errors better.
* Fixed bug in 'implicit loopback'
* Completely removed duplicate interface usage checks from generic network
setup. The NAL should be doing this. Currently socknal is the only one
that supports multiple instances and it doesn't mind at all if 2 different
network instances use the same interfaces, and that's dead useful for
testing.
* Tightened up lonal refcounting; there was a remote chance of a shutdown
race.
extern int ptl_count_acceptor_nis(ptl_ni_t **first_ni);
-extern ptl_err_t ptl_accept(struct socket *sock, __u32 magic, int choose_ni);
+extern ptl_err_t ptl_accept(ptl_ni_t *blind_ni, struct socket *sock, __u32 magic);
extern int ptl_acceptor_timeout(void);
extern int ptl_acceptor_port(void);
#endif
extern int ptl_count_acceptor_nis(ptl_ni_t **first_ni);
-extern ptl_err_t ptl_accept(struct socket *sock, __u32 magic, int choose_ni);
+extern ptl_err_t ptl_accept(ptl_ni_t *blind_ni, struct socket *sock, __u32 magic);
extern int ptl_acceptor_timeout(void);
extern int ptl_acceptor_port(void);
#endif
} lod_iov;
} lo_desc_t;
+/* loopback descriptor types */
#define LOD_IOV 0xeb105
#define LOD_KIOV 0xeb106
ptl_acceptor_timeout());
} else {
/* This might be a generic acceptor connection request... */
- rc = ptl_accept(sock, msg->ibm_magic, 0);
+ rc = ptl_accept(kibnal_data.kib_ni, sock, msg->ibm_magic);
if (rc != PTL_OK)
goto out;
connreq->racr_magic != RANAL_MSG_MAGIC &&
connreq->racr_magic != __swab32(RANAL_MSG_MAGIC)) {
/* Is this a generic acceptor connection request? */
- rc = ptl_accept(sock, connreq->racr_magic, 0);
+ rc = ptl_accept(kranal_data.kra_ni, sock, connreq->racr_magic);
if (rc != PTL_OK) /* nope */
return -EPROTO;
if (route->ksnr_myipaddr != conn->ksnc_myipaddr) {
if (route->ksnr_myipaddr == 0) {
/* route wasn't bound locally yet (the initial route) */
- CWARN("Binding %s %u.%u.%u.%u to %u.%u.%u.%u\n",
- libcfs_nid2str(peer->ksnp_nid),
- HIPQUAD(route->ksnr_ipaddr),
- HIPQUAD(conn->ksnc_myipaddr));
+ CDEBUG(D_NET, "Binding %s %u.%u.%u.%u to %u.%u.%u.%u\n",
+ libcfs_nid2str(peer->ksnp_nid),
+ HIPQUAD(route->ksnr_ipaddr),
+ HIPQUAD(conn->ksnc_myipaddr));
} else {
- CWARN("Rebinding %s %u.%u.%u.%u from "
- "%u.%u.%u.%u to %u.%u.%u.%u\n",
- libcfs_nid2str(peer->ksnp_nid),
- HIPQUAD(route->ksnr_ipaddr),
- HIPQUAD(route->ksnr_myipaddr),
- HIPQUAD(conn->ksnc_myipaddr));
+ CDEBUG(D_NET, "Rebinding %s %u.%u.%u.%u from "
+ "%u.%u.%u.%u to %u.%u.%u.%u\n",
+ libcfs_nid2str(peer->ksnp_nid),
+ HIPQUAD(route->ksnr_ipaddr),
+ HIPQUAD(route->ksnr_myipaddr),
+ HIPQUAD(conn->ksnc_myipaddr));
iface = ksocknal_ip2iface(route->ksnr_peer->ksnp_ni,
route->ksnr_myipaddr);
PORTAL_ALLOC(cr, sizeof(*cr));
if (cr == NULL) {
- CWARN("ENOMEM allocating connection request from"
- "%u.%u.%u.%u\n", HIPQUAD(peer_ip));
+ LCONSOLE_ERROR("Dropping connection request from "
+ "%u.%u.%u.%u: memory exhausted\n",
+ HIPQUAD(peer_ip));
return PTL_FAIL;
}
+ ptl_ni_addref(ni);
+ cr->ksncr_ni = ni;
cr->ksncr_sock = sock;
spin_lock_irqsave(&ksocknal_data.ksnd_connd_lock, flags);
}
int
-ksocknal_create_conn (ksock_route_t *route, struct socket *sock, int type)
+ksocknal_create_conn (ptl_ni_t *ni, ksock_route_t *route,
+ struct socket *sock, int type)
{
rwlock_t *global_lock = &ksocknal_data.ksnd_global_lock;
+ ksock_net_t *net = (ksock_net_t *)ni->ni_data;
__u32 ipaddrs[PTL_MAX_INTERFACES];
int nipaddrs;
- ptl_ni_t *ni;
- ksock_net_t *net;
ptl_nid_t nid;
struct list_head *tmp;
__u64 incarnation;
goto failed_1;
if (route != NULL) {
- ni = route->ksnr_peer->ksnp_ni;
- net = (ksock_net_t *)ni->ni_data;
+ LASSERT(ni == route->ksnr_peer->ksnp_ni);
/* Active connection sends HELLO eagerly */
nipaddrs = ksocknal_local_ipvec(ni, ipaddrs);
- rc = ksocknal_send_hello (conn, ni->ni_nid,
- net->ksnn_incarnation,
- ipaddrs, nipaddrs);
+ rc = ksocknal_send_hello (ni, conn, ipaddrs, nipaddrs);
if (rc != 0)
goto failed_1;
}
* Passive connections use the listener timeout since the peer sends
* eagerly */
nid = (route == NULL) ? PTL_NID_ANY : route->ksnr_peer->ksnp_nid;
- rc = ksocknal_recv_hello (conn, &nid, &incarnation, ipaddrs);
+ rc = ksocknal_recv_hello (ni, conn, &nid, &incarnation, ipaddrs);
if (rc < 0)
goto failed_1;
nipaddrs = rc;
ipaddrs, nipaddrs);
rc = 0;
} else {
- ni = ptl_net2ni(PTL_NIDNET(nid));
-
- if (ni == NULL) {
- CERROR("Refusing connection attempt "
- "(no matching net)\n");
- rc = -ECONNREFUSED;
- goto failed_1;
- }
-
- net = (ksock_net_t *)ni->ni_data;
rc = ksocknal_create_peer(&peer, ni, nid);
-
- /* lose extra ref from ptl_net2ni NB we wait for all the peers
- * to be deleted before ni teardown can complete; i.e. ni can't
- * disappear until all its peer table entries has gone so
- * there's no need to account the peer's refs on ni. */
- ptl_ni_decref(ni);
-
if (rc != 0)
goto failed_1;
write_unlock_irqrestore(global_lock, flags);
nipaddrs = ksocknal_select_ips(peer, ipaddrs, nipaddrs);
- rc = ksocknal_send_hello (conn, ni->ni_nid,
- net->ksnn_incarnation,
- ipaddrs, nipaddrs);
+ rc = ksocknal_send_hello (ni, conn, ipaddrs, nipaddrs);
if (rc < 0)
goto failed_2;
}
ksocknal_connsock_decref(conn);
}
- CWARN("New conn %s %u.%u.%u.%u -> %u.%u.%u.%u/%d"
- " incarnation:"LPD64" sched[%d]/%d\n",
- libcfs_nid2str(nid), HIPQUAD(conn->ksnc_myipaddr),
- HIPQUAD(conn->ksnc_ipaddr), conn->ksnc_port, incarnation,
- (int)(conn->ksnc_scheduler - ksocknal_data.ksnd_schedulers), irq);
+ CDEBUG(D_NET, "New conn %s %u.%u.%u.%u -> %u.%u.%u.%u/%d"
+ " incarnation:"LPD64" sched[%d]/%d\n",
+ libcfs_nid2str(nid), HIPQUAD(conn->ksnc_myipaddr),
+ HIPQUAD(conn->ksnc_ipaddr), conn->ksnc_port, incarnation,
+ (int)(conn->ksnc_scheduler - ksocknal_data.ksnd_schedulers), irq);
ksocknal_conn_decref(conn);
return (0);
if (conn->ksnc_incarnation == incarnation)
continue;
- CWARN("Closing stale conn %s ip:%08x/%d "
- "incarnation:"LPD64"("LPD64")\n",
- libcfs_nid2str(peer->ksnp_nid),
- conn->ksnc_ipaddr, conn->ksnc_port,
- conn->ksnc_incarnation, incarnation);
+ CDEBUG(D_NET, "Closing stale conn %s ip:%08x/%d "
+ "incarnation:"LPD64"("LPD64")\n",
+ libcfs_nid2str(peer->ksnp_nid),
+ conn->ksnc_ipaddr, conn->ksnc_port,
+ conn->ksnc_incarnation, incarnation);
count++;
ksocknal_close_conn_locked (conn, -ESTALE);
typedef struct ksock_connreq
{
struct list_head ksncr_list; /* stash on ksnd_connd_connreqs */
+ ptl_ni_t *ksncr_ni; /* chosen NI */
struct socket *ksncr_sock; /* accepted socket */
} ksock_connreq_t;
extern int ksocknal_add_peer(ptl_ni_t *ni, ptl_nid_t nid, __u32 ip, int port);
extern ksock_peer_t *ksocknal_find_peer_locked (ptl_ni_t *ni, ptl_nid_t nid);
extern ksock_peer_t *ksocknal_find_peer (ptl_ni_t *ni, ptl_nid_t nid);
-extern int ksocknal_create_conn (ksock_route_t *route,
+extern int ksocknal_create_conn (ptl_ni_t *ni, ksock_route_t *route,
struct socket *sock, int type);
extern void ksocknal_close_conn_locked (ksock_conn_t *conn, int why);
extern void ksocknal_terminate_conn (ksock_conn_t *conn);
extern int ksocknal_scheduler (void *arg);
extern int ksocknal_connd (void *arg);
extern int ksocknal_reaper (void *arg);
-extern int ksocknal_send_hello (ksock_conn_t *conn, ptl_nid_t nid,
- __u64 incarnation, __u32 *ipaddrs, int nipaddrs);
-extern int ksocknal_recv_hello (ksock_conn_t *conn, ptl_nid_t *nid,
- __u64 *incarnation, __u32 *ipaddrs);
+extern int ksocknal_send_hello (ptl_ni_t *ni, ksock_conn_t *conn,
+ __u32 *ipaddrs, int nipaddrs);
+extern int ksocknal_recv_hello (ptl_ni_t *ni, ksock_conn_t *conn,
+ ptl_nid_t *nid, __u64 *incarnation,
+ __u32 *ipaddrs);
extern void ksocknal_lib_save_callback(struct socket *sock, ksock_conn_t *conn);
extern void ksocknal_lib_set_callback(struct socket *sock, ksock_conn_t *conn);
LASSERT (rc != -EAGAIN);
if (rc == 0)
- CWARN ("[%p] EOF from %s ip %d.%d.%d.%d:%d\n",
- conn,
- libcfs_nid2str(conn->ksnc_peer->ksnp_nid),
- HIPQUAD(conn->ksnc_ipaddr),
- conn->ksnc_port);
+ CDEBUG (D_NET, "[%p] EOF from %s ip %d.%d.%d.%d:%d\n",
+ conn,
+ libcfs_nid2str(conn->ksnc_peer->ksnp_nid),
+ HIPQUAD(conn->ksnc_ipaddr),
+ conn->ksnc_port);
else if (!conn->ksnc_closing)
CERROR ("[%p] Error %d on read from %s"
" ip %d.%d.%d.%d:%d\n",
}
int
-ksocknal_send_hello (ksock_conn_t *conn,
- ptl_nid_t srcnid, __u64 incarnation,
+ksocknal_send_hello (ptl_ni_t *ni, ksock_conn_t *conn,
__u32 *ipaddrs, int nipaddrs)
{
/* CAVEAT EMPTOR: this byte flips 'ipaddrs' */
+ ksock_net_t *net = (ksock_net_t *)ni->ni_data;
struct socket *sock = conn->ksnc_sock;
ptl_hdr_t hdr;
ptl_magicversion_t *hmv = (ptl_magicversion_t *)&hdr.dest_nid;
hmv->version_major = cpu_to_le16 (PTL_PROTO_TCP_VERSION_MAJOR);
hmv->version_minor = cpu_to_le16 (PTL_PROTO_TCP_VERSION_MINOR);
- hdr.src_nid = cpu_to_le64 (srcnid);
+ hdr.src_nid = cpu_to_le64 (ni->ni_nid);
hdr.type = cpu_to_le32 (PTL_MSG_HELLO);
hdr.payload_length = cpu_to_le32 (nipaddrs * sizeof(*ipaddrs));
hdr.msg.hello.type = cpu_to_le32 (conn->ksnc_type);
- hdr.msg.hello.incarnation = cpu_to_le64 (incarnation);
+ hdr.msg.hello.incarnation = cpu_to_le64 (net->ksnn_incarnation);
for (i = 0; i < nipaddrs; i++) {
ipaddrs[i] = __cpu_to_le32 (ipaddrs[i]);
}
int
-ksocknal_recv_hello (ksock_conn_t *conn, ptl_nid_t *nid,
- __u64 *incarnation, __u32 *ipaddrs)
+ksocknal_recv_hello (ptl_ni_t *ni, ksock_conn_t *conn,
+ ptl_nid_t *nid, __u64 *incarnation, __u32 *ipaddrs)
{
struct socket *sock = conn->ksnc_sock;
int active;
if (!active &&
hmv->magic != le32_to_cpu (PTL_PROTO_TCP_MAGIC)) {
/* Is this a generic acceptor connection request? */
- rc = ptl_accept(sock, hmv->magic, 0);
+ rc = ptl_accept(ni, sock, hmv->magic);
if (rc != PTL_OK)
return -EPROTO;
- /* Yes it is! Start over again now I've skipping it. */
+ /* Yes it is! Start over again now I've skipping the generic
+ * request */
rc = libcfs_sock_read(sock, &hmv->magic,
sizeof (hmv->magic), timeout);
if (rc != 0) {
{
CFS_LIST_HEAD (zombies);
ksock_tx_t *tx;
- ksock_peer_t *peer;
+ ksock_peer_t *peer = route->ksnr_peer;
unsigned long flags;
int type;
struct socket *sock;
write_unlock_irqrestore(&ksocknal_data.ksnd_global_lock, flags);
- rc = ptl_connect(&sock, route->ksnr_peer->ksnp_nid,
+ rc = ptl_connect(&sock, peer->ksnp_nid,
route->ksnr_myipaddr,
route->ksnr_ipaddr, route->ksnr_port);
if (rc != PTL_OK)
goto failed;
- rc = ksocknal_create_conn(route, sock, type);
+ rc = ksocknal_create_conn(peer->ksnp_ni, route, sock, type);
if (rc != 0) {
- ptl_connect_console_error(rc, route->ksnr_peer->ksnp_nid,
+ ptl_connect_console_error(rc, peer->ksnp_nid,
route->ksnr_ipaddr,
route->ksnr_port);
goto failed;
failed:
write_lock_irqsave (&ksocknal_data.ksnd_global_lock, flags);
- peer = route->ksnr_peer;
LASSERT (route->ksnr_connecting);
route->ksnr_connecting = 0;
spin_unlock_irqrestore(&ksocknal_data.ksnd_connd_lock,
flags);
- ksocknal_create_conn(NULL, cr->ksncr_sock, SOCKNAL_CONN_NONE);
+ ksocknal_create_conn(cr->ksncr_ni, NULL,
+ cr->ksncr_sock, SOCKNAL_CONN_NONE);
+ ptl_ni_decref(cr->ksncr_ni);
PORTAL_FREE(cr, sizeof(*cr));
spin_lock_irqsave(&ksocknal_data.ksnd_connd_lock,
return rc;
if (rc == 0)
- return -ECONNABORTED;
+ return -ECONNRESET;
buffer = ((char *)buffer) + rc;
nob -= rc;
/* "normal" errors */
case -ECONNREFUSED:
LCONSOLE_ERROR("Connection to %s at host %u.%u.%u.%u "
- "on port %d was refused; "
+ "on port %d was refused: "
"check that Lustre is running on that node.\n",
libcfs_nid2str(peer_nid),
HIPQUAD(peer_ip), peer_port);
case -EHOSTUNREACH:
case -ENETUNREACH:
LCONSOLE_ERROR("Connection to %s at host %u.%u.%u.%u "
- "was unreachable; the network or that node may "
+ "was unreachable: the network or that node may "
"be down, or Lustre may be misconfigured.\n",
libcfs_nid2str(peer_nid), HIPQUAD(peer_ip));
break;
case -ETIMEDOUT:
LCONSOLE_ERROR("Connection to %s at host %u.%u.%u.%u on "
- "port %d took too long; that node may be hung "
+ "port %d took too long: that node may be hung "
"or experiencing high load.\n",
libcfs_nid2str(peer_nid),
HIPQUAD(peer_ip), peer_port);
break;
case -ECONNRESET:
LCONSOLE_ERROR("Connection to %s at host %u.%u.%u.%u on "
- "port %d was reset; "
- "Is it running a compatible version of Lustre?\n",
+ "port %d was reset: "
+ "is it running a compatible version of Lustre "
+ "and is %s one of its NIDs?\n",
libcfs_nid2str(peer_nid),
- HIPQUAD(peer_ip), peer_port);
+ HIPQUAD(peer_ip), peer_port,
+ libcfs_nid2str(peer_nid));
break;
case -EPROTO:
LCONSOLE_ERROR("Protocol error connecting to %s at host "
"%u.%u.%u.%u on port %d: "
- "Is it running a compatible version of Lustre?\n",
+ "is it running a compatible version of Lustre?\n",
libcfs_nid2str(peer_nid),
HIPQUAD(peer_ip), peer_port);
break;
}
ptl_err_t
-ptl_accept(struct socket *sock, __u32 magic, int choose_ni)
+ptl_accept(ptl_ni_t *blind_ni, struct socket *sock, __u32 magic)
{
ptl_acceptor_connreq_t cr;
__u32 peer_ip;
int rc;
int flip;
ptl_ni_t *ni;
+ char *str;
/* CAVEAT EMPTOR: I may be called by a NAL in any thread's context if I
* passed the new socket "blindly" to the single NI that needed an
- * acceptor. If so, 'choose_ni' is FALSE... */
+ * acceptor. If so, blind_ni != NULL... */
LASSERT (sizeof(cr) <= 16); /* not too big for the stack */
rc = libcfs_sock_getaddr(sock, 1, &peer_ip, &peer_port);
LASSERT (rc == 0); /* we succeeded before */
- if (ptl_accept_magic(magic, PTL_PROTO_TCP_MAGIC)) {
- CERROR("Refusing connection from %u.%u.%u.%u: "
- " 'old' socknal/tcpnal acceptor protocol\n",
- HIPQUAD(peer_ip));
- return PTL_FAIL;
- }
-
- if (ptl_accept_magic(magic, PTL_PROTO_RA_MAGIC)) {
- CERROR("Refusing connection from %u.%u.%u.%u: "
- " 'old' ranal acceptor protocol\n",
- HIPQUAD(peer_ip));
- return PTL_FAIL;
- }
-
- if (ptl_accept_magic(magic, PTL_PROTO_OPENIB_MAGIC)) {
- CERROR("Refusing connection from %u.%u.%u.%u: "
- " 'old' openibnal acceptor protocol\n",
- HIPQUAD(peer_ip));
- return PTL_FAIL;
- }
-
if (!ptl_accept_magic(magic, PTL_PROTO_ACCEPTOR_MAGIC)) {
- CERROR("Refusing connection from %u.%u.%u.%u: "
- " unrecognised magic %08x\n",
- HIPQUAD(peer_ip), magic);
+
+ if (magic == le32_to_cpu(PTL_PROTO_TCP_MAGIC))
+ str = "'old' socknal/tcpnal";
+ else if (ptl_accept_magic(magic, PTL_PROTO_RA_MAGIC))
+ str = "'old' ranal";
+ else if (ptl_accept_magic(magic, PTL_PROTO_OPENIB_MAGIC))
+ str = "'old' openibnal";
+ else
+ str = "unrecognised";
+
+ LCONSOLE_ERROR("Refusing connection from %u.%u.%u.%u magic %08x: "
+ " %s acceptor protocol\n",
+ HIPQUAD(peer_ip), magic, str);
return PTL_FAIL;
}
}
if (cr.acr_version != PTL_PROTO_ACCEPTOR_VERSION) {
- CERROR("Refusing connection from %u.%u.%u.%u: "
- " unrecognised protocol version %d\n",
- HIPQUAD(peer_ip), cr.acr_version);
+ LCONSOLE_ERROR("Refusing connection from %u.%u.%u.%u: "
+ " unrecognised protocol version %d\n",
+ HIPQUAD(peer_ip), cr.acr_version);
return PTL_FAIL;
}
- if (!choose_ni) {
- CDEBUG(D_WARNING, "Skipped %s from %u.%u.%u.%u\n",
- libcfs_nid2str(cr.acr_nid), HIPQUAD(peer_ip));
- /* I got called just to skip the connection request */
- return PTL_OK;
- }
-
ni = ptl_net2ni(PTL_NIDNET(cr.acr_nid));
if (ni == NULL || /* no matching net */
ni->ni_nid != cr.acr_nid) /* right NET, but wrong NID! */ {
if (ni != NULL)
ptl_ni_decref(ni);
- CERROR("Refusing connection from %u.%u.%u.%u for %s: "
- " No matching NI\n",
- HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
+ LCONSOLE_ERROR("Refusing connection from %u.%u.%u.%u for %s: "
+ " No matching NI\n",
+ HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
return PTL_FAIL;
}
if (ni->ni_nal->nal_accept == NULL) {
ptl_ni_decref(ni);
- CERROR("Refusing connection from %u.%u.%u.%u for %s: "
- " NI doesn not accept IP connections\n",
- HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
+ LCONSOLE_ERROR("Refusing connection from %u.%u.%u.%u for %s: "
+ " NI doesn not accept IP connections\n",
+ HIPQUAD(peer_ip), libcfs_nid2str(cr.acr_nid));
return PTL_FAIL;
}
- CDEBUG(D_WARNING, "Accept %s from %u.%u.%u.%u\n",
- libcfs_nid2str(cr.acr_nid), HIPQUAD(peer_ip));
+ CDEBUG(D_NET, "Accept %s from %u.%u.%u.%u%s\n",
+ libcfs_nid2str(cr.acr_nid), HIPQUAD(peer_ip),
+ blind_ni == NULL ? "" : " (blind)");
- rc = ni->ni_nal->nal_accept(ni, sock);
- if (rc != PTL_OK)
- CERROR("NI %s refused connection from %u.%u.%u.%u\n",
- libcfs_nid2str(ni->ni_nid), HIPQUAD(peer_ip));
+ if (blind_ni == NULL) {
+ rc = ni->ni_nal->nal_accept(ni, sock);
+ if (rc != PTL_OK)
+ CERROR("NI %s refused connection from %u.%u.%u.%u\n",
+ libcfs_nid2str(ni->ni_nid), HIPQUAD(peer_ip));
+ } else {
+ /* blind_ni is the only NI that needs me and it was given the
+ * chance to handle this connection request itself in case it
+ * was sent by an "old" socknal. But this connection request
+ * uses the new acceptor protocol and I'm just being called to
+ * verify and skip it */
+ LASSERT (ni == blind_ni);
+ rc = PTL_OK;
+ }
ptl_ni_decref(ni);
return rc;
goto failed;
}
- rc = ptl_accept(newsock, magic, 1);
+ rc = ptl_accept(NULL, newsock, magic);
if (rc != PTL_OK)
goto failed;
struct list_head *tmp;
ptl_ni_t *ni;
unsigned long flags;
+ int islocal = 0;
PTL_LOCK(flags);
+
list_for_each (tmp, &ptl_apini.apini_nis) {
ni = list_entry(tmp, ptl_ni_t, ni_list);
- if (ni->ni_nid == nid)
- return 1;
+ if (ni->ni_nid == nid) {
+ islocal = 1;
+ break;
+ }
}
PTL_UNLOCK(flags);
- return 0;
+ return islocal;
}
void
ptl_shutdown_nalnis (void)
{
int i;
+ int islo;
ptl_ni_t *ni;
unsigned long flags;
ptl_ni_decref_locked(ni); /* drop apini's ref (shutdown on last ref) */
}
+ /* Drop the cached loopback NI. */
+ ptl_ni_decref_locked(ptl_loni);
+ ptl_loni = NULL;
+
/* Now wait for the NI's I just nuked to show up on apini_zombie_nis
* and shut them down in guaranteed thread context */
i = 2;
PTL_UNLOCK(flags);
+ islo = ni->ni_nal->nal_type == LONAL;
+
LASSERT (!in_interrupt());
(ni->ni_nal->nal_shutdown)(ni);
/* can't deref nal anymore now; it might have unregistered
* itself... */
- if (PTL_NETNAL(PTL_NIDNET(ni->ni_nid)) != LONAL)
+ if (!islo)
LCONSOLE(0, "Removed NI %s\n",
libcfs_nid2str(ni->ni_nid));
PTL_UNLOCK(flags);
}
+ ptl_loni = ptl_net2ni(PTL_MKNET(LONAL, 0));
+ LASSERT (ptl_loni != NULL);
+
return PTL_OK;
failed:
}
int
-ptl_nis_conflict(ptl_ni_t *ni1, ptl_ni_t *ni2)
+ptl_net_unique(__u32 net, struct list_head *nilist)
{
- if (PTL_NETNAL(PTL_NIDNET(ni1->ni_nid)) != /* different NALs */
- PTL_NETNAL(PTL_NIDNET(ni2->ni_nid)))
- return 0;
-
- if (ni1 != ni2 &&
- PTL_NIDNET(ni1->ni_nid) == PTL_NIDNET(ni2->ni_nid)) {
- CERROR("Duplicate network: %s\n",
- libcfs_net2str(PTL_NIDNET(ni1->ni_nid)));
- return 1;
- }
-
- if (ni1->ni_interfaces[0] == NULL ||
- ni2->ni_interfaces[0] == NULL) {
- /* one (or both) using all available interfaces */
- if (ni1 != ni2) {
- CERROR("Interface conflict: %s, %s\n",
- libcfs_net2str(PTL_NIDNET(ni1->ni_nid)),
- libcfs_net2str(PTL_NIDNET(ni2->ni_nid)));
- return 1;
- }
- return 0;
- }
-#if 0
- /* leave this commented out so the same interface can be included explicitly in 2
- * networks. */
-
- for (i = 0; i < PTL_MAX_INTERFACES; i++) {
- if (ni1->ni_interfaces[i] == NULL)
- break;
-
- for (j = 0; j < PTL_MAX_INTERFACES; j++) {
- if (ni2->ni_interfaces[j] == NULL)
- break;
+ struct list_head *tmp;
+ ptl_ni_t *ni;
- if (ni1 == ni2 && i == j)
- continue;
+ list_for_each (tmp, nilist) {
+ ni = list_entry(tmp, ptl_ni_t, ni_list);
- if (strcmp(ni1->ni_interfaces[i],
- ni2->ni_interfaces[j]))
- continue;
-
- CERROR("Duplicate interface: %s(%s), %s(%s)\n",
- libcfs_net2str(PTL_NIDNET(ni1->ni_nid)),
- ni1->ni_interfaces[i],
- libcfs_net2str(PTL_NIDNET(ni2->ni_nid)),
- ni2->ni_interfaces[i]);
- return 1;
- }
+ if (PTL_NIDNET(ni->ni_nid) == net)
+ return 0;
}
-#endif
- return 0;
+
+ return 1;
}
-ptl_err_t
-ptl_check_ni_conflicts(ptl_ni_t *ni, struct list_head *nilist)
+ptl_ni_t *
+ptl_new_ni(__u32 net, struct list_head *nilist)
{
- struct list_head *tmp;
- ptl_ni_t *ni2;
+ ptl_ni_t *ni;
- /* Yes! ni _has_ just been added to this list. */
- LASSERT (ni == list_entry(nilist->prev, ptl_ni_t, ni_list));
+ if (!ptl_net_unique(net, nilist)) {
+ LCONSOLE_ERROR("Duplicate network specified: %s\n",
+ libcfs_net2str(net));
+ return NULL;
+ }
- list_for_each (tmp, nilist) {
- ni2 = list_entry(tmp, ptl_ni_t, ni_list);
-
- if (ptl_nis_conflict(ni, ni2))
- return PTL_FAIL;
+ PORTAL_ALLOC(ni, sizeof(*ni));
+ if (ni == NULL) {
+ CERROR("Out of memory creating network %s\n",
+ libcfs_net2str(net));
+ return NULL;
}
- return PTL_OK;
+ /* zero counters/flags, NULL pointers... */
+ memset(ni, 0, sizeof(*ni));
+
+ /* NAL will fill in the address part of the NID */
+ ni->ni_nid = PTL_MKNID(net, 0);
+
+ list_add_tail(&ni->ni_list, nilist);
+ return ni;
}
ptl_err_t
int tokensize = strlen(networks) + 1;
char *tokens;
char *str;
- ptl_ni_t *ni = NULL;
+ ptl_ni_t *ni;
__u32 net;
if (strlen(networks) > PTL_SINGLE_TEXTBUF_NOB) {
/* _WAY_ conservative */
- CERROR("Can't parse networks; string too long\n");
+ LCONSOLE_ERROR("Can't parse networks: string too long\n");
return PTL_FAIL;
}
ptl_apini.apini_network_tokens_nob = tokensize;
memcpy (tokens, networks, tokensize);
str = tokens;
-
- PORTAL_ALLOC(ptl_loni, sizeof(*ptl_loni));
- if (ptl_loni == NULL) {
- CERROR("Can't allocate LO NI\n");
- goto failed;
- }
+
/* Add in the loopback network */
- /* zero counters/flags, NULL pointers... */
- memset(ptl_loni, 0, sizeof(*ptl_loni));
- ptl_loni->ni_nid = PTL_MKNID(PTL_MKNET(LONAL, 0), 0);
- list_add_tail(&ptl_loni->ni_list, nilist);
+ ni = ptl_new_ni(PTL_MKNET(LONAL, 0), nilist);
+ if (ni == NULL)
+ goto failed;
while (str != NULL && *str != 0) {
char *comma = strchr(str, ',');
int niface;
char *iface;
- PORTAL_ALLOC(ni, sizeof(*ni));
- if (ni == NULL) {
- CERROR ("ENOMEM parsing 'networks=\"%s\"'\n", networks);
- goto failed;
- }
- /* zero counters/flags, NULL pointers... */
- memset(ni, 0, sizeof(*ni));
- list_add_tail(&ni->ni_list, nilist);
-
+ /* NB we don't check interface conflicts here; it's the NALs
+ * responsibility (if it cares at all) */
+
if (bracket == NULL ||
(comma != NULL && comma < bracket)) {
+
+ /* no interface list specified */
+
if (comma != NULL)
*comma++ = 0;
net = libcfs_str2net(ptl_trimwhite(str));
goto failed;
}
- ni->ni_nid = PTL_MKNID(net, 0);
- if (ptl_check_ni_conflicts(ni, nilist) != PTL_OK)
+ if (ptl_new_ni(net, nilist) == NULL)
goto failed;
str = comma;
goto failed;
}
- ni->ni_nid = PTL_MKNID(net, 0);
+ ni = ptl_new_ni(net, nilist);
+ if (ni == NULL)
+ goto failed;
niface = 0;
iface = bracket + 1;
}
if (niface == PTL_MAX_INTERFACES) {
- LCONSOLE_ERROR("Too many interfaces for %s\n",
- libcfs_net2str(PTL_NIDNET(ni->ni_nid)));
+ LCONSOLE_ERROR("Too many interfaces for net %s\n",
+ libcfs_net2str(net));
goto failed;
}
iface = comma;
} while (iface != NULL);
- if (ptl_check_ni_conflicts(ni, nilist) != PTL_OK)
- goto failed;
-
str = bracket + 1;
comma = strchr(bracket + 1, ',');
if (comma != NULL) {
ptl_hdr_t *hdr, int type, ptl_process_id_t target,
ptl_libmd_t *md, ptl_size_t offset, ptl_size_t len)
{
- ptl_nid_t gw_nid;
- int routing = 0;
- ptl_err_t rc;
+ unsigned long flags;
+ ptl_nid_t gw_nid;
+ int routing = 0;
+ ptl_err_t rc;
/* CAVEAT EMPTOR! ni != NULL == interface pre-determined (ACK) */
return PTL_FAIL;
}
- if (PTL_NETNAL(PTL_NIDNET(ni->ni_nid)) != LONAL) {
+ if (ni->ni_nal->nal_type != LONAL) {
if (gw_nid != ni->ni_nid) { /* it's not for me */
routing = gw_nid != target.nid; /* will gateway have to forward? */
} else if (allow_destination_aliases || /* force lonal? */
implicit_loopback) {
- ptl_ni_addref(ptl_loni);
- ptl_ni_decref(ni);
+
+ PTL_LOCK(flags);
+ ptl_ni_decref_locked(ni);
ni = ptl_loni;
+ if (ni != NULL)
+ ptl_ni_addref_locked(ni);
+ PTL_UNLOCK(flags);
+
+ if (ni == NULL) /* shutdown in progress */
+ return PTL_FAIL;
+
+ if (implicit_loopback)
+ target.nid = ni->ni_nid;
+
} else { /* barf */
ptl_ni_decref(ni);
CERROR("Attempt to send to self via %s, not LONAL\n",
/* That's "OK I can parse it", not "OK I like it" :) */
}
-ptl_ni_t *
-ptl_nid2ni (ptl_nid_t nid)
-{
- struct list_head *tmp;
- ptl_ni_t *ni;
-
- /* Called holding PTL_LOCK */
-
- list_for_each (tmp, &ptl_apini.apini_nis) {
- ni = list_entry(tmp, ptl_ni_t, ni_list);
-
- /* network type & number match in target NID and ni's NID */
- if (((ni->ni_nid ^ nid)>>32) == 0)
- return ni;
- }
-
- return NULL;
-}
-
ptl_err_t
PtlPut(ptl_handle_md_t mdh, ptl_ack_req_t ack,
ptl_process_id_t target, ptl_pt_index_t portal,
lonal_shutdown(ptl_ni_t *ni)
{
CDEBUG (D_NET, "shutdown\n");
- LASSERT (ni == ptl_loni);
LASSERT (lonal_instanced);
lonal_instanced = 0;
lonal_startup (ptl_ni_t *ni)
{
LASSERT (ni->ni_nal == &ptl_lonal);
- LASSERT (ni == ptl_loni);
LASSERT (!lonal_instanced);
lonal_instanced = 1;
git://git.whamcloud.com / fs / lustre-release.git / commitdiff