Blob Blame History Raw
From: Parav Pandit <parav@mellanox.com>
Date: Wed, 5 Sep 2018 12:54:26 +0300
Subject: RDMA/core: Consider net ns of gid attribute for RoCE
Patch-mainline: v4.20-rc1
Git-commit: 0e9d2c19bff1d351005afb2f990a913e395ba6d4
References: bsc#1103992 FATE#326009

When resolving destination address or route, when net namespace is
unavailable, refer to the net namespace of the netdevice of the SGID
attribute. This is typically the case for requests arriving from the
network for RoCE ports.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Acked-by: Thomas Bogendoerfer <tbogendoerfer@suse.de>
---
 drivers/infiniband/core/addr.c      |   75 ++++++++++++++++++++++++++++++------
 drivers/infiniband/core/cma.c       |    7 +--
 drivers/infiniband/core/core_priv.h |    2 
 drivers/infiniband/core/verbs.c     |    2 
 include/rdma/ib_addr.h              |    3 +
 5 files changed, 73 insertions(+), 16 deletions(-)

--- a/drivers/infiniband/core/addr.c
+++ b/drivers/infiniband/core/addr.c
@@ -62,6 +62,7 @@ struct addr_req {
 			 struct rdma_dev_addr *addr, void *context);
 	unsigned long timeout;
 	struct delayed_work work;
+	bool resolve_by_gid_attr;	/* Consider gid attr in resolve phase */
 	int status;
 	u32 seq;
 };
@@ -518,10 +519,37 @@ static int rdma_set_src_addr_rcu(struct
 	return 0;
 }
 
+static int set_addr_netns_by_gid_rcu(struct rdma_dev_addr *addr)
+{
+	struct net_device *ndev;
+
+	ndev = rdma_read_gid_attr_ndev_rcu(addr->sgid_attr);
+	if (IS_ERR(ndev))
+		return PTR_ERR(ndev);
+
+	/*
+	 * Since we are holding the rcu, reading net and ifindex
+	 * are safe without any additional reference; because
+	 * change_net_namespace() in net/core/dev.c does rcu sync
+	 * after it changes the state to IFF_DOWN and before
+	 * updating netdev fields {net, ifindex}.
+	 */
+	addr->net = dev_net(ndev);
+	addr->bound_dev_if = ndev->ifindex;
+	return 0;
+}
+
+static void rdma_addr_set_net_defaults(struct rdma_dev_addr *addr)
+{
+	addr->net = &init_net;
+	addr->bound_dev_if = 0;
+}
+
 static int addr_resolve(struct sockaddr *src_in,
 			const struct sockaddr *dst_in,
 			struct rdma_dev_addr *addr,
 			bool resolve_neigh,
+			bool resolve_by_gid_attr,
 			u32 seq)
 {
 	struct dst_entry *dst = NULL;
@@ -535,6 +563,23 @@ static int addr_resolve(struct sockaddr
 	}
 
 	rcu_read_lock();
+	if (resolve_by_gid_attr) {
+		if (!addr->sgid_attr) {
+			rcu_read_unlock();
+			pr_warn_ratelimited("%s: missing gid_attr\n", __func__);
+			return -EINVAL;
+		}
+		/*
+		 * If the request is for a specific gid attribute of the
+		 * rdma_dev_addr, derive net from the netdevice of the
+		 * GID attribute.
+		 */
+		ret = set_addr_netns_by_gid_rcu(addr);
+		if (ret) {
+			rcu_read_unlock();
+			return ret;
+		}
+	}
 	if (src_in->sa_family == AF_INET) {
 		ret = addr4_resolve(src_in, dst_in, addr, &rt);
 		dst = &rt->dst;
@@ -543,7 +588,7 @@ static int addr_resolve(struct sockaddr
 	}
 	if (ret) {
 		rcu_read_unlock();
-		return ret;
+		goto done;
 	}
 	ret = rdma_set_src_addr_rcu(addr, &ndev_flags, dst_in, dst);
 	rcu_read_unlock();
@@ -559,6 +604,13 @@ static int addr_resolve(struct sockaddr
 		ip_rt_put(rt);
 	else
 		dst_release(dst);
+done:
+	/*
+	 * Clear the addr net to go back to its original state, only if it was
+	 * derived from GID attribute in this context.
+	 */
+	if (resolve_by_gid_attr)
+		rdma_addr_set_net_defaults(addr);
 	return ret;
 }
 
@@ -573,7 +625,8 @@ static void process_one_req(struct work_
 		src_in = (struct sockaddr *)&req->src_addr;
 		dst_in = (struct sockaddr *)&req->dst_addr;
 		req->status = addr_resolve(src_in, dst_in, req->addr,
-					   true, req->seq);
+					   true, req->resolve_by_gid_attr,
+					   req->seq);
 		if (req->status && time_after_eq(jiffies, req->timeout)) {
 			req->status = -ETIMEDOUT;
 		} else if (req->status == -ENODATA) {
@@ -608,6 +661,7 @@ int rdma_resolve_ip(struct sockaddr *src
 		    struct rdma_dev_addr *addr, int timeout_ms,
 		    void (*callback)(int status, struct sockaddr *src_addr,
 				     struct rdma_dev_addr *addr, void *context),
+		    bool resolve_by_gid_attr,
 		    void *context)
 {
 	struct sockaddr *src_in, *dst_in;
@@ -636,10 +690,12 @@ int rdma_resolve_ip(struct sockaddr *src
 	req->addr = addr;
 	req->callback = callback;
 	req->context = context;
+	req->resolve_by_gid_attr = resolve_by_gid_attr;
 	INIT_DELAYED_WORK(&req->work, process_one_req);
 	req->seq = (u32)atomic_inc_return(&ib_nl_addr_request_seq);
 
-	req->status = addr_resolve(src_in, dst_in, addr, true, req->seq);
+	req->status = addr_resolve(src_in, dst_in, addr, true,
+				   req->resolve_by_gid_attr, req->seq);
 	switch (req->status) {
 	case 0:
 		req->timeout = jiffies;
@@ -683,14 +739,11 @@ int roce_resolve_route_from_path(struct
 	if (!attr || !attr->ndev)
 		return -EINVAL;
 
-	dev_addr.bound_dev_if = attr->ndev->ifindex;
-	/* TODO: Use net from the ib_gid_attr once it is added to it,
-	 * until than, limit itself to init_net.
-	 */
 	dev_addr.net = &init_net;
+	dev_addr.sgid_attr = attr;
 
 	ret = addr_resolve(&sgid._sockaddr, &dgid._sockaddr,
-			   &dev_addr, false, 0);
+			   &dev_addr, false, true, 0);
 	if (ret)
 		return ret;
 
@@ -755,7 +808,7 @@ static void resolve_cb(int status, struc
 
 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
 				 const union ib_gid *dgid,
-				 u8 *dmac, const struct net_device *ndev,
+				 u8 *dmac, const struct ib_gid_attr *sgid_attr,
 				 int *hoplimit)
 {
 	struct rdma_dev_addr dev_addr;
@@ -771,12 +824,12 @@ int rdma_addr_find_l2_eth_by_grh(const u
 	rdma_gid2ip(&dgid_addr._sockaddr, dgid);
 
 	memset(&dev_addr, 0, sizeof(dev_addr));
-	dev_addr.bound_dev_if = ndev->ifindex;
 	dev_addr.net = &init_net;
+	dev_addr.sgid_attr = sgid_attr;
 
 	init_completion(&ctx.comp);
 	ret = rdma_resolve_ip(&sgid_addr._sockaddr, &dgid_addr._sockaddr,
-			      &dev_addr, 1000, resolve_cb, &ctx);
+			      &dev_addr, 1000, resolve_cb, true, &ctx);
 	if (ret)
 		return ret;
 
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -2989,9 +2989,10 @@ int rdma_resolve_addr(struct rdma_cm_id
 		if (dst_addr->sa_family == AF_IB) {
 			ret = cma_resolve_ib_addr(id_priv);
 		} else {
-			ret = rdma_resolve_ip(cma_src_addr(id_priv),
-					      dst_addr, &id->route.addr.dev_addr,
-					      timeout_ms, addr_handler, id_priv);
+			ret = rdma_resolve_ip(cma_src_addr(id_priv), dst_addr,
+					      &id->route.addr.dev_addr,
+					      timeout_ms, addr_handler,
+					      false, id_priv);
 		}
 	}
 	if (ret)
--- a/drivers/infiniband/core/core_priv.h
+++ b/drivers/infiniband/core/core_priv.h
@@ -338,7 +338,7 @@ int rdma_resolve_ip_route(struct sockadd
 
 int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
 				 const union ib_gid *dgid,
-				 u8 *dmac, const struct net_device *ndev,
+				 u8 *dmac, const struct ib_gid_attr *sgid_attr,
 				 int *hoplimit);
 void rdma_copy_src_l2_addr(struct rdma_dev_addr *dev_addr,
 			   const struct net_device *dev);
--- a/drivers/infiniband/core/verbs.c
+++ b/drivers/infiniband/core/verbs.c
@@ -710,7 +710,7 @@ static int ib_resolve_unicast_gid_dmac(s
 
 	ret = rdma_addr_find_l2_eth_by_grh(&sgid_attr->gid, &grh->dgid,
 					   ah_attr->roce.dmac,
-					   sgid_attr->ndev, &hop_limit);
+					   sgid_attr, &hop_limit);
 
 	grh->hop_limit = hop_limit;
 	return ret;
--- a/include/rdma/ib_addr.h
+++ b/include/rdma/ib_addr.h
@@ -95,12 +95,15 @@ int rdma_translate_ip(const struct socka
  * @timeout_ms: Amount of time to wait for the address resolution to complete.
  * @callback: Call invoked once address resolution has completed, timed out,
  *   or been canceled.  A status of 0 indicates success.
+ * @resolve_by_gid_attr:	Resolve the ip based on the GID attribute from
+ *				rdma_dev_addr.
  * @context: User-specified context associated with the call.
  */
 int rdma_resolve_ip(struct sockaddr *src_addr, const struct sockaddr *dst_addr,
 		    struct rdma_dev_addr *addr, int timeout_ms,
 		    void (*callback)(int status, struct sockaddr *src_addr,
 				     struct rdma_dev_addr *addr, void *context),
+		    bool resolve_by_gid_attr,
 		    void *context);
 
 void rdma_addr_cancel(struct rdma_dev_addr *addr);