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        struct rxe_port *port = &rxe->port;
 105        u16 pkey = bth_pkey(pkt);
 106
 107        pkt->pkey_index = 0;
 108
 109        if (!pkey_match(pkey, IB_DEFAULT_PKEY_FULL)) {
 110                pr_warn_ratelimited("bad pkey = 0x%x\n", pkey);
 111                set_bad_pkey_cntr(port);
 112                goto err1;
 113        }
 114
 115        if ((qp_type(qp) == IB_QPT_UD || qp_type(qp) == IB_QPT_GSI) &&
 116            pkt->mask) {
 117                u32 qkey = (qpn == 1) ? GSI_QKEY : qp->attr.qkey;
 118
 119                if (unlikely(deth_qkey(pkt) != qkey)) {
 120                        pr_warn_ratelimited("bad qkey, got 0x%x expected 0x%x for qpn 0x%x\n",
 121                                            deth_qkey(pkt), qkey, qpn);
 122                        set_qkey_viol_cntr(port);
 123                        goto err1;
 124                }
 125        }
 126
 127        return 0;
 128
 129err1:
 130        return -EINVAL;
 131}
 132
 133static int check_addr(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
 134                      struct rxe_qp *qp)
 135{
 136        struct sk_buff *skb = PKT_TO_SKB(pkt);
 137
 138        if (qp_type(qp) != IB_QPT_RC && qp_type(qp) != IB_QPT_UC)
 139                goto done;
 140
 141        if (unlikely(pkt->port_num != qp->attr.port_num)) {
 142                pr_warn_ratelimited("port %d != qp port %d\n",
 143                                    pkt->port_num, qp->attr.port_num);
 144                goto err1;
 145        }
 146
 147        if (skb->protocol == htons(ETH_P_IP)) {
 148                struct in_addr *saddr =
 149                        &qp->pri_av.sgid_addr._sockaddr_in.sin_addr;
 150                struct in_addr *daddr =
 151                        &qp->pri_av.dgid_addr._sockaddr_in.sin_addr;
 152
 153                if (ip_hdr(skb)->daddr != saddr->s_addr) {
 154                        pr_warn_ratelimited("dst addr %pI4 != qp source addr %pI4\n",
 155                                            &ip_hdr(skb)->daddr,
 156                                            &saddr->s_addr);
 157                        goto err1;
 158                }
 159
 160                if (ip_hdr(skb)->saddr != daddr->s_addr) {
 161                        pr_warn_ratelimited("source addr %pI4 != qp dst addr %pI4\n",
 162                                            &ip_hdr(skb)->saddr,
 163                                            &daddr->s_addr);
 164                        goto err1;
 165                }
 166
 167        } else if (skb->protocol == htons(ETH_P_IPV6)) {
 168                struct in6_addr *saddr =
 169                        &qp->pri_av.sgid_addr._sockaddr_in6.sin6_addr;
 170                struct in6_addr *daddr =
 171                        &qp->pri_av.dgid_addr._sockaddr_in6.sin6_addr;
 172
 173                if (memcmp(&ipv6_hdr(skb)->daddr, saddr, sizeof(*saddr))) {
 174                        pr_warn_ratelimited("dst addr %pI6 != qp source addr %pI6\n",
 175                                            &ipv6_hdr(skb)->daddr, saddr);
 176                        goto err1;
 177                }
 178
 179                if (memcmp(&ipv6_hdr(skb)->saddr, daddr, sizeof(*daddr))) {
 180                        pr_warn_ratelimited("source addr %pI6 != qp dst addr %pI6\n",
 181                                            &ipv6_hdr(skb)->saddr, daddr);
 182                        goto err1;
 183                }
 184        }
 185
 186done:
 187        return 0;
 188
 189err1:
 190        return -EINVAL;
 191}
 192
 193static int hdr_check(struct rxe_pkt_info *pkt)
 194{
 195        struct rxe_dev *rxe = pkt->rxe;
 196        struct rxe_port *port = &rxe->port;
 197        struct rxe_qp *qp = NULL;
 198        u32 qpn = bth_qpn(pkt);
 199        int index;
 200        int err;
 201
 202        if (unlikely(bth_tver(pkt) != BTH_TVER)) {
 203                pr_warn_ratelimited("bad tver\n");
 204                goto err1;
 205        }
 206
 207        if (unlikely(qpn == 0)) {
 208                pr_warn_once("QP 0 not supported");
 209                goto err1;
 210        }
 211
 212        if (qpn != IB_MULTICAST_QPN) {
 213                index = (qpn == 1) ? port->qp_gsi_index : qpn;
 214
 215                qp = rxe_pool_get_index(&rxe->qp_pool, index);
 216                if (unlikely(!qp)) {
 217                        pr_warn_ratelimited("no qp matches qpn 0x%x\n", qpn);
 218                        goto err1;
 219                }
 220
 221                err = check_type_state(rxe, pkt, qp);
 222                if (unlikely(err))
 223                        goto err2;
 224
 225                err = check_addr(rxe, pkt, qp);
 226                if (unlikely(err))
 227                        goto err2;
 228
 229                err = check_keys(rxe, pkt, qpn, qp);
 230                if (unlikely(err))
 231                        goto err2;
 232        } else {
 233                if (unlikely((pkt->mask & RXE_GRH_MASK) == 0)) {
 234                        pr_warn_ratelimited("no grh for mcast qpn\n");
 235                        goto err1;
 236                }
 237        }
 238
 239        pkt->qp = qp;
 240        return 0;
 241
 242err2:
 243        rxe_drop_ref(qp);
 244err1:
 245        return -EINVAL;
 246}
 247
 248static inline void rxe_rcv_pkt(struct rxe_pkt_info *pkt, struct sk_buff *skb)
 249{
 250        if (pkt->mask & RXE_REQ_MASK)
 251                rxe_resp_queue_pkt(pkt->qp, skb);
 252        else
 253                rxe_comp_queue_pkt(pkt->qp, skb);
 254}
 255
 256static void rxe_rcv_mcast_pkt(struct rxe_dev *rxe, struct sk_buff *skb)
 257{
 258        struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
 259        struct rxe_mc_grp *mcg;
 260        struct rxe_mc_elem *mce;
 261        struct rxe_qp *qp;
 262        union ib_gid dgid;
 263        struct sk_buff *per_qp_skb;
 264        struct rxe_pkt_info *per_qp_pkt;
 265        int err;
 266
 267        if (skb->protocol == htons(ETH_P_IP))
 268                ipv6_addr_set_v4mapped(ip_hdr(skb)->daddr,
 269                                       (struct in6_addr *)&dgid);
 270        else if (skb->protocol == htons(ETH_P_IPV6))
 271                memcpy(&dgid, &ipv6_hdr(skb)->daddr, sizeof(dgid));
 272
 273        /* lookup mcast group corresponding to mgid, takes a ref */
 274        mcg = rxe_pool_get_key(&rxe->mc_grp_pool, &dgid);
 275        if (!mcg)
 276                goto err1;      /* mcast group not registered */
 277
 278        spin_lock_bh(&mcg->mcg_lock);
 279
 280        list_for_each_entry(mce, &mcg->qp_list, qp_list) {
 281                qp = mce->qp;
 282                pkt = SKB_TO_PKT(skb);
 283
 284                /* validate qp for incoming packet */
 285                err = check_type_state(rxe, pkt, qp);
 286                if (err)
 287                        continue;
 288
 289                err = check_keys(rxe, pkt, bth_qpn(pkt), qp);
 290                if (err)
 291                        continue;
 292
 293                /* for all but the last qp create a new clone of the
 294                 * skb and pass to the qp.
 295                 */
 296                if (mce->qp_list.next != &mcg->qp_list)
 297                        per_qp_skb = skb_clone(skb, GFP_ATOMIC);
 298                else
 299                        per_qp_skb = skb;
 300
 301                if (unlikely(!per_qp_skb))
 302                        continue;
 303
 304                per_qp_pkt = SKB_TO_PKT(per_qp_skb);
 305                per_qp_pkt->qp = qp;
 306                rxe_add_ref(qp);
 307                rxe_rcv_pkt(per_qp_pkt, per_qp_skb);
 308        }
 309
 310        spin_unlock_bh(&mcg->mcg_lock);
 311
 312        rxe_drop_ref(mcg);      /* drop ref from rxe_pool_get_key. */
 313
 314        return;
 315
 316err1:
 317        kfree_skb(skb);
 318}
 319
 320/**
 321 * rxe_chk_dgid - validate destination IP address
 322 * @rxe: rxe device that received packet
 323 * @skb: the received packet buffer
 324 *
 325 * Accept any loopback packets
 326 * Extract IP address from packet and
 327 * Accept if multicast packet
 328 * Accept if matches an SGID table entry
 329 */
 330static int rxe_chk_dgid(struct rxe_dev *rxe, struct sk_buff *skb)
 331{
 332        struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
 333        const struct ib_gid_attr *gid_attr;
 334        union ib_gid dgid;
 335        union ib_gid *pdgid;
 336
 337        if (pkt->mask & RXE_LOOPBACK_MASK)
 338                return 0;
 339
 340        if (skb->protocol == htons(ETH_P_IP)) {
 341                ipv6_addr_set_v4mapped(ip_hdr(skb)->daddr,
 342                                       (struct in6_addr *)&dgid);
 343                pdgid = &dgid;
 344        } else {
 345                pdgid = (union ib_gid *)&ipv6_hdr(skb)->daddr;
 346        }
 347
 348        if (rdma_is_multicast_addr((struct in6_addr *)pdgid))
 349                return 0;
 350
 351        gid_attr = rdma_find_gid_by_port(&rxe->ib_dev, pdgid,
 352                                         IB_GID_TYPE_ROCE_UDP_ENCAP,
 353                                         1, skb->dev);
 354        if (IS_ERR(gid_attr))
 355                return PTR_ERR(gid_attr);
 356
 357        rdma_put_gid_attr(gid_attr);
 358        return 0;
 359}
 360
 361/* rxe_rcv is called from the interface driver */
 362void rxe_rcv(struct sk_buff *skb)
 363{
 364        int err;
 365        struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
 366        struct rxe_dev *rxe = pkt->rxe;
 367        __be32 *icrcp;
 368        u32 calc_icrc, pack_icrc;
 369
 370        pkt->offset = 0;
 371
 372        if (unlikely(skb->len < pkt->offset + RXE_BTH_BYTES))
 373                goto drop;
 374
 375        if (rxe_chk_dgid(rxe, skb) < 0) {
 376                pr_warn_ratelimited("failed checking dgid\n");
 377                goto drop;
 378        }
 379
 380        pkt->opcode = bth_opcode(pkt);
 381        pkt->psn = bth_psn(pkt);
 382        pkt->qp = NULL;
 383        pkt->mask |= rxe_opcode[pkt->opcode].mask;
 384
 385        if (unlikely(skb->len < header_size(pkt)))
 386                goto drop;
 387
 388        err = hdr_check(pkt);
 389        if (unlikely(err))
 390                goto drop;
 391
 392        /* Verify ICRC */
 393        icrcp = (__be32 *)(pkt->hdr + pkt->paylen - RXE_ICRC_SIZE);
 394        pack_icrc = be32_to_cpu(*icrcp);
 395
 396        calc_icrc = rxe_icrc_hdr(pkt, skb);
 397        calc_icrc = rxe_crc32(rxe, calc_icrc, (u8 *)payload_addr(pkt),
 398                              payload_size(pkt) + bth_pad(pkt));
 399        calc_icrc = (__force u32)cpu_to_be32(~calc_icrc);
 400        if (unlikely(calc_icrc != pack_icrc)) {
 401                if (skb->protocol == htons(ETH_P_IPV6))
 402                        pr_warn_ratelimited("bad ICRC from %pI6c\n",
 403                                            &ipv6_hdr(skb)->saddr);
 404                else if (skb->protocol == htons(ETH_P_IP))
 405                        pr_warn_ratelimited("bad ICRC from %pI4\n",
 406                                            &ip_hdr(skb)->saddr);
 407                else
 408                        pr_warn_ratelimited("bad ICRC from unknown\n");
 409
 410                goto drop;
 411        }
 412
 413        rxe_counter_inc(rxe, RXE_CNT_RCVD_PKTS);
 414
 415        if (unlikely(bth_qpn(pkt) == IB_MULTICAST_QPN))
 416                rxe_rcv_mcast_pkt(rxe, skb);
 417        else
 418                rxe_rcv_pkt(pkt, skb);
 419
 420        return;
 421
 422drop:
 423        if (pkt->qp)
 424                rxe_drop_ref(pkt->qp);
 425
 426        kfree_skb(skb);
 427}
 428