1 /* 2 * Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved. 3 * 4 * This software is available to you under a choice of one of two 5 * licenses. You may choose to be licensed under the terms of the GNU 6 * General Public License (GPL) Version 2, available from the file 7 * COPYING in the main directory of this source tree, or the 8 * OpenIB.org BSD license below: 9 * 10 * Redistribution and use in source and binary forms, with or 11 * without modification, are permitted provided that the following 12 * conditions are met: 13 * 14 * - Redistributions of source code must retain the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer. 17 * 18 * - Redistributions in binary form must reproduce the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer in the documentation and/or other materials 21 * provided with the distribution. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 30 * SOFTWARE. 31 */ 32 #include <linux/module.h> 33 #include <linux/list.h> 34 #include <linux/workqueue.h> 35 #include <linux/skbuff.h> 36 #include <linux/timer.h> 37 #include <linux/notifier.h> 38 #include <linux/inetdevice.h> 39 #include <linux/ip.h> 40 #include <linux/tcp.h> 41 #include <linux/if_vlan.h> 42 43 #include <net/neighbour.h> 44 #include <net/netevent.h> 45 #include <net/route.h> 46 #include <net/tcp.h> 47 #include <net/ip6_route.h> 48 #include <net/addrconf.h> 49 50 #include <rdma/ib_addr.h> 51 52 #include <libcxgb_cm.h> 53 #include "iw_cxgb4.h" 54 #include "clip_tbl.h" 55 56 static char *states[] = { 57 "idle", 58 "listen", 59 "connecting", 60 "mpa_wait_req", 61 "mpa_req_sent", 62 "mpa_req_rcvd", 63 "mpa_rep_sent", 64 "fpdu_mode", 65 "aborting", 66 "closing", 67 "moribund", 68 "dead", 69 NULL, 70 }; 71 72 static int nocong; 73 module_param(nocong, int, 0644); 74 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)"); 75 76 static int enable_ecn; 77 module_param(enable_ecn, int, 0644); 78 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)"); 79 80 static int dack_mode = 1; 81 module_param(dack_mode, int, 0644); 82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)"); 83 84 uint c4iw_max_read_depth = 32; 85 module_param(c4iw_max_read_depth, int, 0644); 86 MODULE_PARM_DESC(c4iw_max_read_depth, 87 "Per-connection max ORD/IRD (default=32)"); 88 89 static int enable_tcp_timestamps; 90 module_param(enable_tcp_timestamps, int, 0644); 91 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)"); 92 93 static int enable_tcp_sack; 94 module_param(enable_tcp_sack, int, 0644); 95 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)"); 96 97 static int enable_tcp_window_scaling = 1; 98 module_param(enable_tcp_window_scaling, int, 0644); 99 MODULE_PARM_DESC(enable_tcp_window_scaling, 100 "Enable tcp window scaling (default=1)"); 101 102 int c4iw_debug; 103 module_param(c4iw_debug, int, 0644); 104 MODULE_PARM_DESC(c4iw_debug, "obsolete"); 105 106 static int peer2peer = 1; 107 module_param(peer2peer, int, 0644); 108 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)"); 109 110 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ; 111 module_param(p2p_type, int, 0644); 112 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: " 113 "1=RDMA_READ 0=RDMA_WRITE (default 1)"); 114 115 static int ep_timeout_secs = 60; 116 module_param(ep_timeout_secs, int, 0644); 117 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout " 118 "in seconds (default=60)"); 119 120 static int mpa_rev = 2; 121 module_param(mpa_rev, int, 0644); 122 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, " 123 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft" 124 " compliant (default=2)"); 125 126 static int markers_enabled; 127 module_param(markers_enabled, int, 0644); 128 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)"); 129 130 static int crc_enabled = 1; 131 module_param(crc_enabled, int, 0644); 132 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)"); 133 134 static int rcv_win = 256 * 1024; 135 module_param(rcv_win, int, 0644); 136 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)"); 137 138 static int snd_win = 128 * 1024; 139 module_param(snd_win, int, 0644); 140 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)"); 141 142 static struct workqueue_struct *workq; 143 144 static struct sk_buff_head rxq; 145 146 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp); 147 static void ep_timeout(unsigned long arg); 148 static void connect_reply_upcall(struct c4iw_ep *ep, int status); 149 static int sched(struct c4iw_dev *dev, struct sk_buff *skb); 150 151 static LIST_HEAD(timeout_list); 152 static spinlock_t timeout_lock; 153 154 static void deref_cm_id(struct c4iw_ep_common *epc) 155 { 156 epc->cm_id->rem_ref(epc->cm_id); 157 epc->cm_id = NULL; 158 set_bit(CM_ID_DEREFED, &epc->history); 159 } 160 161 static void ref_cm_id(struct c4iw_ep_common *epc) 162 { 163 set_bit(CM_ID_REFED, &epc->history); 164 epc->cm_id->add_ref(epc->cm_id); 165 } 166 167 static void deref_qp(struct c4iw_ep *ep) 168 { 169 c4iw_qp_rem_ref(&ep->com.qp->ibqp); 170 clear_bit(QP_REFERENCED, &ep->com.flags); 171 set_bit(QP_DEREFED, &ep->com.history); 172 } 173 174 static void ref_qp(struct c4iw_ep *ep) 175 { 176 set_bit(QP_REFERENCED, &ep->com.flags); 177 set_bit(QP_REFED, &ep->com.history); 178 c4iw_qp_add_ref(&ep->com.qp->ibqp); 179 } 180 181 static void start_ep_timer(struct c4iw_ep *ep) 182 { 183 pr_debug("%s ep %p\n", __func__, ep); 184 if (timer_pending(&ep->timer)) { 185 pr_err("%s timer already started! ep %p\n", 186 __func__, ep); 187 return; 188 } 189 clear_bit(TIMEOUT, &ep->com.flags); 190 c4iw_get_ep(&ep->com); 191 ep->timer.expires = jiffies + ep_timeout_secs * HZ; 192 ep->timer.data = (unsigned long)ep; 193 ep->timer.function = ep_timeout; 194 add_timer(&ep->timer); 195 } 196 197 static int stop_ep_timer(struct c4iw_ep *ep) 198 { 199 pr_debug("%s ep %p stopping\n", __func__, ep); 200 del_timer_sync(&ep->timer); 201 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 202 c4iw_put_ep(&ep->com); 203 return 0; 204 } 205 return 1; 206 } 207 208 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb, 209 struct l2t_entry *l2e) 210 { 211 int error = 0; 212 213 if (c4iw_fatal_error(rdev)) { 214 kfree_skb(skb); 215 pr_debug("%s - device in error state - dropping\n", __func__); 216 return -EIO; 217 } 218 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e); 219 if (error < 0) 220 kfree_skb(skb); 221 else if (error == NET_XMIT_DROP) 222 return -ENOMEM; 223 return error < 0 ? error : 0; 224 } 225 226 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb) 227 { 228 int error = 0; 229 230 if (c4iw_fatal_error(rdev)) { 231 kfree_skb(skb); 232 pr_debug("%s - device in error state - dropping\n", __func__); 233 return -EIO; 234 } 235 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb); 236 if (error < 0) 237 kfree_skb(skb); 238 return error < 0 ? error : 0; 239 } 240 241 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb) 242 { 243 u32 len = roundup(sizeof(struct cpl_tid_release), 16); 244 245 skb = get_skb(skb, len, GFP_KERNEL); 246 if (!skb) 247 return; 248 249 cxgb_mk_tid_release(skb, len, hwtid, 0); 250 c4iw_ofld_send(rdev, skb); 251 return; 252 } 253 254 static void set_emss(struct c4iw_ep *ep, u16 opt) 255 { 256 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] - 257 ((AF_INET == ep->com.remote_addr.ss_family) ? 258 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) - 259 sizeof(struct tcphdr); 260 ep->mss = ep->emss; 261 if (TCPOPT_TSTAMP_G(opt)) 262 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4); 263 if (ep->emss < 128) 264 ep->emss = 128; 265 if (ep->emss & 7) 266 pr_debug("Warning: misaligned mtu idx %u mss %u emss=%u\n", 267 TCPOPT_MSS_G(opt), ep->mss, ep->emss); 268 pr_debug("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt), 269 ep->mss, ep->emss); 270 } 271 272 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc) 273 { 274 enum c4iw_ep_state state; 275 276 mutex_lock(&epc->mutex); 277 state = epc->state; 278 mutex_unlock(&epc->mutex); 279 return state; 280 } 281 282 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new) 283 { 284 epc->state = new; 285 } 286 287 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new) 288 { 289 mutex_lock(&epc->mutex); 290 pr_debug("%s - %s -> %s\n", __func__, states[epc->state], states[new]); 291 __state_set(epc, new); 292 mutex_unlock(&epc->mutex); 293 return; 294 } 295 296 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size) 297 { 298 struct sk_buff *skb; 299 unsigned int i; 300 size_t len; 301 302 len = roundup(sizeof(union cpl_wr_size), 16); 303 for (i = 0; i < size; i++) { 304 skb = alloc_skb(len, GFP_KERNEL); 305 if (!skb) 306 goto fail; 307 skb_queue_tail(ep_skb_list, skb); 308 } 309 return 0; 310 fail: 311 skb_queue_purge(ep_skb_list); 312 return -ENOMEM; 313 } 314 315 static void *alloc_ep(int size, gfp_t gfp) 316 { 317 struct c4iw_ep_common *epc; 318 319 epc = kzalloc(size, gfp); 320 if (epc) { 321 kref_init(&epc->kref); 322 mutex_init(&epc->mutex); 323 c4iw_init_wr_wait(&epc->wr_wait); 324 } 325 pr_debug("%s alloc ep %p\n", __func__, epc); 326 return epc; 327 } 328 329 static void remove_ep_tid(struct c4iw_ep *ep) 330 { 331 unsigned long flags; 332 333 spin_lock_irqsave(&ep->com.dev->lock, flags); 334 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0); 335 if (idr_is_empty(&ep->com.dev->hwtid_idr)) 336 wake_up(&ep->com.dev->wait); 337 spin_unlock_irqrestore(&ep->com.dev->lock, flags); 338 } 339 340 static void insert_ep_tid(struct c4iw_ep *ep) 341 { 342 unsigned long flags; 343 344 spin_lock_irqsave(&ep->com.dev->lock, flags); 345 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0); 346 spin_unlock_irqrestore(&ep->com.dev->lock, flags); 347 } 348 349 /* 350 * Atomically lookup the ep ptr given the tid and grab a reference on the ep. 351 */ 352 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid) 353 { 354 struct c4iw_ep *ep; 355 unsigned long flags; 356 357 spin_lock_irqsave(&dev->lock, flags); 358 ep = idr_find(&dev->hwtid_idr, tid); 359 if (ep) 360 c4iw_get_ep(&ep->com); 361 spin_unlock_irqrestore(&dev->lock, flags); 362 return ep; 363 } 364 365 /* 366 * Atomically lookup the ep ptr given the stid and grab a reference on the ep. 367 */ 368 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev, 369 unsigned int stid) 370 { 371 struct c4iw_listen_ep *ep; 372 unsigned long flags; 373 374 spin_lock_irqsave(&dev->lock, flags); 375 ep = idr_find(&dev->stid_idr, stid); 376 if (ep) 377 c4iw_get_ep(&ep->com); 378 spin_unlock_irqrestore(&dev->lock, flags); 379 return ep; 380 } 381 382 void _c4iw_free_ep(struct kref *kref) 383 { 384 struct c4iw_ep *ep; 385 386 ep = container_of(kref, struct c4iw_ep, com.kref); 387 pr_debug("%s ep %p state %s\n", __func__, ep, states[ep->com.state]); 388 if (test_bit(QP_REFERENCED, &ep->com.flags)) 389 deref_qp(ep); 390 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) { 391 if (ep->com.remote_addr.ss_family == AF_INET6) { 392 struct sockaddr_in6 *sin6 = 393 (struct sockaddr_in6 *) 394 &ep->com.local_addr; 395 396 cxgb4_clip_release( 397 ep->com.dev->rdev.lldi.ports[0], 398 (const u32 *)&sin6->sin6_addr.s6_addr, 399 1); 400 } 401 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid); 402 dst_release(ep->dst); 403 cxgb4_l2t_release(ep->l2t); 404 if (ep->mpa_skb) 405 kfree_skb(ep->mpa_skb); 406 } 407 if (!skb_queue_empty(&ep->com.ep_skb_list)) 408 skb_queue_purge(&ep->com.ep_skb_list); 409 kfree(ep); 410 } 411 412 static void release_ep_resources(struct c4iw_ep *ep) 413 { 414 set_bit(RELEASE_RESOURCES, &ep->com.flags); 415 416 /* 417 * If we have a hwtid, then remove it from the idr table 418 * so lookups will no longer find this endpoint. Otherwise 419 * we have a race where one thread finds the ep ptr just 420 * before the other thread is freeing the ep memory. 421 */ 422 if (ep->hwtid != -1) 423 remove_ep_tid(ep); 424 c4iw_put_ep(&ep->com); 425 } 426 427 static int status2errno(int status) 428 { 429 switch (status) { 430 case CPL_ERR_NONE: 431 return 0; 432 case CPL_ERR_CONN_RESET: 433 return -ECONNRESET; 434 case CPL_ERR_ARP_MISS: 435 return -EHOSTUNREACH; 436 case CPL_ERR_CONN_TIMEDOUT: 437 return -ETIMEDOUT; 438 case CPL_ERR_TCAM_FULL: 439 return -ENOMEM; 440 case CPL_ERR_CONN_EXIST: 441 return -EADDRINUSE; 442 default: 443 return -EIO; 444 } 445 } 446 447 /* 448 * Try and reuse skbs already allocated... 449 */ 450 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp) 451 { 452 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) { 453 skb_trim(skb, 0); 454 skb_get(skb); 455 skb_reset_transport_header(skb); 456 } else { 457 skb = alloc_skb(len, gfp); 458 } 459 t4_set_arp_err_handler(skb, NULL, NULL); 460 return skb; 461 } 462 463 static struct net_device *get_real_dev(struct net_device *egress_dev) 464 { 465 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev; 466 } 467 468 static void arp_failure_discard(void *handle, struct sk_buff *skb) 469 { 470 pr_err("ARP failure\n"); 471 kfree_skb(skb); 472 } 473 474 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb) 475 { 476 pr_err("ARP failure during MPA Negotiation - Closing Connection\n"); 477 } 478 479 enum { 480 NUM_FAKE_CPLS = 2, 481 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0, 482 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1, 483 }; 484 485 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb) 486 { 487 struct c4iw_ep *ep; 488 489 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))); 490 release_ep_resources(ep); 491 return 0; 492 } 493 494 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb) 495 { 496 struct c4iw_ep *ep; 497 498 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))); 499 c4iw_put_ep(&ep->parent_ep->com); 500 release_ep_resources(ep); 501 return 0; 502 } 503 504 /* 505 * Fake up a special CPL opcode and call sched() so process_work() will call 506 * _put_ep_safe() in a safe context to free the ep resources. This is needed 507 * because ARP error handlers are called in an ATOMIC context, and 508 * _c4iw_free_ep() needs to block. 509 */ 510 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb, 511 int cpl) 512 { 513 struct cpl_act_establish *rpl = cplhdr(skb); 514 515 /* Set our special ARP_FAILURE opcode */ 516 rpl->ot.opcode = cpl; 517 518 /* 519 * Save ep in the skb->cb area, after where sched() will save the dev 520 * ptr. 521 */ 522 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep; 523 sched(ep->com.dev, skb); 524 } 525 526 /* Handle an ARP failure for an accept */ 527 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb) 528 { 529 struct c4iw_ep *ep = handle; 530 531 pr_err("ARP failure during accept - tid %u - dropping connection\n", 532 ep->hwtid); 533 534 __state_set(&ep->com, DEAD); 535 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE); 536 } 537 538 /* 539 * Handle an ARP failure for an active open. 540 */ 541 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb) 542 { 543 struct c4iw_ep *ep = handle; 544 545 pr_err("ARP failure during connect\n"); 546 connect_reply_upcall(ep, -EHOSTUNREACH); 547 __state_set(&ep->com, DEAD); 548 if (ep->com.remote_addr.ss_family == AF_INET6) { 549 struct sockaddr_in6 *sin6 = 550 (struct sockaddr_in6 *)&ep->com.local_addr; 551 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 552 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 553 } 554 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid); 555 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid); 556 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE); 557 } 558 559 /* 560 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant 561 * and send it along. 562 */ 563 static void abort_arp_failure(void *handle, struct sk_buff *skb) 564 { 565 int ret; 566 struct c4iw_ep *ep = handle; 567 struct c4iw_rdev *rdev = &ep->com.dev->rdev; 568 struct cpl_abort_req *req = cplhdr(skb); 569 570 pr_debug("%s rdev %p\n", __func__, rdev); 571 req->cmd = CPL_ABORT_NO_RST; 572 ret = c4iw_ofld_send(rdev, skb); 573 if (ret) { 574 __state_set(&ep->com, DEAD); 575 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE); 576 } 577 } 578 579 static int send_flowc(struct c4iw_ep *ep) 580 { 581 struct fw_flowc_wr *flowc; 582 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list); 583 int i; 584 u16 vlan = ep->l2t->vlan; 585 int nparams; 586 587 if (WARN_ON(!skb)) 588 return -ENOMEM; 589 590 if (vlan == CPL_L2T_VLAN_NONE) 591 nparams = 8; 592 else 593 nparams = 9; 594 595 flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN); 596 597 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) | 598 FW_FLOWC_WR_NPARAMS_V(nparams)); 599 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN, 600 16)) | FW_WR_FLOWID_V(ep->hwtid)); 601 602 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN; 603 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V 604 (ep->com.dev->rdev.lldi.pf)); 605 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH; 606 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan); 607 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT; 608 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan); 609 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID; 610 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid); 611 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT; 612 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq); 613 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT; 614 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq); 615 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF; 616 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win); 617 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS; 618 flowc->mnemval[7].val = cpu_to_be32(ep->emss); 619 if (nparams == 9) { 620 u16 pri; 621 622 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT; 623 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS; 624 flowc->mnemval[8].val = cpu_to_be32(pri); 625 } else { 626 /* Pad WR to 16 byte boundary */ 627 flowc->mnemval[8].mnemonic = 0; 628 flowc->mnemval[8].val = 0; 629 } 630 for (i = 0; i < 9; i++) { 631 flowc->mnemval[i].r4[0] = 0; 632 flowc->mnemval[i].r4[1] = 0; 633 flowc->mnemval[i].r4[2] = 0; 634 } 635 636 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 637 return c4iw_ofld_send(&ep->com.dev->rdev, skb); 638 } 639 640 static int send_halfclose(struct c4iw_ep *ep) 641 { 642 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list); 643 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16); 644 645 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 646 if (WARN_ON(!skb)) 647 return -ENOMEM; 648 649 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx, 650 NULL, arp_failure_discard); 651 652 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 653 } 654 655 static int send_abort(struct c4iw_ep *ep) 656 { 657 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16); 658 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list); 659 660 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 661 if (WARN_ON(!req_skb)) 662 return -ENOMEM; 663 664 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx, 665 ep, abort_arp_failure); 666 667 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t); 668 } 669 670 static int send_connect(struct c4iw_ep *ep) 671 { 672 struct cpl_act_open_req *req = NULL; 673 struct cpl_t5_act_open_req *t5req = NULL; 674 struct cpl_t6_act_open_req *t6req = NULL; 675 struct cpl_act_open_req6 *req6 = NULL; 676 struct cpl_t5_act_open_req6 *t5req6 = NULL; 677 struct cpl_t6_act_open_req6 *t6req6 = NULL; 678 struct sk_buff *skb; 679 u64 opt0; 680 u32 opt2; 681 unsigned int mtu_idx; 682 u32 wscale; 683 int win, sizev4, sizev6, wrlen; 684 struct sockaddr_in *la = (struct sockaddr_in *) 685 &ep->com.local_addr; 686 struct sockaddr_in *ra = (struct sockaddr_in *) 687 &ep->com.remote_addr; 688 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *) 689 &ep->com.local_addr; 690 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *) 691 &ep->com.remote_addr; 692 int ret; 693 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type; 694 u32 isn = (prandom_u32() & ~7UL) - 1; 695 struct net_device *netdev; 696 u64 params; 697 698 netdev = ep->com.dev->rdev.lldi.ports[0]; 699 700 switch (CHELSIO_CHIP_VERSION(adapter_type)) { 701 case CHELSIO_T4: 702 sizev4 = sizeof(struct cpl_act_open_req); 703 sizev6 = sizeof(struct cpl_act_open_req6); 704 break; 705 case CHELSIO_T5: 706 sizev4 = sizeof(struct cpl_t5_act_open_req); 707 sizev6 = sizeof(struct cpl_t5_act_open_req6); 708 break; 709 case CHELSIO_T6: 710 sizev4 = sizeof(struct cpl_t6_act_open_req); 711 sizev6 = sizeof(struct cpl_t6_act_open_req6); 712 break; 713 default: 714 pr_err("T%d Chip is not supported\n", 715 CHELSIO_CHIP_VERSION(adapter_type)); 716 return -EINVAL; 717 } 718 719 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ? 720 roundup(sizev4, 16) : 721 roundup(sizev6, 16); 722 723 pr_debug("%s ep %p atid %u\n", __func__, ep, ep->atid); 724 725 skb = get_skb(NULL, wrlen, GFP_KERNEL); 726 if (!skb) { 727 pr_err("%s - failed to alloc skb\n", __func__); 728 return -ENOMEM; 729 } 730 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx); 731 732 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx, 733 enable_tcp_timestamps, 734 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1); 735 wscale = cxgb_compute_wscale(rcv_win); 736 737 /* 738 * Specify the largest window that will fit in opt0. The 739 * remainder will be specified in the rx_data_ack. 740 */ 741 win = ep->rcv_win >> 10; 742 if (win > RCV_BUFSIZ_M) 743 win = RCV_BUFSIZ_M; 744 745 opt0 = (nocong ? NO_CONG_F : 0) | 746 KEEP_ALIVE_F | 747 DELACK_F | 748 WND_SCALE_V(wscale) | 749 MSS_IDX_V(mtu_idx) | 750 L2T_IDX_V(ep->l2t->idx) | 751 TX_CHAN_V(ep->tx_chan) | 752 SMAC_SEL_V(ep->smac_idx) | 753 DSCP_V(ep->tos >> 2) | 754 ULP_MODE_V(ULP_MODE_TCPDDP) | 755 RCV_BUFSIZ_V(win); 756 opt2 = RX_CHANNEL_V(0) | 757 CCTRL_ECN_V(enable_ecn) | 758 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid); 759 if (enable_tcp_timestamps) 760 opt2 |= TSTAMPS_EN_F; 761 if (enable_tcp_sack) 762 opt2 |= SACK_EN_F; 763 if (wscale && enable_tcp_window_scaling) 764 opt2 |= WND_SCALE_EN_F; 765 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) { 766 if (peer2peer) 767 isn += 4; 768 769 opt2 |= T5_OPT_2_VALID_F; 770 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE); 771 opt2 |= T5_ISS_F; 772 } 773 774 params = cxgb4_select_ntuple(netdev, ep->l2t); 775 776 if (ep->com.remote_addr.ss_family == AF_INET6) 777 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0], 778 (const u32 *)&la6->sin6_addr.s6_addr, 1); 779 780 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure); 781 782 if (ep->com.remote_addr.ss_family == AF_INET) { 783 switch (CHELSIO_CHIP_VERSION(adapter_type)) { 784 case CHELSIO_T4: 785 req = (struct cpl_act_open_req *)skb_put(skb, wrlen); 786 INIT_TP_WR(req, 0); 787 break; 788 case CHELSIO_T5: 789 t5req = (struct cpl_t5_act_open_req *)skb_put(skb, 790 wrlen); 791 INIT_TP_WR(t5req, 0); 792 req = (struct cpl_act_open_req *)t5req; 793 break; 794 case CHELSIO_T6: 795 t6req = (struct cpl_t6_act_open_req *)skb_put(skb, 796 wrlen); 797 INIT_TP_WR(t6req, 0); 798 req = (struct cpl_act_open_req *)t6req; 799 t5req = (struct cpl_t5_act_open_req *)t6req; 800 break; 801 default: 802 pr_err("T%d Chip is not supported\n", 803 CHELSIO_CHIP_VERSION(adapter_type)); 804 ret = -EINVAL; 805 goto clip_release; 806 } 807 808 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, 809 ((ep->rss_qid<<14) | ep->atid))); 810 req->local_port = la->sin_port; 811 req->peer_port = ra->sin_port; 812 req->local_ip = la->sin_addr.s_addr; 813 req->peer_ip = ra->sin_addr.s_addr; 814 req->opt0 = cpu_to_be64(opt0); 815 816 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) { 817 req->params = cpu_to_be32(params); 818 req->opt2 = cpu_to_be32(opt2); 819 } else { 820 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) { 821 t5req->params = 822 cpu_to_be64(FILTER_TUPLE_V(params)); 823 t5req->rsvd = cpu_to_be32(isn); 824 pr_debug("%s snd_isn %u\n", __func__, t5req->rsvd); 825 t5req->opt2 = cpu_to_be32(opt2); 826 } else { 827 t6req->params = 828 cpu_to_be64(FILTER_TUPLE_V(params)); 829 t6req->rsvd = cpu_to_be32(isn); 830 pr_debug("%s snd_isn %u\n", __func__, t6req->rsvd); 831 t6req->opt2 = cpu_to_be32(opt2); 832 } 833 } 834 } else { 835 switch (CHELSIO_CHIP_VERSION(adapter_type)) { 836 case CHELSIO_T4: 837 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen); 838 INIT_TP_WR(req6, 0); 839 break; 840 case CHELSIO_T5: 841 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb, 842 wrlen); 843 INIT_TP_WR(t5req6, 0); 844 req6 = (struct cpl_act_open_req6 *)t5req6; 845 break; 846 case CHELSIO_T6: 847 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb, 848 wrlen); 849 INIT_TP_WR(t6req6, 0); 850 req6 = (struct cpl_act_open_req6 *)t6req6; 851 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6; 852 break; 853 default: 854 pr_err("T%d Chip is not supported\n", 855 CHELSIO_CHIP_VERSION(adapter_type)); 856 ret = -EINVAL; 857 goto clip_release; 858 } 859 860 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, 861 ((ep->rss_qid<<14)|ep->atid))); 862 req6->local_port = la6->sin6_port; 863 req6->peer_port = ra6->sin6_port; 864 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr)); 865 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8)); 866 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr)); 867 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8)); 868 req6->opt0 = cpu_to_be64(opt0); 869 870 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) { 871 req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev, 872 ep->l2t)); 873 req6->opt2 = cpu_to_be32(opt2); 874 } else { 875 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) { 876 t5req6->params = 877 cpu_to_be64(FILTER_TUPLE_V(params)); 878 t5req6->rsvd = cpu_to_be32(isn); 879 pr_debug("%s snd_isn %u\n", __func__, t5req6->rsvd); 880 t5req6->opt2 = cpu_to_be32(opt2); 881 } else { 882 t6req6->params = 883 cpu_to_be64(FILTER_TUPLE_V(params)); 884 t6req6->rsvd = cpu_to_be32(isn); 885 pr_debug("%s snd_isn %u\n", __func__, t6req6->rsvd); 886 t6req6->opt2 = cpu_to_be32(opt2); 887 } 888 889 } 890 } 891 892 set_bit(ACT_OPEN_REQ, &ep->com.history); 893 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 894 clip_release: 895 if (ret && ep->com.remote_addr.ss_family == AF_INET6) 896 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 897 (const u32 *)&la6->sin6_addr.s6_addr, 1); 898 return ret; 899 } 900 901 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb, 902 u8 mpa_rev_to_use) 903 { 904 int mpalen, wrlen, ret; 905 struct fw_ofld_tx_data_wr *req; 906 struct mpa_message *mpa; 907 struct mpa_v2_conn_params mpa_v2_params; 908 909 pr_debug("%s ep %p tid %u pd_len %d\n", 910 __func__, ep, ep->hwtid, ep->plen); 911 912 BUG_ON(skb_cloned(skb)); 913 914 mpalen = sizeof(*mpa) + ep->plen; 915 if (mpa_rev_to_use == 2) 916 mpalen += sizeof(struct mpa_v2_conn_params); 917 wrlen = roundup(mpalen + sizeof *req, 16); 918 skb = get_skb(skb, wrlen, GFP_KERNEL); 919 if (!skb) { 920 connect_reply_upcall(ep, -ENOMEM); 921 return -ENOMEM; 922 } 923 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 924 925 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen); 926 memset(req, 0, wrlen); 927 req->op_to_immdlen = cpu_to_be32( 928 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) | 929 FW_WR_COMPL_F | 930 FW_WR_IMMDLEN_V(mpalen)); 931 req->flowid_len16 = cpu_to_be32( 932 FW_WR_FLOWID_V(ep->hwtid) | 933 FW_WR_LEN16_V(wrlen >> 4)); 934 req->plen = cpu_to_be32(mpalen); 935 req->tunnel_to_proxy = cpu_to_be32( 936 FW_OFLD_TX_DATA_WR_FLUSH_F | 937 FW_OFLD_TX_DATA_WR_SHOVE_F); 938 939 mpa = (struct mpa_message *)(req + 1); 940 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)); 941 942 mpa->flags = 0; 943 if (crc_enabled) 944 mpa->flags |= MPA_CRC; 945 if (markers_enabled) { 946 mpa->flags |= MPA_MARKERS; 947 ep->mpa_attr.recv_marker_enabled = 1; 948 } else { 949 ep->mpa_attr.recv_marker_enabled = 0; 950 } 951 if (mpa_rev_to_use == 2) 952 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 953 954 mpa->private_data_size = htons(ep->plen); 955 mpa->revision = mpa_rev_to_use; 956 if (mpa_rev_to_use == 1) { 957 ep->tried_with_mpa_v1 = 1; 958 ep->retry_with_mpa_v1 = 0; 959 } 960 961 if (mpa_rev_to_use == 2) { 962 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 963 sizeof (struct mpa_v2_conn_params)); 964 pr_debug("%s initiator ird %u ord %u\n", __func__, ep->ird, 965 ep->ord); 966 mpa_v2_params.ird = htons((u16)ep->ird); 967 mpa_v2_params.ord = htons((u16)ep->ord); 968 969 if (peer2peer) { 970 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 971 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) 972 mpa_v2_params.ord |= 973 htons(MPA_V2_RDMA_WRITE_RTR); 974 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) 975 mpa_v2_params.ord |= 976 htons(MPA_V2_RDMA_READ_RTR); 977 } 978 memcpy(mpa->private_data, &mpa_v2_params, 979 sizeof(struct mpa_v2_conn_params)); 980 981 if (ep->plen) 982 memcpy(mpa->private_data + 983 sizeof(struct mpa_v2_conn_params), 984 ep->mpa_pkt + sizeof(*mpa), ep->plen); 985 } else 986 if (ep->plen) 987 memcpy(mpa->private_data, 988 ep->mpa_pkt + sizeof(*mpa), ep->plen); 989 990 /* 991 * Reference the mpa skb. This ensures the data area 992 * will remain in memory until the hw acks the tx. 993 * Function fw4_ack() will deref it. 994 */ 995 skb_get(skb); 996 t4_set_arp_err_handler(skb, NULL, arp_failure_discard); 997 BUG_ON(ep->mpa_skb); 998 ep->mpa_skb = skb; 999 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1000 if (ret) 1001 return ret; 1002 start_ep_timer(ep); 1003 __state_set(&ep->com, MPA_REQ_SENT); 1004 ep->mpa_attr.initiator = 1; 1005 ep->snd_seq += mpalen; 1006 return ret; 1007 } 1008 1009 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen) 1010 { 1011 int mpalen, wrlen; 1012 struct fw_ofld_tx_data_wr *req; 1013 struct mpa_message *mpa; 1014 struct sk_buff *skb; 1015 struct mpa_v2_conn_params mpa_v2_params; 1016 1017 pr_debug("%s ep %p tid %u pd_len %d\n", 1018 __func__, ep, ep->hwtid, ep->plen); 1019 1020 mpalen = sizeof(*mpa) + plen; 1021 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) 1022 mpalen += sizeof(struct mpa_v2_conn_params); 1023 wrlen = roundup(mpalen + sizeof *req, 16); 1024 1025 skb = get_skb(NULL, wrlen, GFP_KERNEL); 1026 if (!skb) { 1027 pr_err("%s - cannot alloc skb!\n", __func__); 1028 return -ENOMEM; 1029 } 1030 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 1031 1032 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen); 1033 memset(req, 0, wrlen); 1034 req->op_to_immdlen = cpu_to_be32( 1035 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) | 1036 FW_WR_COMPL_F | 1037 FW_WR_IMMDLEN_V(mpalen)); 1038 req->flowid_len16 = cpu_to_be32( 1039 FW_WR_FLOWID_V(ep->hwtid) | 1040 FW_WR_LEN16_V(wrlen >> 4)); 1041 req->plen = cpu_to_be32(mpalen); 1042 req->tunnel_to_proxy = cpu_to_be32( 1043 FW_OFLD_TX_DATA_WR_FLUSH_F | 1044 FW_OFLD_TX_DATA_WR_SHOVE_F); 1045 1046 mpa = (struct mpa_message *)(req + 1); 1047 memset(mpa, 0, sizeof(*mpa)); 1048 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1049 mpa->flags = MPA_REJECT; 1050 mpa->revision = ep->mpa_attr.version; 1051 mpa->private_data_size = htons(plen); 1052 1053 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1054 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1055 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1056 sizeof (struct mpa_v2_conn_params)); 1057 mpa_v2_params.ird = htons(((u16)ep->ird) | 1058 (peer2peer ? MPA_V2_PEER2PEER_MODEL : 1059 0)); 1060 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ? 1061 (p2p_type == 1062 FW_RI_INIT_P2PTYPE_RDMA_WRITE ? 1063 MPA_V2_RDMA_WRITE_RTR : p2p_type == 1064 FW_RI_INIT_P2PTYPE_READ_REQ ? 1065 MPA_V2_RDMA_READ_RTR : 0) : 0)); 1066 memcpy(mpa->private_data, &mpa_v2_params, 1067 sizeof(struct mpa_v2_conn_params)); 1068 1069 if (ep->plen) 1070 memcpy(mpa->private_data + 1071 sizeof(struct mpa_v2_conn_params), pdata, plen); 1072 } else 1073 if (plen) 1074 memcpy(mpa->private_data, pdata, plen); 1075 1076 /* 1077 * Reference the mpa skb again. This ensures the data area 1078 * will remain in memory until the hw acks the tx. 1079 * Function fw4_ack() will deref it. 1080 */ 1081 skb_get(skb); 1082 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 1083 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure); 1084 BUG_ON(ep->mpa_skb); 1085 ep->mpa_skb = skb; 1086 ep->snd_seq += mpalen; 1087 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1088 } 1089 1090 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen) 1091 { 1092 int mpalen, wrlen; 1093 struct fw_ofld_tx_data_wr *req; 1094 struct mpa_message *mpa; 1095 struct sk_buff *skb; 1096 struct mpa_v2_conn_params mpa_v2_params; 1097 1098 pr_debug("%s ep %p tid %u pd_len %d\n", 1099 __func__, ep, ep->hwtid, ep->plen); 1100 1101 mpalen = sizeof(*mpa) + plen; 1102 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) 1103 mpalen += sizeof(struct mpa_v2_conn_params); 1104 wrlen = roundup(mpalen + sizeof *req, 16); 1105 1106 skb = get_skb(NULL, wrlen, GFP_KERNEL); 1107 if (!skb) { 1108 pr_err("%s - cannot alloc skb!\n", __func__); 1109 return -ENOMEM; 1110 } 1111 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 1112 1113 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen); 1114 memset(req, 0, wrlen); 1115 req->op_to_immdlen = cpu_to_be32( 1116 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) | 1117 FW_WR_COMPL_F | 1118 FW_WR_IMMDLEN_V(mpalen)); 1119 req->flowid_len16 = cpu_to_be32( 1120 FW_WR_FLOWID_V(ep->hwtid) | 1121 FW_WR_LEN16_V(wrlen >> 4)); 1122 req->plen = cpu_to_be32(mpalen); 1123 req->tunnel_to_proxy = cpu_to_be32( 1124 FW_OFLD_TX_DATA_WR_FLUSH_F | 1125 FW_OFLD_TX_DATA_WR_SHOVE_F); 1126 1127 mpa = (struct mpa_message *)(req + 1); 1128 memset(mpa, 0, sizeof(*mpa)); 1129 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1130 mpa->flags = 0; 1131 if (ep->mpa_attr.crc_enabled) 1132 mpa->flags |= MPA_CRC; 1133 if (ep->mpa_attr.recv_marker_enabled) 1134 mpa->flags |= MPA_MARKERS; 1135 mpa->revision = ep->mpa_attr.version; 1136 mpa->private_data_size = htons(plen); 1137 1138 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1139 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1140 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1141 sizeof (struct mpa_v2_conn_params)); 1142 mpa_v2_params.ird = htons((u16)ep->ird); 1143 mpa_v2_params.ord = htons((u16)ep->ord); 1144 if (peer2peer && (ep->mpa_attr.p2p_type != 1145 FW_RI_INIT_P2PTYPE_DISABLED)) { 1146 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1147 1148 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) 1149 mpa_v2_params.ord |= 1150 htons(MPA_V2_RDMA_WRITE_RTR); 1151 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) 1152 mpa_v2_params.ord |= 1153 htons(MPA_V2_RDMA_READ_RTR); 1154 } 1155 1156 memcpy(mpa->private_data, &mpa_v2_params, 1157 sizeof(struct mpa_v2_conn_params)); 1158 1159 if (ep->plen) 1160 memcpy(mpa->private_data + 1161 sizeof(struct mpa_v2_conn_params), pdata, plen); 1162 } else 1163 if (plen) 1164 memcpy(mpa->private_data, pdata, plen); 1165 1166 /* 1167 * Reference the mpa skb. This ensures the data area 1168 * will remain in memory until the hw acks the tx. 1169 * Function fw4_ack() will deref it. 1170 */ 1171 skb_get(skb); 1172 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure); 1173 ep->mpa_skb = skb; 1174 __state_set(&ep->com, MPA_REP_SENT); 1175 ep->snd_seq += mpalen; 1176 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1177 } 1178 1179 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb) 1180 { 1181 struct c4iw_ep *ep; 1182 struct cpl_act_establish *req = cplhdr(skb); 1183 unsigned int tid = GET_TID(req); 1184 unsigned int atid = TID_TID_G(ntohl(req->tos_atid)); 1185 struct tid_info *t = dev->rdev.lldi.tids; 1186 int ret; 1187 1188 ep = lookup_atid(t, atid); 1189 1190 pr_debug("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid, 1191 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn)); 1192 1193 mutex_lock(&ep->com.mutex); 1194 dst_confirm(ep->dst); 1195 1196 /* setup the hwtid for this connection */ 1197 ep->hwtid = tid; 1198 cxgb4_insert_tid(t, ep, tid); 1199 insert_ep_tid(ep); 1200 1201 ep->snd_seq = be32_to_cpu(req->snd_isn); 1202 ep->rcv_seq = be32_to_cpu(req->rcv_isn); 1203 1204 set_emss(ep, ntohs(req->tcp_opt)); 1205 1206 /* dealloc the atid */ 1207 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid); 1208 cxgb4_free_atid(t, atid); 1209 set_bit(ACT_ESTAB, &ep->com.history); 1210 1211 /* start MPA negotiation */ 1212 ret = send_flowc(ep); 1213 if (ret) 1214 goto err; 1215 if (ep->retry_with_mpa_v1) 1216 ret = send_mpa_req(ep, skb, 1); 1217 else 1218 ret = send_mpa_req(ep, skb, mpa_rev); 1219 if (ret) 1220 goto err; 1221 mutex_unlock(&ep->com.mutex); 1222 return 0; 1223 err: 1224 mutex_unlock(&ep->com.mutex); 1225 connect_reply_upcall(ep, -ENOMEM); 1226 c4iw_ep_disconnect(ep, 0, GFP_KERNEL); 1227 return 0; 1228 } 1229 1230 static void close_complete_upcall(struct c4iw_ep *ep, int status) 1231 { 1232 struct iw_cm_event event; 1233 1234 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1235 memset(&event, 0, sizeof(event)); 1236 event.event = IW_CM_EVENT_CLOSE; 1237 event.status = status; 1238 if (ep->com.cm_id) { 1239 pr_debug("close complete delivered ep %p cm_id %p tid %u\n", 1240 ep, ep->com.cm_id, ep->hwtid); 1241 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1242 deref_cm_id(&ep->com); 1243 set_bit(CLOSE_UPCALL, &ep->com.history); 1244 } 1245 } 1246 1247 static void peer_close_upcall(struct c4iw_ep *ep) 1248 { 1249 struct iw_cm_event event; 1250 1251 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1252 memset(&event, 0, sizeof(event)); 1253 event.event = IW_CM_EVENT_DISCONNECT; 1254 if (ep->com.cm_id) { 1255 pr_debug("peer close delivered ep %p cm_id %p tid %u\n", 1256 ep, ep->com.cm_id, ep->hwtid); 1257 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1258 set_bit(DISCONN_UPCALL, &ep->com.history); 1259 } 1260 } 1261 1262 static void peer_abort_upcall(struct c4iw_ep *ep) 1263 { 1264 struct iw_cm_event event; 1265 1266 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1267 memset(&event, 0, sizeof(event)); 1268 event.event = IW_CM_EVENT_CLOSE; 1269 event.status = -ECONNRESET; 1270 if (ep->com.cm_id) { 1271 pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep, 1272 ep->com.cm_id, ep->hwtid); 1273 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1274 deref_cm_id(&ep->com); 1275 set_bit(ABORT_UPCALL, &ep->com.history); 1276 } 1277 } 1278 1279 static void connect_reply_upcall(struct c4iw_ep *ep, int status) 1280 { 1281 struct iw_cm_event event; 1282 1283 pr_debug("%s ep %p tid %u status %d\n", 1284 __func__, ep, ep->hwtid, status); 1285 memset(&event, 0, sizeof(event)); 1286 event.event = IW_CM_EVENT_CONNECT_REPLY; 1287 event.status = status; 1288 memcpy(&event.local_addr, &ep->com.local_addr, 1289 sizeof(ep->com.local_addr)); 1290 memcpy(&event.remote_addr, &ep->com.remote_addr, 1291 sizeof(ep->com.remote_addr)); 1292 1293 if ((status == 0) || (status == -ECONNREFUSED)) { 1294 if (!ep->tried_with_mpa_v1) { 1295 /* this means MPA_v2 is used */ 1296 event.ord = ep->ird; 1297 event.ird = ep->ord; 1298 event.private_data_len = ep->plen - 1299 sizeof(struct mpa_v2_conn_params); 1300 event.private_data = ep->mpa_pkt + 1301 sizeof(struct mpa_message) + 1302 sizeof(struct mpa_v2_conn_params); 1303 } else { 1304 /* this means MPA_v1 is used */ 1305 event.ord = cur_max_read_depth(ep->com.dev); 1306 event.ird = cur_max_read_depth(ep->com.dev); 1307 event.private_data_len = ep->plen; 1308 event.private_data = ep->mpa_pkt + 1309 sizeof(struct mpa_message); 1310 } 1311 } 1312 1313 pr_debug("%s ep %p tid %u status %d\n", __func__, ep, 1314 ep->hwtid, status); 1315 set_bit(CONN_RPL_UPCALL, &ep->com.history); 1316 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1317 1318 if (status < 0) 1319 deref_cm_id(&ep->com); 1320 } 1321 1322 static int connect_request_upcall(struct c4iw_ep *ep) 1323 { 1324 struct iw_cm_event event; 1325 int ret; 1326 1327 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1328 memset(&event, 0, sizeof(event)); 1329 event.event = IW_CM_EVENT_CONNECT_REQUEST; 1330 memcpy(&event.local_addr, &ep->com.local_addr, 1331 sizeof(ep->com.local_addr)); 1332 memcpy(&event.remote_addr, &ep->com.remote_addr, 1333 sizeof(ep->com.remote_addr)); 1334 event.provider_data = ep; 1335 if (!ep->tried_with_mpa_v1) { 1336 /* this means MPA_v2 is used */ 1337 event.ord = ep->ord; 1338 event.ird = ep->ird; 1339 event.private_data_len = ep->plen - 1340 sizeof(struct mpa_v2_conn_params); 1341 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) + 1342 sizeof(struct mpa_v2_conn_params); 1343 } else { 1344 /* this means MPA_v1 is used. Send max supported */ 1345 event.ord = cur_max_read_depth(ep->com.dev); 1346 event.ird = cur_max_read_depth(ep->com.dev); 1347 event.private_data_len = ep->plen; 1348 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); 1349 } 1350 c4iw_get_ep(&ep->com); 1351 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id, 1352 &event); 1353 if (ret) 1354 c4iw_put_ep(&ep->com); 1355 set_bit(CONNREQ_UPCALL, &ep->com.history); 1356 c4iw_put_ep(&ep->parent_ep->com); 1357 return ret; 1358 } 1359 1360 static void established_upcall(struct c4iw_ep *ep) 1361 { 1362 struct iw_cm_event event; 1363 1364 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1365 memset(&event, 0, sizeof(event)); 1366 event.event = IW_CM_EVENT_ESTABLISHED; 1367 event.ird = ep->ord; 1368 event.ord = ep->ird; 1369 if (ep->com.cm_id) { 1370 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1371 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1372 set_bit(ESTAB_UPCALL, &ep->com.history); 1373 } 1374 } 1375 1376 static int update_rx_credits(struct c4iw_ep *ep, u32 credits) 1377 { 1378 struct sk_buff *skb; 1379 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16); 1380 u32 credit_dack; 1381 1382 pr_debug("%s ep %p tid %u credits %u\n", 1383 __func__, ep, ep->hwtid, credits); 1384 skb = get_skb(NULL, wrlen, GFP_KERNEL); 1385 if (!skb) { 1386 pr_err("update_rx_credits - cannot alloc skb!\n"); 1387 return 0; 1388 } 1389 1390 /* 1391 * If we couldn't specify the entire rcv window at connection setup 1392 * due to the limit in the number of bits in the RCV_BUFSIZ field, 1393 * then add the overage in to the credits returned. 1394 */ 1395 if (ep->rcv_win > RCV_BUFSIZ_M * 1024) 1396 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024; 1397 1398 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F | 1399 RX_DACK_MODE_V(dack_mode); 1400 1401 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx, 1402 credit_dack); 1403 1404 c4iw_ofld_send(&ep->com.dev->rdev, skb); 1405 return credits; 1406 } 1407 1408 #define RELAXED_IRD_NEGOTIATION 1 1409 1410 /* 1411 * process_mpa_reply - process streaming mode MPA reply 1412 * 1413 * Returns: 1414 * 1415 * 0 upon success indicating a connect request was delivered to the ULP 1416 * or the mpa request is incomplete but valid so far. 1417 * 1418 * 1 if a failure requires the caller to close the connection. 1419 * 1420 * 2 if a failure requires the caller to abort the connection. 1421 */ 1422 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb) 1423 { 1424 struct mpa_message *mpa; 1425 struct mpa_v2_conn_params *mpa_v2_params; 1426 u16 plen; 1427 u16 resp_ird, resp_ord; 1428 u8 rtr_mismatch = 0, insuff_ird = 0; 1429 struct c4iw_qp_attributes attrs; 1430 enum c4iw_qp_attr_mask mask; 1431 int err; 1432 int disconnect = 0; 1433 1434 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1435 1436 /* 1437 * If we get more than the supported amount of private data 1438 * then we must fail this connection. 1439 */ 1440 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) { 1441 err = -EINVAL; 1442 goto err_stop_timer; 1443 } 1444 1445 /* 1446 * copy the new data into our accumulation buffer. 1447 */ 1448 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]), 1449 skb->len); 1450 ep->mpa_pkt_len += skb->len; 1451 1452 /* 1453 * if we don't even have the mpa message, then bail. 1454 */ 1455 if (ep->mpa_pkt_len < sizeof(*mpa)) 1456 return 0; 1457 mpa = (struct mpa_message *) ep->mpa_pkt; 1458 1459 /* Validate MPA header. */ 1460 if (mpa->revision > mpa_rev) { 1461 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n", 1462 __func__, mpa_rev, mpa->revision); 1463 err = -EPROTO; 1464 goto err_stop_timer; 1465 } 1466 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) { 1467 err = -EPROTO; 1468 goto err_stop_timer; 1469 } 1470 1471 plen = ntohs(mpa->private_data_size); 1472 1473 /* 1474 * Fail if there's too much private data. 1475 */ 1476 if (plen > MPA_MAX_PRIVATE_DATA) { 1477 err = -EPROTO; 1478 goto err_stop_timer; 1479 } 1480 1481 /* 1482 * If plen does not account for pkt size 1483 */ 1484 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) { 1485 err = -EPROTO; 1486 goto err_stop_timer; 1487 } 1488 1489 ep->plen = (u8) plen; 1490 1491 /* 1492 * If we don't have all the pdata yet, then bail. 1493 * We'll continue process when more data arrives. 1494 */ 1495 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) 1496 return 0; 1497 1498 if (mpa->flags & MPA_REJECT) { 1499 err = -ECONNREFUSED; 1500 goto err_stop_timer; 1501 } 1502 1503 /* 1504 * Stop mpa timer. If it expired, then 1505 * we ignore the MPA reply. process_timeout() 1506 * will abort the connection. 1507 */ 1508 if (stop_ep_timer(ep)) 1509 return 0; 1510 1511 /* 1512 * If we get here we have accumulated the entire mpa 1513 * start reply message including private data. And 1514 * the MPA header is valid. 1515 */ 1516 __state_set(&ep->com, FPDU_MODE); 1517 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1518 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1519 ep->mpa_attr.version = mpa->revision; 1520 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1521 1522 if (mpa->revision == 2) { 1523 ep->mpa_attr.enhanced_rdma_conn = 1524 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 1525 if (ep->mpa_attr.enhanced_rdma_conn) { 1526 mpa_v2_params = (struct mpa_v2_conn_params *) 1527 (ep->mpa_pkt + sizeof(*mpa)); 1528 resp_ird = ntohs(mpa_v2_params->ird) & 1529 MPA_V2_IRD_ORD_MASK; 1530 resp_ord = ntohs(mpa_v2_params->ord) & 1531 MPA_V2_IRD_ORD_MASK; 1532 pr_debug("%s responder ird %u ord %u ep ird %u ord %u\n", 1533 __func__, 1534 resp_ird, resp_ord, ep->ird, ep->ord); 1535 1536 /* 1537 * This is a double-check. Ideally, below checks are 1538 * not required since ird/ord stuff has been taken 1539 * care of in c4iw_accept_cr 1540 */ 1541 if (ep->ird < resp_ord) { 1542 if (RELAXED_IRD_NEGOTIATION && resp_ord <= 1543 ep->com.dev->rdev.lldi.max_ordird_qp) 1544 ep->ird = resp_ord; 1545 else 1546 insuff_ird = 1; 1547 } else if (ep->ird > resp_ord) { 1548 ep->ird = resp_ord; 1549 } 1550 if (ep->ord > resp_ird) { 1551 if (RELAXED_IRD_NEGOTIATION) 1552 ep->ord = resp_ird; 1553 else 1554 insuff_ird = 1; 1555 } 1556 if (insuff_ird) { 1557 err = -ENOMEM; 1558 ep->ird = resp_ord; 1559 ep->ord = resp_ird; 1560 } 1561 1562 if (ntohs(mpa_v2_params->ird) & 1563 MPA_V2_PEER2PEER_MODEL) { 1564 if (ntohs(mpa_v2_params->ord) & 1565 MPA_V2_RDMA_WRITE_RTR) 1566 ep->mpa_attr.p2p_type = 1567 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 1568 else if (ntohs(mpa_v2_params->ord) & 1569 MPA_V2_RDMA_READ_RTR) 1570 ep->mpa_attr.p2p_type = 1571 FW_RI_INIT_P2PTYPE_READ_REQ; 1572 } 1573 } 1574 } else if (mpa->revision == 1) 1575 if (peer2peer) 1576 ep->mpa_attr.p2p_type = p2p_type; 1577 1578 pr_debug("%s - crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n", 1579 __func__, ep->mpa_attr.crc_enabled, 1580 ep->mpa_attr.recv_marker_enabled, 1581 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, 1582 ep->mpa_attr.p2p_type, p2p_type); 1583 1584 /* 1585 * If responder's RTR does not match with that of initiator, assign 1586 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not 1587 * generated when moving QP to RTS state. 1588 * A TERM message will be sent after QP has moved to RTS state 1589 */ 1590 if ((ep->mpa_attr.version == 2) && peer2peer && 1591 (ep->mpa_attr.p2p_type != p2p_type)) { 1592 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1593 rtr_mismatch = 1; 1594 } 1595 1596 attrs.mpa_attr = ep->mpa_attr; 1597 attrs.max_ird = ep->ird; 1598 attrs.max_ord = ep->ord; 1599 attrs.llp_stream_handle = ep; 1600 attrs.next_state = C4IW_QP_STATE_RTS; 1601 1602 mask = C4IW_QP_ATTR_NEXT_STATE | 1603 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR | 1604 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD; 1605 1606 /* bind QP and TID with INIT_WR */ 1607 err = c4iw_modify_qp(ep->com.qp->rhp, 1608 ep->com.qp, mask, &attrs, 1); 1609 if (err) 1610 goto err; 1611 1612 /* 1613 * If responder's RTR requirement did not match with what initiator 1614 * supports, generate TERM message 1615 */ 1616 if (rtr_mismatch) { 1617 pr_err("%s: RTR mismatch, sending TERM\n", __func__); 1618 attrs.layer_etype = LAYER_MPA | DDP_LLP; 1619 attrs.ecode = MPA_NOMATCH_RTR; 1620 attrs.next_state = C4IW_QP_STATE_TERMINATE; 1621 attrs.send_term = 1; 1622 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 1623 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 1624 err = -ENOMEM; 1625 disconnect = 1; 1626 goto out; 1627 } 1628 1629 /* 1630 * Generate TERM if initiator IRD is not sufficient for responder 1631 * provided ORD. Currently, we do the same behaviour even when 1632 * responder provided IRD is also not sufficient as regards to 1633 * initiator ORD. 1634 */ 1635 if (insuff_ird) { 1636 pr_err("%s: Insufficient IRD, sending TERM\n", __func__); 1637 attrs.layer_etype = LAYER_MPA | DDP_LLP; 1638 attrs.ecode = MPA_INSUFF_IRD; 1639 attrs.next_state = C4IW_QP_STATE_TERMINATE; 1640 attrs.send_term = 1; 1641 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 1642 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 1643 err = -ENOMEM; 1644 disconnect = 1; 1645 goto out; 1646 } 1647 goto out; 1648 err_stop_timer: 1649 stop_ep_timer(ep); 1650 err: 1651 disconnect = 2; 1652 out: 1653 connect_reply_upcall(ep, err); 1654 return disconnect; 1655 } 1656 1657 /* 1658 * process_mpa_request - process streaming mode MPA request 1659 * 1660 * Returns: 1661 * 1662 * 0 upon success indicating a connect request was delivered to the ULP 1663 * or the mpa request is incomplete but valid so far. 1664 * 1665 * 1 if a failure requires the caller to close the connection. 1666 * 1667 * 2 if a failure requires the caller to abort the connection. 1668 */ 1669 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb) 1670 { 1671 struct mpa_message *mpa; 1672 struct mpa_v2_conn_params *mpa_v2_params; 1673 u16 plen; 1674 1675 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1676 1677 /* 1678 * If we get more than the supported amount of private data 1679 * then we must fail this connection. 1680 */ 1681 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) 1682 goto err_stop_timer; 1683 1684 pr_debug("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__); 1685 1686 /* 1687 * Copy the new data into our accumulation buffer. 1688 */ 1689 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]), 1690 skb->len); 1691 ep->mpa_pkt_len += skb->len; 1692 1693 /* 1694 * If we don't even have the mpa message, then bail. 1695 * We'll continue process when more data arrives. 1696 */ 1697 if (ep->mpa_pkt_len < sizeof(*mpa)) 1698 return 0; 1699 1700 pr_debug("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__); 1701 mpa = (struct mpa_message *) ep->mpa_pkt; 1702 1703 /* 1704 * Validate MPA Header. 1705 */ 1706 if (mpa->revision > mpa_rev) { 1707 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n", 1708 __func__, mpa_rev, mpa->revision); 1709 goto err_stop_timer; 1710 } 1711 1712 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) 1713 goto err_stop_timer; 1714 1715 plen = ntohs(mpa->private_data_size); 1716 1717 /* 1718 * Fail if there's too much private data. 1719 */ 1720 if (plen > MPA_MAX_PRIVATE_DATA) 1721 goto err_stop_timer; 1722 1723 /* 1724 * If plen does not account for pkt size 1725 */ 1726 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) 1727 goto err_stop_timer; 1728 ep->plen = (u8) plen; 1729 1730 /* 1731 * If we don't have all the pdata yet, then bail. 1732 */ 1733 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) 1734 return 0; 1735 1736 /* 1737 * If we get here we have accumulated the entire mpa 1738 * start reply message including private data. 1739 */ 1740 ep->mpa_attr.initiator = 0; 1741 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1742 ep->mpa_attr.recv_marker_enabled = markers_enabled; 1743 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1744 ep->mpa_attr.version = mpa->revision; 1745 if (mpa->revision == 1) 1746 ep->tried_with_mpa_v1 = 1; 1747 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1748 1749 if (mpa->revision == 2) { 1750 ep->mpa_attr.enhanced_rdma_conn = 1751 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 1752 if (ep->mpa_attr.enhanced_rdma_conn) { 1753 mpa_v2_params = (struct mpa_v2_conn_params *) 1754 (ep->mpa_pkt + sizeof(*mpa)); 1755 ep->ird = ntohs(mpa_v2_params->ird) & 1756 MPA_V2_IRD_ORD_MASK; 1757 ep->ird = min_t(u32, ep->ird, 1758 cur_max_read_depth(ep->com.dev)); 1759 ep->ord = ntohs(mpa_v2_params->ord) & 1760 MPA_V2_IRD_ORD_MASK; 1761 ep->ord = min_t(u32, ep->ord, 1762 cur_max_read_depth(ep->com.dev)); 1763 pr_debug("%s initiator ird %u ord %u\n", 1764 __func__, ep->ird, ep->ord); 1765 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL) 1766 if (peer2peer) { 1767 if (ntohs(mpa_v2_params->ord) & 1768 MPA_V2_RDMA_WRITE_RTR) 1769 ep->mpa_attr.p2p_type = 1770 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 1771 else if (ntohs(mpa_v2_params->ord) & 1772 MPA_V2_RDMA_READ_RTR) 1773 ep->mpa_attr.p2p_type = 1774 FW_RI_INIT_P2PTYPE_READ_REQ; 1775 } 1776 } 1777 } else if (mpa->revision == 1) 1778 if (peer2peer) 1779 ep->mpa_attr.p2p_type = p2p_type; 1780 1781 pr_debug("%s - crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n", 1782 __func__, 1783 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 1784 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, 1785 ep->mpa_attr.p2p_type); 1786 1787 __state_set(&ep->com, MPA_REQ_RCVD); 1788 1789 /* drive upcall */ 1790 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING); 1791 if (ep->parent_ep->com.state != DEAD) { 1792 if (connect_request_upcall(ep)) 1793 goto err_unlock_parent; 1794 } else { 1795 goto err_unlock_parent; 1796 } 1797 mutex_unlock(&ep->parent_ep->com.mutex); 1798 return 0; 1799 1800 err_unlock_parent: 1801 mutex_unlock(&ep->parent_ep->com.mutex); 1802 goto err_out; 1803 err_stop_timer: 1804 (void)stop_ep_timer(ep); 1805 err_out: 1806 return 2; 1807 } 1808 1809 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb) 1810 { 1811 struct c4iw_ep *ep; 1812 struct cpl_rx_data *hdr = cplhdr(skb); 1813 unsigned int dlen = ntohs(hdr->len); 1814 unsigned int tid = GET_TID(hdr); 1815 __u8 status = hdr->status; 1816 int disconnect = 0; 1817 1818 ep = get_ep_from_tid(dev, tid); 1819 if (!ep) 1820 return 0; 1821 pr_debug("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen); 1822 skb_pull(skb, sizeof(*hdr)); 1823 skb_trim(skb, dlen); 1824 mutex_lock(&ep->com.mutex); 1825 1826 switch (ep->com.state) { 1827 case MPA_REQ_SENT: 1828 update_rx_credits(ep, dlen); 1829 ep->rcv_seq += dlen; 1830 disconnect = process_mpa_reply(ep, skb); 1831 break; 1832 case MPA_REQ_WAIT: 1833 update_rx_credits(ep, dlen); 1834 ep->rcv_seq += dlen; 1835 disconnect = process_mpa_request(ep, skb); 1836 break; 1837 case FPDU_MODE: { 1838 struct c4iw_qp_attributes attrs; 1839 1840 update_rx_credits(ep, dlen); 1841 BUG_ON(!ep->com.qp); 1842 if (status) 1843 pr_err("%s Unexpected streaming data." \ 1844 " qpid %u ep %p state %d tid %u status %d\n", 1845 __func__, ep->com.qp->wq.sq.qid, ep, 1846 ep->com.state, ep->hwtid, status); 1847 attrs.next_state = C4IW_QP_STATE_TERMINATE; 1848 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 1849 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 1850 disconnect = 1; 1851 break; 1852 } 1853 default: 1854 break; 1855 } 1856 mutex_unlock(&ep->com.mutex); 1857 if (disconnect) 1858 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL); 1859 c4iw_put_ep(&ep->com); 1860 return 0; 1861 } 1862 1863 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 1864 { 1865 struct c4iw_ep *ep; 1866 struct cpl_abort_rpl_rss *rpl = cplhdr(skb); 1867 int release = 0; 1868 unsigned int tid = GET_TID(rpl); 1869 1870 ep = get_ep_from_tid(dev, tid); 1871 if (!ep) { 1872 pr_warn("Abort rpl to freed endpoint\n"); 1873 return 0; 1874 } 1875 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1876 mutex_lock(&ep->com.mutex); 1877 switch (ep->com.state) { 1878 case ABORTING: 1879 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 1880 __state_set(&ep->com, DEAD); 1881 release = 1; 1882 break; 1883 default: 1884 pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state); 1885 break; 1886 } 1887 mutex_unlock(&ep->com.mutex); 1888 1889 if (release) 1890 release_ep_resources(ep); 1891 c4iw_put_ep(&ep->com); 1892 return 0; 1893 } 1894 1895 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid) 1896 { 1897 struct sk_buff *skb; 1898 struct fw_ofld_connection_wr *req; 1899 unsigned int mtu_idx; 1900 u32 wscale; 1901 struct sockaddr_in *sin; 1902 int win; 1903 1904 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL); 1905 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req)); 1906 memset(req, 0, sizeof(*req)); 1907 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR)); 1908 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16))); 1909 req->le.filter = cpu_to_be32(cxgb4_select_ntuple( 1910 ep->com.dev->rdev.lldi.ports[0], 1911 ep->l2t)); 1912 sin = (struct sockaddr_in *)&ep->com.local_addr; 1913 req->le.lport = sin->sin_port; 1914 req->le.u.ipv4.lip = sin->sin_addr.s_addr; 1915 sin = (struct sockaddr_in *)&ep->com.remote_addr; 1916 req->le.pport = sin->sin_port; 1917 req->le.u.ipv4.pip = sin->sin_addr.s_addr; 1918 req->tcb.t_state_to_astid = 1919 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) | 1920 FW_OFLD_CONNECTION_WR_ASTID_V(atid)); 1921 req->tcb.cplrxdataack_cplpassacceptrpl = 1922 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F); 1923 req->tcb.tx_max = (__force __be32) jiffies; 1924 req->tcb.rcv_adv = htons(1); 1925 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx, 1926 enable_tcp_timestamps, 1927 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1); 1928 wscale = cxgb_compute_wscale(rcv_win); 1929 1930 /* 1931 * Specify the largest window that will fit in opt0. The 1932 * remainder will be specified in the rx_data_ack. 1933 */ 1934 win = ep->rcv_win >> 10; 1935 if (win > RCV_BUFSIZ_M) 1936 win = RCV_BUFSIZ_M; 1937 1938 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F | 1939 (nocong ? NO_CONG_F : 0) | 1940 KEEP_ALIVE_F | 1941 DELACK_F | 1942 WND_SCALE_V(wscale) | 1943 MSS_IDX_V(mtu_idx) | 1944 L2T_IDX_V(ep->l2t->idx) | 1945 TX_CHAN_V(ep->tx_chan) | 1946 SMAC_SEL_V(ep->smac_idx) | 1947 DSCP_V(ep->tos >> 2) | 1948 ULP_MODE_V(ULP_MODE_TCPDDP) | 1949 RCV_BUFSIZ_V(win)); 1950 req->tcb.opt2 = (__force __be32) (PACE_V(1) | 1951 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) | 1952 RX_CHANNEL_V(0) | 1953 CCTRL_ECN_V(enable_ecn) | 1954 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid)); 1955 if (enable_tcp_timestamps) 1956 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F; 1957 if (enable_tcp_sack) 1958 req->tcb.opt2 |= (__force __be32)SACK_EN_F; 1959 if (wscale && enable_tcp_window_scaling) 1960 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F; 1961 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0); 1962 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2); 1963 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx); 1964 set_bit(ACT_OFLD_CONN, &ep->com.history); 1965 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1966 } 1967 1968 /* 1969 * Some of the error codes above implicitly indicate that there is no TID 1970 * allocated with the result of an ACT_OPEN. We use this predicate to make 1971 * that explicit. 1972 */ 1973 static inline int act_open_has_tid(int status) 1974 { 1975 return (status != CPL_ERR_TCAM_PARITY && 1976 status != CPL_ERR_TCAM_MISS && 1977 status != CPL_ERR_TCAM_FULL && 1978 status != CPL_ERR_CONN_EXIST_SYNRECV && 1979 status != CPL_ERR_CONN_EXIST); 1980 } 1981 1982 static char *neg_adv_str(unsigned int status) 1983 { 1984 switch (status) { 1985 case CPL_ERR_RTX_NEG_ADVICE: 1986 return "Retransmit timeout"; 1987 case CPL_ERR_PERSIST_NEG_ADVICE: 1988 return "Persist timeout"; 1989 case CPL_ERR_KEEPALV_NEG_ADVICE: 1990 return "Keepalive timeout"; 1991 default: 1992 return "Unknown"; 1993 } 1994 } 1995 1996 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi) 1997 { 1998 ep->snd_win = snd_win; 1999 ep->rcv_win = rcv_win; 2000 pr_debug("%s snd_win %d rcv_win %d\n", 2001 __func__, ep->snd_win, ep->rcv_win); 2002 } 2003 2004 #define ACT_OPEN_RETRY_COUNT 2 2005 2006 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip, 2007 struct dst_entry *dst, struct c4iw_dev *cdev, 2008 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos) 2009 { 2010 struct neighbour *n; 2011 int err, step; 2012 struct net_device *pdev; 2013 2014 n = dst_neigh_lookup(dst, peer_ip); 2015 if (!n) 2016 return -ENODEV; 2017 2018 rcu_read_lock(); 2019 err = -ENOMEM; 2020 if (n->dev->flags & IFF_LOOPBACK) { 2021 if (iptype == 4) 2022 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip); 2023 else if (IS_ENABLED(CONFIG_IPV6)) 2024 for_each_netdev(&init_net, pdev) { 2025 if (ipv6_chk_addr(&init_net, 2026 (struct in6_addr *)peer_ip, 2027 pdev, 1)) 2028 break; 2029 } 2030 else 2031 pdev = NULL; 2032 2033 if (!pdev) { 2034 err = -ENODEV; 2035 goto out; 2036 } 2037 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t, 2038 n, pdev, rt_tos2priority(tos)); 2039 if (!ep->l2t) { 2040 dev_put(pdev); 2041 goto out; 2042 } 2043 ep->mtu = pdev->mtu; 2044 ep->tx_chan = cxgb4_port_chan(pdev); 2045 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type, 2046 cxgb4_port_viid(pdev)); 2047 step = cdev->rdev.lldi.ntxq / 2048 cdev->rdev.lldi.nchan; 2049 ep->txq_idx = cxgb4_port_idx(pdev) * step; 2050 step = cdev->rdev.lldi.nrxq / 2051 cdev->rdev.lldi.nchan; 2052 ep->ctrlq_idx = cxgb4_port_idx(pdev); 2053 ep->rss_qid = cdev->rdev.lldi.rxq_ids[ 2054 cxgb4_port_idx(pdev) * step]; 2055 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev)); 2056 dev_put(pdev); 2057 } else { 2058 pdev = get_real_dev(n->dev); 2059 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t, 2060 n, pdev, 0); 2061 if (!ep->l2t) 2062 goto out; 2063 ep->mtu = dst_mtu(dst); 2064 ep->tx_chan = cxgb4_port_chan(pdev); 2065 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type, 2066 cxgb4_port_viid(pdev)); 2067 step = cdev->rdev.lldi.ntxq / 2068 cdev->rdev.lldi.nchan; 2069 ep->txq_idx = cxgb4_port_idx(pdev) * step; 2070 ep->ctrlq_idx = cxgb4_port_idx(pdev); 2071 step = cdev->rdev.lldi.nrxq / 2072 cdev->rdev.lldi.nchan; 2073 ep->rss_qid = cdev->rdev.lldi.rxq_ids[ 2074 cxgb4_port_idx(pdev) * step]; 2075 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev)); 2076 2077 if (clear_mpa_v1) { 2078 ep->retry_with_mpa_v1 = 0; 2079 ep->tried_with_mpa_v1 = 0; 2080 } 2081 } 2082 err = 0; 2083 out: 2084 rcu_read_unlock(); 2085 2086 neigh_release(n); 2087 2088 return err; 2089 } 2090 2091 static int c4iw_reconnect(struct c4iw_ep *ep) 2092 { 2093 int err = 0; 2094 int size = 0; 2095 struct sockaddr_in *laddr = (struct sockaddr_in *) 2096 &ep->com.cm_id->m_local_addr; 2097 struct sockaddr_in *raddr = (struct sockaddr_in *) 2098 &ep->com.cm_id->m_remote_addr; 2099 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *) 2100 &ep->com.cm_id->m_local_addr; 2101 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *) 2102 &ep->com.cm_id->m_remote_addr; 2103 int iptype; 2104 __u8 *ra; 2105 2106 pr_debug("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id); 2107 init_timer(&ep->timer); 2108 c4iw_init_wr_wait(&ep->com.wr_wait); 2109 2110 /* When MPA revision is different on nodes, the node with MPA_rev=2 2111 * tries to reconnect with MPA_rev 1 for the same EP through 2112 * c4iw_reconnect(), where the same EP is assigned with new tid for 2113 * further connection establishment. As we are using the same EP pointer 2114 * for reconnect, few skbs are used during the previous c4iw_connect(), 2115 * which leaves the EP with inadequate skbs for further 2116 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty 2117 * skb_list() during peer_abort(). Allocate skbs which is already used. 2118 */ 2119 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list)); 2120 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) { 2121 err = -ENOMEM; 2122 goto fail1; 2123 } 2124 2125 /* 2126 * Allocate an active TID to initiate a TCP connection. 2127 */ 2128 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep); 2129 if (ep->atid == -1) { 2130 pr_err("%s - cannot alloc atid\n", __func__); 2131 err = -ENOMEM; 2132 goto fail2; 2133 } 2134 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid); 2135 2136 /* find a route */ 2137 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) { 2138 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev, 2139 laddr->sin_addr.s_addr, 2140 raddr->sin_addr.s_addr, 2141 laddr->sin_port, 2142 raddr->sin_port, ep->com.cm_id->tos); 2143 iptype = 4; 2144 ra = (__u8 *)&raddr->sin_addr; 2145 } else { 2146 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi, 2147 get_real_dev, 2148 laddr6->sin6_addr.s6_addr, 2149 raddr6->sin6_addr.s6_addr, 2150 laddr6->sin6_port, 2151 raddr6->sin6_port, 0, 2152 raddr6->sin6_scope_id); 2153 iptype = 6; 2154 ra = (__u8 *)&raddr6->sin6_addr; 2155 } 2156 if (!ep->dst) { 2157 pr_err("%s - cannot find route\n", __func__); 2158 err = -EHOSTUNREACH; 2159 goto fail3; 2160 } 2161 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false, 2162 ep->com.dev->rdev.lldi.adapter_type, 2163 ep->com.cm_id->tos); 2164 if (err) { 2165 pr_err("%s - cannot alloc l2e\n", __func__); 2166 goto fail4; 2167 } 2168 2169 pr_debug("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n", 2170 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid, 2171 ep->l2t->idx); 2172 2173 state_set(&ep->com, CONNECTING); 2174 ep->tos = ep->com.cm_id->tos; 2175 2176 /* send connect request to rnic */ 2177 err = send_connect(ep); 2178 if (!err) 2179 goto out; 2180 2181 cxgb4_l2t_release(ep->l2t); 2182 fail4: 2183 dst_release(ep->dst); 2184 fail3: 2185 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid); 2186 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid); 2187 fail2: 2188 /* 2189 * remember to send notification to upper layer. 2190 * We are in here so the upper layer is not aware that this is 2191 * re-connect attempt and so, upper layer is still waiting for 2192 * response of 1st connect request. 2193 */ 2194 connect_reply_upcall(ep, -ECONNRESET); 2195 fail1: 2196 c4iw_put_ep(&ep->com); 2197 out: 2198 return err; 2199 } 2200 2201 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2202 { 2203 struct c4iw_ep *ep; 2204 struct cpl_act_open_rpl *rpl = cplhdr(skb); 2205 unsigned int atid = TID_TID_G(AOPEN_ATID_G( 2206 ntohl(rpl->atid_status))); 2207 struct tid_info *t = dev->rdev.lldi.tids; 2208 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status)); 2209 struct sockaddr_in *la; 2210 struct sockaddr_in *ra; 2211 struct sockaddr_in6 *la6; 2212 struct sockaddr_in6 *ra6; 2213 int ret = 0; 2214 2215 ep = lookup_atid(t, atid); 2216 la = (struct sockaddr_in *)&ep->com.local_addr; 2217 ra = (struct sockaddr_in *)&ep->com.remote_addr; 2218 la6 = (struct sockaddr_in6 *)&ep->com.local_addr; 2219 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr; 2220 2221 pr_debug("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid, 2222 status, status2errno(status)); 2223 2224 if (cxgb_is_neg_adv(status)) { 2225 pr_debug("%s Connection problems for atid %u status %u (%s)\n", 2226 __func__, atid, status, neg_adv_str(status)); 2227 ep->stats.connect_neg_adv++; 2228 mutex_lock(&dev->rdev.stats.lock); 2229 dev->rdev.stats.neg_adv++; 2230 mutex_unlock(&dev->rdev.stats.lock); 2231 return 0; 2232 } 2233 2234 set_bit(ACT_OPEN_RPL, &ep->com.history); 2235 2236 /* 2237 * Log interesting failures. 2238 */ 2239 switch (status) { 2240 case CPL_ERR_CONN_RESET: 2241 case CPL_ERR_CONN_TIMEDOUT: 2242 break; 2243 case CPL_ERR_TCAM_FULL: 2244 mutex_lock(&dev->rdev.stats.lock); 2245 dev->rdev.stats.tcam_full++; 2246 mutex_unlock(&dev->rdev.stats.lock); 2247 if (ep->com.local_addr.ss_family == AF_INET && 2248 dev->rdev.lldi.enable_fw_ofld_conn) { 2249 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G( 2250 ntohl(rpl->atid_status)))); 2251 if (ret) 2252 goto fail; 2253 return 0; 2254 } 2255 break; 2256 case CPL_ERR_CONN_EXIST: 2257 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) { 2258 set_bit(ACT_RETRY_INUSE, &ep->com.history); 2259 if (ep->com.remote_addr.ss_family == AF_INET6) { 2260 struct sockaddr_in6 *sin6 = 2261 (struct sockaddr_in6 *) 2262 &ep->com.local_addr; 2263 cxgb4_clip_release( 2264 ep->com.dev->rdev.lldi.ports[0], 2265 (const u32 *) 2266 &sin6->sin6_addr.s6_addr, 1); 2267 } 2268 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, 2269 atid); 2270 cxgb4_free_atid(t, atid); 2271 dst_release(ep->dst); 2272 cxgb4_l2t_release(ep->l2t); 2273 c4iw_reconnect(ep); 2274 return 0; 2275 } 2276 break; 2277 default: 2278 if (ep->com.local_addr.ss_family == AF_INET) { 2279 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n", 2280 atid, status, status2errno(status), 2281 &la->sin_addr.s_addr, ntohs(la->sin_port), 2282 &ra->sin_addr.s_addr, ntohs(ra->sin_port)); 2283 } else { 2284 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n", 2285 atid, status, status2errno(status), 2286 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port), 2287 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port)); 2288 } 2289 break; 2290 } 2291 2292 fail: 2293 connect_reply_upcall(ep, status2errno(status)); 2294 state_set(&ep->com, DEAD); 2295 2296 if (ep->com.remote_addr.ss_family == AF_INET6) { 2297 struct sockaddr_in6 *sin6 = 2298 (struct sockaddr_in6 *)&ep->com.local_addr; 2299 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 2300 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 2301 } 2302 if (status && act_open_has_tid(status)) 2303 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl)); 2304 2305 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid); 2306 cxgb4_free_atid(t, atid); 2307 dst_release(ep->dst); 2308 cxgb4_l2t_release(ep->l2t); 2309 c4iw_put_ep(&ep->com); 2310 2311 return 0; 2312 } 2313 2314 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2315 { 2316 struct cpl_pass_open_rpl *rpl = cplhdr(skb); 2317 unsigned int stid = GET_TID(rpl); 2318 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid); 2319 2320 if (!ep) { 2321 pr_debug("%s stid %d lookup failure!\n", __func__, stid); 2322 goto out; 2323 } 2324 pr_debug("%s ep %p status %d error %d\n", __func__, ep, 2325 rpl->status, status2errno(rpl->status)); 2326 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status)); 2327 c4iw_put_ep(&ep->com); 2328 out: 2329 return 0; 2330 } 2331 2332 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2333 { 2334 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb); 2335 unsigned int stid = GET_TID(rpl); 2336 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid); 2337 2338 pr_debug("%s ep %p\n", __func__, ep); 2339 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status)); 2340 c4iw_put_ep(&ep->com); 2341 return 0; 2342 } 2343 2344 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb, 2345 struct cpl_pass_accept_req *req) 2346 { 2347 struct cpl_pass_accept_rpl *rpl; 2348 unsigned int mtu_idx; 2349 u64 opt0; 2350 u32 opt2; 2351 u32 wscale; 2352 struct cpl_t5_pass_accept_rpl *rpl5 = NULL; 2353 int win; 2354 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type; 2355 2356 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2357 BUG_ON(skb_cloned(skb)); 2358 2359 skb_get(skb); 2360 rpl = cplhdr(skb); 2361 if (!is_t4(adapter_type)) { 2362 skb_trim(skb, roundup(sizeof(*rpl5), 16)); 2363 rpl5 = (void *)rpl; 2364 INIT_TP_WR(rpl5, ep->hwtid); 2365 } else { 2366 skb_trim(skb, sizeof(*rpl)); 2367 INIT_TP_WR(rpl, ep->hwtid); 2368 } 2369 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, 2370 ep->hwtid)); 2371 2372 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx, 2373 enable_tcp_timestamps && req->tcpopt.tstamp, 2374 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1); 2375 wscale = cxgb_compute_wscale(rcv_win); 2376 2377 /* 2378 * Specify the largest window that will fit in opt0. The 2379 * remainder will be specified in the rx_data_ack. 2380 */ 2381 win = ep->rcv_win >> 10; 2382 if (win > RCV_BUFSIZ_M) 2383 win = RCV_BUFSIZ_M; 2384 opt0 = (nocong ? NO_CONG_F : 0) | 2385 KEEP_ALIVE_F | 2386 DELACK_F | 2387 WND_SCALE_V(wscale) | 2388 MSS_IDX_V(mtu_idx) | 2389 L2T_IDX_V(ep->l2t->idx) | 2390 TX_CHAN_V(ep->tx_chan) | 2391 SMAC_SEL_V(ep->smac_idx) | 2392 DSCP_V(ep->tos >> 2) | 2393 ULP_MODE_V(ULP_MODE_TCPDDP) | 2394 RCV_BUFSIZ_V(win); 2395 opt2 = RX_CHANNEL_V(0) | 2396 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid); 2397 2398 if (enable_tcp_timestamps && req->tcpopt.tstamp) 2399 opt2 |= TSTAMPS_EN_F; 2400 if (enable_tcp_sack && req->tcpopt.sack) 2401 opt2 |= SACK_EN_F; 2402 if (wscale && enable_tcp_window_scaling) 2403 opt2 |= WND_SCALE_EN_F; 2404 if (enable_ecn) { 2405 const struct tcphdr *tcph; 2406 u32 hlen = ntohl(req->hdr_len); 2407 2408 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5) 2409 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) + 2410 IP_HDR_LEN_G(hlen); 2411 else 2412 tcph = (const void *)(req + 1) + 2413 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen); 2414 if (tcph->ece && tcph->cwr) 2415 opt2 |= CCTRL_ECN_V(1); 2416 } 2417 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) { 2418 u32 isn = (prandom_u32() & ~7UL) - 1; 2419 opt2 |= T5_OPT_2_VALID_F; 2420 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE); 2421 opt2 |= T5_ISS_F; 2422 rpl5 = (void *)rpl; 2423 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16)); 2424 if (peer2peer) 2425 isn += 4; 2426 rpl5->iss = cpu_to_be32(isn); 2427 pr_debug("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss)); 2428 } 2429 2430 rpl->opt0 = cpu_to_be64(opt0); 2431 rpl->opt2 = cpu_to_be32(opt2); 2432 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx); 2433 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure); 2434 2435 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 2436 } 2437 2438 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb) 2439 { 2440 pr_debug("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid); 2441 BUG_ON(skb_cloned(skb)); 2442 skb_trim(skb, sizeof(struct cpl_tid_release)); 2443 release_tid(&dev->rdev, hwtid, skb); 2444 return; 2445 } 2446 2447 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb) 2448 { 2449 struct c4iw_ep *child_ep = NULL, *parent_ep; 2450 struct cpl_pass_accept_req *req = cplhdr(skb); 2451 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid)); 2452 struct tid_info *t = dev->rdev.lldi.tids; 2453 unsigned int hwtid = GET_TID(req); 2454 struct dst_entry *dst; 2455 __u8 local_ip[16], peer_ip[16]; 2456 __be16 local_port, peer_port; 2457 struct sockaddr_in6 *sin6; 2458 int err; 2459 u16 peer_mss = ntohs(req->tcpopt.mss); 2460 int iptype; 2461 unsigned short hdrs; 2462 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid)); 2463 2464 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid); 2465 if (!parent_ep) { 2466 pr_debug("%s connect request on invalid stid %d\n", 2467 __func__, stid); 2468 goto reject; 2469 } 2470 2471 if (state_read(&parent_ep->com) != LISTEN) { 2472 pr_debug("%s - listening ep not in LISTEN\n", __func__); 2473 goto reject; 2474 } 2475 2476 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type, 2477 &iptype, local_ip, peer_ip, &local_port, &peer_port); 2478 2479 /* Find output route */ 2480 if (iptype == 4) { 2481 pr_debug("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n" 2482 , __func__, parent_ep, hwtid, 2483 local_ip, peer_ip, ntohs(local_port), 2484 ntohs(peer_port), peer_mss); 2485 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev, 2486 *(__be32 *)local_ip, *(__be32 *)peer_ip, 2487 local_port, peer_port, tos); 2488 } else { 2489 pr_debug("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n" 2490 , __func__, parent_ep, hwtid, 2491 local_ip, peer_ip, ntohs(local_port), 2492 ntohs(peer_port), peer_mss); 2493 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev, 2494 local_ip, peer_ip, local_port, peer_port, 2495 PASS_OPEN_TOS_G(ntohl(req->tos_stid)), 2496 ((struct sockaddr_in6 *) 2497 &parent_ep->com.local_addr)->sin6_scope_id); 2498 } 2499 if (!dst) { 2500 pr_err("%s - failed to find dst entry!\n", __func__); 2501 goto reject; 2502 } 2503 2504 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL); 2505 if (!child_ep) { 2506 pr_err("%s - failed to allocate ep entry!\n", __func__); 2507 dst_release(dst); 2508 goto reject; 2509 } 2510 2511 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false, 2512 parent_ep->com.dev->rdev.lldi.adapter_type, tos); 2513 if (err) { 2514 pr_err("%s - failed to allocate l2t entry!\n", __func__); 2515 dst_release(dst); 2516 kfree(child_ep); 2517 goto reject; 2518 } 2519 2520 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) + 2521 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0); 2522 if (peer_mss && child_ep->mtu > (peer_mss + hdrs)) 2523 child_ep->mtu = peer_mss + hdrs; 2524 2525 skb_queue_head_init(&child_ep->com.ep_skb_list); 2526 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF)) 2527 goto fail; 2528 2529 state_set(&child_ep->com, CONNECTING); 2530 child_ep->com.dev = dev; 2531 child_ep->com.cm_id = NULL; 2532 2533 if (iptype == 4) { 2534 struct sockaddr_in *sin = (struct sockaddr_in *) 2535 &child_ep->com.local_addr; 2536 2537 sin->sin_family = AF_INET; 2538 sin->sin_port = local_port; 2539 sin->sin_addr.s_addr = *(__be32 *)local_ip; 2540 2541 sin = (struct sockaddr_in *)&child_ep->com.local_addr; 2542 sin->sin_family = AF_INET; 2543 sin->sin_port = ((struct sockaddr_in *) 2544 &parent_ep->com.local_addr)->sin_port; 2545 sin->sin_addr.s_addr = *(__be32 *)local_ip; 2546 2547 sin = (struct sockaddr_in *)&child_ep->com.remote_addr; 2548 sin->sin_family = AF_INET; 2549 sin->sin_port = peer_port; 2550 sin->sin_addr.s_addr = *(__be32 *)peer_ip; 2551 } else { 2552 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr; 2553 sin6->sin6_family = PF_INET6; 2554 sin6->sin6_port = local_port; 2555 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16); 2556 2557 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr; 2558 sin6->sin6_family = PF_INET6; 2559 sin6->sin6_port = ((struct sockaddr_in6 *) 2560 &parent_ep->com.local_addr)->sin6_port; 2561 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16); 2562 2563 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr; 2564 sin6->sin6_family = PF_INET6; 2565 sin6->sin6_port = peer_port; 2566 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16); 2567 } 2568 2569 c4iw_get_ep(&parent_ep->com); 2570 child_ep->parent_ep = parent_ep; 2571 child_ep->tos = tos; 2572 child_ep->dst = dst; 2573 child_ep->hwtid = hwtid; 2574 2575 pr_debug("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__, 2576 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid); 2577 2578 init_timer(&child_ep->timer); 2579 cxgb4_insert_tid(t, child_ep, hwtid); 2580 insert_ep_tid(child_ep); 2581 if (accept_cr(child_ep, skb, req)) { 2582 c4iw_put_ep(&parent_ep->com); 2583 release_ep_resources(child_ep); 2584 } else { 2585 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history); 2586 } 2587 if (iptype == 6) { 2588 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr; 2589 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0], 2590 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 2591 } 2592 goto out; 2593 fail: 2594 c4iw_put_ep(&child_ep->com); 2595 reject: 2596 reject_cr(dev, hwtid, skb); 2597 if (parent_ep) 2598 c4iw_put_ep(&parent_ep->com); 2599 out: 2600 return 0; 2601 } 2602 2603 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb) 2604 { 2605 struct c4iw_ep *ep; 2606 struct cpl_pass_establish *req = cplhdr(skb); 2607 unsigned int tid = GET_TID(req); 2608 int ret; 2609 2610 ep = get_ep_from_tid(dev, tid); 2611 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2612 ep->snd_seq = be32_to_cpu(req->snd_isn); 2613 ep->rcv_seq = be32_to_cpu(req->rcv_isn); 2614 2615 pr_debug("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid, 2616 ntohs(req->tcp_opt)); 2617 2618 set_emss(ep, ntohs(req->tcp_opt)); 2619 2620 dst_confirm(ep->dst); 2621 mutex_lock(&ep->com.mutex); 2622 ep->com.state = MPA_REQ_WAIT; 2623 start_ep_timer(ep); 2624 set_bit(PASS_ESTAB, &ep->com.history); 2625 ret = send_flowc(ep); 2626 mutex_unlock(&ep->com.mutex); 2627 if (ret) 2628 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 2629 c4iw_put_ep(&ep->com); 2630 2631 return 0; 2632 } 2633 2634 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb) 2635 { 2636 struct cpl_peer_close *hdr = cplhdr(skb); 2637 struct c4iw_ep *ep; 2638 struct c4iw_qp_attributes attrs; 2639 int disconnect = 1; 2640 int release = 0; 2641 unsigned int tid = GET_TID(hdr); 2642 int ret; 2643 2644 ep = get_ep_from_tid(dev, tid); 2645 if (!ep) 2646 return 0; 2647 2648 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2649 dst_confirm(ep->dst); 2650 2651 set_bit(PEER_CLOSE, &ep->com.history); 2652 mutex_lock(&ep->com.mutex); 2653 switch (ep->com.state) { 2654 case MPA_REQ_WAIT: 2655 __state_set(&ep->com, CLOSING); 2656 break; 2657 case MPA_REQ_SENT: 2658 __state_set(&ep->com, CLOSING); 2659 connect_reply_upcall(ep, -ECONNRESET); 2660 break; 2661 case MPA_REQ_RCVD: 2662 2663 /* 2664 * We're gonna mark this puppy DEAD, but keep 2665 * the reference on it until the ULP accepts or 2666 * rejects the CR. Also wake up anyone waiting 2667 * in rdma connection migration (see c4iw_accept_cr()). 2668 */ 2669 __state_set(&ep->com, CLOSING); 2670 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid); 2671 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 2672 break; 2673 case MPA_REP_SENT: 2674 __state_set(&ep->com, CLOSING); 2675 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid); 2676 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 2677 break; 2678 case FPDU_MODE: 2679 start_ep_timer(ep); 2680 __state_set(&ep->com, CLOSING); 2681 attrs.next_state = C4IW_QP_STATE_CLOSING; 2682 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2683 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2684 if (ret != -ECONNRESET) { 2685 peer_close_upcall(ep); 2686 disconnect = 1; 2687 } 2688 break; 2689 case ABORTING: 2690 disconnect = 0; 2691 break; 2692 case CLOSING: 2693 __state_set(&ep->com, MORIBUND); 2694 disconnect = 0; 2695 break; 2696 case MORIBUND: 2697 (void)stop_ep_timer(ep); 2698 if (ep->com.cm_id && ep->com.qp) { 2699 attrs.next_state = C4IW_QP_STATE_IDLE; 2700 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2701 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2702 } 2703 close_complete_upcall(ep, 0); 2704 __state_set(&ep->com, DEAD); 2705 release = 1; 2706 disconnect = 0; 2707 break; 2708 case DEAD: 2709 disconnect = 0; 2710 break; 2711 default: 2712 BUG_ON(1); 2713 } 2714 mutex_unlock(&ep->com.mutex); 2715 if (disconnect) 2716 c4iw_ep_disconnect(ep, 0, GFP_KERNEL); 2717 if (release) 2718 release_ep_resources(ep); 2719 c4iw_put_ep(&ep->com); 2720 return 0; 2721 } 2722 2723 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb) 2724 { 2725 struct cpl_abort_req_rss *req = cplhdr(skb); 2726 struct c4iw_ep *ep; 2727 struct sk_buff *rpl_skb; 2728 struct c4iw_qp_attributes attrs; 2729 int ret; 2730 int release = 0; 2731 unsigned int tid = GET_TID(req); 2732 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16); 2733 2734 ep = get_ep_from_tid(dev, tid); 2735 if (!ep) 2736 return 0; 2737 2738 if (cxgb_is_neg_adv(req->status)) { 2739 pr_debug("%s Negative advice on abort- tid %u status %d (%s)\n", 2740 __func__, ep->hwtid, req->status, 2741 neg_adv_str(req->status)); 2742 ep->stats.abort_neg_adv++; 2743 mutex_lock(&dev->rdev.stats.lock); 2744 dev->rdev.stats.neg_adv++; 2745 mutex_unlock(&dev->rdev.stats.lock); 2746 goto deref_ep; 2747 } 2748 pr_debug("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid, 2749 ep->com.state); 2750 set_bit(PEER_ABORT, &ep->com.history); 2751 2752 /* 2753 * Wake up any threads in rdma_init() or rdma_fini(). 2754 * However, this is not needed if com state is just 2755 * MPA_REQ_SENT 2756 */ 2757 if (ep->com.state != MPA_REQ_SENT) 2758 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 2759 2760 mutex_lock(&ep->com.mutex); 2761 switch (ep->com.state) { 2762 case CONNECTING: 2763 c4iw_put_ep(&ep->parent_ep->com); 2764 break; 2765 case MPA_REQ_WAIT: 2766 (void)stop_ep_timer(ep); 2767 break; 2768 case MPA_REQ_SENT: 2769 (void)stop_ep_timer(ep); 2770 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1)) 2771 connect_reply_upcall(ep, -ECONNRESET); 2772 else { 2773 /* 2774 * we just don't send notification upwards because we 2775 * want to retry with mpa_v1 without upper layers even 2776 * knowing it. 2777 * 2778 * do some housekeeping so as to re-initiate the 2779 * connection 2780 */ 2781 pr_debug("%s: mpa_rev=%d. Retrying with mpav1\n", 2782 __func__, mpa_rev); 2783 ep->retry_with_mpa_v1 = 1; 2784 } 2785 break; 2786 case MPA_REP_SENT: 2787 break; 2788 case MPA_REQ_RCVD: 2789 break; 2790 case MORIBUND: 2791 case CLOSING: 2792 stop_ep_timer(ep); 2793 /*FALLTHROUGH*/ 2794 case FPDU_MODE: 2795 if (ep->com.cm_id && ep->com.qp) { 2796 attrs.next_state = C4IW_QP_STATE_ERROR; 2797 ret = c4iw_modify_qp(ep->com.qp->rhp, 2798 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, 2799 &attrs, 1); 2800 if (ret) 2801 pr_err("%s - qp <- error failed!\n", __func__); 2802 } 2803 peer_abort_upcall(ep); 2804 break; 2805 case ABORTING: 2806 break; 2807 case DEAD: 2808 pr_debug("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__); 2809 mutex_unlock(&ep->com.mutex); 2810 goto deref_ep; 2811 default: 2812 BUG_ON(1); 2813 break; 2814 } 2815 dst_confirm(ep->dst); 2816 if (ep->com.state != ABORTING) { 2817 __state_set(&ep->com, DEAD); 2818 /* we don't release if we want to retry with mpa_v1 */ 2819 if (!ep->retry_with_mpa_v1) 2820 release = 1; 2821 } 2822 mutex_unlock(&ep->com.mutex); 2823 2824 rpl_skb = skb_dequeue(&ep->com.ep_skb_list); 2825 if (WARN_ON(!rpl_skb)) { 2826 release = 1; 2827 goto out; 2828 } 2829 2830 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx); 2831 2832 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb); 2833 out: 2834 if (release) 2835 release_ep_resources(ep); 2836 else if (ep->retry_with_mpa_v1) { 2837 if (ep->com.remote_addr.ss_family == AF_INET6) { 2838 struct sockaddr_in6 *sin6 = 2839 (struct sockaddr_in6 *) 2840 &ep->com.local_addr; 2841 cxgb4_clip_release( 2842 ep->com.dev->rdev.lldi.ports[0], 2843 (const u32 *)&sin6->sin6_addr.s6_addr, 2844 1); 2845 } 2846 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid); 2847 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid); 2848 dst_release(ep->dst); 2849 cxgb4_l2t_release(ep->l2t); 2850 c4iw_reconnect(ep); 2851 } 2852 2853 deref_ep: 2854 c4iw_put_ep(&ep->com); 2855 /* Dereferencing ep, referenced in peer_abort_intr() */ 2856 c4iw_put_ep(&ep->com); 2857 return 0; 2858 } 2859 2860 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2861 { 2862 struct c4iw_ep *ep; 2863 struct c4iw_qp_attributes attrs; 2864 struct cpl_close_con_rpl *rpl = cplhdr(skb); 2865 int release = 0; 2866 unsigned int tid = GET_TID(rpl); 2867 2868 ep = get_ep_from_tid(dev, tid); 2869 if (!ep) 2870 return 0; 2871 2872 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2873 BUG_ON(!ep); 2874 2875 /* The cm_id may be null if we failed to connect */ 2876 mutex_lock(&ep->com.mutex); 2877 set_bit(CLOSE_CON_RPL, &ep->com.history); 2878 switch (ep->com.state) { 2879 case CLOSING: 2880 __state_set(&ep->com, MORIBUND); 2881 break; 2882 case MORIBUND: 2883 (void)stop_ep_timer(ep); 2884 if ((ep->com.cm_id) && (ep->com.qp)) { 2885 attrs.next_state = C4IW_QP_STATE_IDLE; 2886 c4iw_modify_qp(ep->com.qp->rhp, 2887 ep->com.qp, 2888 C4IW_QP_ATTR_NEXT_STATE, 2889 &attrs, 1); 2890 } 2891 close_complete_upcall(ep, 0); 2892 __state_set(&ep->com, DEAD); 2893 release = 1; 2894 break; 2895 case ABORTING: 2896 case DEAD: 2897 break; 2898 default: 2899 BUG_ON(1); 2900 break; 2901 } 2902 mutex_unlock(&ep->com.mutex); 2903 if (release) 2904 release_ep_resources(ep); 2905 c4iw_put_ep(&ep->com); 2906 return 0; 2907 } 2908 2909 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb) 2910 { 2911 struct cpl_rdma_terminate *rpl = cplhdr(skb); 2912 unsigned int tid = GET_TID(rpl); 2913 struct c4iw_ep *ep; 2914 struct c4iw_qp_attributes attrs; 2915 2916 ep = get_ep_from_tid(dev, tid); 2917 BUG_ON(!ep); 2918 2919 if (ep && ep->com.qp) { 2920 pr_warn("TERM received tid %u qpid %u\n", 2921 tid, ep->com.qp->wq.sq.qid); 2922 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2923 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2924 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2925 } else 2926 pr_warn("TERM received tid %u no ep/qp\n", tid); 2927 c4iw_put_ep(&ep->com); 2928 2929 return 0; 2930 } 2931 2932 /* 2933 * Upcall from the adapter indicating data has been transmitted. 2934 * For us its just the single MPA request or reply. We can now free 2935 * the skb holding the mpa message. 2936 */ 2937 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb) 2938 { 2939 struct c4iw_ep *ep; 2940 struct cpl_fw4_ack *hdr = cplhdr(skb); 2941 u8 credits = hdr->credits; 2942 unsigned int tid = GET_TID(hdr); 2943 2944 2945 ep = get_ep_from_tid(dev, tid); 2946 if (!ep) 2947 return 0; 2948 pr_debug("%s ep %p tid %u credits %u\n", 2949 __func__, ep, ep->hwtid, credits); 2950 if (credits == 0) { 2951 pr_debug("%s 0 credit ack ep %p tid %u state %u\n", 2952 __func__, ep, ep->hwtid, state_read(&ep->com)); 2953 goto out; 2954 } 2955 2956 dst_confirm(ep->dst); 2957 if (ep->mpa_skb) { 2958 pr_debug("%s last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n", 2959 __func__, ep, ep->hwtid, 2960 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0); 2961 mutex_lock(&ep->com.mutex); 2962 kfree_skb(ep->mpa_skb); 2963 ep->mpa_skb = NULL; 2964 if (test_bit(STOP_MPA_TIMER, &ep->com.flags)) 2965 stop_ep_timer(ep); 2966 mutex_unlock(&ep->com.mutex); 2967 } 2968 out: 2969 c4iw_put_ep(&ep->com); 2970 return 0; 2971 } 2972 2973 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len) 2974 { 2975 int abort; 2976 struct c4iw_ep *ep = to_ep(cm_id); 2977 2978 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2979 2980 mutex_lock(&ep->com.mutex); 2981 if (ep->com.state != MPA_REQ_RCVD) { 2982 mutex_unlock(&ep->com.mutex); 2983 c4iw_put_ep(&ep->com); 2984 return -ECONNRESET; 2985 } 2986 set_bit(ULP_REJECT, &ep->com.history); 2987 if (mpa_rev == 0) 2988 abort = 1; 2989 else 2990 abort = send_mpa_reject(ep, pdata, pdata_len); 2991 mutex_unlock(&ep->com.mutex); 2992 2993 stop_ep_timer(ep); 2994 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL); 2995 c4iw_put_ep(&ep->com); 2996 return 0; 2997 } 2998 2999 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 3000 { 3001 int err; 3002 struct c4iw_qp_attributes attrs; 3003 enum c4iw_qp_attr_mask mask; 3004 struct c4iw_ep *ep = to_ep(cm_id); 3005 struct c4iw_dev *h = to_c4iw_dev(cm_id->device); 3006 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn); 3007 int abort = 0; 3008 3009 pr_debug("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 3010 3011 mutex_lock(&ep->com.mutex); 3012 if (ep->com.state != MPA_REQ_RCVD) { 3013 err = -ECONNRESET; 3014 goto err_out; 3015 } 3016 3017 BUG_ON(!qp); 3018 3019 set_bit(ULP_ACCEPT, &ep->com.history); 3020 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) || 3021 (conn_param->ird > cur_max_read_depth(ep->com.dev))) { 3022 err = -EINVAL; 3023 goto err_abort; 3024 } 3025 3026 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 3027 if (conn_param->ord > ep->ird) { 3028 if (RELAXED_IRD_NEGOTIATION) { 3029 conn_param->ord = ep->ird; 3030 } else { 3031 ep->ird = conn_param->ird; 3032 ep->ord = conn_param->ord; 3033 send_mpa_reject(ep, conn_param->private_data, 3034 conn_param->private_data_len); 3035 err = -ENOMEM; 3036 goto err_abort; 3037 } 3038 } 3039 if (conn_param->ird < ep->ord) { 3040 if (RELAXED_IRD_NEGOTIATION && 3041 ep->ord <= h->rdev.lldi.max_ordird_qp) { 3042 conn_param->ird = ep->ord; 3043 } else { 3044 err = -ENOMEM; 3045 goto err_abort; 3046 } 3047 } 3048 } 3049 ep->ird = conn_param->ird; 3050 ep->ord = conn_param->ord; 3051 3052 if (ep->mpa_attr.version == 1) { 3053 if (peer2peer && ep->ird == 0) 3054 ep->ird = 1; 3055 } else { 3056 if (peer2peer && 3057 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) && 3058 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0) 3059 ep->ird = 1; 3060 } 3061 3062 pr_debug("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord); 3063 3064 ep->com.cm_id = cm_id; 3065 ref_cm_id(&ep->com); 3066 ep->com.qp = qp; 3067 ref_qp(ep); 3068 3069 /* bind QP to EP and move to RTS */ 3070 attrs.mpa_attr = ep->mpa_attr; 3071 attrs.max_ird = ep->ird; 3072 attrs.max_ord = ep->ord; 3073 attrs.llp_stream_handle = ep; 3074 attrs.next_state = C4IW_QP_STATE_RTS; 3075 3076 /* bind QP and TID with INIT_WR */ 3077 mask = C4IW_QP_ATTR_NEXT_STATE | 3078 C4IW_QP_ATTR_LLP_STREAM_HANDLE | 3079 C4IW_QP_ATTR_MPA_ATTR | 3080 C4IW_QP_ATTR_MAX_IRD | 3081 C4IW_QP_ATTR_MAX_ORD; 3082 3083 err = c4iw_modify_qp(ep->com.qp->rhp, 3084 ep->com.qp, mask, &attrs, 1); 3085 if (err) 3086 goto err_deref_cm_id; 3087 3088 set_bit(STOP_MPA_TIMER, &ep->com.flags); 3089 err = send_mpa_reply(ep, conn_param->private_data, 3090 conn_param->private_data_len); 3091 if (err) 3092 goto err_deref_cm_id; 3093 3094 __state_set(&ep->com, FPDU_MODE); 3095 established_upcall(ep); 3096 mutex_unlock(&ep->com.mutex); 3097 c4iw_put_ep(&ep->com); 3098 return 0; 3099 err_deref_cm_id: 3100 deref_cm_id(&ep->com); 3101 err_abort: 3102 abort = 1; 3103 err_out: 3104 mutex_unlock(&ep->com.mutex); 3105 if (abort) 3106 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 3107 c4iw_put_ep(&ep->com); 3108 return err; 3109 } 3110 3111 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id) 3112 { 3113 struct in_device *ind; 3114 int found = 0; 3115 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr; 3116 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr; 3117 3118 ind = in_dev_get(dev->rdev.lldi.ports[0]); 3119 if (!ind) 3120 return -EADDRNOTAVAIL; 3121 for_primary_ifa(ind) { 3122 laddr->sin_addr.s_addr = ifa->ifa_address; 3123 raddr->sin_addr.s_addr = ifa->ifa_address; 3124 found = 1; 3125 break; 3126 } 3127 endfor_ifa(ind); 3128 in_dev_put(ind); 3129 return found ? 0 : -EADDRNOTAVAIL; 3130 } 3131 3132 static int get_lladdr(struct net_device *dev, struct in6_addr *addr, 3133 unsigned char banned_flags) 3134 { 3135 struct inet6_dev *idev; 3136 int err = -EADDRNOTAVAIL; 3137 3138 rcu_read_lock(); 3139 idev = __in6_dev_get(dev); 3140 if (idev != NULL) { 3141 struct inet6_ifaddr *ifp; 3142 3143 read_lock_bh(&idev->lock); 3144 list_for_each_entry(ifp, &idev->addr_list, if_list) { 3145 if (ifp->scope == IFA_LINK && 3146 !(ifp->flags & banned_flags)) { 3147 memcpy(addr, &ifp->addr, 16); 3148 err = 0; 3149 break; 3150 } 3151 } 3152 read_unlock_bh(&idev->lock); 3153 } 3154 rcu_read_unlock(); 3155 return err; 3156 } 3157 3158 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id) 3159 { 3160 struct in6_addr uninitialized_var(addr); 3161 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr; 3162 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr; 3163 3164 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) { 3165 memcpy(la6->sin6_addr.s6_addr, &addr, 16); 3166 memcpy(ra6->sin6_addr.s6_addr, &addr, 16); 3167 return 0; 3168 } 3169 return -EADDRNOTAVAIL; 3170 } 3171 3172 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 3173 { 3174 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 3175 struct c4iw_ep *ep; 3176 int err = 0; 3177 struct sockaddr_in *laddr; 3178 struct sockaddr_in *raddr; 3179 struct sockaddr_in6 *laddr6; 3180 struct sockaddr_in6 *raddr6; 3181 __u8 *ra; 3182 int iptype; 3183 3184 if ((conn_param->ord > cur_max_read_depth(dev)) || 3185 (conn_param->ird > cur_max_read_depth(dev))) { 3186 err = -EINVAL; 3187 goto out; 3188 } 3189 ep = alloc_ep(sizeof(*ep), GFP_KERNEL); 3190 if (!ep) { 3191 pr_err("%s - cannot alloc ep\n", __func__); 3192 err = -ENOMEM; 3193 goto out; 3194 } 3195 3196 skb_queue_head_init(&ep->com.ep_skb_list); 3197 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) { 3198 err = -ENOMEM; 3199 goto fail1; 3200 } 3201 3202 init_timer(&ep->timer); 3203 ep->plen = conn_param->private_data_len; 3204 if (ep->plen) 3205 memcpy(ep->mpa_pkt + sizeof(struct mpa_message), 3206 conn_param->private_data, ep->plen); 3207 ep->ird = conn_param->ird; 3208 ep->ord = conn_param->ord; 3209 3210 if (peer2peer && ep->ord == 0) 3211 ep->ord = 1; 3212 3213 ep->com.cm_id = cm_id; 3214 ref_cm_id(&ep->com); 3215 ep->com.dev = dev; 3216 ep->com.qp = get_qhp(dev, conn_param->qpn); 3217 if (!ep->com.qp) { 3218 pr_debug("%s qpn 0x%x not found!\n", __func__, conn_param->qpn); 3219 err = -EINVAL; 3220 goto fail2; 3221 } 3222 ref_qp(ep); 3223 pr_debug("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn, 3224 ep->com.qp, cm_id); 3225 3226 /* 3227 * Allocate an active TID to initiate a TCP connection. 3228 */ 3229 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep); 3230 if (ep->atid == -1) { 3231 pr_err("%s - cannot alloc atid\n", __func__); 3232 err = -ENOMEM; 3233 goto fail2; 3234 } 3235 insert_handle(dev, &dev->atid_idr, ep, ep->atid); 3236 3237 memcpy(&ep->com.local_addr, &cm_id->m_local_addr, 3238 sizeof(ep->com.local_addr)); 3239 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr, 3240 sizeof(ep->com.remote_addr)); 3241 3242 laddr = (struct sockaddr_in *)&ep->com.local_addr; 3243 raddr = (struct sockaddr_in *)&ep->com.remote_addr; 3244 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr; 3245 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr; 3246 3247 if (cm_id->m_remote_addr.ss_family == AF_INET) { 3248 iptype = 4; 3249 ra = (__u8 *)&raddr->sin_addr; 3250 3251 /* 3252 * Handle loopback requests to INADDR_ANY. 3253 */ 3254 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) { 3255 err = pick_local_ipaddrs(dev, cm_id); 3256 if (err) 3257 goto fail2; 3258 } 3259 3260 /* find a route */ 3261 pr_debug("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n", 3262 __func__, &laddr->sin_addr, ntohs(laddr->sin_port), 3263 ra, ntohs(raddr->sin_port)); 3264 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev, 3265 laddr->sin_addr.s_addr, 3266 raddr->sin_addr.s_addr, 3267 laddr->sin_port, 3268 raddr->sin_port, cm_id->tos); 3269 } else { 3270 iptype = 6; 3271 ra = (__u8 *)&raddr6->sin6_addr; 3272 3273 /* 3274 * Handle loopback requests to INADDR_ANY. 3275 */ 3276 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) { 3277 err = pick_local_ip6addrs(dev, cm_id); 3278 if (err) 3279 goto fail2; 3280 } 3281 3282 /* find a route */ 3283 pr_debug("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n", 3284 __func__, laddr6->sin6_addr.s6_addr, 3285 ntohs(laddr6->sin6_port), 3286 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port)); 3287 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev, 3288 laddr6->sin6_addr.s6_addr, 3289 raddr6->sin6_addr.s6_addr, 3290 laddr6->sin6_port, 3291 raddr6->sin6_port, 0, 3292 raddr6->sin6_scope_id); 3293 } 3294 if (!ep->dst) { 3295 pr_err("%s - cannot find route\n", __func__); 3296 err = -EHOSTUNREACH; 3297 goto fail3; 3298 } 3299 3300 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true, 3301 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos); 3302 if (err) { 3303 pr_err("%s - cannot alloc l2e\n", __func__); 3304 goto fail4; 3305 } 3306 3307 pr_debug("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n", 3308 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid, 3309 ep->l2t->idx); 3310 3311 state_set(&ep->com, CONNECTING); 3312 ep->tos = cm_id->tos; 3313 3314 /* send connect request to rnic */ 3315 err = send_connect(ep); 3316 if (!err) 3317 goto out; 3318 3319 cxgb4_l2t_release(ep->l2t); 3320 fail4: 3321 dst_release(ep->dst); 3322 fail3: 3323 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid); 3324 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid); 3325 fail2: 3326 skb_queue_purge(&ep->com.ep_skb_list); 3327 deref_cm_id(&ep->com); 3328 fail1: 3329 c4iw_put_ep(&ep->com); 3330 out: 3331 return err; 3332 } 3333 3334 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep) 3335 { 3336 int err; 3337 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) 3338 &ep->com.local_addr; 3339 3340 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) { 3341 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0], 3342 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3343 if (err) 3344 return err; 3345 } 3346 c4iw_init_wr_wait(&ep->com.wr_wait); 3347 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0], 3348 ep->stid, &sin6->sin6_addr, 3349 sin6->sin6_port, 3350 ep->com.dev->rdev.lldi.rxq_ids[0]); 3351 if (!err) 3352 err = c4iw_wait_for_reply(&ep->com.dev->rdev, 3353 &ep->com.wr_wait, 3354 0, 0, __func__); 3355 else if (err > 0) 3356 err = net_xmit_errno(err); 3357 if (err) { 3358 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 3359 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3360 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n", 3361 err, ep->stid, 3362 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port)); 3363 } 3364 return err; 3365 } 3366 3367 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep) 3368 { 3369 int err; 3370 struct sockaddr_in *sin = (struct sockaddr_in *) 3371 &ep->com.local_addr; 3372 3373 if (dev->rdev.lldi.enable_fw_ofld_conn) { 3374 do { 3375 err = cxgb4_create_server_filter( 3376 ep->com.dev->rdev.lldi.ports[0], ep->stid, 3377 sin->sin_addr.s_addr, sin->sin_port, 0, 3378 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0); 3379 if (err == -EBUSY) { 3380 if (c4iw_fatal_error(&ep->com.dev->rdev)) { 3381 err = -EIO; 3382 break; 3383 } 3384 set_current_state(TASK_UNINTERRUPTIBLE); 3385 schedule_timeout(usecs_to_jiffies(100)); 3386 } 3387 } while (err == -EBUSY); 3388 } else { 3389 c4iw_init_wr_wait(&ep->com.wr_wait); 3390 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0], 3391 ep->stid, sin->sin_addr.s_addr, sin->sin_port, 3392 0, ep->com.dev->rdev.lldi.rxq_ids[0]); 3393 if (!err) 3394 err = c4iw_wait_for_reply(&ep->com.dev->rdev, 3395 &ep->com.wr_wait, 3396 0, 0, __func__); 3397 else if (err > 0) 3398 err = net_xmit_errno(err); 3399 } 3400 if (err) 3401 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n" 3402 , err, ep->stid, 3403 &sin->sin_addr, ntohs(sin->sin_port)); 3404 return err; 3405 } 3406 3407 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog) 3408 { 3409 int err = 0; 3410 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 3411 struct c4iw_listen_ep *ep; 3412 3413 might_sleep(); 3414 3415 ep = alloc_ep(sizeof(*ep), GFP_KERNEL); 3416 if (!ep) { 3417 pr_err("%s - cannot alloc ep\n", __func__); 3418 err = -ENOMEM; 3419 goto fail1; 3420 } 3421 skb_queue_head_init(&ep->com.ep_skb_list); 3422 pr_debug("%s ep %p\n", __func__, ep); 3423 ep->com.cm_id = cm_id; 3424 ref_cm_id(&ep->com); 3425 ep->com.dev = dev; 3426 ep->backlog = backlog; 3427 memcpy(&ep->com.local_addr, &cm_id->m_local_addr, 3428 sizeof(ep->com.local_addr)); 3429 3430 /* 3431 * Allocate a server TID. 3432 */ 3433 if (dev->rdev.lldi.enable_fw_ofld_conn && 3434 ep->com.local_addr.ss_family == AF_INET) 3435 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids, 3436 cm_id->m_local_addr.ss_family, ep); 3437 else 3438 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids, 3439 cm_id->m_local_addr.ss_family, ep); 3440 3441 if (ep->stid == -1) { 3442 pr_err("%s - cannot alloc stid\n", __func__); 3443 err = -ENOMEM; 3444 goto fail2; 3445 } 3446 insert_handle(dev, &dev->stid_idr, ep, ep->stid); 3447 3448 memcpy(&ep->com.local_addr, &cm_id->m_local_addr, 3449 sizeof(ep->com.local_addr)); 3450 3451 state_set(&ep->com, LISTEN); 3452 if (ep->com.local_addr.ss_family == AF_INET) 3453 err = create_server4(dev, ep); 3454 else 3455 err = create_server6(dev, ep); 3456 if (!err) { 3457 cm_id->provider_data = ep; 3458 goto out; 3459 } 3460 3461 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, 3462 ep->com.local_addr.ss_family); 3463 fail2: 3464 deref_cm_id(&ep->com); 3465 c4iw_put_ep(&ep->com); 3466 fail1: 3467 out: 3468 return err; 3469 } 3470 3471 int c4iw_destroy_listen(struct iw_cm_id *cm_id) 3472 { 3473 int err; 3474 struct c4iw_listen_ep *ep = to_listen_ep(cm_id); 3475 3476 pr_debug("%s ep %p\n", __func__, ep); 3477 3478 might_sleep(); 3479 state_set(&ep->com, DEAD); 3480 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn && 3481 ep->com.local_addr.ss_family == AF_INET) { 3482 err = cxgb4_remove_server_filter( 3483 ep->com.dev->rdev.lldi.ports[0], ep->stid, 3484 ep->com.dev->rdev.lldi.rxq_ids[0], 0); 3485 } else { 3486 struct sockaddr_in6 *sin6; 3487 c4iw_init_wr_wait(&ep->com.wr_wait); 3488 err = cxgb4_remove_server( 3489 ep->com.dev->rdev.lldi.ports[0], ep->stid, 3490 ep->com.dev->rdev.lldi.rxq_ids[0], 0); 3491 if (err) 3492 goto done; 3493 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait, 3494 0, 0, __func__); 3495 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr; 3496 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 3497 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3498 } 3499 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid); 3500 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, 3501 ep->com.local_addr.ss_family); 3502 done: 3503 deref_cm_id(&ep->com); 3504 c4iw_put_ep(&ep->com); 3505 return err; 3506 } 3507 3508 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) 3509 { 3510 int ret = 0; 3511 int close = 0; 3512 int fatal = 0; 3513 struct c4iw_rdev *rdev; 3514 3515 mutex_lock(&ep->com.mutex); 3516 3517 pr_debug("%s ep %p state %s, abrupt %d\n", __func__, ep, 3518 states[ep->com.state], abrupt); 3519 3520 /* 3521 * Ref the ep here in case we have fatal errors causing the 3522 * ep to be released and freed. 3523 */ 3524 c4iw_get_ep(&ep->com); 3525 3526 rdev = &ep->com.dev->rdev; 3527 if (c4iw_fatal_error(rdev)) { 3528 fatal = 1; 3529 close_complete_upcall(ep, -EIO); 3530 ep->com.state = DEAD; 3531 } 3532 switch (ep->com.state) { 3533 case MPA_REQ_WAIT: 3534 case MPA_REQ_SENT: 3535 case MPA_REQ_RCVD: 3536 case MPA_REP_SENT: 3537 case FPDU_MODE: 3538 case CONNECTING: 3539 close = 1; 3540 if (abrupt) 3541 ep->com.state = ABORTING; 3542 else { 3543 ep->com.state = CLOSING; 3544 3545 /* 3546 * if we close before we see the fw4_ack() then we fix 3547 * up the timer state since we're reusing it. 3548 */ 3549 if (ep->mpa_skb && 3550 test_bit(STOP_MPA_TIMER, &ep->com.flags)) { 3551 clear_bit(STOP_MPA_TIMER, &ep->com.flags); 3552 stop_ep_timer(ep); 3553 } 3554 start_ep_timer(ep); 3555 } 3556 set_bit(CLOSE_SENT, &ep->com.flags); 3557 break; 3558 case CLOSING: 3559 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) { 3560 close = 1; 3561 if (abrupt) { 3562 (void)stop_ep_timer(ep); 3563 ep->com.state = ABORTING; 3564 } else 3565 ep->com.state = MORIBUND; 3566 } 3567 break; 3568 case MORIBUND: 3569 case ABORTING: 3570 case DEAD: 3571 pr_debug("%s ignoring disconnect ep %p state %u\n", 3572 __func__, ep, ep->com.state); 3573 break; 3574 default: 3575 BUG(); 3576 break; 3577 } 3578 3579 if (close) { 3580 if (abrupt) { 3581 set_bit(EP_DISC_ABORT, &ep->com.history); 3582 close_complete_upcall(ep, -ECONNRESET); 3583 ret = send_abort(ep); 3584 } else { 3585 set_bit(EP_DISC_CLOSE, &ep->com.history); 3586 ret = send_halfclose(ep); 3587 } 3588 if (ret) { 3589 set_bit(EP_DISC_FAIL, &ep->com.history); 3590 if (!abrupt) { 3591 stop_ep_timer(ep); 3592 close_complete_upcall(ep, -EIO); 3593 } 3594 if (ep->com.qp) { 3595 struct c4iw_qp_attributes attrs; 3596 3597 attrs.next_state = C4IW_QP_STATE_ERROR; 3598 ret = c4iw_modify_qp(ep->com.qp->rhp, 3599 ep->com.qp, 3600 C4IW_QP_ATTR_NEXT_STATE, 3601 &attrs, 1); 3602 if (ret) 3603 pr_err("%s - qp <- error failed!\n", 3604 __func__); 3605 } 3606 fatal = 1; 3607 } 3608 } 3609 mutex_unlock(&ep->com.mutex); 3610 c4iw_put_ep(&ep->com); 3611 if (fatal) 3612 release_ep_resources(ep); 3613 return ret; 3614 } 3615 3616 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb, 3617 struct cpl_fw6_msg_ofld_connection_wr_rpl *req) 3618 { 3619 struct c4iw_ep *ep; 3620 int atid = be32_to_cpu(req->tid); 3621 3622 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids, 3623 (__force u32) req->tid); 3624 if (!ep) 3625 return; 3626 3627 switch (req->retval) { 3628 case FW_ENOMEM: 3629 set_bit(ACT_RETRY_NOMEM, &ep->com.history); 3630 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) { 3631 send_fw_act_open_req(ep, atid); 3632 return; 3633 } 3634 case FW_EADDRINUSE: 3635 set_bit(ACT_RETRY_INUSE, &ep->com.history); 3636 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) { 3637 send_fw_act_open_req(ep, atid); 3638 return; 3639 } 3640 break; 3641 default: 3642 pr_info("%s unexpected ofld conn wr retval %d\n", 3643 __func__, req->retval); 3644 break; 3645 } 3646 pr_err("active ofld_connect_wr failure %d atid %d\n", 3647 req->retval, atid); 3648 mutex_lock(&dev->rdev.stats.lock); 3649 dev->rdev.stats.act_ofld_conn_fails++; 3650 mutex_unlock(&dev->rdev.stats.lock); 3651 connect_reply_upcall(ep, status2errno(req->retval)); 3652 state_set(&ep->com, DEAD); 3653 if (ep->com.remote_addr.ss_family == AF_INET6) { 3654 struct sockaddr_in6 *sin6 = 3655 (struct sockaddr_in6 *)&ep->com.local_addr; 3656 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 3657 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3658 } 3659 remove_handle(dev, &dev->atid_idr, atid); 3660 cxgb4_free_atid(dev->rdev.lldi.tids, atid); 3661 dst_release(ep->dst); 3662 cxgb4_l2t_release(ep->l2t); 3663 c4iw_put_ep(&ep->com); 3664 } 3665 3666 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb, 3667 struct cpl_fw6_msg_ofld_connection_wr_rpl *req) 3668 { 3669 struct sk_buff *rpl_skb; 3670 struct cpl_pass_accept_req *cpl; 3671 int ret; 3672 3673 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie; 3674 BUG_ON(!rpl_skb); 3675 if (req->retval) { 3676 pr_debug("%s passive open failure %d\n", __func__, req->retval); 3677 mutex_lock(&dev->rdev.stats.lock); 3678 dev->rdev.stats.pas_ofld_conn_fails++; 3679 mutex_unlock(&dev->rdev.stats.lock); 3680 kfree_skb(rpl_skb); 3681 } else { 3682 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb); 3683 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 3684 (__force u32) htonl( 3685 (__force u32) req->tid))); 3686 ret = pass_accept_req(dev, rpl_skb); 3687 if (!ret) 3688 kfree_skb(rpl_skb); 3689 } 3690 return; 3691 } 3692 3693 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb) 3694 { 3695 struct cpl_fw6_msg *rpl = cplhdr(skb); 3696 struct cpl_fw6_msg_ofld_connection_wr_rpl *req; 3697 3698 switch (rpl->type) { 3699 case FW6_TYPE_CQE: 3700 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]); 3701 break; 3702 case FW6_TYPE_OFLD_CONNECTION_WR_RPL: 3703 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data; 3704 switch (req->t_state) { 3705 case TCP_SYN_SENT: 3706 active_ofld_conn_reply(dev, skb, req); 3707 break; 3708 case TCP_SYN_RECV: 3709 passive_ofld_conn_reply(dev, skb, req); 3710 break; 3711 default: 3712 pr_err("%s unexpected ofld conn wr state %d\n", 3713 __func__, req->t_state); 3714 break; 3715 } 3716 break; 3717 } 3718 return 0; 3719 } 3720 3721 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos) 3722 { 3723 __be32 l2info; 3724 __be16 hdr_len, vlantag, len; 3725 u16 eth_hdr_len; 3726 int tcp_hdr_len, ip_hdr_len; 3727 u8 intf; 3728 struct cpl_rx_pkt *cpl = cplhdr(skb); 3729 struct cpl_pass_accept_req *req; 3730 struct tcp_options_received tmp_opt; 3731 struct c4iw_dev *dev; 3732 enum chip_type type; 3733 3734 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *))); 3735 /* Store values from cpl_rx_pkt in temporary location. */ 3736 vlantag = cpl->vlan; 3737 len = cpl->len; 3738 l2info = cpl->l2info; 3739 hdr_len = cpl->hdr_len; 3740 intf = cpl->iff; 3741 3742 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header)); 3743 3744 /* 3745 * We need to parse the TCP options from SYN packet. 3746 * to generate cpl_pass_accept_req. 3747 */ 3748 memset(&tmp_opt, 0, sizeof(tmp_opt)); 3749 tcp_clear_options(&tmp_opt); 3750 tcp_parse_options(skb, &tmp_opt, 0, NULL); 3751 3752 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req)); 3753 memset(req, 0, sizeof(*req)); 3754 req->l2info = cpu_to_be16(SYN_INTF_V(intf) | 3755 SYN_MAC_IDX_V(RX_MACIDX_G( 3756 be32_to_cpu(l2info))) | 3757 SYN_XACT_MATCH_F); 3758 type = dev->rdev.lldi.adapter_type; 3759 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len)); 3760 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len)); 3761 req->hdr_len = 3762 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info)))); 3763 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) { 3764 eth_hdr_len = is_t4(type) ? 3765 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) : 3766 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info)); 3767 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) | 3768 IP_HDR_LEN_V(ip_hdr_len) | 3769 ETH_HDR_LEN_V(eth_hdr_len)); 3770 } else { /* T6 and later */ 3771 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info)); 3772 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) | 3773 T6_IP_HDR_LEN_V(ip_hdr_len) | 3774 T6_ETH_HDR_LEN_V(eth_hdr_len)); 3775 } 3776 req->vlan = vlantag; 3777 req->len = len; 3778 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) | 3779 PASS_OPEN_TOS_V(tos)); 3780 req->tcpopt.mss = htons(tmp_opt.mss_clamp); 3781 if (tmp_opt.wscale_ok) 3782 req->tcpopt.wsf = tmp_opt.snd_wscale; 3783 req->tcpopt.tstamp = tmp_opt.saw_tstamp; 3784 if (tmp_opt.sack_ok) 3785 req->tcpopt.sack = 1; 3786 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0)); 3787 return; 3788 } 3789 3790 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb, 3791 __be32 laddr, __be16 lport, 3792 __be32 raddr, __be16 rport, 3793 u32 rcv_isn, u32 filter, u16 window, 3794 u32 rss_qid, u8 port_id) 3795 { 3796 struct sk_buff *req_skb; 3797 struct fw_ofld_connection_wr *req; 3798 struct cpl_pass_accept_req *cpl = cplhdr(skb); 3799 int ret; 3800 3801 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL); 3802 if (!req_skb) 3803 return; 3804 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req)); 3805 memset(req, 0, sizeof(*req)); 3806 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F); 3807 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16))); 3808 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F); 3809 req->le.filter = (__force __be32) filter; 3810 req->le.lport = lport; 3811 req->le.pport = rport; 3812 req->le.u.ipv4.lip = laddr; 3813 req->le.u.ipv4.pip = raddr; 3814 req->tcb.rcv_nxt = htonl(rcv_isn + 1); 3815 req->tcb.rcv_adv = htons(window); 3816 req->tcb.t_state_to_astid = 3817 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) | 3818 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) | 3819 FW_OFLD_CONNECTION_WR_ASTID_V( 3820 PASS_OPEN_TID_G(ntohl(cpl->tos_stid)))); 3821 3822 /* 3823 * We store the qid in opt2 which will be used by the firmware 3824 * to send us the wr response. 3825 */ 3826 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid)); 3827 3828 /* 3829 * We initialize the MSS index in TCB to 0xF. 3830 * So that when driver sends cpl_pass_accept_rpl 3831 * TCB picks up the correct value. If this was 0 3832 * TP will ignore any value > 0 for MSS index. 3833 */ 3834 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF)); 3835 req->cookie = (uintptr_t)skb; 3836 3837 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id); 3838 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb); 3839 if (ret < 0) { 3840 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__, 3841 ret); 3842 kfree_skb(skb); 3843 kfree_skb(req_skb); 3844 } 3845 } 3846 3847 /* 3848 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt 3849 * messages when a filter is being used instead of server to 3850 * redirect a syn packet. When packets hit filter they are redirected 3851 * to the offload queue and driver tries to establish the connection 3852 * using firmware work request. 3853 */ 3854 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb) 3855 { 3856 int stid; 3857 unsigned int filter; 3858 struct ethhdr *eh = NULL; 3859 struct vlan_ethhdr *vlan_eh = NULL; 3860 struct iphdr *iph; 3861 struct tcphdr *tcph; 3862 struct rss_header *rss = (void *)skb->data; 3863 struct cpl_rx_pkt *cpl = (void *)skb->data; 3864 struct cpl_pass_accept_req *req = (void *)(rss + 1); 3865 struct l2t_entry *e; 3866 struct dst_entry *dst; 3867 struct c4iw_ep *lep = NULL; 3868 u16 window; 3869 struct port_info *pi; 3870 struct net_device *pdev; 3871 u16 rss_qid, eth_hdr_len; 3872 int step; 3873 u32 tx_chan; 3874 struct neighbour *neigh; 3875 3876 /* Drop all non-SYN packets */ 3877 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F))) 3878 goto reject; 3879 3880 /* 3881 * Drop all packets which did not hit the filter. 3882 * Unlikely to happen. 3883 */ 3884 if (!(rss->filter_hit && rss->filter_tid)) 3885 goto reject; 3886 3887 /* 3888 * Calculate the server tid from filter hit index from cpl_rx_pkt. 3889 */ 3890 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val); 3891 3892 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid); 3893 if (!lep) { 3894 pr_debug("%s connect request on invalid stid %d\n", 3895 __func__, stid); 3896 goto reject; 3897 } 3898 3899 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) { 3900 case CHELSIO_T4: 3901 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info)); 3902 break; 3903 case CHELSIO_T5: 3904 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info)); 3905 break; 3906 case CHELSIO_T6: 3907 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info)); 3908 break; 3909 default: 3910 pr_err("T%d Chip is not supported\n", 3911 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)); 3912 goto reject; 3913 } 3914 3915 if (eth_hdr_len == ETH_HLEN) { 3916 eh = (struct ethhdr *)(req + 1); 3917 iph = (struct iphdr *)(eh + 1); 3918 } else { 3919 vlan_eh = (struct vlan_ethhdr *)(req + 1); 3920 iph = (struct iphdr *)(vlan_eh + 1); 3921 skb->vlan_tci = ntohs(cpl->vlan); 3922 } 3923 3924 if (iph->version != 0x4) 3925 goto reject; 3926 3927 tcph = (struct tcphdr *)(iph + 1); 3928 skb_set_network_header(skb, (void *)iph - (void *)rss); 3929 skb_set_transport_header(skb, (void *)tcph - (void *)rss); 3930 skb_get(skb); 3931 3932 pr_debug("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__, 3933 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr), 3934 ntohs(tcph->source), iph->tos); 3935 3936 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev, 3937 iph->daddr, iph->saddr, tcph->dest, 3938 tcph->source, iph->tos); 3939 if (!dst) { 3940 pr_err("%s - failed to find dst entry!\n", 3941 __func__); 3942 goto reject; 3943 } 3944 neigh = dst_neigh_lookup_skb(dst, skb); 3945 3946 if (!neigh) { 3947 pr_err("%s - failed to allocate neigh!\n", 3948 __func__); 3949 goto free_dst; 3950 } 3951 3952 if (neigh->dev->flags & IFF_LOOPBACK) { 3953 pdev = ip_dev_find(&init_net, iph->daddr); 3954 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh, 3955 pdev, 0); 3956 pi = (struct port_info *)netdev_priv(pdev); 3957 tx_chan = cxgb4_port_chan(pdev); 3958 dev_put(pdev); 3959 } else { 3960 pdev = get_real_dev(neigh->dev); 3961 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh, 3962 pdev, 0); 3963 pi = (struct port_info *)netdev_priv(pdev); 3964 tx_chan = cxgb4_port_chan(pdev); 3965 } 3966 neigh_release(neigh); 3967 if (!e) { 3968 pr_err("%s - failed to allocate l2t entry!\n", 3969 __func__); 3970 goto free_dst; 3971 } 3972 3973 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan; 3974 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step]; 3975 window = (__force u16) htons((__force u16)tcph->window); 3976 3977 /* Calcuate filter portion for LE region. */ 3978 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple( 3979 dev->rdev.lldi.ports[0], 3980 e)); 3981 3982 /* 3983 * Synthesize the cpl_pass_accept_req. We have everything except the 3984 * TID. Once firmware sends a reply with TID we update the TID field 3985 * in cpl and pass it through the regular cpl_pass_accept_req path. 3986 */ 3987 build_cpl_pass_accept_req(skb, stid, iph->tos); 3988 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr, 3989 tcph->source, ntohl(tcph->seq), filter, window, 3990 rss_qid, pi->port_id); 3991 cxgb4_l2t_release(e); 3992 free_dst: 3993 dst_release(dst); 3994 reject: 3995 if (lep) 3996 c4iw_put_ep(&lep->com); 3997 return 0; 3998 } 3999 4000 /* 4001 * These are the real handlers that are called from a 4002 * work queue. 4003 */ 4004 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = { 4005 [CPL_ACT_ESTABLISH] = act_establish, 4006 [CPL_ACT_OPEN_RPL] = act_open_rpl, 4007 [CPL_RX_DATA] = rx_data, 4008 [CPL_ABORT_RPL_RSS] = abort_rpl, 4009 [CPL_ABORT_RPL] = abort_rpl, 4010 [CPL_PASS_OPEN_RPL] = pass_open_rpl, 4011 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl, 4012 [CPL_PASS_ACCEPT_REQ] = pass_accept_req, 4013 [CPL_PASS_ESTABLISH] = pass_establish, 4014 [CPL_PEER_CLOSE] = peer_close, 4015 [CPL_ABORT_REQ_RSS] = peer_abort, 4016 [CPL_CLOSE_CON_RPL] = close_con_rpl, 4017 [CPL_RDMA_TERMINATE] = terminate, 4018 [CPL_FW4_ACK] = fw4_ack, 4019 [CPL_FW6_MSG] = deferred_fw6_msg, 4020 [CPL_RX_PKT] = rx_pkt, 4021 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe, 4022 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe 4023 }; 4024 4025 static void process_timeout(struct c4iw_ep *ep) 4026 { 4027 struct c4iw_qp_attributes attrs; 4028 int abort = 1; 4029 4030 mutex_lock(&ep->com.mutex); 4031 pr_debug("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid, 4032 ep->com.state); 4033 set_bit(TIMEDOUT, &ep->com.history); 4034 switch (ep->com.state) { 4035 case MPA_REQ_SENT: 4036 connect_reply_upcall(ep, -ETIMEDOUT); 4037 break; 4038 case MPA_REQ_WAIT: 4039 case MPA_REQ_RCVD: 4040 case MPA_REP_SENT: 4041 case FPDU_MODE: 4042 break; 4043 case CLOSING: 4044 case MORIBUND: 4045 if (ep->com.cm_id && ep->com.qp) { 4046 attrs.next_state = C4IW_QP_STATE_ERROR; 4047 c4iw_modify_qp(ep->com.qp->rhp, 4048 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, 4049 &attrs, 1); 4050 } 4051 close_complete_upcall(ep, -ETIMEDOUT); 4052 break; 4053 case ABORTING: 4054 case DEAD: 4055 4056 /* 4057 * These states are expected if the ep timed out at the same 4058 * time as another thread was calling stop_ep_timer(). 4059 * So we silently do nothing for these states. 4060 */ 4061 abort = 0; 4062 break; 4063 default: 4064 WARN(1, "%s unexpected state ep %p tid %u state %u\n", 4065 __func__, ep, ep->hwtid, ep->com.state); 4066 abort = 0; 4067 } 4068 mutex_unlock(&ep->com.mutex); 4069 if (abort) 4070 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 4071 c4iw_put_ep(&ep->com); 4072 } 4073 4074 static void process_timedout_eps(void) 4075 { 4076 struct c4iw_ep *ep; 4077 4078 spin_lock_irq(&timeout_lock); 4079 while (!list_empty(&timeout_list)) { 4080 struct list_head *tmp; 4081 4082 tmp = timeout_list.next; 4083 list_del(tmp); 4084 tmp->next = NULL; 4085 tmp->prev = NULL; 4086 spin_unlock_irq(&timeout_lock); 4087 ep = list_entry(tmp, struct c4iw_ep, entry); 4088 process_timeout(ep); 4089 spin_lock_irq(&timeout_lock); 4090 } 4091 spin_unlock_irq(&timeout_lock); 4092 } 4093 4094 static void process_work(struct work_struct *work) 4095 { 4096 struct sk_buff *skb = NULL; 4097 struct c4iw_dev *dev; 4098 struct cpl_act_establish *rpl; 4099 unsigned int opcode; 4100 int ret; 4101 4102 process_timedout_eps(); 4103 while ((skb = skb_dequeue(&rxq))) { 4104 rpl = cplhdr(skb); 4105 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *))); 4106 opcode = rpl->ot.opcode; 4107 4108 BUG_ON(!work_handlers[opcode]); 4109 ret = work_handlers[opcode](dev, skb); 4110 if (!ret) 4111 kfree_skb(skb); 4112 process_timedout_eps(); 4113 } 4114 } 4115 4116 static DECLARE_WORK(skb_work, process_work); 4117 4118 static void ep_timeout(unsigned long arg) 4119 { 4120 struct c4iw_ep *ep = (struct c4iw_ep *)arg; 4121 int kickit = 0; 4122 4123 spin_lock(&timeout_lock); 4124 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 4125 /* 4126 * Only insert if it is not already on the list. 4127 */ 4128 if (!ep->entry.next) { 4129 list_add_tail(&ep->entry, &timeout_list); 4130 kickit = 1; 4131 } 4132 } 4133 spin_unlock(&timeout_lock); 4134 if (kickit) 4135 queue_work(workq, &skb_work); 4136 } 4137 4138 /* 4139 * All the CM events are handled on a work queue to have a safe context. 4140 */ 4141 static int sched(struct c4iw_dev *dev, struct sk_buff *skb) 4142 { 4143 4144 /* 4145 * Save dev in the skb->cb area. 4146 */ 4147 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev; 4148 4149 /* 4150 * Queue the skb and schedule the worker thread. 4151 */ 4152 skb_queue_tail(&rxq, skb); 4153 queue_work(workq, &skb_work); 4154 return 0; 4155 } 4156 4157 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 4158 { 4159 struct cpl_set_tcb_rpl *rpl = cplhdr(skb); 4160 4161 if (rpl->status != CPL_ERR_NONE) { 4162 pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n", 4163 rpl->status, GET_TID(rpl)); 4164 } 4165 kfree_skb(skb); 4166 return 0; 4167 } 4168 4169 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb) 4170 { 4171 struct cpl_fw6_msg *rpl = cplhdr(skb); 4172 struct c4iw_wr_wait *wr_waitp; 4173 int ret; 4174 4175 pr_debug("%s type %u\n", __func__, rpl->type); 4176 4177 switch (rpl->type) { 4178 case FW6_TYPE_WR_RPL: 4179 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff); 4180 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1]; 4181 pr_debug("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret); 4182 if (wr_waitp) 4183 c4iw_wake_up(wr_waitp, ret ? -ret : 0); 4184 kfree_skb(skb); 4185 break; 4186 case FW6_TYPE_CQE: 4187 case FW6_TYPE_OFLD_CONNECTION_WR_RPL: 4188 sched(dev, skb); 4189 break; 4190 default: 4191 pr_err("%s unexpected fw6 msg type %u\n", 4192 __func__, rpl->type); 4193 kfree_skb(skb); 4194 break; 4195 } 4196 return 0; 4197 } 4198 4199 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb) 4200 { 4201 struct cpl_abort_req_rss *req = cplhdr(skb); 4202 struct c4iw_ep *ep; 4203 unsigned int tid = GET_TID(req); 4204 4205 ep = get_ep_from_tid(dev, tid); 4206 /* This EP will be dereferenced in peer_abort() */ 4207 if (!ep) { 4208 pr_warn("Abort on non-existent endpoint, tid %d\n", tid); 4209 kfree_skb(skb); 4210 return 0; 4211 } 4212 if (cxgb_is_neg_adv(req->status)) { 4213 pr_debug("%s Negative advice on abort- tid %u status %d (%s)\n", 4214 __func__, ep->hwtid, req->status, 4215 neg_adv_str(req->status)); 4216 goto out; 4217 } 4218 pr_debug("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid, 4219 ep->com.state); 4220 4221 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 4222 out: 4223 sched(dev, skb); 4224 return 0; 4225 } 4226 4227 /* 4228 * Most upcalls from the T4 Core go to sched() to 4229 * schedule the processing on a work queue. 4230 */ 4231 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = { 4232 [CPL_ACT_ESTABLISH] = sched, 4233 [CPL_ACT_OPEN_RPL] = sched, 4234 [CPL_RX_DATA] = sched, 4235 [CPL_ABORT_RPL_RSS] = sched, 4236 [CPL_ABORT_RPL] = sched, 4237 [CPL_PASS_OPEN_RPL] = sched, 4238 [CPL_CLOSE_LISTSRV_RPL] = sched, 4239 [CPL_PASS_ACCEPT_REQ] = sched, 4240 [CPL_PASS_ESTABLISH] = sched, 4241 [CPL_PEER_CLOSE] = sched, 4242 [CPL_CLOSE_CON_RPL] = sched, 4243 [CPL_ABORT_REQ_RSS] = peer_abort_intr, 4244 [CPL_RDMA_TERMINATE] = sched, 4245 [CPL_FW4_ACK] = sched, 4246 [CPL_SET_TCB_RPL] = set_tcb_rpl, 4247 [CPL_FW6_MSG] = fw6_msg, 4248 [CPL_RX_PKT] = sched 4249 }; 4250 4251 int __init c4iw_cm_init(void) 4252 { 4253 spin_lock_init(&timeout_lock); 4254 skb_queue_head_init(&rxq); 4255 4256 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM); 4257 if (!workq) 4258 return -ENOMEM; 4259 4260 return 0; 4261 } 4262 4263 void c4iw_cm_term(void) 4264 { 4265 WARN_ON(!list_empty(&timeout_list)); 4266 flush_workqueue(workq); 4267 destroy_workqueue(workq); 4268 } 4269