• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
3  * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *	- Redistributions of source code must retain the above
16  *	  copyright notice, this list of conditions and the following
17  *	  disclaimer.
18  *
19  *	- Redistributions in binary form must reproduce the above
20  *	  copyright notice, this list of conditions and the following
21  *	  disclaimer in the documentation and/or other materials
22  *	  provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 #include <linux/skbuff.h>
35 
36 #include "rxe.h"
37 #include "rxe_loc.h"
38 
39 /* 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)40 static int check_type_state(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
41 			    struct rxe_qp *qp)
42 {
43 	unsigned int pkt_type;
44 
45 	if (unlikely(!qp->valid))
46 		goto err1;
47 
48 	pkt_type = pkt->opcode & 0xe0;
49 
50 	switch (qp_type(qp)) {
51 	case IB_QPT_RC:
52 		if (unlikely(pkt_type != IB_OPCODE_RC)) {
53 			pr_warn_ratelimited("bad qp type\n");
54 			goto err1;
55 		}
56 		break;
57 	case IB_QPT_UC:
58 		if (unlikely(pkt_type != IB_OPCODE_UC)) {
59 			pr_warn_ratelimited("bad qp type\n");
60 			goto err1;
61 		}
62 		break;
63 	case IB_QPT_UD:
64 	case IB_QPT_SMI:
65 	case IB_QPT_GSI:
66 		if (unlikely(pkt_type != IB_OPCODE_UD)) {
67 			pr_warn_ratelimited("bad qp type\n");
68 			goto err1;
69 		}
70 		break;
71 	default:
72 		pr_warn_ratelimited("unsupported qp type\n");
73 		goto err1;
74 	}
75 
76 	if (pkt->mask & RXE_REQ_MASK) {
77 		if (unlikely(qp->resp.state != QP_STATE_READY))
78 			goto err1;
79 	} else if (unlikely(qp->req.state < QP_STATE_READY ||
80 				qp->req.state > QP_STATE_DRAINED)) {
81 		goto err1;
82 	}
83 
84 	return 0;
85 
86 err1:
87 	return -EINVAL;
88 }
89 
set_bad_pkey_cntr(struct rxe_port * port)90 static void set_bad_pkey_cntr(struct rxe_port *port)
91 {
92 	spin_lock_bh(&port->port_lock);
93 	port->attr.bad_pkey_cntr = min((u32)0xffff,
94 				       port->attr.bad_pkey_cntr + 1);
95 	spin_unlock_bh(&port->port_lock);
96 }
97 
set_qkey_viol_cntr(struct rxe_port * port)98 static void set_qkey_viol_cntr(struct rxe_port *port)
99 {
100 	spin_lock_bh(&port->port_lock);
101 	port->attr.qkey_viol_cntr = min((u32)0xffff,
102 					port->attr.qkey_viol_cntr + 1);
103 	spin_unlock_bh(&port->port_lock);
104 }
105 
check_keys(struct rxe_dev * rxe,struct rxe_pkt_info * pkt,u32 qpn,struct rxe_qp * qp)106 static int check_keys(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
107 		      u32 qpn, struct rxe_qp *qp)
108 {
109 	int i;
110 	int found_pkey = 0;
111 	struct rxe_port *port = &rxe->port;
112 	u16 pkey = bth_pkey(pkt);
113 
114 	pkt->pkey_index = 0;
115 
116 	if (qpn == 1) {
117 		for (i = 0; i < port->attr.pkey_tbl_len; i++) {
118 			if (pkey_match(pkey, port->pkey_tbl[i])) {
119 				pkt->pkey_index = i;
120 				found_pkey = 1;
121 				break;
122 			}
123 		}
124 
125 		if (!found_pkey) {
126 			pr_warn_ratelimited("bad pkey = 0x%x\n", pkey);
127 			set_bad_pkey_cntr(port);
128 			goto err1;
129 		}
130 	} else {
131 		if (unlikely(!pkey_match(pkey,
132 					 port->pkey_tbl[qp->attr.pkey_index]
133 					))) {
134 			pr_warn_ratelimited("bad pkey = 0x%0x\n", pkey);
135 			set_bad_pkey_cntr(port);
136 			goto err1;
137 		}
138 		pkt->pkey_index = qp->attr.pkey_index;
139 	}
140 
141 	if ((qp_type(qp) == IB_QPT_UD || qp_type(qp) == IB_QPT_GSI) &&
142 	    pkt->mask) {
143 		u32 qkey = (qpn == 1) ? GSI_QKEY : qp->attr.qkey;
144 
145 		if (unlikely(deth_qkey(pkt) != qkey)) {
146 			pr_warn_ratelimited("bad qkey, got 0x%x expected 0x%x for qpn 0x%x\n",
147 					    deth_qkey(pkt), qkey, qpn);
148 			set_qkey_viol_cntr(port);
149 			goto err1;
150 		}
151 	}
152 
153 	return 0;
154 
155 err1:
156 	return -EINVAL;
157 }
158 
check_addr(struct rxe_dev * rxe,struct rxe_pkt_info * pkt,struct rxe_qp * qp)159 static int check_addr(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
160 		      struct rxe_qp *qp)
161 {
162 	struct sk_buff *skb = PKT_TO_SKB(pkt);
163 
164 	if (qp_type(qp) != IB_QPT_RC && qp_type(qp) != IB_QPT_UC)
165 		goto done;
166 
167 	if (unlikely(pkt->port_num != qp->attr.port_num)) {
168 		pr_warn_ratelimited("port %d != qp port %d\n",
169 				    pkt->port_num, qp->attr.port_num);
170 		goto err1;
171 	}
172 
173 	if (skb->protocol == htons(ETH_P_IP)) {
174 		struct in_addr *saddr =
175 			&qp->pri_av.sgid_addr._sockaddr_in.sin_addr;
176 		struct in_addr *daddr =
177 			&qp->pri_av.dgid_addr._sockaddr_in.sin_addr;
178 
179 		if (ip_hdr(skb)->daddr != saddr->s_addr) {
180 			pr_warn_ratelimited("dst addr %pI4 != qp source addr %pI4\n",
181 					    &ip_hdr(skb)->daddr,
182 					    &saddr->s_addr);
183 			goto err1;
184 		}
185 
186 		if (ip_hdr(skb)->saddr != daddr->s_addr) {
187 			pr_warn_ratelimited("source addr %pI4 != qp dst addr %pI4\n",
188 					    &ip_hdr(skb)->saddr,
189 					    &daddr->s_addr);
190 			goto err1;
191 		}
192 
193 	} else if (skb->protocol == htons(ETH_P_IPV6)) {
194 		struct in6_addr *saddr =
195 			&qp->pri_av.sgid_addr._sockaddr_in6.sin6_addr;
196 		struct in6_addr *daddr =
197 			&qp->pri_av.dgid_addr._sockaddr_in6.sin6_addr;
198 
199 		if (memcmp(&ipv6_hdr(skb)->daddr, saddr, sizeof(*saddr))) {
200 			pr_warn_ratelimited("dst addr %pI6 != qp source addr %pI6\n",
201 					    &ipv6_hdr(skb)->daddr, saddr);
202 			goto err1;
203 		}
204 
205 		if (memcmp(&ipv6_hdr(skb)->saddr, daddr, sizeof(*daddr))) {
206 			pr_warn_ratelimited("source addr %pI6 != qp dst addr %pI6\n",
207 					    &ipv6_hdr(skb)->saddr, daddr);
208 			goto err1;
209 		}
210 	}
211 
212 done:
213 	return 0;
214 
215 err1:
216 	return -EINVAL;
217 }
218 
hdr_check(struct rxe_pkt_info * pkt)219 static int hdr_check(struct rxe_pkt_info *pkt)
220 {
221 	struct rxe_dev *rxe = pkt->rxe;
222 	struct rxe_port *port = &rxe->port;
223 	struct rxe_qp *qp = NULL;
224 	u32 qpn = bth_qpn(pkt);
225 	int index;
226 	int err;
227 
228 	if (unlikely(bth_tver(pkt) != BTH_TVER)) {
229 		pr_warn_ratelimited("bad tver\n");
230 		goto err1;
231 	}
232 
233 	if (unlikely(qpn == 0)) {
234 		pr_warn_once("QP 0 not supported");
235 		goto err1;
236 	}
237 
238 	if (qpn != IB_MULTICAST_QPN) {
239 		index = (qpn == 1) ? port->qp_gsi_index : qpn;
240 
241 		qp = rxe_pool_get_index(&rxe->qp_pool, index);
242 		if (unlikely(!qp)) {
243 			pr_warn_ratelimited("no qp matches qpn 0x%x\n", qpn);
244 			goto err1;
245 		}
246 
247 		err = check_type_state(rxe, pkt, qp);
248 		if (unlikely(err))
249 			goto err2;
250 
251 		err = check_addr(rxe, pkt, qp);
252 		if (unlikely(err))
253 			goto err2;
254 
255 		err = check_keys(rxe, pkt, qpn, qp);
256 		if (unlikely(err))
257 			goto err2;
258 	} else {
259 		if (unlikely((pkt->mask & RXE_GRH_MASK) == 0)) {
260 			pr_warn_ratelimited("no grh for mcast qpn\n");
261 			goto err1;
262 		}
263 	}
264 
265 	pkt->qp = qp;
266 	return 0;
267 
268 err2:
269 	rxe_drop_ref(qp);
270 err1:
271 	return -EINVAL;
272 }
273 
rxe_rcv_pkt(struct rxe_pkt_info * pkt,struct sk_buff * skb)274 static inline void rxe_rcv_pkt(struct rxe_pkt_info *pkt, struct sk_buff *skb)
275 {
276 	if (pkt->mask & RXE_REQ_MASK)
277 		rxe_resp_queue_pkt(pkt->qp, skb);
278 	else
279 		rxe_comp_queue_pkt(pkt->qp, skb);
280 }
281 
rxe_rcv_mcast_pkt(struct rxe_dev * rxe,struct sk_buff * skb)282 static void rxe_rcv_mcast_pkt(struct rxe_dev *rxe, struct sk_buff *skb)
283 {
284 	struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
285 	struct rxe_mc_grp *mcg;
286 	struct rxe_mc_elem *mce;
287 	struct rxe_qp *qp;
288 	union ib_gid dgid;
289 	struct sk_buff *per_qp_skb;
290 	struct rxe_pkt_info *per_qp_pkt;
291 	int err;
292 
293 	if (skb->protocol == htons(ETH_P_IP))
294 		ipv6_addr_set_v4mapped(ip_hdr(skb)->daddr,
295 				       (struct in6_addr *)&dgid);
296 	else if (skb->protocol == htons(ETH_P_IPV6))
297 		memcpy(&dgid, &ipv6_hdr(skb)->daddr, sizeof(dgid));
298 
299 	/* lookup mcast group corresponding to mgid, takes a ref */
300 	mcg = rxe_pool_get_key(&rxe->mc_grp_pool, &dgid);
301 	if (!mcg)
302 		goto err1;	/* mcast group not registered */
303 
304 	spin_lock_bh(&mcg->mcg_lock);
305 
306 	list_for_each_entry(mce, &mcg->qp_list, qp_list) {
307 		qp = mce->qp;
308 
309 		/* validate qp for incoming packet */
310 		err = check_type_state(rxe, pkt, qp);
311 		if (err)
312 			continue;
313 
314 		err = check_keys(rxe, pkt, bth_qpn(pkt), qp);
315 		if (err)
316 			continue;
317 
318 		/* for all but the last qp create a new clone of the
319 		 * skb and pass to the qp. If an error occurs in the
320 		 * checks for the last qp in the list we need to
321 		 * free the skb since it hasn't been passed on to
322 		 * rxe_rcv_pkt() which would free it later.
323 		 */
324 		if (mce->qp_list.next != &mcg->qp_list) {
325 			per_qp_skb = skb_clone(skb, GFP_ATOMIC);
326 		} else {
327 			per_qp_skb = skb;
328 			/* show we have consumed the skb */
329 			skb = NULL;
330 		}
331 
332 		if (unlikely(!per_qp_skb))
333 			continue;
334 
335 		per_qp_pkt = SKB_TO_PKT(per_qp_skb);
336 		per_qp_pkt->qp = qp;
337 		rxe_add_ref(qp);
338 		rxe_rcv_pkt(per_qp_pkt, per_qp_skb);
339 	}
340 
341 	spin_unlock_bh(&mcg->mcg_lock);
342 
343 	rxe_drop_ref(mcg);	/* drop ref from rxe_pool_get_key. */
344 
345 err1:
346 	/* free skb if not consumed */
347 	kfree_skb(skb);
348 }
349 
rxe_match_dgid(struct rxe_dev * rxe,struct sk_buff * skb)350 static int rxe_match_dgid(struct rxe_dev *rxe, struct sk_buff *skb)
351 {
352 	struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
353 	const struct ib_gid_attr *gid_attr;
354 	union ib_gid dgid;
355 	union ib_gid *pdgid;
356 
357 	if (pkt->mask & RXE_LOOPBACK_MASK)
358 		return 0;
359 
360 	if (skb->protocol == htons(ETH_P_IP)) {
361 		ipv6_addr_set_v4mapped(ip_hdr(skb)->daddr,
362 				       (struct in6_addr *)&dgid);
363 		pdgid = &dgid;
364 	} else {
365 		pdgid = (union ib_gid *)&ipv6_hdr(skb)->daddr;
366 	}
367 
368 	gid_attr = rdma_find_gid_by_port(&rxe->ib_dev, pdgid,
369 					 IB_GID_TYPE_ROCE_UDP_ENCAP,
370 					 1, skb->dev);
371 	if (IS_ERR(gid_attr))
372 		return PTR_ERR(gid_attr);
373 
374 	rdma_put_gid_attr(gid_attr);
375 	return 0;
376 }
377 
378 /* rxe_rcv is called from the interface driver */
rxe_rcv(struct sk_buff * skb)379 void rxe_rcv(struct sk_buff *skb)
380 {
381 	int err;
382 	struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
383 	struct rxe_dev *rxe = pkt->rxe;
384 	__be32 *icrcp;
385 	u32 calc_icrc, pack_icrc;
386 
387 	pkt->offset = 0;
388 
389 	if (unlikely(skb->len < pkt->offset + RXE_BTH_BYTES))
390 		goto drop;
391 
392 	if (rxe_match_dgid(rxe, skb) < 0) {
393 		pr_warn_ratelimited("failed matching dgid\n");
394 		goto drop;
395 	}
396 
397 	pkt->opcode = bth_opcode(pkt);
398 	pkt->psn = bth_psn(pkt);
399 	pkt->qp = NULL;
400 	pkt->mask |= rxe_opcode[pkt->opcode].mask;
401 
402 	if (unlikely(skb->len < header_size(pkt)))
403 		goto drop;
404 
405 	err = hdr_check(pkt);
406 	if (unlikely(err))
407 		goto drop;
408 
409 	/* Verify ICRC */
410 	icrcp = (__be32 *)(pkt->hdr + pkt->paylen - RXE_ICRC_SIZE);
411 	pack_icrc = be32_to_cpu(*icrcp);
412 
413 	calc_icrc = rxe_icrc_hdr(pkt, skb);
414 	calc_icrc = rxe_crc32(rxe, calc_icrc, (u8 *)payload_addr(pkt),
415 			      payload_size(pkt) + bth_pad(pkt));
416 	calc_icrc = (__force u32)cpu_to_be32(~calc_icrc);
417 	if (unlikely(calc_icrc != pack_icrc)) {
418 		if (skb->protocol == htons(ETH_P_IPV6))
419 			pr_warn_ratelimited("bad ICRC from %pI6c\n",
420 					    &ipv6_hdr(skb)->saddr);
421 		else if (skb->protocol == htons(ETH_P_IP))
422 			pr_warn_ratelimited("bad ICRC from %pI4\n",
423 					    &ip_hdr(skb)->saddr);
424 		else
425 			pr_warn_ratelimited("bad ICRC from unknown\n");
426 
427 		goto drop;
428 	}
429 
430 	rxe_counter_inc(rxe, RXE_CNT_RCVD_PKTS);
431 
432 	if (unlikely(bth_qpn(pkt) == IB_MULTICAST_QPN))
433 		rxe_rcv_mcast_pkt(rxe, skb);
434 	else
435 		rxe_rcv_pkt(pkt, skb);
436 
437 	return;
438 
439 drop:
440 	if (pkt->qp)
441 		rxe_drop_ref(pkt->qp);
442 
443 	kfree_skb(skb);
444 }
445