From faec2f7b96b555055d0aa6cc6b83a537270bed52 Mon Sep 17 00:00:00 2001 From: Sean Hefty Date: Thu, 15 Feb 2007 17:00:17 -0800 Subject: IB/sa: Track multicast join/leave requests The IB SA tracks multicast join/leave requests on a per port basis and does not do any reference counting: if two users of the same port join the same group, and one leaves that group, then the SA will remove the port from the group even though there is one user who wants to stay a member left. Therefore, in order to support multiple users of the same multicast group from the same port, we need to perform reference counting locally. To do this, add an multicast submodule to ib_sa to perform reference counting of multicast join/leave operations. Modify ib_ipoib (the only in-kernel user of multicast) to use the new interface. Signed-off-by: Roland Dreier --- include/rdma/ib_addr.h | 6 ++ include/rdma/ib_sa.h | 159 +++++++++++++++++++++---------------------------- 2 files changed, 74 insertions(+), 91 deletions(-) (limited to 'include') diff --git a/include/rdma/ib_addr.h b/include/rdma/ib_addr.h index c094e5012862..c36750ff6ae8 100644 --- a/include/rdma/ib_addr.h +++ b/include/rdma/ib_addr.h @@ -110,6 +110,12 @@ static inline void ib_addr_set_pkey(struct rdma_dev_addr *dev_addr, u16 pkey) dev_addr->broadcast[9] = (unsigned char) pkey; } +static inline void ib_addr_get_mgid(struct rdma_dev_addr *dev_addr, + union ib_gid *gid) +{ + memcpy(gid, dev_addr->broadcast + 4, sizeof *gid); +} + static inline void ib_addr_get_sgid(struct rdma_dev_addr *dev_addr, union ib_gid *gid) { diff --git a/include/rdma/ib_sa.h b/include/rdma/ib_sa.h index 97715b0c20b6..5e26b2f53f86 100644 --- a/include/rdma/ib_sa.h +++ b/include/rdma/ib_sa.h @@ -285,18 +285,6 @@ int ib_sa_path_rec_get(struct ib_sa_client *client, void *context, struct ib_sa_query **query); -int ib_sa_mcmember_rec_query(struct ib_sa_client *client, - struct ib_device *device, u8 port_num, - u8 method, - struct ib_sa_mcmember_rec *rec, - ib_sa_comp_mask comp_mask, - int timeout_ms, gfp_t gfp_mask, - void (*callback)(int status, - struct ib_sa_mcmember_rec *resp, - void *context), - void *context, - struct ib_sa_query **query); - int ib_sa_service_rec_query(struct ib_sa_client *client, struct ib_device *device, u8 port_num, u8 method, @@ -309,93 +297,82 @@ int ib_sa_service_rec_query(struct ib_sa_client *client, void *context, struct ib_sa_query **sa_query); +struct ib_sa_multicast { + struct ib_sa_mcmember_rec rec; + ib_sa_comp_mask comp_mask; + int (*callback)(int status, + struct ib_sa_multicast *multicast); + void *context; +}; + /** - * ib_sa_mcmember_rec_set - Start an MCMember set query - * @client:SA client - * @device:device to send query on - * @port_num: port number to send query on - * @rec:MCMember Record to send in query - * @comp_mask:component mask to send in query - * @timeout_ms:time to wait for response - * @gfp_mask:GFP mask to use for internal allocations - * @callback:function called when query completes, times out or is - * canceled - * @context:opaque user context passed to callback - * @sa_query:query context, used to cancel query + * ib_sa_join_multicast - Initiates a join request to the specified multicast + * group. + * @client: SA client + * @device: Device associated with the multicast group. + * @port_num: Port on the specified device to associate with the multicast + * group. + * @rec: SA multicast member record specifying group attributes. + * @comp_mask: Component mask indicating which group attributes of %rec are + * valid. + * @gfp_mask: GFP mask for memory allocations. + * @callback: User callback invoked once the join operation completes. + * @context: User specified context stored with the ib_sa_multicast structure. * - * Send an MCMember Set query to the SA (eg to join a multicast - * group). The callback function will be called when the query - * completes (or fails); status is 0 for a successful response, -EINTR - * if the query is canceled, -ETIMEDOUT is the query timed out, or - * -EIO if an error occurred sending the query. The resp parameter of - * the callback is only valid if status is 0. + * This call initiates a multicast join request with the SA for the specified + * multicast group. If the join operation is started successfully, it returns + * an ib_sa_multicast structure that is used to track the multicast operation. + * Users must free this structure by calling ib_free_multicast, even if the + * join operation later fails. (The callback status is non-zero.) * - * If the return value of ib_sa_mcmember_rec_set() is negative, it is - * an error code. Otherwise it is a query ID that can be used to - * cancel the query. + * If the join operation fails; status will be non-zero, with the following + * failures possible: + * -ETIMEDOUT: The request timed out. + * -EIO: An error occurred sending the query. + * -EINVAL: The MCMemberRecord values differed from the existing group's. + * -ENETRESET: Indicates that an fatal error has occurred on the multicast + * group, and the user must rejoin the group to continue using it. */ -static inline int -ib_sa_mcmember_rec_set(struct ib_sa_client *client, - struct ib_device *device, u8 port_num, - struct ib_sa_mcmember_rec *rec, - ib_sa_comp_mask comp_mask, - int timeout_ms, gfp_t gfp_mask, - void (*callback)(int status, - struct ib_sa_mcmember_rec *resp, - void *context), - void *context, - struct ib_sa_query **query) -{ - return ib_sa_mcmember_rec_query(client, device, port_num, - IB_MGMT_METHOD_SET, - rec, comp_mask, - timeout_ms, gfp_mask, callback, - context, query); -} +struct ib_sa_multicast *ib_sa_join_multicast(struct ib_sa_client *client, + struct ib_device *device, u8 port_num, + struct ib_sa_mcmember_rec *rec, + ib_sa_comp_mask comp_mask, gfp_t gfp_mask, + int (*callback)(int status, + struct ib_sa_multicast + *multicast), + void *context); /** - * ib_sa_mcmember_rec_delete - Start an MCMember delete query - * @client:SA client - * @device:device to send query on - * @port_num: port number to send query on - * @rec:MCMember Record to send in query - * @comp_mask:component mask to send in query - * @timeout_ms:time to wait for response - * @gfp_mask:GFP mask to use for internal allocations - * @callback:function called when query completes, times out or is - * canceled - * @context:opaque user context passed to callback - * @sa_query:query context, used to cancel query - * - * Send an MCMember Delete query to the SA (eg to leave a multicast - * group). The callback function will be called when the query - * completes (or fails); status is 0 for a successful response, -EINTR - * if the query is canceled, -ETIMEDOUT is the query timed out, or - * -EIO if an error occurred sending the query. The resp parameter of - * the callback is only valid if status is 0. + * ib_free_multicast - Frees the multicast tracking structure, and releases + * any reference on the multicast group. + * @multicast: Multicast tracking structure allocated by ib_join_multicast. * - * If the return value of ib_sa_mcmember_rec_delete() is negative, it - * is an error code. Otherwise it is a query ID that can be used to - * cancel the query. + * This call blocks until the multicast identifier is destroyed. It may + * not be called from within the multicast callback; however, returning a non- + * zero value from the callback will result in destroying the multicast + * tracking structure. + */ +void ib_sa_free_multicast(struct ib_sa_multicast *multicast); + +/** + * ib_get_mcmember_rec - Looks up a multicast member record by its MGID and + * returns it if found. + * @device: Device associated with the multicast group. + * @port_num: Port on the specified device to associate with the multicast + * group. + * @mgid: MGID of multicast group. + * @rec: Location to copy SA multicast member record. */ -static inline int -ib_sa_mcmember_rec_delete(struct ib_sa_client *client, - struct ib_device *device, u8 port_num, - struct ib_sa_mcmember_rec *rec, - ib_sa_comp_mask comp_mask, - int timeout_ms, gfp_t gfp_mask, - void (*callback)(int status, - struct ib_sa_mcmember_rec *resp, - void *context), - void *context, - struct ib_sa_query **query) -{ - return ib_sa_mcmember_rec_query(client, device, port_num, - IB_SA_METHOD_DELETE, - rec, comp_mask, - timeout_ms, gfp_mask, callback, - context, query); -} +int ib_sa_get_mcmember_rec(struct ib_device *device, u8 port_num, + union ib_gid *mgid, struct ib_sa_mcmember_rec *rec); + +/** + * ib_init_ah_from_mcmember - Initialize address handle attributes based on + * an SA multicast member record. + */ +int ib_init_ah_from_mcmember(struct ib_device *device, u8 port_num, + struct ib_sa_mcmember_rec *rec, + struct ib_ah_attr *ah_attr); /** * ib_init_ah_from_path - Initialize address handle attributes based on an SA -- cgit v1.2.3 From c8f6a362bf3eb28ade6027b49bb160a336dd51c0 Mon Sep 17 00:00:00 2001 From: Sean Hefty Date: Thu, 15 Feb 2007 17:00:18 -0800 Subject: RDMA/cma: Add multicast communication support Extend rdma_cm to support multicast communication. Multicast support is added to the existing RDMA_PS_UDP port space, as well as a new RDMA_PS_IPOIB port space. The latter port space allows joining the multicast groups used by IPoIB, which enables offloading IPoIB traffic to a separate QP. The port space determines the signature used in the MGID when joining the group. The newly added RDMA_PS_IPOIB also allows for unicast operations, similar to RDMA_PS_UDP. Supporting the RDMA_PS_IPOIB requires changing how UD QPs are initialized, since we can no longer assume that the qkey is constant. This requires saving the Q_Key to use when attaching to a device, so that it is available when creating the QP. The Q_Key information is exported to the user through the existing rdma_init_qp_attr() interface. Multicast support is also exported to userspace through the rdma_ucm. Signed-off-by: Roland Dreier --- drivers/infiniband/core/cma.c | 359 ++++++++++++++++++++++++++++++++++++----- drivers/infiniband/core/ucma.c | 204 ++++++++++++++++++++++- include/rdma/rdma_cm.h | 21 ++- include/rdma/rdma_cm_ib.h | 4 +- include/rdma/rdma_user_cm.h | 13 +- 5 files changed, 549 insertions(+), 52 deletions(-) (limited to 'include') diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index db88e609bf42..f8d69b3fa307 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -71,6 +71,7 @@ static struct workqueue_struct *cma_wq; static DEFINE_IDR(sdp_ps); static DEFINE_IDR(tcp_ps); static DEFINE_IDR(udp_ps); +static DEFINE_IDR(ipoib_ps); static int next_port; struct cma_device { @@ -116,6 +117,7 @@ struct rdma_id_private { struct list_head list; struct list_head listen_list; struct cma_device *cma_dev; + struct list_head mc_list; enum cma_state state; spinlock_t lock; @@ -134,10 +136,23 @@ struct rdma_id_private { } cm_id; u32 seq_num; + u32 qkey; u32 qp_num; u8 srq; }; +struct cma_multicast { + struct rdma_id_private *id_priv; + union { + struct ib_sa_multicast *ib; + } multicast; + struct list_head list; + void *context; + struct sockaddr addr; + u8 pad[sizeof(struct sockaddr_in6) - + sizeof(struct sockaddr)]; +}; + struct cma_work { struct work_struct work; struct rdma_id_private *id; @@ -243,6 +258,11 @@ static inline void sdp_set_ip_ver(struct sdp_hh *hh, u8 ip_ver) hh->ip_version = (ip_ver << 4) | (hh->ip_version & 0xF); } +static inline int cma_is_ud_ps(enum rdma_port_space ps) +{ + return (ps == RDMA_PS_UDP || ps == RDMA_PS_IPOIB); +} + static void cma_attach_to_dev(struct rdma_id_private *id_priv, struct cma_device *cma_dev) { @@ -265,19 +285,41 @@ static void cma_detach_from_dev(struct rdma_id_private *id_priv) id_priv->cma_dev = NULL; } +static int cma_set_qkey(struct ib_device *device, u8 port_num, + enum rdma_port_space ps, + struct rdma_dev_addr *dev_addr, u32 *qkey) +{ + struct ib_sa_mcmember_rec rec; + int ret = 0; + + switch (ps) { + case RDMA_PS_UDP: + *qkey = RDMA_UDP_QKEY; + break; + case RDMA_PS_IPOIB: + ib_addr_get_mgid(dev_addr, &rec.mgid); + ret = ib_sa_get_mcmember_rec(device, port_num, &rec.mgid, &rec); + *qkey = be32_to_cpu(rec.qkey); + break; + default: + break; + } + return ret; +} + static int cma_acquire_dev(struct rdma_id_private *id_priv) { - enum rdma_node_type dev_type = id_priv->id.route.addr.dev_addr.dev_type; + struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr; struct cma_device *cma_dev; union ib_gid gid; int ret = -ENODEV; - switch (rdma_node_get_transport(dev_type)) { + switch (rdma_node_get_transport(dev_addr->dev_type)) { case RDMA_TRANSPORT_IB: - ib_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid); + ib_addr_get_sgid(dev_addr, &gid); break; case RDMA_TRANSPORT_IWARP: - iw_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid); + iw_addr_get_sgid(dev_addr, &gid); break; default: return -ENODEV; @@ -287,7 +329,12 @@ static int cma_acquire_dev(struct rdma_id_private *id_priv) ret = ib_find_cached_gid(cma_dev->device, &gid, &id_priv->id.port_num, NULL); if (!ret) { - cma_attach_to_dev(id_priv, cma_dev); + ret = cma_set_qkey(cma_dev->device, + id_priv->id.port_num, + id_priv->id.ps, dev_addr, + &id_priv->qkey); + if (!ret) + cma_attach_to_dev(id_priv, cma_dev); break; } } @@ -325,40 +372,50 @@ struct rdma_cm_id *rdma_create_id(rdma_cm_event_handler event_handler, init_waitqueue_head(&id_priv->wait_remove); atomic_set(&id_priv->dev_remove, 0); INIT_LIST_HEAD(&id_priv->listen_list); + INIT_LIST_HEAD(&id_priv->mc_list); get_random_bytes(&id_priv->seq_num, sizeof id_priv->seq_num); return &id_priv->id; } EXPORT_SYMBOL(rdma_create_id); -static int cma_init_ib_qp(struct rdma_id_private *id_priv, struct ib_qp *qp) +static int cma_init_ud_qp(struct rdma_id_private *id_priv, struct ib_qp *qp) { struct ib_qp_attr qp_attr; - struct rdma_dev_addr *dev_addr; - int ret; + int qp_attr_mask, ret; - dev_addr = &id_priv->id.route.addr.dev_addr; - ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num, - ib_addr_get_pkey(dev_addr), - &qp_attr.pkey_index); + qp_attr.qp_state = IB_QPS_INIT; + ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask); if (ret) return ret; - qp_attr.qp_state = IB_QPS_INIT; - qp_attr.qp_access_flags = 0; - qp_attr.port_num = id_priv->id.port_num; - return ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_ACCESS_FLAGS | - IB_QP_PKEY_INDEX | IB_QP_PORT); + ret = ib_modify_qp(qp, &qp_attr, qp_attr_mask); + if (ret) + return ret; + + qp_attr.qp_state = IB_QPS_RTR; + ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE); + if (ret) + return ret; + + qp_attr.qp_state = IB_QPS_RTS; + qp_attr.sq_psn = 0; + ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN); + + return ret; } -static int cma_init_iw_qp(struct rdma_id_private *id_priv, struct ib_qp *qp) +static int cma_init_conn_qp(struct rdma_id_private *id_priv, struct ib_qp *qp) { struct ib_qp_attr qp_attr; + int qp_attr_mask, ret; qp_attr.qp_state = IB_QPS_INIT; - qp_attr.qp_access_flags = IB_ACCESS_LOCAL_WRITE; + ret = rdma_init_qp_attr(&id_priv->id, &qp_attr, &qp_attr_mask); + if (ret) + return ret; - return ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_ACCESS_FLAGS); + return ib_modify_qp(qp, &qp_attr, qp_attr_mask); } int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd, @@ -376,18 +433,10 @@ int rdma_create_qp(struct rdma_cm_id *id, struct ib_pd *pd, if (IS_ERR(qp)) return PTR_ERR(qp); - switch (rdma_node_get_transport(id->device->node_type)) { - case RDMA_TRANSPORT_IB: - ret = cma_init_ib_qp(id_priv, qp); - break; - case RDMA_TRANSPORT_IWARP: - ret = cma_init_iw_qp(id_priv, qp); - break; - default: - ret = -ENOSYS; - break; - } - + if (cma_is_ud_ps(id_priv->id.ps)) + ret = cma_init_ud_qp(id_priv, qp); + else + ret = cma_init_conn_qp(id_priv, qp); if (ret) goto err; @@ -460,23 +509,55 @@ static int cma_modify_qp_err(struct rdma_cm_id *id) return ib_modify_qp(id->qp, &qp_attr, IB_QP_STATE); } +static int cma_ib_init_qp_attr(struct rdma_id_private *id_priv, + struct ib_qp_attr *qp_attr, int *qp_attr_mask) +{ + struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr; + int ret; + + ret = ib_find_cached_pkey(id_priv->id.device, id_priv->id.port_num, + ib_addr_get_pkey(dev_addr), + &qp_attr->pkey_index); + if (ret) + return ret; + + qp_attr->port_num = id_priv->id.port_num; + *qp_attr_mask = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT; + + if (cma_is_ud_ps(id_priv->id.ps)) { + qp_attr->qkey = id_priv->qkey; + *qp_attr_mask |= IB_QP_QKEY; + } else { + qp_attr->qp_access_flags = 0; + *qp_attr_mask |= IB_QP_ACCESS_FLAGS; + } + return 0; +} + int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr, int *qp_attr_mask) { struct rdma_id_private *id_priv; - int ret; + int ret = 0; id_priv = container_of(id, struct rdma_id_private, id); switch (rdma_node_get_transport(id_priv->id.device->node_type)) { case RDMA_TRANSPORT_IB: - ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr, - qp_attr_mask); + if (!id_priv->cm_id.ib || cma_is_ud_ps(id_priv->id.ps)) + ret = cma_ib_init_qp_attr(id_priv, qp_attr, qp_attr_mask); + else + ret = ib_cm_init_qp_attr(id_priv->cm_id.ib, qp_attr, + qp_attr_mask); if (qp_attr->qp_state == IB_QPS_RTR) qp_attr->rq_psn = id_priv->seq_num; break; case RDMA_TRANSPORT_IWARP: - ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr, - qp_attr_mask); + if (!id_priv->cm_id.iw) { + qp_attr->qp_access_flags = IB_ACCESS_LOCAL_WRITE; + *qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS; + } else + ret = iw_cm_init_qp_attr(id_priv->cm_id.iw, qp_attr, + qp_attr_mask); break; default: ret = -ENOSYS; @@ -698,6 +779,19 @@ static void cma_release_port(struct rdma_id_private *id_priv) mutex_unlock(&lock); } +static void cma_leave_mc_groups(struct rdma_id_private *id_priv) +{ + struct cma_multicast *mc; + + while (!list_empty(&id_priv->mc_list)) { + mc = container_of(id_priv->mc_list.next, + struct cma_multicast, list); + list_del(&mc->list); + ib_sa_free_multicast(mc->multicast.ib); + kfree(mc); + } +} + void rdma_destroy_id(struct rdma_cm_id *id) { struct rdma_id_private *id_priv; @@ -722,6 +816,7 @@ void rdma_destroy_id(struct rdma_cm_id *id) default: break; } + cma_leave_mc_groups(id_priv); mutex_lock(&lock); cma_detach_from_dev(id_priv); } @@ -972,7 +1067,7 @@ static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event) memset(&event, 0, sizeof event); offset = cma_user_data_offset(listen_id->id.ps); event.event = RDMA_CM_EVENT_CONNECT_REQUEST; - if (listen_id->id.ps == RDMA_PS_UDP) { + if (cma_is_ud_ps(listen_id->id.ps)) { conn_id = cma_new_udp_id(&listen_id->id, ib_event); event.param.ud.private_data = ib_event->private_data + offset; event.param.ud.private_data_len = @@ -1725,7 +1820,7 @@ static int cma_alloc_port(struct idr *ps, struct rdma_id_private *id_priv, struct rdma_bind_list *bind_list; int port, ret; - bind_list = kzalloc(sizeof *bind_list, GFP_KERNEL); + bind_list = kmalloc(sizeof *bind_list, GFP_KERNEL); if (!bind_list) return -ENOMEM; @@ -1847,6 +1942,9 @@ static int cma_get_port(struct rdma_id_private *id_priv) case RDMA_PS_UDP: ps = &udp_ps; break; + case RDMA_PS_IPOIB: + ps = &ipoib_ps; + break; default: return -EPROTONOSUPPORT; } @@ -1961,7 +2059,7 @@ static int cma_sidr_rep_handler(struct ib_cm_id *cm_id, event.status = ib_event->param.sidr_rep_rcvd.status; break; } - if (rep->qkey != RDMA_UD_QKEY) { + if (id_priv->qkey != rep->qkey) { event.event = RDMA_CM_EVENT_UNREACHABLE; event.status = -EINVAL; break; @@ -2160,7 +2258,7 @@ int rdma_connect(struct rdma_cm_id *id, struct rdma_conn_param *conn_param) switch (rdma_node_get_transport(id->device->node_type)) { case RDMA_TRANSPORT_IB: - if (id->ps == RDMA_PS_UDP) + if (cma_is_ud_ps(id->ps)) ret = cma_resolve_ib_udp(id_priv, conn_param); else ret = cma_connect_ib(id_priv, conn_param); @@ -2256,7 +2354,7 @@ static int cma_send_sidr_rep(struct rdma_id_private *id_priv, rep.status = status; if (status == IB_SIDR_SUCCESS) { rep.qp_num = id_priv->qp_num; - rep.qkey = RDMA_UD_QKEY; + rep.qkey = id_priv->qkey; } rep.private_data = private_data; rep.private_data_len = private_data_len; @@ -2280,7 +2378,7 @@ int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param) switch (rdma_node_get_transport(id->device->node_type)) { case RDMA_TRANSPORT_IB: - if (id->ps == RDMA_PS_UDP) + if (cma_is_ud_ps(id->ps)) ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS, conn_param->private_data, conn_param->private_data_len); @@ -2341,7 +2439,7 @@ int rdma_reject(struct rdma_cm_id *id, const void *private_data, switch (rdma_node_get_transport(id->device->node_type)) { case RDMA_TRANSPORT_IB: - if (id->ps == RDMA_PS_UDP) + if (cma_is_ud_ps(id->ps)) ret = cma_send_sidr_rep(id_priv, IB_SIDR_REJECT, private_data, private_data_len); else @@ -2392,6 +2490,178 @@ out: } EXPORT_SYMBOL(rdma_disconnect); +static int cma_ib_mc_handler(int status, struct ib_sa_multicast *multicast) +{ + struct rdma_id_private *id_priv; + struct cma_multicast *mc = multicast->context; + struct rdma_cm_event event; + int ret; + + id_priv = mc->id_priv; + atomic_inc(&id_priv->dev_remove); + if (!cma_comp(id_priv, CMA_ADDR_BOUND) && + !cma_comp(id_priv, CMA_ADDR_RESOLVED)) + goto out; + + if (!status && id_priv->id.qp) + status = ib_attach_mcast(id_priv->id.qp, &multicast->rec.mgid, + multicast->rec.mlid); + + memset(&event, 0, sizeof event); + event.status = status; + event.param.ud.private_data = mc->context; + if (!status) { + event.event = RDMA_CM_EVENT_MULTICAST_JOIN; + ib_init_ah_from_mcmember(id_priv->id.device, + id_priv->id.port_num, &multicast->rec, + &event.param.ud.ah_attr); + event.param.ud.qp_num = 0xFFFFFF; + event.param.ud.qkey = be32_to_cpu(multicast->rec.qkey); + } else + event.event = RDMA_CM_EVENT_MULTICAST_ERROR; + + ret = id_priv->id.event_handler(&id_priv->id, &event); + if (ret) { + cma_exch(id_priv, CMA_DESTROYING); + cma_release_remove(id_priv); + rdma_destroy_id(&id_priv->id); + return 0; + } +out: + cma_release_remove(id_priv); + return 0; +} + +static void cma_set_mgid(struct rdma_id_private *id_priv, + struct sockaddr *addr, union ib_gid *mgid) +{ + unsigned char mc_map[MAX_ADDR_LEN]; + struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr; + struct sockaddr_in *sin = (struct sockaddr_in *) addr; + struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) addr; + + 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) == + 0xFF10A01B)) { + /* IPv6 address is an SA assigned MGID. */ + memcpy(mgid, &sin6->sin6_addr, sizeof *mgid); + } else { + ip_ib_mc_map(sin->sin_addr.s_addr, mc_map); + if (id_priv->id.ps == RDMA_PS_UDP) + mc_map[7] = 0x01; /* Use RDMA CM signature */ + mc_map[8] = ib_addr_get_pkey(dev_addr) >> 8; + mc_map[9] = (unsigned char) ib_addr_get_pkey(dev_addr); + *mgid = *(union ib_gid *) (mc_map + 4); + } +} + +static int cma_join_ib_multicast(struct rdma_id_private *id_priv, + struct cma_multicast *mc) +{ + struct ib_sa_mcmember_rec rec; + struct rdma_dev_addr *dev_addr = &id_priv->id.route.addr.dev_addr; + ib_sa_comp_mask comp_mask; + int ret; + + ib_addr_get_mgid(dev_addr, &rec.mgid); + ret = ib_sa_get_mcmember_rec(id_priv->id.device, id_priv->id.port_num, + &rec.mgid, &rec); + if (ret) + return ret; + + cma_set_mgid(id_priv, &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); + rec.pkey = cpu_to_be16(ib_addr_get_pkey(dev_addr)); + rec.join_state = 1; + + comp_mask = IB_SA_MCMEMBER_REC_MGID | IB_SA_MCMEMBER_REC_PORT_GID | + IB_SA_MCMEMBER_REC_PKEY | IB_SA_MCMEMBER_REC_JOIN_STATE | + IB_SA_MCMEMBER_REC_QKEY | IB_SA_MCMEMBER_REC_SL | + IB_SA_MCMEMBER_REC_FLOW_LABEL | + IB_SA_MCMEMBER_REC_TRAFFIC_CLASS; + + mc->multicast.ib = ib_sa_join_multicast(&sa_client, id_priv->id.device, + id_priv->id.port_num, &rec, + comp_mask, GFP_KERNEL, + cma_ib_mc_handler, mc); + if (IS_ERR(mc->multicast.ib)) + return PTR_ERR(mc->multicast.ib); + + return 0; +} + +int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr, + void *context) +{ + struct rdma_id_private *id_priv; + struct cma_multicast *mc; + int ret; + + id_priv = container_of(id, struct rdma_id_private, id); + if (!cma_comp(id_priv, CMA_ADDR_BOUND) && + !cma_comp(id_priv, CMA_ADDR_RESOLVED)) + return -EINVAL; + + mc = kmalloc(sizeof *mc, GFP_KERNEL); + if (!mc) + return -ENOMEM; + + memcpy(&mc->addr, addr, ip_addr_size(addr)); + mc->context = context; + mc->id_priv = id_priv; + + spin_lock(&id_priv->lock); + list_add(&mc->list, &id_priv->mc_list); + spin_unlock(&id_priv->lock); + + switch (rdma_node_get_transport(id->device->node_type)) { + case RDMA_TRANSPORT_IB: + ret = cma_join_ib_multicast(id_priv, mc); + break; + default: + ret = -ENOSYS; + break; + } + + if (ret) { + spin_lock_irq(&id_priv->lock); + list_del(&mc->list); + spin_unlock_irq(&id_priv->lock); + kfree(mc); + } + return ret; +} +EXPORT_SYMBOL(rdma_join_multicast); + +void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr) +{ + struct rdma_id_private *id_priv; + struct cma_multicast *mc; + + id_priv = container_of(id, struct rdma_id_private, id); + spin_lock_irq(&id_priv->lock); + list_for_each_entry(mc, &id_priv->mc_list, list) { + if (!memcmp(&mc->addr, addr, ip_addr_size(addr))) { + list_del(&mc->list); + spin_unlock_irq(&id_priv->lock); + + if (id->qp) + ib_detach_mcast(id->qp, + &mc->multicast.ib->rec.mgid, + mc->multicast.ib->rec.mlid); + ib_sa_free_multicast(mc->multicast.ib); + kfree(mc); + return; + } + } + spin_unlock_irq(&id_priv->lock); +} +EXPORT_SYMBOL(rdma_leave_multicast); + static void cma_add_one(struct ib_device *device) { struct cma_device *cma_dev; @@ -2522,6 +2792,7 @@ static void cma_cleanup(void) idr_destroy(&sdp_ps); idr_destroy(&tcp_ps); idr_destroy(&udp_ps); + idr_destroy(&ipoib_ps); } module_init(cma_init); diff --git a/drivers/infiniband/core/ucma.c b/drivers/infiniband/core/ucma.c index 6b81b98961c7..b516b93b8550 100644 --- a/drivers/infiniband/core/ucma.c +++ b/drivers/infiniband/core/ucma.c @@ -70,10 +70,24 @@ struct ucma_context { u64 uid; struct list_head list; + struct list_head mc_list; +}; + +struct ucma_multicast { + struct ucma_context *ctx; + int id; + int events_reported; + + u64 uid; + struct list_head list; + struct sockaddr addr; + u8 pad[sizeof(struct sockaddr_in6) - + sizeof(struct sockaddr)]; }; struct ucma_event { struct ucma_context *ctx; + struct ucma_multicast *mc; struct list_head list; struct rdma_cm_id *cm_id; struct rdma_ucm_event_resp resp; @@ -81,6 +95,7 @@ struct ucma_event { static DEFINE_MUTEX(mut); static DEFINE_IDR(ctx_idr); +static DEFINE_IDR(multicast_idr); static inline struct ucma_context *_ucma_find_context(int id, struct ucma_file *file) @@ -124,6 +139,7 @@ static struct ucma_context *ucma_alloc_ctx(struct ucma_file *file) atomic_set(&ctx->ref, 1); init_completion(&ctx->comp); + INIT_LIST_HEAD(&ctx->mc_list); ctx->file = file; do { @@ -147,6 +163,37 @@ error: return NULL; } +static struct ucma_multicast* ucma_alloc_multicast(struct ucma_context *ctx) +{ + struct ucma_multicast *mc; + int ret; + + mc = kzalloc(sizeof(*mc), GFP_KERNEL); + if (!mc) + return NULL; + + do { + ret = idr_pre_get(&multicast_idr, GFP_KERNEL); + if (!ret) + goto error; + + mutex_lock(&mut); + ret = idr_get_new(&multicast_idr, mc, &mc->id); + mutex_unlock(&mut); + } while (ret == -EAGAIN); + + if (ret) + goto error; + + mc->ctx = ctx; + list_add_tail(&mc->list, &ctx->mc_list); + return mc; + +error: + kfree(mc); + return NULL; +} + static void ucma_copy_conn_event(struct rdma_ucm_conn_param *dst, struct rdma_conn_param *src) { @@ -180,8 +227,19 @@ static void ucma_set_event_context(struct ucma_context *ctx, struct ucma_event *uevent) { uevent->ctx = ctx; - uevent->resp.uid = ctx->uid; - uevent->resp.id = ctx->id; + switch (event->event) { + case RDMA_CM_EVENT_MULTICAST_JOIN: + case RDMA_CM_EVENT_MULTICAST_ERROR: + uevent->mc = (struct ucma_multicast *) + event->param.ud.private_data; + uevent->resp.uid = uevent->mc->uid; + uevent->resp.id = uevent->mc->id; + break; + default: + uevent->resp.uid = ctx->uid; + uevent->resp.id = ctx->id; + break; + } } static int ucma_event_handler(struct rdma_cm_id *cm_id, @@ -199,7 +257,7 @@ static int ucma_event_handler(struct rdma_cm_id *cm_id, ucma_set_event_context(ctx, event, uevent); uevent->resp.event = event->event; uevent->resp.status = event->status; - if (cm_id->ps == RDMA_PS_UDP) + if (cm_id->ps == RDMA_PS_UDP || cm_id->ps == RDMA_PS_IPOIB) ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud); else ucma_copy_conn_event(&uevent->resp.param.conn, @@ -290,6 +348,8 @@ static ssize_t ucma_get_event(struct ucma_file *file, const char __user *inbuf, list_del(&uevent->list); uevent->ctx->events_reported++; + if (uevent->mc) + uevent->mc->events_reported++; kfree(uevent); done: mutex_unlock(&file->mut); @@ -342,6 +402,19 @@ err1: return ret; } +static void ucma_cleanup_multicast(struct ucma_context *ctx) +{ + struct ucma_multicast *mc, *tmp; + + mutex_lock(&mut); + list_for_each_entry_safe(mc, tmp, &ctx->mc_list, list) { + list_del(&mc->list); + idr_remove(&multicast_idr, mc->id); + kfree(mc); + } + mutex_unlock(&mut); +} + static void ucma_cleanup_events(struct ucma_context *ctx) { struct ucma_event *uevent, *tmp; @@ -360,6 +433,19 @@ static void ucma_cleanup_events(struct ucma_context *ctx) } } +static void ucma_cleanup_mc_events(struct ucma_multicast *mc) +{ + struct ucma_event *uevent, *tmp; + + list_for_each_entry_safe(uevent, tmp, &mc->ctx->file->event_list, list) { + if (uevent->mc != mc) + continue; + + list_del(&uevent->list); + kfree(uevent); + } +} + static int ucma_free_ctx(struct ucma_context *ctx) { int events_reported; @@ -367,6 +453,8 @@ static int ucma_free_ctx(struct ucma_context *ctx) /* No new events will be generated after destroying the id. */ rdma_destroy_id(ctx->cm_id); + ucma_cleanup_multicast(ctx); + /* Cleanup events not yet reported to the user. */ mutex_lock(&ctx->file->mut); ucma_cleanup_events(ctx); @@ -731,6 +819,114 @@ static ssize_t ucma_notify(struct ucma_file *file, const char __user *inbuf, return ret; } +static ssize_t ucma_join_multicast(struct ucma_file *file, + const char __user *inbuf, + int in_len, int out_len) +{ + struct rdma_ucm_join_mcast cmd; + struct rdma_ucm_create_id_resp resp; + struct ucma_context *ctx; + struct ucma_multicast *mc; + int ret; + + if (out_len < sizeof(resp)) + return -ENOSPC; + + if (copy_from_user(&cmd, inbuf, sizeof(cmd))) + return -EFAULT; + + ctx = ucma_get_ctx(file, cmd.id); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + mutex_lock(&file->mut); + mc = ucma_alloc_multicast(ctx); + if (IS_ERR(mc)) { + ret = PTR_ERR(mc); + goto err1; + } + + mc->uid = cmd.uid; + memcpy(&mc->addr, &cmd.addr, sizeof cmd.addr); + ret = rdma_join_multicast(ctx->cm_id, &mc->addr, mc); + if (ret) + goto err2; + + resp.id = mc->id; + if (copy_to_user((void __user *)(unsigned long)cmd.response, + &resp, sizeof(resp))) { + ret = -EFAULT; + goto err3; + } + + mutex_unlock(&file->mut); + ucma_put_ctx(ctx); + return 0; + +err3: + rdma_leave_multicast(ctx->cm_id, &mc->addr); + ucma_cleanup_mc_events(mc); +err2: + mutex_lock(&mut); + idr_remove(&multicast_idr, mc->id); + mutex_unlock(&mut); + list_del(&mc->list); + kfree(mc); +err1: + mutex_unlock(&file->mut); + ucma_put_ctx(ctx); + return ret; +} + +static ssize_t ucma_leave_multicast(struct ucma_file *file, + const char __user *inbuf, + int in_len, int out_len) +{ + struct rdma_ucm_destroy_id cmd; + struct rdma_ucm_destroy_id_resp resp; + struct ucma_multicast *mc; + int ret = 0; + + if (out_len < sizeof(resp)) + return -ENOSPC; + + if (copy_from_user(&cmd, inbuf, sizeof(cmd))) + return -EFAULT; + + mutex_lock(&mut); + mc = idr_find(&multicast_idr, cmd.id); + if (!mc) + mc = ERR_PTR(-ENOENT); + else if (mc->ctx->file != file) + mc = ERR_PTR(-EINVAL); + else { + idr_remove(&multicast_idr, mc->id); + atomic_inc(&mc->ctx->ref); + } + mutex_unlock(&mut); + + if (IS_ERR(mc)) { + ret = PTR_ERR(mc); + goto out; + } + + rdma_leave_multicast(mc->ctx->cm_id, &mc->addr); + mutex_lock(&mc->ctx->file->mut); + ucma_cleanup_mc_events(mc); + list_del(&mc->list); + mutex_unlock(&mc->ctx->file->mut); + + ucma_put_ctx(mc->ctx); + resp.events_reported = mc->events_reported; + kfree(mc); + + if (copy_to_user((void __user *)(unsigned long)cmd.response, + &resp, sizeof(resp))) + ret = -EFAULT; +out: + return ret; +} + static ssize_t (*ucma_cmd_table[])(struct ucma_file *file, const char __user *inbuf, int in_len, int out_len) = { @@ -750,6 +946,8 @@ static ssize_t (*ucma_cmd_table[])(struct ucma_file *file, [RDMA_USER_CM_CMD_GET_OPTION] = NULL, [RDMA_USER_CM_CMD_SET_OPTION] = NULL, [RDMA_USER_CM_CMD_NOTIFY] = ucma_notify, + [RDMA_USER_CM_CMD_JOIN_MCAST] = ucma_join_multicast, + [RDMA_USER_CM_CMD_LEAVE_MCAST] = ucma_leave_multicast, }; static ssize_t ucma_write(struct file *filp, const char __user *buf, diff --git a/include/rdma/rdma_cm.h b/include/rdma/rdma_cm.h index 36cd8a8526a0..2d6a7705eae7 100644 --- a/include/rdma/rdma_cm.h +++ b/include/rdma/rdma_cm.h @@ -52,10 +52,13 @@ enum rdma_cm_event_type { RDMA_CM_EVENT_ESTABLISHED, RDMA_CM_EVENT_DISCONNECTED, RDMA_CM_EVENT_DEVICE_REMOVAL, + RDMA_CM_EVENT_MULTICAST_JOIN, + RDMA_CM_EVENT_MULTICAST_ERROR }; enum rdma_port_space { RDMA_PS_SDP = 0x0001, + RDMA_PS_IPOIB= 0x0002, RDMA_PS_TCP = 0x0106, RDMA_PS_UDP = 0x0111, RDMA_PS_SCTP = 0x0183 @@ -294,5 +297,21 @@ int rdma_reject(struct rdma_cm_id *id, const void *private_data, */ int rdma_disconnect(struct rdma_cm_id *id); -#endif /* RDMA_CM_H */ +/** + * rdma_join_multicast - Join the multicast group specified by the given + * address. + * @id: Communication identifier associated with the request. + * @addr: Multicast address identifying the group to join. + * @context: User-defined context associated with the join request, returned + * to the user through the private_data pointer in multicast events. + */ +int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr, + void *context); +/** + * rdma_leave_multicast - Leave the multicast group specified by the given + * address. + */ +void rdma_leave_multicast(struct rdma_cm_id *id, struct sockaddr *addr); + +#endif /* RDMA_CM_H */ diff --git a/include/rdma/rdma_cm_ib.h b/include/rdma/rdma_cm_ib.h index 9b176df1d667..950424b38f16 100644 --- a/include/rdma/rdma_cm_ib.h +++ b/include/rdma/rdma_cm_ib.h @@ -44,7 +44,7 @@ int rdma_set_ib_paths(struct rdma_cm_id *id, struct ib_sa_path_rec *path_rec, int num_paths); -/* Global qkey for UD QPs and multicast groups. */ -#define RDMA_UD_QKEY 0x01234567 +/* Global qkey for UDP QPs and multicast groups. */ +#define RDMA_UDP_QKEY 0x01234567 #endif /* RDMA_CM_IB_H */ diff --git a/include/rdma/rdma_user_cm.h b/include/rdma/rdma_user_cm.h index 9572ab8eeac1..f632b0c007c9 100644 --- a/include/rdma/rdma_user_cm.h +++ b/include/rdma/rdma_user_cm.h @@ -38,7 +38,7 @@ #include #include -#define RDMA_USER_CM_ABI_VERSION 3 +#define RDMA_USER_CM_ABI_VERSION 4 #define RDMA_MAX_PRIVATE_DATA 256 @@ -58,7 +58,9 @@ enum { RDMA_USER_CM_CMD_GET_EVENT, RDMA_USER_CM_CMD_GET_OPTION, RDMA_USER_CM_CMD_SET_OPTION, - RDMA_USER_CM_CMD_NOTIFY + RDMA_USER_CM_CMD_NOTIFY, + RDMA_USER_CM_CMD_JOIN_MCAST, + RDMA_USER_CM_CMD_LEAVE_MCAST }; /* @@ -188,6 +190,13 @@ struct rdma_ucm_notify { __u32 event; }; +struct rdma_ucm_join_mcast { + __u64 response; /* rdma_ucm_create_id_resp */ + __u64 uid; + struct sockaddr_in6 addr; + __u32 id; +}; + struct rdma_ucm_get_event { __u64 response; }; -- cgit v1.2.3