From 1c9b281997b5876c0c8ed62506b56db89d262b57 Mon Sep 17 00:00:00 2001 From: Jason Gunthorpe Date: Thu, 19 Nov 2009 12:55:21 -0800 Subject: RDMA/cma: Correct detection of SA Created MGID RDMA CM treats AF_INET6 addresses that are either 0 or prefixed with FF1x:A01B::/32 as MGIDs, but the detection for the prefix was buggy; fix it up. Signed-off-by: Jason Gunthorpe Signed-off-by: Sean Hefty Signed-off-by: Roland Dreier --- drivers/infiniband/core/cma.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/infiniband/core/cma.c') diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index 0753178..8bb2cf4 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -2687,7 +2687,7 @@ static void cma_set_mgid(struct rdma_id_private *id_priv, if (cma_any_addr(addr)) { memset(mgid, 0, sizeof *mgid); } else if ((addr->sa_family == AF_INET6) && - ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFF10A01B) == + ((be32_to_cpu(sin6->sin6_addr.s6_addr32[0]) & 0xFFF0FFFF) == 0xFF10A01B)) { /* IPv6 address is an SA assigned MGID. */ memcpy(mgid, &sin6->sin6_addr, sizeof *mgid); -- cgit v1.1 From e2e626972e652d18520f84d69fc06cfa307d11ff Mon Sep 17 00:00:00 2001 From: Jason Gunthorpe Date: Thu, 19 Nov 2009 12:55:22 -0800 Subject: RDMA/cma: Fix AF_INET6 support in multicast joining If joining to an AF_INET6 address, we need to map the address to a MGID in the same way as the IP stack. The old code would just fall through to the IPv4 case and generate garbage. Signed-off-by: Jason Gunthorpe Signed-off-by: Sean Hefty Signed-off-by: Roland Dreier --- drivers/infiniband/core/cma.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'drivers/infiniband/core/cma.c') diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index 8bb2cf4..052b4c0 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -2691,6 +2691,11 @@ static void cma_set_mgid(struct rdma_id_private *id_priv, 0xFF10A01B)) { /* IPv6 address is an SA assigned MGID. */ memcpy(mgid, &sin6->sin6_addr, sizeof *mgid); + } else if ((addr->sa_family == AF_INET6)) { + ipv6_ib_mc_map(&sin6->sin6_addr, dev_addr->broadcast, mc_map); + if (id_priv->id.ps == RDMA_PS_UDP) + mc_map[7] = 0x01; /* Use RDMA CM signature */ + *mgid = *(union ib_gid *) (mc_map + 4); } else { ip_ib_mc_map(sin->sin_addr.s_addr, dev_addr->broadcast, mc_map); if (id_priv->id.ps == RDMA_PS_UDP) -- cgit v1.1 From 6266ed6e4164466177238b11ecb825a3a108a3e4 Mon Sep 17 00:00:00 2001 From: Sean Hefty Date: Thu, 19 Nov 2009 12:55:22 -0800 Subject: RDMA/cma: Replace net_device pointer with index Provide the device interface when resolving route information to ensure that the correct outbound device is used. This will also simplify processing of sin6_scope_id for IPv6 support. Based on work from: David Wilder Jason Gunthorpe Signed-off-by: Sean Hefty Signed-off-by: Roland Dreier --- drivers/infiniband/core/cma.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/infiniband/core/cma.c') diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index 052b4c0..699ad12 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -2820,7 +2820,7 @@ static int cma_netdev_change(struct net_device *ndev, struct rdma_id_private *id dev_addr = &id_priv->id.route.addr.dev_addr; - if ((dev_addr->src_dev == ndev) && + if ((dev_addr->bound_dev_if == ndev->ifindex) && memcmp(dev_addr->src_dev_addr, ndev->dev_addr, ndev->addr_len)) { printk(KERN_INFO "RDMA CM addr change for ndev %s used by id %p\n", ndev->name, &id_priv->id); -- cgit v1.1 From c4315d85f9b76834289fd503796c01b8311c4b84 Mon Sep 17 00:00:00 2001 From: Sean Hefty Date: Thu, 19 Nov 2009 12:57:18 -0800 Subject: IB/addr: Store net_device type instead of translating to RDMA transport The struct rdma_dev_addr stores net_device address information: the source device address, destination hardware address, and broadcast address. For consistency, store the net_device type rather than converting it to the rdma_node_type. The type indicates the format of the various hardware addresses, which is what we're concerned with, and not the RDMA node type that the address may map to. Signed-off-by: Sean Hefty Signed-off-by: Roland Dreier --- drivers/infiniband/core/cma.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'drivers/infiniband/core/cma.c') diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index 699ad12..b305b5c 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -330,11 +330,11 @@ static int cma_acquire_dev(struct rdma_id_private *id_priv) union ib_gid gid; int ret = -ENODEV; - switch (rdma_node_get_transport(dev_addr->dev_type)) { - case RDMA_TRANSPORT_IB: + switch (dev_addr->dev_type) { + case ARPHRD_INFINIBAND: ib_addr_get_sgid(dev_addr, &gid); break; - case RDMA_TRANSPORT_IWARP: + case ARPHRD_ETHER: iw_addr_get_sgid(dev_addr, &gid); break; default: -- cgit v1.1 From 6f8372b69c3198e06cecb1df2cb9682d0c55e657 Mon Sep 17 00:00:00 2001 From: Sean Hefty Date: Thu, 19 Nov 2009 13:26:06 -0800 Subject: RDMA/cm: fix loopback address support The RDMA CM is intended to support the use of a loopback address when establishing a connection; however, the behavior of the CM when loopback addresses are used is confusing and does not always work, depending on whether loopback was specified by the server, the client, or both. The defined behavior of rdma_bind_addr is to associate an RDMA device with an rdma_cm_id, as long as the user specified a non- zero address. (ie they weren't just trying to reserve a port) Currently, if the loopback address is passed to rdam_bind_addr, no device is associated with the rdma_cm_id. Fix this. If a loopback address is specified by the client as the destination address for a connection, it will fail to establish a connection. This is true even if the server is listing across all addresses or on the loopback address itself. The issue is that the server tries to translate the IP address carried in the REQ message to a local net_device address, which fails. The translation is not needed in this case, since the REQ carries the actual HW address that should be used. Finally, cleanup loopback support to be more transport neutral. Replace separate calls to get/set the sgid and dgid from the device address to a single call that behaves correctly depending on the format of the device address. And support both IPv4 and IPv6 address formats. Signed-off-by: Sean Hefty [ Fixed RDS build by s/ib_addr_get/rdma_addr_get/ - Roland ] Signed-off-by: Roland Dreier --- drivers/infiniband/core/cma.c | 77 ++++++++++++++++++++++++------------------- 1 file changed, 43 insertions(+), 34 deletions(-) (limited to 'drivers/infiniband/core/cma.c') diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index b305b5c..38867a4 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -330,17 +330,7 @@ static int cma_acquire_dev(struct rdma_id_private *id_priv) union ib_gid gid; int ret = -ENODEV; - switch (dev_addr->dev_type) { - case ARPHRD_INFINIBAND: - ib_addr_get_sgid(dev_addr, &gid); - break; - case ARPHRD_ETHER: - iw_addr_get_sgid(dev_addr, &gid); - break; - default: - return -ENODEV; - } - + rdma_addr_get_sgid(dev_addr, &gid); list_for_each_entry(cma_dev, &dev_list, list) { ret = ib_find_cached_gid(cma_dev->device, &gid, &id_priv->id.port_num, NULL); @@ -1032,11 +1022,17 @@ static struct rdma_id_private *cma_new_conn_id(struct rdma_cm_id *listen_id, if (rt->num_paths == 2) rt->path_rec[1] = *ib_event->param.req_rcvd.alternate_path; - ib_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid); - ret = rdma_translate_ip((struct sockaddr *) &id->route.addr.src_addr, - &id->route.addr.dev_addr); - if (ret) - goto destroy_id; + if (cma_any_addr((struct sockaddr *) &rt->addr.src_addr)) { + rt->addr.dev_addr.dev_type = ARPHRD_INFINIBAND; + rdma_addr_set_sgid(&rt->addr.dev_addr, &rt->path_rec[0].sgid); + ib_addr_set_pkey(&rt->addr.dev_addr, rt->path_rec[0].pkey); + } else { + ret = rdma_translate_ip((struct sockaddr *) &rt->addr.src_addr, + &rt->addr.dev_addr); + if (ret) + goto destroy_id; + } + rdma_addr_set_dgid(&rt->addr.dev_addr, &rt->path_rec[0].dgid); id_priv = container_of(id, struct rdma_id_private, id); id_priv->state = CMA_CONNECT; @@ -1071,10 +1067,12 @@ static struct rdma_id_private *cma_new_udp_id(struct rdma_cm_id *listen_id, cma_save_net_info(&id->route.addr, &listen_id->route.addr, ip_ver, port, src, dst); - ret = rdma_translate_ip((struct sockaddr *) &id->route.addr.src_addr, - &id->route.addr.dev_addr); - if (ret) - goto err; + if (!cma_any_addr((struct sockaddr *) &id->route.addr.src_addr)) { + ret = rdma_translate_ip((struct sockaddr *) &id->route.addr.src_addr, + &id->route.addr.dev_addr); + if (ret) + goto err; + } id_priv = container_of(id, struct rdma_id_private, id); id_priv->state = CMA_CONNECT; @@ -1565,8 +1563,8 @@ static int cma_query_ib_route(struct rdma_id_private *id_priv, int timeout_ms, struct sockaddr_in6 *sin6; memset(&path_rec, 0, sizeof path_rec); - ib_addr_get_sgid(&addr->dev_addr, &path_rec.sgid); - ib_addr_get_dgid(&addr->dev_addr, &path_rec.dgid); + rdma_addr_get_sgid(&addr->dev_addr, &path_rec.sgid); + rdma_addr_get_dgid(&addr->dev_addr, &path_rec.dgid); path_rec.pkey = cpu_to_be16(ib_addr_get_pkey(&addr->dev_addr)); path_rec.numb_path = 1; path_rec.reversible = 1; @@ -1781,7 +1779,11 @@ port_found: if (ret) goto out; - ib_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid); + id_priv->id.route.addr.dev_addr.dev_type = + (rdma_node_get_transport(cma_dev->device->node_type) == RDMA_TRANSPORT_IB) ? + ARPHRD_INFINIBAND : ARPHRD_ETHER; + + rdma_addr_set_sgid(&id_priv->id.route.addr.dev_addr, &gid); ib_addr_set_pkey(&id_priv->id.route.addr.dev_addr, pkey); id_priv->id.port_num = p; cma_attach_to_dev(id_priv, cma_dev); @@ -1839,7 +1841,7 @@ out: static int cma_resolve_loopback(struct rdma_id_private *id_priv) { struct cma_work *work; - struct sockaddr_in *src_in, *dst_in; + struct sockaddr *src, *dst; union ib_gid gid; int ret; @@ -1853,14 +1855,19 @@ static int cma_resolve_loopback(struct rdma_id_private *id_priv) goto err; } - ib_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid); - ib_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid); + rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid); + rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid); - if (cma_zero_addr((struct sockaddr *) &id_priv->id.route.addr.src_addr)) { - src_in = (struct sockaddr_in *)&id_priv->id.route.addr.src_addr; - dst_in = (struct sockaddr_in *)&id_priv->id.route.addr.dst_addr; - src_in->sin_family = dst_in->sin_family; - src_in->sin_addr.s_addr = dst_in->sin_addr.s_addr; + src = (struct sockaddr *) &id_priv->id.route.addr.src_addr; + if (cma_zero_addr(src)) { + dst = (struct sockaddr *) &id_priv->id.route.addr.dst_addr; + if ((src->sa_family = dst->sa_family) == AF_INET) { + ((struct sockaddr_in *) src)->sin_addr.s_addr = + ((struct sockaddr_in *) dst)->sin_addr.s_addr; + } else { + ipv6_addr_copy(&((struct sockaddr_in6 *) src)->sin6_addr, + &((struct sockaddr_in6 *) dst)->sin6_addr); + } } work->id = id_priv; @@ -2089,7 +2096,9 @@ int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr) if (!cma_comp_exch(id_priv, CMA_IDLE, CMA_ADDR_BOUND)) return -EINVAL; - if (!cma_any_addr(addr)) { + if (cma_loopback_addr(addr)) { + ret = cma_bind_loopback(id_priv); + } else if (!cma_zero_addr(addr)) { ret = rdma_translate_ip(addr, &id->route.addr.dev_addr); if (ret) goto err1; @@ -2108,7 +2117,7 @@ int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr) return 0; err2: - if (!cma_any_addr(addr)) { + if (id_priv->cma_dev) { mutex_lock(&lock); cma_detach_from_dev(id_priv); mutex_unlock(&lock); @@ -2721,7 +2730,7 @@ static int cma_join_ib_multicast(struct rdma_id_private *id_priv, cma_set_mgid(id_priv, (struct sockaddr *) &mc->addr, &rec.mgid); if (id_priv->id.ps == RDMA_PS_UDP) rec.qkey = cpu_to_be32(RDMA_UDP_QKEY); - ib_addr_get_sgid(dev_addr, &rec.port_gid); + rdma_addr_get_sgid(dev_addr, &rec.port_gid); rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr)); rec.join_state = 1; -- cgit v1.1 From d14714df61681cfecf945a58436edf197327e87f Mon Sep 17 00:00:00 2001 From: Sean Hefty Date: Thu, 19 Nov 2009 16:46:25 -0800 Subject: IB/addr: Fix IPv6 routing lookup Include link scope as part of address resolution. Combine local and remote address resolution into a single, simpler code path. Fix error checking in the IPv6 routing lookups. Based on work from: David Wilder Jason Gunthorpe Signed-off-by: Sean Hefty [ Fix up cma_check_linklocal() for !IPV6 case. - Roland ] Signed-off-by: Roland Dreier --- drivers/infiniband/core/cma.c | 47 ++++++++++++++++++++++++++++++------------- 1 file changed, 33 insertions(+), 14 deletions(-) (limited to 'drivers/infiniband/core/cma.c') diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index 38867a4..fbdd731 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -1472,15 +1472,6 @@ static void cma_listen_on_all(struct rdma_id_private *id_priv) mutex_unlock(&lock); } -static int cma_bind_any(struct rdma_cm_id *id, sa_family_t af) -{ - struct sockaddr_storage addr_in; - - memset(&addr_in, 0, sizeof addr_in); - addr_in.ss_family = af; - return rdma_bind_addr(id, (struct sockaddr *) &addr_in); -} - int rdma_listen(struct rdma_cm_id *id, int backlog) { struct rdma_id_private *id_priv; @@ -1488,7 +1479,8 @@ int rdma_listen(struct rdma_cm_id *id, int backlog) id_priv = container_of(id, struct rdma_id_private, id); if (id_priv->state == CMA_IDLE) { - ret = cma_bind_any(id, AF_INET); + ((struct sockaddr *) &id->route.addr.src_addr)->sa_family = AF_INET; + ret = rdma_bind_addr(id, (struct sockaddr *) &id->route.addr.src_addr); if (ret) return ret; } @@ -1885,10 +1877,14 @@ err: static int cma_bind_addr(struct rdma_cm_id *id, struct sockaddr *src_addr, struct sockaddr *dst_addr) { - if (src_addr && src_addr->sa_family) - return rdma_bind_addr(id, src_addr); - else - return cma_bind_any(id, dst_addr->sa_family); + if (!src_addr || !src_addr->sa_family) { + src_addr = (struct sockaddr *) &id->route.addr.src_addr; + if ((src_addr->sa_family = dst_addr->sa_family) == AF_INET6) { + ((struct sockaddr_in6 *) src_addr)->sin6_scope_id = + ((struct sockaddr_in6 *) dst_addr)->sin6_scope_id; + } + } + return rdma_bind_addr(id, src_addr); } int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr, @@ -2084,6 +2080,25 @@ static int cma_get_port(struct rdma_id_private *id_priv) return ret; } +static int cma_check_linklocal(struct rdma_dev_addr *dev_addr, + struct sockaddr *addr) +{ +#if defined(CONFIG_IPv6) || defined(CONFIG_IPV6_MODULE) + struct sockaddr_in6 *sin6; + + if (addr->sa_family != AF_INET6) + return 0; + + sin6 = (struct sockaddr_in6 *) addr; + if ((ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL) && + !sin6->sin6_scope_id) + return -EINVAL; + + dev_addr->bound_dev_if = sin6->sin6_scope_id; +#endif + return 0; +} + int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr) { struct rdma_id_private *id_priv; @@ -2096,6 +2111,10 @@ int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr) if (!cma_comp_exch(id_priv, CMA_IDLE, CMA_ADDR_BOUND)) return -EINVAL; + ret = cma_check_linklocal(&id->route.addr.dev_addr, addr); + if (ret) + goto err1; + if (cma_loopback_addr(addr)) { ret = cma_bind_loopback(id_priv); } else if (!cma_zero_addr(addr)) { -- cgit v1.1 From fd4582a3999e03fa9eae315bf14c88fd32d44035 Mon Sep 17 00:00:00 2001 From: "Robert P. J. Day" Date: Wed, 6 Jan 2010 13:16:30 -0800 Subject: IB/addr: Correct CONFIG_IPv6 to CONFIG_IPV6 Correct misspelled "CONFIG_IPv6" that was introduced in commit d14714df ("IB/addr: Fix IPv6 routing lookup"). The config variable should be all uppercase. Signed-off-by: Robert P. J. Day [ This was my fault when I munged the original patch. - Roland ] Signed-off-by: Roland Dreier --- drivers/infiniband/core/cma.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/infiniband/core/cma.c') diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index fbdd731..cc9b594 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -2083,7 +2083,7 @@ static int cma_get_port(struct rdma_id_private *id_priv) static int cma_check_linklocal(struct rdma_dev_addr *dev_addr, struct sockaddr *addr) { -#if defined(CONFIG_IPv6) || defined(CONFIG_IPV6_MODULE) +#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) struct sockaddr_in6 *sin6; if (addr->sa_family != AF_INET6) -- cgit v1.1 From 8523c0480979080e8088e40f25459e5b2d19f621 Mon Sep 17 00:00:00 2001 From: Sean Hefty Date: Mon, 8 Feb 2010 16:41:15 -0800 Subject: RDMA/cm: Revert association of an RDMA device when binding to loopback Revert the following change from commit 6f8372b6 ("RDMA/cm: fix loopback address support") The defined behavior of rdma_bind_addr is to associate an RDMA device with an rdma_cm_id, as long as the user specified a non- zero address. (ie they weren't just trying to reserve a port) Currently, if the loopback address is passed to rdma_bind_addr, no device is associated with the rdma_cm_id. Fix this. It turns out that important apps such as Open MPI depend on rdma_bind_addr() NOT associating any RDMA device when binding to a loopback address. Open MPI is being updated to deal with this, but at least until a new Open MPI release is available, maintain the previous behavior: allow rdma_bind_addr() to succeed, but do not bind to a device. Signed-off-by: Sean Hefty Acked-by: Steve Wise Signed-off-by: Roland Dreier --- drivers/infiniband/core/cma.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) (limited to 'drivers/infiniband/core/cma.c') diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index cc9b594..875e34e 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -2115,9 +2115,7 @@ int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr) if (ret) goto err1; - if (cma_loopback_addr(addr)) { - ret = cma_bind_loopback(id_priv); - } else if (!cma_zero_addr(addr)) { + if (!cma_any_addr(addr)) { ret = rdma_translate_ip(addr, &id->route.addr.dev_addr); if (ret) goto err1; -- cgit v1.1