[PATCH] IB: Add user-supplied context to userspace CM ABI

- Add user specified context to all uCM events. Users will not retrieve
any events associated with the context after destroying the corresponding
cm_id.
- Provide the ib_cm_init_qp_attr() call to userspace clients of the CM.
This call may be used to set QP attributes properly before modifying the QP.
- Fixes some error handling synchonization and cleanup issues.
- Performs some minor code cleanup.

Signed-off-by: Sean Hefty <[e-mail]>
Signed-off-by: Roland Dreier <[e-mail]>

drivers/infiniband/core/ucm.c

···
 
 static struct semaphore ctx_id_mutex;
 static struct idr       ctx_id_table;
-static int              ctx_id_rover = 0;
 
 static struct ib_ucm_context *ib_ucm_ctx_get(struct ib_ucm_file *file, int id)
 {
···
 		wake_up(&ctx->wait);
 }
 
-static ssize_t ib_ucm_destroy_ctx(struct ib_ucm_file *file, int id)
+static inline int ib_ucm_new_cm_id(int event)
 {
-	struct ib_ucm_context *ctx;
+	return event == IB_CM_REQ_RECEIVED || event == IB_CM_SIDR_REQ_RECEIVED;
+}
+
+static void ib_ucm_cleanup_events(struct ib_ucm_context *ctx)
+{
 	struct ib_ucm_event *uevent;
 
-	down(&ctx_id_mutex);
-	ctx = idr_find(&ctx_id_table, id);
-	if (!ctx)
-		ctx = ERR_PTR(-ENOENT);
-	else if (ctx->file != file)
-		ctx = ERR_PTR(-EINVAL);
-	else
-		idr_remove(&ctx_id_table, ctx->id);
-	up(&ctx_id_mutex);
-
-	if (IS_ERR(ctx))
-		return PTR_ERR(ctx);
-
-	atomic_dec(&ctx->ref);
-	wait_event(ctx->wait, !atomic_read(&ctx->ref));
-
-	/* No new events will be generated after destroying the cm_id. */
-	if (!IS_ERR(ctx->cm_id))
-		ib_destroy_cm_id(ctx->cm_id);
-
-	/* Cleanup events not yet reported to the user. */
-	down(&file->mutex);
+	down(&ctx->file->mutex);
 	list_del(&ctx->file_list);
 	while (!list_empty(&ctx->events)) {
 
···
 		list_del(&uevent->ctx_list);
 
 		/* clear incoming connections. */
-		if (uevent->cm_id)
+		if (ib_ucm_new_cm_id(uevent->resp.event))
 			ib_destroy_cm_id(uevent->cm_id);
 
 		kfree(uevent);
 	}
-	up(&file->mutex);
-
-	kfree(ctx);
-	return 0;
+	up(&ctx->file->mutex);
 }
 
 static struct ib_ucm_context *ib_ucm_ctx_alloc(struct ib_ucm_file *file)
···
 	if (!ctx)
 		return NULL;
 
+	memset(ctx, 0, sizeof *ctx);
 	atomic_set(&ctx->ref, 1);
 	init_waitqueue_head(&ctx->wait);
 	ctx->file = file;
-
 	INIT_LIST_HEAD(&ctx->events);
 
-	list_add_tail(&ctx->file_list, &file->ctxs);
+	do {
+		result = idr_pre_get(&ctx_id_table, GFP_KERNEL);
+		if (!result)
+			goto error;
 
-	ctx_id_rover = (ctx_id_rover + 1) & INT_MAX;
-retry:
-	result = idr_pre_get(&ctx_id_table, GFP_KERNEL);
-	if (!result)
-		goto error;
+		down(&ctx_id_mutex);
+		result = idr_get_new(&ctx_id_table, ctx, &ctx->id);
+		up(&ctx_id_mutex);
+	} while (result == -EAGAIN);
 
-	down(&ctx_id_mutex);
-	result = idr_get_new_above(&ctx_id_table, ctx, ctx_id_rover, &ctx->id);
-	up(&ctx_id_mutex);
-
-	if (result == -EAGAIN)
-		goto retry;
 	if (result)
 		goto error;
 
+	list_add_tail(&ctx->file_list, &file->ctxs);
 	ucm_dbg("Allocated CM ID <%d>\n", ctx->id);
-
 	return ctx;
-error:
-	list_del(&ctx->file_list);
-	kfree(ctx);
 
+error:
+	kfree(ctx);
 	return NULL;
 }
 /*
···
 		kpath->packet_life_time_selector;
 }
 
-static void ib_ucm_event_req_get(struct ib_ucm_context *ctx,
-				 struct ib_ucm_req_event_resp *ureq,
+static void ib_ucm_event_req_get(struct ib_ucm_req_event_resp *ureq,
 				 struct ib_cm_req_event_param *kreq)
 {
-	ureq->listen_id = ctx->id;
-
 	ureq->remote_ca_guid             = kreq->remote_ca_guid;
 	ureq->remote_qkey                = kreq->remote_qkey;
 	ureq->remote_qpn                 = kreq->remote_qpn;
···
 	urep->srq                 = krep->srq;
 }
 
-static void ib_ucm_event_sidr_req_get(struct ib_ucm_context *ctx,
-				      struct ib_ucm_sidr_req_event_resp *ureq,
-				      struct ib_cm_sidr_req_event_param *kreq)
-{
-	ureq->listen_id = ctx->id;
-	ureq->pkey      = kreq->pkey;
-}
-
 static void ib_ucm_event_sidr_rep_get(struct ib_ucm_sidr_rep_event_resp *urep,
 				      struct ib_cm_sidr_rep_event_param *krep)
 {
···
 	urep->qpn    = krep->qpn;
 };
 
-static int ib_ucm_event_process(struct ib_ucm_context *ctx,
-				struct ib_cm_event *evt,
+static int ib_ucm_event_process(struct ib_cm_event *evt,
 				struct ib_ucm_event *uvt)
 {
 	void *info = NULL;
 
 	switch (evt->event) {
 	case IB_CM_REQ_RECEIVED:
-		ib_ucm_event_req_get(ctx, &uvt->resp.u.req_resp,
+		ib_ucm_event_req_get(&uvt->resp.u.req_resp,
 				     &evt->param.req_rcvd);
 		uvt->data_len      = IB_CM_REQ_PRIVATE_DATA_SIZE;
 		uvt->resp.present  = IB_UCM_PRES_PRIMARY;
···
 		info	      = evt->param.apr_rcvd.apr_info;
 		break;
 	case IB_CM_SIDR_REQ_RECEIVED:
-		ib_ucm_event_sidr_req_get(ctx, &uvt->resp.u.sidr_req_resp,
-					  &evt->param.sidr_req_rcvd);
+		uvt->resp.u.sidr_req_resp.pkey = 
+					evt->param.sidr_req_rcvd.pkey;
 		uvt->data_len = IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE;
 		break;
 	case IB_CM_SIDR_REP_RECEIVED:
···
 	struct ib_ucm_event *uevent;
 	struct ib_ucm_context *ctx;
 	int result = 0;
-	int id;
 
 	ctx = cm_id->context;
-
-	if (event->event == IB_CM_REQ_RECEIVED ||
-	    event->event == IB_CM_SIDR_REQ_RECEIVED)
-		id = IB_UCM_CM_ID_INVALID;
-	else
-		id = ctx->id;
 
 	uevent = kmalloc(sizeof(*uevent), GFP_KERNEL);
 	if (!uevent)
 		goto err1;
 
 	memset(uevent, 0, sizeof(*uevent));
-	uevent->resp.id    = id;
+	uevent->ctx = ctx;
+	uevent->cm_id = cm_id;
+	uevent->resp.uid = ctx->uid;
+	uevent->resp.id = ctx->id;
 	uevent->resp.event = event->event;
 
-	result = ib_ucm_event_process(ctx, event, uevent);
+	result = ib_ucm_event_process(event, uevent);
 	if (result)
 		goto err2;
-
-	uevent->ctx   = ctx;
-	uevent->cm_id = (id == IB_UCM_CM_ID_INVALID) ? cm_id : NULL;
 
 	down(&ctx->file->mutex);
 	list_add_tail(&uevent->file_list, &ctx->file->events);
···
 	kfree(uevent);
 err1:
 	/* Destroy new cm_id's */
-	return (id == IB_UCM_CM_ID_INVALID);
+	return ib_ucm_new_cm_id(event->event);
 }
 
 static ssize_t ib_ucm_event(struct ib_ucm_file *file,
···
 {
 	struct ib_ucm_context *ctx;
 	struct ib_ucm_event_get cmd;
-	struct ib_ucm_event *uevent = NULL;
+	struct ib_ucm_event *uevent;
 	int result = 0;
 	DEFINE_WAIT(wait);
 
···
 	 * wait
 	 */
 	down(&file->mutex);
-
 	while (list_empty(&file->events)) {
 
 		if (file->filp->f_flags & O_NONBLOCK) {
···
 
 	uevent = list_entry(file->events.next, struct ib_ucm_event, file_list);
 
-	if (!uevent->cm_id)
-		goto user;
+	if (ib_ucm_new_cm_id(uevent->resp.event)) {
+		ctx = ib_ucm_ctx_alloc(file);
+		if (!ctx) {
+			result = -ENOMEM;
+			goto done;
+		}
 
-	ctx = ib_ucm_ctx_alloc(file);
-	if (!ctx) {
-		result = -ENOMEM;
-		goto done;
+		ctx->cm_id = uevent->cm_id;
+		ctx->cm_id->context = ctx;
+		uevent->resp.id = ctx->id;
 	}
 
-	ctx->cm_id          = uevent->cm_id;
-	ctx->cm_id->context = ctx;
-
-	uevent->resp.id = ctx->id;
-
-user:
 	if (copy_to_user((void __user *)(unsigned long)cmd.response,
 			 &uevent->resp, sizeof(uevent->resp))) {
 		result = -EFAULT;
···
 	}
 
 	if (uevent->data) {
-
 		if (cmd.data_len < uevent->data_len) {
 			result = -ENOMEM;
 			goto done;
 		}
-
 		if (copy_to_user((void __user *)(unsigned long)cmd.data,
 				 uevent->data, uevent->data_len)) {
 			result = -EFAULT;
···
 	}
 
 	if (uevent->info) {
-
 		if (cmd.info_len < uevent->info_len) {
 			result = -ENOMEM;
 			goto done;
 		}
-
 		if (copy_to_user((void __user *)(unsigned long)cmd.info,
 				 uevent->info, uevent->info_len)) {
 			result = -EFAULT;
···
 
 	list_del(&uevent->file_list);
 	list_del(&uevent->ctx_list);
+	uevent->ctx->events_reported++;
 
 	kfree(uevent->data);
 	kfree(uevent->info);
···
 	if (!ctx)
 		return -ENOMEM;
 
+	ctx->uid = cmd.uid;
 	ctx->cm_id = ib_create_cm_id(ib_ucm_event_handler, ctx);
 	if (IS_ERR(ctx->cm_id)) {
 		result = PTR_ERR(ctx->cm_id);
···
 	return 0;
 
 err:
-	ib_ucm_destroy_ctx(file, ctx->id);
+	down(&ctx_id_mutex);
+	idr_remove(&ctx_id_table, ctx->id);
+	up(&ctx_id_mutex);
+
+	if (!IS_ERR(ctx->cm_id))
+		ib_destroy_cm_id(ctx->cm_id);
+
+	kfree(ctx);
 	return result;
 }
 
···
 				 int in_len, int out_len)
 {
 	struct ib_ucm_destroy_id cmd;
+	struct ib_ucm_destroy_id_resp resp;
+	struct ib_ucm_context *ctx;
+	int result = 0;
+
+	if (out_len < sizeof(resp))
+		return -ENOSPC;
 
 	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
 		return -EFAULT;
 
-	return ib_ucm_destroy_ctx(file, cmd.id);
+	down(&ctx_id_mutex);
+	ctx = idr_find(&ctx_id_table, cmd.id);
+	if (!ctx)
+		ctx = ERR_PTR(-ENOENT);
+	else if (ctx->file != file)
+		ctx = ERR_PTR(-EINVAL);
+	else
+		idr_remove(&ctx_id_table, ctx->id);
+	up(&ctx_id_mutex);
+
+	if (IS_ERR(ctx))
+		return PTR_ERR(ctx);
+
+	atomic_dec(&ctx->ref);
+	wait_event(ctx->wait, !atomic_read(&ctx->ref));
+
+	/* No new events will be generated after destroying the cm_id. */
+	ib_destroy_cm_id(ctx->cm_id);
+	/* Cleanup events not yet reported to the user. */
+	ib_ucm_cleanup_events(ctx);
+
+	resp.events_reported = ctx->events_reported;
+	if (copy_to_user((void __user *)(unsigned long)cmd.response,
+			 &resp, sizeof(resp)))
+		result = -EFAULT;
+
+	kfree(ctx);
+	return result;
 }
 
 static ssize_t ib_ucm_attr_id(struct ib_ucm_file *file,
···
 			 &resp, sizeof(resp)))
 		result = -EFAULT;
 
+	ib_ucm_ctx_put(ctx);
+	return result;
+}
+
+static void ib_ucm_copy_ah_attr(struct ib_ucm_ah_attr *dest_attr,
+				struct ib_ah_attr *src_attr)
+{
+	memcpy(dest_attr->grh_dgid, src_attr->grh.dgid.raw,
+	       sizeof src_attr->grh.dgid);
+	dest_attr->grh_flow_label = src_attr->grh.flow_label;
+	dest_attr->grh_sgid_index = src_attr->grh.sgid_index;
+	dest_attr->grh_hop_limit = src_attr->grh.hop_limit;
+	dest_attr->grh_traffic_class = src_attr->grh.traffic_class;
+
+	dest_attr->dlid = src_attr->dlid;
+	dest_attr->sl = src_attr->sl;
+	dest_attr->src_path_bits = src_attr->src_path_bits;
+	dest_attr->static_rate = src_attr->static_rate;
+	dest_attr->is_global = (src_attr->ah_flags & IB_AH_GRH);
+	dest_attr->port_num = src_attr->port_num;
+}
+
+static void ib_ucm_copy_qp_attr(struct ib_ucm_init_qp_attr_resp *dest_attr,
+				struct ib_qp_attr *src_attr)
+{
+	dest_attr->cur_qp_state = src_attr->cur_qp_state;
+	dest_attr->path_mtu = src_attr->path_mtu;
+	dest_attr->path_mig_state = src_attr->path_mig_state;
+	dest_attr->qkey = src_attr->qkey;
+	dest_attr->rq_psn = src_attr->rq_psn;
+	dest_attr->sq_psn = src_attr->sq_psn;
+	dest_attr->dest_qp_num = src_attr->dest_qp_num;
+	dest_attr->qp_access_flags = src_attr->qp_access_flags;
+
+	dest_attr->max_send_wr = src_attr->cap.max_send_wr;
+	dest_attr->max_recv_wr = src_attr->cap.max_recv_wr;
+	dest_attr->max_send_sge = src_attr->cap.max_send_sge;
+	dest_attr->max_recv_sge = src_attr->cap.max_recv_sge;
+	dest_attr->max_inline_data = src_attr->cap.max_inline_data;
+
+	ib_ucm_copy_ah_attr(&dest_attr->ah_attr, &src_attr->ah_attr);
+	ib_ucm_copy_ah_attr(&dest_attr->alt_ah_attr, &src_attr->alt_ah_attr);
+
+	dest_attr->pkey_index = src_attr->pkey_index;
+	dest_attr->alt_pkey_index = src_attr->alt_pkey_index;
+	dest_attr->en_sqd_async_notify = src_attr->en_sqd_async_notify;
+	dest_attr->sq_draining = src_attr->sq_draining;
+	dest_attr->max_rd_atomic = src_attr->max_rd_atomic;
+	dest_attr->max_dest_rd_atomic = src_attr->max_dest_rd_atomic;
+	dest_attr->min_rnr_timer = src_attr->min_rnr_timer;
+	dest_attr->port_num = src_attr->port_num;
+	dest_attr->timeout = src_attr->timeout;
+	dest_attr->retry_cnt = src_attr->retry_cnt;
+	dest_attr->rnr_retry = src_attr->rnr_retry;
+	dest_attr->alt_port_num = src_attr->alt_port_num;
+	dest_attr->alt_timeout = src_attr->alt_timeout;
+}
+
+static ssize_t ib_ucm_init_qp_attr(struct ib_ucm_file *file,
+				   const char __user *inbuf,
+				   int in_len, int out_len)
+{
+	struct ib_ucm_init_qp_attr_resp resp;
+	struct ib_ucm_init_qp_attr cmd;
+	struct ib_ucm_context *ctx;
+	struct ib_qp_attr qp_attr;
+	int result = 0;
+
+	if (out_len < sizeof(resp))
+		return -ENOSPC;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	ctx = ib_ucm_ctx_get(file, cmd.id);
+	if (IS_ERR(ctx))
+		return PTR_ERR(ctx);
+
+	resp.qp_attr_mask = 0;
+	memset(&qp_attr, 0, sizeof qp_attr);
+	qp_attr.qp_state = cmd.qp_state;
+	result = ib_cm_init_qp_attr(ctx->cm_id, &qp_attr, &resp.qp_attr_mask);
+	if (result)
+		goto out;
+
+	ib_ucm_copy_qp_attr(&resp, &qp_attr);
+
+	if (copy_to_user((void __user *)(unsigned long)cmd.response,
+			 &resp, sizeof(resp)))
+		result = -EFAULT;
+
+out:
 	ib_ucm_ctx_put(ctx);
 	return result;
 }
···
 
 	ctx = ib_ucm_ctx_get(file, cmd.id);
 	if (!IS_ERR(ctx)) {
+		ctx->uid = cmd.uid;
 		result = ib_send_cm_rep(ctx->cm_id, &param);
 		ib_ucm_ctx_put(ctx);
 	} else
···
 	[IB_USER_CM_CMD_SEND_SIDR_REQ] = ib_ucm_send_sidr_req,
 	[IB_USER_CM_CMD_SEND_SIDR_REP] = ib_ucm_send_sidr_rep,
 	[IB_USER_CM_CMD_EVENT]	       = ib_ucm_event,
+	[IB_USER_CM_CMD_INIT_QP_ATTR]  = ib_ucm_init_qp_attr,
 };
 
 static ssize_t ib_ucm_write(struct file *filp, const char __user *buf,
···
 
 	down(&file->mutex);
 	while (!list_empty(&file->ctxs)) {
-
 		ctx = list_entry(file->ctxs.next,
 				 struct ib_ucm_context, file_list);
-
 		up(&file->mutex);
-		ib_ucm_destroy_ctx(file, ctx->id);
+
+		down(&ctx_id_mutex);
+		idr_remove(&ctx_id_table, ctx->id);
+		up(&ctx_id_mutex);
+
+		ib_destroy_cm_id(ctx->cm_id);
+		ib_ucm_cleanup_events(ctx);
+		kfree(ctx);
+
 		down(&file->mutex);
 	}
 	up(&file->mutex);

drivers/infiniband/core/ucm.h

···
 /*
  * Copyright (c) 2005 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
···
 #include <rdma/ib_cm.h>
 #include <rdma/ib_user_cm.h>
 
-#define IB_UCM_CM_ID_INVALID 0xffffffff
-
 struct ib_ucm_file {
 	struct semaphore mutex;
 	struct file *filp;
···
 	int                 id;
 	wait_queue_head_t   wait;
 	atomic_t            ref;
+	int		    events_reported;
 
 	struct ib_ucm_file *file;
 	struct ib_cm_id    *cm_id;
+	__u64		   uid;
 
 	struct list_head    events;    /* list of pending events. */
 	struct list_head    file_list; /* member in file ctx list */
···
 	struct list_head file_list; /* member in file event list */
 	struct list_head ctx_list;  /* member in ctx event list */
 
+	struct ib_cm_id *cm_id;
 	struct ib_ucm_event_resp resp;
 	void *data;
 	void *info;
 	int data_len;
 	int info_len;
-	/*
-	 * new connection identifiers needs to be saved until
-	 * userspace can get a handle on them.
-	 */
-	struct ib_cm_id *cm_id;
 };
 
 #endif /* UCM_H */

include/rdma/ib_user_cm.h

···
 /*
  * Copyright (c) 2005 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
···
 
 #include <linux/types.h>
 
-#define IB_USER_CM_ABI_VERSION 1
+#define IB_USER_CM_ABI_VERSION 2
 
 enum {
 	IB_USER_CM_CMD_CREATE_ID,
···
 	IB_USER_CM_CMD_SEND_SIDR_REP,
 
 	IB_USER_CM_CMD_EVENT,
+	IB_USER_CM_CMD_INIT_QP_ATTR,
 };
 /*
  * command ABI structures.
···
 };
 
 struct ib_ucm_create_id {
+	__u64 uid;
 	__u64 response;
 };
 
···
 };
 
 struct ib_ucm_destroy_id {
+	__u64 response;
 	__u32 id;
+};
+
+struct ib_ucm_destroy_id_resp {
+	__u32 events_reported;
 };
 
 struct ib_ucm_attr_id {
···
 	__be64 service_mask;
 	__be32 local_id;
 	__be32 remote_id;
+};
+
+struct ib_ucm_init_qp_attr {
+	__u64 response;
+	__u32 id;
+	__u32 qp_state;
+};
+
+struct ib_ucm_ah_attr {
+	__u8	grh_dgid[16];
+	__u32	grh_flow_label;
+	__u16	dlid;
+	__u16	reserved;
+	__u8	grh_sgid_index;
+	__u8	grh_hop_limit;
+	__u8	grh_traffic_class;
+	__u8	sl;
+	__u8	src_path_bits;
+	__u8	static_rate;
+	__u8	is_global;
+	__u8	port_num;
+};
+
+struct ib_ucm_init_qp_attr_resp {
+	__u32	qp_attr_mask;
+	__u32	qp_state;
+	__u32	cur_qp_state;
+	__u32	path_mtu;
+	__u32	path_mig_state;
+	__u32	qkey;
+	__u32	rq_psn;
+	__u32	sq_psn;
+	__u32	dest_qp_num;
+	__u32	qp_access_flags;
+
+	struct ib_ucm_ah_attr	ah_attr;
+	struct ib_ucm_ah_attr	alt_ah_attr;
+
+	/* ib_qp_cap */
+	__u32	max_send_wr;
+	__u32	max_recv_wr;
+	__u32	max_send_sge;
+	__u32	max_recv_sge;
+	__u32	max_inline_data;
+
+	__u16	pkey_index;
+	__u16	alt_pkey_index;
+	__u8	en_sqd_async_notify;
+	__u8	sq_draining;
+	__u8	max_rd_atomic;
+	__u8	max_dest_rd_atomic;
+	__u8	min_rnr_timer;
+	__u8	port_num;
+	__u8	timeout;
+	__u8	retry_cnt;
+	__u8	rnr_retry;
+	__u8	alt_port_num;
+	__u8	alt_timeout;
 };
 
 struct ib_ucm_listen {
···
 };
 
 struct ib_ucm_rep {
+	__u64 uid;
 	__u64 data;
 	__u32 id;
 	__u32 qpn;
···
 };
 
 struct ib_ucm_req_event_resp {
-	__u32                  listen_id;
 	/* device */
 	/* port */
 	struct ib_ucm_path_rec primary_path;
···
 };
 
 struct ib_ucm_sidr_req_event_resp {
-	__u32 listen_id;
 	/* device */
 	/* port */
 	__u16 pkey;
···
 #define IB_UCM_PRES_ALTERNATE 0x08
 
 struct ib_ucm_event_resp {
+	__u64 uid;
 	__u32 id;
 	__u32 event;
 	__u32 present;