_time, /* how long (secs) to wait (resp) */
struct sockaddr_in *to, /* destination */
struct sockaddr_in *from_who, /* responder's port/address */
uint_t auth) /* type of auth wanted. */
{
int s;
char hostname[MAXHOSTNAMELEN];
struct sockaddr_in from; /* us. */
socklen_t from_len;
XDR xmit_xdrs, rcv_xdrs; /* xdr memory */
AUTH *xmit_auth; /* our chosen auth cookie */
gid_t fake_gids = 1; /* fake gids list for auth_unix */
caddr_t trm_msg, rcv_msg; /* outgoing/incoming rpc mesgs */
struct rpc_msg reply; /* our reply msg header */
int trm_len, rcv_len;
struct rpc_err rpc_error; /* to store RPC errors in on rcv. */
static uint_t xid; /* current xid */
uint_t xmit_len; /* How much of the buffer we used */
int nrefreshes = 2; /* # of times to refresh cred */
int flags = 0; /* send flags */
uint_t xdelay;
int errors, preserve_errno;
uint32_t timeout;
socklen_t optlen;
xmit_auth = NULL;
trm_len = mac_get_mtu();
trm_msg = bkmem_alloc(trm_len);
rcv_msg = bkmem_alloc(NFSBUF_SIZE);
if (trm_msg == NULL || rcv_msg == NULL) {
errno = ENOMEM;
rpc_error.re_status = RPC_CANTSEND;
goto gt_error;
}
if ((s = socket(PF_INET, SOCK_DGRAM, 0)) < 0) {
rpc_error.re_status = RPC_CANTSEND;
goto gt_error;
}
if (dontroute) {
(void) setsockopt(s, SOL_SOCKET, SO_DONTROUTE,
(const void *)&dontroute, sizeof (dontroute));
}
if (to->sin_addr.s_addr == cached_destination.s_addr) {
optlen = sizeof (timeout);
(void) getsockopt(s, SOL_SOCKET, SO_RCVTIMEO, (void *)&timeout,
&optlen);
} else {
cached_destination.s_addr = htonl(INADDR_ANY);
}
/* Bind our endpoint. */
from.sin_family = AF_INET;
ipv4_getipaddr(&from.sin_addr);
from.sin_addr.s_addr = htonl(from.sin_addr.s_addr);
from.sin_port = get_source_port(B_TRUE);
if (bind(s, (struct sockaddr *)&from, sizeof (from)) < 0) {
rpc_error.re_status = RPC_CANTSEND;
goto gt_error;
}
bzero((caddr_t)&rpc_error, sizeof (struct rpc_err));
/* initialize reply's rpc_msg struct, so we can decode later. */
reply.acpted_rply.ar_verf = _null_auth; /* struct copy */
reply.acpted_rply.ar_results.where = ret;
reply.acpted_rply.ar_results.proc = out_xdr;
if (ntohs(to->sin_port) == 0) {
/* snag the udp port we need. */
if ((to->sin_port = (in_port_t)bpmap_getport(prog, vers,
&(rpc_error.re_status), to, NULL)) == 0)
goto gt_error;
to->sin_port = htons(to->sin_port);
}
/* generate xid - increment */
if (xid == 0)
xid = (uint_t)(prom_gettime() / 1000) + 1;
else
xid++;
/* set up outgoing pkt as xdr modified. */
xdrmem_create(&xmit_xdrs, trm_msg, trm_len, XDR_ENCODE);
/* setup rpc header */
if (rpc_hdr(&xmit_xdrs, xid, prog, vers, proc) != TRUE) {
dprintf("brpc_call: cannot setup rpc header.\n");
rpc_error.re_status = RPC_FAILED;
goto gt_error;
}
/* setup authentication */
switch (auth) {
case AUTH_NONE:
xmit_auth = authnone_create();
break;
case AUTH_UNIX:
/*
* Assumes we've configured the stack and thus know our
* IP address/hostname, either by using DHCP or rarp/bootparams.
*/
gethostname(hostname, sizeof (hostname));
xmit_auth = authunix_create(hostname, 0, 1, 1, &fake_gids);
break;
default:
dprintf("brpc_call: Unsupported authentication type: %d\n",
auth);
rpc_error.re_status = RPC_AUTHERROR;
goto gt_error;
/*NOTREACHED*/
}
/*
* rpc_hdr puts everything in the xmit buffer for the header
* EXCEPT the proc. Put it, and our authentication info into
* it now, serializing as we go. We will be at the place where
* we left off.
*/
xmit_xdrs.x_op = XDR_ENCODE;
if ((XDR_PUTINT32(&xmit_xdrs, (int32_t *)&proc) == FALSE) ||
(AUTH_MARSHALL(xmit_auth, &xmit_xdrs, NULL) == FALSE) ||
((*in_xdr)(&xmit_xdrs, args) == FALSE)) {
rpc_error.re_status = RPC_CANTENCODEARGS;
goto gt_error;
} else
xmit_len = (int)XDR_GETPOS(&xmit_xdrs); /* for sendto */
/*
* Right now the outgoing packet should be all serialized and
* ready to go... Set up timers.
*/
xdelay = (rexmit == 0) ? RPC_REXMIT_MSEC : (rexmit * 1000);
(void) setsockopt(s, SOL_SOCKET, SO_RCVTIMEO, (void *)&xdelay,
sizeof (xdelay));
wait_time = (wait_time == 0) ? RPC_RCVWAIT_MSEC : (wait_time * 1000);
wait_time += prom_gettime();
/*
* send out the request. The first item in the receive buffer will
* be the xid. Check if it is correct.
*/
errors = 0;
rpc_error.re_status = RPC_TIMEDOUT;
do {
if (sendto(s, trm_msg, xmit_len, flags, (struct sockaddr *)to,
sizeof (struct sockaddr_in)) < 0) {
/*
* If errno is set to ETIMEDOUT, return
* with RPC status as RPC_TIMEDOUT. Calling
* funciton will take care of this error by
* retrying the RPC call.
*/
if (errno == ETIMEDOUT) {
rpc_error.re_status = RPC_TIMEDOUT;
} else {
rpc_error.re_status = RPC_CANTSEND;
}
goto gt_error;
}
from_len = sizeof (struct sockaddr_in);
while ((rcv_len = recvfrom(s, rcv_msg, NFSBUF_SIZE,
MSG_DONTWAIT, (struct sockaddr *)from_who,
&from_len)) > 0 || errors < RPC_ALLOWABLE_ERRORS) {
if (rcv_len < 0) {
if (errno == EWOULDBLOCK ||
errno == ETIMEDOUT) {
break; /* timeout */
}
rpc_error.re_status = RPC_CANTRECV;
goto gt_error;
}
if (ntohl(*((uint32_t *)(rcv_msg))) != xid) {
dprintf("brpc_call: xid: 0x%x != 0x%x\n",
*(uint32_t *)(rcv_msg), xid);
continue;
}
/*
* Let's deserialize the data into our 'ret' buffer.
*/
xdrmem_create(&rcv_xdrs, rcv_msg, rcv_len, XDR_DECODE);
if (xdr_replymsg(&rcv_xdrs, &reply) == FALSE) {
rpc_error.re_status = RPC_CANTDECODERES;
goto gt_error;
}
_seterr_reply(&reply, &rpc_error);
switch (rpc_error.re_status) {
case RPC_SUCCESS:
/*
* XXX - validate for unix and none
* always return true.
*/
if (AUTH_VALIDATE(xmit_auth,
&reply.acpted_rply.ar_verf) == FALSE) {
rpc_error.re_status = RPC_AUTHERROR;
rpc_error.re_why = AUTH_INVALIDRESP;
errors++;
}
if (reply.acpted_rply.ar_verf.oa_base !=
0) {
xmit_xdrs.x_op = XDR_FREE;
(void) xdr_opaque_auth(
&xmit_xdrs,
&reply.acpted_rply.ar_verf);
}
break;
case RPC_AUTHERROR:
/*
* Let's see if our credentials need
* refreshing
*/
if (nrefreshes > 0 && AUTH_REFRESH(xmit_auth,
NULL, NULL)) {
nrefreshes--;
}
errors++;
break;
case RPC_PROCUNAVAIL:
/*
* Might be a silly portmapper implementation
* erroneously responding to our rpc broadcast
* indirect portmapper call. For this
* particular case, we don't increment the
* error counter because we want to keep
* sifting for successful replies...
*/
if (to->sin_addr.s_addr !=
ntohl(INADDR_BROADCAST))
errors++;
break;
case RPC_PROGVERSMISMATCH:
/*
* Successfully talked to server, but they
* don't speak our lingo.
*/
goto gt_error;
default:
/* Just keep trying till there's no data... */
errors++;
break;
}
if (rpc_error.re_status != RPC_SUCCESS) {
dprintf("brpc_call: from: %s, error: ",
inet_ntoa(from_who->sin_addr));
rpc_disperr(&rpc_error);
} else
break;
}
/*
* If we're having trouble reassembling datagrams, let the
* application know ASAP so that it can take the appropriate
* actions.
*/
} while (rpc_error.re_status != RPC_SUCCESS && errno != ETIMEDOUT &&
prom_gettime() < wait_time);
gt_error:
if (xmit_auth != NULL)
AUTH_DESTROY(xmit_auth);
if (trm_msg != NULL)
bkmem_free(trm_msg, trm_len);
if (rcv_msg != NULL)
bkmem_free(rcv_msg, NFSBUF_SIZE);
if (rpc_error.re_status != RPC_SUCCESS)
rpc_disperr(&rpc_error);
/*
* socket calls reset errno. Since we want to hold onto the errno
* value if it is ETIMEDOUT to communicate to our caller that this
* RPC_TIMEDOUT situation is due to a stack problem (we're getting
* a reply, but the stack simply can't assemble it.), we need to
* preserve errno's value over the socket_close().
*/
preserve_errno = (errno == ETIMEDOUT) ? errno : 0;
(void) socket_close(s);
errno = preserve_errno;
return (rpc_error.re_status);
}
/*
* Handle a IP datagram addressed to our MAC address or to the link
* layer broadcast address. Also respond to ARP requests. Generates
* inetgrams as long as there's data and the mac level IP timeout timer
* hasn't expired. As soon as there is no data, we try for
* IBD_INPUT_ATTEMPTS for more, then exit the loop, even if there is time
* left, since we expect to have data waiting for us when we're called, we just
* don't know how much.
*
* We workaround slow proms (some proms have hard sleeps for as much as 3msec)
* even though there are is data waiting.
*
* Returns the total number of MEDIA_LVL frames placed on the socket.
* Caller is expected to free up the inetgram resources.
*/
static int
ibd_input(int index)
{
struct inetgram *inp;
ipoib_ptxhdr_t *eh;
int frames = 0; /* successful frames */
int attempts = 0; /* failed attempts after success */
int16_t len = 0, data_len;
uint32_t timeout, reltime;
uint32_t pre_pr, post_pr; /* prom_read interval */
#ifdef DEBUG
int failures = 0; /* total failures */
int total_attempts = 0; /* total prom_read */
int no_data = 0; /* no data in prom */
int arps = 0; /* arp requests processed */
uint32_t tot_pr = 0; /* prom_read time */
uint32_t tot_pc = 0; /* inetgram creation time */
uint32_t pre_pc;
uint32_t now;
#endif /* DEBUG */
if (!initialized)
prom_panic("IPoIB device is not initialized.");
if ((reltime = sockets[index].in_timeout) == 0)
reltime = mac_state.mac_in_timeout;
timeout = prom_gettime() + reltime;
do {
if (frames > IBD_MAX_FRAMES) {
/* someone is trying a denial of service attack */
break;
}
/*
* The following is being paranoid about possible bugs
* where prom_read() returns a nonzero length, even when
* it's not read a packet; it zeroes out the header to
* compensate. Paranoia from calvin prom (V2) days.
*/
bzero(mac_state.mac_buf, sizeof (ipoib_ptxhdr_t));
/*
* Prom_read() will return 0 or -2 if no data is present. A
* return value of -1 means an error has occurred. We adjust
* the timeout by calling the time spent in prom_read() "free".
* prom_read() returns the number of bytes actually read, but
* will only copy "len" bytes into our buffer. Adjust in
* case the MTU is wrong.
*/
pre_pr = prom_gettime();
len = prom_read(mac_state.mac_dev, mac_state.mac_buf,
mac_state.mac_mtu, 0, NETWORK);
post_pr = prom_gettime();
timeout += (post_pr - pre_pr);
#ifdef DEBUG
tot_pr += (post_pr - pre_pr);
total_attempts++;
#endif /* DEBUG */
if (len > mac_state.mac_mtu) {
dprintf("ibd_input: adjusting MTU %d -> %d\n",
mac_state.mac_mtu, len);
bkmem_free(mac_state.mac_buf, mac_state.mac_mtu);
mac_state.mac_mtu = len;
mac_state.mac_buf = bkmem_alloc(mac_state.mac_mtu);
if (mac_state.mac_buf == NULL) {
prom_panic("ibd_input: Cannot reallocate "
"netbuf memory.");
}
len = 0; /* pretend there was no data */
}
if (len == -1) {
#ifdef DEBUG
failures++;
#endif /* DEBUG */
break;
}
if (len == 0 || len == -2) {
if (frames != 0)
attempts++;
#ifdef DEBUG
no_data++;
#endif /* DEBUG */
continue;
}
eh = (ipoib_ptxhdr_t *)mac_state.mac_buf;
if (eh->ipoib_rhdr.ipoib_type == ntohs(ETHERTYPE_IP) &&
len >= (sizeof (ipoib_ptxhdr_t) + sizeof (struct ip))) {
int offset;
#ifdef DEBUG
pre_pc = prom_gettime();
#endif /* DEBUG */
inp = (struct inetgram *)bkmem_zalloc(
sizeof (struct inetgram));
if (inp == NULL) {
errno = ENOMEM;
return (frames == 0 ? -1 : frames);
}
offset = sizeof (ipoib_ptxhdr_t);
data_len = len - offset;
inp->igm_mp = allocb(data_len, 0);
if (inp->igm_mp == NULL) {
errno = ENOMEM;
bkmem_free((caddr_t)inp,
sizeof (struct inetgram));
return (frames == 0 ? -1 : frames);
}
bcopy((caddr_t)(mac_state.mac_buf + offset),
inp->igm_mp->b_rptr, data_len);
inp->igm_mp->b_wptr += data_len;
inp->igm_level = NETWORK_LVL;
add_grams(&sockets[index].inq, inp);
frames++;
attempts = 0;
#ifdef DEBUG
tot_pc += prom_gettime() - pre_pc;
#endif /* DEBUG */
continue;
}
if (eh->ipoib_rhdr.ipoib_type == ntohs(ETHERTYPE_ARP) &&
len >= sizeof (struct arp_packet)) {
struct in_addr ip;
struct ibd_arp *ea;
#ifdef DEBUG
printf("ibd_input: ARP message received\n");
arps++;
#endif /* DEBUG */
ea = (struct ibd_arp *)(mac_state.mac_buf +
sizeof (ipoib_ptxhdr_t));
if (ea->arp_pro != ntohs(ETHERTYPE_IP))
continue;
ipv4_getipaddr(&ip);
ip.s_addr = ntohl(ip.s_addr);
if (ea->arp_op == ntohs(ARPOP_REQUEST) &&
ip.s_addr != INADDR_ANY &&
(bcmp((caddr_t)ea->arp_tpa, (caddr_t)&ip,
sizeof (struct in_addr)) == 0)) {
ea->arp_op = htons(ARPOP_REPLY);
bcopy((caddr_t)&ea->arp_sha,
(caddr_t)&eh->ipoib_dest, IPOIB_ADDRL);
bcopy((caddr_t)&ea->arp_sha,
(caddr_t)&ea->arp_tha, IPOIB_ADDRL);
bcopy((caddr_t)ea->arp_spa,
(caddr_t)ea->arp_tpa,
sizeof (struct in_addr));
bcopy(mac_state.mac_addr_buf,
(caddr_t)&ea->arp_sha,
mac_state.mac_addr_len);
bcopy((caddr_t)&ip, (caddr_t)ea->arp_spa,
sizeof (struct in_addr));
(void) prom_write(mac_state.mac_dev,
mac_state.mac_buf,
sizeof (struct arp_packet), 0, NETWORK);
/* don't charge for ARP replies */
timeout += reltime;
}
}
} while (attempts < IBD_INPUT_ATTEMPTS &&
#ifdef DEBUG
(now = prom_gettime()) < timeout);
#else
prom_gettime() < timeout);
#endif /* DEBUG */
#ifdef DEBUG
printf("ibd_input(%d): T/S/N/A/F/P/M: %d/%d/%d/%d/%d/%d/%d "
"T/O: %d < %d = %s\n", index, total_attempts, frames, no_data,
arps, failures, tot_pr, tot_pc, now, timeout,
(now < timeout) ? "TRUE" : "FALSE");
#endif /* DEBUG */
return (frames);
}
static struct memlist *
memlist_alloc()
{
return ((struct memlist *)bkmem_alloc(sizeof (struct memlist)));
}