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 static int peer2peer = 1; 103 module_param(peer2peer, int, 0644); 104 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)"); 105 106 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ; 107 module_param(p2p_type, int, 0644); 108 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: " 109 "1=RDMA_READ 0=RDMA_WRITE (default 1)"); 110 111 static int ep_timeout_secs = 60; 112 module_param(ep_timeout_secs, int, 0644); 113 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout " 114 "in seconds (default=60)"); 115 116 static int mpa_rev = 2; 117 module_param(mpa_rev, int, 0644); 118 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, " 119 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft" 120 " compliant (default=2)"); 121 122 static int markers_enabled; 123 module_param(markers_enabled, int, 0644); 124 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)"); 125 126 static int crc_enabled = 1; 127 module_param(crc_enabled, int, 0644); 128 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)"); 129 130 static int rcv_win = 256 * 1024; 131 module_param(rcv_win, int, 0644); 132 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)"); 133 134 static int snd_win = 128 * 1024; 135 module_param(snd_win, int, 0644); 136 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)"); 137 138 static struct workqueue_struct *workq; 139 140 static struct sk_buff_head rxq; 141 142 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp); 143 static void ep_timeout(struct timer_list *t); 144 static void connect_reply_upcall(struct c4iw_ep *ep, int status); 145 static int sched(struct c4iw_dev *dev, struct sk_buff *skb); 146 147 static LIST_HEAD(timeout_list); 148 static spinlock_t timeout_lock; 149 150 static void deref_cm_id(struct c4iw_ep_common *epc) 151 { 152 epc->cm_id->rem_ref(epc->cm_id); 153 epc->cm_id = NULL; 154 set_bit(CM_ID_DEREFED, &epc->history); 155 } 156 157 static void ref_cm_id(struct c4iw_ep_common *epc) 158 { 159 set_bit(CM_ID_REFED, &epc->history); 160 epc->cm_id->add_ref(epc->cm_id); 161 } 162 163 static void deref_qp(struct c4iw_ep *ep) 164 { 165 c4iw_qp_rem_ref(&ep->com.qp->ibqp); 166 clear_bit(QP_REFERENCED, &ep->com.flags); 167 set_bit(QP_DEREFED, &ep->com.history); 168 } 169 170 static void ref_qp(struct c4iw_ep *ep) 171 { 172 set_bit(QP_REFERENCED, &ep->com.flags); 173 set_bit(QP_REFED, &ep->com.history); 174 c4iw_qp_add_ref(&ep->com.qp->ibqp); 175 } 176 177 static void start_ep_timer(struct c4iw_ep *ep) 178 { 179 pr_debug("ep %p\n", ep); 180 if (timer_pending(&ep->timer)) { 181 pr_err("%s timer already started! ep %p\n", 182 __func__, ep); 183 return; 184 } 185 clear_bit(TIMEOUT, &ep->com.flags); 186 c4iw_get_ep(&ep->com); 187 ep->timer.expires = jiffies + ep_timeout_secs * HZ; 188 add_timer(&ep->timer); 189 } 190 191 static int stop_ep_timer(struct c4iw_ep *ep) 192 { 193 pr_debug("ep %p stopping\n", ep); 194 del_timer_sync(&ep->timer); 195 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 196 c4iw_put_ep(&ep->com); 197 return 0; 198 } 199 return 1; 200 } 201 202 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb, 203 struct l2t_entry *l2e) 204 { 205 int error = 0; 206 207 if (c4iw_fatal_error(rdev)) { 208 kfree_skb(skb); 209 pr_err("%s - device in error state - dropping\n", __func__); 210 return -EIO; 211 } 212 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e); 213 if (error < 0) 214 kfree_skb(skb); 215 else if (error == NET_XMIT_DROP) 216 return -ENOMEM; 217 return error < 0 ? error : 0; 218 } 219 220 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb) 221 { 222 int error = 0; 223 224 if (c4iw_fatal_error(rdev)) { 225 kfree_skb(skb); 226 pr_err("%s - device in error state - dropping\n", __func__); 227 return -EIO; 228 } 229 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb); 230 if (error < 0) 231 kfree_skb(skb); 232 return error < 0 ? error : 0; 233 } 234 235 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb) 236 { 237 u32 len = roundup(sizeof(struct cpl_tid_release), 16); 238 239 skb = get_skb(skb, len, GFP_KERNEL); 240 if (!skb) 241 return; 242 243 cxgb_mk_tid_release(skb, len, hwtid, 0); 244 c4iw_ofld_send(rdev, skb); 245 return; 246 } 247 248 static void set_emss(struct c4iw_ep *ep, u16 opt) 249 { 250 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] - 251 ((AF_INET == ep->com.remote_addr.ss_family) ? 252 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) - 253 sizeof(struct tcphdr); 254 ep->mss = ep->emss; 255 if (TCPOPT_TSTAMP_G(opt)) 256 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4); 257 if (ep->emss < 128) 258 ep->emss = 128; 259 if (ep->emss & 7) 260 pr_debug("Warning: misaligned mtu idx %u mss %u emss=%u\n", 261 TCPOPT_MSS_G(opt), ep->mss, ep->emss); 262 pr_debug("mss_idx %u mss %u emss=%u\n", TCPOPT_MSS_G(opt), ep->mss, 263 ep->emss); 264 } 265 266 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc) 267 { 268 enum c4iw_ep_state state; 269 270 mutex_lock(&epc->mutex); 271 state = epc->state; 272 mutex_unlock(&epc->mutex); 273 return state; 274 } 275 276 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new) 277 { 278 epc->state = new; 279 } 280 281 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new) 282 { 283 mutex_lock(&epc->mutex); 284 pr_debug("%s -> %s\n", states[epc->state], states[new]); 285 __state_set(epc, new); 286 mutex_unlock(&epc->mutex); 287 return; 288 } 289 290 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size) 291 { 292 struct sk_buff *skb; 293 unsigned int i; 294 size_t len; 295 296 len = roundup(sizeof(union cpl_wr_size), 16); 297 for (i = 0; i < size; i++) { 298 skb = alloc_skb(len, GFP_KERNEL); 299 if (!skb) 300 goto fail; 301 skb_queue_tail(ep_skb_list, skb); 302 } 303 return 0; 304 fail: 305 skb_queue_purge(ep_skb_list); 306 return -ENOMEM; 307 } 308 309 static void *alloc_ep(int size, gfp_t gfp) 310 { 311 struct c4iw_ep_common *epc; 312 313 epc = kzalloc(size, gfp); 314 if (epc) { 315 epc->wr_waitp = c4iw_alloc_wr_wait(gfp); 316 if (!epc->wr_waitp) { 317 kfree(epc); 318 epc = NULL; 319 goto out; 320 } 321 kref_init(&epc->kref); 322 mutex_init(&epc->mutex); 323 c4iw_init_wr_wait(epc->wr_waitp); 324 } 325 pr_debug("alloc ep %p\n", epc); 326 out: 327 return epc; 328 } 329 330 static void remove_ep_tid(struct c4iw_ep *ep) 331 { 332 unsigned long flags; 333 334 spin_lock_irqsave(&ep->com.dev->lock, flags); 335 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0); 336 if (idr_is_empty(&ep->com.dev->hwtid_idr)) 337 wake_up(&ep->com.dev->wait); 338 spin_unlock_irqrestore(&ep->com.dev->lock, flags); 339 } 340 341 static void insert_ep_tid(struct c4iw_ep *ep) 342 { 343 unsigned long flags; 344 345 spin_lock_irqsave(&ep->com.dev->lock, flags); 346 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0); 347 spin_unlock_irqrestore(&ep->com.dev->lock, flags); 348 } 349 350 /* 351 * Atomically lookup the ep ptr given the tid and grab a reference on the ep. 352 */ 353 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid) 354 { 355 struct c4iw_ep *ep; 356 unsigned long flags; 357 358 spin_lock_irqsave(&dev->lock, flags); 359 ep = idr_find(&dev->hwtid_idr, tid); 360 if (ep) 361 c4iw_get_ep(&ep->com); 362 spin_unlock_irqrestore(&dev->lock, flags); 363 return ep; 364 } 365 366 /* 367 * Atomically lookup the ep ptr given the stid and grab a reference on the ep. 368 */ 369 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev, 370 unsigned int stid) 371 { 372 struct c4iw_listen_ep *ep; 373 unsigned long flags; 374 375 spin_lock_irqsave(&dev->lock, flags); 376 ep = idr_find(&dev->stid_idr, stid); 377 if (ep) 378 c4iw_get_ep(&ep->com); 379 spin_unlock_irqrestore(&dev->lock, flags); 380 return ep; 381 } 382 383 void _c4iw_free_ep(struct kref *kref) 384 { 385 struct c4iw_ep *ep; 386 387 ep = container_of(kref, struct c4iw_ep, com.kref); 388 pr_debug("ep %p state %s\n", ep, states[ep->com.state]); 389 if (test_bit(QP_REFERENCED, &ep->com.flags)) 390 deref_qp(ep); 391 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) { 392 if (ep->com.remote_addr.ss_family == AF_INET6) { 393 struct sockaddr_in6 *sin6 = 394 (struct sockaddr_in6 *) 395 &ep->com.local_addr; 396 397 cxgb4_clip_release( 398 ep->com.dev->rdev.lldi.ports[0], 399 (const u32 *)&sin6->sin6_addr.s6_addr, 400 1); 401 } 402 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid, 403 ep->com.local_addr.ss_family); 404 dst_release(ep->dst); 405 cxgb4_l2t_release(ep->l2t); 406 kfree_skb(ep->mpa_skb); 407 } 408 if (!skb_queue_empty(&ep->com.ep_skb_list)) 409 skb_queue_purge(&ep->com.ep_skb_list); 410 c4iw_put_wr_wait(ep->com.wr_waitp); 411 kfree(ep); 412 } 413 414 static void release_ep_resources(struct c4iw_ep *ep) 415 { 416 set_bit(RELEASE_RESOURCES, &ep->com.flags); 417 418 /* 419 * If we have a hwtid, then remove it from the idr table 420 * so lookups will no longer find this endpoint. Otherwise 421 * we have a race where one thread finds the ep ptr just 422 * before the other thread is freeing the ep memory. 423 */ 424 if (ep->hwtid != -1) 425 remove_ep_tid(ep); 426 c4iw_put_ep(&ep->com); 427 } 428 429 static int status2errno(int status) 430 { 431 switch (status) { 432 case CPL_ERR_NONE: 433 return 0; 434 case CPL_ERR_CONN_RESET: 435 return -ECONNRESET; 436 case CPL_ERR_ARP_MISS: 437 return -EHOSTUNREACH; 438 case CPL_ERR_CONN_TIMEDOUT: 439 return -ETIMEDOUT; 440 case CPL_ERR_TCAM_FULL: 441 return -ENOMEM; 442 case CPL_ERR_CONN_EXIST: 443 return -EADDRINUSE; 444 default: 445 return -EIO; 446 } 447 } 448 449 /* 450 * Try and reuse skbs already allocated... 451 */ 452 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp) 453 { 454 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) { 455 skb_trim(skb, 0); 456 skb_get(skb); 457 skb_reset_transport_header(skb); 458 } else { 459 skb = alloc_skb(len, gfp); 460 } 461 t4_set_arp_err_handler(skb, NULL, NULL); 462 return skb; 463 } 464 465 static struct net_device *get_real_dev(struct net_device *egress_dev) 466 { 467 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev; 468 } 469 470 static void arp_failure_discard(void *handle, struct sk_buff *skb) 471 { 472 pr_err("ARP failure\n"); 473 kfree_skb(skb); 474 } 475 476 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb) 477 { 478 pr_err("ARP failure during MPA Negotiation - Closing Connection\n"); 479 } 480 481 enum { 482 NUM_FAKE_CPLS = 2, 483 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0, 484 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1, 485 }; 486 487 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb) 488 { 489 struct c4iw_ep *ep; 490 491 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))); 492 release_ep_resources(ep); 493 kfree_skb(skb); 494 return 0; 495 } 496 497 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb) 498 { 499 struct c4iw_ep *ep; 500 501 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))); 502 c4iw_put_ep(&ep->parent_ep->com); 503 release_ep_resources(ep); 504 kfree_skb(skb); 505 return 0; 506 } 507 508 /* 509 * Fake up a special CPL opcode and call sched() so process_work() will call 510 * _put_ep_safe() in a safe context to free the ep resources. This is needed 511 * because ARP error handlers are called in an ATOMIC context, and 512 * _c4iw_free_ep() needs to block. 513 */ 514 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb, 515 int cpl) 516 { 517 struct cpl_act_establish *rpl = cplhdr(skb); 518 519 /* Set our special ARP_FAILURE opcode */ 520 rpl->ot.opcode = cpl; 521 522 /* 523 * Save ep in the skb->cb area, after where sched() will save the dev 524 * ptr. 525 */ 526 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep; 527 sched(ep->com.dev, skb); 528 } 529 530 /* Handle an ARP failure for an accept */ 531 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb) 532 { 533 struct c4iw_ep *ep = handle; 534 535 pr_err("ARP failure during accept - tid %u - dropping connection\n", 536 ep->hwtid); 537 538 __state_set(&ep->com, DEAD); 539 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE); 540 } 541 542 /* 543 * Handle an ARP failure for an active open. 544 */ 545 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb) 546 { 547 struct c4iw_ep *ep = handle; 548 549 pr_err("ARP failure during connect\n"); 550 connect_reply_upcall(ep, -EHOSTUNREACH); 551 __state_set(&ep->com, DEAD); 552 if (ep->com.remote_addr.ss_family == AF_INET6) { 553 struct sockaddr_in6 *sin6 = 554 (struct sockaddr_in6 *)&ep->com.local_addr; 555 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 556 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 557 } 558 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid); 559 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid); 560 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE); 561 } 562 563 /* 564 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant 565 * and send it along. 566 */ 567 static void abort_arp_failure(void *handle, struct sk_buff *skb) 568 { 569 int ret; 570 struct c4iw_ep *ep = handle; 571 struct c4iw_rdev *rdev = &ep->com.dev->rdev; 572 struct cpl_abort_req *req = cplhdr(skb); 573 574 pr_debug("rdev %p\n", rdev); 575 req->cmd = CPL_ABORT_NO_RST; 576 skb_get(skb); 577 ret = c4iw_ofld_send(rdev, skb); 578 if (ret) { 579 __state_set(&ep->com, DEAD); 580 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE); 581 } else 582 kfree_skb(skb); 583 } 584 585 static int send_flowc(struct c4iw_ep *ep) 586 { 587 struct fw_flowc_wr *flowc; 588 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list); 589 u16 vlan = ep->l2t->vlan; 590 int nparams; 591 int flowclen, flowclen16; 592 593 if (WARN_ON(!skb)) 594 return -ENOMEM; 595 596 if (vlan == CPL_L2T_VLAN_NONE) 597 nparams = 9; 598 else 599 nparams = 10; 600 601 flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]); 602 flowclen16 = DIV_ROUND_UP(flowclen, 16); 603 flowclen = flowclen16 * 16; 604 605 flowc = __skb_put(skb, flowclen); 606 memset(flowc, 0, flowclen); 607 608 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) | 609 FW_FLOWC_WR_NPARAMS_V(nparams)); 610 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(flowclen16) | 611 FW_WR_FLOWID_V(ep->hwtid)); 612 613 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN; 614 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V 615 (ep->com.dev->rdev.lldi.pf)); 616 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH; 617 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan); 618 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT; 619 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan); 620 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID; 621 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid); 622 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT; 623 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq); 624 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT; 625 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq); 626 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF; 627 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win); 628 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS; 629 flowc->mnemval[7].val = cpu_to_be32(ep->emss); 630 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_RCV_SCALE; 631 flowc->mnemval[8].val = cpu_to_be32(ep->snd_wscale); 632 if (nparams == 10) { 633 u16 pri; 634 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT; 635 flowc->mnemval[9].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS; 636 flowc->mnemval[9].val = cpu_to_be32(pri); 637 } 638 639 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 640 return c4iw_ofld_send(&ep->com.dev->rdev, skb); 641 } 642 643 static int send_halfclose(struct c4iw_ep *ep) 644 { 645 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list); 646 u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16); 647 648 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 649 if (WARN_ON(!skb)) 650 return -ENOMEM; 651 652 cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx, 653 NULL, arp_failure_discard); 654 655 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 656 } 657 658 static void read_tcb(struct c4iw_ep *ep) 659 { 660 struct sk_buff *skb; 661 struct cpl_get_tcb *req; 662 int wrlen = roundup(sizeof(*req), 16); 663 664 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL); 665 if (WARN_ON(!skb)) 666 return; 667 668 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx); 669 req = (struct cpl_get_tcb *) skb_put(skb, wrlen); 670 memset(req, 0, wrlen); 671 INIT_TP_WR(req, ep->hwtid); 672 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_GET_TCB, ep->hwtid)); 673 req->reply_ctrl = htons(REPLY_CHAN_V(0) | QUEUENO_V(ep->rss_qid)); 674 675 /* 676 * keep a ref on the ep so the tcb is not unlocked before this 677 * cpl completes. The ref is released in read_tcb_rpl(). 678 */ 679 c4iw_get_ep(&ep->com); 680 if (WARN_ON(c4iw_ofld_send(&ep->com.dev->rdev, skb))) 681 c4iw_put_ep(&ep->com); 682 } 683 684 static int send_abort_req(struct c4iw_ep *ep) 685 { 686 u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16); 687 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list); 688 689 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 690 if (WARN_ON(!req_skb)) 691 return -ENOMEM; 692 693 cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx, 694 ep, abort_arp_failure); 695 696 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t); 697 } 698 699 static int send_abort(struct c4iw_ep *ep) 700 { 701 if (!ep->com.qp || !ep->com.qp->srq) { 702 send_abort_req(ep); 703 return 0; 704 } 705 set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags); 706 read_tcb(ep); 707 return 0; 708 } 709 710 static int send_connect(struct c4iw_ep *ep) 711 { 712 struct cpl_act_open_req *req = NULL; 713 struct cpl_t5_act_open_req *t5req = NULL; 714 struct cpl_t6_act_open_req *t6req = NULL; 715 struct cpl_act_open_req6 *req6 = NULL; 716 struct cpl_t5_act_open_req6 *t5req6 = NULL; 717 struct cpl_t6_act_open_req6 *t6req6 = NULL; 718 struct sk_buff *skb; 719 u64 opt0; 720 u32 opt2; 721 unsigned int mtu_idx; 722 u32 wscale; 723 int win, sizev4, sizev6, wrlen; 724 struct sockaddr_in *la = (struct sockaddr_in *) 725 &ep->com.local_addr; 726 struct sockaddr_in *ra = (struct sockaddr_in *) 727 &ep->com.remote_addr; 728 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *) 729 &ep->com.local_addr; 730 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *) 731 &ep->com.remote_addr; 732 int ret; 733 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type; 734 u32 isn = (prandom_u32() & ~7UL) - 1; 735 struct net_device *netdev; 736 u64 params; 737 738 netdev = ep->com.dev->rdev.lldi.ports[0]; 739 740 switch (CHELSIO_CHIP_VERSION(adapter_type)) { 741 case CHELSIO_T4: 742 sizev4 = sizeof(struct cpl_act_open_req); 743 sizev6 = sizeof(struct cpl_act_open_req6); 744 break; 745 case CHELSIO_T5: 746 sizev4 = sizeof(struct cpl_t5_act_open_req); 747 sizev6 = sizeof(struct cpl_t5_act_open_req6); 748 break; 749 case CHELSIO_T6: 750 sizev4 = sizeof(struct cpl_t6_act_open_req); 751 sizev6 = sizeof(struct cpl_t6_act_open_req6); 752 break; 753 default: 754 pr_err("T%d Chip is not supported\n", 755 CHELSIO_CHIP_VERSION(adapter_type)); 756 return -EINVAL; 757 } 758 759 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ? 760 roundup(sizev4, 16) : 761 roundup(sizev6, 16); 762 763 pr_debug("ep %p atid %u\n", ep, ep->atid); 764 765 skb = get_skb(NULL, wrlen, GFP_KERNEL); 766 if (!skb) { 767 pr_err("%s - failed to alloc skb\n", __func__); 768 return -ENOMEM; 769 } 770 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx); 771 772 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx, 773 enable_tcp_timestamps, 774 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1); 775 wscale = cxgb_compute_wscale(rcv_win); 776 777 /* 778 * Specify the largest window that will fit in opt0. The 779 * remainder will be specified in the rx_data_ack. 780 */ 781 win = ep->rcv_win >> 10; 782 if (win > RCV_BUFSIZ_M) 783 win = RCV_BUFSIZ_M; 784 785 opt0 = (nocong ? NO_CONG_F : 0) | 786 KEEP_ALIVE_F | 787 DELACK_F | 788 WND_SCALE_V(wscale) | 789 MSS_IDX_V(mtu_idx) | 790 L2T_IDX_V(ep->l2t->idx) | 791 TX_CHAN_V(ep->tx_chan) | 792 SMAC_SEL_V(ep->smac_idx) | 793 DSCP_V(ep->tos >> 2) | 794 ULP_MODE_V(ULP_MODE_TCPDDP) | 795 RCV_BUFSIZ_V(win); 796 opt2 = RX_CHANNEL_V(0) | 797 CCTRL_ECN_V(enable_ecn) | 798 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid); 799 if (enable_tcp_timestamps) 800 opt2 |= TSTAMPS_EN_F; 801 if (enable_tcp_sack) 802 opt2 |= SACK_EN_F; 803 if (wscale && enable_tcp_window_scaling) 804 opt2 |= WND_SCALE_EN_F; 805 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) { 806 if (peer2peer) 807 isn += 4; 808 809 opt2 |= T5_OPT_2_VALID_F; 810 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE); 811 opt2 |= T5_ISS_F; 812 } 813 814 params = cxgb4_select_ntuple(netdev, ep->l2t); 815 816 if (ep->com.remote_addr.ss_family == AF_INET6) 817 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0], 818 (const u32 *)&la6->sin6_addr.s6_addr, 1); 819 820 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure); 821 822 if (ep->com.remote_addr.ss_family == AF_INET) { 823 switch (CHELSIO_CHIP_VERSION(adapter_type)) { 824 case CHELSIO_T4: 825 req = skb_put(skb, wrlen); 826 INIT_TP_WR(req, 0); 827 break; 828 case CHELSIO_T5: 829 t5req = skb_put(skb, wrlen); 830 INIT_TP_WR(t5req, 0); 831 req = (struct cpl_act_open_req *)t5req; 832 break; 833 case CHELSIO_T6: 834 t6req = skb_put(skb, wrlen); 835 INIT_TP_WR(t6req, 0); 836 req = (struct cpl_act_open_req *)t6req; 837 t5req = (struct cpl_t5_act_open_req *)t6req; 838 break; 839 default: 840 pr_err("T%d Chip is not supported\n", 841 CHELSIO_CHIP_VERSION(adapter_type)); 842 ret = -EINVAL; 843 goto clip_release; 844 } 845 846 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, 847 ((ep->rss_qid<<14) | ep->atid))); 848 req->local_port = la->sin_port; 849 req->peer_port = ra->sin_port; 850 req->local_ip = la->sin_addr.s_addr; 851 req->peer_ip = ra->sin_addr.s_addr; 852 req->opt0 = cpu_to_be64(opt0); 853 854 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) { 855 req->params = cpu_to_be32(params); 856 req->opt2 = cpu_to_be32(opt2); 857 } else { 858 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) { 859 t5req->params = 860 cpu_to_be64(FILTER_TUPLE_V(params)); 861 t5req->rsvd = cpu_to_be32(isn); 862 pr_debug("snd_isn %u\n", t5req->rsvd); 863 t5req->opt2 = cpu_to_be32(opt2); 864 } else { 865 t6req->params = 866 cpu_to_be64(FILTER_TUPLE_V(params)); 867 t6req->rsvd = cpu_to_be32(isn); 868 pr_debug("snd_isn %u\n", t6req->rsvd); 869 t6req->opt2 = cpu_to_be32(opt2); 870 } 871 } 872 } else { 873 switch (CHELSIO_CHIP_VERSION(adapter_type)) { 874 case CHELSIO_T4: 875 req6 = skb_put(skb, wrlen); 876 INIT_TP_WR(req6, 0); 877 break; 878 case CHELSIO_T5: 879 t5req6 = skb_put(skb, wrlen); 880 INIT_TP_WR(t5req6, 0); 881 req6 = (struct cpl_act_open_req6 *)t5req6; 882 break; 883 case CHELSIO_T6: 884 t6req6 = skb_put(skb, wrlen); 885 INIT_TP_WR(t6req6, 0); 886 req6 = (struct cpl_act_open_req6 *)t6req6; 887 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6; 888 break; 889 default: 890 pr_err("T%d Chip is not supported\n", 891 CHELSIO_CHIP_VERSION(adapter_type)); 892 ret = -EINVAL; 893 goto clip_release; 894 } 895 896 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, 897 ((ep->rss_qid<<14)|ep->atid))); 898 req6->local_port = la6->sin6_port; 899 req6->peer_port = ra6->sin6_port; 900 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr)); 901 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8)); 902 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr)); 903 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8)); 904 req6->opt0 = cpu_to_be64(opt0); 905 906 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) { 907 req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev, 908 ep->l2t)); 909 req6->opt2 = cpu_to_be32(opt2); 910 } else { 911 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) { 912 t5req6->params = 913 cpu_to_be64(FILTER_TUPLE_V(params)); 914 t5req6->rsvd = cpu_to_be32(isn); 915 pr_debug("snd_isn %u\n", t5req6->rsvd); 916 t5req6->opt2 = cpu_to_be32(opt2); 917 } else { 918 t6req6->params = 919 cpu_to_be64(FILTER_TUPLE_V(params)); 920 t6req6->rsvd = cpu_to_be32(isn); 921 pr_debug("snd_isn %u\n", t6req6->rsvd); 922 t6req6->opt2 = cpu_to_be32(opt2); 923 } 924 925 } 926 } 927 928 set_bit(ACT_OPEN_REQ, &ep->com.history); 929 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 930 clip_release: 931 if (ret && ep->com.remote_addr.ss_family == AF_INET6) 932 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 933 (const u32 *)&la6->sin6_addr.s6_addr, 1); 934 return ret; 935 } 936 937 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb, 938 u8 mpa_rev_to_use) 939 { 940 int mpalen, wrlen, ret; 941 struct fw_ofld_tx_data_wr *req; 942 struct mpa_message *mpa; 943 struct mpa_v2_conn_params mpa_v2_params; 944 945 pr_debug("ep %p tid %u pd_len %d\n", 946 ep, ep->hwtid, ep->plen); 947 948 mpalen = sizeof(*mpa) + ep->plen; 949 if (mpa_rev_to_use == 2) 950 mpalen += sizeof(struct mpa_v2_conn_params); 951 wrlen = roundup(mpalen + sizeof *req, 16); 952 skb = get_skb(skb, wrlen, GFP_KERNEL); 953 if (!skb) { 954 connect_reply_upcall(ep, -ENOMEM); 955 return -ENOMEM; 956 } 957 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 958 959 req = skb_put_zero(skb, wrlen); 960 req->op_to_immdlen = cpu_to_be32( 961 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) | 962 FW_WR_COMPL_F | 963 FW_WR_IMMDLEN_V(mpalen)); 964 req->flowid_len16 = cpu_to_be32( 965 FW_WR_FLOWID_V(ep->hwtid) | 966 FW_WR_LEN16_V(wrlen >> 4)); 967 req->plen = cpu_to_be32(mpalen); 968 req->tunnel_to_proxy = cpu_to_be32( 969 FW_OFLD_TX_DATA_WR_FLUSH_F | 970 FW_OFLD_TX_DATA_WR_SHOVE_F); 971 972 mpa = (struct mpa_message *)(req + 1); 973 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)); 974 975 mpa->flags = 0; 976 if (crc_enabled) 977 mpa->flags |= MPA_CRC; 978 if (markers_enabled) { 979 mpa->flags |= MPA_MARKERS; 980 ep->mpa_attr.recv_marker_enabled = 1; 981 } else { 982 ep->mpa_attr.recv_marker_enabled = 0; 983 } 984 if (mpa_rev_to_use == 2) 985 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 986 987 mpa->private_data_size = htons(ep->plen); 988 mpa->revision = mpa_rev_to_use; 989 if (mpa_rev_to_use == 1) { 990 ep->tried_with_mpa_v1 = 1; 991 ep->retry_with_mpa_v1 = 0; 992 } 993 994 if (mpa_rev_to_use == 2) { 995 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 996 sizeof (struct mpa_v2_conn_params)); 997 pr_debug("initiator ird %u ord %u\n", ep->ird, 998 ep->ord); 999 mpa_v2_params.ird = htons((u16)ep->ird); 1000 mpa_v2_params.ord = htons((u16)ep->ord); 1001 1002 if (peer2peer) { 1003 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1004 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) 1005 mpa_v2_params.ord |= 1006 htons(MPA_V2_RDMA_WRITE_RTR); 1007 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) 1008 mpa_v2_params.ord |= 1009 htons(MPA_V2_RDMA_READ_RTR); 1010 } 1011 memcpy(mpa->private_data, &mpa_v2_params, 1012 sizeof(struct mpa_v2_conn_params)); 1013 1014 if (ep->plen) 1015 memcpy(mpa->private_data + 1016 sizeof(struct mpa_v2_conn_params), 1017 ep->mpa_pkt + sizeof(*mpa), ep->plen); 1018 } else 1019 if (ep->plen) 1020 memcpy(mpa->private_data, 1021 ep->mpa_pkt + sizeof(*mpa), ep->plen); 1022 1023 /* 1024 * Reference the mpa skb. This ensures the data area 1025 * will remain in memory until the hw acks the tx. 1026 * Function fw4_ack() will deref it. 1027 */ 1028 skb_get(skb); 1029 t4_set_arp_err_handler(skb, NULL, arp_failure_discard); 1030 ep->mpa_skb = skb; 1031 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1032 if (ret) 1033 return ret; 1034 start_ep_timer(ep); 1035 __state_set(&ep->com, MPA_REQ_SENT); 1036 ep->mpa_attr.initiator = 1; 1037 ep->snd_seq += mpalen; 1038 return ret; 1039 } 1040 1041 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen) 1042 { 1043 int mpalen, wrlen; 1044 struct fw_ofld_tx_data_wr *req; 1045 struct mpa_message *mpa; 1046 struct sk_buff *skb; 1047 struct mpa_v2_conn_params mpa_v2_params; 1048 1049 pr_debug("ep %p tid %u pd_len %d\n", 1050 ep, ep->hwtid, ep->plen); 1051 1052 mpalen = sizeof(*mpa) + plen; 1053 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) 1054 mpalen += sizeof(struct mpa_v2_conn_params); 1055 wrlen = roundup(mpalen + sizeof *req, 16); 1056 1057 skb = get_skb(NULL, wrlen, GFP_KERNEL); 1058 if (!skb) { 1059 pr_err("%s - cannot alloc skb!\n", __func__); 1060 return -ENOMEM; 1061 } 1062 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 1063 1064 req = skb_put_zero(skb, wrlen); 1065 req->op_to_immdlen = cpu_to_be32( 1066 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) | 1067 FW_WR_COMPL_F | 1068 FW_WR_IMMDLEN_V(mpalen)); 1069 req->flowid_len16 = cpu_to_be32( 1070 FW_WR_FLOWID_V(ep->hwtid) | 1071 FW_WR_LEN16_V(wrlen >> 4)); 1072 req->plen = cpu_to_be32(mpalen); 1073 req->tunnel_to_proxy = cpu_to_be32( 1074 FW_OFLD_TX_DATA_WR_FLUSH_F | 1075 FW_OFLD_TX_DATA_WR_SHOVE_F); 1076 1077 mpa = (struct mpa_message *)(req + 1); 1078 memset(mpa, 0, sizeof(*mpa)); 1079 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1080 mpa->flags = MPA_REJECT; 1081 mpa->revision = ep->mpa_attr.version; 1082 mpa->private_data_size = htons(plen); 1083 1084 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1085 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1086 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1087 sizeof (struct mpa_v2_conn_params)); 1088 mpa_v2_params.ird = htons(((u16)ep->ird) | 1089 (peer2peer ? MPA_V2_PEER2PEER_MODEL : 1090 0)); 1091 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ? 1092 (p2p_type == 1093 FW_RI_INIT_P2PTYPE_RDMA_WRITE ? 1094 MPA_V2_RDMA_WRITE_RTR : p2p_type == 1095 FW_RI_INIT_P2PTYPE_READ_REQ ? 1096 MPA_V2_RDMA_READ_RTR : 0) : 0)); 1097 memcpy(mpa->private_data, &mpa_v2_params, 1098 sizeof(struct mpa_v2_conn_params)); 1099 1100 if (ep->plen) 1101 memcpy(mpa->private_data + 1102 sizeof(struct mpa_v2_conn_params), pdata, plen); 1103 } else 1104 if (plen) 1105 memcpy(mpa->private_data, pdata, plen); 1106 1107 /* 1108 * Reference the mpa skb again. This ensures the data area 1109 * will remain in memory until the hw acks the tx. 1110 * Function fw4_ack() will deref it. 1111 */ 1112 skb_get(skb); 1113 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 1114 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure); 1115 ep->mpa_skb = skb; 1116 ep->snd_seq += mpalen; 1117 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1118 } 1119 1120 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen) 1121 { 1122 int mpalen, wrlen; 1123 struct fw_ofld_tx_data_wr *req; 1124 struct mpa_message *mpa; 1125 struct sk_buff *skb; 1126 struct mpa_v2_conn_params mpa_v2_params; 1127 1128 pr_debug("ep %p tid %u pd_len %d\n", 1129 ep, ep->hwtid, ep->plen); 1130 1131 mpalen = sizeof(*mpa) + plen; 1132 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) 1133 mpalen += sizeof(struct mpa_v2_conn_params); 1134 wrlen = roundup(mpalen + sizeof *req, 16); 1135 1136 skb = get_skb(NULL, wrlen, GFP_KERNEL); 1137 if (!skb) { 1138 pr_err("%s - cannot alloc skb!\n", __func__); 1139 return -ENOMEM; 1140 } 1141 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 1142 1143 req = skb_put_zero(skb, wrlen); 1144 req->op_to_immdlen = cpu_to_be32( 1145 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) | 1146 FW_WR_COMPL_F | 1147 FW_WR_IMMDLEN_V(mpalen)); 1148 req->flowid_len16 = cpu_to_be32( 1149 FW_WR_FLOWID_V(ep->hwtid) | 1150 FW_WR_LEN16_V(wrlen >> 4)); 1151 req->plen = cpu_to_be32(mpalen); 1152 req->tunnel_to_proxy = cpu_to_be32( 1153 FW_OFLD_TX_DATA_WR_FLUSH_F | 1154 FW_OFLD_TX_DATA_WR_SHOVE_F); 1155 1156 mpa = (struct mpa_message *)(req + 1); 1157 memset(mpa, 0, sizeof(*mpa)); 1158 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1159 mpa->flags = 0; 1160 if (ep->mpa_attr.crc_enabled) 1161 mpa->flags |= MPA_CRC; 1162 if (ep->mpa_attr.recv_marker_enabled) 1163 mpa->flags |= MPA_MARKERS; 1164 mpa->revision = ep->mpa_attr.version; 1165 mpa->private_data_size = htons(plen); 1166 1167 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1168 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1169 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1170 sizeof (struct mpa_v2_conn_params)); 1171 mpa_v2_params.ird = htons((u16)ep->ird); 1172 mpa_v2_params.ord = htons((u16)ep->ord); 1173 if (peer2peer && (ep->mpa_attr.p2p_type != 1174 FW_RI_INIT_P2PTYPE_DISABLED)) { 1175 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1176 1177 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) 1178 mpa_v2_params.ord |= 1179 htons(MPA_V2_RDMA_WRITE_RTR); 1180 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) 1181 mpa_v2_params.ord |= 1182 htons(MPA_V2_RDMA_READ_RTR); 1183 } 1184 1185 memcpy(mpa->private_data, &mpa_v2_params, 1186 sizeof(struct mpa_v2_conn_params)); 1187 1188 if (ep->plen) 1189 memcpy(mpa->private_data + 1190 sizeof(struct mpa_v2_conn_params), pdata, plen); 1191 } else 1192 if (plen) 1193 memcpy(mpa->private_data, pdata, plen); 1194 1195 /* 1196 * Reference the mpa skb. This ensures the data area 1197 * will remain in memory until the hw acks the tx. 1198 * Function fw4_ack() will deref it. 1199 */ 1200 skb_get(skb); 1201 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure); 1202 ep->mpa_skb = skb; 1203 __state_set(&ep->com, MPA_REP_SENT); 1204 ep->snd_seq += mpalen; 1205 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1206 } 1207 1208 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb) 1209 { 1210 struct c4iw_ep *ep; 1211 struct cpl_act_establish *req = cplhdr(skb); 1212 unsigned short tcp_opt = ntohs(req->tcp_opt); 1213 unsigned int tid = GET_TID(req); 1214 unsigned int atid = TID_TID_G(ntohl(req->tos_atid)); 1215 struct tid_info *t = dev->rdev.lldi.tids; 1216 int ret; 1217 1218 ep = lookup_atid(t, atid); 1219 1220 pr_debug("ep %p tid %u snd_isn %u rcv_isn %u\n", ep, tid, 1221 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn)); 1222 1223 mutex_lock(&ep->com.mutex); 1224 dst_confirm(ep->dst); 1225 1226 /* setup the hwtid for this connection */ 1227 ep->hwtid = tid; 1228 cxgb4_insert_tid(t, ep, tid, ep->com.local_addr.ss_family); 1229 insert_ep_tid(ep); 1230 1231 ep->snd_seq = be32_to_cpu(req->snd_isn); 1232 ep->rcv_seq = be32_to_cpu(req->rcv_isn); 1233 ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt); 1234 1235 set_emss(ep, tcp_opt); 1236 1237 /* dealloc the atid */ 1238 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid); 1239 cxgb4_free_atid(t, atid); 1240 set_bit(ACT_ESTAB, &ep->com.history); 1241 1242 /* start MPA negotiation */ 1243 ret = send_flowc(ep); 1244 if (ret) 1245 goto err; 1246 if (ep->retry_with_mpa_v1) 1247 ret = send_mpa_req(ep, skb, 1); 1248 else 1249 ret = send_mpa_req(ep, skb, mpa_rev); 1250 if (ret) 1251 goto err; 1252 mutex_unlock(&ep->com.mutex); 1253 return 0; 1254 err: 1255 mutex_unlock(&ep->com.mutex); 1256 connect_reply_upcall(ep, -ENOMEM); 1257 c4iw_ep_disconnect(ep, 0, GFP_KERNEL); 1258 return 0; 1259 } 1260 1261 static void close_complete_upcall(struct c4iw_ep *ep, int status) 1262 { 1263 struct iw_cm_event event; 1264 1265 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 1266 memset(&event, 0, sizeof(event)); 1267 event.event = IW_CM_EVENT_CLOSE; 1268 event.status = status; 1269 if (ep->com.cm_id) { 1270 pr_debug("close complete delivered ep %p cm_id %p tid %u\n", 1271 ep, ep->com.cm_id, ep->hwtid); 1272 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1273 deref_cm_id(&ep->com); 1274 set_bit(CLOSE_UPCALL, &ep->com.history); 1275 } 1276 } 1277 1278 static void peer_close_upcall(struct c4iw_ep *ep) 1279 { 1280 struct iw_cm_event event; 1281 1282 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 1283 memset(&event, 0, sizeof(event)); 1284 event.event = IW_CM_EVENT_DISCONNECT; 1285 if (ep->com.cm_id) { 1286 pr_debug("peer close delivered ep %p cm_id %p tid %u\n", 1287 ep, ep->com.cm_id, ep->hwtid); 1288 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1289 set_bit(DISCONN_UPCALL, &ep->com.history); 1290 } 1291 } 1292 1293 static void peer_abort_upcall(struct c4iw_ep *ep) 1294 { 1295 struct iw_cm_event event; 1296 1297 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 1298 memset(&event, 0, sizeof(event)); 1299 event.event = IW_CM_EVENT_CLOSE; 1300 event.status = -ECONNRESET; 1301 if (ep->com.cm_id) { 1302 pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep, 1303 ep->com.cm_id, ep->hwtid); 1304 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1305 deref_cm_id(&ep->com); 1306 set_bit(ABORT_UPCALL, &ep->com.history); 1307 } 1308 } 1309 1310 static void connect_reply_upcall(struct c4iw_ep *ep, int status) 1311 { 1312 struct iw_cm_event event; 1313 1314 pr_debug("ep %p tid %u status %d\n", 1315 ep, ep->hwtid, status); 1316 memset(&event, 0, sizeof(event)); 1317 event.event = IW_CM_EVENT_CONNECT_REPLY; 1318 event.status = status; 1319 memcpy(&event.local_addr, &ep->com.local_addr, 1320 sizeof(ep->com.local_addr)); 1321 memcpy(&event.remote_addr, &ep->com.remote_addr, 1322 sizeof(ep->com.remote_addr)); 1323 1324 if ((status == 0) || (status == -ECONNREFUSED)) { 1325 if (!ep->tried_with_mpa_v1) { 1326 /* this means MPA_v2 is used */ 1327 event.ord = ep->ird; 1328 event.ird = ep->ord; 1329 event.private_data_len = ep->plen - 1330 sizeof(struct mpa_v2_conn_params); 1331 event.private_data = ep->mpa_pkt + 1332 sizeof(struct mpa_message) + 1333 sizeof(struct mpa_v2_conn_params); 1334 } else { 1335 /* this means MPA_v1 is used */ 1336 event.ord = cur_max_read_depth(ep->com.dev); 1337 event.ird = cur_max_read_depth(ep->com.dev); 1338 event.private_data_len = ep->plen; 1339 event.private_data = ep->mpa_pkt + 1340 sizeof(struct mpa_message); 1341 } 1342 } 1343 1344 pr_debug("ep %p tid %u status %d\n", ep, 1345 ep->hwtid, status); 1346 set_bit(CONN_RPL_UPCALL, &ep->com.history); 1347 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1348 1349 if (status < 0) 1350 deref_cm_id(&ep->com); 1351 } 1352 1353 static int connect_request_upcall(struct c4iw_ep *ep) 1354 { 1355 struct iw_cm_event event; 1356 int ret; 1357 1358 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 1359 memset(&event, 0, sizeof(event)); 1360 event.event = IW_CM_EVENT_CONNECT_REQUEST; 1361 memcpy(&event.local_addr, &ep->com.local_addr, 1362 sizeof(ep->com.local_addr)); 1363 memcpy(&event.remote_addr, &ep->com.remote_addr, 1364 sizeof(ep->com.remote_addr)); 1365 event.provider_data = ep; 1366 if (!ep->tried_with_mpa_v1) { 1367 /* this means MPA_v2 is used */ 1368 event.ord = ep->ord; 1369 event.ird = ep->ird; 1370 event.private_data_len = ep->plen - 1371 sizeof(struct mpa_v2_conn_params); 1372 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) + 1373 sizeof(struct mpa_v2_conn_params); 1374 } else { 1375 /* this means MPA_v1 is used. Send max supported */ 1376 event.ord = cur_max_read_depth(ep->com.dev); 1377 event.ird = cur_max_read_depth(ep->com.dev); 1378 event.private_data_len = ep->plen; 1379 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); 1380 } 1381 c4iw_get_ep(&ep->com); 1382 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id, 1383 &event); 1384 if (ret) 1385 c4iw_put_ep(&ep->com); 1386 set_bit(CONNREQ_UPCALL, &ep->com.history); 1387 c4iw_put_ep(&ep->parent_ep->com); 1388 return ret; 1389 } 1390 1391 static void established_upcall(struct c4iw_ep *ep) 1392 { 1393 struct iw_cm_event event; 1394 1395 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 1396 memset(&event, 0, sizeof(event)); 1397 event.event = IW_CM_EVENT_ESTABLISHED; 1398 event.ird = ep->ord; 1399 event.ord = ep->ird; 1400 if (ep->com.cm_id) { 1401 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 1402 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1403 set_bit(ESTAB_UPCALL, &ep->com.history); 1404 } 1405 } 1406 1407 static int update_rx_credits(struct c4iw_ep *ep, u32 credits) 1408 { 1409 struct sk_buff *skb; 1410 u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16); 1411 u32 credit_dack; 1412 1413 pr_debug("ep %p tid %u credits %u\n", 1414 ep, ep->hwtid, credits); 1415 skb = get_skb(NULL, wrlen, GFP_KERNEL); 1416 if (!skb) { 1417 pr_err("update_rx_credits - cannot alloc skb!\n"); 1418 return 0; 1419 } 1420 1421 /* 1422 * If we couldn't specify the entire rcv window at connection setup 1423 * due to the limit in the number of bits in the RCV_BUFSIZ field, 1424 * then add the overage in to the credits returned. 1425 */ 1426 if (ep->rcv_win > RCV_BUFSIZ_M * 1024) 1427 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024; 1428 1429 credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F | 1430 RX_DACK_MODE_V(dack_mode); 1431 1432 cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx, 1433 credit_dack); 1434 1435 c4iw_ofld_send(&ep->com.dev->rdev, skb); 1436 return credits; 1437 } 1438 1439 #define RELAXED_IRD_NEGOTIATION 1 1440 1441 /* 1442 * process_mpa_reply - process streaming mode MPA reply 1443 * 1444 * Returns: 1445 * 1446 * 0 upon success indicating a connect request was delivered to the ULP 1447 * or the mpa request is incomplete but valid so far. 1448 * 1449 * 1 if a failure requires the caller to close the connection. 1450 * 1451 * 2 if a failure requires the caller to abort the connection. 1452 */ 1453 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb) 1454 { 1455 struct mpa_message *mpa; 1456 struct mpa_v2_conn_params *mpa_v2_params; 1457 u16 plen; 1458 u16 resp_ird, resp_ord; 1459 u8 rtr_mismatch = 0, insuff_ird = 0; 1460 struct c4iw_qp_attributes attrs; 1461 enum c4iw_qp_attr_mask mask; 1462 int err; 1463 int disconnect = 0; 1464 1465 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 1466 1467 /* 1468 * If we get more than the supported amount of private data 1469 * then we must fail this connection. 1470 */ 1471 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) { 1472 err = -EINVAL; 1473 goto err_stop_timer; 1474 } 1475 1476 /* 1477 * copy the new data into our accumulation buffer. 1478 */ 1479 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]), 1480 skb->len); 1481 ep->mpa_pkt_len += skb->len; 1482 1483 /* 1484 * if we don't even have the mpa message, then bail. 1485 */ 1486 if (ep->mpa_pkt_len < sizeof(*mpa)) 1487 return 0; 1488 mpa = (struct mpa_message *) ep->mpa_pkt; 1489 1490 /* Validate MPA header. */ 1491 if (mpa->revision > mpa_rev) { 1492 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n", 1493 __func__, mpa_rev, mpa->revision); 1494 err = -EPROTO; 1495 goto err_stop_timer; 1496 } 1497 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) { 1498 err = -EPROTO; 1499 goto err_stop_timer; 1500 } 1501 1502 plen = ntohs(mpa->private_data_size); 1503 1504 /* 1505 * Fail if there's too much private data. 1506 */ 1507 if (plen > MPA_MAX_PRIVATE_DATA) { 1508 err = -EPROTO; 1509 goto err_stop_timer; 1510 } 1511 1512 /* 1513 * If plen does not account for pkt size 1514 */ 1515 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) { 1516 err = -EPROTO; 1517 goto err_stop_timer; 1518 } 1519 1520 ep->plen = (u8) plen; 1521 1522 /* 1523 * If we don't have all the pdata yet, then bail. 1524 * We'll continue process when more data arrives. 1525 */ 1526 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) 1527 return 0; 1528 1529 if (mpa->flags & MPA_REJECT) { 1530 err = -ECONNREFUSED; 1531 goto err_stop_timer; 1532 } 1533 1534 /* 1535 * Stop mpa timer. If it expired, then 1536 * we ignore the MPA reply. process_timeout() 1537 * will abort the connection. 1538 */ 1539 if (stop_ep_timer(ep)) 1540 return 0; 1541 1542 /* 1543 * If we get here we have accumulated the entire mpa 1544 * start reply message including private data. And 1545 * the MPA header is valid. 1546 */ 1547 __state_set(&ep->com, FPDU_MODE); 1548 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1549 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1550 ep->mpa_attr.version = mpa->revision; 1551 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1552 1553 if (mpa->revision == 2) { 1554 ep->mpa_attr.enhanced_rdma_conn = 1555 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 1556 if (ep->mpa_attr.enhanced_rdma_conn) { 1557 mpa_v2_params = (struct mpa_v2_conn_params *) 1558 (ep->mpa_pkt + sizeof(*mpa)); 1559 resp_ird = ntohs(mpa_v2_params->ird) & 1560 MPA_V2_IRD_ORD_MASK; 1561 resp_ord = ntohs(mpa_v2_params->ord) & 1562 MPA_V2_IRD_ORD_MASK; 1563 pr_debug("responder ird %u ord %u ep ird %u ord %u\n", 1564 resp_ird, resp_ord, ep->ird, ep->ord); 1565 1566 /* 1567 * This is a double-check. Ideally, below checks are 1568 * not required since ird/ord stuff has been taken 1569 * care of in c4iw_accept_cr 1570 */ 1571 if (ep->ird < resp_ord) { 1572 if (RELAXED_IRD_NEGOTIATION && resp_ord <= 1573 ep->com.dev->rdev.lldi.max_ordird_qp) 1574 ep->ird = resp_ord; 1575 else 1576 insuff_ird = 1; 1577 } else if (ep->ird > resp_ord) { 1578 ep->ird = resp_ord; 1579 } 1580 if (ep->ord > resp_ird) { 1581 if (RELAXED_IRD_NEGOTIATION) 1582 ep->ord = resp_ird; 1583 else 1584 insuff_ird = 1; 1585 } 1586 if (insuff_ird) { 1587 err = -ENOMEM; 1588 ep->ird = resp_ord; 1589 ep->ord = resp_ird; 1590 } 1591 1592 if (ntohs(mpa_v2_params->ird) & 1593 MPA_V2_PEER2PEER_MODEL) { 1594 if (ntohs(mpa_v2_params->ord) & 1595 MPA_V2_RDMA_WRITE_RTR) 1596 ep->mpa_attr.p2p_type = 1597 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 1598 else if (ntohs(mpa_v2_params->ord) & 1599 MPA_V2_RDMA_READ_RTR) 1600 ep->mpa_attr.p2p_type = 1601 FW_RI_INIT_P2PTYPE_READ_REQ; 1602 } 1603 } 1604 } else if (mpa->revision == 1) 1605 if (peer2peer) 1606 ep->mpa_attr.p2p_type = p2p_type; 1607 1608 pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n", 1609 ep->mpa_attr.crc_enabled, 1610 ep->mpa_attr.recv_marker_enabled, 1611 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, 1612 ep->mpa_attr.p2p_type, p2p_type); 1613 1614 /* 1615 * If responder's RTR does not match with that of initiator, assign 1616 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not 1617 * generated when moving QP to RTS state. 1618 * A TERM message will be sent after QP has moved to RTS state 1619 */ 1620 if ((ep->mpa_attr.version == 2) && peer2peer && 1621 (ep->mpa_attr.p2p_type != p2p_type)) { 1622 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1623 rtr_mismatch = 1; 1624 } 1625 1626 attrs.mpa_attr = ep->mpa_attr; 1627 attrs.max_ird = ep->ird; 1628 attrs.max_ord = ep->ord; 1629 attrs.llp_stream_handle = ep; 1630 attrs.next_state = C4IW_QP_STATE_RTS; 1631 1632 mask = C4IW_QP_ATTR_NEXT_STATE | 1633 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR | 1634 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD; 1635 1636 /* bind QP and TID with INIT_WR */ 1637 err = c4iw_modify_qp(ep->com.qp->rhp, 1638 ep->com.qp, mask, &attrs, 1); 1639 if (err) 1640 goto err; 1641 1642 /* 1643 * If responder's RTR requirement did not match with what initiator 1644 * supports, generate TERM message 1645 */ 1646 if (rtr_mismatch) { 1647 pr_err("%s: RTR mismatch, sending TERM\n", __func__); 1648 attrs.layer_etype = LAYER_MPA | DDP_LLP; 1649 attrs.ecode = MPA_NOMATCH_RTR; 1650 attrs.next_state = C4IW_QP_STATE_TERMINATE; 1651 attrs.send_term = 1; 1652 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 1653 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 1654 err = -ENOMEM; 1655 disconnect = 1; 1656 goto out; 1657 } 1658 1659 /* 1660 * Generate TERM if initiator IRD is not sufficient for responder 1661 * provided ORD. Currently, we do the same behaviour even when 1662 * responder provided IRD is also not sufficient as regards to 1663 * initiator ORD. 1664 */ 1665 if (insuff_ird) { 1666 pr_err("%s: Insufficient IRD, sending TERM\n", __func__); 1667 attrs.layer_etype = LAYER_MPA | DDP_LLP; 1668 attrs.ecode = MPA_INSUFF_IRD; 1669 attrs.next_state = C4IW_QP_STATE_TERMINATE; 1670 attrs.send_term = 1; 1671 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 1672 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 1673 err = -ENOMEM; 1674 disconnect = 1; 1675 goto out; 1676 } 1677 goto out; 1678 err_stop_timer: 1679 stop_ep_timer(ep); 1680 err: 1681 disconnect = 2; 1682 out: 1683 connect_reply_upcall(ep, err); 1684 return disconnect; 1685 } 1686 1687 /* 1688 * process_mpa_request - process streaming mode MPA request 1689 * 1690 * Returns: 1691 * 1692 * 0 upon success indicating a connect request was delivered to the ULP 1693 * or the mpa request is incomplete but valid so far. 1694 * 1695 * 1 if a failure requires the caller to close the connection. 1696 * 1697 * 2 if a failure requires the caller to abort the connection. 1698 */ 1699 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb) 1700 { 1701 struct mpa_message *mpa; 1702 struct mpa_v2_conn_params *mpa_v2_params; 1703 u16 plen; 1704 1705 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 1706 1707 /* 1708 * If we get more than the supported amount of private data 1709 * then we must fail this connection. 1710 */ 1711 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) 1712 goto err_stop_timer; 1713 1714 pr_debug("enter (%s line %u)\n", __FILE__, __LINE__); 1715 1716 /* 1717 * Copy the new data into our accumulation buffer. 1718 */ 1719 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]), 1720 skb->len); 1721 ep->mpa_pkt_len += skb->len; 1722 1723 /* 1724 * If we don't even have the mpa message, then bail. 1725 * We'll continue process when more data arrives. 1726 */ 1727 if (ep->mpa_pkt_len < sizeof(*mpa)) 1728 return 0; 1729 1730 pr_debug("enter (%s line %u)\n", __FILE__, __LINE__); 1731 mpa = (struct mpa_message *) ep->mpa_pkt; 1732 1733 /* 1734 * Validate MPA Header. 1735 */ 1736 if (mpa->revision > mpa_rev) { 1737 pr_err("%s MPA version mismatch. Local = %d, Received = %d\n", 1738 __func__, mpa_rev, mpa->revision); 1739 goto err_stop_timer; 1740 } 1741 1742 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) 1743 goto err_stop_timer; 1744 1745 plen = ntohs(mpa->private_data_size); 1746 1747 /* 1748 * Fail if there's too much private data. 1749 */ 1750 if (plen > MPA_MAX_PRIVATE_DATA) 1751 goto err_stop_timer; 1752 1753 /* 1754 * If plen does not account for pkt size 1755 */ 1756 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) 1757 goto err_stop_timer; 1758 ep->plen = (u8) plen; 1759 1760 /* 1761 * If we don't have all the pdata yet, then bail. 1762 */ 1763 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) 1764 return 0; 1765 1766 /* 1767 * If we get here we have accumulated the entire mpa 1768 * start reply message including private data. 1769 */ 1770 ep->mpa_attr.initiator = 0; 1771 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1772 ep->mpa_attr.recv_marker_enabled = markers_enabled; 1773 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1774 ep->mpa_attr.version = mpa->revision; 1775 if (mpa->revision == 1) 1776 ep->tried_with_mpa_v1 = 1; 1777 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1778 1779 if (mpa->revision == 2) { 1780 ep->mpa_attr.enhanced_rdma_conn = 1781 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 1782 if (ep->mpa_attr.enhanced_rdma_conn) { 1783 mpa_v2_params = (struct mpa_v2_conn_params *) 1784 (ep->mpa_pkt + sizeof(*mpa)); 1785 ep->ird = ntohs(mpa_v2_params->ird) & 1786 MPA_V2_IRD_ORD_MASK; 1787 ep->ird = min_t(u32, ep->ird, 1788 cur_max_read_depth(ep->com.dev)); 1789 ep->ord = ntohs(mpa_v2_params->ord) & 1790 MPA_V2_IRD_ORD_MASK; 1791 ep->ord = min_t(u32, ep->ord, 1792 cur_max_read_depth(ep->com.dev)); 1793 pr_debug("initiator ird %u ord %u\n", 1794 ep->ird, ep->ord); 1795 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL) 1796 if (peer2peer) { 1797 if (ntohs(mpa_v2_params->ord) & 1798 MPA_V2_RDMA_WRITE_RTR) 1799 ep->mpa_attr.p2p_type = 1800 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 1801 else if (ntohs(mpa_v2_params->ord) & 1802 MPA_V2_RDMA_READ_RTR) 1803 ep->mpa_attr.p2p_type = 1804 FW_RI_INIT_P2PTYPE_READ_REQ; 1805 } 1806 } 1807 } else if (mpa->revision == 1) 1808 if (peer2peer) 1809 ep->mpa_attr.p2p_type = p2p_type; 1810 1811 pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n", 1812 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 1813 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, 1814 ep->mpa_attr.p2p_type); 1815 1816 __state_set(&ep->com, MPA_REQ_RCVD); 1817 1818 /* drive upcall */ 1819 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING); 1820 if (ep->parent_ep->com.state != DEAD) { 1821 if (connect_request_upcall(ep)) 1822 goto err_unlock_parent; 1823 } else { 1824 goto err_unlock_parent; 1825 } 1826 mutex_unlock(&ep->parent_ep->com.mutex); 1827 return 0; 1828 1829 err_unlock_parent: 1830 mutex_unlock(&ep->parent_ep->com.mutex); 1831 goto err_out; 1832 err_stop_timer: 1833 (void)stop_ep_timer(ep); 1834 err_out: 1835 return 2; 1836 } 1837 1838 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb) 1839 { 1840 struct c4iw_ep *ep; 1841 struct cpl_rx_data *hdr = cplhdr(skb); 1842 unsigned int dlen = ntohs(hdr->len); 1843 unsigned int tid = GET_TID(hdr); 1844 __u8 status = hdr->status; 1845 int disconnect = 0; 1846 1847 ep = get_ep_from_tid(dev, tid); 1848 if (!ep) 1849 return 0; 1850 pr_debug("ep %p tid %u dlen %u\n", ep, ep->hwtid, dlen); 1851 skb_pull(skb, sizeof(*hdr)); 1852 skb_trim(skb, dlen); 1853 mutex_lock(&ep->com.mutex); 1854 1855 switch (ep->com.state) { 1856 case MPA_REQ_SENT: 1857 update_rx_credits(ep, dlen); 1858 ep->rcv_seq += dlen; 1859 disconnect = process_mpa_reply(ep, skb); 1860 break; 1861 case MPA_REQ_WAIT: 1862 update_rx_credits(ep, dlen); 1863 ep->rcv_seq += dlen; 1864 disconnect = process_mpa_request(ep, skb); 1865 break; 1866 case FPDU_MODE: { 1867 struct c4iw_qp_attributes attrs; 1868 1869 update_rx_credits(ep, dlen); 1870 if (status) 1871 pr_err("%s Unexpected streaming data." \ 1872 " qpid %u ep %p state %d tid %u status %d\n", 1873 __func__, ep->com.qp->wq.sq.qid, ep, 1874 ep->com.state, ep->hwtid, status); 1875 attrs.next_state = C4IW_QP_STATE_TERMINATE; 1876 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 1877 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 1878 disconnect = 1; 1879 break; 1880 } 1881 default: 1882 break; 1883 } 1884 mutex_unlock(&ep->com.mutex); 1885 if (disconnect) 1886 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL); 1887 c4iw_put_ep(&ep->com); 1888 return 0; 1889 } 1890 1891 static void complete_cached_srq_buffers(struct c4iw_ep *ep, u32 srqidx) 1892 { 1893 enum chip_type adapter_type; 1894 1895 adapter_type = ep->com.dev->rdev.lldi.adapter_type; 1896 1897 /* 1898 * If this TCB had a srq buffer cached, then we must complete 1899 * it. For user mode, that means saving the srqidx in the 1900 * user/kernel status page for this qp. For kernel mode, just 1901 * synthesize the CQE now. 1902 */ 1903 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T5 && srqidx) { 1904 if (ep->com.qp->ibqp.uobject) 1905 t4_set_wq_in_error(&ep->com.qp->wq, srqidx); 1906 else 1907 c4iw_flush_srqidx(ep->com.qp, srqidx); 1908 } 1909 } 1910 1911 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 1912 { 1913 u32 srqidx; 1914 struct c4iw_ep *ep; 1915 struct cpl_abort_rpl_rss6 *rpl = cplhdr(skb); 1916 int release = 0; 1917 unsigned int tid = GET_TID(rpl); 1918 1919 ep = get_ep_from_tid(dev, tid); 1920 if (!ep) { 1921 pr_warn("Abort rpl to freed endpoint\n"); 1922 return 0; 1923 } 1924 1925 if (ep->com.qp && ep->com.qp->srq) { 1926 srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(rpl->srqidx_status)); 1927 complete_cached_srq_buffers(ep, srqidx ? srqidx : ep->srqe_idx); 1928 } 1929 1930 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 1931 mutex_lock(&ep->com.mutex); 1932 switch (ep->com.state) { 1933 case ABORTING: 1934 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET); 1935 __state_set(&ep->com, DEAD); 1936 release = 1; 1937 break; 1938 default: 1939 pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state); 1940 break; 1941 } 1942 mutex_unlock(&ep->com.mutex); 1943 1944 if (release) { 1945 close_complete_upcall(ep, -ECONNRESET); 1946 release_ep_resources(ep); 1947 } 1948 c4iw_put_ep(&ep->com); 1949 return 0; 1950 } 1951 1952 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid) 1953 { 1954 struct sk_buff *skb; 1955 struct fw_ofld_connection_wr *req; 1956 unsigned int mtu_idx; 1957 u32 wscale; 1958 struct sockaddr_in *sin; 1959 int win; 1960 1961 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL); 1962 req = __skb_put_zero(skb, sizeof(*req)); 1963 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR)); 1964 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16))); 1965 req->le.filter = cpu_to_be32(cxgb4_select_ntuple( 1966 ep->com.dev->rdev.lldi.ports[0], 1967 ep->l2t)); 1968 sin = (struct sockaddr_in *)&ep->com.local_addr; 1969 req->le.lport = sin->sin_port; 1970 req->le.u.ipv4.lip = sin->sin_addr.s_addr; 1971 sin = (struct sockaddr_in *)&ep->com.remote_addr; 1972 req->le.pport = sin->sin_port; 1973 req->le.u.ipv4.pip = sin->sin_addr.s_addr; 1974 req->tcb.t_state_to_astid = 1975 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) | 1976 FW_OFLD_CONNECTION_WR_ASTID_V(atid)); 1977 req->tcb.cplrxdataack_cplpassacceptrpl = 1978 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F); 1979 req->tcb.tx_max = (__force __be32) jiffies; 1980 req->tcb.rcv_adv = htons(1); 1981 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx, 1982 enable_tcp_timestamps, 1983 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1); 1984 wscale = cxgb_compute_wscale(rcv_win); 1985 1986 /* 1987 * Specify the largest window that will fit in opt0. The 1988 * remainder will be specified in the rx_data_ack. 1989 */ 1990 win = ep->rcv_win >> 10; 1991 if (win > RCV_BUFSIZ_M) 1992 win = RCV_BUFSIZ_M; 1993 1994 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F | 1995 (nocong ? NO_CONG_F : 0) | 1996 KEEP_ALIVE_F | 1997 DELACK_F | 1998 WND_SCALE_V(wscale) | 1999 MSS_IDX_V(mtu_idx) | 2000 L2T_IDX_V(ep->l2t->idx) | 2001 TX_CHAN_V(ep->tx_chan) | 2002 SMAC_SEL_V(ep->smac_idx) | 2003 DSCP_V(ep->tos >> 2) | 2004 ULP_MODE_V(ULP_MODE_TCPDDP) | 2005 RCV_BUFSIZ_V(win)); 2006 req->tcb.opt2 = (__force __be32) (PACE_V(1) | 2007 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) | 2008 RX_CHANNEL_V(0) | 2009 CCTRL_ECN_V(enable_ecn) | 2010 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid)); 2011 if (enable_tcp_timestamps) 2012 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F; 2013 if (enable_tcp_sack) 2014 req->tcb.opt2 |= (__force __be32)SACK_EN_F; 2015 if (wscale && enable_tcp_window_scaling) 2016 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F; 2017 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0); 2018 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2); 2019 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx); 2020 set_bit(ACT_OFLD_CONN, &ep->com.history); 2021 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 2022 } 2023 2024 /* 2025 * Some of the error codes above implicitly indicate that there is no TID 2026 * allocated with the result of an ACT_OPEN. We use this predicate to make 2027 * that explicit. 2028 */ 2029 static inline int act_open_has_tid(int status) 2030 { 2031 return (status != CPL_ERR_TCAM_PARITY && 2032 status != CPL_ERR_TCAM_MISS && 2033 status != CPL_ERR_TCAM_FULL && 2034 status != CPL_ERR_CONN_EXIST_SYNRECV && 2035 status != CPL_ERR_CONN_EXIST); 2036 } 2037 2038 static char *neg_adv_str(unsigned int status) 2039 { 2040 switch (status) { 2041 case CPL_ERR_RTX_NEG_ADVICE: 2042 return "Retransmit timeout"; 2043 case CPL_ERR_PERSIST_NEG_ADVICE: 2044 return "Persist timeout"; 2045 case CPL_ERR_KEEPALV_NEG_ADVICE: 2046 return "Keepalive timeout"; 2047 default: 2048 return "Unknown"; 2049 } 2050 } 2051 2052 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi) 2053 { 2054 ep->snd_win = snd_win; 2055 ep->rcv_win = rcv_win; 2056 pr_debug("snd_win %d rcv_win %d\n", 2057 ep->snd_win, ep->rcv_win); 2058 } 2059 2060 #define ACT_OPEN_RETRY_COUNT 2 2061 2062 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip, 2063 struct dst_entry *dst, struct c4iw_dev *cdev, 2064 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos) 2065 { 2066 struct neighbour *n; 2067 int err, step; 2068 struct net_device *pdev; 2069 2070 n = dst_neigh_lookup(dst, peer_ip); 2071 if (!n) 2072 return -ENODEV; 2073 2074 rcu_read_lock(); 2075 err = -ENOMEM; 2076 if (n->dev->flags & IFF_LOOPBACK) { 2077 if (iptype == 4) 2078 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip); 2079 else if (IS_ENABLED(CONFIG_IPV6)) 2080 for_each_netdev(&init_net, pdev) { 2081 if (ipv6_chk_addr(&init_net, 2082 (struct in6_addr *)peer_ip, 2083 pdev, 1)) 2084 break; 2085 } 2086 else 2087 pdev = NULL; 2088 2089 if (!pdev) { 2090 err = -ENODEV; 2091 goto out; 2092 } 2093 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t, 2094 n, pdev, rt_tos2priority(tos)); 2095 if (!ep->l2t) { 2096 dev_put(pdev); 2097 goto out; 2098 } 2099 ep->mtu = pdev->mtu; 2100 ep->tx_chan = cxgb4_port_chan(pdev); 2101 ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx; 2102 step = cdev->rdev.lldi.ntxq / 2103 cdev->rdev.lldi.nchan; 2104 ep->txq_idx = cxgb4_port_idx(pdev) * step; 2105 step = cdev->rdev.lldi.nrxq / 2106 cdev->rdev.lldi.nchan; 2107 ep->ctrlq_idx = cxgb4_port_idx(pdev); 2108 ep->rss_qid = cdev->rdev.lldi.rxq_ids[ 2109 cxgb4_port_idx(pdev) * step]; 2110 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev)); 2111 dev_put(pdev); 2112 } else { 2113 pdev = get_real_dev(n->dev); 2114 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t, 2115 n, pdev, rt_tos2priority(tos)); 2116 if (!ep->l2t) 2117 goto out; 2118 ep->mtu = dst_mtu(dst); 2119 ep->tx_chan = cxgb4_port_chan(pdev); 2120 ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx; 2121 step = cdev->rdev.lldi.ntxq / 2122 cdev->rdev.lldi.nchan; 2123 ep->txq_idx = cxgb4_port_idx(pdev) * step; 2124 ep->ctrlq_idx = cxgb4_port_idx(pdev); 2125 step = cdev->rdev.lldi.nrxq / 2126 cdev->rdev.lldi.nchan; 2127 ep->rss_qid = cdev->rdev.lldi.rxq_ids[ 2128 cxgb4_port_idx(pdev) * step]; 2129 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev)); 2130 2131 if (clear_mpa_v1) { 2132 ep->retry_with_mpa_v1 = 0; 2133 ep->tried_with_mpa_v1 = 0; 2134 } 2135 } 2136 err = 0; 2137 out: 2138 rcu_read_unlock(); 2139 2140 neigh_release(n); 2141 2142 return err; 2143 } 2144 2145 static int c4iw_reconnect(struct c4iw_ep *ep) 2146 { 2147 int err = 0; 2148 int size = 0; 2149 struct sockaddr_in *laddr = (struct sockaddr_in *) 2150 &ep->com.cm_id->m_local_addr; 2151 struct sockaddr_in *raddr = (struct sockaddr_in *) 2152 &ep->com.cm_id->m_remote_addr; 2153 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *) 2154 &ep->com.cm_id->m_local_addr; 2155 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *) 2156 &ep->com.cm_id->m_remote_addr; 2157 int iptype; 2158 __u8 *ra; 2159 2160 pr_debug("qp %p cm_id %p\n", ep->com.qp, ep->com.cm_id); 2161 c4iw_init_wr_wait(ep->com.wr_waitp); 2162 2163 /* When MPA revision is different on nodes, the node with MPA_rev=2 2164 * tries to reconnect with MPA_rev 1 for the same EP through 2165 * c4iw_reconnect(), where the same EP is assigned with new tid for 2166 * further connection establishment. As we are using the same EP pointer 2167 * for reconnect, few skbs are used during the previous c4iw_connect(), 2168 * which leaves the EP with inadequate skbs for further 2169 * c4iw_reconnect(), Further causing a crash due to an empty 2170 * skb_list() during peer_abort(). Allocate skbs which is already used. 2171 */ 2172 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list)); 2173 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) { 2174 err = -ENOMEM; 2175 goto fail1; 2176 } 2177 2178 /* 2179 * Allocate an active TID to initiate a TCP connection. 2180 */ 2181 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep); 2182 if (ep->atid == -1) { 2183 pr_err("%s - cannot alloc atid\n", __func__); 2184 err = -ENOMEM; 2185 goto fail2; 2186 } 2187 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid); 2188 2189 /* find a route */ 2190 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) { 2191 ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev, 2192 laddr->sin_addr.s_addr, 2193 raddr->sin_addr.s_addr, 2194 laddr->sin_port, 2195 raddr->sin_port, ep->com.cm_id->tos); 2196 iptype = 4; 2197 ra = (__u8 *)&raddr->sin_addr; 2198 } else { 2199 ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi, 2200 get_real_dev, 2201 laddr6->sin6_addr.s6_addr, 2202 raddr6->sin6_addr.s6_addr, 2203 laddr6->sin6_port, 2204 raddr6->sin6_port, 2205 ep->com.cm_id->tos, 2206 raddr6->sin6_scope_id); 2207 iptype = 6; 2208 ra = (__u8 *)&raddr6->sin6_addr; 2209 } 2210 if (!ep->dst) { 2211 pr_err("%s - cannot find route\n", __func__); 2212 err = -EHOSTUNREACH; 2213 goto fail3; 2214 } 2215 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false, 2216 ep->com.dev->rdev.lldi.adapter_type, 2217 ep->com.cm_id->tos); 2218 if (err) { 2219 pr_err("%s - cannot alloc l2e\n", __func__); 2220 goto fail4; 2221 } 2222 2223 pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n", 2224 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid, 2225 ep->l2t->idx); 2226 2227 state_set(&ep->com, CONNECTING); 2228 ep->tos = ep->com.cm_id->tos; 2229 2230 /* send connect request to rnic */ 2231 err = send_connect(ep); 2232 if (!err) 2233 goto out; 2234 2235 cxgb4_l2t_release(ep->l2t); 2236 fail4: 2237 dst_release(ep->dst); 2238 fail3: 2239 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid); 2240 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid); 2241 fail2: 2242 /* 2243 * remember to send notification to upper layer. 2244 * We are in here so the upper layer is not aware that this is 2245 * re-connect attempt and so, upper layer is still waiting for 2246 * response of 1st connect request. 2247 */ 2248 connect_reply_upcall(ep, -ECONNRESET); 2249 fail1: 2250 c4iw_put_ep(&ep->com); 2251 out: 2252 return err; 2253 } 2254 2255 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2256 { 2257 struct c4iw_ep *ep; 2258 struct cpl_act_open_rpl *rpl = cplhdr(skb); 2259 unsigned int atid = TID_TID_G(AOPEN_ATID_G( 2260 ntohl(rpl->atid_status))); 2261 struct tid_info *t = dev->rdev.lldi.tids; 2262 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status)); 2263 struct sockaddr_in *la; 2264 struct sockaddr_in *ra; 2265 struct sockaddr_in6 *la6; 2266 struct sockaddr_in6 *ra6; 2267 int ret = 0; 2268 2269 ep = lookup_atid(t, atid); 2270 la = (struct sockaddr_in *)&ep->com.local_addr; 2271 ra = (struct sockaddr_in *)&ep->com.remote_addr; 2272 la6 = (struct sockaddr_in6 *)&ep->com.local_addr; 2273 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr; 2274 2275 pr_debug("ep %p atid %u status %u errno %d\n", ep, atid, 2276 status, status2errno(status)); 2277 2278 if (cxgb_is_neg_adv(status)) { 2279 pr_debug("Connection problems for atid %u status %u (%s)\n", 2280 atid, status, neg_adv_str(status)); 2281 ep->stats.connect_neg_adv++; 2282 mutex_lock(&dev->rdev.stats.lock); 2283 dev->rdev.stats.neg_adv++; 2284 mutex_unlock(&dev->rdev.stats.lock); 2285 return 0; 2286 } 2287 2288 set_bit(ACT_OPEN_RPL, &ep->com.history); 2289 2290 /* 2291 * Log interesting failures. 2292 */ 2293 switch (status) { 2294 case CPL_ERR_CONN_RESET: 2295 case CPL_ERR_CONN_TIMEDOUT: 2296 break; 2297 case CPL_ERR_TCAM_FULL: 2298 mutex_lock(&dev->rdev.stats.lock); 2299 dev->rdev.stats.tcam_full++; 2300 mutex_unlock(&dev->rdev.stats.lock); 2301 if (ep->com.local_addr.ss_family == AF_INET && 2302 dev->rdev.lldi.enable_fw_ofld_conn) { 2303 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G( 2304 ntohl(rpl->atid_status)))); 2305 if (ret) 2306 goto fail; 2307 return 0; 2308 } 2309 break; 2310 case CPL_ERR_CONN_EXIST: 2311 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) { 2312 set_bit(ACT_RETRY_INUSE, &ep->com.history); 2313 if (ep->com.remote_addr.ss_family == AF_INET6) { 2314 struct sockaddr_in6 *sin6 = 2315 (struct sockaddr_in6 *) 2316 &ep->com.local_addr; 2317 cxgb4_clip_release( 2318 ep->com.dev->rdev.lldi.ports[0], 2319 (const u32 *) 2320 &sin6->sin6_addr.s6_addr, 1); 2321 } 2322 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, 2323 atid); 2324 cxgb4_free_atid(t, atid); 2325 dst_release(ep->dst); 2326 cxgb4_l2t_release(ep->l2t); 2327 c4iw_reconnect(ep); 2328 return 0; 2329 } 2330 break; 2331 default: 2332 if (ep->com.local_addr.ss_family == AF_INET) { 2333 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n", 2334 atid, status, status2errno(status), 2335 &la->sin_addr.s_addr, ntohs(la->sin_port), 2336 &ra->sin_addr.s_addr, ntohs(ra->sin_port)); 2337 } else { 2338 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n", 2339 atid, status, status2errno(status), 2340 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port), 2341 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port)); 2342 } 2343 break; 2344 } 2345 2346 fail: 2347 connect_reply_upcall(ep, status2errno(status)); 2348 state_set(&ep->com, DEAD); 2349 2350 if (ep->com.remote_addr.ss_family == AF_INET6) { 2351 struct sockaddr_in6 *sin6 = 2352 (struct sockaddr_in6 *)&ep->com.local_addr; 2353 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 2354 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 2355 } 2356 if (status && act_open_has_tid(status)) 2357 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl), 2358 ep->com.local_addr.ss_family); 2359 2360 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid); 2361 cxgb4_free_atid(t, atid); 2362 dst_release(ep->dst); 2363 cxgb4_l2t_release(ep->l2t); 2364 c4iw_put_ep(&ep->com); 2365 2366 return 0; 2367 } 2368 2369 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2370 { 2371 struct cpl_pass_open_rpl *rpl = cplhdr(skb); 2372 unsigned int stid = GET_TID(rpl); 2373 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid); 2374 2375 if (!ep) { 2376 pr_warn("%s stid %d lookup failure!\n", __func__, stid); 2377 goto out; 2378 } 2379 pr_debug("ep %p status %d error %d\n", ep, 2380 rpl->status, status2errno(rpl->status)); 2381 c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status)); 2382 c4iw_put_ep(&ep->com); 2383 out: 2384 return 0; 2385 } 2386 2387 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2388 { 2389 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb); 2390 unsigned int stid = GET_TID(rpl); 2391 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid); 2392 2393 if (!ep) { 2394 pr_warn("%s stid %d lookup failure!\n", __func__, stid); 2395 goto out; 2396 } 2397 pr_debug("ep %p\n", ep); 2398 c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status)); 2399 c4iw_put_ep(&ep->com); 2400 out: 2401 return 0; 2402 } 2403 2404 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb, 2405 struct cpl_pass_accept_req *req) 2406 { 2407 struct cpl_pass_accept_rpl *rpl; 2408 unsigned int mtu_idx; 2409 u64 opt0; 2410 u32 opt2; 2411 u32 wscale; 2412 struct cpl_t5_pass_accept_rpl *rpl5 = NULL; 2413 int win; 2414 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type; 2415 2416 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 2417 2418 skb_get(skb); 2419 rpl = cplhdr(skb); 2420 if (!is_t4(adapter_type)) { 2421 skb_trim(skb, roundup(sizeof(*rpl5), 16)); 2422 rpl5 = (void *)rpl; 2423 INIT_TP_WR(rpl5, ep->hwtid); 2424 } else { 2425 skb_trim(skb, sizeof(*rpl)); 2426 INIT_TP_WR(rpl, ep->hwtid); 2427 } 2428 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, 2429 ep->hwtid)); 2430 2431 cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx, 2432 enable_tcp_timestamps && req->tcpopt.tstamp, 2433 (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1); 2434 wscale = cxgb_compute_wscale(rcv_win); 2435 2436 /* 2437 * Specify the largest window that will fit in opt0. The 2438 * remainder will be specified in the rx_data_ack. 2439 */ 2440 win = ep->rcv_win >> 10; 2441 if (win > RCV_BUFSIZ_M) 2442 win = RCV_BUFSIZ_M; 2443 opt0 = (nocong ? NO_CONG_F : 0) | 2444 KEEP_ALIVE_F | 2445 DELACK_F | 2446 WND_SCALE_V(wscale) | 2447 MSS_IDX_V(mtu_idx) | 2448 L2T_IDX_V(ep->l2t->idx) | 2449 TX_CHAN_V(ep->tx_chan) | 2450 SMAC_SEL_V(ep->smac_idx) | 2451 DSCP_V(ep->tos >> 2) | 2452 ULP_MODE_V(ULP_MODE_TCPDDP) | 2453 RCV_BUFSIZ_V(win); 2454 opt2 = RX_CHANNEL_V(0) | 2455 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid); 2456 2457 if (enable_tcp_timestamps && req->tcpopt.tstamp) 2458 opt2 |= TSTAMPS_EN_F; 2459 if (enable_tcp_sack && req->tcpopt.sack) 2460 opt2 |= SACK_EN_F; 2461 if (wscale && enable_tcp_window_scaling) 2462 opt2 |= WND_SCALE_EN_F; 2463 if (enable_ecn) { 2464 const struct tcphdr *tcph; 2465 u32 hlen = ntohl(req->hdr_len); 2466 2467 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5) 2468 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) + 2469 IP_HDR_LEN_G(hlen); 2470 else 2471 tcph = (const void *)(req + 1) + 2472 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen); 2473 if (tcph->ece && tcph->cwr) 2474 opt2 |= CCTRL_ECN_V(1); 2475 } 2476 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) { 2477 u32 isn = (prandom_u32() & ~7UL) - 1; 2478 opt2 |= T5_OPT_2_VALID_F; 2479 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE); 2480 opt2 |= T5_ISS_F; 2481 rpl5 = (void *)rpl; 2482 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16)); 2483 if (peer2peer) 2484 isn += 4; 2485 rpl5->iss = cpu_to_be32(isn); 2486 pr_debug("iss %u\n", be32_to_cpu(rpl5->iss)); 2487 } 2488 2489 rpl->opt0 = cpu_to_be64(opt0); 2490 rpl->opt2 = cpu_to_be32(opt2); 2491 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx); 2492 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure); 2493 2494 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 2495 } 2496 2497 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb) 2498 { 2499 pr_debug("c4iw_dev %p tid %u\n", dev, hwtid); 2500 skb_trim(skb, sizeof(struct cpl_tid_release)); 2501 release_tid(&dev->rdev, hwtid, skb); 2502 return; 2503 } 2504 2505 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb) 2506 { 2507 struct c4iw_ep *child_ep = NULL, *parent_ep; 2508 struct cpl_pass_accept_req *req = cplhdr(skb); 2509 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid)); 2510 struct tid_info *t = dev->rdev.lldi.tids; 2511 unsigned int hwtid = GET_TID(req); 2512 struct dst_entry *dst; 2513 __u8 local_ip[16], peer_ip[16]; 2514 __be16 local_port, peer_port; 2515 struct sockaddr_in6 *sin6; 2516 int err; 2517 u16 peer_mss = ntohs(req->tcpopt.mss); 2518 int iptype; 2519 unsigned short hdrs; 2520 u8 tos; 2521 2522 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid); 2523 if (!parent_ep) { 2524 pr_err("%s connect request on invalid stid %d\n", 2525 __func__, stid); 2526 goto reject; 2527 } 2528 2529 if (state_read(&parent_ep->com) != LISTEN) { 2530 pr_err("%s - listening ep not in LISTEN\n", __func__); 2531 goto reject; 2532 } 2533 2534 if (parent_ep->com.cm_id->tos_set) 2535 tos = parent_ep->com.cm_id->tos; 2536 else 2537 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid)); 2538 2539 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type, 2540 &iptype, local_ip, peer_ip, &local_port, &peer_port); 2541 2542 /* Find output route */ 2543 if (iptype == 4) { 2544 pr_debug("parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n" 2545 , parent_ep, hwtid, 2546 local_ip, peer_ip, ntohs(local_port), 2547 ntohs(peer_port), peer_mss); 2548 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev, 2549 *(__be32 *)local_ip, *(__be32 *)peer_ip, 2550 local_port, peer_port, tos); 2551 } else { 2552 pr_debug("parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n" 2553 , parent_ep, hwtid, 2554 local_ip, peer_ip, ntohs(local_port), 2555 ntohs(peer_port), peer_mss); 2556 dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev, 2557 local_ip, peer_ip, local_port, peer_port, 2558 tos, 2559 ((struct sockaddr_in6 *) 2560 &parent_ep->com.local_addr)->sin6_scope_id); 2561 } 2562 if (!dst) { 2563 pr_err("%s - failed to find dst entry!\n", __func__); 2564 goto reject; 2565 } 2566 2567 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL); 2568 if (!child_ep) { 2569 pr_err("%s - failed to allocate ep entry!\n", __func__); 2570 dst_release(dst); 2571 goto reject; 2572 } 2573 2574 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false, 2575 parent_ep->com.dev->rdev.lldi.adapter_type, tos); 2576 if (err) { 2577 pr_err("%s - failed to allocate l2t entry!\n", __func__); 2578 dst_release(dst); 2579 kfree(child_ep); 2580 goto reject; 2581 } 2582 2583 hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) + 2584 sizeof(struct tcphdr) + 2585 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0); 2586 if (peer_mss && child_ep->mtu > (peer_mss + hdrs)) 2587 child_ep->mtu = peer_mss + hdrs; 2588 2589 skb_queue_head_init(&child_ep->com.ep_skb_list); 2590 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF)) 2591 goto fail; 2592 2593 state_set(&child_ep->com, CONNECTING); 2594 child_ep->com.dev = dev; 2595 child_ep->com.cm_id = NULL; 2596 2597 if (iptype == 4) { 2598 struct sockaddr_in *sin = (struct sockaddr_in *) 2599 &child_ep->com.local_addr; 2600 2601 sin->sin_family = AF_INET; 2602 sin->sin_port = local_port; 2603 sin->sin_addr.s_addr = *(__be32 *)local_ip; 2604 2605 sin = (struct sockaddr_in *)&child_ep->com.local_addr; 2606 sin->sin_family = AF_INET; 2607 sin->sin_port = ((struct sockaddr_in *) 2608 &parent_ep->com.local_addr)->sin_port; 2609 sin->sin_addr.s_addr = *(__be32 *)local_ip; 2610 2611 sin = (struct sockaddr_in *)&child_ep->com.remote_addr; 2612 sin->sin_family = AF_INET; 2613 sin->sin_port = peer_port; 2614 sin->sin_addr.s_addr = *(__be32 *)peer_ip; 2615 } else { 2616 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr; 2617 sin6->sin6_family = PF_INET6; 2618 sin6->sin6_port = local_port; 2619 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16); 2620 2621 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr; 2622 sin6->sin6_family = PF_INET6; 2623 sin6->sin6_port = ((struct sockaddr_in6 *) 2624 &parent_ep->com.local_addr)->sin6_port; 2625 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16); 2626 2627 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr; 2628 sin6->sin6_family = PF_INET6; 2629 sin6->sin6_port = peer_port; 2630 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16); 2631 } 2632 2633 c4iw_get_ep(&parent_ep->com); 2634 child_ep->parent_ep = parent_ep; 2635 child_ep->tos = tos; 2636 child_ep->dst = dst; 2637 child_ep->hwtid = hwtid; 2638 2639 pr_debug("tx_chan %u smac_idx %u rss_qid %u\n", 2640 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid); 2641 2642 timer_setup(&child_ep->timer, ep_timeout, 0); 2643 cxgb4_insert_tid(t, child_ep, hwtid, 2644 child_ep->com.local_addr.ss_family); 2645 insert_ep_tid(child_ep); 2646 if (accept_cr(child_ep, skb, req)) { 2647 c4iw_put_ep(&parent_ep->com); 2648 release_ep_resources(child_ep); 2649 } else { 2650 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history); 2651 } 2652 if (iptype == 6) { 2653 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr; 2654 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0], 2655 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 2656 } 2657 goto out; 2658 fail: 2659 c4iw_put_ep(&child_ep->com); 2660 reject: 2661 reject_cr(dev, hwtid, skb); 2662 out: 2663 if (parent_ep) 2664 c4iw_put_ep(&parent_ep->com); 2665 return 0; 2666 } 2667 2668 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb) 2669 { 2670 struct c4iw_ep *ep; 2671 struct cpl_pass_establish *req = cplhdr(skb); 2672 unsigned int tid = GET_TID(req); 2673 int ret; 2674 u16 tcp_opt = ntohs(req->tcp_opt); 2675 2676 ep = get_ep_from_tid(dev, tid); 2677 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 2678 ep->snd_seq = be32_to_cpu(req->snd_isn); 2679 ep->rcv_seq = be32_to_cpu(req->rcv_isn); 2680 ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt); 2681 2682 pr_debug("ep %p hwtid %u tcp_opt 0x%02x\n", ep, tid, tcp_opt); 2683 2684 set_emss(ep, tcp_opt); 2685 2686 dst_confirm(ep->dst); 2687 mutex_lock(&ep->com.mutex); 2688 ep->com.state = MPA_REQ_WAIT; 2689 start_ep_timer(ep); 2690 set_bit(PASS_ESTAB, &ep->com.history); 2691 ret = send_flowc(ep); 2692 mutex_unlock(&ep->com.mutex); 2693 if (ret) 2694 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 2695 c4iw_put_ep(&ep->com); 2696 2697 return 0; 2698 } 2699 2700 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb) 2701 { 2702 struct cpl_peer_close *hdr = cplhdr(skb); 2703 struct c4iw_ep *ep; 2704 struct c4iw_qp_attributes attrs; 2705 int disconnect = 1; 2706 int release = 0; 2707 unsigned int tid = GET_TID(hdr); 2708 int ret; 2709 2710 ep = get_ep_from_tid(dev, tid); 2711 if (!ep) 2712 return 0; 2713 2714 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 2715 dst_confirm(ep->dst); 2716 2717 set_bit(PEER_CLOSE, &ep->com.history); 2718 mutex_lock(&ep->com.mutex); 2719 switch (ep->com.state) { 2720 case MPA_REQ_WAIT: 2721 __state_set(&ep->com, CLOSING); 2722 break; 2723 case MPA_REQ_SENT: 2724 __state_set(&ep->com, CLOSING); 2725 connect_reply_upcall(ep, -ECONNRESET); 2726 break; 2727 case MPA_REQ_RCVD: 2728 2729 /* 2730 * We're gonna mark this puppy DEAD, but keep 2731 * the reference on it until the ULP accepts or 2732 * rejects the CR. Also wake up anyone waiting 2733 * in rdma connection migration (see c4iw_accept_cr()). 2734 */ 2735 __state_set(&ep->com, CLOSING); 2736 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid); 2737 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET); 2738 break; 2739 case MPA_REP_SENT: 2740 __state_set(&ep->com, CLOSING); 2741 pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid); 2742 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET); 2743 break; 2744 case FPDU_MODE: 2745 start_ep_timer(ep); 2746 __state_set(&ep->com, CLOSING); 2747 attrs.next_state = C4IW_QP_STATE_CLOSING; 2748 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2749 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2750 if (ret != -ECONNRESET) { 2751 peer_close_upcall(ep); 2752 disconnect = 1; 2753 } 2754 break; 2755 case ABORTING: 2756 disconnect = 0; 2757 break; 2758 case CLOSING: 2759 __state_set(&ep->com, MORIBUND); 2760 disconnect = 0; 2761 break; 2762 case MORIBUND: 2763 (void)stop_ep_timer(ep); 2764 if (ep->com.cm_id && ep->com.qp) { 2765 attrs.next_state = C4IW_QP_STATE_IDLE; 2766 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2767 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2768 } 2769 close_complete_upcall(ep, 0); 2770 __state_set(&ep->com, DEAD); 2771 release = 1; 2772 disconnect = 0; 2773 break; 2774 case DEAD: 2775 disconnect = 0; 2776 break; 2777 default: 2778 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state); 2779 } 2780 mutex_unlock(&ep->com.mutex); 2781 if (disconnect) 2782 c4iw_ep_disconnect(ep, 0, GFP_KERNEL); 2783 if (release) 2784 release_ep_resources(ep); 2785 c4iw_put_ep(&ep->com); 2786 return 0; 2787 } 2788 2789 static void finish_peer_abort(struct c4iw_dev *dev, struct c4iw_ep *ep) 2790 { 2791 complete_cached_srq_buffers(ep, ep->srqe_idx); 2792 if (ep->com.cm_id && ep->com.qp) { 2793 struct c4iw_qp_attributes attrs; 2794 2795 attrs.next_state = C4IW_QP_STATE_ERROR; 2796 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2797 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2798 } 2799 peer_abort_upcall(ep); 2800 release_ep_resources(ep); 2801 c4iw_put_ep(&ep->com); 2802 } 2803 2804 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb) 2805 { 2806 struct cpl_abort_req_rss6 *req = cplhdr(skb); 2807 struct c4iw_ep *ep; 2808 struct sk_buff *rpl_skb; 2809 struct c4iw_qp_attributes attrs; 2810 int ret; 2811 int release = 0; 2812 unsigned int tid = GET_TID(req); 2813 u8 status; 2814 u32 srqidx; 2815 2816 u32 len = roundup(sizeof(struct cpl_abort_rpl), 16); 2817 2818 ep = get_ep_from_tid(dev, tid); 2819 if (!ep) 2820 return 0; 2821 2822 status = ABORT_RSS_STATUS_G(be32_to_cpu(req->srqidx_status)); 2823 2824 if (cxgb_is_neg_adv(status)) { 2825 pr_debug("Negative advice on abort- tid %u status %d (%s)\n", 2826 ep->hwtid, status, neg_adv_str(status)); 2827 ep->stats.abort_neg_adv++; 2828 mutex_lock(&dev->rdev.stats.lock); 2829 dev->rdev.stats.neg_adv++; 2830 mutex_unlock(&dev->rdev.stats.lock); 2831 goto deref_ep; 2832 } 2833 2834 pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid, 2835 ep->com.state); 2836 set_bit(PEER_ABORT, &ep->com.history); 2837 2838 /* 2839 * Wake up any threads in rdma_init() or rdma_fini(). 2840 * However, this is not needed if com state is just 2841 * MPA_REQ_SENT 2842 */ 2843 if (ep->com.state != MPA_REQ_SENT) 2844 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET); 2845 2846 mutex_lock(&ep->com.mutex); 2847 switch (ep->com.state) { 2848 case CONNECTING: 2849 c4iw_put_ep(&ep->parent_ep->com); 2850 break; 2851 case MPA_REQ_WAIT: 2852 (void)stop_ep_timer(ep); 2853 break; 2854 case MPA_REQ_SENT: 2855 (void)stop_ep_timer(ep); 2856 if (status != CPL_ERR_CONN_RESET || mpa_rev == 1 || 2857 (mpa_rev == 2 && ep->tried_with_mpa_v1)) 2858 connect_reply_upcall(ep, -ECONNRESET); 2859 else { 2860 /* 2861 * we just don't send notification upwards because we 2862 * want to retry with mpa_v1 without upper layers even 2863 * knowing it. 2864 * 2865 * do some housekeeping so as to re-initiate the 2866 * connection 2867 */ 2868 pr_info("%s: mpa_rev=%d. Retrying with mpav1\n", 2869 __func__, mpa_rev); 2870 ep->retry_with_mpa_v1 = 1; 2871 } 2872 break; 2873 case MPA_REP_SENT: 2874 break; 2875 case MPA_REQ_RCVD: 2876 break; 2877 case MORIBUND: 2878 case CLOSING: 2879 stop_ep_timer(ep); 2880 /*FALLTHROUGH*/ 2881 case FPDU_MODE: 2882 if (ep->com.qp && ep->com.qp->srq) { 2883 srqidx = ABORT_RSS_SRQIDX_G( 2884 be32_to_cpu(req->srqidx_status)); 2885 if (srqidx) { 2886 complete_cached_srq_buffers(ep, 2887 req->srqidx_status); 2888 } else { 2889 /* Hold ep ref until finish_peer_abort() */ 2890 c4iw_get_ep(&ep->com); 2891 __state_set(&ep->com, ABORTING); 2892 set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags); 2893 read_tcb(ep); 2894 break; 2895 2896 } 2897 } 2898 2899 if (ep->com.cm_id && ep->com.qp) { 2900 attrs.next_state = C4IW_QP_STATE_ERROR; 2901 ret = c4iw_modify_qp(ep->com.qp->rhp, 2902 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, 2903 &attrs, 1); 2904 if (ret) 2905 pr_err("%s - qp <- error failed!\n", __func__); 2906 } 2907 peer_abort_upcall(ep); 2908 break; 2909 case ABORTING: 2910 break; 2911 case DEAD: 2912 pr_warn("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__); 2913 mutex_unlock(&ep->com.mutex); 2914 goto deref_ep; 2915 default: 2916 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state); 2917 break; 2918 } 2919 dst_confirm(ep->dst); 2920 if (ep->com.state != ABORTING) { 2921 __state_set(&ep->com, DEAD); 2922 /* we don't release if we want to retry with mpa_v1 */ 2923 if (!ep->retry_with_mpa_v1) 2924 release = 1; 2925 } 2926 mutex_unlock(&ep->com.mutex); 2927 2928 rpl_skb = skb_dequeue(&ep->com.ep_skb_list); 2929 if (WARN_ON(!rpl_skb)) { 2930 release = 1; 2931 goto out; 2932 } 2933 2934 cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx); 2935 2936 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb); 2937 out: 2938 if (release) 2939 release_ep_resources(ep); 2940 else if (ep->retry_with_mpa_v1) { 2941 if (ep->com.remote_addr.ss_family == AF_INET6) { 2942 struct sockaddr_in6 *sin6 = 2943 (struct sockaddr_in6 *) 2944 &ep->com.local_addr; 2945 cxgb4_clip_release( 2946 ep->com.dev->rdev.lldi.ports[0], 2947 (const u32 *)&sin6->sin6_addr.s6_addr, 2948 1); 2949 } 2950 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid); 2951 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid, 2952 ep->com.local_addr.ss_family); 2953 dst_release(ep->dst); 2954 cxgb4_l2t_release(ep->l2t); 2955 c4iw_reconnect(ep); 2956 } 2957 2958 deref_ep: 2959 c4iw_put_ep(&ep->com); 2960 /* Dereferencing ep, referenced in peer_abort_intr() */ 2961 c4iw_put_ep(&ep->com); 2962 return 0; 2963 } 2964 2965 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2966 { 2967 struct c4iw_ep *ep; 2968 struct c4iw_qp_attributes attrs; 2969 struct cpl_close_con_rpl *rpl = cplhdr(skb); 2970 int release = 0; 2971 unsigned int tid = GET_TID(rpl); 2972 2973 ep = get_ep_from_tid(dev, tid); 2974 if (!ep) 2975 return 0; 2976 2977 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 2978 2979 /* The cm_id may be null if we failed to connect */ 2980 mutex_lock(&ep->com.mutex); 2981 set_bit(CLOSE_CON_RPL, &ep->com.history); 2982 switch (ep->com.state) { 2983 case CLOSING: 2984 __state_set(&ep->com, MORIBUND); 2985 break; 2986 case MORIBUND: 2987 (void)stop_ep_timer(ep); 2988 if ((ep->com.cm_id) && (ep->com.qp)) { 2989 attrs.next_state = C4IW_QP_STATE_IDLE; 2990 c4iw_modify_qp(ep->com.qp->rhp, 2991 ep->com.qp, 2992 C4IW_QP_ATTR_NEXT_STATE, 2993 &attrs, 1); 2994 } 2995 close_complete_upcall(ep, 0); 2996 __state_set(&ep->com, DEAD); 2997 release = 1; 2998 break; 2999 case ABORTING: 3000 case DEAD: 3001 break; 3002 default: 3003 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state); 3004 break; 3005 } 3006 mutex_unlock(&ep->com.mutex); 3007 if (release) 3008 release_ep_resources(ep); 3009 c4iw_put_ep(&ep->com); 3010 return 0; 3011 } 3012 3013 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb) 3014 { 3015 struct cpl_rdma_terminate *rpl = cplhdr(skb); 3016 unsigned int tid = GET_TID(rpl); 3017 struct c4iw_ep *ep; 3018 struct c4iw_qp_attributes attrs; 3019 3020 ep = get_ep_from_tid(dev, tid); 3021 3022 if (ep) { 3023 if (ep->com.qp) { 3024 pr_warn("TERM received tid %u qpid %u\n", tid, 3025 ep->com.qp->wq.sq.qid); 3026 attrs.next_state = C4IW_QP_STATE_TERMINATE; 3027 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 3028 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 3029 } 3030 3031 c4iw_put_ep(&ep->com); 3032 } else 3033 pr_warn("TERM received tid %u no ep/qp\n", tid); 3034 3035 return 0; 3036 } 3037 3038 /* 3039 * Upcall from the adapter indicating data has been transmitted. 3040 * For us its just the single MPA request or reply. We can now free 3041 * the skb holding the mpa message. 3042 */ 3043 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb) 3044 { 3045 struct c4iw_ep *ep; 3046 struct cpl_fw4_ack *hdr = cplhdr(skb); 3047 u8 credits = hdr->credits; 3048 unsigned int tid = GET_TID(hdr); 3049 3050 3051 ep = get_ep_from_tid(dev, tid); 3052 if (!ep) 3053 return 0; 3054 pr_debug("ep %p tid %u credits %u\n", 3055 ep, ep->hwtid, credits); 3056 if (credits == 0) { 3057 pr_debug("0 credit ack ep %p tid %u state %u\n", 3058 ep, ep->hwtid, state_read(&ep->com)); 3059 goto out; 3060 } 3061 3062 dst_confirm(ep->dst); 3063 if (ep->mpa_skb) { 3064 pr_debug("last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n", 3065 ep, ep->hwtid, state_read(&ep->com), 3066 ep->mpa_attr.initiator ? 1 : 0); 3067 mutex_lock(&ep->com.mutex); 3068 kfree_skb(ep->mpa_skb); 3069 ep->mpa_skb = NULL; 3070 if (test_bit(STOP_MPA_TIMER, &ep->com.flags)) 3071 stop_ep_timer(ep); 3072 mutex_unlock(&ep->com.mutex); 3073 } 3074 out: 3075 c4iw_put_ep(&ep->com); 3076 return 0; 3077 } 3078 3079 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len) 3080 { 3081 int abort; 3082 struct c4iw_ep *ep = to_ep(cm_id); 3083 3084 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 3085 3086 mutex_lock(&ep->com.mutex); 3087 if (ep->com.state != MPA_REQ_RCVD) { 3088 mutex_unlock(&ep->com.mutex); 3089 c4iw_put_ep(&ep->com); 3090 return -ECONNRESET; 3091 } 3092 set_bit(ULP_REJECT, &ep->com.history); 3093 if (mpa_rev == 0) 3094 abort = 1; 3095 else 3096 abort = send_mpa_reject(ep, pdata, pdata_len); 3097 mutex_unlock(&ep->com.mutex); 3098 3099 stop_ep_timer(ep); 3100 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL); 3101 c4iw_put_ep(&ep->com); 3102 return 0; 3103 } 3104 3105 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 3106 { 3107 int err; 3108 struct c4iw_qp_attributes attrs; 3109 enum c4iw_qp_attr_mask mask; 3110 struct c4iw_ep *ep = to_ep(cm_id); 3111 struct c4iw_dev *h = to_c4iw_dev(cm_id->device); 3112 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn); 3113 int abort = 0; 3114 3115 pr_debug("ep %p tid %u\n", ep, ep->hwtid); 3116 3117 mutex_lock(&ep->com.mutex); 3118 if (ep->com.state != MPA_REQ_RCVD) { 3119 err = -ECONNRESET; 3120 goto err_out; 3121 } 3122 3123 if (!qp) { 3124 err = -EINVAL; 3125 goto err_out; 3126 } 3127 3128 set_bit(ULP_ACCEPT, &ep->com.history); 3129 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) || 3130 (conn_param->ird > cur_max_read_depth(ep->com.dev))) { 3131 err = -EINVAL; 3132 goto err_abort; 3133 } 3134 3135 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 3136 if (conn_param->ord > ep->ird) { 3137 if (RELAXED_IRD_NEGOTIATION) { 3138 conn_param->ord = ep->ird; 3139 } else { 3140 ep->ird = conn_param->ird; 3141 ep->ord = conn_param->ord; 3142 send_mpa_reject(ep, conn_param->private_data, 3143 conn_param->private_data_len); 3144 err = -ENOMEM; 3145 goto err_abort; 3146 } 3147 } 3148 if (conn_param->ird < ep->ord) { 3149 if (RELAXED_IRD_NEGOTIATION && 3150 ep->ord <= h->rdev.lldi.max_ordird_qp) { 3151 conn_param->ird = ep->ord; 3152 } else { 3153 err = -ENOMEM; 3154 goto err_abort; 3155 } 3156 } 3157 } 3158 ep->ird = conn_param->ird; 3159 ep->ord = conn_param->ord; 3160 3161 if (ep->mpa_attr.version == 1) { 3162 if (peer2peer && ep->ird == 0) 3163 ep->ird = 1; 3164 } else { 3165 if (peer2peer && 3166 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) && 3167 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0) 3168 ep->ird = 1; 3169 } 3170 3171 pr_debug("ird %d ord %d\n", ep->ird, ep->ord); 3172 3173 ep->com.cm_id = cm_id; 3174 ref_cm_id(&ep->com); 3175 ep->com.qp = qp; 3176 ref_qp(ep); 3177 3178 /* bind QP to EP and move to RTS */ 3179 attrs.mpa_attr = ep->mpa_attr; 3180 attrs.max_ird = ep->ird; 3181 attrs.max_ord = ep->ord; 3182 attrs.llp_stream_handle = ep; 3183 attrs.next_state = C4IW_QP_STATE_RTS; 3184 3185 /* bind QP and TID with INIT_WR */ 3186 mask = C4IW_QP_ATTR_NEXT_STATE | 3187 C4IW_QP_ATTR_LLP_STREAM_HANDLE | 3188 C4IW_QP_ATTR_MPA_ATTR | 3189 C4IW_QP_ATTR_MAX_IRD | 3190 C4IW_QP_ATTR_MAX_ORD; 3191 3192 err = c4iw_modify_qp(ep->com.qp->rhp, 3193 ep->com.qp, mask, &attrs, 1); 3194 if (err) 3195 goto err_deref_cm_id; 3196 3197 set_bit(STOP_MPA_TIMER, &ep->com.flags); 3198 err = send_mpa_reply(ep, conn_param->private_data, 3199 conn_param->private_data_len); 3200 if (err) 3201 goto err_deref_cm_id; 3202 3203 __state_set(&ep->com, FPDU_MODE); 3204 established_upcall(ep); 3205 mutex_unlock(&ep->com.mutex); 3206 c4iw_put_ep(&ep->com); 3207 return 0; 3208 err_deref_cm_id: 3209 deref_cm_id(&ep->com); 3210 err_abort: 3211 abort = 1; 3212 err_out: 3213 mutex_unlock(&ep->com.mutex); 3214 if (abort) 3215 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 3216 c4iw_put_ep(&ep->com); 3217 return err; 3218 } 3219 3220 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id) 3221 { 3222 struct in_device *ind; 3223 int found = 0; 3224 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr; 3225 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr; 3226 3227 ind = in_dev_get(dev->rdev.lldi.ports[0]); 3228 if (!ind) 3229 return -EADDRNOTAVAIL; 3230 for_primary_ifa(ind) { 3231 laddr->sin_addr.s_addr = ifa->ifa_address; 3232 raddr->sin_addr.s_addr = ifa->ifa_address; 3233 found = 1; 3234 break; 3235 } 3236 endfor_ifa(ind); 3237 in_dev_put(ind); 3238 return found ? 0 : -EADDRNOTAVAIL; 3239 } 3240 3241 static int get_lladdr(struct net_device *dev, struct in6_addr *addr, 3242 unsigned char banned_flags) 3243 { 3244 struct inet6_dev *idev; 3245 int err = -EADDRNOTAVAIL; 3246 3247 rcu_read_lock(); 3248 idev = __in6_dev_get(dev); 3249 if (idev != NULL) { 3250 struct inet6_ifaddr *ifp; 3251 3252 read_lock_bh(&idev->lock); 3253 list_for_each_entry(ifp, &idev->addr_list, if_list) { 3254 if (ifp->scope == IFA_LINK && 3255 !(ifp->flags & banned_flags)) { 3256 memcpy(addr, &ifp->addr, 16); 3257 err = 0; 3258 break; 3259 } 3260 } 3261 read_unlock_bh(&idev->lock); 3262 } 3263 rcu_read_unlock(); 3264 return err; 3265 } 3266 3267 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id) 3268 { 3269 struct in6_addr uninitialized_var(addr); 3270 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr; 3271 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr; 3272 3273 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) { 3274 memcpy(la6->sin6_addr.s6_addr, &addr, 16); 3275 memcpy(ra6->sin6_addr.s6_addr, &addr, 16); 3276 return 0; 3277 } 3278 return -EADDRNOTAVAIL; 3279 } 3280 3281 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 3282 { 3283 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 3284 struct c4iw_ep *ep; 3285 int err = 0; 3286 struct sockaddr_in *laddr; 3287 struct sockaddr_in *raddr; 3288 struct sockaddr_in6 *laddr6; 3289 struct sockaddr_in6 *raddr6; 3290 __u8 *ra; 3291 int iptype; 3292 3293 if ((conn_param->ord > cur_max_read_depth(dev)) || 3294 (conn_param->ird > cur_max_read_depth(dev))) { 3295 err = -EINVAL; 3296 goto out; 3297 } 3298 ep = alloc_ep(sizeof(*ep), GFP_KERNEL); 3299 if (!ep) { 3300 pr_err("%s - cannot alloc ep\n", __func__); 3301 err = -ENOMEM; 3302 goto out; 3303 } 3304 3305 skb_queue_head_init(&ep->com.ep_skb_list); 3306 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) { 3307 err = -ENOMEM; 3308 goto fail1; 3309 } 3310 3311 timer_setup(&ep->timer, ep_timeout, 0); 3312 ep->plen = conn_param->private_data_len; 3313 if (ep->plen) 3314 memcpy(ep->mpa_pkt + sizeof(struct mpa_message), 3315 conn_param->private_data, ep->plen); 3316 ep->ird = conn_param->ird; 3317 ep->ord = conn_param->ord; 3318 3319 if (peer2peer && ep->ord == 0) 3320 ep->ord = 1; 3321 3322 ep->com.cm_id = cm_id; 3323 ref_cm_id(&ep->com); 3324 cm_id->provider_data = ep; 3325 ep->com.dev = dev; 3326 ep->com.qp = get_qhp(dev, conn_param->qpn); 3327 if (!ep->com.qp) { 3328 pr_warn("%s qpn 0x%x not found!\n", __func__, conn_param->qpn); 3329 err = -EINVAL; 3330 goto fail2; 3331 } 3332 ref_qp(ep); 3333 pr_debug("qpn 0x%x qp %p cm_id %p\n", conn_param->qpn, 3334 ep->com.qp, cm_id); 3335 3336 /* 3337 * Allocate an active TID to initiate a TCP connection. 3338 */ 3339 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep); 3340 if (ep->atid == -1) { 3341 pr_err("%s - cannot alloc atid\n", __func__); 3342 err = -ENOMEM; 3343 goto fail2; 3344 } 3345 insert_handle(dev, &dev->atid_idr, ep, ep->atid); 3346 3347 memcpy(&ep->com.local_addr, &cm_id->m_local_addr, 3348 sizeof(ep->com.local_addr)); 3349 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr, 3350 sizeof(ep->com.remote_addr)); 3351 3352 laddr = (struct sockaddr_in *)&ep->com.local_addr; 3353 raddr = (struct sockaddr_in *)&ep->com.remote_addr; 3354 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr; 3355 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr; 3356 3357 if (cm_id->m_remote_addr.ss_family == AF_INET) { 3358 iptype = 4; 3359 ra = (__u8 *)&raddr->sin_addr; 3360 3361 /* 3362 * Handle loopback requests to INADDR_ANY. 3363 */ 3364 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) { 3365 err = pick_local_ipaddrs(dev, cm_id); 3366 if (err) 3367 goto fail2; 3368 } 3369 3370 /* find a route */ 3371 pr_debug("saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n", 3372 &laddr->sin_addr, ntohs(laddr->sin_port), 3373 ra, ntohs(raddr->sin_port)); 3374 ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev, 3375 laddr->sin_addr.s_addr, 3376 raddr->sin_addr.s_addr, 3377 laddr->sin_port, 3378 raddr->sin_port, cm_id->tos); 3379 } else { 3380 iptype = 6; 3381 ra = (__u8 *)&raddr6->sin6_addr; 3382 3383 /* 3384 * Handle loopback requests to INADDR_ANY. 3385 */ 3386 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) { 3387 err = pick_local_ip6addrs(dev, cm_id); 3388 if (err) 3389 goto fail2; 3390 } 3391 3392 /* find a route */ 3393 pr_debug("saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n", 3394 laddr6->sin6_addr.s6_addr, 3395 ntohs(laddr6->sin6_port), 3396 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port)); 3397 ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev, 3398 laddr6->sin6_addr.s6_addr, 3399 raddr6->sin6_addr.s6_addr, 3400 laddr6->sin6_port, 3401 raddr6->sin6_port, cm_id->tos, 3402 raddr6->sin6_scope_id); 3403 } 3404 if (!ep->dst) { 3405 pr_err("%s - cannot find route\n", __func__); 3406 err = -EHOSTUNREACH; 3407 goto fail3; 3408 } 3409 3410 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true, 3411 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos); 3412 if (err) { 3413 pr_err("%s - cannot alloc l2e\n", __func__); 3414 goto fail4; 3415 } 3416 3417 pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n", 3418 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid, 3419 ep->l2t->idx); 3420 3421 state_set(&ep->com, CONNECTING); 3422 ep->tos = cm_id->tos; 3423 3424 /* send connect request to rnic */ 3425 err = send_connect(ep); 3426 if (!err) 3427 goto out; 3428 3429 cxgb4_l2t_release(ep->l2t); 3430 fail4: 3431 dst_release(ep->dst); 3432 fail3: 3433 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid); 3434 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid); 3435 fail2: 3436 skb_queue_purge(&ep->com.ep_skb_list); 3437 deref_cm_id(&ep->com); 3438 fail1: 3439 c4iw_put_ep(&ep->com); 3440 out: 3441 return err; 3442 } 3443 3444 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep) 3445 { 3446 int err; 3447 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) 3448 &ep->com.local_addr; 3449 3450 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) { 3451 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0], 3452 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3453 if (err) 3454 return err; 3455 } 3456 c4iw_init_wr_wait(ep->com.wr_waitp); 3457 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0], 3458 ep->stid, &sin6->sin6_addr, 3459 sin6->sin6_port, 3460 ep->com.dev->rdev.lldi.rxq_ids[0]); 3461 if (!err) 3462 err = c4iw_wait_for_reply(&ep->com.dev->rdev, 3463 ep->com.wr_waitp, 3464 0, 0, __func__); 3465 else if (err > 0) 3466 err = net_xmit_errno(err); 3467 if (err) { 3468 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 3469 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3470 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n", 3471 err, ep->stid, 3472 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port)); 3473 } 3474 return err; 3475 } 3476 3477 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep) 3478 { 3479 int err; 3480 struct sockaddr_in *sin = (struct sockaddr_in *) 3481 &ep->com.local_addr; 3482 3483 if (dev->rdev.lldi.enable_fw_ofld_conn) { 3484 do { 3485 err = cxgb4_create_server_filter( 3486 ep->com.dev->rdev.lldi.ports[0], ep->stid, 3487 sin->sin_addr.s_addr, sin->sin_port, 0, 3488 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0); 3489 if (err == -EBUSY) { 3490 if (c4iw_fatal_error(&ep->com.dev->rdev)) { 3491 err = -EIO; 3492 break; 3493 } 3494 set_current_state(TASK_UNINTERRUPTIBLE); 3495 schedule_timeout(usecs_to_jiffies(100)); 3496 } 3497 } while (err == -EBUSY); 3498 } else { 3499 c4iw_init_wr_wait(ep->com.wr_waitp); 3500 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0], 3501 ep->stid, sin->sin_addr.s_addr, sin->sin_port, 3502 0, ep->com.dev->rdev.lldi.rxq_ids[0]); 3503 if (!err) 3504 err = c4iw_wait_for_reply(&ep->com.dev->rdev, 3505 ep->com.wr_waitp, 3506 0, 0, __func__); 3507 else if (err > 0) 3508 err = net_xmit_errno(err); 3509 } 3510 if (err) 3511 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n" 3512 , err, ep->stid, 3513 &sin->sin_addr, ntohs(sin->sin_port)); 3514 return err; 3515 } 3516 3517 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog) 3518 { 3519 int err = 0; 3520 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 3521 struct c4iw_listen_ep *ep; 3522 3523 might_sleep(); 3524 3525 ep = alloc_ep(sizeof(*ep), GFP_KERNEL); 3526 if (!ep) { 3527 pr_err("%s - cannot alloc ep\n", __func__); 3528 err = -ENOMEM; 3529 goto fail1; 3530 } 3531 skb_queue_head_init(&ep->com.ep_skb_list); 3532 pr_debug("ep %p\n", ep); 3533 ep->com.cm_id = cm_id; 3534 ref_cm_id(&ep->com); 3535 ep->com.dev = dev; 3536 ep->backlog = backlog; 3537 memcpy(&ep->com.local_addr, &cm_id->m_local_addr, 3538 sizeof(ep->com.local_addr)); 3539 3540 /* 3541 * Allocate a server TID. 3542 */ 3543 if (dev->rdev.lldi.enable_fw_ofld_conn && 3544 ep->com.local_addr.ss_family == AF_INET) 3545 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids, 3546 cm_id->m_local_addr.ss_family, ep); 3547 else 3548 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids, 3549 cm_id->m_local_addr.ss_family, ep); 3550 3551 if (ep->stid == -1) { 3552 pr_err("%s - cannot alloc stid\n", __func__); 3553 err = -ENOMEM; 3554 goto fail2; 3555 } 3556 insert_handle(dev, &dev->stid_idr, ep, ep->stid); 3557 3558 state_set(&ep->com, LISTEN); 3559 if (ep->com.local_addr.ss_family == AF_INET) 3560 err = create_server4(dev, ep); 3561 else 3562 err = create_server6(dev, ep); 3563 if (!err) { 3564 cm_id->provider_data = ep; 3565 goto out; 3566 } 3567 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid); 3568 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, 3569 ep->com.local_addr.ss_family); 3570 fail2: 3571 deref_cm_id(&ep->com); 3572 c4iw_put_ep(&ep->com); 3573 fail1: 3574 out: 3575 return err; 3576 } 3577 3578 int c4iw_destroy_listen(struct iw_cm_id *cm_id) 3579 { 3580 int err; 3581 struct c4iw_listen_ep *ep = to_listen_ep(cm_id); 3582 3583 pr_debug("ep %p\n", ep); 3584 3585 might_sleep(); 3586 state_set(&ep->com, DEAD); 3587 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn && 3588 ep->com.local_addr.ss_family == AF_INET) { 3589 err = cxgb4_remove_server_filter( 3590 ep->com.dev->rdev.lldi.ports[0], ep->stid, 3591 ep->com.dev->rdev.lldi.rxq_ids[0], 0); 3592 } else { 3593 struct sockaddr_in6 *sin6; 3594 c4iw_init_wr_wait(ep->com.wr_waitp); 3595 err = cxgb4_remove_server( 3596 ep->com.dev->rdev.lldi.ports[0], ep->stid, 3597 ep->com.dev->rdev.lldi.rxq_ids[0], 0); 3598 if (err) 3599 goto done; 3600 err = c4iw_wait_for_reply(&ep->com.dev->rdev, ep->com.wr_waitp, 3601 0, 0, __func__); 3602 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr; 3603 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 3604 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3605 } 3606 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid); 3607 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, 3608 ep->com.local_addr.ss_family); 3609 done: 3610 deref_cm_id(&ep->com); 3611 c4iw_put_ep(&ep->com); 3612 return err; 3613 } 3614 3615 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) 3616 { 3617 int ret = 0; 3618 int close = 0; 3619 int fatal = 0; 3620 struct c4iw_rdev *rdev; 3621 3622 mutex_lock(&ep->com.mutex); 3623 3624 pr_debug("ep %p state %s, abrupt %d\n", ep, 3625 states[ep->com.state], abrupt); 3626 3627 /* 3628 * Ref the ep here in case we have fatal errors causing the 3629 * ep to be released and freed. 3630 */ 3631 c4iw_get_ep(&ep->com); 3632 3633 rdev = &ep->com.dev->rdev; 3634 if (c4iw_fatal_error(rdev)) { 3635 fatal = 1; 3636 close_complete_upcall(ep, -EIO); 3637 ep->com.state = DEAD; 3638 } 3639 switch (ep->com.state) { 3640 case MPA_REQ_WAIT: 3641 case MPA_REQ_SENT: 3642 case MPA_REQ_RCVD: 3643 case MPA_REP_SENT: 3644 case FPDU_MODE: 3645 case CONNECTING: 3646 close = 1; 3647 if (abrupt) 3648 ep->com.state = ABORTING; 3649 else { 3650 ep->com.state = CLOSING; 3651 3652 /* 3653 * if we close before we see the fw4_ack() then we fix 3654 * up the timer state since we're reusing it. 3655 */ 3656 if (ep->mpa_skb && 3657 test_bit(STOP_MPA_TIMER, &ep->com.flags)) { 3658 clear_bit(STOP_MPA_TIMER, &ep->com.flags); 3659 stop_ep_timer(ep); 3660 } 3661 start_ep_timer(ep); 3662 } 3663 set_bit(CLOSE_SENT, &ep->com.flags); 3664 break; 3665 case CLOSING: 3666 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) { 3667 close = 1; 3668 if (abrupt) { 3669 (void)stop_ep_timer(ep); 3670 ep->com.state = ABORTING; 3671 } else 3672 ep->com.state = MORIBUND; 3673 } 3674 break; 3675 case MORIBUND: 3676 case ABORTING: 3677 case DEAD: 3678 pr_debug("ignoring disconnect ep %p state %u\n", 3679 ep, ep->com.state); 3680 break; 3681 default: 3682 WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state); 3683 break; 3684 } 3685 3686 if (close) { 3687 if (abrupt) { 3688 set_bit(EP_DISC_ABORT, &ep->com.history); 3689 ret = send_abort(ep); 3690 } else { 3691 set_bit(EP_DISC_CLOSE, &ep->com.history); 3692 ret = send_halfclose(ep); 3693 } 3694 if (ret) { 3695 set_bit(EP_DISC_FAIL, &ep->com.history); 3696 if (!abrupt) { 3697 stop_ep_timer(ep); 3698 close_complete_upcall(ep, -EIO); 3699 } 3700 if (ep->com.qp) { 3701 struct c4iw_qp_attributes attrs; 3702 3703 attrs.next_state = C4IW_QP_STATE_ERROR; 3704 ret = c4iw_modify_qp(ep->com.qp->rhp, 3705 ep->com.qp, 3706 C4IW_QP_ATTR_NEXT_STATE, 3707 &attrs, 1); 3708 if (ret) 3709 pr_err("%s - qp <- error failed!\n", 3710 __func__); 3711 } 3712 fatal = 1; 3713 } 3714 } 3715 mutex_unlock(&ep->com.mutex); 3716 c4iw_put_ep(&ep->com); 3717 if (fatal) 3718 release_ep_resources(ep); 3719 return ret; 3720 } 3721 3722 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb, 3723 struct cpl_fw6_msg_ofld_connection_wr_rpl *req) 3724 { 3725 struct c4iw_ep *ep; 3726 int atid = be32_to_cpu(req->tid); 3727 3728 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids, 3729 (__force u32) req->tid); 3730 if (!ep) 3731 return; 3732 3733 switch (req->retval) { 3734 case FW_ENOMEM: 3735 set_bit(ACT_RETRY_NOMEM, &ep->com.history); 3736 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) { 3737 send_fw_act_open_req(ep, atid); 3738 return; 3739 } 3740 /* fall through */ 3741 case FW_EADDRINUSE: 3742 set_bit(ACT_RETRY_INUSE, &ep->com.history); 3743 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) { 3744 send_fw_act_open_req(ep, atid); 3745 return; 3746 } 3747 break; 3748 default: 3749 pr_info("%s unexpected ofld conn wr retval %d\n", 3750 __func__, req->retval); 3751 break; 3752 } 3753 pr_err("active ofld_connect_wr failure %d atid %d\n", 3754 req->retval, atid); 3755 mutex_lock(&dev->rdev.stats.lock); 3756 dev->rdev.stats.act_ofld_conn_fails++; 3757 mutex_unlock(&dev->rdev.stats.lock); 3758 connect_reply_upcall(ep, status2errno(req->retval)); 3759 state_set(&ep->com, DEAD); 3760 if (ep->com.remote_addr.ss_family == AF_INET6) { 3761 struct sockaddr_in6 *sin6 = 3762 (struct sockaddr_in6 *)&ep->com.local_addr; 3763 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 3764 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3765 } 3766 remove_handle(dev, &dev->atid_idr, atid); 3767 cxgb4_free_atid(dev->rdev.lldi.tids, atid); 3768 dst_release(ep->dst); 3769 cxgb4_l2t_release(ep->l2t); 3770 c4iw_put_ep(&ep->com); 3771 } 3772 3773 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb, 3774 struct cpl_fw6_msg_ofld_connection_wr_rpl *req) 3775 { 3776 struct sk_buff *rpl_skb; 3777 struct cpl_pass_accept_req *cpl; 3778 int ret; 3779 3780 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie; 3781 if (req->retval) { 3782 pr_err("%s passive open failure %d\n", __func__, req->retval); 3783 mutex_lock(&dev->rdev.stats.lock); 3784 dev->rdev.stats.pas_ofld_conn_fails++; 3785 mutex_unlock(&dev->rdev.stats.lock); 3786 kfree_skb(rpl_skb); 3787 } else { 3788 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb); 3789 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 3790 (__force u32) htonl( 3791 (__force u32) req->tid))); 3792 ret = pass_accept_req(dev, rpl_skb); 3793 if (!ret) 3794 kfree_skb(rpl_skb); 3795 } 3796 return; 3797 } 3798 3799 static inline u64 t4_tcb_get_field64(__be64 *tcb, u16 word) 3800 { 3801 u64 tlo = be64_to_cpu(tcb[((31 - word) / 2)]); 3802 u64 thi = be64_to_cpu(tcb[((31 - word) / 2) - 1]); 3803 u64 t; 3804 u32 shift = 32; 3805 3806 t = (thi << shift) | (tlo >> shift); 3807 3808 return t; 3809 } 3810 3811 static inline u32 t4_tcb_get_field32(__be64 *tcb, u16 word, u32 mask, u32 shift) 3812 { 3813 u32 v; 3814 u64 t = be64_to_cpu(tcb[(31 - word) / 2]); 3815 3816 if (word & 0x1) 3817 shift += 32; 3818 v = (t >> shift) & mask; 3819 return v; 3820 } 3821 3822 static int read_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 3823 { 3824 struct cpl_get_tcb_rpl *rpl = cplhdr(skb); 3825 __be64 *tcb = (__be64 *)(rpl + 1); 3826 unsigned int tid = GET_TID(rpl); 3827 struct c4iw_ep *ep; 3828 u64 t_flags_64; 3829 u32 rx_pdu_out; 3830 3831 ep = get_ep_from_tid(dev, tid); 3832 if (!ep) 3833 return 0; 3834 /* Examine the TF_RX_PDU_OUT (bit 49 of the t_flags) in order to 3835 * determine if there's a rx PDU feedback event pending. 3836 * 3837 * If that bit is set, it means we'll need to re-read the TCB's 3838 * rq_start value. The final value is the one present in a TCB 3839 * with the TF_RX_PDU_OUT bit cleared. 3840 */ 3841 3842 t_flags_64 = t4_tcb_get_field64(tcb, TCB_T_FLAGS_W); 3843 rx_pdu_out = (t_flags_64 & TF_RX_PDU_OUT_V(1)) >> TF_RX_PDU_OUT_S; 3844 3845 c4iw_put_ep(&ep->com); /* from get_ep_from_tid() */ 3846 c4iw_put_ep(&ep->com); /* from read_tcb() */ 3847 3848 /* If TF_RX_PDU_OUT bit is set, re-read the TCB */ 3849 if (rx_pdu_out) { 3850 if (++ep->rx_pdu_out_cnt >= 2) { 3851 WARN_ONCE(1, "tcb re-read() reached the guard limit, finishing the cleanup\n"); 3852 goto cleanup; 3853 } 3854 read_tcb(ep); 3855 return 0; 3856 } 3857 3858 ep->srqe_idx = t4_tcb_get_field32(tcb, TCB_RQ_START_W, TCB_RQ_START_W, 3859 TCB_RQ_START_S); 3860 cleanup: 3861 pr_debug("ep %p tid %u %016x\n", ep, ep->hwtid, ep->srqe_idx); 3862 3863 if (test_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) 3864 finish_peer_abort(dev, ep); 3865 else if (test_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) 3866 send_abort_req(ep); 3867 else 3868 WARN_ONCE(1, "unexpected state!"); 3869 3870 return 0; 3871 } 3872 3873 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb) 3874 { 3875 struct cpl_fw6_msg *rpl = cplhdr(skb); 3876 struct cpl_fw6_msg_ofld_connection_wr_rpl *req; 3877 3878 switch (rpl->type) { 3879 case FW6_TYPE_CQE: 3880 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]); 3881 break; 3882 case FW6_TYPE_OFLD_CONNECTION_WR_RPL: 3883 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data; 3884 switch (req->t_state) { 3885 case TCP_SYN_SENT: 3886 active_ofld_conn_reply(dev, skb, req); 3887 break; 3888 case TCP_SYN_RECV: 3889 passive_ofld_conn_reply(dev, skb, req); 3890 break; 3891 default: 3892 pr_err("%s unexpected ofld conn wr state %d\n", 3893 __func__, req->t_state); 3894 break; 3895 } 3896 break; 3897 } 3898 return 0; 3899 } 3900 3901 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos) 3902 { 3903 __be32 l2info; 3904 __be16 hdr_len, vlantag, len; 3905 u16 eth_hdr_len; 3906 int tcp_hdr_len, ip_hdr_len; 3907 u8 intf; 3908 struct cpl_rx_pkt *cpl = cplhdr(skb); 3909 struct cpl_pass_accept_req *req; 3910 struct tcp_options_received tmp_opt; 3911 struct c4iw_dev *dev; 3912 enum chip_type type; 3913 3914 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *))); 3915 /* Store values from cpl_rx_pkt in temporary location. */ 3916 vlantag = cpl->vlan; 3917 len = cpl->len; 3918 l2info = cpl->l2info; 3919 hdr_len = cpl->hdr_len; 3920 intf = cpl->iff; 3921 3922 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header)); 3923 3924 /* 3925 * We need to parse the TCP options from SYN packet. 3926 * to generate cpl_pass_accept_req. 3927 */ 3928 memset(&tmp_opt, 0, sizeof(tmp_opt)); 3929 tcp_clear_options(&tmp_opt); 3930 tcp_parse_options(&init_net, skb, &tmp_opt, 0, NULL); 3931 3932 req = __skb_push(skb, sizeof(*req)); 3933 memset(req, 0, sizeof(*req)); 3934 req->l2info = cpu_to_be16(SYN_INTF_V(intf) | 3935 SYN_MAC_IDX_V(RX_MACIDX_G( 3936 be32_to_cpu(l2info))) | 3937 SYN_XACT_MATCH_F); 3938 type = dev->rdev.lldi.adapter_type; 3939 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len)); 3940 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len)); 3941 req->hdr_len = 3942 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info)))); 3943 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) { 3944 eth_hdr_len = is_t4(type) ? 3945 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) : 3946 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info)); 3947 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) | 3948 IP_HDR_LEN_V(ip_hdr_len) | 3949 ETH_HDR_LEN_V(eth_hdr_len)); 3950 } else { /* T6 and later */ 3951 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info)); 3952 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) | 3953 T6_IP_HDR_LEN_V(ip_hdr_len) | 3954 T6_ETH_HDR_LEN_V(eth_hdr_len)); 3955 } 3956 req->vlan = vlantag; 3957 req->len = len; 3958 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) | 3959 PASS_OPEN_TOS_V(tos)); 3960 req->tcpopt.mss = htons(tmp_opt.mss_clamp); 3961 if (tmp_opt.wscale_ok) 3962 req->tcpopt.wsf = tmp_opt.snd_wscale; 3963 req->tcpopt.tstamp = tmp_opt.saw_tstamp; 3964 if (tmp_opt.sack_ok) 3965 req->tcpopt.sack = 1; 3966 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0)); 3967 return; 3968 } 3969 3970 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb, 3971 __be32 laddr, __be16 lport, 3972 __be32 raddr, __be16 rport, 3973 u32 rcv_isn, u32 filter, u16 window, 3974 u32 rss_qid, u8 port_id) 3975 { 3976 struct sk_buff *req_skb; 3977 struct fw_ofld_connection_wr *req; 3978 struct cpl_pass_accept_req *cpl = cplhdr(skb); 3979 int ret; 3980 3981 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL); 3982 if (!req_skb) 3983 return; 3984 req = __skb_put_zero(req_skb, sizeof(*req)); 3985 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F); 3986 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16))); 3987 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F); 3988 req->le.filter = (__force __be32) filter; 3989 req->le.lport = lport; 3990 req->le.pport = rport; 3991 req->le.u.ipv4.lip = laddr; 3992 req->le.u.ipv4.pip = raddr; 3993 req->tcb.rcv_nxt = htonl(rcv_isn + 1); 3994 req->tcb.rcv_adv = htons(window); 3995 req->tcb.t_state_to_astid = 3996 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) | 3997 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) | 3998 FW_OFLD_CONNECTION_WR_ASTID_V( 3999 PASS_OPEN_TID_G(ntohl(cpl->tos_stid)))); 4000 4001 /* 4002 * We store the qid in opt2 which will be used by the firmware 4003 * to send us the wr response. 4004 */ 4005 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid)); 4006 4007 /* 4008 * We initialize the MSS index in TCB to 0xF. 4009 * So that when driver sends cpl_pass_accept_rpl 4010 * TCB picks up the correct value. If this was 0 4011 * TP will ignore any value > 0 for MSS index. 4012 */ 4013 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF)); 4014 req->cookie = (uintptr_t)skb; 4015 4016 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id); 4017 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb); 4018 if (ret < 0) { 4019 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__, 4020 ret); 4021 kfree_skb(skb); 4022 kfree_skb(req_skb); 4023 } 4024 } 4025 4026 /* 4027 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt 4028 * messages when a filter is being used instead of server to 4029 * redirect a syn packet. When packets hit filter they are redirected 4030 * to the offload queue and driver tries to establish the connection 4031 * using firmware work request. 4032 */ 4033 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb) 4034 { 4035 int stid; 4036 unsigned int filter; 4037 struct ethhdr *eh = NULL; 4038 struct vlan_ethhdr *vlan_eh = NULL; 4039 struct iphdr *iph; 4040 struct tcphdr *tcph; 4041 struct rss_header *rss = (void *)skb->data; 4042 struct cpl_rx_pkt *cpl = (void *)skb->data; 4043 struct cpl_pass_accept_req *req = (void *)(rss + 1); 4044 struct l2t_entry *e; 4045 struct dst_entry *dst; 4046 struct c4iw_ep *lep = NULL; 4047 u16 window; 4048 struct port_info *pi; 4049 struct net_device *pdev; 4050 u16 rss_qid, eth_hdr_len; 4051 int step; 4052 struct neighbour *neigh; 4053 4054 /* Drop all non-SYN packets */ 4055 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F))) 4056 goto reject; 4057 4058 /* 4059 * Drop all packets which did not hit the filter. 4060 * Unlikely to happen. 4061 */ 4062 if (!(rss->filter_hit && rss->filter_tid)) 4063 goto reject; 4064 4065 /* 4066 * Calculate the server tid from filter hit index from cpl_rx_pkt. 4067 */ 4068 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val); 4069 4070 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid); 4071 if (!lep) { 4072 pr_warn("%s connect request on invalid stid %d\n", 4073 __func__, stid); 4074 goto reject; 4075 } 4076 4077 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) { 4078 case CHELSIO_T4: 4079 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info)); 4080 break; 4081 case CHELSIO_T5: 4082 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info)); 4083 break; 4084 case CHELSIO_T6: 4085 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info)); 4086 break; 4087 default: 4088 pr_err("T%d Chip is not supported\n", 4089 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)); 4090 goto reject; 4091 } 4092 4093 if (eth_hdr_len == ETH_HLEN) { 4094 eh = (struct ethhdr *)(req + 1); 4095 iph = (struct iphdr *)(eh + 1); 4096 } else { 4097 vlan_eh = (struct vlan_ethhdr *)(req + 1); 4098 iph = (struct iphdr *)(vlan_eh + 1); 4099 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(cpl->vlan)); 4100 } 4101 4102 if (iph->version != 0x4) 4103 goto reject; 4104 4105 tcph = (struct tcphdr *)(iph + 1); 4106 skb_set_network_header(skb, (void *)iph - (void *)rss); 4107 skb_set_transport_header(skb, (void *)tcph - (void *)rss); 4108 skb_get(skb); 4109 4110 pr_debug("lip 0x%x lport %u pip 0x%x pport %u tos %d\n", 4111 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr), 4112 ntohs(tcph->source), iph->tos); 4113 4114 dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev, 4115 iph->daddr, iph->saddr, tcph->dest, 4116 tcph->source, iph->tos); 4117 if (!dst) { 4118 pr_err("%s - failed to find dst entry!\n", __func__); 4119 goto reject; 4120 } 4121 neigh = dst_neigh_lookup_skb(dst, skb); 4122 4123 if (!neigh) { 4124 pr_err("%s - failed to allocate neigh!\n", __func__); 4125 goto free_dst; 4126 } 4127 4128 if (neigh->dev->flags & IFF_LOOPBACK) { 4129 pdev = ip_dev_find(&init_net, iph->daddr); 4130 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh, 4131 pdev, 0); 4132 pi = (struct port_info *)netdev_priv(pdev); 4133 dev_put(pdev); 4134 } else { 4135 pdev = get_real_dev(neigh->dev); 4136 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh, 4137 pdev, 0); 4138 pi = (struct port_info *)netdev_priv(pdev); 4139 } 4140 neigh_release(neigh); 4141 if (!e) { 4142 pr_err("%s - failed to allocate l2t entry!\n", 4143 __func__); 4144 goto free_dst; 4145 } 4146 4147 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan; 4148 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step]; 4149 window = (__force u16) htons((__force u16)tcph->window); 4150 4151 /* Calcuate filter portion for LE region. */ 4152 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple( 4153 dev->rdev.lldi.ports[0], 4154 e)); 4155 4156 /* 4157 * Synthesize the cpl_pass_accept_req. We have everything except the 4158 * TID. Once firmware sends a reply with TID we update the TID field 4159 * in cpl and pass it through the regular cpl_pass_accept_req path. 4160 */ 4161 build_cpl_pass_accept_req(skb, stid, iph->tos); 4162 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr, 4163 tcph->source, ntohl(tcph->seq), filter, window, 4164 rss_qid, pi->port_id); 4165 cxgb4_l2t_release(e); 4166 free_dst: 4167 dst_release(dst); 4168 reject: 4169 if (lep) 4170 c4iw_put_ep(&lep->com); 4171 return 0; 4172 } 4173 4174 /* 4175 * These are the real handlers that are called from a 4176 * work queue. 4177 */ 4178 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = { 4179 [CPL_ACT_ESTABLISH] = act_establish, 4180 [CPL_ACT_OPEN_RPL] = act_open_rpl, 4181 [CPL_RX_DATA] = rx_data, 4182 [CPL_ABORT_RPL_RSS] = abort_rpl, 4183 [CPL_ABORT_RPL] = abort_rpl, 4184 [CPL_PASS_OPEN_RPL] = pass_open_rpl, 4185 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl, 4186 [CPL_PASS_ACCEPT_REQ] = pass_accept_req, 4187 [CPL_PASS_ESTABLISH] = pass_establish, 4188 [CPL_PEER_CLOSE] = peer_close, 4189 [CPL_ABORT_REQ_RSS] = peer_abort, 4190 [CPL_CLOSE_CON_RPL] = close_con_rpl, 4191 [CPL_RDMA_TERMINATE] = terminate, 4192 [CPL_FW4_ACK] = fw4_ack, 4193 [CPL_GET_TCB_RPL] = read_tcb_rpl, 4194 [CPL_FW6_MSG] = deferred_fw6_msg, 4195 [CPL_RX_PKT] = rx_pkt, 4196 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe, 4197 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe 4198 }; 4199 4200 static void process_timeout(struct c4iw_ep *ep) 4201 { 4202 struct c4iw_qp_attributes attrs; 4203 int abort = 1; 4204 4205 mutex_lock(&ep->com.mutex); 4206 pr_debug("ep %p tid %u state %d\n", ep, ep->hwtid, ep->com.state); 4207 set_bit(TIMEDOUT, &ep->com.history); 4208 switch (ep->com.state) { 4209 case MPA_REQ_SENT: 4210 connect_reply_upcall(ep, -ETIMEDOUT); 4211 break; 4212 case MPA_REQ_WAIT: 4213 case MPA_REQ_RCVD: 4214 case MPA_REP_SENT: 4215 case FPDU_MODE: 4216 break; 4217 case CLOSING: 4218 case MORIBUND: 4219 if (ep->com.cm_id && ep->com.qp) { 4220 attrs.next_state = C4IW_QP_STATE_ERROR; 4221 c4iw_modify_qp(ep->com.qp->rhp, 4222 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, 4223 &attrs, 1); 4224 } 4225 close_complete_upcall(ep, -ETIMEDOUT); 4226 break; 4227 case ABORTING: 4228 case DEAD: 4229 4230 /* 4231 * These states are expected if the ep timed out at the same 4232 * time as another thread was calling stop_ep_timer(). 4233 * So we silently do nothing for these states. 4234 */ 4235 abort = 0; 4236 break; 4237 default: 4238 WARN(1, "%s unexpected state ep %p tid %u state %u\n", 4239 __func__, ep, ep->hwtid, ep->com.state); 4240 abort = 0; 4241 } 4242 mutex_unlock(&ep->com.mutex); 4243 if (abort) 4244 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 4245 c4iw_put_ep(&ep->com); 4246 } 4247 4248 static void process_timedout_eps(void) 4249 { 4250 struct c4iw_ep *ep; 4251 4252 spin_lock_irq(&timeout_lock); 4253 while (!list_empty(&timeout_list)) { 4254 struct list_head *tmp; 4255 4256 tmp = timeout_list.next; 4257 list_del(tmp); 4258 tmp->next = NULL; 4259 tmp->prev = NULL; 4260 spin_unlock_irq(&timeout_lock); 4261 ep = list_entry(tmp, struct c4iw_ep, entry); 4262 process_timeout(ep); 4263 spin_lock_irq(&timeout_lock); 4264 } 4265 spin_unlock_irq(&timeout_lock); 4266 } 4267 4268 static void process_work(struct work_struct *work) 4269 { 4270 struct sk_buff *skb = NULL; 4271 struct c4iw_dev *dev; 4272 struct cpl_act_establish *rpl; 4273 unsigned int opcode; 4274 int ret; 4275 4276 process_timedout_eps(); 4277 while ((skb = skb_dequeue(&rxq))) { 4278 rpl = cplhdr(skb); 4279 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *))); 4280 opcode = rpl->ot.opcode; 4281 4282 if (opcode >= ARRAY_SIZE(work_handlers) || 4283 !work_handlers[opcode]) { 4284 pr_err("No handler for opcode 0x%x.\n", opcode); 4285 kfree_skb(skb); 4286 } else { 4287 ret = work_handlers[opcode](dev, skb); 4288 if (!ret) 4289 kfree_skb(skb); 4290 } 4291 process_timedout_eps(); 4292 } 4293 } 4294 4295 static DECLARE_WORK(skb_work, process_work); 4296 4297 static void ep_timeout(struct timer_list *t) 4298 { 4299 struct c4iw_ep *ep = from_timer(ep, t, timer); 4300 int kickit = 0; 4301 4302 spin_lock(&timeout_lock); 4303 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 4304 /* 4305 * Only insert if it is not already on the list. 4306 */ 4307 if (!ep->entry.next) { 4308 list_add_tail(&ep->entry, &timeout_list); 4309 kickit = 1; 4310 } 4311 } 4312 spin_unlock(&timeout_lock); 4313 if (kickit) 4314 queue_work(workq, &skb_work); 4315 } 4316 4317 /* 4318 * All the CM events are handled on a work queue to have a safe context. 4319 */ 4320 static int sched(struct c4iw_dev *dev, struct sk_buff *skb) 4321 { 4322 4323 /* 4324 * Save dev in the skb->cb area. 4325 */ 4326 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev; 4327 4328 /* 4329 * Queue the skb and schedule the worker thread. 4330 */ 4331 skb_queue_tail(&rxq, skb); 4332 queue_work(workq, &skb_work); 4333 return 0; 4334 } 4335 4336 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 4337 { 4338 struct cpl_set_tcb_rpl *rpl = cplhdr(skb); 4339 4340 if (rpl->status != CPL_ERR_NONE) { 4341 pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n", 4342 rpl->status, GET_TID(rpl)); 4343 } 4344 kfree_skb(skb); 4345 return 0; 4346 } 4347 4348 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb) 4349 { 4350 struct cpl_fw6_msg *rpl = cplhdr(skb); 4351 struct c4iw_wr_wait *wr_waitp; 4352 int ret; 4353 4354 pr_debug("type %u\n", rpl->type); 4355 4356 switch (rpl->type) { 4357 case FW6_TYPE_WR_RPL: 4358 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff); 4359 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1]; 4360 pr_debug("wr_waitp %p ret %u\n", wr_waitp, ret); 4361 if (wr_waitp) 4362 c4iw_wake_up_deref(wr_waitp, ret ? -ret : 0); 4363 kfree_skb(skb); 4364 break; 4365 case FW6_TYPE_CQE: 4366 case FW6_TYPE_OFLD_CONNECTION_WR_RPL: 4367 sched(dev, skb); 4368 break; 4369 default: 4370 pr_err("%s unexpected fw6 msg type %u\n", 4371 __func__, rpl->type); 4372 kfree_skb(skb); 4373 break; 4374 } 4375 return 0; 4376 } 4377 4378 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb) 4379 { 4380 struct cpl_abort_req_rss *req = cplhdr(skb); 4381 struct c4iw_ep *ep; 4382 unsigned int tid = GET_TID(req); 4383 4384 ep = get_ep_from_tid(dev, tid); 4385 /* This EP will be dereferenced in peer_abort() */ 4386 if (!ep) { 4387 pr_warn("Abort on non-existent endpoint, tid %d\n", tid); 4388 kfree_skb(skb); 4389 return 0; 4390 } 4391 if (cxgb_is_neg_adv(req->status)) { 4392 pr_debug("Negative advice on abort- tid %u status %d (%s)\n", 4393 ep->hwtid, req->status, 4394 neg_adv_str(req->status)); 4395 goto out; 4396 } 4397 pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid, ep->com.state); 4398 4399 c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET); 4400 out: 4401 sched(dev, skb); 4402 return 0; 4403 } 4404 4405 /* 4406 * Most upcalls from the T4 Core go to sched() to 4407 * schedule the processing on a work queue. 4408 */ 4409 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = { 4410 [CPL_ACT_ESTABLISH] = sched, 4411 [CPL_ACT_OPEN_RPL] = sched, 4412 [CPL_RX_DATA] = sched, 4413 [CPL_ABORT_RPL_RSS] = sched, 4414 [CPL_ABORT_RPL] = sched, 4415 [CPL_PASS_OPEN_RPL] = sched, 4416 [CPL_CLOSE_LISTSRV_RPL] = sched, 4417 [CPL_PASS_ACCEPT_REQ] = sched, 4418 [CPL_PASS_ESTABLISH] = sched, 4419 [CPL_PEER_CLOSE] = sched, 4420 [CPL_CLOSE_CON_RPL] = sched, 4421 [CPL_ABORT_REQ_RSS] = peer_abort_intr, 4422 [CPL_RDMA_TERMINATE] = sched, 4423 [CPL_FW4_ACK] = sched, 4424 [CPL_SET_TCB_RPL] = set_tcb_rpl, 4425 [CPL_GET_TCB_RPL] = sched, 4426 [CPL_FW6_MSG] = fw6_msg, 4427 [CPL_RX_PKT] = sched 4428 }; 4429 4430 int __init c4iw_cm_init(void) 4431 { 4432 spin_lock_init(&timeout_lock); 4433 skb_queue_head_init(&rxq); 4434 4435 workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM); 4436 if (!workq) 4437 return -ENOMEM; 4438 4439 return 0; 4440 } 4441 4442 void c4iw_cm_term(void) 4443 { 4444 WARN_ON(!list_empty(&timeout_list)); 4445 flush_workqueue(workq); 4446 destroy_workqueue(workq); 4447 } 4448