1 /*
2 * Copyright (c) 2006 Oracle. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 *
32 */
33 #include <linux/kernel.h>
34 #include <linux/in.h>
35 #include <linux/if.h>
36 #include <linux/netdevice.h>
37 #include <linux/inetdevice.h>
38 #include <linux/if_arp.h>
39 #include <linux/delay.h>
40 #include <linux/slab.h>
41 #include <linux/module.h>
42
43 #include "rds.h"
44 #include "iw.h"
45
46 unsigned int fastreg_pool_size = RDS_FASTREG_POOL_SIZE;
47 unsigned int fastreg_message_size = RDS_FASTREG_SIZE + 1; /* +1 allows for unaligned MRs */
48
49 module_param(fastreg_pool_size, int, 0444);
50 MODULE_PARM_DESC(fastreg_pool_size, " Max number of fastreg MRs per device");
51 module_param(fastreg_message_size, int, 0444);
52 MODULE_PARM_DESC(fastreg_message_size, " Max size of a RDMA transfer (fastreg MRs)");
53
54 struct list_head rds_iw_devices;
55
56 /* NOTE: if also grabbing iwdev lock, grab this first */
57 DEFINE_SPINLOCK(iw_nodev_conns_lock);
58 LIST_HEAD(iw_nodev_conns);
59
rds_iw_add_one(struct ib_device * device)60 static void rds_iw_add_one(struct ib_device *device)
61 {
62 struct rds_iw_device *rds_iwdev;
63 struct ib_device_attr *dev_attr;
64
65 /* Only handle iwarp devices */
66 if (device->node_type != RDMA_NODE_RNIC)
67 return;
68
69 dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL);
70 if (!dev_attr)
71 return;
72
73 if (ib_query_device(device, dev_attr)) {
74 rdsdebug("Query device failed for %s\n", device->name);
75 goto free_attr;
76 }
77
78 rds_iwdev = kmalloc(sizeof *rds_iwdev, GFP_KERNEL);
79 if (!rds_iwdev)
80 goto free_attr;
81
82 spin_lock_init(&rds_iwdev->spinlock);
83
84 rds_iwdev->dma_local_lkey = !!(dev_attr->device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY);
85 rds_iwdev->max_wrs = dev_attr->max_qp_wr;
86 rds_iwdev->max_sge = min(dev_attr->max_sge, RDS_IW_MAX_SGE);
87
88 rds_iwdev->dev = device;
89 rds_iwdev->pd = ib_alloc_pd(device);
90 if (IS_ERR(rds_iwdev->pd))
91 goto free_dev;
92
93 if (!rds_iwdev->dma_local_lkey) {
94 rds_iwdev->mr = ib_get_dma_mr(rds_iwdev->pd,
95 IB_ACCESS_REMOTE_READ |
96 IB_ACCESS_REMOTE_WRITE |
97 IB_ACCESS_LOCAL_WRITE);
98 if (IS_ERR(rds_iwdev->mr))
99 goto err_pd;
100 } else
101 rds_iwdev->mr = NULL;
102
103 rds_iwdev->mr_pool = rds_iw_create_mr_pool(rds_iwdev);
104 if (IS_ERR(rds_iwdev->mr_pool)) {
105 rds_iwdev->mr_pool = NULL;
106 goto err_mr;
107 }
108
109 INIT_LIST_HEAD(&rds_iwdev->cm_id_list);
110 INIT_LIST_HEAD(&rds_iwdev->conn_list);
111 list_add_tail(&rds_iwdev->list, &rds_iw_devices);
112
113 ib_set_client_data(device, &rds_iw_client, rds_iwdev);
114
115 goto free_attr;
116
117 err_mr:
118 if (rds_iwdev->mr)
119 ib_dereg_mr(rds_iwdev->mr);
120 err_pd:
121 ib_dealloc_pd(rds_iwdev->pd);
122 free_dev:
123 kfree(rds_iwdev);
124 free_attr:
125 kfree(dev_attr);
126 }
127
rds_iw_remove_one(struct ib_device * device)128 static void rds_iw_remove_one(struct ib_device *device)
129 {
130 struct rds_iw_device *rds_iwdev;
131 struct rds_iw_cm_id *i_cm_id, *next;
132
133 rds_iwdev = ib_get_client_data(device, &rds_iw_client);
134 if (!rds_iwdev)
135 return;
136
137 spin_lock_irq(&rds_iwdev->spinlock);
138 list_for_each_entry_safe(i_cm_id, next, &rds_iwdev->cm_id_list, list) {
139 list_del(&i_cm_id->list);
140 kfree(i_cm_id);
141 }
142 spin_unlock_irq(&rds_iwdev->spinlock);
143
144 rds_iw_destroy_conns(rds_iwdev);
145
146 if (rds_iwdev->mr_pool)
147 rds_iw_destroy_mr_pool(rds_iwdev->mr_pool);
148
149 if (rds_iwdev->mr)
150 ib_dereg_mr(rds_iwdev->mr);
151
152 while (ib_dealloc_pd(rds_iwdev->pd)) {
153 rdsdebug("Failed to dealloc pd %p\n", rds_iwdev->pd);
154 msleep(1);
155 }
156
157 list_del(&rds_iwdev->list);
158 kfree(rds_iwdev);
159 }
160
161 struct ib_client rds_iw_client = {
162 .name = "rds_iw",
163 .add = rds_iw_add_one,
164 .remove = rds_iw_remove_one
165 };
166
rds_iw_conn_info_visitor(struct rds_connection * conn,void * buffer)167 static int rds_iw_conn_info_visitor(struct rds_connection *conn,
168 void *buffer)
169 {
170 struct rds_info_rdma_connection *iinfo = buffer;
171 struct rds_iw_connection *ic;
172
173 /* We will only ever look at IB transports */
174 if (conn->c_trans != &rds_iw_transport)
175 return 0;
176
177 iinfo->src_addr = conn->c_laddr;
178 iinfo->dst_addr = conn->c_faddr;
179
180 memset(&iinfo->src_gid, 0, sizeof(iinfo->src_gid));
181 memset(&iinfo->dst_gid, 0, sizeof(iinfo->dst_gid));
182 if (rds_conn_state(conn) == RDS_CONN_UP) {
183 struct rds_iw_device *rds_iwdev;
184 struct rdma_dev_addr *dev_addr;
185
186 ic = conn->c_transport_data;
187 dev_addr = &ic->i_cm_id->route.addr.dev_addr;
188
189 rdma_addr_get_sgid(dev_addr, (union ib_gid *) &iinfo->src_gid);
190 rdma_addr_get_dgid(dev_addr, (union ib_gid *) &iinfo->dst_gid);
191
192 rds_iwdev = ib_get_client_data(ic->i_cm_id->device, &rds_iw_client);
193 iinfo->max_send_wr = ic->i_send_ring.w_nr;
194 iinfo->max_recv_wr = ic->i_recv_ring.w_nr;
195 iinfo->max_send_sge = rds_iwdev->max_sge;
196 rds_iw_get_mr_info(rds_iwdev, iinfo);
197 }
198 return 1;
199 }
200
rds_iw_ic_info(struct socket * sock,unsigned int len,struct rds_info_iterator * iter,struct rds_info_lengths * lens)201 static void rds_iw_ic_info(struct socket *sock, unsigned int len,
202 struct rds_info_iterator *iter,
203 struct rds_info_lengths *lens)
204 {
205 rds_for_each_conn_info(sock, len, iter, lens,
206 rds_iw_conn_info_visitor,
207 sizeof(struct rds_info_rdma_connection));
208 }
209
210
211 /*
212 * Early RDS/IB was built to only bind to an address if there is an IPoIB
213 * device with that address set.
214 *
215 * If it were me, I'd advocate for something more flexible. Sending and
216 * receiving should be device-agnostic. Transports would try and maintain
217 * connections between peers who have messages queued. Userspace would be
218 * allowed to influence which paths have priority. We could call userspace
219 * asserting this policy "routing".
220 */
rds_iw_laddr_check(__be32 addr)221 static int rds_iw_laddr_check(__be32 addr)
222 {
223 int ret;
224 struct rdma_cm_id *cm_id;
225 struct sockaddr_in sin;
226
227 /* Create a CMA ID and try to bind it. This catches both
228 * IB and iWARP capable NICs.
229 */
230 cm_id = rdma_create_id(NULL, NULL, RDMA_PS_TCP, IB_QPT_RC);
231 if (IS_ERR(cm_id))
232 return PTR_ERR(cm_id);
233
234 memset(&sin, 0, sizeof(sin));
235 sin.sin_family = AF_INET;
236 sin.sin_addr.s_addr = addr;
237
238 /* rdma_bind_addr will only succeed for IB & iWARP devices */
239 ret = rdma_bind_addr(cm_id, (struct sockaddr *)&sin);
240 /* due to this, we will claim to support IB devices unless we
241 check node_type. */
242 if (ret || cm_id->device->node_type != RDMA_NODE_RNIC)
243 ret = -EADDRNOTAVAIL;
244
245 rdsdebug("addr %pI4 ret %d node type %d\n",
246 &addr, ret,
247 cm_id->device ? cm_id->device->node_type : -1);
248
249 rdma_destroy_id(cm_id);
250
251 return ret;
252 }
253
rds_iw_exit(void)254 void rds_iw_exit(void)
255 {
256 rds_info_deregister_func(RDS_INFO_IWARP_CONNECTIONS, rds_iw_ic_info);
257 rds_iw_destroy_nodev_conns();
258 ib_unregister_client(&rds_iw_client);
259 rds_iw_sysctl_exit();
260 rds_iw_recv_exit();
261 rds_trans_unregister(&rds_iw_transport);
262 }
263
264 struct rds_transport rds_iw_transport = {
265 .laddr_check = rds_iw_laddr_check,
266 .xmit_complete = rds_iw_xmit_complete,
267 .xmit = rds_iw_xmit,
268 .xmit_rdma = rds_iw_xmit_rdma,
269 .recv = rds_iw_recv,
270 .conn_alloc = rds_iw_conn_alloc,
271 .conn_free = rds_iw_conn_free,
272 .conn_connect = rds_iw_conn_connect,
273 .conn_shutdown = rds_iw_conn_shutdown,
274 .inc_copy_to_user = rds_iw_inc_copy_to_user,
275 .inc_free = rds_iw_inc_free,
276 .cm_initiate_connect = rds_iw_cm_initiate_connect,
277 .cm_handle_connect = rds_iw_cm_handle_connect,
278 .cm_connect_complete = rds_iw_cm_connect_complete,
279 .stats_info_copy = rds_iw_stats_info_copy,
280 .exit = rds_iw_exit,
281 .get_mr = rds_iw_get_mr,
282 .sync_mr = rds_iw_sync_mr,
283 .free_mr = rds_iw_free_mr,
284 .flush_mrs = rds_iw_flush_mrs,
285 .t_owner = THIS_MODULE,
286 .t_name = "iwarp",
287 .t_type = RDS_TRANS_IWARP,
288 .t_prefer_loopback = 1,
289 };
290
rds_iw_init(void)291 int rds_iw_init(void)
292 {
293 int ret;
294
295 INIT_LIST_HEAD(&rds_iw_devices);
296
297 ret = ib_register_client(&rds_iw_client);
298 if (ret)
299 goto out;
300
301 ret = rds_iw_sysctl_init();
302 if (ret)
303 goto out_ibreg;
304
305 ret = rds_iw_recv_init();
306 if (ret)
307 goto out_sysctl;
308
309 ret = rds_trans_register(&rds_iw_transport);
310 if (ret)
311 goto out_recv;
312
313 rds_info_register_func(RDS_INFO_IWARP_CONNECTIONS, rds_iw_ic_info);
314
315 goto out;
316
317 out_recv:
318 rds_iw_recv_exit();
319 out_sysctl:
320 rds_iw_sysctl_exit();
321 out_ibreg:
322 ib_unregister_client(&rds_iw_client);
323 out:
324 return ret;
325 }
326
327 MODULE_LICENSE("GPL");
328
329