kib_data_t kibnal_data;
kib_tunables_t kibnal_tunables;
-#ifdef CONFIG_SYSCTL
#define IBNAL_SYSCTL 202
-#define IBNAL_SYSCTL_TIMEOUT 1
+enum {
+ IBNAL_SYSCTL_TIMEOUT=1,
+ IBNAL_SYSCTL_LISTENER_TIMEOUT,
+ IBNAL_SYSCTL_BACKLOG,
+ IBNAL_SYSCTL_PORT
+};
static ctl_table kibnal_ctl_table[] = {
{IBNAL_SYSCTL_TIMEOUT, "timeout",
&kibnal_tunables.kib_io_timeout, sizeof (int),
0644, NULL, &proc_dointvec},
+ {IBNAL_SYSCTL_LISTENER_TIMEOUT, "listener_timeout",
+ &kibnal_tunables.kib_listener_timeout, sizeof(int),
+ 0644, NULL, &proc_dointvec},
+ {IBNAL_SYSCTL_BACKLOG, "backlog",
+ &kibnal_tunables.kib_backlog, sizeof(int),
+ 0644, NULL, kibnal_listener_procint},
+ {IBNAL_SYSCTL_PORT, "port",
+ &kibnal_tunables.kib_port, sizeof(int),
+ 0644, NULL, kibnal_listener_procint},
{ 0 }
};
{IBNAL_SYSCTL, "openibnal", NULL, 0, 0555, kibnal_ctl_table},
{ 0 }
};
-#endif
+
+__u32
+kibnal_cksum (void *ptr, int nob)
+{
+ char *c = ptr;
+ __u32 sum = 0;
+
+ while (nob-- > 0)
+ sum = ((sum << 1) | (sum >> 31)) + *c++;
+
+ /* ensure I don't return 0 (== no checksum) */
+ return (sum == 0) ? 1 : sum;
+}
void
-print_service(struct ib_common_attrib_service *service, char *tag, int rc)
+kibnal_init_msg(kib_msg_t *msg, int type, int body_nob)
{
- char name[32];
+ msg->ibm_type = type;
+ msg->ibm_nob = offsetof(kib_msg_t, ibm_u) + body_nob;
+}
- if (service == NULL)
- {
- CWARN("tag : %s\n"
- "status : %d (NULL)\n", tag, rc);
- return;
+void
+kibnal_pack_msg(kib_msg_t *msg, int credits, ptl_nid_t dstnid, __u64 dststamp)
+{
+ /* CAVEAT EMPTOR! all message fields not set here should have been
+ * initialised previously. */
+ msg->ibm_magic = IBNAL_MSG_MAGIC;
+ msg->ibm_version = IBNAL_MSG_VERSION;
+ /* ibm_type */
+ msg->ibm_credits = credits;
+ /* ibm_nob */
+ msg->ibm_cksum = 0;
+ msg->ibm_srcnid = kibnal_lib.libnal_ni.ni_pid.nid;
+ msg->ibm_srcstamp = kibnal_data.kib_incarnation;
+ msg->ibm_dstnid = dstnid;
+ msg->ibm_dststamp = dststamp;
+#if IBNAL_CKSUM
+ /* NB ibm_cksum zero while computing cksum */
+ msg->ibm_cksum = kibnal_cksum(msg, msg->ibm_nob);
+#endif
+}
+
+int
+kibnal_unpack_msg(kib_msg_t *msg, int nob)
+{
+ const int hdr_size = offsetof(kib_msg_t, ibm_u);
+ __u32 msg_cksum;
+ int flip;
+ int msg_nob;
+
+ if (nob < 6) {
+ CERROR("Short message: %d\n", nob);
+ return -EPROTO;
+ }
+
+ if (msg->ibm_magic == IBNAL_MSG_MAGIC) {
+ flip = 0;
+ } else if (msg->ibm_magic == __swab32(IBNAL_MSG_MAGIC)) {
+ flip = 1;
+ } else {
+ CERROR("Bad magic: %08x\n", msg->ibm_magic);
+ return -EPROTO;
+ }
+
+ if (msg->ibm_version !=
+ (flip ? __swab16(IBNAL_MSG_VERSION) : IBNAL_MSG_VERSION)) {
+ CERROR("Bad version: %d\n", msg->ibm_version);
+ return -EPROTO;
+ }
+
+ if (nob < hdr_size) {
+ CERROR("Short message: %d\n", nob);
+ return -EPROTO;
+ }
+
+ msg_nob = flip ? __swab32(msg->ibm_nob) : msg->ibm_nob;
+ if (msg_nob > nob) {
+ CERROR("Short message: got %d, wanted %d\n", nob, msg_nob);
+ return -EPROTO;
+ }
+
+ /* checksum must be computed with ibm_cksum zero and BEFORE anything
+ * gets flipped */
+ msg_cksum = flip ? __swab32(msg->ibm_cksum) : msg->ibm_cksum;
+ msg->ibm_cksum = 0;
+ if (msg_cksum != 0 &&
+ msg_cksum != kibnal_cksum(msg, msg_nob)) {
+ CERROR("Bad checksum\n");
+ return -EPROTO;
+ }
+ msg->ibm_cksum = msg_cksum;
+
+ if (flip) {
+ /* leave magic unflipped as a clue to peer endianness */
+ __swab16s(&msg->ibm_version);
+ LASSERT (sizeof(msg->ibm_type) == 1);
+ LASSERT (sizeof(msg->ibm_credits) == 1);
+ msg->ibm_nob = msg_nob;
+ __swab64s(&msg->ibm_srcnid);
+ __swab64s(&msg->ibm_srcstamp);
+ __swab64s(&msg->ibm_dstnid);
+ __swab64s(&msg->ibm_dststamp);
}
- strncpy (name, service->service_name, sizeof(name)-1);
- name[sizeof(name)-1] = 0;
- CWARN("tag : %s\n"
- "status : %d\n"
- "service id: "LPX64"\n"
- "name : %s\n"
- "NID : "LPX64"\n", tag, rc,
- service->service_id, name,
- *kibnal_service_nid_field(service));
+ if (msg->ibm_srcnid == PTL_NID_ANY) {
+ CERROR("Bad src nid: "LPX64"\n", msg->ibm_srcnid);
+ return -EPROTO;
+ }
+
+ switch (msg->ibm_type) {
+ default:
+ CERROR("Unknown message type %x\n", msg->ibm_type);
+ return -EPROTO;
+
+ case IBNAL_MSG_SVCQRY:
+ case IBNAL_MSG_NOOP:
+ break;
+
+ case IBNAL_MSG_SVCRSP:
+ if (msg_nob < hdr_size + sizeof(msg->ibm_u.svcrsp)) {
+ CERROR("Short SVCRSP: %d(%d)\n", msg_nob,
+ (int)(hdr_size + sizeof(msg->ibm_u.svcrsp)));
+ return -EPROTO;
+ }
+ if (flip) {
+ __swab64s(&msg->ibm_u.svcrsp.ibsr_svc_id);
+ __swab16s(&msg->ibm_u.svcrsp.ibsr_svc_pkey);
+ }
+ break;
+
+ case IBNAL_MSG_CONNREQ:
+ case IBNAL_MSG_CONNACK:
+ if (msg_nob < hdr_size + sizeof(msg->ibm_u.connparams)) {
+ CERROR("Short CONNREQ: %d(%d)\n", msg_nob,
+ (int)(hdr_size + sizeof(msg->ibm_u.connparams)));
+ return -EPROTO;
+ }
+ if (flip)
+ __swab32s(&msg->ibm_u.connparams.ibcp_queue_depth);
+ break;
+
+ case IBNAL_MSG_IMMEDIATE:
+ if (msg_nob < offsetof(kib_msg_t, ibm_u.immediate.ibim_payload[0])) {
+ CERROR("Short IMMEDIATE: %d(%d)\n", msg_nob,
+ (int)offsetof(kib_msg_t, ibm_u.immediate.ibim_payload[0]));
+ return -EPROTO;
+ }
+ break;
+
+ case IBNAL_MSG_PUT_RDMA:
+ case IBNAL_MSG_GET_RDMA:
+ if (msg_nob < hdr_size + sizeof(msg->ibm_u.rdma)) {
+ CERROR("Short RDMA req: %d(%d)\n", msg_nob,
+ (int)(hdr_size + sizeof(msg->ibm_u.rdma)));
+ return -EPROTO;
+ }
+ if (flip) {
+ __swab32s(&msg->ibm_u.rdma.ibrm_desc.rd_key);
+ __swab32s(&msg->ibm_u.rdma.ibrm_desc.rd_nob);
+ __swab64s(&msg->ibm_u.rdma.ibrm_desc.rd_addr);
+ }
+ break;
+
+ case IBNAL_MSG_PUT_DONE:
+ case IBNAL_MSG_GET_DONE:
+ if (msg_nob < hdr_size + sizeof(msg->ibm_u.completion)) {
+ CERROR("Short RDMA completion: %d(%d)\n", msg_nob,
+ (int)(hdr_size + sizeof(msg->ibm_u.completion)));
+ return -EPROTO;
+ }
+ if (flip)
+ __swab32s(&msg->ibm_u.completion.ibcm_status);
+ break;
+ }
+ return 0;
}
-void
-kibnal_service_setunset_done (tTS_IB_CLIENT_QUERY_TID tid, int status,
- struct ib_common_attrib_service *service, void *arg)
+int
+kibnal_sock_write (struct socket *sock, void *buffer, int nob)
{
- *(int *)arg = status;
- up (&kibnal_data.kib_nid_signal);
+ int rc;
+ mm_segment_t oldmm = get_fs();
+ struct iovec iov = {
+ .iov_base = buffer,
+ .iov_len = nob
+ };
+ struct msghdr msg = {
+ .msg_name = NULL,
+ .msg_namelen = 0,
+ .msg_iov = &iov,
+ .msg_iovlen = 1,
+ .msg_control = NULL,
+ .msg_controllen = 0,
+ .msg_flags = MSG_DONTWAIT
+ };
+
+ /* We've set up the socket's send buffer to be large enough for
+ * everything we send, so a single non-blocking send should
+ * complete without error. */
+
+ set_fs(KERNEL_DS);
+ rc = sock_sendmsg(sock, &msg, iov.iov_len);
+ set_fs(oldmm);
+
+ if (rc == nob)
+ return 0;
+
+ if (rc >= 0)
+ return -EAGAIN;
+
+ return rc;
}
-#if IBNAL_CHECK_ADVERT
-void
-kibnal_check_advert (void)
+int
+kibnal_sock_read (struct socket *sock, void *buffer, int nob, int timeout)
{
- struct ib_common_attrib_service *svc;
- __u64 tid;
- int rc;
- int rc2;
+ int rc;
+ mm_segment_t oldmm = get_fs();
+ long ticks = timeout * HZ;
+ unsigned long then;
+ struct timeval tv;
- PORTAL_ALLOC(svc, sizeof(*svc));
- if (svc == NULL)
- return;
+ LASSERT (nob > 0);
+ LASSERT (ticks > 0);
+
+ for (;;) {
+ struct iovec iov = {
+ .iov_base = buffer,
+ .iov_len = nob
+ };
+ struct msghdr msg = {
+ .msg_name = NULL,
+ .msg_namelen = 0,
+ .msg_iov = &iov,
+ .msg_iovlen = 1,
+ .msg_control = NULL,
+ .msg_controllen = 0,
+ .msg_flags = 0
+ };
+
+ /* Set receive timeout to remaining time */
+ tv = (struct timeval) {
+ .tv_sec = ticks / HZ,
+ .tv_usec = ((ticks % HZ) * 1000000) / HZ
+ };
+ set_fs(KERNEL_DS);
+ rc = sock_setsockopt(sock, SOL_SOCKET, SO_RCVTIMEO,
+ (char *)&tv, sizeof(tv));
+ set_fs(oldmm);
+ if (rc != 0) {
+ CERROR("Can't set socket recv timeout %d: %d\n",
+ timeout, rc);
+ return rc;
+ }
+
+ set_fs(KERNEL_DS);
+ then = jiffies;
+ rc = sock_recvmsg(sock, &msg, iov.iov_len, 0);
+ ticks -= jiffies - then;
+ set_fs(oldmm);
+
+ if (rc < 0)
+ return rc;
- memset (svc, 0, sizeof (*svc));
- kibnal_set_service_keys(svc, kibnal_data.kib_nid);
+ if (rc == 0)
+ return -ECONNABORTED;
- rc = ib_service_get (kibnal_data.kib_device,
- kibnal_data.kib_port,
- svc,
- KIBNAL_SERVICE_KEY_MASK,
- kibnal_tunables.kib_io_timeout * HZ,
- kibnal_service_setunset_done, &rc2,
- &tid);
+ buffer = ((char *)buffer) + rc;
+ nob -= rc;
+ if (nob == 0)
+ return 0;
+
+ if (ticks <= 0)
+ return -ETIMEDOUT;
+ }
+}
+
+int
+kibnal_create_sock(struct socket **sockp)
+{
+ struct socket *sock;
+ int rc;
+ int option;
+ mm_segment_t oldmm = get_fs();
+
+ rc = sock_create(PF_INET, SOCK_STREAM, 0, &sock);
if (rc != 0) {
- CERROR ("Immediate error %d checking SM service\n", rc);
- } else {
- down (&kibnal_data.kib_nid_signal);
- rc = rc2;
+ CERROR("Can't create socket: %d\n", rc);
+ return rc;
+ }
+
+ /* Ensure sends will not block */
+ option = 2 * sizeof(kib_msg_t);
+ set_fs(KERNEL_DS);
+ rc = sock_setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
+ (char *)&option, sizeof(option));
+ set_fs(oldmm);
+ if (rc != 0) {
+ CERROR("Can't set send buffer %d: %d\n", option, rc);
+ goto failed;
+ }
+
+ option = 1;
+ set_fs(KERNEL_DS);
+ rc = sock_setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
+ (char *)&option, sizeof(option));
+ set_fs(oldmm);
+ if (rc != 0) {
+ CERROR("Can't set SO_REUSEADDR: %d\n", rc);
+ goto failed;
+ }
+
+ *sockp = sock;
+ return 0;
+
+ failed:
+ sock_release(sock);
+ return rc;
+}
+
+void
+kibnal_pause(int ticks)
+{
+ set_current_state(TASK_UNINTERRUPTIBLE);
+ schedule_timeout(ticks);
+}
+
+int
+kibnal_connect_sock(kib_peer_t *peer, struct socket **sockp)
+{
+ struct sockaddr_in locaddr;
+ struct sockaddr_in srvaddr;
+ struct socket *sock;
+ unsigned int port;
+ int rc;
+
+ for (port = 1023; port >= 512; port--) {
+ memset(&locaddr, 0, sizeof(locaddr));
+ locaddr.sin_family = AF_INET;
+ locaddr.sin_port = htons(port);
+ locaddr.sin_addr.s_addr = htonl(INADDR_ANY);
+
+ memset (&srvaddr, 0, sizeof (srvaddr));
+ srvaddr.sin_family = AF_INET;
+ srvaddr.sin_port = htons (peer->ibp_port);
+ srvaddr.sin_addr.s_addr = htonl (peer->ibp_ip);
+
+ rc = kibnal_create_sock(&sock);
if (rc != 0)
- CERROR ("Error %d checking SM service\n", rc);
+ return rc;
+
+ rc = sock->ops->bind(sock,
+ (struct sockaddr *)&locaddr, sizeof(locaddr));
+ if (rc != 0) {
+ sock_release(sock);
+
+ if (rc == -EADDRINUSE) {
+ CDEBUG(D_NET, "Port %d already in use\n", port);
+ continue;
+ }
+
+ CERROR("Can't bind to reserved port %d: %d\n", port, rc);
+ return rc;
+ }
+
+ rc = sock->ops->connect(sock,
+ (struct sockaddr *)&srvaddr, sizeof(srvaddr),
+ 0);
+ if (rc == 0) {
+ *sockp = sock;
+ return 0;
+ }
+
+ sock_release(sock);
+
+ if (rc != -EADDRNOTAVAIL) {
+ CERROR("Can't connect port %d to %u.%u.%u.%u/%d: %d\n",
+ port, HIPQUAD(peer->ibp_ip), peer->ibp_port, rc);
+ return rc;
+ }
+
+ CDEBUG(D_NET, "Port %d not available for %u.%u.%u.%u/%d\n",
+ port, HIPQUAD(peer->ibp_ip), peer->ibp_port);
}
- PORTAL_FREE(svc, sizeof(*svc));
+ /* all ports busy */
+ return -EHOSTUNREACH;
}
-#endif
int
-kibnal_advertise (void)
+kibnal_make_svcqry (kib_conn_t *conn)
{
- struct ib_common_attrib_service *svc;
- __u64 tid;
- int rc;
- int rc2;
+ kib_peer_t *peer = conn->ibc_peer;
+ kib_msg_t *msg;
+ struct socket *sock;
+ int rc;
+ int nob;
- LASSERT (kibnal_data.kib_nid != PTL_NID_ANY);
+ LASSERT (conn->ibc_connreq != NULL);
+ msg = &conn->ibc_connreq->cr_msg;
- PORTAL_ALLOC(svc, sizeof(*svc));
- if (svc == NULL)
- return (-ENOMEM);
+ kibnal_init_msg(msg, IBNAL_MSG_SVCQRY, 0);
+ kibnal_pack_msg(msg, 0, peer->ibp_nid, 0);
- memset (svc, 0, sizeof (*svc));
+ rc = kibnal_connect_sock(peer, &sock);
+ if (rc != 0)
+ return rc;
- svc->service_id = kibnal_data.kib_service_id;
+ rc = kibnal_sock_write(sock, msg, msg->ibm_nob);
+ if (rc != 0) {
+ CERROR("Error %d sending svcqry to "
+ LPX64"@%u.%u.%u.%u/%d\n", rc,
+ peer->ibp_nid, HIPQUAD(peer->ibp_ip), peer->ibp_port);
+ goto out;
+ }
- rc = ib_cached_gid_get(kibnal_data.kib_device,
- kibnal_data.kib_port,
- 0,
- svc->service_gid);
+ nob = offsetof(kib_msg_t, ibm_u) + sizeof(msg->ibm_u.svcrsp);
+ rc = kibnal_sock_read(sock, msg, nob, kibnal_tunables.kib_io_timeout);
if (rc != 0) {
- CERROR ("Can't get port %d GID: %d\n",
- kibnal_data.kib_port, rc);
+ CERROR("Error %d receiving svcrsp from "
+ LPX64"@%u.%u.%u.%u/%d\n", rc,
+ peer->ibp_nid, HIPQUAD(peer->ibp_ip), peer->ibp_port);
goto out;
}
-
- rc = ib_cached_pkey_get(kibnal_data.kib_device,
- kibnal_data.kib_port,
- 0,
- &svc->service_pkey);
+
+ rc = kibnal_unpack_msg(msg, nob);
if (rc != 0) {
- CERROR ("Can't get port %d PKEY: %d\n",
- kibnal_data.kib_port, rc);
+ CERROR("Error %d unpacking svcrsp from "
+ LPX64"@%u.%u.%u.%u/%d\n", rc,
+ peer->ibp_nid, HIPQUAD(peer->ibp_ip), peer->ibp_port);
+ goto out;
+ }
+
+ if (msg->ibm_type != IBNAL_MSG_SVCRSP) {
+ CERROR("Unexpected response type %d from "
+ LPX64"@%u.%u.%u.%u/%d\n", msg->ibm_type,
+ peer->ibp_nid, HIPQUAD(peer->ibp_ip), peer->ibp_port);
+ rc = -EPROTO;
goto out;
}
- svc->service_lease = 0xffffffff;
+ if (msg->ibm_dstnid != kibnal_lib.libnal_ni.ni_pid.nid ||
+ msg->ibm_dststamp != kibnal_data.kib_incarnation) {
+ CERROR("Unexpected dst NID/stamp "LPX64"/"LPX64" from "
+ LPX64"@%u.%u.%u.%u/%d\n",
+ msg->ibm_dstnid, msg->ibm_dststamp,
+ peer->ibp_nid, HIPQUAD(peer->ibp_ip), peer->ibp_port);
+ rc = -EPROTO;
+ goto out;
+ }
- kibnal_set_service_keys(svc, kibnal_data.kib_nid);
+ if (msg->ibm_srcnid != peer->ibp_nid) {
+ CERROR("Unexpected src NID "LPX64" from "
+ LPX64"@%u.%u.%u.%u/%d\n", msg->ibm_srcnid,
+ peer->ibp_nid, HIPQUAD(peer->ibp_ip), peer->ibp_port);
+ rc = -EPROTO;
+ goto out;
+ }
- CDEBUG(D_NET, "Advertising service id "LPX64" %s:"LPX64"\n",
- svc->service_id,
- svc->service_name, *kibnal_service_nid_field(svc));
+ conn->ibc_incarnation = msg->ibm_srcstamp;
+ conn->ibc_connreq->cr_svcrsp = msg->ibm_u.svcrsp;
+ out:
+ sock_release(sock);
+ return rc;
+}
- rc = ib_service_set (kibnal_data.kib_device,
- kibnal_data.kib_port,
- svc,
- IB_SA_SERVICE_COMP_MASK_ID |
- IB_SA_SERVICE_COMP_MASK_GID |
- IB_SA_SERVICE_COMP_MASK_PKEY |
- IB_SA_SERVICE_COMP_MASK_LEASE |
- KIBNAL_SERVICE_KEY_MASK,
- kibnal_tunables.kib_io_timeout * HZ,
- kibnal_service_setunset_done, &rc2, &tid);
+void
+kibnal_handle_svcqry (struct socket *sock)
+{
+ struct sockaddr_in addr;
+ __u32 peer_ip;
+ unsigned int peer_port;
+ kib_msg_t *msg;
+ __u64 srcnid;
+ __u64 srcstamp;
+ int len;
+ int rc;
+
+ len = sizeof(addr);
+ rc = sock->ops->getname(sock, (struct sockaddr *)&addr, &len, 2);
+ if (rc != 0) {
+ CERROR("Can't get peer's IP: %d\n", rc);
+ return;
+ }
+
+ peer_ip = ntohl(addr.sin_addr.s_addr);
+ peer_port = ntohs(addr.sin_port);
+
+ if (peer_port >= 1024) {
+ CERROR("Refusing unprivileged connection from %u.%u.%u.%u/%d\n",
+ HIPQUAD(peer_ip), peer_port);
+ return;
+ }
+ PORTAL_ALLOC(msg, sizeof(*msg));
+ if (msg == NULL) {
+ CERROR("Can't allocate msgs for %u.%u.%u.%u/%d\n",
+ HIPQUAD(peer_ip), peer_port);
+ goto out;
+ }
+
+ rc = kibnal_sock_read(sock, msg, offsetof(kib_msg_t, ibm_u),
+ kibnal_tunables.kib_listener_timeout);
if (rc != 0) {
- CERROR ("Immediate error %d advertising NID "LPX64"\n",
- rc, kibnal_data.kib_nid);
+ CERROR("Error %d receiving svcqry from %u.%u.%u.%u/%d\n",
+ rc, HIPQUAD(peer_ip), peer_port);
+ goto out;
+ }
+
+ rc = kibnal_unpack_msg(msg, offsetof(kib_msg_t, ibm_u));
+ if (rc != 0) {
+ CERROR("Error %d unpacking svcqry from %u.%u.%u.%u/%d\n",
+ rc, HIPQUAD(peer_ip), peer_port);
+ goto out;
+ }
+
+ if (msg->ibm_type != IBNAL_MSG_SVCQRY) {
+ CERROR("Unexpected message %d from %u.%u.%u.%u/%d\n",
+ msg->ibm_type, HIPQUAD(peer_ip), peer_port);
+ goto out;
+ }
+
+ if (msg->ibm_dstnid != kibnal_lib.libnal_ni.ni_pid.nid) {
+ CERROR("Unexpected dstnid "LPX64"(expected "LPX64" "
+ "from %u.%u.%u.%u/%d\n", msg->ibm_dstnid,
+ kibnal_lib.libnal_ni.ni_pid.nid,
+ HIPQUAD(peer_ip), peer_port);
goto out;
}
- down (&kibnal_data.kib_nid_signal);
+ srcnid = msg->ibm_srcnid;
+ srcstamp = msg->ibm_srcstamp;
+
+ kibnal_init_msg(msg, IBNAL_MSG_SVCRSP, sizeof(msg->ibm_u.svcrsp));
- rc = rc2;
- if (rc != 0)
- CERROR ("Error %d advertising NID "LPX64"\n",
- rc, kibnal_data.kib_nid);
+ msg->ibm_u.svcrsp.ibsr_svc_id = kibnal_data.kib_svc_id;
+ memcpy(msg->ibm_u.svcrsp.ibsr_svc_gid, kibnal_data.kib_svc_gid,
+ sizeof(kibnal_data.kib_svc_gid));
+ msg->ibm_u.svcrsp.ibsr_svc_pkey = kibnal_data.kib_svc_pkey;
+
+ kibnal_pack_msg(msg, 0, srcnid, srcstamp);
+
+ rc = kibnal_sock_write (sock, msg, msg->ibm_nob);
+ if (rc != 0) {
+ CERROR("Error %d replying to svcqry from %u.%u.%u.%u/%d\n",
+ rc, HIPQUAD(peer_ip), peer_port);
+ goto out;
+ }
+
out:
- PORTAL_FREE(svc, sizeof(*svc));
- return (rc);
+ PORTAL_FREE(msg, sizeof(*msg));
}
void
-kibnal_unadvertise (int expect_success)
+kibnal_free_acceptsock (kib_acceptsock_t *as)
{
- struct ib_common_attrib_service *svc;
- __u64 tid;
- int rc;
- int rc2;
+ sock_release(as->ibas_sock);
+ PORTAL_FREE(as, sizeof(*as));
+}
+
+int
+kibnal_ip_listener(void *arg)
+{
+ struct sockaddr_in addr;
+ wait_queue_t wait;
+ struct socket *sock;
+ kib_acceptsock_t *as;
+ int port;
+ char name[16];
+ int rc;
+ unsigned long flags;
+
+ /* Parent thread holds kib_nid_mutex, and is, or is about to
+ * block on kib_listener_signal */
+
+ port = kibnal_tunables.kib_port;
+ snprintf(name, sizeof(name), "kibnal_lstn%03d", port);
+ kportal_daemonize(name);
+ kportal_blockallsigs();
+
+ init_waitqueue_entry(&wait, current);
+
+ rc = kibnal_create_sock(&sock);
+ if (rc != 0)
+ goto out_0;
+
+ memset(&addr, 0, sizeof(addr));
+ addr.sin_family = AF_INET;
+ addr.sin_port = htons(port);
+ addr.sin_addr.s_addr = INADDR_ANY;
+
+ rc = sock->ops->bind(sock, (struct sockaddr *)&addr, sizeof(addr));
+ if (rc != 0) {
+ CERROR("Can't bind to port %d\n", port);
+ goto out_1;
+ }
+
+ rc = sock->ops->listen(sock, kibnal_tunables.kib_backlog);
+ if (rc != 0) {
+ CERROR("Can't set listen backlog %d: %d\n",
+ kibnal_tunables.kib_backlog, rc);
+ goto out_1;
+ }
+
+ LASSERT (kibnal_data.kib_listener_sock == NULL);
+ kibnal_data.kib_listener_sock = sock;
+
+ /* unblock waiting parent */
+ LASSERT (kibnal_data.kib_listener_shutdown == 0);
+ up(&kibnal_data.kib_listener_signal);
+
+ /* Wake me any time something happens on my socket */
+ add_wait_queue(sock->sk->sk_sleep, &wait);
+ as = NULL;
+
+ while (kibnal_data.kib_listener_shutdown == 0) {
+
+ if (as == NULL) {
+ PORTAL_ALLOC(as, sizeof(*as));
+ if (as == NULL) {
+ CERROR("Out of Memory: pausing...\n");
+ kibnal_pause(HZ);
+ continue;
+ }
+ as->ibas_sock = NULL;
+ }
+
+ if (as->ibas_sock == NULL) {
+ as->ibas_sock = sock_alloc();
+ if (as->ibas_sock == NULL) {
+ CERROR("Can't allocate socket: pausing...\n");
+ kibnal_pause(HZ);
+ continue;
+ }
+ /* XXX this should add a ref to sock->ops->owner, if
+ * TCP could be a module */
+ as->ibas_sock->type = sock->type;
+ as->ibas_sock->ops = sock->ops;
+ }
+
+ set_current_state(TASK_INTERRUPTIBLE);
+
+ rc = sock->ops->accept(sock, as->ibas_sock, O_NONBLOCK);
+
+ /* Sleep for socket activity? */
+ if (rc == -EAGAIN &&
+ kibnal_data.kib_listener_shutdown == 0)
+ schedule();
+
+ set_current_state(TASK_RUNNING);
+
+ if (rc == 0) {
+ spin_lock_irqsave(&kibnal_data.kib_connd_lock, flags);
+
+ list_add_tail(&as->ibas_list,
+ &kibnal_data.kib_connd_acceptq);
+
+ spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags);
+ wake_up(&kibnal_data.kib_connd_waitq);
- LASSERT (kibnal_data.kib_nid != PTL_NID_ANY);
+ as = NULL;
+ continue;
+ }
+
+ if (rc != -EAGAIN) {
+ CERROR("Accept failed: %d, pausing...\n", rc);
+ kibnal_pause(HZ);
+ }
+ }
+
+ if (as != NULL) {
+ if (as->ibas_sock != NULL)
+ sock_release(as->ibas_sock);
+ PORTAL_FREE(as, sizeof(*as));
+ }
+
+ rc = 0;
+ remove_wait_queue(sock->sk->sk_sleep, &wait);
+ out_1:
+ sock_release(sock);
+ kibnal_data.kib_listener_sock = NULL;
+ out_0:
+ /* set completion status and unblock thread waiting for me
+ * (parent on startup failure, executioner on normal shutdown) */
+ kibnal_data.kib_listener_shutdown = rc;
+ up(&kibnal_data.kib_listener_signal);
+
+ return 0;
+}
+
+int
+kibnal_start_ip_listener (void)
+{
+ long pid;
+ int rc;
+
+ CDEBUG(D_NET, "Starting listener\n");
+
+ /* Called holding kib_nid_mutex: listener stopped */
+ LASSERT (kibnal_data.kib_listener_sock == NULL);
+
+ kibnal_data.kib_listener_shutdown = 0;
+ pid = kernel_thread(kibnal_ip_listener, NULL, 0);
+ if (pid < 0) {
+ CERROR("Can't spawn listener: %ld\n", pid);
+ return (int)pid;
+ }
+
+ /* Block until listener has started up. */
+ down(&kibnal_data.kib_listener_signal);
+
+ rc = kibnal_data.kib_listener_shutdown;
+ LASSERT ((rc != 0) == (kibnal_data.kib_listener_sock == NULL));
+
+ CDEBUG((rc == 0) ? D_WARNING : D_ERROR,
+ "Listener %s: pid:%ld port:%d backlog:%d\n",
+ (rc == 0) ? "started OK" : "startup failed",
+ pid, kibnal_tunables.kib_port, kibnal_tunables.kib_backlog);
+
+ return rc;
+}
+
+void
+kibnal_stop_ip_listener(int clear_acceptq)
+{
+ struct list_head zombie_accepts;
+ kib_acceptsock_t *as;
+ unsigned long flags;
+
+ CDEBUG(D_NET, "Stopping listener\n");
+
+ /* Called holding kib_nid_mutex: listener running */
+ LASSERT (kibnal_data.kib_listener_sock != NULL);
- PORTAL_ALLOC(svc, sizeof(*svc));
- if (svc == NULL)
+ kibnal_data.kib_listener_shutdown = 1;
+ wake_up_all(kibnal_data.kib_listener_sock->sk->sk_sleep);
+
+ /* Block until listener has torn down. */
+ down(&kibnal_data.kib_listener_signal);
+
+ LASSERT (kibnal_data.kib_listener_sock == NULL);
+ CDEBUG(D_WARNING, "Listener stopped\n");
+
+ if (!clear_acceptq)
return;
- memset (svc, 0, sizeof(*svc));
+ /* Close any unhandled accepts */
+ spin_lock_irqsave(&kibnal_data.kib_connd_lock, flags);
+
+ list_add(&zombie_accepts, &kibnal_data.kib_connd_acceptq);
+ list_del_init(&kibnal_data.kib_connd_acceptq);
+
+ spin_unlock_irqrestore(&kibnal_data.kib_connd_lock, flags);
+
+ while (!list_empty(&zombie_accepts)) {
+ as = list_entry(zombie_accepts.next,
+ kib_acceptsock_t, ibas_list);
+ list_del(&as->ibas_list);
+ kibnal_free_acceptsock(as);
+ }
+}
+
+int
+kibnal_listener_procint(ctl_table *table, int write, struct file *filp,
+ void *buffer, size_t *lenp)
+{
+ int *tunable = (int *)table->data;
+ int old_val;
+ int rc;
- kibnal_set_service_keys(svc, kibnal_data.kib_nid);
+ /* No race with nal initialisation since the nal is setup all the time
+ * it's loaded. When that changes, change this! */
+ LASSERT (kibnal_data.kib_init == IBNAL_INIT_ALL);
- CDEBUG(D_NET, "Unadvertising service %s:"LPX64"\n",
- svc->service_name, *kibnal_service_nid_field(svc));
+ down(&kibnal_data.kib_nid_mutex);
- rc = ib_service_delete (kibnal_data.kib_device,
- kibnal_data.kib_port,
- svc,
- KIBNAL_SERVICE_KEY_MASK,
- kibnal_tunables.kib_io_timeout * HZ,
- kibnal_service_setunset_done, &rc2, &tid);
+ LASSERT (tunable == &kibnal_tunables.kib_port ||
+ tunable == &kibnal_tunables.kib_backlog);
+ old_val = *tunable;
+
+ rc = proc_dointvec(table, write, filp, buffer, lenp);
+
+ if (write &&
+ (*tunable != old_val ||
+ kibnal_data.kib_listener_sock == NULL)) {
+
+ if (kibnal_data.kib_listener_sock != NULL)
+ kibnal_stop_ip_listener(0);
+
+ rc = kibnal_start_ip_listener();
+ if (rc != 0) {
+ CERROR("Unable to restart listener with new tunable:"
+ " reverting to old value\n");
+ *tunable = old_val;
+ kibnal_start_ip_listener();
+ }
+ }
+
+ up(&kibnal_data.kib_nid_mutex);
+
+ LASSERT (kibnal_data.kib_init == IBNAL_INIT_ALL);
+ return rc;
+}
+
+int
+kibnal_start_ib_listener (void)
+{
+ int rc;
+
+ LASSERT (kibnal_data.kib_listen_handle == NULL);
+
+ kibnal_data.kib_svc_id = ib_cm_service_assign();
+ CDEBUG(D_NET, "svc id "LPX64"\n", kibnal_data.kib_svc_id);
+
+ rc = ib_cached_gid_get(kibnal_data.kib_device,
+ kibnal_data.kib_port, 0,
+ kibnal_data.kib_svc_gid);
if (rc != 0) {
- CERROR ("Immediate error %d unadvertising NID "LPX64"\n",
- rc, kibnal_data.kib_nid);
- goto out;
+ CERROR("Can't get port %d GID: %d\n",
+ kibnal_data.kib_port, rc);
+ return rc;
+ }
+
+ rc = ib_cached_pkey_get(kibnal_data.kib_device,
+ kibnal_data.kib_port, 0,
+ &kibnal_data.kib_svc_pkey);
+ if (rc != 0) {
+ CERROR ("Can't get port %d PKEY: %d\n",
+ kibnal_data.kib_port, rc);
+ return rc;
+ }
+
+ rc = ib_cm_listen(kibnal_data.kib_svc_id,
+ TS_IB_CM_SERVICE_EXACT_MASK,
+ kibnal_passive_conn_callback, NULL,
+ &kibnal_data.kib_listen_handle);
+ if (rc != 0) {
+ kibnal_data.kib_listen_handle = NULL;
+ CERROR ("Can't create IB listener: %d\n", rc);
+ return rc;
}
+
+ LASSERT (kibnal_data.kib_listen_handle != NULL);
+ return 0;
+}
- down (&kibnal_data.kib_nid_signal);
+void
+kibnal_stop_ib_listener (void)
+{
+ int rc;
- if ((rc2 == 0) == !!expect_success)
- goto out; /* success: rc == 0 */
-
- if (expect_success)
- CERROR("Error %d unadvertising NID "LPX64"\n",
- rc, kibnal_data.kib_nid);
- else
- CWARN("Removed conflicting NID "LPX64"\n",
- kibnal_data.kib_nid);
- out:
- PORTAL_FREE(svc, sizeof(*svc));
+ LASSERT (kibnal_data.kib_listen_handle != NULL);
+
+ rc = ib_cm_listen_stop (kibnal_data.kib_listen_handle);
+ if (rc != 0)
+ CERROR("Error stopping IB listener: %d\n", rc);
+
+ kibnal_data.kib_listen_handle = NULL;
}
int
-kibnal_set_mynid(ptl_nid_t nid)
+kibnal_set_mynid (ptl_nid_t nid)
{
- struct timeval tv;
- lib_ni_t *ni = &kibnal_lib.libnal_ni;
- int rc;
+ lib_ni_t *ni = &kibnal_lib.libnal_ni;
+ int rc;
CDEBUG(D_IOCTL, "setting mynid to "LPX64" (old nid="LPX64")\n",
nid, ni->ni_pid.nid);
- do_gettimeofday(&tv);
-
down (&kibnal_data.kib_nid_mutex);
if (nid == kibnal_data.kib_nid) {
CDEBUG(D_NET, "NID "LPX64"("LPX64")\n",
kibnal_data.kib_nid, nid);
-
- if (kibnal_data.kib_nid != PTL_NID_ANY) {
- kibnal_unadvertise (1);
-
- rc = ib_cm_listen_stop (kibnal_data.kib_listen_handle);
- if (rc != 0)
- CERROR ("Error %d stopping listener\n", rc);
- }
-
- kibnal_data.kib_nid = ni->ni_pid.nid = nid;
- kibnal_data.kib_incarnation = (((__u64)tv.tv_sec) * 1000000) + tv.tv_usec;
+ if (kibnal_data.kib_listener_sock != NULL)
+ kibnal_stop_ip_listener(1);
+ if (kibnal_data.kib_listen_handle != NULL)
+ kibnal_stop_ib_listener();
+
+ ni->ni_pid.nid = nid;
+ kibnal_data.kib_incarnation++;
+ mb();
/* Delete all existing peers and their connections after new
- * NID/incarnation set to ensure no old connections in our brave
- * new world. */
+ * NID/incarnation set to ensure no old connections in our brave new
+ * world. */
kibnal_del_peer (PTL_NID_ANY, 0);
- if (kibnal_data.kib_nid == PTL_NID_ANY) {
- /* No new NID to install */
- up (&kibnal_data.kib_nid_mutex);
- return (0);
- }
-
- /* remove any previous advert (crashed node etc) */
- kibnal_unadvertise(0);
-
- /* Assign new service number */
- kibnal_data.kib_service_id = ib_cm_service_assign();
- CDEBUG(D_NET, "service_id "LPX64"\n", kibnal_data.kib_service_id);
+ if (ni->ni_pid.nid != PTL_NID_ANY) {
+ /* got a new NID to install */
+ rc = kibnal_start_ib_listener();
+ if (rc != 0) {
+ CERROR("Can't start IB listener: %d\n", rc);
+ goto failed_0;
+ }
- rc = ib_cm_listen(kibnal_data.kib_service_id,
- TS_IB_CM_SERVICE_EXACT_MASK,
- kibnal_passive_conn_callback, NULL,
- &kibnal_data.kib_listen_handle);
- if (rc == 0) {
- rc = kibnal_advertise();
- if (rc == 0) {
-#if IBNAL_CHECK_ADVERT
- kibnal_check_advert();
-#endif
- up (&kibnal_data.kib_nid_mutex);
- return (0);
+ rc = kibnal_start_ip_listener();
+ if (rc != 0) {
+ CERROR("Can't start IP listener: %d\n", rc);
+ goto failed_1;
}
-
- ib_cm_listen_stop(kibnal_data.kib_listen_handle);
- /* remove any peers that sprung up while I failed to
- * advertise myself */
- kibnal_del_peer (PTL_NID_ANY, 0);
}
- kibnal_data.kib_nid = PTL_NID_ANY;
- up (&kibnal_data.kib_nid_mutex);
- return (rc);
+ up(&kibnal_data.kib_nid_mutex);
+ return 0;
+
+ failed_1:
+ kibnal_stop_ib_listener();
+ failed_0:
+ ni->ni_pid.nid = PTL_NID_ANY;
+ kibnal_data.kib_incarnation++;
+ mb();
+ kibnal_del_peer (PTL_NID_ANY, 0);
+ up(&kibnal_data.kib_nid_mutex);
+ return rc;
}
kib_peer_t *
peer->ibp_reconnect_interval = IBNAL_MIN_RECONNECT_INTERVAL;
atomic_inc (&kibnal_data.kib_npeers);
+ CDEBUG(D_NET, "peer %p "LPX64"\n", peer, nid);
+
return (peer);
}
}
int
-kibnal_get_peer_info (int index, ptl_nid_t *nidp, int *persistencep)
+kibnal_get_peer_info (int index, ptl_nid_t *nidp, __u32 *ipp, int *portp,
+ int *persistencep)
{
kib_peer_t *peer;
struct list_head *ptmp;
continue;
*nidp = peer->ibp_nid;
+ *ipp = peer->ibp_ip;
+ *portp = peer->ibp_port;
*persistencep = peer->ibp_persistence;
read_unlock (&kibnal_data.kib_global_lock);
}
int
-kibnal_add_persistent_peer (ptl_nid_t nid)
+kibnal_add_persistent_peer (ptl_nid_t nid, __u32 ip, int port)
{
unsigned long flags;
kib_peer_t *peer;
kibnal_nid2peerlist (nid));
}
+ peer->ibp_ip = ip;
+ peer->ibp_port = port;
peer->ibp_persistence++;
write_unlock_irqrestore (&kibnal_data.kib_global_lock, flags);
if (peer->ibp_persistence != 0)
return;
- list_for_each_safe (ctmp, cnxt, &peer->ibp_conns) {
- conn = list_entry(ctmp, kib_conn_t, ibc_list);
+ if (list_empty(&peer->ibp_conns)) {
+ kibnal_unlink_peer_locked(peer);
+ } else {
+ list_for_each_safe (ctmp, cnxt, &peer->ibp_conns) {
+ conn = list_entry(ctmp, kib_conn_t, ibc_list);
- kibnal_close_conn_locked (conn, 0);
+ kibnal_close_conn_locked (conn, 0);
+ }
+ /* NB peer is no longer persistent; closing its last conn
+ * unlinked it. */
}
-
- /* NB peer unlinks itself when last conn is closed */
+ /* NB peer now unlinked; might even be freed if the peer table had the
+ * last ref on it. */
}
int
/* I just nuked the last connection on shutdown; wake up
* everyone so they can exit. */
wake_up_all(&kibnal_data.kib_sched_waitq);
- wake_up_all(&kibnal_data.kib_connd_waitq);
+ wake_up_all(&kibnal_data.kib_reaper_waitq);
}
}
/* last ref only goes on zombies */
LASSERT (conn->ibc_state == IBNAL_CONN_ZOMBIE);
- spin_lock_irqsave (&kibnal_data.kib_connd_lock, flags);
+ spin_lock_irqsave (&kibnal_data.kib_reaper_lock, flags);
- list_add (&conn->ibc_list, &kibnal_data.kib_connd_conns);
- wake_up (&kibnal_data.kib_connd_waitq);
+ list_add (&conn->ibc_list, &kibnal_data.kib_reaper_conns);
+ wake_up (&kibnal_data.kib_reaper_waitq);
- spin_unlock_irqrestore (&kibnal_data.kib_connd_lock, flags);
+ spin_unlock_irqrestore (&kibnal_data.kib_reaper_lock, flags);
}
int
switch(pcfg->pcfg_command) {
case NAL_CMD_GET_PEER: {
ptl_nid_t nid = 0;
+ __u32 ip = 0;
+ int port = 0;
int share_count = 0;
rc = kibnal_get_peer_info(pcfg->pcfg_count,
- &nid, &share_count);
+ &nid, &ip, &port, &share_count);
pcfg->pcfg_nid = nid;
pcfg->pcfg_size = 0;
- pcfg->pcfg_id = 0;
- pcfg->pcfg_misc = 0;
+ pcfg->pcfg_id = ip;
+ pcfg->pcfg_misc = port;
pcfg->pcfg_count = 0;
pcfg->pcfg_wait = share_count;
break;
}
case NAL_CMD_ADD_PEER: {
- rc = kibnal_add_persistent_peer (pcfg->pcfg_nid);
+ rc = kibnal_add_persistent_peer (pcfg->pcfg_nid,
+ pcfg->pcfg_id, /* IP */
+ pcfg->pcfg_misc); /* port */
break;
}
case NAL_CMD_DEL_PEER: {
libcfs_nal_cmd_unregister(OPENIBNAL);
/* No new peers */
- /* resetting my NID to unadvertises me, removes my
+ /* resetting my NID unadvertises me, removes my
* listener and nukes all current peers */
kibnal_set_mynid (PTL_NID_ANY);
LASSERT (atomic_read (&kibnal_data.kib_nconns) == 0);
LASSERT (list_empty (&kibnal_data.kib_sched_rxq));
LASSERT (list_empty (&kibnal_data.kib_sched_txq));
- LASSERT (list_empty (&kibnal_data.kib_connd_conns));
+ LASSERT (list_empty (&kibnal_data.kib_reaper_conns));
LASSERT (list_empty (&kibnal_data.kib_connd_peers));
+ LASSERT (list_empty (&kibnal_data.kib_connd_acceptq));
/* flag threads to terminate; wake and wait for them to die */
kibnal_data.kib_shutdown = 1;
wake_up_all (&kibnal_data.kib_sched_waitq);
+ wake_up_all (&kibnal_data.kib_reaper_waitq);
wake_up_all (&kibnal_data.kib_connd_waitq);
i = 2;
ptl_ni_limits_t *requested_limits,
ptl_ni_limits_t *actual_limits)
{
+ struct timeval tv;
ptl_process_id_t process_id;
int pkmem = atomic_read(&portal_kmemory);
int rc;
memset (&kibnal_data, 0, sizeof (kibnal_data)); /* zero pointers, flags etc */
+ do_gettimeofday(&tv);
+ kibnal_data.kib_incarnation = (((__u64)tv.tv_sec) * 1000000) + tv.tv_usec;
+
init_MUTEX (&kibnal_data.kib_nid_mutex);
- init_MUTEX_LOCKED (&kibnal_data.kib_nid_signal);
- kibnal_data.kib_nid = PTL_NID_ANY;
+ init_MUTEX_LOCKED (&kibnal_data.kib_listener_signal);
rwlock_init(&kibnal_data.kib_global_lock);
for (i = 0; i < kibnal_data.kib_peer_hash_size; i++)
INIT_LIST_HEAD(&kibnal_data.kib_peers[i]);
+ spin_lock_init (&kibnal_data.kib_reaper_lock);
+ INIT_LIST_HEAD (&kibnal_data.kib_reaper_conns);
+ init_waitqueue_head (&kibnal_data.kib_reaper_waitq);
+
spin_lock_init (&kibnal_data.kib_connd_lock);
+ INIT_LIST_HEAD (&kibnal_data.kib_connd_acceptq);
INIT_LIST_HEAD (&kibnal_data.kib_connd_peers);
- INIT_LIST_HEAD (&kibnal_data.kib_connd_conns);
init_waitqueue_head (&kibnal_data.kib_connd_waitq);
spin_lock_init (&kibnal_data.kib_sched_lock);
process_id.pid = requested_pid;
- process_id.nid = kibnal_data.kib_nid;
+ process_id.nid = PTL_NID_ANY; /* don't know my NID yet */
rc = lib_init(&kibnal_lib, nal, process_id,
requested_limits, actual_limits);
/*****************************************************/
for (i = 0; i < IBNAL_N_SCHED; i++) {
- rc = kibnal_thread_start (kibnal_scheduler, (void *)i);
+ rc = kibnal_thread_start (kibnal_scheduler,
+ (void *)((unsigned long)i));
if (rc != 0) {
CERROR("Can't spawn openibnal scheduler[%d]: %d\n",
i, rc);
}
}
- rc = kibnal_thread_start (kibnal_connd, NULL);
+ for (i = 0; i < IBNAL_N_CONND; i++) {
+ rc = kibnal_thread_start (kibnal_connd,
+ (void *)((unsigned long)i));
+ if (rc != 0) {
+ CERROR("Can't spawn openibnal connd[%d]: %d\n",
+ i, rc);
+ goto failed;
+ }
+ }
+
+ rc = kibnal_thread_start (kibnal_reaper, NULL);
if (rc != 0) {
- CERROR ("Can't spawn openibnal connd: %d\n", rc);
+ CERROR ("Can't spawn openibnal reaper: %d\n", rc);
goto failed;
}
void __exit
kibnal_module_fini (void)
{
-#ifdef CONFIG_SYSCTL
if (kibnal_tunables.kib_sysctl != NULL)
unregister_sysctl_table (kibnal_tunables.kib_sysctl);
-#endif
PtlNIFini(kibnal_ni);
ptl_unregister_nal(OPENIBNAL);
int rc;
/* the following must be sizeof(int) for proc_dointvec() */
- LASSERT(sizeof (kibnal_tunables.kib_io_timeout) == sizeof (int));
+ LASSERT (sizeof(kibnal_tunables.kib_io_timeout) == sizeof(int));
+ LASSERT (sizeof(kibnal_tunables.kib_listener_timeout) == sizeof(int));
+ LASSERT (sizeof(kibnal_tunables.kib_backlog) == sizeof(int));
+ LASSERT (sizeof(kibnal_tunables.kib_port) == sizeof(int));
kibnal_api.nal_ni_init = kibnal_api_startup;
kibnal_api.nal_ni_fini = kibnal_api_shutdown;
/* Initialise dynamic tunables to defaults once only */
kibnal_tunables.kib_io_timeout = IBNAL_IO_TIMEOUT;
+ kibnal_tunables.kib_listener_timeout = IBNAL_LISTENER_TIMEOUT;
+ kibnal_tunables.kib_backlog = IBNAL_BACKLOG;
+ kibnal_tunables.kib_port = IBNAL_PORT;
rc = ptl_register_nal(OPENIBNAL, &kibnal_api);
if (rc != PTL_OK) {
return (-ENODEV);
}
-#ifdef CONFIG_SYSCTL
- /* Press on regardless even if registering sysctl doesn't work */
kibnal_tunables.kib_sysctl =
register_sysctl_table (kibnal_top_ctl_table, 0);
-#endif
+ if (kibnal_tunables.kib_sysctl == NULL) {
+ CERROR("Can't register sysctl table\n");
+ PtlNIFini(kibnal_ni);
+ ptl_unregister_nal(OPENIBNAL);
+ return (-ENOMEM);
+ }
+
return (0);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff