Commit a9bb7912 authored by Hefty, Sean's avatar Hefty, Sean Committed by Roland Dreier

RDMA/cma: Add an ID_REUSEADDR option

Lustre requires that clients bind to a privileged port number before
connecting to a remote server.  On larger clusters (typically more
than about 1000 nodes), the number of privileged ports is exhausted,
resulting in lustre being unusable.

To handle this, we add support for reusable addresses to the rdma_cm.
This mimics the behavior of the socket option SO_REUSEADDR.  A user
may set an rdma_cm_id to reuse an address before calling
rdma_bind_addr() (explicitly or implicitly).  If set, other
rdma_cm_id's may be bound to the same address, provided that they all
have reuse enabled, and there are no active listens.

If rdma_listen() is called on an rdma_cm_id that has reuse enabled, it
will only succeed if there are no other id's bound to that same
address.  The reuse option is exported to user space.  The behavior of
the kernel reuse implementation was verified against that given by
sockets.

This patch is derived from a path by Ira Weiny <weiny2@llnl.gov>
Signed-off-by: default avatarSean Hefty <sean.hefty@intel.com>
Signed-off-by: default avatarRoland Dreier <roland@purestorage.com>
parent 43b752da
...@@ -148,6 +148,7 @@ struct rdma_id_private { ...@@ -148,6 +148,7 @@ struct rdma_id_private {
u32 qp_num; u32 qp_num;
u8 srq; u8 srq;
u8 tos; u8 tos;
u8 reuseaddr;
}; };
struct cma_multicast { struct cma_multicast {
...@@ -1579,50 +1580,6 @@ static void cma_listen_on_all(struct rdma_id_private *id_priv) ...@@ -1579,50 +1580,6 @@ static void cma_listen_on_all(struct rdma_id_private *id_priv)
mutex_unlock(&lock); mutex_unlock(&lock);
} }
int rdma_listen(struct rdma_cm_id *id, int backlog)
{
struct rdma_id_private *id_priv;
int ret;
id_priv = container_of(id, struct rdma_id_private, id);
if (id_priv->state == CMA_IDLE) {
((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;
}
if (!cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_LISTEN))
return -EINVAL;
id_priv->backlog = backlog;
if (id->device) {
switch (rdma_node_get_transport(id->device->node_type)) {
case RDMA_TRANSPORT_IB:
ret = cma_ib_listen(id_priv);
if (ret)
goto err;
break;
case RDMA_TRANSPORT_IWARP:
ret = cma_iw_listen(id_priv, backlog);
if (ret)
goto err;
break;
default:
ret = -ENOSYS;
goto err;
}
} else
cma_listen_on_all(id_priv);
return 0;
err:
id_priv->backlog = 0;
cma_comp_exch(id_priv, CMA_LISTEN, CMA_ADDR_BOUND);
return ret;
}
EXPORT_SYMBOL(rdma_listen);
void rdma_set_service_type(struct rdma_cm_id *id, int tos) void rdma_set_service_type(struct rdma_cm_id *id, int tos)
{ {
struct rdma_id_private *id_priv; struct rdma_id_private *id_priv;
...@@ -2105,6 +2062,25 @@ int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr, ...@@ -2105,6 +2062,25 @@ int rdma_resolve_addr(struct rdma_cm_id *id, struct sockaddr *src_addr,
} }
EXPORT_SYMBOL(rdma_resolve_addr); EXPORT_SYMBOL(rdma_resolve_addr);
int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse)
{
struct rdma_id_private *id_priv;
unsigned long flags;
int ret;
id_priv = container_of(id, struct rdma_id_private, id);
spin_lock_irqsave(&id_priv->lock, flags);
if (id_priv->state == CMA_IDLE) {
id_priv->reuseaddr = reuse;
ret = 0;
} else {
ret = -EINVAL;
}
spin_unlock_irqrestore(&id_priv->lock, flags);
return ret;
}
EXPORT_SYMBOL(rdma_set_reuseaddr);
static void cma_bind_port(struct rdma_bind_list *bind_list, static void cma_bind_port(struct rdma_bind_list *bind_list,
struct rdma_id_private *id_priv) struct rdma_id_private *id_priv)
{ {
...@@ -2180,31 +2156,29 @@ static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv) ...@@ -2180,31 +2156,29 @@ static int cma_alloc_any_port(struct idr *ps, struct rdma_id_private *id_priv)
return -EADDRNOTAVAIL; return -EADDRNOTAVAIL;
} }
static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv) /*
* Check that the requested port is available. This is called when trying to
* bind to a specific port, or when trying to listen on a bound port. In
* the latter case, the provided id_priv may already be on the bind_list, but
* we still need to check that it's okay to start listening.
*/
static int cma_check_port(struct rdma_bind_list *bind_list,
struct rdma_id_private *id_priv, uint8_t reuseaddr)
{ {
struct rdma_id_private *cur_id; struct rdma_id_private *cur_id;
struct sockaddr *addr, *cur_addr; struct sockaddr *addr, *cur_addr;
struct rdma_bind_list *bind_list;
struct hlist_node *node; struct hlist_node *node;
unsigned short snum;
addr = (struct sockaddr *) &id_priv->id.route.addr.src_addr; addr = (struct sockaddr *) &id_priv->id.route.addr.src_addr;
snum = ntohs(cma_port(addr)); if (cma_any_addr(addr) && !reuseaddr)
if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
return -EACCES;
bind_list = idr_find(ps, snum);
if (!bind_list)
return cma_alloc_port(ps, id_priv, snum);
/*
* We don't support binding to any address if anyone is bound to
* a specific address on the same port.
*/
if (cma_any_addr(addr))
return -EADDRNOTAVAIL; return -EADDRNOTAVAIL;
hlist_for_each_entry(cur_id, node, &bind_list->owners, node) { hlist_for_each_entry(cur_id, node, &bind_list->owners, node) {
if (id_priv == cur_id)
continue;
if ((cur_id->state == CMA_LISTEN) ||
!reuseaddr || !cur_id->reuseaddr) {
cur_addr = (struct sockaddr *) &cur_id->id.route.addr.src_addr; cur_addr = (struct sockaddr *) &cur_id->id.route.addr.src_addr;
if (cma_any_addr(cur_addr)) if (cma_any_addr(cur_addr))
return -EADDRNOTAVAIL; return -EADDRNOTAVAIL;
...@@ -2212,9 +2186,41 @@ static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv) ...@@ -2212,9 +2186,41 @@ static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
if (!cma_addr_cmp(addr, cur_addr)) if (!cma_addr_cmp(addr, cur_addr))
return -EADDRINUSE; return -EADDRINUSE;
} }
}
return 0;
}
static int cma_use_port(struct idr *ps, struct rdma_id_private *id_priv)
{
struct rdma_bind_list *bind_list;
unsigned short snum;
int ret;
snum = ntohs(cma_port((struct sockaddr *) &id_priv->id.route.addr.src_addr));
if (snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
return -EACCES;
bind_list = idr_find(ps, snum);
if (!bind_list) {
ret = cma_alloc_port(ps, id_priv, snum);
} else {
ret = cma_check_port(bind_list, id_priv, id_priv->reuseaddr);
if (!ret)
cma_bind_port(bind_list, id_priv); cma_bind_port(bind_list, id_priv);
return 0; }
return ret;
}
static int cma_bind_listen(struct rdma_id_private *id_priv)
{
struct rdma_bind_list *bind_list = id_priv->bind_list;
int ret = 0;
mutex_lock(&lock);
if (bind_list->owners.first->next)
ret = cma_check_port(bind_list, id_priv, 0);
mutex_unlock(&lock);
return ret;
} }
static int cma_get_port(struct rdma_id_private *id_priv) static int cma_get_port(struct rdma_id_private *id_priv)
...@@ -2268,6 +2274,56 @@ static int cma_check_linklocal(struct rdma_dev_addr *dev_addr, ...@@ -2268,6 +2274,56 @@ static int cma_check_linklocal(struct rdma_dev_addr *dev_addr,
return 0; return 0;
} }
int rdma_listen(struct rdma_cm_id *id, int backlog)
{
struct rdma_id_private *id_priv;
int ret;
id_priv = container_of(id, struct rdma_id_private, id);
if (id_priv->state == CMA_IDLE) {
((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;
}
if (!cma_comp_exch(id_priv, CMA_ADDR_BOUND, CMA_LISTEN))
return -EINVAL;
if (id_priv->reuseaddr) {
ret = cma_bind_listen(id_priv);
if (ret)
goto err;
}
id_priv->backlog = backlog;
if (id->device) {
switch (rdma_node_get_transport(id->device->node_type)) {
case RDMA_TRANSPORT_IB:
ret = cma_ib_listen(id_priv);
if (ret)
goto err;
break;
case RDMA_TRANSPORT_IWARP:
ret = cma_iw_listen(id_priv, backlog);
if (ret)
goto err;
break;
default:
ret = -ENOSYS;
goto err;
}
} else
cma_listen_on_all(id_priv);
return 0;
err:
id_priv->backlog = 0;
cma_comp_exch(id_priv, CMA_LISTEN, CMA_ADDR_BOUND);
return ret;
}
EXPORT_SYMBOL(rdma_listen);
int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr) int rdma_bind_addr(struct rdma_cm_id *id, struct sockaddr *addr)
{ {
struct rdma_id_private *id_priv; struct rdma_id_private *id_priv;
......
...@@ -883,6 +883,13 @@ static int ucma_set_option_id(struct ucma_context *ctx, int optname, ...@@ -883,6 +883,13 @@ static int ucma_set_option_id(struct ucma_context *ctx, int optname,
} }
rdma_set_service_type(ctx->cm_id, *((u8 *) optval)); rdma_set_service_type(ctx->cm_id, *((u8 *) optval));
break; break;
case RDMA_OPTION_ID_REUSEADDR:
if (optlen != sizeof(int)) {
ret = -EINVAL;
break;
}
ret = rdma_set_reuseaddr(ctx->cm_id, *((int *) optval) ? 1 : 0);
break;
default: default:
ret = -ENOSYS; ret = -ENOSYS;
} }
......
...@@ -329,4 +329,14 @@ void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr); ...@@ -329,4 +329,14 @@ void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr);
*/ */
void rdma_set_service_type(struct rdma_cm_id *id, int tos); void rdma_set_service_type(struct rdma_cm_id *id, int tos);
/**
* rdma_set_reuseaddr - Allow the reuse of local addresses when binding
* the rdma_cm_id.
* @id: Communication identifier to configure.
* @reuse: Value indicating if the bound address is reusable.
*
* Reuse must be set before an address is bound to the id.
*/
int rdma_set_reuseaddr(struct rdma_cm_id *id, int reuse);
#endif /* RDMA_CM_H */ #endif /* RDMA_CM_H */
...@@ -222,6 +222,7 @@ enum { ...@@ -222,6 +222,7 @@ enum {
/* Option details */ /* Option details */
enum { enum {
RDMA_OPTION_ID_TOS = 0, RDMA_OPTION_ID_TOS = 0,
RDMA_OPTION_ID_REUSEADDR = 1,
RDMA_OPTION_IB_PATH = 1 RDMA_OPTION_IB_PATH = 1
}; };
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment