LXR linux/drivers/infiniband/sw/rxe/rxe

   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
  39static int check_type_state(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
  40                            struct rxe_qp *qp)
  41{
  42        if (unlikely(!qp->valid))
  43                goto err1;
  44
  45        switch (qp_type(qp)) {
  46        case IB_QPT_RC:
  47                if (unlikely((pkt->opcode & IB_OPCODE_RC) != 0)) {
  48                        pr_warn_ratelimited("bad qp type\n");
  49                        goto err1;
  50                }
  51                break;
  52        case IB_QPT_UC:
  53                if (unlikely(!(pkt->opcode & IB_OPCODE_UC))) {
  54                        pr_warn_ratelimited("bad qp type\n");
  55                        goto err1;
  56                }
  57                break;
  58        case IB_QPT_UD:
  59        case IB_QPT_SMI:
  60        case IB_QPT_GSI:
  61                if (unlikely(!(pkt->opcode & IB_OPCODE_UD))) {
  62                        pr_warn_ratelimited("bad qp type\n");
  63                        goto err1;
  64                }
  65                break;
  66        default:
  67                pr_warn_ratelimited("unsupported qp type\n");
  68                goto err1;
  69        }
  70
  71        if (pkt->mask & RXE_REQ_MASK) {
  72                if (unlikely(qp->resp.state != QP_STATE_READY))
  73                        goto err1;
  74        } else if (unlikely(qp->req.state < QP_STATE_READY ||
  75                                qp->req.state > QP_STATE_DRAINED)) {
  76                goto err1;
  77        }
  78
  79        return 0;
  80
  81err1:
  82        return -EINVAL;
  83}
  84
  85static void set_bad_pkey_cntr(struct rxe_port *port)
  86{
  87        spin_lock_bh(&port->port_lock);
  88        port->attr.bad_pkey_cntr = min((u32)0xffff,
  89                                       port->attr.bad_pkey_cntr + 1);
  90        spin_unlock_bh(&port->port_lock);
  91}
  92
  93static void set_qkey_viol_cntr(struct rxe_port *port)
  94{
  95        spin_lock_bh(&port->port_lock);
  96        port->attr.qkey_viol_cntr = min((u32)0xffff,
  97                                        port->attr.qkey_viol_cntr + 1);
  98        spin_unlock_bh(&port->port_lock);
  99}
 100
 101static int check_keys(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
 102                      u32 qpn, struct rxe_qp *qp)
 103{
 104        int i;
 105        int found_pkey = 0;
 106        struct rxe_port *port = &rxe->port;
 107        u16 pkey = bth_pkey(pkt);
 108
 109        pkt->pkey_index = 0;
 110
 111        if (qpn == 1) {
 112                for (i = 0; i < port->attr.pkey_tbl_len; i++) {
 113                        if (pkey_match(pkey, port->pkey_tbl[i])) {
 114                                pkt->pkey_index = i;
 115                                found_pkey = 1;
 116                                break;
 117                        }
 118                }
 119
 120                if (!found_pkey) {
 121                        pr_warn_ratelimited("bad pkey = 0x%x\n", pkey);
 122                        set_bad_pkey_cntr(port);
 123                        goto err1;
 124                }
 125        } else {
 126                if (unlikely(!pkey_match(pkey,
 127                                         port->pkey_tbl[qp->attr.pkey_index]
 128                                        ))) {
 129                        pr_warn_ratelimited("bad pkey = 0x%0x\n", pkey);
 130                        set_bad_pkey_cntr(port);
 131                        goto err1;
 132                }
 133                pkt->pkey_index = qp->attr.pkey_index;
 134        }
 135
 136        if ((qp_type(qp) == IB_QPT_UD || qp_type(qp) == IB_QPT_GSI) &&
 137            pkt->mask) {
 138                u32 qkey = (qpn == 1) ? GSI_QKEY : qp->attr.qkey;
 139
 140                if (unlikely(deth_qkey(pkt) != qkey)) {
 141                        pr_warn_ratelimited("bad qkey, got 0x%x expected 0x%x for qpn 0x%x\n",
 142                                            deth_qkey(pkt), qkey, qpn);
 143                        set_qkey_viol_cntr(port);
 144                        goto err1;
 145                }
 146        }
 147
 148        return 0;
 149
 150err1:
 151        return -EINVAL;
 152}
 153
 154static int check_addr(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
 155                      struct rxe_qp *qp)
 156{
 157        struct sk_buff *skb = PKT_TO_SKB(pkt);
 158
 159        if (qp_type(qp) != IB_QPT_RC && qp_type(qp) != IB_QPT_UC)
 160                goto done;
 161
 162        if (unlikely(pkt->port_num != qp->attr.port_num)) {
 163                pr_warn_ratelimited("port %d != qp port %d\n",
 164                                    pkt->port_num, qp->attr.port_num);
 165                goto err1;
 166        }
 167
 168        if (skb->protocol == htons(ETH_P_IP)) {
 169                struct in_addr *saddr =
 170                        &qp->pri_av.sgid_addr._sockaddr_in.sin_addr;
 171                struct in_addr *daddr =
 172                        &qp->pri_av.dgid_addr._sockaddr_in.sin_addr;
 173
 174                if (ip_hdr(skb)->daddr != saddr->s_addr) {
 175                        pr_warn_ratelimited("dst addr %pI4 != qp source addr %pI4\n",
 176                                            &ip_hdr(skb)->daddr,
 177                                            &saddr->s_addr);
 178                        goto err1;
 179                }
 180
 181                if (ip_hdr(skb)->saddr != daddr->s_addr) {
 182                        pr_warn_ratelimited("source addr %pI4 != qp dst addr %pI4\n",
 183                                            &ip_hdr(skb)->saddr,
 184                                            &daddr->s_addr);
 185                        goto err1;
 186                }
 187
 188        } else if (skb->protocol == htons(ETH_P_IPV6)) {
 189                struct in6_addr *saddr =
 190                        &qp->pri_av.sgid_addr._sockaddr_in6.sin6_addr;
 191                struct in6_addr *daddr =
 192                        &qp->pri_av.dgid_addr._sockaddr_in6.sin6_addr;
 193
 194                if (memcmp(&ipv6_hdr(skb)->daddr, saddr, sizeof(*saddr))) {
 195                        pr_warn_ratelimited("dst addr %pI6 != qp source addr %pI6\n",
 196                                            &ipv6_hdr(skb)->daddr, saddr);
 197                        goto err1;
 198                }
 199
 200                if (memcmp(&ipv6_hdr(skb)->saddr, daddr, sizeof(*daddr))) {
 201                        pr_warn_ratelimited("source addr %pI6 != qp dst addr %pI6\n",
 202                                            &ipv6_hdr(skb)->saddr, daddr);
 203                        goto err1;
 204                }
 205        }
 206
 207done:
 208        return 0;
 209
 210err1:
 211        return -EINVAL;
 212}
 213
 214static int hdr_check(struct rxe_pkt_info *pkt)
 215{
 216        struct rxe_dev *rxe = pkt->rxe;
 217        struct rxe_port *port = &rxe->port;
 218        struct rxe_qp *qp = NULL;
 219        u32 qpn = bth_qpn(pkt);
 220        int index;
 221        int err;
 222
 223        if (unlikely(bth_tver(pkt) != BTH_TVER)) {
 224                pr_warn_ratelimited("bad tver\n");
 225                goto err1;
 226        }
 227
 228        if (unlikely(qpn == 0)) {
 229                pr_warn_once("QP 0 not supported");
 230                goto err1;
 231        }
 232
 233        if (qpn != IB_MULTICAST_QPN) {
 234                index = (qpn == 1) ? port->qp_gsi_index : qpn;
 235
 236                qp = rxe_pool_get_index(&rxe->qp_pool, index);
 237                if (unlikely(!qp)) {
 238                        pr_warn_ratelimited("no qp matches qpn 0x%x\n", qpn);
 239                        goto err1;
 240                }
 241
 242                err = check_type_state(rxe, pkt, qp);
 243                if (unlikely(err))
 244                        goto err2;
 245
 246                err = check_addr(rxe, pkt, qp);
 247                if (unlikely(err))
 248                        goto err2;
 249
 250                err = check_keys(rxe, pkt, qpn, qp);
 251                if (unlikely(err))
 252                        goto err2;
 253        } else {
 254                if (unlikely((pkt->mask & RXE_GRH_MASK) == 0)) {
 255                        pr_warn_ratelimited("no grh for mcast qpn\n");
 256                        goto err1;
 257                }
 258        }
 259
 260        pkt->qp = qp;
 261        return 0;
 262
 263err2:
 264        rxe_drop_ref(qp);
 265err1:
 266        return -EINVAL;
 267}
 268
 269static inline void rxe_rcv_pkt(struct rxe_pkt_info *pkt, struct sk_buff *skb)
 270{
 271        if (pkt->mask & RXE_REQ_MASK)
 272                rxe_resp_queue_pkt(pkt->qp, skb);
 273        else
 274                rxe_comp_queue_pkt(pkt->qp, skb);
 275}
 276
 277static void rxe_rcv_mcast_pkt(struct rxe_dev *rxe, struct sk_buff *skb)
 278{
 279        struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
 280        struct rxe_mc_grp *mcg;
 281        struct rxe_mc_elem *mce;
 282        struct rxe_qp *qp;
 283        union ib_gid dgid;
 284        int err;
 285
 286        if (skb->protocol == htons(ETH_P_IP))
 287                ipv6_addr_set_v4mapped(ip_hdr(skb)->daddr,
 288                                       (struct in6_addr *)&dgid);
 289        else if (skb->protocol == htons(ETH_P_IPV6))
 290                memcpy(&dgid, &ipv6_hdr(skb)->daddr, sizeof(dgid));
 291
 292        /* lookup mcast group corresponding to mgid, takes a ref */
 293        mcg = rxe_pool_get_key(&rxe->mc_grp_pool, &dgid);
 294        if (!mcg)
 295                goto err1;      /* mcast group not registered */
 296
 297        spin_lock_bh(&mcg->mcg_lock);
 298
 299        list_for_each_entry(mce, &mcg->qp_list, qp_list) {
 300                qp = mce->qp;
 301                pkt = SKB_TO_PKT(skb);
 302
 303                /* validate qp for incoming packet */
 304                err = check_type_state(rxe, pkt, qp);
 305                if (err)
 306                        continue;
 307
 308                err = check_keys(rxe, pkt, bth_qpn(pkt), qp);
 309                if (err)
 310                        continue;
 311
 312                /* if *not* the last qp in the list
 313                 * increase the users of the skb then post to the next qp
 314                 */
 315                if (mce->qp_list.next != &mcg->qp_list)
 316                        skb_get(skb);
 317
 318                pkt->qp = qp;
 319                rxe_add_ref(qp);
 320                rxe_rcv_pkt(pkt, skb);
 321        }
 322
 323        spin_unlock_bh(&mcg->mcg_lock);
 324
 325        rxe_drop_ref(mcg);      /* drop ref from rxe_pool_get_key. */
 326
 327err1:
 328        kfree_skb(skb);
 329}
 330
 331static int rxe_match_dgid(struct rxe_dev *rxe, struct sk_buff *skb)
 332{
 333        const struct ib_gid_attr *gid_attr;
 334        union ib_gid dgid;
 335        union ib_gid *pdgid;
 336
 337        if (skb->protocol == htons(ETH_P_IP)) {
 338                ipv6_addr_set_v4mapped(ip_hdr(skb)->daddr,
 339                                       (struct in6_addr *)&dgid);
 340                pdgid = &dgid;
 341        } else {
 342                pdgid = (union ib_gid *)&ipv6_hdr(skb)->daddr;
 343        }
 344
 345        gid_attr = rdma_find_gid_by_port(&rxe->ib_dev, pdgid,
 346                                         IB_GID_TYPE_ROCE_UDP_ENCAP,
 347                                         1, skb->dev);
 348        if (IS_ERR(gid_attr))
 349                return PTR_ERR(gid_attr);
 350
 351        rdma_put_gid_attr(gid_attr);
 352        return 0;
 353}
 354
 355/* rxe_rcv is called from the interface driver */
 356void rxe_rcv(struct sk_buff *skb)
 357{
 358        int err;
 359        struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
 360        struct rxe_dev *rxe = pkt->rxe;
 361        __be32 *icrcp;
 362        u32 calc_icrc, pack_icrc;
 363
 364        pkt->offset = 0;
 365
 366        if (unlikely(skb->len < pkt->offset + RXE_BTH_BYTES))
 367                goto drop;
 368
 369        if (unlikely(rxe_match_dgid(rxe, skb) < 0)) {
 370                pr_warn_ratelimited("failed matching dgid\n");
 371                goto drop;
 372        }
 373
 374        pkt->opcode = bth_opcode(pkt);
 375        pkt->psn = bth_psn(pkt);
 376        pkt->qp = NULL;
 377        pkt->mask |= rxe_opcode[pkt->opcode].mask;
 378
 379        if (unlikely(skb->len < header_size(pkt)))
 380                goto drop;
 381
 382        err = hdr_check(pkt);
 383        if (unlikely(err))
 384                goto drop;
 385
 386        /* Verify ICRC */
 387        icrcp = (__be32 *)(pkt->hdr + pkt->paylen - RXE_ICRC_SIZE);
 388        pack_icrc = be32_to_cpu(*icrcp);
 389
 390        calc_icrc = rxe_icrc_hdr(pkt, skb);
 391        calc_icrc = rxe_crc32(rxe, calc_icrc, (u8 *)payload_addr(pkt),
 392                              payload_size(pkt) + bth_pad(pkt));
 393        calc_icrc = (__force u32)cpu_to_be32(~calc_icrc);
 394        if (unlikely(calc_icrc != pack_icrc)) {
 395                if (skb->protocol == htons(ETH_P_IPV6))
 396                        pr_warn_ratelimited("bad ICRC from %pI6c\n",
 397                                            &ipv6_hdr(skb)->saddr);
 398                else if (skb->protocol == htons(ETH_P_IP))
 399                        pr_warn_ratelimited("bad ICRC from %pI4\n",
 400                                            &ip_hdr(skb)->saddr);
 401                else
 402                        pr_warn_ratelimited("bad ICRC from unknown\n");
 403
 404                goto drop;
 405        }
 406
 407        rxe_counter_inc(rxe, RXE_CNT_RCVD_PKTS);
 408
 409        if (unlikely(bth_qpn(pkt) == IB_MULTICAST_QPN))
 410                rxe_rcv_mcast_pkt(rxe, skb);
 411        else
 412                rxe_rcv_pkt(pkt, skb);
 413
 414        return;
 415
 416drop:
 417        if (pkt->qp)
 418                rxe_drop_ref(pkt->qp);
 419
 420        kfree_skb(skb);
 421}
 422