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
42 #include "rds.h"
43 #include "iw.h"
44
45 unsigned int fastreg_pool_size = RDS_FASTREG_POOL_SIZE;
46 unsigned int fastreg_message_size = RDS_FASTREG_SIZE + 1; /* +1 allows for unaligned MRs */
47
48 module_param(fastreg_pool_size, int, 0444);
49 MODULE_PARM_DESC(fastreg_pool_size, " Max number of fastreg MRs per device");
50 module_param(fastreg_message_size, int, 0444);
51 MODULE_PARM_DESC(fastreg_message_size, " Max size of a RDMA transfer (fastreg MRs)");
52
53 struct list_head rds_iw_devices;
54
55 /* NOTE: if also grabbing iwdev lock, grab this first */
56 DEFINE_SPINLOCK(iw_nodev_conns_lock);
57 LIST_HEAD(iw_nodev_conns);
58
rds_iw_add_one(struct ib_device * device)59 static void rds_iw_add_one(struct ib_device *device)
60 {
61 struct rds_iw_device *rds_iwdev;
62 struct ib_device_attr *dev_attr;
63
64 /* Only handle iwarp devices */
65 if (device->node_type != RDMA_NODE_RNIC)
66 return;
67
68 dev_attr = kmalloc(sizeof *dev_attr, GFP_KERNEL);
69 if (!dev_attr)
70 return;
71
72 if (ib_query_device(device, dev_attr)) {
73 rdsdebug("Query device failed for %s\n", device->name);
74 goto free_attr;
75 }
76
77 rds_iwdev = kmalloc(sizeof *rds_iwdev, GFP_KERNEL);
78 if (!rds_iwdev)
79 goto free_attr;
80
81 spin_lock_init(&rds_iwdev->spinlock);
82
83 rds_iwdev->dma_local_lkey = !!(dev_attr->device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY);
84 rds_iwdev->max_wrs = dev_attr->max_qp_wr;
85 rds_iwdev->max_sge = min(dev_attr->max_sge, RDS_IW_MAX_SGE);
86
87 rds_iwdev->dev = device;
88 rds_iwdev->pd = ib_alloc_pd(device);
89 if (IS_ERR(rds_iwdev->pd))
90 goto free_dev;
91
92 if (!rds_iwdev->dma_local_lkey) {
93 rds_iwdev->mr = ib_get_dma_mr(rds_iwdev->pd,
94 IB_ACCESS_REMOTE_READ |
95 IB_ACCESS_REMOTE_WRITE |
96 IB_ACCESS_LOCAL_WRITE);
97 if (IS_ERR(rds_iwdev->mr))
98 goto err_pd;
99 } else
100 rds_iwdev->mr = NULL;
101
102 rds_iwdev->mr_pool = rds_iw_create_mr_pool(rds_iwdev);
103 if (IS_ERR(rds_iwdev->mr_pool)) {
104 rds_iwdev->mr_pool = NULL;
105 goto err_mr;
106 }
107
108 INIT_LIST_HEAD(&rds_iwdev->cm_id_list);
109 INIT_LIST_HEAD(&rds_iwdev->conn_list);
110 list_add_tail(&rds_iwdev->list, &rds_iw_devices);
111
112 ib_set_client_data(device, &rds_iw_client, rds_iwdev);
113
114 goto free_attr;
115
116 err_mr:
117 if (rds_iwdev->mr)
118 ib_dereg_mr(rds_iwdev->mr);
119 err_pd:
120 ib_dealloc_pd(rds_iwdev->pd);
121 free_dev:
122 kfree(rds_iwdev);
123 free_attr:
124 kfree(dev_attr);
125 }
126
rds_iw_remove_one(struct ib_device * device)127 static void rds_iw_remove_one(struct ib_device *device)
128 {
129 struct rds_iw_device *rds_iwdev;
130 struct rds_iw_cm_id *i_cm_id, *next;
131
132 rds_iwdev = ib_get_client_data(device, &rds_iw_client);
133 if (!rds_iwdev)
134 return;
135
136 spin_lock_irq(&rds_iwdev->spinlock);
137 list_for_each_entry_safe(i_cm_id, next, &rds_iwdev->cm_id_list, list) {
138 list_del(&i_cm_id->list);
139 kfree(i_cm_id);
140 }
141 spin_unlock_irq(&rds_iwdev->spinlock);
142
143 rds_iw_destroy_conns(rds_iwdev);
144
145 if (rds_iwdev->mr_pool)
146 rds_iw_destroy_mr_pool(rds_iwdev->mr_pool);
147
148 if (rds_iwdev->mr)
149 ib_dereg_mr(rds_iwdev->mr);
150
151 while (ib_dealloc_pd(rds_iwdev->pd)) {
152 rdsdebug("Failed to dealloc pd %p\n", rds_iwdev->pd);
153 msleep(1);
154 }
155
156 list_del(&rds_iwdev->list);
157 kfree(rds_iwdev);
158 }
159
160 struct ib_client rds_iw_client = {
161 .name = "rds_iw",
162 .add = rds_iw_add_one,
163 .remove = rds_iw_remove_one
164 };
165
rds_iw_conn_info_visitor(struct rds_connection * conn,void * buffer)166 static int rds_iw_conn_info_visitor(struct rds_connection *conn,
167 void *buffer)
168 {
169 struct rds_info_rdma_connection *iinfo = buffer;
170 struct rds_iw_connection *ic;
171
172 /* We will only ever look at IB transports */
173 if (conn->c_trans != &rds_iw_transport)
174 return 0;
175
176 iinfo->src_addr = conn->c_laddr;
177 iinfo->dst_addr = conn->c_faddr;
178
179 memset(&iinfo->src_gid, 0, sizeof(iinfo->src_gid));
180 memset(&iinfo->dst_gid, 0, sizeof(iinfo->dst_gid));
181 if (rds_conn_state(conn) == RDS_CONN_UP) {
182 struct rds_iw_device *rds_iwdev;
183 struct rdma_dev_addr *dev_addr;
184
185 ic = conn->c_transport_data;
186 dev_addr = &ic->i_cm_id->route.addr.dev_addr;
187
188 rdma_addr_get_sgid(dev_addr, (union ib_gid *) &iinfo->src_gid);
189 rdma_addr_get_dgid(dev_addr, (union ib_gid *) &iinfo->dst_gid);
190
191 rds_iwdev = ib_get_client_data(ic->i_cm_id->device, &rds_iw_client);
192 iinfo->max_send_wr = ic->i_send_ring.w_nr;
193 iinfo->max_recv_wr = ic->i_recv_ring.w_nr;
194 iinfo->max_send_sge = rds_iwdev->max_sge;
195 rds_iw_get_mr_info(rds_iwdev, iinfo);
196 }
197 return 1;
198 }
199
rds_iw_ic_info(struct socket * sock,unsigned int len,struct rds_info_iterator * iter,struct rds_info_lengths * lens)200 static void rds_iw_ic_info(struct socket *sock, unsigned int len,
201 struct rds_info_iterator *iter,
202 struct rds_info_lengths *lens)
203 {
204 rds_for_each_conn_info(sock, len, iter, lens,
205 rds_iw_conn_info_visitor,
206 sizeof(struct rds_info_rdma_connection));
207 }
208
209
210 /*
211 * Early RDS/IB was built to only bind to an address if there is an IPoIB
212 * device with that address set.
213 *
214 * If it were me, I'd advocate for something more flexible. Sending and
215 * receiving should be device-agnostic. Transports would try and maintain
216 * connections between peers who have messages queued. Userspace would be
217 * allowed to influence which paths have priority. We could call userspace
218 * asserting this policy "routing".
219 */
rds_iw_laddr_check(__be32 addr)220 static int rds_iw_laddr_check(__be32 addr)
221 {
222 int ret;
223 struct rdma_cm_id *cm_id;
224 struct sockaddr_in sin;
225
226 /* Create a CMA ID and try to bind it. This catches both
227 * IB and iWARP capable NICs.
228 */
229 cm_id = rdma_create_id(NULL, NULL, RDMA_PS_TCP);
230 if (IS_ERR(cm_id))
231 return PTR_ERR(cm_id);
232
233 memset(&sin, 0, sizeof(sin));
234 sin.sin_family = AF_INET;
235 sin.sin_addr.s_addr = addr;
236
237 /* rdma_bind_addr will only succeed for IB & iWARP devices */
238 ret = rdma_bind_addr(cm_id, (struct sockaddr *)&sin);
239 /* due to this, we will claim to support IB devices unless we
240 check node_type. */
241 if (ret || cm_id->device->node_type != RDMA_NODE_RNIC)
242 ret = -EADDRNOTAVAIL;
243
244 rdsdebug("addr %pI4 ret %d node type %d\n",
245 &addr, ret,
246 cm_id->device ? cm_id->device->node_type : -1);
247
248 rdma_destroy_id(cm_id);
249
250 return ret;
251 }
252
rds_iw_exit(void)253 void rds_iw_exit(void)
254 {
255 rds_info_deregister_func(RDS_INFO_IWARP_CONNECTIONS, rds_iw_ic_info);
256 rds_iw_destroy_nodev_conns();
257 ib_unregister_client(&rds_iw_client);
258 rds_iw_sysctl_exit();
259 rds_iw_recv_exit();
260 rds_trans_unregister(&rds_iw_transport);
261 }
262
263 struct rds_transport rds_iw_transport = {
264 .laddr_check = rds_iw_laddr_check,
265 .xmit_complete = rds_iw_xmit_complete,
266 .xmit = rds_iw_xmit,
267 .xmit_rdma = rds_iw_xmit_rdma,
268 .recv = rds_iw_recv,
269 .conn_alloc = rds_iw_conn_alloc,
270 .conn_free = rds_iw_conn_free,
271 .conn_connect = rds_iw_conn_connect,
272 .conn_shutdown = rds_iw_conn_shutdown,
273 .inc_copy_to_user = rds_iw_inc_copy_to_user,
274 .inc_free = rds_iw_inc_free,
275 .cm_initiate_connect = rds_iw_cm_initiate_connect,
276 .cm_handle_connect = rds_iw_cm_handle_connect,
277 .cm_connect_complete = rds_iw_cm_connect_complete,
278 .stats_info_copy = rds_iw_stats_info_copy,
279 .exit = rds_iw_exit,
280 .get_mr = rds_iw_get_mr,
281 .sync_mr = rds_iw_sync_mr,
282 .free_mr = rds_iw_free_mr,
283 .flush_mrs = rds_iw_flush_mrs,
284 .t_owner = THIS_MODULE,
285 .t_name = "iwarp",
286 .t_type = RDS_TRANS_IWARP,
287 .t_prefer_loopback = 1,
288 };
289
rds_iw_init(void)290 int rds_iw_init(void)
291 {
292 int ret;
293
294 INIT_LIST_HEAD(&rds_iw_devices);
295
296 ret = ib_register_client(&rds_iw_client);
297 if (ret)
298 goto out;
299
300 ret = rds_iw_sysctl_init();
301 if (ret)
302 goto out_ibreg;
303
304 ret = rds_iw_recv_init();
305 if (ret)
306 goto out_sysctl;
307
308 ret = rds_trans_register(&rds_iw_transport);
309 if (ret)
310 goto out_recv;
311
312 rds_info_register_func(RDS_INFO_IWARP_CONNECTIONS, rds_iw_ic_info);
313
314 goto out;
315
316 out_recv:
317 rds_iw_recv_exit();
318 out_sysctl:
319 rds_iw_sysctl_exit();
320 out_ibreg:
321 ib_unregister_client(&rds_iw_client);
322 out:
323 return ret;
324 }
325
326 MODULE_LICENSE("GPL");
327
328