drivers/target/iscsi/cxgbit/cxgbit_cm.c

   1 /*
   2  * Copyright (c) 2016 Chelsio Communications, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  */
   8
   9 #include <linux/module.h>
  10 #include <linux/list.h>
  11 #include <linux/workqueue.h>
  12 #include <linux/skbuff.h>
  13 #include <linux/timer.h>
  14 #include <linux/notifier.h>
  15 #include <linux/inetdevice.h>
  16 #include <linux/ip.h>
  17 #include <linux/tcp.h>
  18 #include <linux/if_vlan.h>
  19
  20 #include <net/neighbour.h>
  21 #include <net/netevent.h>
  22 #include <net/route.h>
  23 #include <net/tcp.h>
  24 #include <net/ip6_route.h>
  25 #include <net/addrconf.h>
  26
  27 #include <libcxgb_cm.h>
  28 #include "cxgbit.h"
  29 #include "clip_tbl.h"
  30
  31 static void cxgbit_init_wr_wait(struct cxgbit_wr_wait *wr_waitp)
  32 {
  33         wr_waitp->ret = 0;
  34         reinit_completion(&wr_waitp->completion);
  35 }
  36
  37 static void
  38 cxgbit_wake_up(struct cxgbit_wr_wait *wr_waitp, const char *func, u8 ret)
  39 {
  40         if (ret == CPL_ERR_NONE)
  41                 wr_waitp->ret = 0;
  42         else
  43                 wr_waitp->ret = -EIO;
  44
  45         if (wr_waitp->ret)
  46                 pr_err("%s: err:%u", func, ret);
  47
  48         complete(&wr_waitp->completion);
  49 }
  50
  51 static int
  52 cxgbit_wait_for_reply(struct cxgbit_device *cdev,
  53                       struct cxgbit_wr_wait *wr_waitp, u32 tid, u32 timeout,
  54                       const char *func)
  55 {
  56         int ret;
  57
  58         if (!test_bit(CDEV_STATE_UP, &cdev->flags)) {
  59                 wr_waitp->ret = -EIO;
  60                 goto out;
  61         }
  62
  63         ret = wait_for_completion_timeout(&wr_waitp->completion, timeout * HZ);
  64         if (!ret) {
  65                 pr_info("%s - Device %s not responding tid %u\n",
  66                         func, pci_name(cdev->lldi.pdev), tid);
  67                 wr_waitp->ret = -ETIMEDOUT;
  68         }
  69 out:
  70         if (wr_waitp->ret)
  71                 pr_info("%s: FW reply %d tid %u\n",
  72                         pci_name(cdev->lldi.pdev), wr_waitp->ret, tid);
  73         return wr_waitp->ret;
  74 }
  75
  76 static int cxgbit_np_hashfn(const struct cxgbit_np *cnp)
  77 {
  78         return ((unsigned long)cnp >> 10) & (NP_INFO_HASH_SIZE - 1);
  79 }
  80
  81 static struct np_info *
  82 cxgbit_np_hash_add(struct cxgbit_device *cdev, struct cxgbit_np *cnp,
  83                    unsigned int stid)
  84 {
  85         struct np_info *p = kzalloc(sizeof(*p), GFP_KERNEL);
  86
  87         if (p) {
  88                 int bucket = cxgbit_np_hashfn(cnp);
  89
  90                 p->cnp = cnp;
  91                 p->stid = stid;
  92                 spin_lock(&cdev->np_lock);
  93                 p->next = cdev->np_hash_tab[bucket];
  94                 cdev->np_hash_tab[bucket] = p;
  95                 spin_unlock(&cdev->np_lock);
  96         }
  97
  98         return p;
  99 }
 100
 101 static int
 102 cxgbit_np_hash_find(struct cxgbit_device *cdev, struct cxgbit_np *cnp)
 103 {
 104         int stid = -1, bucket = cxgbit_np_hashfn(cnp);
 105         struct np_info *p;
 106
 107         spin_lock(&cdev->np_lock);
 108         for (p = cdev->np_hash_tab[bucket]; p; p = p->next) {
 109                 if (p->cnp == cnp) {
 110                         stid = p->stid;
 111                         break;
 112                 }
 113         }
 114         spin_unlock(&cdev->np_lock);
 115
 116         return stid;
 117 }
 118
 119 static int cxgbit_np_hash_del(struct cxgbit_device *cdev, struct cxgbit_np *cnp)
 120 {
 121         int stid = -1, bucket = cxgbit_np_hashfn(cnp);
 122         struct np_info *p, **prev = &cdev->np_hash_tab[bucket];
 123
 124         spin_lock(&cdev->np_lock);
 125         for (p = *prev; p; prev = &p->next, p = p->next) {
 126                 if (p->cnp == cnp) {
 127                         stid = p->stid;
 128                         *prev = p->next;
 129                         kfree(p);
 130                         break;
 131                 }
 132         }
 133         spin_unlock(&cdev->np_lock);
 134
 135         return stid;
 136 }
 137
 138 void _cxgbit_free_cnp(struct kref *kref)
 139 {
 140         struct cxgbit_np *cnp;
 141
 142         cnp = container_of(kref, struct cxgbit_np, kref);
 143         kfree(cnp);
 144 }
 145
 146 static int
 147 cxgbit_create_server6(struct cxgbit_device *cdev, unsigned int stid,
 148                       struct cxgbit_np *cnp)
 149 {
 150         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
 151                                      &cnp->com.local_addr;
 152         int addr_type;
 153         int ret;
 154
 155         pr_debug("%s: dev = %s; stid = %u; sin6_port = %u\n",
 156                  __func__, cdev->lldi.ports[0]->name, stid, sin6->sin6_port);
 157
 158         addr_type = ipv6_addr_type((const struct in6_addr *)
 159                                    &sin6->sin6_addr);
 160         if (addr_type != IPV6_ADDR_ANY) {
 161                 ret = cxgb4_clip_get(cdev->lldi.ports[0],
 162                                      (const u32 *)&sin6->sin6_addr.s6_addr, 1);
 163                 if (ret) {
 164                         pr_err("Unable to find clip table entry. laddr %pI6. Error:%d.\n",
 165                                sin6->sin6_addr.s6_addr, ret);
 166                         return -ENOMEM;
 167                 }
 168         }
 169
 170         cxgbit_get_cnp(cnp);
 171         cxgbit_init_wr_wait(&cnp->com.wr_wait);
 172
 173         ret = cxgb4_create_server6(cdev->lldi.ports[0],
 174                                    stid, &sin6->sin6_addr,
 175                                    sin6->sin6_port,
 176                                    cdev->lldi.rxq_ids[0]);
 177         if (!ret)
 178                 ret = cxgbit_wait_for_reply(cdev, &cnp->com.wr_wait,
 179                                             0, 10, __func__);
 180         else if (ret > 0)
 181                 ret = net_xmit_errno(ret);
 182         else
 183                 cxgbit_put_cnp(cnp);
 184
 185         if (ret) {
 186                 if (ret != -ETIMEDOUT)
 187                         cxgb4_clip_release(cdev->lldi.ports[0],
 188                                    (const u32 *)&sin6->sin6_addr.s6_addr, 1);
 189
 190                 pr_err("create server6 err %d stid %d laddr %pI6 lport %d\n",
 191                        ret, stid, sin6->sin6_addr.s6_addr,
 192                        ntohs(sin6->sin6_port));
 193         }
 194
 195         return ret;
 196 }
 197
 198 static int
 199 cxgbit_create_server4(struct cxgbit_device *cdev, unsigned int stid,
 200                       struct cxgbit_np *cnp)
 201 {
 202         struct sockaddr_in *sin = (struct sockaddr_in *)
 203                                    &cnp->com.local_addr;
 204         int ret;
 205
 206         pr_debug("%s: dev = %s; stid = %u; sin_port = %u\n",
 207                  __func__, cdev->lldi.ports[0]->name, stid, sin->sin_port);
 208
 209         cxgbit_get_cnp(cnp);
 210         cxgbit_init_wr_wait(&cnp->com.wr_wait);
 211
 212         ret = cxgb4_create_server(cdev->lldi.ports[0],
 213                                   stid, sin->sin_addr.s_addr,
 214                                   sin->sin_port, 0,
 215                                   cdev->lldi.rxq_ids[0]);
 216         if (!ret)
 217                 ret = cxgbit_wait_for_reply(cdev,
 218                                             &cnp->com.wr_wait,
 219                                             0, 10, __func__);
 220         else if (ret > 0)
 221                 ret = net_xmit_errno(ret);
 222         else
 223                 cxgbit_put_cnp(cnp);
 224
 225         if (ret)
 226                 pr_err("create server failed err %d stid %d laddr %pI4 lport %d\n",
 227                        ret, stid, &sin->sin_addr, ntohs(sin->sin_port));
 228         return ret;
 229 }
 230
 231 struct cxgbit_device *cxgbit_find_device(struct net_device *ndev, u8 *port_id)
 232 {
 233         struct cxgbit_device *cdev;
 234         u8 i;
 235
 236         list_for_each_entry(cdev, &cdev_list_head, list) {
 237                 struct cxgb4_lld_info *lldi = &cdev->lldi;
 238
 239                 for (i = 0; i < lldi->nports; i++) {
 240                         if (lldi->ports[i] == ndev) {
 241                                 if (port_id)
 242                                         *port_id = i;
 243                                 return cdev;
 244                         }
 245                 }
 246         }
 247
 248         return NULL;
 249 }
 250
 251 static struct net_device *cxgbit_get_real_dev(struct net_device *ndev)
 252 {
 253         if (ndev->priv_flags & IFF_BONDING) {
 254                 pr_err("Bond devices are not supported. Interface:%s\n",
 255                        ndev->name);
 256                 return NULL;
 257         }
 258
 259         if (is_vlan_dev(ndev))
 260                 return vlan_dev_real_dev(ndev);
 261
 262         return ndev;
 263 }
 264
 265 static struct net_device *cxgbit_ipv4_netdev(__be32 saddr)
 266 {
 267         struct net_device *ndev;
 268
 269         ndev = __ip_dev_find(&init_net, saddr, false);
 270         if (!ndev)
 271                 return NULL;
 272
 273         return cxgbit_get_real_dev(ndev);
 274 }
 275
 276 static struct net_device *cxgbit_ipv6_netdev(struct in6_addr *addr6)
 277 {
 278         struct net_device *ndev = NULL;
 279         bool found = false;
 280
 281         if (IS_ENABLED(CONFIG_IPV6)) {
 282                 for_each_netdev_rcu(&init_net, ndev)
 283                         if (ipv6_chk_addr(&init_net, addr6, ndev, 1)) {
 284                                 found = true;
 285                                 break;
 286                         }
 287         }
 288         if (!found)
 289                 return NULL;
 290         return cxgbit_get_real_dev(ndev);
 291 }
 292
 293 static struct cxgbit_device *cxgbit_find_np_cdev(struct cxgbit_np *cnp)
 294 {
 295         struct sockaddr_storage *sockaddr = &cnp->com.local_addr;
 296         int ss_family = sockaddr->ss_family;
 297         struct net_device *ndev = NULL;
 298         struct cxgbit_device *cdev = NULL;
 299
 300         rcu_read_lock();
 301         if (ss_family == AF_INET) {
 302                 struct sockaddr_in *sin;
 303
 304                 sin = (struct sockaddr_in *)sockaddr;
 305                 ndev = cxgbit_ipv4_netdev(sin->sin_addr.s_addr);
 306         } else if (ss_family == AF_INET6) {
 307                 struct sockaddr_in6 *sin6;
 308
 309                 sin6 = (struct sockaddr_in6 *)sockaddr;
 310                 ndev = cxgbit_ipv6_netdev(&sin6->sin6_addr);
 311         }
 312         if (!ndev)
 313                 goto out;
 314
 315         cdev = cxgbit_find_device(ndev, NULL);
 316 out:
 317         rcu_read_unlock();
 318         return cdev;
 319 }
 320
 321 static bool cxgbit_inaddr_any(struct cxgbit_np *cnp)
 322 {
 323         struct sockaddr_storage *sockaddr = &cnp->com.local_addr;
 324         int ss_family = sockaddr->ss_family;
 325         int addr_type;
 326
 327         if (ss_family == AF_INET) {
 328                 struct sockaddr_in *sin;
 329
 330                 sin = (struct sockaddr_in *)sockaddr;
 331                 if (sin->sin_addr.s_addr == htonl(INADDR_ANY))
 332                         return true;
 333         } else if (ss_family == AF_INET6) {
 334                 struct sockaddr_in6 *sin6;
 335
 336                 sin6 = (struct sockaddr_in6 *)sockaddr;
 337                 addr_type = ipv6_addr_type((const struct in6_addr *)
 338                                 &sin6->sin6_addr);
 339                 if (addr_type == IPV6_ADDR_ANY)
 340                         return true;
 341         }
 342         return false;
 343 }
 344
 345 static int
 346 __cxgbit_setup_cdev_np(struct cxgbit_device *cdev, struct cxgbit_np *cnp)
 347 {
 348         int stid, ret;
 349         int ss_family = cnp->com.local_addr.ss_family;
 350
 351         if (!test_bit(CDEV_STATE_UP, &cdev->flags))
 352                 return -EINVAL;
 353
 354         stid = cxgb4_alloc_stid(cdev->lldi.tids, ss_family, cnp);
 355         if (stid < 0)
 356                 return -EINVAL;
 357
 358         if (!cxgbit_np_hash_add(cdev, cnp, stid)) {
 359                 cxgb4_free_stid(cdev->lldi.tids, stid, ss_family);
 360                 return -EINVAL;
 361         }
 362
 363         if (ss_family == AF_INET)
 364                 ret = cxgbit_create_server4(cdev, stid, cnp);
 365         else
 366                 ret = cxgbit_create_server6(cdev, stid, cnp);
 367
 368         if (ret) {
 369                 if (ret != -ETIMEDOUT)
 370                         cxgb4_free_stid(cdev->lldi.tids, stid,
 371                                         ss_family);
 372                 cxgbit_np_hash_del(cdev, cnp);
 373                 return ret;
 374         }
 375         return ret;
 376 }
 377
 378 static int cxgbit_setup_cdev_np(struct cxgbit_np *cnp)
 379 {
 380         struct cxgbit_device *cdev;
 381         int ret = -1;
 382
 383         mutex_lock(&cdev_list_lock);
 384         cdev = cxgbit_find_np_cdev(cnp);
 385         if (!cdev)
 386                 goto out;
 387
 388         if (cxgbit_np_hash_find(cdev, cnp) >= 0)
 389                 goto out;
 390
 391         if (__cxgbit_setup_cdev_np(cdev, cnp))
 392                 goto out;
 393
 394         cnp->com.cdev = cdev;
 395         ret = 0;
 396 out:
 397         mutex_unlock(&cdev_list_lock);
 398         return ret;
 399 }
 400
 401 static int cxgbit_setup_all_np(struct cxgbit_np *cnp)
 402 {
 403         struct cxgbit_device *cdev;
 404         int ret;
 405         u32 count = 0;
 406
 407         mutex_lock(&cdev_list_lock);
 408         list_for_each_entry(cdev, &cdev_list_head, list) {
 409                 if (cxgbit_np_hash_find(cdev, cnp) >= 0) {
 410                         mutex_unlock(&cdev_list_lock);
 411                         return -1;
 412                 }
 413         }
 414
 415         list_for_each_entry(cdev, &cdev_list_head, list) {
 416                 ret = __cxgbit_setup_cdev_np(cdev, cnp);
 417                 if (ret == -ETIMEDOUT)
 418                         break;
 419                 if (ret != 0)
 420                         continue;
 421                 count++;
 422         }
 423         mutex_unlock(&cdev_list_lock);
 424
 425         return count ? 0 : -1;
 426 }
 427
 428 int cxgbit_setup_np(struct iscsi_np *np, struct sockaddr_storage *ksockaddr)
 429 {
 430         struct cxgbit_np *cnp;
 431         int ret;
 432
 433         if ((ksockaddr->ss_family != AF_INET) &&
 434             (ksockaddr->ss_family != AF_INET6))
 435                 return -EINVAL;
 436
 437         cnp = kzalloc(sizeof(*cnp), GFP_KERNEL);
 438         if (!cnp)
 439                 return -ENOMEM;
 440
 441         init_waitqueue_head(&cnp->accept_wait);
 442         init_completion(&cnp->com.wr_wait.completion);
 443         init_completion(&cnp->accept_comp);
 444         INIT_LIST_HEAD(&cnp->np_accept_list);
 445         spin_lock_init(&cnp->np_accept_lock);
 446         kref_init(&cnp->kref);
 447         memcpy(&np->np_sockaddr, ksockaddr,
 448                sizeof(struct sockaddr_storage));
 449         memcpy(&cnp->com.local_addr, &np->np_sockaddr,
 450                sizeof(cnp->com.local_addr));
 451
 452         cnp->np = np;
 453         cnp->com.cdev = NULL;
 454
 455         if (cxgbit_inaddr_any(cnp))
 456                 ret = cxgbit_setup_all_np(cnp);
 457         else
 458                 ret = cxgbit_setup_cdev_np(cnp);
 459
 460         if (ret) {
 461                 cxgbit_put_cnp(cnp);
 462                 return -EINVAL;
 463         }
 464
 465         np->np_context = cnp;
 466         cnp->com.state = CSK_STATE_LISTEN;
 467         return 0;
 468 }
 469
 470 static void
 471 cxgbit_set_conn_info(struct iscsi_np *np, struct iscsi_conn *conn,
 472                      struct cxgbit_sock *csk)
 473 {
 474         conn->login_family = np->np_sockaddr.ss_family;
 475         conn->login_sockaddr = csk->com.remote_addr;
 476         conn->local_sockaddr = csk->com.local_addr;
 477 }
 478
 479 int cxgbit_accept_np(struct iscsi_np *np, struct iscsi_conn *conn)
 480 {
 481         struct cxgbit_np *cnp = np->np_context;
 482         struct cxgbit_sock *csk;
 483         int ret = 0;
 484
 485 accept_wait:
 486         ret = wait_for_completion_interruptible(&cnp->accept_comp);
 487         if (ret)
 488                 return -ENODEV;
 489
 490         spin_lock_bh(&np->np_thread_lock);
 491         if (np->np_thread_state >= ISCSI_NP_THREAD_RESET) {
 492                 spin_unlock_bh(&np->np_thread_lock);
 493                 /**
 494                  * No point in stalling here when np_thread
 495                  * is in state RESET/SHUTDOWN/EXIT - bail
 496                  **/
 497                 return -ENODEV;
 498         }
 499         spin_unlock_bh(&np->np_thread_lock);
 500
 501         spin_lock_bh(&cnp->np_accept_lock);
 502         if (list_empty(&cnp->np_accept_list)) {
 503                 spin_unlock_bh(&cnp->np_accept_lock);
 504                 goto accept_wait;
 505         }
 506
 507         csk = list_first_entry(&cnp->np_accept_list,
 508                                struct cxgbit_sock,
 509                                accept_node);
 510
 511         list_del_init(&csk->accept_node);
 512         spin_unlock_bh(&cnp->np_accept_lock);
 513         conn->context = csk;
 514         csk->conn = conn;
 515
 516         cxgbit_set_conn_info(np, conn, csk);
 517         return 0;
 518 }
 519
 520 static int
 521 __cxgbit_free_cdev_np(struct cxgbit_device *cdev, struct cxgbit_np *cnp)
 522 {
 523         int stid, ret;
 524         bool ipv6 = false;
 525
 526         stid = cxgbit_np_hash_del(cdev, cnp);
 527         if (stid < 0)
 528                 return -EINVAL;
 529         if (!test_bit(CDEV_STATE_UP, &cdev->flags))
 530                 return -EINVAL;
 531
 532         if (cnp->np->np_sockaddr.ss_family == AF_INET6)
 533                 ipv6 = true;
 534
 535         cxgbit_get_cnp(cnp);
 536         cxgbit_init_wr_wait(&cnp->com.wr_wait);
 537         ret = cxgb4_remove_server(cdev->lldi.ports[0], stid,
 538                                   cdev->lldi.rxq_ids[0], ipv6);
 539
 540         if (ret > 0)
 541                 ret = net_xmit_errno(ret);
 542
 543         if (ret) {
 544                 cxgbit_put_cnp(cnp);
 545                 return ret;
 546         }
 547
 548         ret = cxgbit_wait_for_reply(cdev, &cnp->com.wr_wait,
 549                                     0, 10, __func__);
 550         if (ret == -ETIMEDOUT)
 551                 return ret;
 552
 553         if (ipv6 && cnp->com.cdev) {
 554                 struct sockaddr_in6 *sin6;
 555
 556                 sin6 = (struct sockaddr_in6 *)&cnp->com.local_addr;
 557                 cxgb4_clip_release(cdev->lldi.ports[0],
 558                                    (const u32 *)&sin6->sin6_addr.s6_addr,
 559                                    1);
 560         }
 561
 562         cxgb4_free_stid(cdev->lldi.tids, stid,
 563                         cnp->com.local_addr.ss_family);
 564         return 0;
 565 }
 566
 567 static void cxgbit_free_all_np(struct cxgbit_np *cnp)
 568 {
 569         struct cxgbit_device *cdev;
 570         int ret;
 571
 572         mutex_lock(&cdev_list_lock);
 573         list_for_each_entry(cdev, &cdev_list_head, list) {
 574                 ret = __cxgbit_free_cdev_np(cdev, cnp);
 575                 if (ret == -ETIMEDOUT)
 576                         break;
 577         }
 578         mutex_unlock(&cdev_list_lock);
 579 }
 580
 581 static void cxgbit_free_cdev_np(struct cxgbit_np *cnp)
 582 {
 583         struct cxgbit_device *cdev;
 584         bool found = false;
 585
 586         mutex_lock(&cdev_list_lock);
 587         list_for_each_entry(cdev, &cdev_list_head, list) {
 588                 if (cdev == cnp->com.cdev) {
 589                         found = true;
 590                         break;
 591                 }
 592         }
 593         if (!found)
 594                 goto out;
 595
 596         __cxgbit_free_cdev_np(cdev, cnp);
 597 out:
 598         mutex_unlock(&cdev_list_lock);
 599 }
 600
 601 static void __cxgbit_free_conn(struct cxgbit_sock *csk);
 602
 603 void cxgbit_free_np(struct iscsi_np *np)
 604 {
 605         struct cxgbit_np *cnp = np->np_context;
 606         struct cxgbit_sock *csk, *tmp;
 607
 608         cnp->com.state = CSK_STATE_DEAD;
 609         if (cnp->com.cdev)
 610                 cxgbit_free_cdev_np(cnp);
 611         else
 612                 cxgbit_free_all_np(cnp);
 613
 614         spin_lock_bh(&cnp->np_accept_lock);
 615         list_for_each_entry_safe(csk, tmp, &cnp->np_accept_list, accept_node) {
 616                 list_del_init(&csk->accept_node);
 617                 __cxgbit_free_conn(csk);
 618         }
 619         spin_unlock_bh(&cnp->np_accept_lock);
 620
 621         np->np_context = NULL;
 622         cxgbit_put_cnp(cnp);
 623 }
 624
 625 static void cxgbit_send_halfclose(struct cxgbit_sock *csk)
 626 {
 627         struct sk_buff *skb;
 628         u32 len = roundup(sizeof(struct cpl_close_con_req), 16);
 629
 630         skb = alloc_skb(len, GFP_ATOMIC);
 631         if (!skb)
 632                 return;
 633
 634         cxgb_mk_close_con_req(skb, len, csk->tid, csk->txq_idx,
 635                               NULL, NULL);
 636
 637         cxgbit_skcb_flags(skb) |= SKCBF_TX_FLAG_COMPL;
 638         __skb_queue_tail(&csk->txq, skb);
 639         cxgbit_push_tx_frames(csk);
 640 }
 641
 642 static void cxgbit_arp_failure_discard(void *handle, struct sk_buff *skb)
 643 {
 644         struct cxgbit_sock *csk = handle;
 645
 646         pr_debug("%s cxgbit_device %p\n", __func__, handle);
 647         kfree_skb(skb);
 648         cxgbit_put_csk(csk);
 649 }
 650
 651 static void cxgbit_abort_arp_failure(void *handle, struct sk_buff *skb)
 652 {
 653         struct cxgbit_device *cdev = handle;
 654         struct cpl_abort_req *req = cplhdr(skb);
 655
 656         pr_debug("%s cdev %p\n", __func__, cdev);
 657         req->cmd = CPL_ABORT_NO_RST;
 658         cxgbit_ofld_send(cdev, skb);
 659 }
 660
 661 static int cxgbit_send_abort_req(struct cxgbit_sock *csk)
 662 {
 663         struct sk_buff *skb;
 664         u32 len = roundup(sizeof(struct cpl_abort_req), 16);
 665
 666         pr_debug("%s: csk %p tid %u; state %d\n",
 667                  __func__, csk, csk->tid, csk->com.state);
 668
 669         __skb_queue_purge(&csk->txq);
 670
 671         if (!test_and_set_bit(CSK_TX_DATA_SENT, &csk->com.flags))
 672                 cxgbit_send_tx_flowc_wr(csk);
 673
 674         skb = __skb_dequeue(&csk->skbq);
 675         cxgb_mk_abort_req(skb, len, csk->tid, csk->txq_idx,
 676                           csk->com.cdev, cxgbit_abort_arp_failure);
 677
 678         return cxgbit_l2t_send(csk->com.cdev, skb, csk->l2t);
 679 }
 680
 681 static void
 682 __cxgbit_abort_conn(struct cxgbit_sock *csk, struct sk_buff *skb)
 683 {
 684         __kfree_skb(skb);
 685
 686         if (csk->com.state != CSK_STATE_ESTABLISHED)
 687                 goto no_abort;
 688
 689         set_bit(CSK_ABORT_RPL_WAIT, &csk->com.flags);
 690         csk->com.state = CSK_STATE_ABORTING;
 691
 692         cxgbit_send_abort_req(csk);
 693
 694         return;
 695
 696 no_abort:
 697         cxgbit_wake_up(&csk->com.wr_wait, __func__, CPL_ERR_NONE);
 698         cxgbit_put_csk(csk);
 699 }
 700
 701 void cxgbit_abort_conn(struct cxgbit_sock *csk)
 702 {
 703         struct sk_buff *skb = alloc_skb(0, GFP_KERNEL | __GFP_NOFAIL);
 704
 705         cxgbit_get_csk(csk);
 706         cxgbit_init_wr_wait(&csk->com.wr_wait);
 707
 708         spin_lock_bh(&csk->lock);
 709         if (csk->lock_owner) {
 710                 cxgbit_skcb_rx_backlog_fn(skb) = __cxgbit_abort_conn;
 711                 __skb_queue_tail(&csk->backlogq, skb);
 712         } else {
 713                 __cxgbit_abort_conn(csk, skb);
 714         }
 715         spin_unlock_bh(&csk->lock);
 716
 717         cxgbit_wait_for_reply(csk->com.cdev, &csk->com.wr_wait,
 718                               csk->tid, 600, __func__);
 719 }
 720
 721 static void __cxgbit_free_conn(struct cxgbit_sock *csk)
 722 {
 723         struct iscsi_conn *conn = csk->conn;
 724         bool release = false;
 725
 726         pr_debug("%s: state %d\n",
 727                  __func__, csk->com.state);
 728
 729         spin_lock_bh(&csk->lock);
 730         switch (csk->com.state) {
 731         case CSK_STATE_ESTABLISHED:
 732                 if (conn && (conn->conn_state == TARG_CONN_STATE_IN_LOGOUT)) {
 733                         csk->com.state = CSK_STATE_CLOSING;
 734                         cxgbit_send_halfclose(csk);
 735                 } else {
 736                         csk->com.state = CSK_STATE_ABORTING;
 737                         cxgbit_send_abort_req(csk);
 738                 }
 739                 break;
 740         case CSK_STATE_CLOSING:
 741                 csk->com.state = CSK_STATE_MORIBUND;
 742                 cxgbit_send_halfclose(csk);
 743                 break;
 744         case CSK_STATE_DEAD:
 745                 release = true;
 746                 break;
 747         default:
 748                 pr_err("%s: csk %p; state %d\n",
 749                        __func__, csk, csk->com.state);
 750         }
 751         spin_unlock_bh(&csk->lock);
 752
 753         if (release)
 754                 cxgbit_put_csk(csk);
 755 }
 756
 757 void cxgbit_free_conn(struct iscsi_conn *conn)
 758 {
 759         __cxgbit_free_conn(conn->context);
 760 }
 761
 762 static void cxgbit_set_emss(struct cxgbit_sock *csk, u16 opt)
 763 {
 764         csk->emss = csk->com.cdev->lldi.mtus[TCPOPT_MSS_G(opt)] -
 765                         ((csk->com.remote_addr.ss_family == AF_INET) ?
 766                         sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
 767                         sizeof(struct tcphdr);
 768         csk->mss = csk->emss;
 769         if (TCPOPT_TSTAMP_G(opt))
 770                 csk->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
 771         if (csk->emss < 128)
 772                 csk->emss = 128;
 773         if (csk->emss & 7)
 774                 pr_info("Warning: misaligned mtu idx %u mss %u emss=%u\n",
 775                         TCPOPT_MSS_G(opt), csk->mss, csk->emss);
 776         pr_debug("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
 777                  csk->mss, csk->emss);
 778 }
 779
 780 static void cxgbit_free_skb(struct cxgbit_sock *csk)
 781 {
 782         struct sk_buff *skb;
 783
 784         __skb_queue_purge(&csk->txq);
 785         __skb_queue_purge(&csk->rxq);
 786         __skb_queue_purge(&csk->backlogq);
 787         __skb_queue_purge(&csk->ppodq);
 788         __skb_queue_purge(&csk->skbq);
 789
 790         while ((skb = cxgbit_sock_dequeue_wr(csk)))
 791                 kfree_skb(skb);
 792
 793         __kfree_skb(csk->lro_hskb);
 794 }
 795
 796 void _cxgbit_free_csk(struct kref *kref)
 797 {
 798         struct cxgbit_sock *csk;
 799         struct cxgbit_device *cdev;
 800
 801         csk = container_of(kref, struct cxgbit_sock, kref);
 802
 803         pr_debug("%s csk %p state %d\n", __func__, csk, csk->com.state);
 804
 805         if (csk->com.local_addr.ss_family == AF_INET6) {
 806                 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
 807                                              &csk->com.local_addr;
 808                 cxgb4_clip_release(csk->com.cdev->lldi.ports[0],
 809                                    (const u32 *)
 810                                    &sin6->sin6_addr.s6_addr, 1);
 811         }
 812
 813         cxgb4_remove_tid(csk->com.cdev->lldi.tids, 0, csk->tid,
 814                          csk->com.local_addr.ss_family);
 815         dst_release(csk->dst);
 816         cxgb4_l2t_release(csk->l2t);
 817
 818         cdev = csk->com.cdev;
 819         spin_lock_bh(&cdev->cskq.lock);
 820         list_del(&csk->list);
 821         spin_unlock_bh(&cdev->cskq.lock);
 822
 823         cxgbit_free_skb(csk);
 824         cxgbit_put_cnp(csk->cnp);
 825         cxgbit_put_cdev(cdev);
 826
 827         kfree(csk);
 828 }
 829
 830 static void cxgbit_set_tcp_window(struct cxgbit_sock *csk, struct port_info *pi)
 831 {
 832         unsigned int linkspeed;
 833         u8 scale;
 834
 835         linkspeed = pi->link_cfg.speed;
 836         scale = linkspeed / SPEED_10000;
 837
 838 #define CXGBIT_10G_RCV_WIN (256 * 1024)
 839         csk->rcv_win = CXGBIT_10G_RCV_WIN;
 840         if (scale)
 841                 csk->rcv_win *= scale;
 842
 843 #define CXGBIT_10G_SND_WIN (256 * 1024)
 844         csk->snd_win = CXGBIT_10G_SND_WIN;
 845         if (scale)
 846                 csk->snd_win *= scale;
 847
 848         pr_debug("%s snd_win %d rcv_win %d\n",
 849                  __func__, csk->snd_win, csk->rcv_win);
 850 }
 851
 852 #ifdef CONFIG_CHELSIO_T4_DCB
 853 static u8 cxgbit_get_iscsi_dcb_state(struct net_device *ndev)
 854 {
 855         return ndev->dcbnl_ops->getstate(ndev);
 856 }
 857
 858 static int cxgbit_select_priority(int pri_mask)
 859 {
 860         if (!pri_mask)
 861                 return 0;
 862
 863         return (ffs(pri_mask) - 1);
 864 }
 865
 866 static u8 cxgbit_get_iscsi_dcb_priority(struct net_device *ndev, u16 local_port)
 867 {
 868         int ret;
 869         u8 caps;
 870
 871         struct dcb_app iscsi_dcb_app = {
 872                 .protocol = local_port
 873         };
 874
 875         ret = (int)ndev->dcbnl_ops->getcap(ndev, DCB_CAP_ATTR_DCBX, &caps);
 876
 877         if (ret)
 878                 return 0;
 879
 880         if (caps & DCB_CAP_DCBX_VER_IEEE) {
 881                 iscsi_dcb_app.selector = IEEE_8021QAZ_APP_SEL_ANY;
 882
 883                 ret = dcb_ieee_getapp_mask(ndev, &iscsi_dcb_app);
 884
 885         } else if (caps & DCB_CAP_DCBX_VER_CEE) {
 886                 iscsi_dcb_app.selector = DCB_APP_IDTYPE_PORTNUM;
 887
 888                 ret = dcb_getapp(ndev, &iscsi_dcb_app);
 889         }
 890
 891         pr_info("iSCSI priority is set to %u\n", cxgbit_select_priority(ret));
 892
 893         return cxgbit_select_priority(ret);
 894 }
 895 #endif
 896
 897 static int
 898 cxgbit_offload_init(struct cxgbit_sock *csk, int iptype, __u8 *peer_ip,
 899                     u16 local_port, struct dst_entry *dst,
 900                     struct cxgbit_device *cdev)
 901 {
 902         struct neighbour *n;
 903         int ret, step;
 904         struct net_device *ndev;
 905         u16 rxq_idx, port_id;
 906 #ifdef CONFIG_CHELSIO_T4_DCB
 907         u8 priority = 0;
 908 #endif
 909
 910         n = dst_neigh_lookup(dst, peer_ip);
 911         if (!n)
 912                 return -ENODEV;
 913
 914         rcu_read_lock();
 915         if (!(n->nud_state & NUD_VALID))
 916                 neigh_event_send(n, NULL);
 917
 918         ret = -ENOMEM;
 919         if (n->dev->flags & IFF_LOOPBACK) {
 920                 if (iptype == 4)
 921                         ndev = cxgbit_ipv4_netdev(*(__be32 *)peer_ip);
 922                 else if (IS_ENABLED(CONFIG_IPV6))
 923                         ndev = cxgbit_ipv6_netdev((struct in6_addr *)peer_ip);
 924                 else
 925                         ndev = NULL;
 926
 927                 if (!ndev) {
 928                         ret = -ENODEV;
 929                         goto out;
 930                 }
 931
 932                 csk->l2t = cxgb4_l2t_get(cdev->lldi.l2t,
 933                                          n, ndev, 0);
 934                 if (!csk->l2t)
 935                         goto out;
 936                 csk->mtu = ndev->mtu;
 937                 csk->tx_chan = cxgb4_port_chan(ndev);
 938                 csk->smac_idx = cxgb4_tp_smt_idx(cdev->lldi.adapter_type,
 939                                                  cxgb4_port_viid(ndev));
 940                 step = cdev->lldi.ntxq /
 941                         cdev->lldi.nchan;
 942                 csk->txq_idx = cxgb4_port_idx(ndev) * step;
 943                 step = cdev->lldi.nrxq /
 944                         cdev->lldi.nchan;
 945                 csk->ctrlq_idx = cxgb4_port_idx(ndev);
 946                 csk->rss_qid = cdev->lldi.rxq_ids[
 947                                 cxgb4_port_idx(ndev) * step];
 948                 csk->port_id = cxgb4_port_idx(ndev);
 949                 cxgbit_set_tcp_window(csk,
 950                                       (struct port_info *)netdev_priv(ndev));
 951         } else {
 952                 ndev = cxgbit_get_real_dev(n->dev);
 953                 if (!ndev) {
 954                         ret = -ENODEV;
 955                         goto out;
 956                 }
 957
 958 #ifdef CONFIG_CHELSIO_T4_DCB
 959                 if (cxgbit_get_iscsi_dcb_state(ndev))
 960                         priority = cxgbit_get_iscsi_dcb_priority(ndev,
 961                                                                  local_port);
 962
 963                 csk->dcb_priority = priority;
 964
 965                 csk->l2t = cxgb4_l2t_get(cdev->lldi.l2t, n, ndev, priority);
 966 #else
 967                 csk->l2t = cxgb4_l2t_get(cdev->lldi.l2t, n, ndev, 0);
 968 #endif
 969                 if (!csk->l2t)
 970                         goto out;
 971                 port_id = cxgb4_port_idx(ndev);
 972                 csk->mtu = dst_mtu(dst);
 973                 csk->tx_chan = cxgb4_port_chan(ndev);
 974                 csk->smac_idx = cxgb4_tp_smt_idx(cdev->lldi.adapter_type,
 975                                                  cxgb4_port_viid(ndev));
 976                 step = cdev->lldi.ntxq /
 977                         cdev->lldi.nports;
 978                 csk->txq_idx = (port_id * step) +
 979                                 (cdev->selectq[port_id][0]++ % step);
 980                 csk->ctrlq_idx = cxgb4_port_idx(ndev);
 981                 step = cdev->lldi.nrxq /
 982                         cdev->lldi.nports;
 983                 rxq_idx = (port_id * step) +
 984                                 (cdev->selectq[port_id][1]++ % step);
 985                 csk->rss_qid = cdev->lldi.rxq_ids[rxq_idx];
 986                 csk->port_id = port_id;
 987                 cxgbit_set_tcp_window(csk,
 988                                       (struct port_info *)netdev_priv(ndev));
 989         }
 990         ret = 0;
 991 out:
 992         rcu_read_unlock();
 993         neigh_release(n);
 994         return ret;
 995 }
 996
 997 int cxgbit_ofld_send(struct cxgbit_device *cdev, struct sk_buff *skb)
 998 {
 999         int ret = 0;
1000
1001         if (!test_bit(CDEV_STATE_UP, &cdev->flags)) {
1002                 kfree_skb(skb);
1003                 pr_err("%s - device not up - dropping\n", __func__);
1004                 return -EIO;
1005         }
1006
1007         ret = cxgb4_ofld_send(cdev->lldi.ports[0], skb);
1008         if (ret < 0)
1009                 kfree_skb(skb);
1010         return ret < 0 ? ret : 0;
1011 }
1012
1013 static void cxgbit_release_tid(struct cxgbit_device *cdev, u32 tid)
1014 {
1015         u32 len = roundup(sizeof(struct cpl_tid_release), 16);
1016         struct sk_buff *skb;
1017
1018         skb = alloc_skb(len, GFP_ATOMIC);
1019         if (!skb)
1020                 return;
1021
1022         cxgb_mk_tid_release(skb, len, tid, 0);
1023         cxgbit_ofld_send(cdev, skb);
1024 }
1025
1026 int
1027 cxgbit_l2t_send(struct cxgbit_device *cdev, struct sk_buff *skb,
1028                 struct l2t_entry *l2e)
1029 {
1030         int ret = 0;
1031
1032         if (!test_bit(CDEV_STATE_UP, &cdev->flags)) {
1033                 kfree_skb(skb);
1034                 pr_err("%s - device not up - dropping\n", __func__);
1035                 return -EIO;
1036         }
1037
1038         ret = cxgb4_l2t_send(cdev->lldi.ports[0], skb, l2e);
1039         if (ret < 0)
1040                 kfree_skb(skb);
1041         return ret < 0 ? ret : 0;
1042 }
1043
1044 static void cxgbit_send_rx_credits(struct cxgbit_sock *csk, struct sk_buff *skb)
1045 {
1046         if (csk->com.state != CSK_STATE_ESTABLISHED) {
1047                 __kfree_skb(skb);
1048                 return;
1049         }
1050
1051         cxgbit_ofld_send(csk->com.cdev, skb);
1052 }
1053
1054 /*
1055  * CPL connection rx data ack: host ->
1056  * Send RX credits through an RX_DATA_ACK CPL message.
1057  * Returns the number of credits sent.
1058  */
1059 int cxgbit_rx_data_ack(struct cxgbit_sock *csk)
1060 {
1061         struct sk_buff *skb;
1062         u32 len = roundup(sizeof(struct cpl_rx_data_ack), 16);
1063         u32 credit_dack;
1064
1065         skb = alloc_skb(len, GFP_KERNEL);
1066         if (!skb)
1067                 return -1;
1068
1069         credit_dack = RX_DACK_CHANGE_F | RX_DACK_MODE_V(1) |
1070                       RX_CREDITS_V(csk->rx_credits);
1071
1072         cxgb_mk_rx_data_ack(skb, len, csk->tid, csk->ctrlq_idx,
1073                             credit_dack);
1074
1075         csk->rx_credits = 0;
1076
1077         spin_lock_bh(&csk->lock);
1078         if (csk->lock_owner) {
1079                 cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_send_rx_credits;
1080                 __skb_queue_tail(&csk->backlogq, skb);
1081                 spin_unlock_bh(&csk->lock);
1082                 return 0;
1083         }
1084
1085         cxgbit_send_rx_credits(csk, skb);
1086         spin_unlock_bh(&csk->lock);
1087
1088         return 0;
1089 }
1090
1091 #define FLOWC_WR_NPARAMS_MIN    9
1092 #define FLOWC_WR_NPARAMS_MAX    11
1093 static int cxgbit_alloc_csk_skb(struct cxgbit_sock *csk)
1094 {
1095         struct sk_buff *skb;
1096         u32 len, flowclen;
1097         u8 i;
1098
1099         flowclen = offsetof(struct fw_flowc_wr,
1100                             mnemval[FLOWC_WR_NPARAMS_MAX]);
1101
1102         len = max_t(u32, sizeof(struct cpl_abort_req),
1103                     sizeof(struct cpl_abort_rpl));
1104
1105         len = max(len, flowclen);
1106         len = roundup(len, 16);
1107
1108         for (i = 0; i < 3; i++) {
1109                 skb = alloc_skb(len, GFP_ATOMIC);
1110                 if (!skb)
1111                         goto out;
1112                 __skb_queue_tail(&csk->skbq, skb);
1113         }
1114
1115         skb = alloc_skb(LRO_SKB_MIN_HEADROOM, GFP_ATOMIC);
1116         if (!skb)
1117                 goto out;
1118
1119         memset(skb->data, 0, LRO_SKB_MIN_HEADROOM);
1120         csk->lro_hskb = skb;
1121
1122         return 0;
1123 out:
1124         __skb_queue_purge(&csk->skbq);
1125         return -ENOMEM;
1126 }
1127
1128 static void
1129 cxgbit_pass_accept_rpl(struct cxgbit_sock *csk, struct cpl_pass_accept_req *req)
1130 {
1131         struct sk_buff *skb;
1132         const struct tcphdr *tcph;
1133         struct cpl_t5_pass_accept_rpl *rpl5;
1134         struct cxgb4_lld_info *lldi = &csk->com.cdev->lldi;
1135         unsigned int len = roundup(sizeof(*rpl5), 16);
1136         unsigned int mtu_idx;
1137         u64 opt0;
1138         u32 opt2, hlen;
1139         u32 wscale;
1140         u32 win;
1141
1142         pr_debug("%s csk %p tid %u\n", __func__, csk, csk->tid);
1143
1144         skb = alloc_skb(len, GFP_ATOMIC);
1145         if (!skb) {
1146                 cxgbit_put_csk(csk);
1147                 return;
1148         }
1149
1150         rpl5 = __skb_put_zero(skb, len);
1151
1152         INIT_TP_WR(rpl5, csk->tid);
1153         OPCODE_TID(rpl5) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1154                                                      csk->tid));
1155         cxgb_best_mtu(csk->com.cdev->lldi.mtus, csk->mtu, &mtu_idx,
1156                       req->tcpopt.tstamp,
1157                       (csk->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1158         wscale = cxgb_compute_wscale(csk->rcv_win);
1159         /*
1160          * Specify the largest window that will fit in opt0. The
1161          * remainder will be specified in the rx_data_ack.
1162          */
1163         win = csk->rcv_win >> 10;
1164         if (win > RCV_BUFSIZ_M)
1165                 win = RCV_BUFSIZ_M;
1166         opt0 =  TCAM_BYPASS_F |
1167                 WND_SCALE_V(wscale) |
1168                 MSS_IDX_V(mtu_idx) |
1169                 L2T_IDX_V(csk->l2t->idx) |
1170                 TX_CHAN_V(csk->tx_chan) |
1171                 SMAC_SEL_V(csk->smac_idx) |
1172                 DSCP_V(csk->tos >> 2) |
1173                 ULP_MODE_V(ULP_MODE_ISCSI) |
1174                 RCV_BUFSIZ_V(win);
1175
1176         opt2 = RX_CHANNEL_V(0) |
1177                 RSS_QUEUE_VALID_F | RSS_QUEUE_V(csk->rss_qid);
1178
1179         if (!is_t5(lldi->adapter_type))
1180                 opt2 |= RX_FC_DISABLE_F;
1181
1182         if (req->tcpopt.tstamp)
1183                 opt2 |= TSTAMPS_EN_F;
1184         if (req->tcpopt.sack)
1185                 opt2 |= SACK_EN_F;
1186         if (wscale)
1187                 opt2 |= WND_SCALE_EN_F;
1188
1189         hlen = ntohl(req->hdr_len);
1190
1191         if (is_t5(lldi->adapter_type))
1192                 tcph = (struct tcphdr *)((u8 *)(req + 1) +
1193                        ETH_HDR_LEN_G(hlen) + IP_HDR_LEN_G(hlen));
1194         else
1195                 tcph = (struct tcphdr *)((u8 *)(req + 1) +
1196                        T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen));
1197
1198         if (tcph->ece && tcph->cwr)
1199                 opt2 |= CCTRL_ECN_V(1);
1200
1201         opt2 |= RX_COALESCE_V(3);
1202         opt2 |= CONG_CNTRL_V(CONG_ALG_NEWRENO);
1203
1204         opt2 |= T5_ISS_F;
1205         rpl5->iss = cpu_to_be32((prandom_u32() & ~7UL) - 1);
1206
1207         opt2 |= T5_OPT_2_VALID_F;
1208
1209         rpl5->opt0 = cpu_to_be64(opt0);
1210         rpl5->opt2 = cpu_to_be32(opt2);
1211         set_wr_txq(skb, CPL_PRIORITY_SETUP, csk->ctrlq_idx);
1212         t4_set_arp_err_handler(skb, csk, cxgbit_arp_failure_discard);
1213         cxgbit_l2t_send(csk->com.cdev, skb, csk->l2t);
1214 }
1215
1216 static void
1217 cxgbit_pass_accept_req(struct cxgbit_device *cdev, struct sk_buff *skb)
1218 {
1219         struct cxgbit_sock *csk = NULL;
1220         struct cxgbit_np *cnp;
1221         struct cpl_pass_accept_req *req = cplhdr(skb);
1222         unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1223         struct tid_info *t = cdev->lldi.tids;
1224         unsigned int tid = GET_TID(req);
1225         u16 peer_mss = ntohs(req->tcpopt.mss);
1226         unsigned short hdrs;
1227
1228         struct dst_entry *dst;
1229         __u8 local_ip[16], peer_ip[16];
1230         __be16 local_port, peer_port;
1231         int ret;
1232         int iptype;
1233
1234         pr_debug("%s: cdev = %p; stid = %u; tid = %u\n",
1235                  __func__, cdev, stid, tid);
1236
1237         cnp = lookup_stid(t, stid);
1238         if (!cnp) {
1239                 pr_err("%s connect request on invalid stid %d\n",
1240                        __func__, stid);
1241                 goto rel_skb;
1242         }
1243
1244         if (cnp->com.state != CSK_STATE_LISTEN) {
1245                 pr_err("%s - listening parent not in CSK_STATE_LISTEN\n",
1246                        __func__);
1247                 goto reject;
1248         }
1249
1250         csk = lookup_tid(t, tid);
1251         if (csk) {
1252                 pr_err("%s csk not null tid %u\n",
1253                        __func__, tid);
1254                 goto rel_skb;
1255         }
1256
1257         cxgb_get_4tuple(req, cdev->lldi.adapter_type, &iptype, local_ip,
1258                         peer_ip, &local_port, &peer_port);
1259
1260         /* Find output route */
1261         if (iptype == 4)  {
1262                 pr_debug("%s parent sock %p tid %u laddr %pI4 raddr %pI4 "
1263                          "lport %d rport %d peer_mss %d\n"
1264                          , __func__, cnp, tid,
1265                          local_ip, peer_ip, ntohs(local_port),
1266                          ntohs(peer_port), peer_mss);
1267                 dst = cxgb_find_route(&cdev->lldi, cxgbit_get_real_dev,
1268                                       *(__be32 *)local_ip,
1269                                       *(__be32 *)peer_ip,
1270                                       local_port, peer_port,
1271                                       PASS_OPEN_TOS_G(ntohl(req->tos_stid)));
1272         } else {
1273                 pr_debug("%s parent sock %p tid %u laddr %pI6 raddr %pI6 "
1274                          "lport %d rport %d peer_mss %d\n"
1275                          , __func__, cnp, tid,
1276                          local_ip, peer_ip, ntohs(local_port),
1277                          ntohs(peer_port), peer_mss);
1278                 dst = cxgb_find_route6(&cdev->lldi, cxgbit_get_real_dev,
1279                                        local_ip, peer_ip,
1280                                        local_port, peer_port,
1281                                        PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
1282                                        ((struct sockaddr_in6 *)
1283                                         &cnp->com.local_addr)->sin6_scope_id);
1284         }
1285         if (!dst) {
1286                 pr_err("%s - failed to find dst entry!\n",
1287                        __func__);
1288                 goto reject;
1289         }
1290
1291         csk = kzalloc(sizeof(*csk), GFP_ATOMIC);
1292         if (!csk) {
1293                 dst_release(dst);
1294                 goto rel_skb;
1295         }
1296
1297         ret = cxgbit_offload_init(csk, iptype, peer_ip, ntohs(local_port),
1298                                   dst, cdev);
1299         if (ret) {
1300                 pr_err("%s - failed to allocate l2t entry!\n",
1301                        __func__);
1302                 dst_release(dst);
1303                 kfree(csk);
1304                 goto reject;
1305         }
1306
1307         kref_init(&csk->kref);
1308         init_completion(&csk->com.wr_wait.completion);
1309
1310         INIT_LIST_HEAD(&csk->accept_node);
1311
1312         hdrs = (iptype == 4 ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
1313                 sizeof(struct tcphdr) + (req->tcpopt.tstamp ? 12 : 0);
1314         if (peer_mss && csk->mtu > (peer_mss + hdrs))
1315                 csk->mtu = peer_mss + hdrs;
1316
1317         csk->com.state = CSK_STATE_CONNECTING;
1318         csk->com.cdev = cdev;
1319         csk->cnp = cnp;
1320         csk->tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
1321         csk->dst = dst;
1322         csk->tid = tid;
1323         csk->wr_cred = cdev->lldi.wr_cred -
1324                         DIV_ROUND_UP(sizeof(struct cpl_abort_req), 16);
1325         csk->wr_max_cred = csk->wr_cred;
1326         csk->wr_una_cred = 0;
1327
1328         if (iptype == 4) {
1329                 struct sockaddr_in *sin = (struct sockaddr_in *)
1330                                           &csk->com.local_addr;
1331                 sin->sin_family = AF_INET;
1332                 sin->sin_port = local_port;
1333                 sin->sin_addr.s_addr = *(__be32 *)local_ip;
1334
1335                 sin = (struct sockaddr_in *)&csk->com.remote_addr;
1336                 sin->sin_family = AF_INET;
1337                 sin->sin_port = peer_port;
1338                 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
1339         } else {
1340                 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
1341                                             &csk->com.local_addr;
1342
1343                 sin6->sin6_family = PF_INET6;
1344                 sin6->sin6_port = local_port;
1345                 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
1346                 cxgb4_clip_get(cdev->lldi.ports[0],
1347                                (const u32 *)&sin6->sin6_addr.s6_addr,
1348                                1);
1349
1350                 sin6 = (struct sockaddr_in6 *)&csk->com.remote_addr;
1351                 sin6->sin6_family = PF_INET6;
1352                 sin6->sin6_port = peer_port;
1353                 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
1354         }
1355
1356         skb_queue_head_init(&csk->rxq);
1357         skb_queue_head_init(&csk->txq);
1358         skb_queue_head_init(&csk->ppodq);
1359         skb_queue_head_init(&csk->backlogq);
1360         skb_queue_head_init(&csk->skbq);
1361         cxgbit_sock_reset_wr_list(csk);
1362         spin_lock_init(&csk->lock);
1363         init_waitqueue_head(&csk->waitq);
1364         init_waitqueue_head(&csk->ack_waitq);
1365         csk->lock_owner = false;
1366
1367         if (cxgbit_alloc_csk_skb(csk)) {
1368                 dst_release(dst);
1369                 kfree(csk);
1370                 goto rel_skb;
1371         }
1372
1373         cxgbit_get_cnp(cnp);
1374         cxgbit_get_cdev(cdev);
1375
1376         spin_lock(&cdev->cskq.lock);
1377         list_add_tail(&csk->list, &cdev->cskq.list);
1378         spin_unlock(&cdev->cskq.lock);
1379         cxgb4_insert_tid(t, csk, tid, csk->com.local_addr.ss_family);
1380         cxgbit_pass_accept_rpl(csk, req);
1381         goto rel_skb;
1382
1383 reject:
1384         cxgbit_release_tid(cdev, tid);
1385 rel_skb:
1386         __kfree_skb(skb);
1387 }
1388
1389 static u32
1390 cxgbit_tx_flowc_wr_credits(struct cxgbit_sock *csk, u32 *nparamsp,
1391                            u32 *flowclenp)
1392 {
1393         u32 nparams, flowclen16, flowclen;
1394
1395         nparams = FLOWC_WR_NPARAMS_MIN;
1396
1397         if (csk->snd_wscale)
1398                 nparams++;
1399
1400 #ifdef CONFIG_CHELSIO_T4_DCB
1401         nparams++;
1402 #endif
1403         flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]);
1404         flowclen16 = DIV_ROUND_UP(flowclen, 16);
1405         flowclen = flowclen16 * 16;
1406         /*
1407          * Return the number of 16-byte credits used by the flowc request.
1408          * Pass back the nparams and actual flowc length if requested.
1409          */
1410         if (nparamsp)
1411                 *nparamsp = nparams;
1412         if (flowclenp)
1413                 *flowclenp = flowclen;
1414         return flowclen16;
1415 }
1416
1417 u32 cxgbit_send_tx_flowc_wr(struct cxgbit_sock *csk)
1418 {
1419         struct cxgbit_device *cdev = csk->com.cdev;
1420         struct fw_flowc_wr *flowc;
1421         u32 nparams, flowclen16, flowclen;
1422         struct sk_buff *skb;
1423         u8 index;
1424
1425 #ifdef CONFIG_CHELSIO_T4_DCB
1426         u16 vlan = ((struct l2t_entry *)csk->l2t)->vlan;
1427 #endif
1428
1429         flowclen16 = cxgbit_tx_flowc_wr_credits(csk, &nparams, &flowclen);
1430
1431         skb = __skb_dequeue(&csk->skbq);
1432         flowc = __skb_put_zero(skb, flowclen);
1433
1434         flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
1435                                            FW_FLOWC_WR_NPARAMS_V(nparams));
1436         flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(flowclen16) |
1437                                           FW_WR_FLOWID_V(csk->tid));
1438         flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
1439         flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
1440                                             (csk->com.cdev->lldi.pf));
1441         flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
1442         flowc->mnemval[1].val = cpu_to_be32(csk->tx_chan);
1443         flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
1444         flowc->mnemval[2].val = cpu_to_be32(csk->tx_chan);
1445         flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
1446         flowc->mnemval[3].val = cpu_to_be32(csk->rss_qid);
1447         flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
1448         flowc->mnemval[4].val = cpu_to_be32(csk->snd_nxt);
1449         flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
1450         flowc->mnemval[5].val = cpu_to_be32(csk->rcv_nxt);
1451         flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
1452         flowc->mnemval[6].val = cpu_to_be32(csk->snd_win);
1453         flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
1454         flowc->mnemval[7].val = cpu_to_be32(csk->emss);
1455
1456         flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_TXDATAPLEN_MAX;
1457         if (test_bit(CDEV_ISO_ENABLE, &cdev->flags))
1458                 flowc->mnemval[8].val = cpu_to_be32(CXGBIT_MAX_ISO_PAYLOAD);
1459         else
1460                 flowc->mnemval[8].val = cpu_to_be32(16384);
1461
1462         index = 9;
1463
1464         if (csk->snd_wscale) {
1465                 flowc->mnemval[index].mnemonic = FW_FLOWC_MNEM_RCV_SCALE;
1466                 flowc->mnemval[index].val = cpu_to_be32(csk->snd_wscale);
1467                 index++;
1468         }
1469
1470 #ifdef CONFIG_CHELSIO_T4_DCB
1471         flowc->mnemval[index].mnemonic = FW_FLOWC_MNEM_DCBPRIO;
1472         if (vlan == VLAN_NONE) {
1473                 pr_warn("csk %u without VLAN Tag on DCB Link\n", csk->tid);
1474                 flowc->mnemval[index].val = cpu_to_be32(0);
1475         } else
1476                 flowc->mnemval[index].val = cpu_to_be32(
1477                                 (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT);
1478 #endif
1479
1480         pr_debug("%s: csk %p; tx_chan = %u; rss_qid = %u; snd_seq = %u;"
1481                  " rcv_seq = %u; snd_win = %u; emss = %u\n",
1482                  __func__, csk, csk->tx_chan, csk->rss_qid, csk->snd_nxt,
1483                  csk->rcv_nxt, csk->snd_win, csk->emss);
1484         set_wr_txq(skb, CPL_PRIORITY_DATA, csk->txq_idx);
1485         cxgbit_ofld_send(csk->com.cdev, skb);
1486         return flowclen16;
1487 }
1488
1489 int cxgbit_setup_conn_digest(struct cxgbit_sock *csk)
1490 {
1491         struct sk_buff *skb;
1492         struct cpl_set_tcb_field *req;
1493         u8 hcrc = csk->submode & CXGBIT_SUBMODE_HCRC;
1494         u8 dcrc = csk->submode & CXGBIT_SUBMODE_DCRC;
1495         unsigned int len = roundup(sizeof(*req), 16);
1496         int ret;
1497
1498         skb = alloc_skb(len, GFP_KERNEL);
1499         if (!skb)
1500                 return -ENOMEM;
1501
1502         /*  set up ulp submode */
1503         req = __skb_put_zero(skb, len);
1504
1505         INIT_TP_WR(req, csk->tid);
1506         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, csk->tid));
1507         req->reply_ctrl = htons(NO_REPLY_V(0) | QUEUENO_V(csk->rss_qid));
1508         req->word_cookie = htons(0);
1509         req->mask = cpu_to_be64(0x3 << 4);
1510         req->val = cpu_to_be64(((hcrc ? ULP_CRC_HEADER : 0) |
1511                                 (dcrc ? ULP_CRC_DATA : 0)) << 4);
1512         set_wr_txq(skb, CPL_PRIORITY_CONTROL, csk->ctrlq_idx);
1513
1514         cxgbit_get_csk(csk);
1515         cxgbit_init_wr_wait(&csk->com.wr_wait);
1516
1517         cxgbit_ofld_send(csk->com.cdev, skb);
1518
1519         ret = cxgbit_wait_for_reply(csk->com.cdev,
1520                                     &csk->com.wr_wait,
1521                                     csk->tid, 5, __func__);
1522         if (ret)
1523                 return -1;
1524
1525         return 0;
1526 }
1527
1528 int cxgbit_setup_conn_pgidx(struct cxgbit_sock *csk, u32 pg_idx)
1529 {
1530         struct sk_buff *skb;
1531         struct cpl_set_tcb_field *req;
1532         unsigned int len = roundup(sizeof(*req), 16);
1533         int ret;
1534
1535         skb = alloc_skb(len, GFP_KERNEL);
1536         if (!skb)
1537                 return -ENOMEM;
1538
1539         req = __skb_put_zero(skb, len);
1540
1541         INIT_TP_WR(req, csk->tid);
1542         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, csk->tid));
1543         req->reply_ctrl = htons(NO_REPLY_V(0) | QUEUENO_V(csk->rss_qid));
1544         req->word_cookie = htons(0);
1545         req->mask = cpu_to_be64(0x3 << 8);
1546         req->val = cpu_to_be64(pg_idx << 8);
1547         set_wr_txq(skb, CPL_PRIORITY_CONTROL, csk->ctrlq_idx);
1548
1549         cxgbit_get_csk(csk);
1550         cxgbit_init_wr_wait(&csk->com.wr_wait);
1551
1552         cxgbit_ofld_send(csk->com.cdev, skb);
1553
1554         ret = cxgbit_wait_for_reply(csk->com.cdev,
1555                                     &csk->com.wr_wait,
1556                                     csk->tid, 5, __func__);
1557         if (ret)
1558                 return -1;
1559
1560         return 0;
1561 }
1562
1563 static void
1564 cxgbit_pass_open_rpl(struct cxgbit_device *cdev, struct sk_buff *skb)
1565 {
1566         struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1567         struct tid_info *t = cdev->lldi.tids;
1568         unsigned int stid = GET_TID(rpl);
1569         struct cxgbit_np *cnp = lookup_stid(t, stid);
1570
1571         pr_debug("%s: cnp = %p; stid = %u; status = %d\n",
1572                  __func__, cnp, stid, rpl->status);
1573
1574         if (!cnp) {
1575                 pr_info("%s stid %d lookup failure\n", __func__, stid);
1576                 goto rel_skb;
1577         }
1578
1579         cxgbit_wake_up(&cnp->com.wr_wait, __func__, rpl->status);
1580         cxgbit_put_cnp(cnp);
1581 rel_skb:
1582         __kfree_skb(skb);
1583 }
1584
1585 static void
1586 cxgbit_close_listsrv_rpl(struct cxgbit_device *cdev, struct sk_buff *skb)
1587 {
1588         struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1589         struct tid_info *t = cdev->lldi.tids;
1590         unsigned int stid = GET_TID(rpl);
1591         struct cxgbit_np *cnp = lookup_stid(t, stid);
1592
1593         pr_debug("%s: cnp = %p; stid = %u; status = %d\n",
1594                  __func__, cnp, stid, rpl->status);
1595
1596         if (!cnp) {
1597                 pr_info("%s stid %d lookup failure\n", __func__, stid);
1598                 goto rel_skb;
1599         }
1600
1601         cxgbit_wake_up(&cnp->com.wr_wait, __func__, rpl->status);
1602         cxgbit_put_cnp(cnp);
1603 rel_skb:
1604         __kfree_skb(skb);
1605 }
1606
1607 static void
1608 cxgbit_pass_establish(struct cxgbit_device *cdev, struct sk_buff *skb)
1609 {
1610         struct cpl_pass_establish *req = cplhdr(skb);
1611         struct tid_info *t = cdev->lldi.tids;
1612         unsigned int tid = GET_TID(req);
1613         struct cxgbit_sock *csk;
1614         struct cxgbit_np *cnp;
1615         u16 tcp_opt = be16_to_cpu(req->tcp_opt);
1616         u32 snd_isn = be32_to_cpu(req->snd_isn);
1617         u32 rcv_isn = be32_to_cpu(req->rcv_isn);
1618
1619         csk = lookup_tid(t, tid);
1620         if (unlikely(!csk)) {
1621                 pr_err("can't find connection for tid %u.\n", tid);
1622                 goto rel_skb;
1623         }
1624         cnp = csk->cnp;
1625
1626         pr_debug("%s: csk %p; tid %u; cnp %p\n",
1627                  __func__, csk, tid, cnp);
1628
1629         csk->write_seq = snd_isn;
1630         csk->snd_una = snd_isn;
1631         csk->snd_nxt = snd_isn;
1632
1633         csk->rcv_nxt = rcv_isn;
1634
1635         if (csk->rcv_win > (RCV_BUFSIZ_M << 10))
1636                 csk->rx_credits = (csk->rcv_win - (RCV_BUFSIZ_M << 10));
1637
1638         csk->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
1639         cxgbit_set_emss(csk, tcp_opt);
1640         dst_confirm(csk->dst);
1641         csk->com.state = CSK_STATE_ESTABLISHED;
1642         spin_lock_bh(&cnp->np_accept_lock);
1643         list_add_tail(&csk->accept_node, &cnp->np_accept_list);
1644         spin_unlock_bh(&cnp->np_accept_lock);
1645         complete(&cnp->accept_comp);
1646 rel_skb:
1647         __kfree_skb(skb);
1648 }
1649
1650 static void cxgbit_queue_rx_skb(struct cxgbit_sock *csk, struct sk_buff *skb)
1651 {
1652         cxgbit_skcb_flags(skb) = 0;
1653         spin_lock_bh(&csk->rxq.lock);
1654         __skb_queue_tail(&csk->rxq, skb);
1655         spin_unlock_bh(&csk->rxq.lock);
1656         wake_up(&csk->waitq);
1657 }
1658
1659 static void cxgbit_peer_close(struct cxgbit_sock *csk, struct sk_buff *skb)
1660 {
1661         pr_debug("%s: csk %p; tid %u; state %d\n",
1662                  __func__, csk, csk->tid, csk->com.state);
1663
1664         switch (csk->com.state) {
1665         case CSK_STATE_ESTABLISHED:
1666                 csk->com.state = CSK_STATE_CLOSING;
1667                 cxgbit_queue_rx_skb(csk, skb);
1668                 return;
1669         case CSK_STATE_CLOSING:
1670                 /* simultaneous close */
1671                 csk->com.state = CSK_STATE_MORIBUND;
1672                 break;
1673         case CSK_STATE_MORIBUND:
1674                 csk->com.state = CSK_STATE_DEAD;
1675                 cxgbit_put_csk(csk);
1676                 break;
1677         case CSK_STATE_ABORTING:
1678                 break;
1679         default:
1680                 pr_info("%s: cpl_peer_close in bad state %d\n",
1681                         __func__, csk->com.state);
1682         }
1683
1684         __kfree_skb(skb);
1685 }
1686
1687 static void cxgbit_close_con_rpl(struct cxgbit_sock *csk, struct sk_buff *skb)
1688 {
1689         pr_debug("%s: csk %p; tid %u; state %d\n",
1690                  __func__, csk, csk->tid, csk->com.state);
1691
1692         switch (csk->com.state) {
1693         case CSK_STATE_CLOSING:
1694                 csk->com.state = CSK_STATE_MORIBUND;
1695                 break;
1696         case CSK_STATE_MORIBUND:
1697                 csk->com.state = CSK_STATE_DEAD;
1698                 cxgbit_put_csk(csk);
1699                 break;
1700         case CSK_STATE_ABORTING:
1701         case CSK_STATE_DEAD:
1702                 break;
1703         default:
1704                 pr_info("%s: cpl_close_con_rpl in bad state %d\n",
1705                         __func__, csk->com.state);
1706         }
1707
1708         __kfree_skb(skb);
1709 }
1710
1711 static void cxgbit_abort_req_rss(struct cxgbit_sock *csk, struct sk_buff *skb)
1712 {
1713         struct cpl_abort_req_rss *hdr = cplhdr(skb);
1714         unsigned int tid = GET_TID(hdr);
1715         struct sk_buff *rpl_skb;
1716         bool release = false;
1717         bool wakeup_thread = false;
1718         u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
1719
1720         pr_debug("%s: csk %p; tid %u; state %d\n",
1721                  __func__, csk, tid, csk->com.state);
1722
1723         if (cxgb_is_neg_adv(hdr->status)) {
1724                 pr_err("%s: got neg advise %d on tid %u\n",
1725                        __func__, hdr->status, tid);
1726                 goto rel_skb;
1727         }
1728
1729         switch (csk->com.state) {
1730         case CSK_STATE_CONNECTING:
1731         case CSK_STATE_MORIBUND:
1732                 csk->com.state = CSK_STATE_DEAD;
1733                 release = true;
1734                 break;
1735         case CSK_STATE_ESTABLISHED:
1736                 csk->com.state = CSK_STATE_DEAD;
1737                 wakeup_thread = true;
1738                 break;
1739         case CSK_STATE_CLOSING:
1740                 csk->com.state = CSK_STATE_DEAD;
1741                 if (!csk->conn)
1742                         release = true;
1743                 break;
1744         case CSK_STATE_ABORTING:
1745                 break;
1746         default:
1747                 pr_info("%s: cpl_abort_req_rss in bad state %d\n",
1748                         __func__, csk->com.state);
1749                 csk->com.state = CSK_STATE_DEAD;
1750         }
1751
1752         __skb_queue_purge(&csk->txq);
1753
1754         if (!test_and_set_bit(CSK_TX_DATA_SENT, &csk->com.flags))
1755                 cxgbit_send_tx_flowc_wr(csk);
1756
1757         rpl_skb = __skb_dequeue(&csk->skbq);
1758
1759         cxgb_mk_abort_rpl(rpl_skb, len, csk->tid, csk->txq_idx);
1760         cxgbit_ofld_send(csk->com.cdev, rpl_skb);
1761
1762         if (wakeup_thread) {
1763                 cxgbit_queue_rx_skb(csk, skb);
1764                 return;
1765         }
1766
1767         if (release)
1768                 cxgbit_put_csk(csk);
1769 rel_skb:
1770         __kfree_skb(skb);
1771 }
1772
1773 static void cxgbit_abort_rpl_rss(struct cxgbit_sock *csk, struct sk_buff *skb)
1774 {
1775         struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1776
1777         pr_debug("%s: csk %p; tid %u; state %d\n",
1778                  __func__, csk, csk->tid, csk->com.state);
1779
1780         switch (csk->com.state) {
1781         case CSK_STATE_ABORTING:
1782                 csk->com.state = CSK_STATE_DEAD;
1783                 if (test_bit(CSK_ABORT_RPL_WAIT, &csk->com.flags))
1784                         cxgbit_wake_up(&csk->com.wr_wait, __func__,
1785                                        rpl->status);
1786                 cxgbit_put_csk(csk);
1787                 break;
1788         default:
1789                 pr_info("%s: cpl_abort_rpl_rss in state %d\n",
1790                         __func__, csk->com.state);
1791         }
1792
1793         __kfree_skb(skb);
1794 }
1795
1796 static bool cxgbit_credit_err(const struct cxgbit_sock *csk)
1797 {
1798         const struct sk_buff *skb = csk->wr_pending_head;
1799         u32 credit = 0;
1800
1801         if (unlikely(csk->wr_cred > csk->wr_max_cred)) {
1802                 pr_err("csk 0x%p, tid %u, credit %u > %u\n",
1803                        csk, csk->tid, csk->wr_cred, csk->wr_max_cred);
1804                 return true;
1805         }
1806
1807         while (skb) {
1808                 credit += (__force u32)skb->csum;
1809                 skb = cxgbit_skcb_tx_wr_next(skb);
1810         }
1811
1812         if (unlikely((csk->wr_cred + credit) != csk->wr_max_cred)) {
1813                 pr_err("csk 0x%p, tid %u, credit %u + %u != %u.\n",
1814                        csk, csk->tid, csk->wr_cred,
1815                        credit, csk->wr_max_cred);
1816
1817                 return true;
1818         }
1819
1820         return false;
1821 }
1822
1823 static void cxgbit_fw4_ack(struct cxgbit_sock *csk, struct sk_buff *skb)
1824 {
1825         struct cpl_fw4_ack *rpl = (struct cpl_fw4_ack *)cplhdr(skb);
1826         u32 credits = rpl->credits;
1827         u32 snd_una = ntohl(rpl->snd_una);
1828
1829         csk->wr_cred += credits;
1830         if (csk->wr_una_cred > (csk->wr_max_cred - csk->wr_cred))
1831                 csk->wr_una_cred = csk->wr_max_cred - csk->wr_cred;
1832
1833         while (credits) {
1834                 struct sk_buff *p = cxgbit_sock_peek_wr(csk);
1835                 u32 csum;
1836
1837                 if (unlikely(!p)) {
1838                         pr_err("csk 0x%p,%u, cr %u,%u+%u, empty.\n",
1839                                csk, csk->tid, credits,
1840                                csk->wr_cred, csk->wr_una_cred);
1841                         break;
1842                 }
1843
1844                 csum = (__force u32)p->csum;
1845                 if (unlikely(credits < csum)) {
1846                         pr_warn("csk 0x%p,%u, cr %u,%u+%u, < %u.\n",
1847                                 csk,  csk->tid,
1848                                 credits, csk->wr_cred, csk->wr_una_cred,
1849                                 csum);
1850                         p->csum = (__force __wsum)(csum - credits);
1851                         break;
1852                 }
1853
1854                 cxgbit_sock_dequeue_wr(csk);
1855                 credits -= csum;
1856                 kfree_skb(p);
1857         }
1858
1859         if (unlikely(cxgbit_credit_err(csk))) {
1860                 cxgbit_queue_rx_skb(csk, skb);
1861                 return;
1862         }
1863
1864         if (rpl->seq_vld & CPL_FW4_ACK_FLAGS_SEQVAL) {
1865                 if (unlikely(before(snd_una, csk->snd_una))) {
1866                         pr_warn("csk 0x%p,%u, snd_una %u/%u.",
1867                                 csk, csk->tid, snd_una,
1868                                 csk->snd_una);
1869                         goto rel_skb;
1870                 }
1871
1872                 if (csk->snd_una != snd_una) {
1873                         csk->snd_una = snd_una;
1874                         dst_confirm(csk->dst);
1875                         wake_up(&csk->ack_waitq);
1876                 }
1877         }
1878
1879         if (skb_queue_len(&csk->txq))
1880                 cxgbit_push_tx_frames(csk);
1881
1882 rel_skb:
1883         __kfree_skb(skb);
1884 }
1885
1886 static void cxgbit_set_tcb_rpl(struct cxgbit_device *cdev, struct sk_buff *skb)
1887 {
1888         struct cxgbit_sock *csk;
1889         struct cpl_set_tcb_rpl *rpl = (struct cpl_set_tcb_rpl *)skb->data;
1890         unsigned int tid = GET_TID(rpl);
1891         struct cxgb4_lld_info *lldi = &cdev->lldi;
1892         struct tid_info *t = lldi->tids;
1893
1894         csk = lookup_tid(t, tid);
1895         if (unlikely(!csk)) {
1896                 pr_err("can't find connection for tid %u.\n", tid);
1897                 goto rel_skb;
1898         } else {
1899                 cxgbit_wake_up(&csk->com.wr_wait, __func__, rpl->status);
1900         }
1901
1902         cxgbit_put_csk(csk);
1903 rel_skb:
1904         __kfree_skb(skb);
1905 }
1906
1907 static void cxgbit_rx_data(struct cxgbit_device *cdev, struct sk_buff *skb)
1908 {
1909         struct cxgbit_sock *csk;
1910         struct cpl_rx_data *cpl = cplhdr(skb);
1911         unsigned int tid = GET_TID(cpl);
1912         struct cxgb4_lld_info *lldi = &cdev->lldi;
1913         struct tid_info *t = lldi->tids;
1914
1915         csk = lookup_tid(t, tid);
1916         if (unlikely(!csk)) {
1917                 pr_err("can't find conn. for tid %u.\n", tid);
1918                 goto rel_skb;
1919         }
1920
1921         cxgbit_queue_rx_skb(csk, skb);
1922         return;
1923 rel_skb:
1924         __kfree_skb(skb);
1925 }
1926
1927 static void
1928 __cxgbit_process_rx_cpl(struct cxgbit_sock *csk, struct sk_buff *skb)
1929 {
1930         spin_lock(&csk->lock);
1931         if (csk->lock_owner) {
1932                 __skb_queue_tail(&csk->backlogq, skb);
1933                 spin_unlock(&csk->lock);
1934                 return;
1935         }
1936
1937         cxgbit_skcb_rx_backlog_fn(skb)(csk, skb);
1938         spin_unlock(&csk->lock);
1939 }
1940
1941 static void cxgbit_process_rx_cpl(struct cxgbit_sock *csk, struct sk_buff *skb)
1942 {
1943         cxgbit_get_csk(csk);
1944         __cxgbit_process_rx_cpl(csk, skb);
1945         cxgbit_put_csk(csk);
1946 }
1947
1948 static void cxgbit_rx_cpl(struct cxgbit_device *cdev, struct sk_buff *skb)
1949 {
1950         struct cxgbit_sock *csk;
1951         struct cpl_tx_data *cpl = cplhdr(skb);
1952         struct cxgb4_lld_info *lldi = &cdev->lldi;
1953         struct tid_info *t = lldi->tids;
1954         unsigned int tid = GET_TID(cpl);
1955         u8 opcode = cxgbit_skcb_rx_opcode(skb);
1956         bool ref = true;
1957
1958         switch (opcode) {
1959         case CPL_FW4_ACK:
1960                         cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_fw4_ack;
1961                         ref = false;
1962                         break;
1963         case CPL_PEER_CLOSE:
1964                         cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_peer_close;
1965                         break;
1966         case CPL_CLOSE_CON_RPL:
1967                         cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_close_con_rpl;
1968                         break;
1969         case CPL_ABORT_REQ_RSS:
1970                         cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_abort_req_rss;
1971                         break;
1972         case CPL_ABORT_RPL_RSS:
1973                         cxgbit_skcb_rx_backlog_fn(skb) = cxgbit_abort_rpl_rss;
1974                         break;
1975         default:
1976                 goto rel_skb;
1977         }
1978
1979         csk = lookup_tid(t, tid);
1980         if (unlikely(!csk)) {
1981                 pr_err("can't find conn. for tid %u.\n", tid);
1982                 goto rel_skb;
1983         }
1984
1985         if (ref)
1986                 cxgbit_process_rx_cpl(csk, skb);
1987         else
1988                 __cxgbit_process_rx_cpl(csk, skb);
1989
1990         return;
1991 rel_skb:
1992         __kfree_skb(skb);
1993 }
1994
1995 cxgbit_cplhandler_func cxgbit_cplhandlers[NUM_CPL_CMDS] = {
1996         [CPL_PASS_OPEN_RPL]     = cxgbit_pass_open_rpl,
1997         [CPL_CLOSE_LISTSRV_RPL] = cxgbit_close_listsrv_rpl,
1998         [CPL_PASS_ACCEPT_REQ]   = cxgbit_pass_accept_req,
1999         [CPL_PASS_ESTABLISH]    = cxgbit_pass_establish,
2000         [CPL_SET_TCB_RPL]       = cxgbit_set_tcb_rpl,
2001         [CPL_RX_DATA]           = cxgbit_rx_data,
2002         [CPL_FW4_ACK]           = cxgbit_rx_cpl,
2003         [CPL_PEER_CLOSE]        = cxgbit_rx_cpl,
2004         [CPL_CLOSE_CON_RPL]     = cxgbit_rx_cpl,
2005         [CPL_ABORT_REQ_RSS]     = cxgbit_rx_cpl,
2006         [CPL_ABORT_RPL_RSS]     = cxgbit_rx_cpl,
2007 };