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 unsigned long flags;
18
19 if (unlikely(!qp->valid))
20 return -EINVAL;
21
22 pkt_type = pkt->opcode & 0xe0;
23
24 switch (qp_type(qp)) {
25 case IB_QPT_RC:
26 if (unlikely(pkt_type != IB_OPCODE_RC))
27 return -EINVAL;
28 break;
29 case IB_QPT_UC:
30 if (unlikely(pkt_type != IB_OPCODE_UC))
31 return -EINVAL;
32 break;
33 case IB_QPT_UD:
34 case IB_QPT_GSI:
35 if (unlikely(pkt_type != IB_OPCODE_UD))
36 return -EINVAL;
37 break;
38 default:
39 return -EINVAL;
40 }
41
42 spin_lock_irqsave(&qp->state_lock, flags);
43 if (pkt->mask & RXE_REQ_MASK) {
44 if (unlikely(qp_state(qp) < IB_QPS_RTR)) {
45 spin_unlock_irqrestore(&qp->state_lock, flags);
46 return -EINVAL;
47 }
48 } else {
49 if (unlikely(qp_state(qp) < IB_QPS_RTS)) {
50 spin_unlock_irqrestore(&qp->state_lock, flags);
51 return -EINVAL;
52 }
53 }
54 spin_unlock_irqrestore(&qp->state_lock, flags);
55
56 return 0;
57 }
58
set_bad_pkey_cntr(struct rxe_port * port)59 static void set_bad_pkey_cntr(struct rxe_port *port)
60 {
61 spin_lock_bh(&port->port_lock);
62 port->attr.bad_pkey_cntr = min((u32)0xffff,
63 port->attr.bad_pkey_cntr + 1);
64 spin_unlock_bh(&port->port_lock);
65 }
66
set_qkey_viol_cntr(struct rxe_port * port)67 static void set_qkey_viol_cntr(struct rxe_port *port)
68 {
69 spin_lock_bh(&port->port_lock);
70 port->attr.qkey_viol_cntr = min((u32)0xffff,
71 port->attr.qkey_viol_cntr + 1);
72 spin_unlock_bh(&port->port_lock);
73 }
74
check_keys(struct rxe_dev * rxe,struct rxe_pkt_info * pkt,u32 qpn,struct rxe_qp * qp)75 static int check_keys(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
76 u32 qpn, struct rxe_qp *qp)
77 {
78 struct rxe_port *port = &rxe->port;
79 u16 pkey = bth_pkey(pkt);
80
81 pkt->pkey_index = 0;
82
83 if (!pkey_match(pkey, IB_DEFAULT_PKEY_FULL)) {
84 set_bad_pkey_cntr(port);
85 return -EINVAL;
86 }
87
88 if (qp_type(qp) == IB_QPT_UD || qp_type(qp) == IB_QPT_GSI) {
89 u32 qkey = (qpn == 1) ? GSI_QKEY : qp->attr.qkey;
90
91 if (unlikely(deth_qkey(pkt) != qkey)) {
92 set_qkey_viol_cntr(port);
93 return -EINVAL;
94 }
95 }
96
97 return 0;
98 }
99
check_addr(struct rxe_dev * rxe,struct rxe_pkt_info * pkt,struct rxe_qp * qp)100 static int check_addr(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
101 struct rxe_qp *qp)
102 {
103 struct sk_buff *skb = PKT_TO_SKB(pkt);
104
105 if (qp_type(qp) != IB_QPT_RC && qp_type(qp) != IB_QPT_UC)
106 return 0;
107
108 if (unlikely(pkt->port_num != qp->attr.port_num))
109 return -EINVAL;
110
111 if (skb->protocol == htons(ETH_P_IP)) {
112 struct in_addr *saddr =
113 &qp->pri_av.sgid_addr._sockaddr_in.sin_addr;
114 struct in_addr *daddr =
115 &qp->pri_av.dgid_addr._sockaddr_in.sin_addr;
116
117 if ((ip_hdr(skb)->daddr != saddr->s_addr) ||
118 (ip_hdr(skb)->saddr != daddr->s_addr))
119 return -EINVAL;
120
121 } else if (skb->protocol == htons(ETH_P_IPV6)) {
122 struct in6_addr *saddr =
123 &qp->pri_av.sgid_addr._sockaddr_in6.sin6_addr;
124 struct in6_addr *daddr =
125 &qp->pri_av.dgid_addr._sockaddr_in6.sin6_addr;
126
127 if (memcmp(&ipv6_hdr(skb)->daddr, saddr, sizeof(*saddr)) ||
128 memcmp(&ipv6_hdr(skb)->saddr, daddr, sizeof(*daddr)))
129 return -EINVAL;
130 }
131
132 return 0;
133 }
134
hdr_check(struct rxe_pkt_info * pkt)135 static int hdr_check(struct rxe_pkt_info *pkt)
136 {
137 struct rxe_dev *rxe = pkt->rxe;
138 struct rxe_port *port = &rxe->port;
139 struct rxe_qp *qp = NULL;
140 u32 qpn = bth_qpn(pkt);
141 int index;
142 int err;
143
144 if (unlikely(bth_tver(pkt) != BTH_TVER))
145 goto err1;
146
147 if (unlikely(qpn == 0))
148 goto err1;
149
150 if (qpn != IB_MULTICAST_QPN) {
151 index = (qpn == 1) ? port->qp_gsi_index : qpn;
152
153 qp = rxe_pool_get_index(&rxe->qp_pool, index);
154 if (unlikely(!qp))
155 goto err1;
156
157 err = check_type_state(rxe, pkt, qp);
158 if (unlikely(err))
159 goto err2;
160
161 err = check_addr(rxe, pkt, qp);
162 if (unlikely(err))
163 goto err2;
164
165 err = check_keys(rxe, pkt, qpn, qp);
166 if (unlikely(err))
167 goto err2;
168 } else {
169 if (unlikely((pkt->mask & RXE_GRH_MASK) == 0))
170 goto err1;
171 }
172
173 pkt->qp = qp;
174 return 0;
175
176 err2:
177 rxe_put(qp);
178 err1:
179 return -EINVAL;
180 }
181
rxe_rcv_pkt(struct rxe_pkt_info * pkt,struct sk_buff * skb)182 static inline void rxe_rcv_pkt(struct rxe_pkt_info *pkt, struct sk_buff *skb)
183 {
184 if (pkt->mask & RXE_REQ_MASK)
185 rxe_resp_queue_pkt(pkt->qp, skb);
186 else
187 rxe_comp_queue_pkt(pkt->qp, skb);
188 }
189
rxe_rcv_mcast_pkt(struct rxe_dev * rxe,struct sk_buff * skb)190 static void rxe_rcv_mcast_pkt(struct rxe_dev *rxe, struct sk_buff *skb)
191 {
192 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
193 struct rxe_mcg *mcg;
194 struct rxe_mca *mca;
195 struct rxe_qp *qp;
196 union ib_gid dgid;
197 int err;
198
199 if (skb->protocol == htons(ETH_P_IP))
200 ipv6_addr_set_v4mapped(ip_hdr(skb)->daddr,
201 (struct in6_addr *)&dgid);
202 else if (skb->protocol == htons(ETH_P_IPV6))
203 memcpy(&dgid, &ipv6_hdr(skb)->daddr, sizeof(dgid));
204
205 /* lookup mcast group corresponding to mgid, takes a ref */
206 mcg = rxe_lookup_mcg(rxe, &dgid);
207 if (!mcg)
208 goto drop; /* mcast group not registered */
209
210 spin_lock_bh(&rxe->mcg_lock);
211
212 /* this is unreliable datagram service so we let
213 * failures to deliver a multicast packet to a
214 * single QP happen and just move on and try
215 * the rest of them on the list
216 */
217 list_for_each_entry(mca, &mcg->qp_list, qp_list) {
218 qp = mca->qp;
219
220 /* validate qp for incoming packet */
221 err = check_type_state(rxe, pkt, qp);
222 if (err)
223 continue;
224
225 err = check_keys(rxe, pkt, bth_qpn(pkt), qp);
226 if (err)
227 continue;
228
229 /* for all but the last QP create a new clone of the
230 * skb and pass to the QP. Pass the original skb to
231 * the last QP in the list.
232 */
233 if (mca->qp_list.next != &mcg->qp_list) {
234 struct sk_buff *cskb;
235 struct rxe_pkt_info *cpkt;
236
237 cskb = skb_clone(skb, GFP_ATOMIC);
238 if (unlikely(!cskb))
239 continue;
240
241 if (WARN_ON(!ib_device_try_get(&rxe->ib_dev))) {
242 kfree_skb(cskb);
243 break;
244 }
245
246 cpkt = SKB_TO_PKT(cskb);
247 cpkt->qp = qp;
248 rxe_get(qp);
249 rxe_rcv_pkt(cpkt, cskb);
250 } else {
251 pkt->qp = qp;
252 rxe_get(qp);
253 rxe_rcv_pkt(pkt, skb);
254 skb = NULL; /* mark consumed */
255 }
256 }
257
258 spin_unlock_bh(&rxe->mcg_lock);
259
260 kref_put(&mcg->ref_cnt, rxe_cleanup_mcg);
261
262 if (likely(!skb))
263 return;
264
265 /* This only occurs if one of the checks fails on the last
266 * QP in the list above
267 */
268
269 drop:
270 kfree_skb(skb);
271 ib_device_put(&rxe->ib_dev);
272 }
273
274 /**
275 * rxe_chk_dgid - validate destination IP address
276 * @rxe: rxe device that received packet
277 * @skb: the received packet buffer
278 *
279 * Accept any loopback packets
280 * Extract IP address from packet and
281 * Accept if multicast packet
282 * Accept if matches an SGID table entry
283 */
rxe_chk_dgid(struct rxe_dev * rxe,struct sk_buff * skb)284 static int rxe_chk_dgid(struct rxe_dev *rxe, struct sk_buff *skb)
285 {
286 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
287 const struct ib_gid_attr *gid_attr;
288 union ib_gid dgid;
289 union ib_gid *pdgid;
290
291 if (pkt->mask & RXE_LOOPBACK_MASK)
292 return 0;
293
294 if (skb->protocol == htons(ETH_P_IP)) {
295 ipv6_addr_set_v4mapped(ip_hdr(skb)->daddr,
296 (struct in6_addr *)&dgid);
297 pdgid = &dgid;
298 } else {
299 pdgid = (union ib_gid *)&ipv6_hdr(skb)->daddr;
300 }
301
302 if (rdma_is_multicast_addr((struct in6_addr *)pdgid))
303 return 0;
304
305 gid_attr = rdma_find_gid_by_port(&rxe->ib_dev, pdgid,
306 IB_GID_TYPE_ROCE_UDP_ENCAP,
307 1, skb->dev);
308 if (IS_ERR(gid_attr))
309 return PTR_ERR(gid_attr);
310
311 rdma_put_gid_attr(gid_attr);
312 return 0;
313 }
314
315 /* rxe_rcv is called from the interface driver */
rxe_rcv(struct sk_buff * skb)316 void rxe_rcv(struct sk_buff *skb)
317 {
318 int err;
319 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
320 struct rxe_dev *rxe = pkt->rxe;
321
322 if (unlikely(skb->len < RXE_BTH_BYTES))
323 goto drop;
324
325 if (rxe_chk_dgid(rxe, skb) < 0)
326 goto drop;
327
328 pkt->opcode = bth_opcode(pkt);
329 pkt->psn = bth_psn(pkt);
330 pkt->qp = NULL;
331 pkt->mask |= rxe_opcode[pkt->opcode].mask;
332
333 if (unlikely(skb->len < header_size(pkt)))
334 goto drop;
335
336 err = hdr_check(pkt);
337 if (unlikely(err))
338 goto drop;
339
340 err = rxe_icrc_check(skb, pkt);
341 if (unlikely(err))
342 goto drop;
343
344 rxe_counter_inc(rxe, RXE_CNT_RCVD_PKTS);
345
346 if (unlikely(bth_qpn(pkt) == IB_MULTICAST_QPN))
347 rxe_rcv_mcast_pkt(rxe, skb);
348 else
349 rxe_rcv_pkt(pkt, skb);
350
351 return;
352
353 drop:
354 if (pkt->qp)
355 rxe_put(pkt->qp);
356
357 kfree_skb(skb);
358 ib_device_put(&rxe->ib_dev);
359 }
360