From: Parav Pandit <parav@xxxxxxxxxxxx>
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@xxxxxxxxxxxx>
Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxxxx>
---
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(-)
diff --git a/drivers/infiniband/core/addr.c b/drivers/infiniband/core/addr.c
index 846d1bb2c9c9..5675152eba04 100644
--- 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;
};
@@ -520,10 +521,37 @@ rdma_set_src_addr_rcu(struct rdma_dev_addr *dev_addr,
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;
@@ -537,6 +565,23 @@ static int addr_resolve(struct sockaddr *src_in,
}
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;
@@ -545,7 +590,7 @@ static int addr_resolve(struct sockaddr *src_in,
}
if (ret) {
rcu_read_unlock();
- return ret;
+ goto done;
}
ret = rdma_set_src_addr_rcu(addr, &ndev_flags, dst_in, dst);
rcu_read_unlock();
@@ -561,6 +606,13 @@ static int addr_resolve(struct sockaddr *src_in,
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;
}
@@ -575,7 +627,8 @@ static void process_one_req(struct work_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) {
@@ -610,6 +663,7 @@ 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)
{
struct sockaddr *src_in, *dst_in;
@@ -638,10 +692,12 @@ int rdma_resolve_ip(struct sockaddr *src_addr, const struct sockaddr *dst_addr,
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;
@@ -685,14 +741,11 @@ int roce_resolve_route_from_path(struct sa_path_rec *rec,
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;
@@ -757,7 +810,7 @@ static void resolve_cb(int status, struct sockaddr *src_addr,
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;
@@ -773,12 +826,12 @@ int rdma_addr_find_l2_eth_by_grh(const union ib_gid *sgid,
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;
diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
index e9ee8ff642b6..e941deb1770d 100644
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -2989,9 +2989,10 @@ int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
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)
diff --git a/drivers/infiniband/core/core_priv.h b/drivers/infiniband/core/core_priv.h
index 33f50e1929e7..d7399d5b1cb6 100644
--- a/drivers/infiniband/core/core_priv.h
+++ b/drivers/infiniband/core/core_priv.h
@@ -338,7 +338,7 @@ int rdma_resolve_ip_route(struct sockaddr *src_addr,
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);
diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
index 6ee03d6089eb..c36be384fe34 100644
--- a/drivers/infiniband/core/verbs.c
+++ b/drivers/infiniband/core/verbs.c
@@ -710,7 +710,7 @@ static int ib_resolve_unicast_gid_dmac(struct ib_device *device,
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;
diff --git a/include/rdma/ib_addr.h b/include/rdma/ib_addr.h
index 676514a930ab..2e33b1529015 100644
--- a/include/rdma/ib_addr.h
+++ b/include/rdma/ib_addr.h
@@ -95,12 +95,15 @@ int rdma_translate_ip(const struct sockaddr *addr,
* @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);
--
2.14.4
