1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /*
3 * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
4 * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved.
5 */
6
7 #include <linux/skbuff.h>
8
9 #include "rxe.h"
10 #include "rxe_loc.h"
11
12 /* check that QP matches packet opcode type and is in a valid state */
check_type_state(struct rxe_dev * rxe,struct rxe_pkt_info * pkt,struct rxe_qp * qp)13 static int check_type_state(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
14 struct rxe_qp *qp)
15 {
16 unsigned int pkt_type;
17
18 if (unlikely(!qp->valid))
19 return -EINVAL;
20
21 pkt_type = pkt->opcode & 0xe0;
22
23 switch (qp_type(qp)) {
24 case IB_QPT_RC:
25 if (unlikely(pkt_type != IB_OPCODE_RC))
26 return -EINVAL;
27 break;
28 case IB_QPT_UC:
29 if (unlikely(pkt_type != IB_OPCODE_UC))
30 return -EINVAL;
31 break;
32 case IB_QPT_UD:
33 case IB_QPT_GSI:
34 if (unlikely(pkt_type != IB_OPCODE_UD))
35 return -EINVAL;
36 break;
37 default:
38 return -EINVAL;
39 }
40
41 if (pkt->mask & RXE_REQ_MASK) {
42 if (unlikely(qp->resp.state != QP_STATE_READY))
43 return -EINVAL;
44 } else if (unlikely(qp->req.state < QP_STATE_READY ||
45 qp->req.state > QP_STATE_DRAINED))
46 return -EINVAL;
47
48 return 0;
49 }
50
set_bad_pkey_cntr(struct rxe_port * port)51 static void set_bad_pkey_cntr(struct rxe_port *port)
52 {
53 spin_lock_bh(&port->port_lock);
54 port->attr.bad_pkey_cntr = min((u32)0xffff,
55 port->attr.bad_pkey_cntr + 1);
56 spin_unlock_bh(&port->port_lock);
57 }
58
set_qkey_viol_cntr(struct rxe_port * port)59 static void set_qkey_viol_cntr(struct rxe_port *port)
60 {
61 spin_lock_bh(&port->port_lock);
62 port->attr.qkey_viol_cntr = min((u32)0xffff,
63 port->attr.qkey_viol_cntr + 1);
64 spin_unlock_bh(&port->port_lock);
65 }
66
check_keys(struct rxe_dev * rxe,struct rxe_pkt_info * pkt,u32 qpn,struct rxe_qp * qp)67 static int check_keys(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
68 u32 qpn, struct rxe_qp *qp)
69 {
70 struct rxe_port *port = &rxe->port;
71 u16 pkey = bth_pkey(pkt);
72
73 pkt->pkey_index = 0;
74
75 if (!pkey_match(pkey, IB_DEFAULT_PKEY_FULL)) {
76 set_bad_pkey_cntr(port);
77 return -EINVAL;
78 }
79
80 if (qp_type(qp) == IB_QPT_UD || qp_type(qp) == IB_QPT_GSI) {
81 u32 qkey = (qpn == 1) ? GSI_QKEY : qp->attr.qkey;
82
83 if (unlikely(deth_qkey(pkt) != qkey)) {
84 set_qkey_viol_cntr(port);
85 return -EINVAL;
86 }
87 }
88
89 return 0;
90 }
91
check_addr(struct rxe_dev * rxe,struct rxe_pkt_info * pkt,struct rxe_qp * qp)92 static int check_addr(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
93 struct rxe_qp *qp)
94 {
95 struct sk_buff *skb = PKT_TO_SKB(pkt);
96
97 if (qp_type(qp) != IB_QPT_RC && qp_type(qp) != IB_QPT_UC)
98 return 0;
99
100 if (unlikely(pkt->port_num != qp->attr.port_num))
101 return -EINVAL;
102
103 if (skb->protocol == htons(ETH_P_IP)) {
104 struct in_addr *saddr =
105 &qp->pri_av.sgid_addr._sockaddr_in.sin_addr;
106 struct in_addr *daddr =
107 &qp->pri_av.dgid_addr._sockaddr_in.sin_addr;
108
109 if ((ip_hdr(skb)->daddr != saddr->s_addr) ||
110 (ip_hdr(skb)->saddr != daddr->s_addr))
111 return -EINVAL;
112
113 } else if (skb->protocol == htons(ETH_P_IPV6)) {
114 struct in6_addr *saddr =
115 &qp->pri_av.sgid_addr._sockaddr_in6.sin6_addr;
116 struct in6_addr *daddr =
117 &qp->pri_av.dgid_addr._sockaddr_in6.sin6_addr;
118
119 if (memcmp(&ipv6_hdr(skb)->daddr, saddr, sizeof(*saddr)) ||
120 memcmp(&ipv6_hdr(skb)->saddr, daddr, sizeof(*daddr)))
121 return -EINVAL;
122 }
123
124 return 0;
125 }
126
hdr_check(struct rxe_pkt_info * pkt)127 static int hdr_check(struct rxe_pkt_info *pkt)
128 {
129 struct rxe_dev *rxe = pkt->rxe;
130 struct rxe_port *port = &rxe->port;
131 struct rxe_qp *qp = NULL;
132 u32 qpn = bth_qpn(pkt);
133 int index;
134 int err;
135
136 if (unlikely(bth_tver(pkt) != BTH_TVER))
137 goto err1;
138
139 if (unlikely(qpn == 0))
140 goto err1;
141
142 if (qpn != IB_MULTICAST_QPN) {
143 index = (qpn == 1) ? port->qp_gsi_index : qpn;
144
145 qp = rxe_pool_get_index(&rxe->qp_pool, index);
146 if (unlikely(!qp))
147 goto err1;
148
149 err = check_type_state(rxe, pkt, qp);
150 if (unlikely(err))
151 goto err2;
152
153 err = check_addr(rxe, pkt, qp);
154 if (unlikely(err))
155 goto err2;
156
157 err = check_keys(rxe, pkt, qpn, qp);
158 if (unlikely(err))
159 goto err2;
160 } else {
161 if (unlikely((pkt->mask & RXE_GRH_MASK) == 0))
162 goto err1;
163 }
164
165 pkt->qp = qp;
166 return 0;
167
168 err2:
169 rxe_put(qp);
170 err1:
171 return -EINVAL;
172 }
173
rxe_rcv_pkt(struct rxe_pkt_info * pkt,struct sk_buff * skb)174 static inline void rxe_rcv_pkt(struct rxe_pkt_info *pkt, struct sk_buff *skb)
175 {
176 if (pkt->mask & RXE_REQ_MASK)
177 rxe_resp_queue_pkt(pkt->qp, skb);
178 else
179 rxe_comp_queue_pkt(pkt->qp, skb);
180 }
181
rxe_rcv_mcast_pkt(struct rxe_dev * rxe,struct sk_buff * skb)182 static void rxe_rcv_mcast_pkt(struct rxe_dev *rxe, struct sk_buff *skb)
183 {
184 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
185 struct rxe_mcg *mcg;
186 struct rxe_mca *mca;
187 struct rxe_qp *qp;
188 union ib_gid dgid;
189 int err;
190
191 if (skb->protocol == htons(ETH_P_IP))
192 ipv6_addr_set_v4mapped(ip_hdr(skb)->daddr,
193 (struct in6_addr *)&dgid);
194 else if (skb->protocol == htons(ETH_P_IPV6))
195 memcpy(&dgid, &ipv6_hdr(skb)->daddr, sizeof(dgid));
196
197 /* lookup mcast group corresponding to mgid, takes a ref */
198 mcg = rxe_lookup_mcg(rxe, &dgid);
199 if (!mcg)
200 goto drop; /* mcast group not registered */
201
202 spin_lock_bh(&rxe->mcg_lock);
203
204 /* this is unreliable datagram service so we let
205 * failures to deliver a multicast packet to a
206 * single QP happen and just move on and try
207 * the rest of them on the list
208 */
209 list_for_each_entry(mca, &mcg->qp_list, qp_list) {
210 qp = mca->qp;
211
212 /* validate qp for incoming packet */
213 err = check_type_state(rxe, pkt, qp);
214 if (err)
215 continue;
216
217 err = check_keys(rxe, pkt, bth_qpn(pkt), qp);
218 if (err)
219 continue;
220
221 /* for all but the last QP create a new clone of the
222 * skb and pass to the QP. Pass the original skb to
223 * the last QP in the list.
224 */
225 if (mca->qp_list.next != &mcg->qp_list) {
226 struct sk_buff *cskb;
227 struct rxe_pkt_info *cpkt;
228
229 cskb = skb_clone(skb, GFP_ATOMIC);
230 if (unlikely(!cskb))
231 continue;
232
233 if (WARN_ON(!ib_device_try_get(&rxe->ib_dev))) {
234 kfree_skb(cskb);
235 break;
236 }
237
238 cpkt = SKB_TO_PKT(cskb);
239 cpkt->qp = qp;
240 rxe_get(qp);
241 rxe_rcv_pkt(cpkt, cskb);
242 } else {
243 pkt->qp = qp;
244 rxe_get(qp);
245 rxe_rcv_pkt(pkt, skb);
246 skb = NULL; /* mark consumed */
247 }
248 }
249
250 spin_unlock_bh(&rxe->mcg_lock);
251
252 kref_put(&mcg->ref_cnt, rxe_cleanup_mcg);
253
254 if (likely(!skb))
255 return;
256
257 /* This only occurs if one of the checks fails on the last
258 * QP in the list above
259 */
260
261 drop:
262 kfree_skb(skb);
263 ib_device_put(&rxe->ib_dev);
264 }
265
266 /**
267 * rxe_chk_dgid - validate destination IP address
268 * @rxe: rxe device that received packet
269 * @skb: the received packet buffer
270 *
271 * Accept any loopback packets
272 * Extract IP address from packet and
273 * Accept if multicast packet
274 * Accept if matches an SGID table entry
275 */
rxe_chk_dgid(struct rxe_dev * rxe,struct sk_buff * skb)276 static int rxe_chk_dgid(struct rxe_dev *rxe, struct sk_buff *skb)
277 {
278 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
279 const struct ib_gid_attr *gid_attr;
280 union ib_gid dgid;
281 union ib_gid *pdgid;
282
283 if (pkt->mask & RXE_LOOPBACK_MASK)
284 return 0;
285
286 if (skb->protocol == htons(ETH_P_IP)) {
287 ipv6_addr_set_v4mapped(ip_hdr(skb)->daddr,
288 (struct in6_addr *)&dgid);
289 pdgid = &dgid;
290 } else {
291 pdgid = (union ib_gid *)&ipv6_hdr(skb)->daddr;
292 }
293
294 if (rdma_is_multicast_addr((struct in6_addr *)pdgid))
295 return 0;
296
297 gid_attr = rdma_find_gid_by_port(&rxe->ib_dev, pdgid,
298 IB_GID_TYPE_ROCE_UDP_ENCAP,
299 1, skb->dev);
300 if (IS_ERR(gid_attr))
301 return PTR_ERR(gid_attr);
302
303 rdma_put_gid_attr(gid_attr);
304 return 0;
305 }
306
307 /* rxe_rcv is called from the interface driver */
rxe_rcv(struct sk_buff * skb)308 void rxe_rcv(struct sk_buff *skb)
309 {
310 int err;
311 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
312 struct rxe_dev *rxe = pkt->rxe;
313
314 if (unlikely(skb->len < RXE_BTH_BYTES))
315 goto drop;
316
317 if (rxe_chk_dgid(rxe, skb) < 0)
318 goto drop;
319
320 pkt->opcode = bth_opcode(pkt);
321 pkt->psn = bth_psn(pkt);
322 pkt->qp = NULL;
323 pkt->mask |= rxe_opcode[pkt->opcode].mask;
324
325 if (unlikely(skb->len < header_size(pkt)))
326 goto drop;
327
328 err = hdr_check(pkt);
329 if (unlikely(err))
330 goto drop;
331
332 err = rxe_icrc_check(skb, pkt);
333 if (unlikely(err))
334 goto drop;
335
336 rxe_counter_inc(rxe, RXE_CNT_RCVD_PKTS);
337
338 if (unlikely(bth_qpn(pkt) == IB_MULTICAST_QPN))
339 rxe_rcv_mcast_pkt(rxe, skb);
340 else
341 rxe_rcv_pkt(pkt, skb);
342
343 return;
344
345 drop:
346 if (pkt->qp)
347 rxe_put(pkt->qp);
348
349 kfree_skb(skb);
350 ib_device_put(&rxe->ib_dev);
351 }
352