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 "iw_cxgb4.h" 53 #include "clip_tbl.h" 54 55 static char *states[] = { 56 "idle", 57 "listen", 58 "connecting", 59 "mpa_wait_req", 60 "mpa_req_sent", 61 "mpa_req_rcvd", 62 "mpa_rep_sent", 63 "fpdu_mode", 64 "aborting", 65 "closing", 66 "moribund", 67 "dead", 68 NULL, 69 }; 70 71 static int nocong; 72 module_param(nocong, int, 0644); 73 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)"); 74 75 static int enable_ecn; 76 module_param(enable_ecn, int, 0644); 77 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)"); 78 79 static int dack_mode = 1; 80 module_param(dack_mode, int, 0644); 81 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)"); 82 83 uint c4iw_max_read_depth = 32; 84 module_param(c4iw_max_read_depth, int, 0644); 85 MODULE_PARM_DESC(c4iw_max_read_depth, 86 "Per-connection max ORD/IRD (default=32)"); 87 88 static int enable_tcp_timestamps; 89 module_param(enable_tcp_timestamps, int, 0644); 90 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)"); 91 92 static int enable_tcp_sack; 93 module_param(enable_tcp_sack, int, 0644); 94 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)"); 95 96 static int enable_tcp_window_scaling = 1; 97 module_param(enable_tcp_window_scaling, int, 0644); 98 MODULE_PARM_DESC(enable_tcp_window_scaling, 99 "Enable tcp window scaling (default=1)"); 100 101 int c4iw_debug; 102 module_param(c4iw_debug, int, 0644); 103 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)"); 104 105 static int peer2peer = 1; 106 module_param(peer2peer, int, 0644); 107 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)"); 108 109 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ; 110 module_param(p2p_type, int, 0644); 111 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: " 112 "1=RDMA_READ 0=RDMA_WRITE (default 1)"); 113 114 static int ep_timeout_secs = 60; 115 module_param(ep_timeout_secs, int, 0644); 116 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout " 117 "in seconds (default=60)"); 118 119 static int mpa_rev = 2; 120 module_param(mpa_rev, int, 0644); 121 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, " 122 "1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft" 123 " compliant (default=2)"); 124 125 static int markers_enabled; 126 module_param(markers_enabled, int, 0644); 127 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)"); 128 129 static int crc_enabled = 1; 130 module_param(crc_enabled, int, 0644); 131 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)"); 132 133 static int rcv_win = 256 * 1024; 134 module_param(rcv_win, int, 0644); 135 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)"); 136 137 static int snd_win = 128 * 1024; 138 module_param(snd_win, int, 0644); 139 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)"); 140 141 static struct workqueue_struct *workq; 142 143 static struct sk_buff_head rxq; 144 145 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp); 146 static void ep_timeout(unsigned long arg); 147 static void connect_reply_upcall(struct c4iw_ep *ep, int status); 148 149 static LIST_HEAD(timeout_list); 150 static spinlock_t timeout_lock; 151 152 static void deref_qp(struct c4iw_ep *ep) 153 { 154 c4iw_qp_rem_ref(&ep->com.qp->ibqp); 155 clear_bit(QP_REFERENCED, &ep->com.flags); 156 } 157 158 static void ref_qp(struct c4iw_ep *ep) 159 { 160 set_bit(QP_REFERENCED, &ep->com.flags); 161 c4iw_qp_add_ref(&ep->com.qp->ibqp); 162 } 163 164 static void start_ep_timer(struct c4iw_ep *ep) 165 { 166 PDBG("%s ep %p\n", __func__, ep); 167 if (timer_pending(&ep->timer)) { 168 pr_err("%s timer already started! ep %p\n", 169 __func__, ep); 170 return; 171 } 172 clear_bit(TIMEOUT, &ep->com.flags); 173 c4iw_get_ep(&ep->com); 174 ep->timer.expires = jiffies + ep_timeout_secs * HZ; 175 ep->timer.data = (unsigned long)ep; 176 ep->timer.function = ep_timeout; 177 add_timer(&ep->timer); 178 } 179 180 static int stop_ep_timer(struct c4iw_ep *ep) 181 { 182 PDBG("%s ep %p stopping\n", __func__, ep); 183 del_timer_sync(&ep->timer); 184 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 185 c4iw_put_ep(&ep->com); 186 return 0; 187 } 188 return 1; 189 } 190 191 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb, 192 struct l2t_entry *l2e) 193 { 194 int error = 0; 195 196 if (c4iw_fatal_error(rdev)) { 197 kfree_skb(skb); 198 PDBG("%s - device in error state - dropping\n", __func__); 199 return -EIO; 200 } 201 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e); 202 if (error < 0) 203 kfree_skb(skb); 204 return error < 0 ? error : 0; 205 } 206 207 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb) 208 { 209 int error = 0; 210 211 if (c4iw_fatal_error(rdev)) { 212 kfree_skb(skb); 213 PDBG("%s - device in error state - dropping\n", __func__); 214 return -EIO; 215 } 216 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb); 217 if (error < 0) 218 kfree_skb(skb); 219 return error < 0 ? error : 0; 220 } 221 222 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb) 223 { 224 struct cpl_tid_release *req; 225 226 skb = get_skb(skb, sizeof *req, GFP_KERNEL); 227 if (!skb) 228 return; 229 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req)); 230 INIT_TP_WR(req, hwtid); 231 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid)); 232 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0); 233 c4iw_ofld_send(rdev, skb); 234 return; 235 } 236 237 static void set_emss(struct c4iw_ep *ep, u16 opt) 238 { 239 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] - 240 ((AF_INET == ep->com.remote_addr.ss_family) ? 241 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) - 242 sizeof(struct tcphdr); 243 ep->mss = ep->emss; 244 if (TCPOPT_TSTAMP_G(opt)) 245 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4); 246 if (ep->emss < 128) 247 ep->emss = 128; 248 if (ep->emss & 7) 249 PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n", 250 TCPOPT_MSS_G(opt), ep->mss, ep->emss); 251 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt), 252 ep->mss, ep->emss); 253 } 254 255 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc) 256 { 257 enum c4iw_ep_state state; 258 259 mutex_lock(&epc->mutex); 260 state = epc->state; 261 mutex_unlock(&epc->mutex); 262 return state; 263 } 264 265 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new) 266 { 267 epc->state = new; 268 } 269 270 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new) 271 { 272 mutex_lock(&epc->mutex); 273 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]); 274 __state_set(epc, new); 275 mutex_unlock(&epc->mutex); 276 return; 277 } 278 279 static void *alloc_ep(int size, gfp_t gfp) 280 { 281 struct c4iw_ep_common *epc; 282 283 epc = kzalloc(size, gfp); 284 if (epc) { 285 kref_init(&epc->kref); 286 mutex_init(&epc->mutex); 287 c4iw_init_wr_wait(&epc->wr_wait); 288 } 289 PDBG("%s alloc ep %p\n", __func__, epc); 290 return epc; 291 } 292 293 void _c4iw_free_ep(struct kref *kref) 294 { 295 struct c4iw_ep *ep; 296 297 ep = container_of(kref, struct c4iw_ep, com.kref); 298 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]); 299 if (test_bit(QP_REFERENCED, &ep->com.flags)) 300 deref_qp(ep); 301 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) { 302 if (ep->com.remote_addr.ss_family == AF_INET6) { 303 struct sockaddr_in6 *sin6 = 304 (struct sockaddr_in6 *) 305 &ep->com.mapped_local_addr; 306 307 cxgb4_clip_release( 308 ep->com.dev->rdev.lldi.ports[0], 309 (const u32 *)&sin6->sin6_addr.s6_addr, 310 1); 311 } 312 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid); 313 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid); 314 dst_release(ep->dst); 315 cxgb4_l2t_release(ep->l2t); 316 } 317 if (test_bit(RELEASE_MAPINFO, &ep->com.flags)) { 318 print_addr(&ep->com, __func__, "remove_mapinfo/mapping"); 319 iwpm_remove_mapinfo(&ep->com.local_addr, 320 &ep->com.mapped_local_addr); 321 iwpm_remove_mapping(&ep->com.local_addr, RDMA_NL_C4IW); 322 } 323 kfree(ep); 324 } 325 326 static void release_ep_resources(struct c4iw_ep *ep) 327 { 328 set_bit(RELEASE_RESOURCES, &ep->com.flags); 329 c4iw_put_ep(&ep->com); 330 } 331 332 static int status2errno(int status) 333 { 334 switch (status) { 335 case CPL_ERR_NONE: 336 return 0; 337 case CPL_ERR_CONN_RESET: 338 return -ECONNRESET; 339 case CPL_ERR_ARP_MISS: 340 return -EHOSTUNREACH; 341 case CPL_ERR_CONN_TIMEDOUT: 342 return -ETIMEDOUT; 343 case CPL_ERR_TCAM_FULL: 344 return -ENOMEM; 345 case CPL_ERR_CONN_EXIST: 346 return -EADDRINUSE; 347 default: 348 return -EIO; 349 } 350 } 351 352 /* 353 * Try and reuse skbs already allocated... 354 */ 355 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp) 356 { 357 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) { 358 skb_trim(skb, 0); 359 skb_get(skb); 360 skb_reset_transport_header(skb); 361 } else { 362 skb = alloc_skb(len, gfp); 363 } 364 t4_set_arp_err_handler(skb, NULL, NULL); 365 return skb; 366 } 367 368 static struct net_device *get_real_dev(struct net_device *egress_dev) 369 { 370 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev; 371 } 372 373 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev) 374 { 375 int i; 376 377 egress_dev = get_real_dev(egress_dev); 378 for (i = 0; i < dev->rdev.lldi.nports; i++) 379 if (dev->rdev.lldi.ports[i] == egress_dev) 380 return 1; 381 return 0; 382 } 383 384 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip, 385 __u8 *peer_ip, __be16 local_port, 386 __be16 peer_port, u8 tos, 387 __u32 sin6_scope_id) 388 { 389 struct dst_entry *dst = NULL; 390 391 if (IS_ENABLED(CONFIG_IPV6)) { 392 struct flowi6 fl6; 393 394 memset(&fl6, 0, sizeof(fl6)); 395 memcpy(&fl6.daddr, peer_ip, 16); 396 memcpy(&fl6.saddr, local_ip, 16); 397 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL) 398 fl6.flowi6_oif = sin6_scope_id; 399 dst = ip6_route_output(&init_net, NULL, &fl6); 400 if (!dst) 401 goto out; 402 if (!our_interface(dev, ip6_dst_idev(dst)->dev) && 403 !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) { 404 dst_release(dst); 405 dst = NULL; 406 } 407 } 408 409 out: 410 return dst; 411 } 412 413 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip, 414 __be32 peer_ip, __be16 local_port, 415 __be16 peer_port, u8 tos) 416 { 417 struct rtable *rt; 418 struct flowi4 fl4; 419 struct neighbour *n; 420 421 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip, 422 peer_port, local_port, IPPROTO_TCP, 423 tos, 0); 424 if (IS_ERR(rt)) 425 return NULL; 426 n = dst_neigh_lookup(&rt->dst, &peer_ip); 427 if (!n) 428 return NULL; 429 if (!our_interface(dev, n->dev) && 430 !(n->dev->flags & IFF_LOOPBACK)) { 431 neigh_release(n); 432 dst_release(&rt->dst); 433 return NULL; 434 } 435 neigh_release(n); 436 return &rt->dst; 437 } 438 439 static void arp_failure_discard(void *handle, struct sk_buff *skb) 440 { 441 PDBG("%s c4iw_dev %p\n", __func__, handle); 442 kfree_skb(skb); 443 } 444 445 /* 446 * Handle an ARP failure for an active open. 447 */ 448 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb) 449 { 450 struct c4iw_ep *ep = handle; 451 452 printk(KERN_ERR MOD "ARP failure duing connect\n"); 453 kfree_skb(skb); 454 connect_reply_upcall(ep, -EHOSTUNREACH); 455 state_set(&ep->com, DEAD); 456 if (ep->com.remote_addr.ss_family == AF_INET6) { 457 struct sockaddr_in6 *sin6 = 458 (struct sockaddr_in6 *)&ep->com.mapped_local_addr; 459 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 460 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 461 } 462 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid); 463 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid); 464 dst_release(ep->dst); 465 cxgb4_l2t_release(ep->l2t); 466 c4iw_put_ep(&ep->com); 467 } 468 469 /* 470 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant 471 * and send it along. 472 */ 473 static void abort_arp_failure(void *handle, struct sk_buff *skb) 474 { 475 struct c4iw_rdev *rdev = handle; 476 struct cpl_abort_req *req = cplhdr(skb); 477 478 PDBG("%s rdev %p\n", __func__, rdev); 479 req->cmd = CPL_ABORT_NO_RST; 480 c4iw_ofld_send(rdev, skb); 481 } 482 483 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb) 484 { 485 unsigned int flowclen = 80; 486 struct fw_flowc_wr *flowc; 487 int i; 488 489 skb = get_skb(skb, flowclen, GFP_KERNEL); 490 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen); 491 492 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) | 493 FW_FLOWC_WR_NPARAMS_V(8)); 494 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(flowclen, 495 16)) | FW_WR_FLOWID_V(ep->hwtid)); 496 497 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN; 498 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V 499 (ep->com.dev->rdev.lldi.pf)); 500 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH; 501 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan); 502 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT; 503 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan); 504 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID; 505 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid); 506 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT; 507 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq); 508 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT; 509 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq); 510 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF; 511 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win); 512 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS; 513 flowc->mnemval[7].val = cpu_to_be32(ep->emss); 514 /* Pad WR to 16 byte boundary */ 515 flowc->mnemval[8].mnemonic = 0; 516 flowc->mnemval[8].val = 0; 517 for (i = 0; i < 9; i++) { 518 flowc->mnemval[i].r4[0] = 0; 519 flowc->mnemval[i].r4[1] = 0; 520 flowc->mnemval[i].r4[2] = 0; 521 } 522 523 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 524 c4iw_ofld_send(&ep->com.dev->rdev, skb); 525 } 526 527 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp) 528 { 529 struct cpl_close_con_req *req; 530 struct sk_buff *skb; 531 int wrlen = roundup(sizeof *req, 16); 532 533 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 534 skb = get_skb(NULL, wrlen, gfp); 535 if (!skb) { 536 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__); 537 return -ENOMEM; 538 } 539 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 540 t4_set_arp_err_handler(skb, NULL, arp_failure_discard); 541 req = (struct cpl_close_con_req *) skb_put(skb, wrlen); 542 memset(req, 0, wrlen); 543 INIT_TP_WR(req, ep->hwtid); 544 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, 545 ep->hwtid)); 546 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 547 } 548 549 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp) 550 { 551 struct cpl_abort_req *req; 552 int wrlen = roundup(sizeof *req, 16); 553 554 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 555 skb = get_skb(skb, wrlen, gfp); 556 if (!skb) { 557 printk(KERN_ERR MOD "%s - failed to alloc skb.\n", 558 __func__); 559 return -ENOMEM; 560 } 561 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 562 t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure); 563 req = (struct cpl_abort_req *) skb_put(skb, wrlen); 564 memset(req, 0, wrlen); 565 INIT_TP_WR(req, ep->hwtid); 566 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid)); 567 req->cmd = CPL_ABORT_SEND_RST; 568 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 569 } 570 571 /* 572 * c4iw_form_pm_msg - Form a port mapper message with mapping info 573 */ 574 static void c4iw_form_pm_msg(struct c4iw_ep *ep, 575 struct iwpm_sa_data *pm_msg) 576 { 577 memcpy(&pm_msg->loc_addr, &ep->com.local_addr, 578 sizeof(ep->com.local_addr)); 579 memcpy(&pm_msg->rem_addr, &ep->com.remote_addr, 580 sizeof(ep->com.remote_addr)); 581 } 582 583 /* 584 * c4iw_form_reg_msg - Form a port mapper message with dev info 585 */ 586 static void c4iw_form_reg_msg(struct c4iw_dev *dev, 587 struct iwpm_dev_data *pm_msg) 588 { 589 memcpy(pm_msg->dev_name, dev->ibdev.name, IWPM_DEVNAME_SIZE); 590 memcpy(pm_msg->if_name, dev->rdev.lldi.ports[0]->name, 591 IWPM_IFNAME_SIZE); 592 } 593 594 static void c4iw_record_pm_msg(struct c4iw_ep *ep, 595 struct iwpm_sa_data *pm_msg) 596 { 597 memcpy(&ep->com.mapped_local_addr, &pm_msg->mapped_loc_addr, 598 sizeof(ep->com.mapped_local_addr)); 599 memcpy(&ep->com.mapped_remote_addr, &pm_msg->mapped_rem_addr, 600 sizeof(ep->com.mapped_remote_addr)); 601 } 602 603 static int get_remote_addr(struct c4iw_ep *parent_ep, struct c4iw_ep *child_ep) 604 { 605 int ret; 606 607 print_addr(&parent_ep->com, __func__, "get_remote_addr parent_ep "); 608 print_addr(&child_ep->com, __func__, "get_remote_addr child_ep "); 609 610 ret = iwpm_get_remote_info(&parent_ep->com.mapped_local_addr, 611 &child_ep->com.mapped_remote_addr, 612 &child_ep->com.remote_addr, RDMA_NL_C4IW); 613 if (ret) 614 PDBG("Unable to find remote peer addr info - err %d\n", ret); 615 616 return ret; 617 } 618 619 static void best_mtu(const unsigned short *mtus, unsigned short mtu, 620 unsigned int *idx, int use_ts, int ipv6) 621 { 622 unsigned short hdr_size = (ipv6 ? 623 sizeof(struct ipv6hdr) : 624 sizeof(struct iphdr)) + 625 sizeof(struct tcphdr) + 626 (use_ts ? 627 round_up(TCPOLEN_TIMESTAMP, 4) : 0); 628 unsigned short data_size = mtu - hdr_size; 629 630 cxgb4_best_aligned_mtu(mtus, hdr_size, data_size, 8, idx); 631 } 632 633 static int send_connect(struct c4iw_ep *ep) 634 { 635 struct cpl_act_open_req *req; 636 struct cpl_t5_act_open_req *t5_req; 637 struct cpl_act_open_req6 *req6; 638 struct cpl_t5_act_open_req6 *t5_req6; 639 struct sk_buff *skb; 640 u64 opt0; 641 u32 opt2; 642 unsigned int mtu_idx; 643 int wscale; 644 int wrlen; 645 int sizev4 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ? 646 sizeof(struct cpl_act_open_req) : 647 sizeof(struct cpl_t5_act_open_req); 648 int sizev6 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ? 649 sizeof(struct cpl_act_open_req6) : 650 sizeof(struct cpl_t5_act_open_req6); 651 struct sockaddr_in *la = (struct sockaddr_in *) 652 &ep->com.mapped_local_addr; 653 struct sockaddr_in *ra = (struct sockaddr_in *) 654 &ep->com.mapped_remote_addr; 655 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *) 656 &ep->com.mapped_local_addr; 657 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *) 658 &ep->com.mapped_remote_addr; 659 int win; 660 int ret; 661 662 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ? 663 roundup(sizev4, 16) : 664 roundup(sizev6, 16); 665 666 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid); 667 668 skb = get_skb(NULL, wrlen, GFP_KERNEL); 669 if (!skb) { 670 printk(KERN_ERR MOD "%s - failed to alloc skb.\n", 671 __func__); 672 return -ENOMEM; 673 } 674 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx); 675 676 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx, 677 enable_tcp_timestamps, 678 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1); 679 wscale = compute_wscale(rcv_win); 680 681 /* 682 * Specify the largest window that will fit in opt0. The 683 * remainder will be specified in the rx_data_ack. 684 */ 685 win = ep->rcv_win >> 10; 686 if (win > RCV_BUFSIZ_M) 687 win = RCV_BUFSIZ_M; 688 689 opt0 = (nocong ? NO_CONG_F : 0) | 690 KEEP_ALIVE_F | 691 DELACK_F | 692 WND_SCALE_V(wscale) | 693 MSS_IDX_V(mtu_idx) | 694 L2T_IDX_V(ep->l2t->idx) | 695 TX_CHAN_V(ep->tx_chan) | 696 SMAC_SEL_V(ep->smac_idx) | 697 DSCP_V(ep->tos) | 698 ULP_MODE_V(ULP_MODE_TCPDDP) | 699 RCV_BUFSIZ_V(win); 700 opt2 = RX_CHANNEL_V(0) | 701 CCTRL_ECN_V(enable_ecn) | 702 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid); 703 if (enable_tcp_timestamps) 704 opt2 |= TSTAMPS_EN_F; 705 if (enable_tcp_sack) 706 opt2 |= SACK_EN_F; 707 if (wscale && enable_tcp_window_scaling) 708 opt2 |= WND_SCALE_EN_F; 709 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) { 710 opt2 |= T5_OPT_2_VALID_F; 711 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE); 712 opt2 |= T5_ISS_F; 713 } 714 715 if (ep->com.remote_addr.ss_family == AF_INET6) 716 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0], 717 (const u32 *)&la6->sin6_addr.s6_addr, 1); 718 719 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure); 720 721 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) { 722 if (ep->com.remote_addr.ss_family == AF_INET) { 723 req = (struct cpl_act_open_req *) skb_put(skb, wrlen); 724 INIT_TP_WR(req, 0); 725 OPCODE_TID(req) = cpu_to_be32( 726 MK_OPCODE_TID(CPL_ACT_OPEN_REQ, 727 ((ep->rss_qid << 14) | ep->atid))); 728 req->local_port = la->sin_port; 729 req->peer_port = ra->sin_port; 730 req->local_ip = la->sin_addr.s_addr; 731 req->peer_ip = ra->sin_addr.s_addr; 732 req->opt0 = cpu_to_be64(opt0); 733 req->params = cpu_to_be32(cxgb4_select_ntuple( 734 ep->com.dev->rdev.lldi.ports[0], 735 ep->l2t)); 736 req->opt2 = cpu_to_be32(opt2); 737 } else { 738 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen); 739 740 INIT_TP_WR(req6, 0); 741 OPCODE_TID(req6) = cpu_to_be32( 742 MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, 743 ((ep->rss_qid<<14)|ep->atid))); 744 req6->local_port = la6->sin6_port; 745 req6->peer_port = ra6->sin6_port; 746 req6->local_ip_hi = *((__be64 *) 747 (la6->sin6_addr.s6_addr)); 748 req6->local_ip_lo = *((__be64 *) 749 (la6->sin6_addr.s6_addr + 8)); 750 req6->peer_ip_hi = *((__be64 *) 751 (ra6->sin6_addr.s6_addr)); 752 req6->peer_ip_lo = *((__be64 *) 753 (ra6->sin6_addr.s6_addr + 8)); 754 req6->opt0 = cpu_to_be64(opt0); 755 req6->params = cpu_to_be32(cxgb4_select_ntuple( 756 ep->com.dev->rdev.lldi.ports[0], 757 ep->l2t)); 758 req6->opt2 = cpu_to_be32(opt2); 759 } 760 } else { 761 u32 isn = (prandom_u32() & ~7UL) - 1; 762 763 if (peer2peer) 764 isn += 4; 765 766 if (ep->com.remote_addr.ss_family == AF_INET) { 767 t5_req = (struct cpl_t5_act_open_req *) 768 skb_put(skb, wrlen); 769 INIT_TP_WR(t5_req, 0); 770 OPCODE_TID(t5_req) = cpu_to_be32( 771 MK_OPCODE_TID(CPL_ACT_OPEN_REQ, 772 ((ep->rss_qid << 14) | ep->atid))); 773 t5_req->local_port = la->sin_port; 774 t5_req->peer_port = ra->sin_port; 775 t5_req->local_ip = la->sin_addr.s_addr; 776 t5_req->peer_ip = ra->sin_addr.s_addr; 777 t5_req->opt0 = cpu_to_be64(opt0); 778 t5_req->params = cpu_to_be64(FILTER_TUPLE_V( 779 cxgb4_select_ntuple( 780 ep->com.dev->rdev.lldi.ports[0], 781 ep->l2t))); 782 t5_req->rsvd = cpu_to_be32(isn); 783 PDBG("%s snd_isn %u\n", __func__, 784 be32_to_cpu(t5_req->rsvd)); 785 t5_req->opt2 = cpu_to_be32(opt2); 786 } else { 787 t5_req6 = (struct cpl_t5_act_open_req6 *) 788 skb_put(skb, wrlen); 789 INIT_TP_WR(t5_req6, 0); 790 OPCODE_TID(t5_req6) = cpu_to_be32( 791 MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, 792 ((ep->rss_qid<<14)|ep->atid))); 793 t5_req6->local_port = la6->sin6_port; 794 t5_req6->peer_port = ra6->sin6_port; 795 t5_req6->local_ip_hi = *((__be64 *) 796 (la6->sin6_addr.s6_addr)); 797 t5_req6->local_ip_lo = *((__be64 *) 798 (la6->sin6_addr.s6_addr + 8)); 799 t5_req6->peer_ip_hi = *((__be64 *) 800 (ra6->sin6_addr.s6_addr)); 801 t5_req6->peer_ip_lo = *((__be64 *) 802 (ra6->sin6_addr.s6_addr + 8)); 803 t5_req6->opt0 = cpu_to_be64(opt0); 804 t5_req6->params = cpu_to_be64(FILTER_TUPLE_V( 805 cxgb4_select_ntuple( 806 ep->com.dev->rdev.lldi.ports[0], 807 ep->l2t))); 808 t5_req6->rsvd = cpu_to_be32(isn); 809 PDBG("%s snd_isn %u\n", __func__, 810 be32_to_cpu(t5_req6->rsvd)); 811 t5_req6->opt2 = cpu_to_be32(opt2); 812 } 813 } 814 815 set_bit(ACT_OPEN_REQ, &ep->com.history); 816 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 817 if (ret && ep->com.remote_addr.ss_family == AF_INET6) 818 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 819 (const u32 *)&la6->sin6_addr.s6_addr, 1); 820 return ret; 821 } 822 823 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb, 824 u8 mpa_rev_to_use) 825 { 826 int mpalen, wrlen; 827 struct fw_ofld_tx_data_wr *req; 828 struct mpa_message *mpa; 829 struct mpa_v2_conn_params mpa_v2_params; 830 831 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen); 832 833 BUG_ON(skb_cloned(skb)); 834 835 mpalen = sizeof(*mpa) + ep->plen; 836 if (mpa_rev_to_use == 2) 837 mpalen += sizeof(struct mpa_v2_conn_params); 838 wrlen = roundup(mpalen + sizeof *req, 16); 839 skb = get_skb(skb, wrlen, GFP_KERNEL); 840 if (!skb) { 841 connect_reply_upcall(ep, -ENOMEM); 842 return; 843 } 844 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 845 846 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen); 847 memset(req, 0, wrlen); 848 req->op_to_immdlen = cpu_to_be32( 849 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) | 850 FW_WR_COMPL_F | 851 FW_WR_IMMDLEN_V(mpalen)); 852 req->flowid_len16 = cpu_to_be32( 853 FW_WR_FLOWID_V(ep->hwtid) | 854 FW_WR_LEN16_V(wrlen >> 4)); 855 req->plen = cpu_to_be32(mpalen); 856 req->tunnel_to_proxy = cpu_to_be32( 857 FW_OFLD_TX_DATA_WR_FLUSH_F | 858 FW_OFLD_TX_DATA_WR_SHOVE_F); 859 860 mpa = (struct mpa_message *)(req + 1); 861 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)); 862 mpa->flags = (crc_enabled ? MPA_CRC : 0) | 863 (markers_enabled ? MPA_MARKERS : 0) | 864 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0); 865 mpa->private_data_size = htons(ep->plen); 866 mpa->revision = mpa_rev_to_use; 867 if (mpa_rev_to_use == 1) { 868 ep->tried_with_mpa_v1 = 1; 869 ep->retry_with_mpa_v1 = 0; 870 } 871 872 if (mpa_rev_to_use == 2) { 873 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 874 sizeof (struct mpa_v2_conn_params)); 875 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird, 876 ep->ord); 877 mpa_v2_params.ird = htons((u16)ep->ird); 878 mpa_v2_params.ord = htons((u16)ep->ord); 879 880 if (peer2peer) { 881 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 882 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) 883 mpa_v2_params.ord |= 884 htons(MPA_V2_RDMA_WRITE_RTR); 885 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) 886 mpa_v2_params.ord |= 887 htons(MPA_V2_RDMA_READ_RTR); 888 } 889 memcpy(mpa->private_data, &mpa_v2_params, 890 sizeof(struct mpa_v2_conn_params)); 891 892 if (ep->plen) 893 memcpy(mpa->private_data + 894 sizeof(struct mpa_v2_conn_params), 895 ep->mpa_pkt + sizeof(*mpa), ep->plen); 896 } else 897 if (ep->plen) 898 memcpy(mpa->private_data, 899 ep->mpa_pkt + sizeof(*mpa), ep->plen); 900 901 /* 902 * Reference the mpa skb. This ensures the data area 903 * will remain in memory until the hw acks the tx. 904 * Function fw4_ack() will deref it. 905 */ 906 skb_get(skb); 907 t4_set_arp_err_handler(skb, NULL, arp_failure_discard); 908 BUG_ON(ep->mpa_skb); 909 ep->mpa_skb = skb; 910 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 911 start_ep_timer(ep); 912 __state_set(&ep->com, MPA_REQ_SENT); 913 ep->mpa_attr.initiator = 1; 914 ep->snd_seq += mpalen; 915 return; 916 } 917 918 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen) 919 { 920 int mpalen, wrlen; 921 struct fw_ofld_tx_data_wr *req; 922 struct mpa_message *mpa; 923 struct sk_buff *skb; 924 struct mpa_v2_conn_params mpa_v2_params; 925 926 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen); 927 928 mpalen = sizeof(*mpa) + plen; 929 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) 930 mpalen += sizeof(struct mpa_v2_conn_params); 931 wrlen = roundup(mpalen + sizeof *req, 16); 932 933 skb = get_skb(NULL, wrlen, GFP_KERNEL); 934 if (!skb) { 935 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__); 936 return -ENOMEM; 937 } 938 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 939 940 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen); 941 memset(req, 0, wrlen); 942 req->op_to_immdlen = cpu_to_be32( 943 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) | 944 FW_WR_COMPL_F | 945 FW_WR_IMMDLEN_V(mpalen)); 946 req->flowid_len16 = cpu_to_be32( 947 FW_WR_FLOWID_V(ep->hwtid) | 948 FW_WR_LEN16_V(wrlen >> 4)); 949 req->plen = cpu_to_be32(mpalen); 950 req->tunnel_to_proxy = cpu_to_be32( 951 FW_OFLD_TX_DATA_WR_FLUSH_F | 952 FW_OFLD_TX_DATA_WR_SHOVE_F); 953 954 mpa = (struct mpa_message *)(req + 1); 955 memset(mpa, 0, sizeof(*mpa)); 956 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 957 mpa->flags = MPA_REJECT; 958 mpa->revision = ep->mpa_attr.version; 959 mpa->private_data_size = htons(plen); 960 961 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 962 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 963 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 964 sizeof (struct mpa_v2_conn_params)); 965 mpa_v2_params.ird = htons(((u16)ep->ird) | 966 (peer2peer ? MPA_V2_PEER2PEER_MODEL : 967 0)); 968 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ? 969 (p2p_type == 970 FW_RI_INIT_P2PTYPE_RDMA_WRITE ? 971 MPA_V2_RDMA_WRITE_RTR : p2p_type == 972 FW_RI_INIT_P2PTYPE_READ_REQ ? 973 MPA_V2_RDMA_READ_RTR : 0) : 0)); 974 memcpy(mpa->private_data, &mpa_v2_params, 975 sizeof(struct mpa_v2_conn_params)); 976 977 if (ep->plen) 978 memcpy(mpa->private_data + 979 sizeof(struct mpa_v2_conn_params), pdata, plen); 980 } else 981 if (plen) 982 memcpy(mpa->private_data, pdata, plen); 983 984 /* 985 * Reference the mpa skb again. This ensures the data area 986 * will remain in memory until the hw acks the tx. 987 * Function fw4_ack() will deref it. 988 */ 989 skb_get(skb); 990 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 991 t4_set_arp_err_handler(skb, NULL, arp_failure_discard); 992 BUG_ON(ep->mpa_skb); 993 ep->mpa_skb = skb; 994 ep->snd_seq += mpalen; 995 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 996 } 997 998 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen) 999 { 1000 int mpalen, wrlen; 1001 struct fw_ofld_tx_data_wr *req; 1002 struct mpa_message *mpa; 1003 struct sk_buff *skb; 1004 struct mpa_v2_conn_params mpa_v2_params; 1005 1006 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen); 1007 1008 mpalen = sizeof(*mpa) + plen; 1009 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) 1010 mpalen += sizeof(struct mpa_v2_conn_params); 1011 wrlen = roundup(mpalen + sizeof *req, 16); 1012 1013 skb = get_skb(NULL, wrlen, GFP_KERNEL); 1014 if (!skb) { 1015 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__); 1016 return -ENOMEM; 1017 } 1018 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 1019 1020 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen); 1021 memset(req, 0, wrlen); 1022 req->op_to_immdlen = cpu_to_be32( 1023 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) | 1024 FW_WR_COMPL_F | 1025 FW_WR_IMMDLEN_V(mpalen)); 1026 req->flowid_len16 = cpu_to_be32( 1027 FW_WR_FLOWID_V(ep->hwtid) | 1028 FW_WR_LEN16_V(wrlen >> 4)); 1029 req->plen = cpu_to_be32(mpalen); 1030 req->tunnel_to_proxy = cpu_to_be32( 1031 FW_OFLD_TX_DATA_WR_FLUSH_F | 1032 FW_OFLD_TX_DATA_WR_SHOVE_F); 1033 1034 mpa = (struct mpa_message *)(req + 1); 1035 memset(mpa, 0, sizeof(*mpa)); 1036 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1037 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) | 1038 (markers_enabled ? MPA_MARKERS : 0); 1039 mpa->revision = ep->mpa_attr.version; 1040 mpa->private_data_size = htons(plen); 1041 1042 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1043 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1044 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1045 sizeof (struct mpa_v2_conn_params)); 1046 mpa_v2_params.ird = htons((u16)ep->ird); 1047 mpa_v2_params.ord = htons((u16)ep->ord); 1048 if (peer2peer && (ep->mpa_attr.p2p_type != 1049 FW_RI_INIT_P2PTYPE_DISABLED)) { 1050 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1051 1052 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) 1053 mpa_v2_params.ord |= 1054 htons(MPA_V2_RDMA_WRITE_RTR); 1055 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) 1056 mpa_v2_params.ord |= 1057 htons(MPA_V2_RDMA_READ_RTR); 1058 } 1059 1060 memcpy(mpa->private_data, &mpa_v2_params, 1061 sizeof(struct mpa_v2_conn_params)); 1062 1063 if (ep->plen) 1064 memcpy(mpa->private_data + 1065 sizeof(struct mpa_v2_conn_params), pdata, plen); 1066 } else 1067 if (plen) 1068 memcpy(mpa->private_data, pdata, plen); 1069 1070 /* 1071 * Reference the mpa skb. This ensures the data area 1072 * will remain in memory until the hw acks the tx. 1073 * Function fw4_ack() will deref it. 1074 */ 1075 skb_get(skb); 1076 t4_set_arp_err_handler(skb, NULL, arp_failure_discard); 1077 ep->mpa_skb = skb; 1078 __state_set(&ep->com, MPA_REP_SENT); 1079 ep->snd_seq += mpalen; 1080 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1081 } 1082 1083 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb) 1084 { 1085 struct c4iw_ep *ep; 1086 struct cpl_act_establish *req = cplhdr(skb); 1087 unsigned int tid = GET_TID(req); 1088 unsigned int atid = TID_TID_G(ntohl(req->tos_atid)); 1089 struct tid_info *t = dev->rdev.lldi.tids; 1090 1091 ep = lookup_atid(t, atid); 1092 1093 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid, 1094 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn)); 1095 1096 mutex_lock(&ep->com.mutex); 1097 dst_confirm(ep->dst); 1098 1099 /* setup the hwtid for this connection */ 1100 ep->hwtid = tid; 1101 cxgb4_insert_tid(t, ep, tid); 1102 insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid); 1103 1104 ep->snd_seq = be32_to_cpu(req->snd_isn); 1105 ep->rcv_seq = be32_to_cpu(req->rcv_isn); 1106 1107 set_emss(ep, ntohs(req->tcp_opt)); 1108 1109 /* dealloc the atid */ 1110 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid); 1111 cxgb4_free_atid(t, atid); 1112 set_bit(ACT_ESTAB, &ep->com.history); 1113 1114 /* start MPA negotiation */ 1115 send_flowc(ep, NULL); 1116 if (ep->retry_with_mpa_v1) 1117 send_mpa_req(ep, skb, 1); 1118 else 1119 send_mpa_req(ep, skb, mpa_rev); 1120 mutex_unlock(&ep->com.mutex); 1121 return 0; 1122 } 1123 1124 static void close_complete_upcall(struct c4iw_ep *ep, int status) 1125 { 1126 struct iw_cm_event event; 1127 1128 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1129 memset(&event, 0, sizeof(event)); 1130 event.event = IW_CM_EVENT_CLOSE; 1131 event.status = status; 1132 if (ep->com.cm_id) { 1133 PDBG("close complete delivered ep %p cm_id %p tid %u\n", 1134 ep, ep->com.cm_id, ep->hwtid); 1135 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1136 ep->com.cm_id->rem_ref(ep->com.cm_id); 1137 ep->com.cm_id = NULL; 1138 set_bit(CLOSE_UPCALL, &ep->com.history); 1139 } 1140 } 1141 1142 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp) 1143 { 1144 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1145 __state_set(&ep->com, ABORTING); 1146 set_bit(ABORT_CONN, &ep->com.history); 1147 return send_abort(ep, skb, gfp); 1148 } 1149 1150 static void peer_close_upcall(struct c4iw_ep *ep) 1151 { 1152 struct iw_cm_event event; 1153 1154 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1155 memset(&event, 0, sizeof(event)); 1156 event.event = IW_CM_EVENT_DISCONNECT; 1157 if (ep->com.cm_id) { 1158 PDBG("peer close delivered ep %p cm_id %p tid %u\n", 1159 ep, ep->com.cm_id, ep->hwtid); 1160 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1161 set_bit(DISCONN_UPCALL, &ep->com.history); 1162 } 1163 } 1164 1165 static void peer_abort_upcall(struct c4iw_ep *ep) 1166 { 1167 struct iw_cm_event event; 1168 1169 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1170 memset(&event, 0, sizeof(event)); 1171 event.event = IW_CM_EVENT_CLOSE; 1172 event.status = -ECONNRESET; 1173 if (ep->com.cm_id) { 1174 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep, 1175 ep->com.cm_id, ep->hwtid); 1176 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1177 ep->com.cm_id->rem_ref(ep->com.cm_id); 1178 ep->com.cm_id = NULL; 1179 set_bit(ABORT_UPCALL, &ep->com.history); 1180 } 1181 } 1182 1183 static void connect_reply_upcall(struct c4iw_ep *ep, int status) 1184 { 1185 struct iw_cm_event event; 1186 1187 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status); 1188 memset(&event, 0, sizeof(event)); 1189 event.event = IW_CM_EVENT_CONNECT_REPLY; 1190 event.status = status; 1191 memcpy(&event.local_addr, &ep->com.local_addr, 1192 sizeof(ep->com.local_addr)); 1193 memcpy(&event.remote_addr, &ep->com.remote_addr, 1194 sizeof(ep->com.remote_addr)); 1195 1196 if ((status == 0) || (status == -ECONNREFUSED)) { 1197 if (!ep->tried_with_mpa_v1) { 1198 /* this means MPA_v2 is used */ 1199 event.private_data_len = ep->plen - 1200 sizeof(struct mpa_v2_conn_params); 1201 event.private_data = ep->mpa_pkt + 1202 sizeof(struct mpa_message) + 1203 sizeof(struct mpa_v2_conn_params); 1204 } else { 1205 /* this means MPA_v1 is used */ 1206 event.private_data_len = ep->plen; 1207 event.private_data = ep->mpa_pkt + 1208 sizeof(struct mpa_message); 1209 } 1210 } 1211 1212 PDBG("%s ep %p tid %u status %d\n", __func__, ep, 1213 ep->hwtid, status); 1214 set_bit(CONN_RPL_UPCALL, &ep->com.history); 1215 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1216 1217 if (status < 0) { 1218 ep->com.cm_id->rem_ref(ep->com.cm_id); 1219 ep->com.cm_id = NULL; 1220 } 1221 } 1222 1223 static int connect_request_upcall(struct c4iw_ep *ep) 1224 { 1225 struct iw_cm_event event; 1226 int ret; 1227 1228 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1229 memset(&event, 0, sizeof(event)); 1230 event.event = IW_CM_EVENT_CONNECT_REQUEST; 1231 memcpy(&event.local_addr, &ep->com.local_addr, 1232 sizeof(ep->com.local_addr)); 1233 memcpy(&event.remote_addr, &ep->com.remote_addr, 1234 sizeof(ep->com.remote_addr)); 1235 event.provider_data = ep; 1236 if (!ep->tried_with_mpa_v1) { 1237 /* this means MPA_v2 is used */ 1238 event.ord = ep->ord; 1239 event.ird = ep->ird; 1240 event.private_data_len = ep->plen - 1241 sizeof(struct mpa_v2_conn_params); 1242 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) + 1243 sizeof(struct mpa_v2_conn_params); 1244 } else { 1245 /* this means MPA_v1 is used. Send max supported */ 1246 event.ord = cur_max_read_depth(ep->com.dev); 1247 event.ird = cur_max_read_depth(ep->com.dev); 1248 event.private_data_len = ep->plen; 1249 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); 1250 } 1251 c4iw_get_ep(&ep->com); 1252 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id, 1253 &event); 1254 if (ret) 1255 c4iw_put_ep(&ep->com); 1256 set_bit(CONNREQ_UPCALL, &ep->com.history); 1257 c4iw_put_ep(&ep->parent_ep->com); 1258 return ret; 1259 } 1260 1261 static void established_upcall(struct c4iw_ep *ep) 1262 { 1263 struct iw_cm_event event; 1264 1265 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1266 memset(&event, 0, sizeof(event)); 1267 event.event = IW_CM_EVENT_ESTABLISHED; 1268 event.ird = ep->ird; 1269 event.ord = ep->ord; 1270 if (ep->com.cm_id) { 1271 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1272 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1273 set_bit(ESTAB_UPCALL, &ep->com.history); 1274 } 1275 } 1276 1277 static int update_rx_credits(struct c4iw_ep *ep, u32 credits) 1278 { 1279 struct cpl_rx_data_ack *req; 1280 struct sk_buff *skb; 1281 int wrlen = roundup(sizeof *req, 16); 1282 1283 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits); 1284 skb = get_skb(NULL, wrlen, GFP_KERNEL); 1285 if (!skb) { 1286 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n"); 1287 return 0; 1288 } 1289 1290 /* 1291 * If we couldn't specify the entire rcv window at connection setup 1292 * due to the limit in the number of bits in the RCV_BUFSIZ field, 1293 * then add the overage in to the credits returned. 1294 */ 1295 if (ep->rcv_win > RCV_BUFSIZ_M * 1024) 1296 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024; 1297 1298 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen); 1299 memset(req, 0, wrlen); 1300 INIT_TP_WR(req, ep->hwtid); 1301 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK, 1302 ep->hwtid)); 1303 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK_F | 1304 RX_DACK_CHANGE_F | 1305 RX_DACK_MODE_V(dack_mode)); 1306 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx); 1307 c4iw_ofld_send(&ep->com.dev->rdev, skb); 1308 return credits; 1309 } 1310 1311 #define RELAXED_IRD_NEGOTIATION 1 1312 1313 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb) 1314 { 1315 struct mpa_message *mpa; 1316 struct mpa_v2_conn_params *mpa_v2_params; 1317 u16 plen; 1318 u16 resp_ird, resp_ord; 1319 u8 rtr_mismatch = 0, insuff_ird = 0; 1320 struct c4iw_qp_attributes attrs; 1321 enum c4iw_qp_attr_mask mask; 1322 int err; 1323 int disconnect = 0; 1324 1325 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1326 1327 /* 1328 * Stop mpa timer. If it expired, then 1329 * we ignore the MPA reply. process_timeout() 1330 * will abort the connection. 1331 */ 1332 if (stop_ep_timer(ep)) 1333 return 0; 1334 1335 /* 1336 * If we get more than the supported amount of private data 1337 * then we must fail this connection. 1338 */ 1339 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) { 1340 err = -EINVAL; 1341 goto err; 1342 } 1343 1344 /* 1345 * copy the new data into our accumulation buffer. 1346 */ 1347 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]), 1348 skb->len); 1349 ep->mpa_pkt_len += skb->len; 1350 1351 /* 1352 * if we don't even have the mpa message, then bail. 1353 */ 1354 if (ep->mpa_pkt_len < sizeof(*mpa)) 1355 return 0; 1356 mpa = (struct mpa_message *) ep->mpa_pkt; 1357 1358 /* Validate MPA header. */ 1359 if (mpa->revision > mpa_rev) { 1360 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d," 1361 " Received = %d\n", __func__, mpa_rev, mpa->revision); 1362 err = -EPROTO; 1363 goto err; 1364 } 1365 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) { 1366 err = -EPROTO; 1367 goto err; 1368 } 1369 1370 plen = ntohs(mpa->private_data_size); 1371 1372 /* 1373 * Fail if there's too much private data. 1374 */ 1375 if (plen > MPA_MAX_PRIVATE_DATA) { 1376 err = -EPROTO; 1377 goto err; 1378 } 1379 1380 /* 1381 * If plen does not account for pkt size 1382 */ 1383 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) { 1384 err = -EPROTO; 1385 goto err; 1386 } 1387 1388 ep->plen = (u8) plen; 1389 1390 /* 1391 * If we don't have all the pdata yet, then bail. 1392 * We'll continue process when more data arrives. 1393 */ 1394 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) 1395 return 0; 1396 1397 if (mpa->flags & MPA_REJECT) { 1398 err = -ECONNREFUSED; 1399 goto err; 1400 } 1401 1402 /* 1403 * If we get here we have accumulated the entire mpa 1404 * start reply message including private data. And 1405 * the MPA header is valid. 1406 */ 1407 __state_set(&ep->com, FPDU_MODE); 1408 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1409 ep->mpa_attr.recv_marker_enabled = markers_enabled; 1410 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1411 ep->mpa_attr.version = mpa->revision; 1412 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1413 1414 if (mpa->revision == 2) { 1415 ep->mpa_attr.enhanced_rdma_conn = 1416 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 1417 if (ep->mpa_attr.enhanced_rdma_conn) { 1418 mpa_v2_params = (struct mpa_v2_conn_params *) 1419 (ep->mpa_pkt + sizeof(*mpa)); 1420 resp_ird = ntohs(mpa_v2_params->ird) & 1421 MPA_V2_IRD_ORD_MASK; 1422 resp_ord = ntohs(mpa_v2_params->ord) & 1423 MPA_V2_IRD_ORD_MASK; 1424 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n", 1425 __func__, resp_ird, resp_ord, ep->ird, ep->ord); 1426 1427 /* 1428 * This is a double-check. Ideally, below checks are 1429 * not required since ird/ord stuff has been taken 1430 * care of in c4iw_accept_cr 1431 */ 1432 if (ep->ird < resp_ord) { 1433 if (RELAXED_IRD_NEGOTIATION && resp_ord <= 1434 ep->com.dev->rdev.lldi.max_ordird_qp) 1435 ep->ird = resp_ord; 1436 else 1437 insuff_ird = 1; 1438 } else if (ep->ird > resp_ord) { 1439 ep->ird = resp_ord; 1440 } 1441 if (ep->ord > resp_ird) { 1442 if (RELAXED_IRD_NEGOTIATION) 1443 ep->ord = resp_ird; 1444 else 1445 insuff_ird = 1; 1446 } 1447 if (insuff_ird) { 1448 err = -ENOMEM; 1449 ep->ird = resp_ord; 1450 ep->ord = resp_ird; 1451 } 1452 1453 if (ntohs(mpa_v2_params->ird) & 1454 MPA_V2_PEER2PEER_MODEL) { 1455 if (ntohs(mpa_v2_params->ord) & 1456 MPA_V2_RDMA_WRITE_RTR) 1457 ep->mpa_attr.p2p_type = 1458 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 1459 else if (ntohs(mpa_v2_params->ord) & 1460 MPA_V2_RDMA_READ_RTR) 1461 ep->mpa_attr.p2p_type = 1462 FW_RI_INIT_P2PTYPE_READ_REQ; 1463 } 1464 } 1465 } else if (mpa->revision == 1) 1466 if (peer2peer) 1467 ep->mpa_attr.p2p_type = p2p_type; 1468 1469 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, " 1470 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = " 1471 "%d\n", __func__, ep->mpa_attr.crc_enabled, 1472 ep->mpa_attr.recv_marker_enabled, 1473 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, 1474 ep->mpa_attr.p2p_type, p2p_type); 1475 1476 /* 1477 * If responder's RTR does not match with that of initiator, assign 1478 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not 1479 * generated when moving QP to RTS state. 1480 * A TERM message will be sent after QP has moved to RTS state 1481 */ 1482 if ((ep->mpa_attr.version == 2) && peer2peer && 1483 (ep->mpa_attr.p2p_type != p2p_type)) { 1484 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1485 rtr_mismatch = 1; 1486 } 1487 1488 attrs.mpa_attr = ep->mpa_attr; 1489 attrs.max_ird = ep->ird; 1490 attrs.max_ord = ep->ord; 1491 attrs.llp_stream_handle = ep; 1492 attrs.next_state = C4IW_QP_STATE_RTS; 1493 1494 mask = C4IW_QP_ATTR_NEXT_STATE | 1495 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR | 1496 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD; 1497 1498 /* bind QP and TID with INIT_WR */ 1499 err = c4iw_modify_qp(ep->com.qp->rhp, 1500 ep->com.qp, mask, &attrs, 1); 1501 if (err) 1502 goto err; 1503 1504 /* 1505 * If responder's RTR requirement did not match with what initiator 1506 * supports, generate TERM message 1507 */ 1508 if (rtr_mismatch) { 1509 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__); 1510 attrs.layer_etype = LAYER_MPA | DDP_LLP; 1511 attrs.ecode = MPA_NOMATCH_RTR; 1512 attrs.next_state = C4IW_QP_STATE_TERMINATE; 1513 attrs.send_term = 1; 1514 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 1515 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 1516 err = -ENOMEM; 1517 disconnect = 1; 1518 goto out; 1519 } 1520 1521 /* 1522 * Generate TERM if initiator IRD is not sufficient for responder 1523 * provided ORD. Currently, we do the same behaviour even when 1524 * responder provided IRD is also not sufficient as regards to 1525 * initiator ORD. 1526 */ 1527 if (insuff_ird) { 1528 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n", 1529 __func__); 1530 attrs.layer_etype = LAYER_MPA | DDP_LLP; 1531 attrs.ecode = MPA_INSUFF_IRD; 1532 attrs.next_state = C4IW_QP_STATE_TERMINATE; 1533 attrs.send_term = 1; 1534 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 1535 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 1536 err = -ENOMEM; 1537 disconnect = 1; 1538 goto out; 1539 } 1540 goto out; 1541 err: 1542 __state_set(&ep->com, ABORTING); 1543 send_abort(ep, skb, GFP_KERNEL); 1544 out: 1545 connect_reply_upcall(ep, err); 1546 return disconnect; 1547 } 1548 1549 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb) 1550 { 1551 struct mpa_message *mpa; 1552 struct mpa_v2_conn_params *mpa_v2_params; 1553 u16 plen; 1554 1555 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1556 1557 /* 1558 * If we get more than the supported amount of private data 1559 * then we must fail this connection. 1560 */ 1561 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) { 1562 (void)stop_ep_timer(ep); 1563 abort_connection(ep, skb, GFP_KERNEL); 1564 return; 1565 } 1566 1567 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__); 1568 1569 /* 1570 * Copy the new data into our accumulation buffer. 1571 */ 1572 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]), 1573 skb->len); 1574 ep->mpa_pkt_len += skb->len; 1575 1576 /* 1577 * If we don't even have the mpa message, then bail. 1578 * We'll continue process when more data arrives. 1579 */ 1580 if (ep->mpa_pkt_len < sizeof(*mpa)) 1581 return; 1582 1583 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__); 1584 mpa = (struct mpa_message *) ep->mpa_pkt; 1585 1586 /* 1587 * Validate MPA Header. 1588 */ 1589 if (mpa->revision > mpa_rev) { 1590 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d," 1591 " Received = %d\n", __func__, mpa_rev, mpa->revision); 1592 (void)stop_ep_timer(ep); 1593 abort_connection(ep, skb, GFP_KERNEL); 1594 return; 1595 } 1596 1597 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) { 1598 (void)stop_ep_timer(ep); 1599 abort_connection(ep, skb, GFP_KERNEL); 1600 return; 1601 } 1602 1603 plen = ntohs(mpa->private_data_size); 1604 1605 /* 1606 * Fail if there's too much private data. 1607 */ 1608 if (plen > MPA_MAX_PRIVATE_DATA) { 1609 (void)stop_ep_timer(ep); 1610 abort_connection(ep, skb, GFP_KERNEL); 1611 return; 1612 } 1613 1614 /* 1615 * If plen does not account for pkt size 1616 */ 1617 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) { 1618 (void)stop_ep_timer(ep); 1619 abort_connection(ep, skb, GFP_KERNEL); 1620 return; 1621 } 1622 ep->plen = (u8) plen; 1623 1624 /* 1625 * If we don't have all the pdata yet, then bail. 1626 */ 1627 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) 1628 return; 1629 1630 /* 1631 * If we get here we have accumulated the entire mpa 1632 * start reply message including private data. 1633 */ 1634 ep->mpa_attr.initiator = 0; 1635 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1636 ep->mpa_attr.recv_marker_enabled = markers_enabled; 1637 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1638 ep->mpa_attr.version = mpa->revision; 1639 if (mpa->revision == 1) 1640 ep->tried_with_mpa_v1 = 1; 1641 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1642 1643 if (mpa->revision == 2) { 1644 ep->mpa_attr.enhanced_rdma_conn = 1645 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 1646 if (ep->mpa_attr.enhanced_rdma_conn) { 1647 mpa_v2_params = (struct mpa_v2_conn_params *) 1648 (ep->mpa_pkt + sizeof(*mpa)); 1649 ep->ird = ntohs(mpa_v2_params->ird) & 1650 MPA_V2_IRD_ORD_MASK; 1651 ep->ord = ntohs(mpa_v2_params->ord) & 1652 MPA_V2_IRD_ORD_MASK; 1653 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird, 1654 ep->ord); 1655 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL) 1656 if (peer2peer) { 1657 if (ntohs(mpa_v2_params->ord) & 1658 MPA_V2_RDMA_WRITE_RTR) 1659 ep->mpa_attr.p2p_type = 1660 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 1661 else if (ntohs(mpa_v2_params->ord) & 1662 MPA_V2_RDMA_READ_RTR) 1663 ep->mpa_attr.p2p_type = 1664 FW_RI_INIT_P2PTYPE_READ_REQ; 1665 } 1666 } 1667 } else if (mpa->revision == 1) 1668 if (peer2peer) 1669 ep->mpa_attr.p2p_type = p2p_type; 1670 1671 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, " 1672 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__, 1673 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 1674 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, 1675 ep->mpa_attr.p2p_type); 1676 1677 /* 1678 * If the endpoint timer already expired, then we ignore 1679 * the start request. process_timeout() will abort 1680 * the connection. 1681 */ 1682 if (!stop_ep_timer(ep)) { 1683 __state_set(&ep->com, MPA_REQ_RCVD); 1684 1685 /* drive upcall */ 1686 mutex_lock_nested(&ep->parent_ep->com.mutex, 1687 SINGLE_DEPTH_NESTING); 1688 if (ep->parent_ep->com.state != DEAD) { 1689 if (connect_request_upcall(ep)) 1690 abort_connection(ep, skb, GFP_KERNEL); 1691 } else { 1692 abort_connection(ep, skb, GFP_KERNEL); 1693 } 1694 mutex_unlock(&ep->parent_ep->com.mutex); 1695 } 1696 return; 1697 } 1698 1699 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb) 1700 { 1701 struct c4iw_ep *ep; 1702 struct cpl_rx_data *hdr = cplhdr(skb); 1703 unsigned int dlen = ntohs(hdr->len); 1704 unsigned int tid = GET_TID(hdr); 1705 struct tid_info *t = dev->rdev.lldi.tids; 1706 __u8 status = hdr->status; 1707 int disconnect = 0; 1708 1709 ep = lookup_tid(t, tid); 1710 if (!ep) 1711 return 0; 1712 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen); 1713 skb_pull(skb, sizeof(*hdr)); 1714 skb_trim(skb, dlen); 1715 mutex_lock(&ep->com.mutex); 1716 1717 /* update RX credits */ 1718 update_rx_credits(ep, dlen); 1719 1720 switch (ep->com.state) { 1721 case MPA_REQ_SENT: 1722 ep->rcv_seq += dlen; 1723 disconnect = process_mpa_reply(ep, skb); 1724 break; 1725 case MPA_REQ_WAIT: 1726 ep->rcv_seq += dlen; 1727 process_mpa_request(ep, skb); 1728 break; 1729 case FPDU_MODE: { 1730 struct c4iw_qp_attributes attrs; 1731 BUG_ON(!ep->com.qp); 1732 if (status) 1733 pr_err("%s Unexpected streaming data." \ 1734 " qpid %u ep %p state %d tid %u status %d\n", 1735 __func__, ep->com.qp->wq.sq.qid, ep, 1736 ep->com.state, ep->hwtid, status); 1737 attrs.next_state = C4IW_QP_STATE_TERMINATE; 1738 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 1739 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 1740 disconnect = 1; 1741 break; 1742 } 1743 default: 1744 break; 1745 } 1746 mutex_unlock(&ep->com.mutex); 1747 if (disconnect) 1748 c4iw_ep_disconnect(ep, 0, GFP_KERNEL); 1749 return 0; 1750 } 1751 1752 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 1753 { 1754 struct c4iw_ep *ep; 1755 struct cpl_abort_rpl_rss *rpl = cplhdr(skb); 1756 int release = 0; 1757 unsigned int tid = GET_TID(rpl); 1758 struct tid_info *t = dev->rdev.lldi.tids; 1759 1760 ep = lookup_tid(t, tid); 1761 if (!ep) { 1762 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n"); 1763 return 0; 1764 } 1765 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 1766 mutex_lock(&ep->com.mutex); 1767 switch (ep->com.state) { 1768 case ABORTING: 1769 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 1770 __state_set(&ep->com, DEAD); 1771 release = 1; 1772 break; 1773 default: 1774 printk(KERN_ERR "%s ep %p state %d\n", 1775 __func__, ep, ep->com.state); 1776 break; 1777 } 1778 mutex_unlock(&ep->com.mutex); 1779 1780 if (release) 1781 release_ep_resources(ep); 1782 return 0; 1783 } 1784 1785 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid) 1786 { 1787 struct sk_buff *skb; 1788 struct fw_ofld_connection_wr *req; 1789 unsigned int mtu_idx; 1790 int wscale; 1791 struct sockaddr_in *sin; 1792 int win; 1793 1794 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL); 1795 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req)); 1796 memset(req, 0, sizeof(*req)); 1797 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR)); 1798 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16))); 1799 req->le.filter = cpu_to_be32(cxgb4_select_ntuple( 1800 ep->com.dev->rdev.lldi.ports[0], 1801 ep->l2t)); 1802 sin = (struct sockaddr_in *)&ep->com.mapped_local_addr; 1803 req->le.lport = sin->sin_port; 1804 req->le.u.ipv4.lip = sin->sin_addr.s_addr; 1805 sin = (struct sockaddr_in *)&ep->com.mapped_remote_addr; 1806 req->le.pport = sin->sin_port; 1807 req->le.u.ipv4.pip = sin->sin_addr.s_addr; 1808 req->tcb.t_state_to_astid = 1809 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) | 1810 FW_OFLD_CONNECTION_WR_ASTID_V(atid)); 1811 req->tcb.cplrxdataack_cplpassacceptrpl = 1812 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F); 1813 req->tcb.tx_max = (__force __be32) jiffies; 1814 req->tcb.rcv_adv = htons(1); 1815 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx, 1816 enable_tcp_timestamps, 1817 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1); 1818 wscale = compute_wscale(rcv_win); 1819 1820 /* 1821 * Specify the largest window that will fit in opt0. The 1822 * remainder will be specified in the rx_data_ack. 1823 */ 1824 win = ep->rcv_win >> 10; 1825 if (win > RCV_BUFSIZ_M) 1826 win = RCV_BUFSIZ_M; 1827 1828 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F | 1829 (nocong ? NO_CONG_F : 0) | 1830 KEEP_ALIVE_F | 1831 DELACK_F | 1832 WND_SCALE_V(wscale) | 1833 MSS_IDX_V(mtu_idx) | 1834 L2T_IDX_V(ep->l2t->idx) | 1835 TX_CHAN_V(ep->tx_chan) | 1836 SMAC_SEL_V(ep->smac_idx) | 1837 DSCP_V(ep->tos) | 1838 ULP_MODE_V(ULP_MODE_TCPDDP) | 1839 RCV_BUFSIZ_V(win)); 1840 req->tcb.opt2 = (__force __be32) (PACE_V(1) | 1841 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) | 1842 RX_CHANNEL_V(0) | 1843 CCTRL_ECN_V(enable_ecn) | 1844 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid)); 1845 if (enable_tcp_timestamps) 1846 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F; 1847 if (enable_tcp_sack) 1848 req->tcb.opt2 |= (__force __be32)SACK_EN_F; 1849 if (wscale && enable_tcp_window_scaling) 1850 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F; 1851 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0); 1852 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2); 1853 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx); 1854 set_bit(ACT_OFLD_CONN, &ep->com.history); 1855 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 1856 } 1857 1858 /* 1859 * Return whether a failed active open has allocated a TID 1860 */ 1861 static inline int act_open_has_tid(int status) 1862 { 1863 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST && 1864 status != CPL_ERR_ARP_MISS; 1865 } 1866 1867 /* Returns whether a CPL status conveys negative advice. 1868 */ 1869 static int is_neg_adv(unsigned int status) 1870 { 1871 return status == CPL_ERR_RTX_NEG_ADVICE || 1872 status == CPL_ERR_PERSIST_NEG_ADVICE || 1873 status == CPL_ERR_KEEPALV_NEG_ADVICE; 1874 } 1875 1876 static char *neg_adv_str(unsigned int status) 1877 { 1878 switch (status) { 1879 case CPL_ERR_RTX_NEG_ADVICE: 1880 return "Retransmit timeout"; 1881 case CPL_ERR_PERSIST_NEG_ADVICE: 1882 return "Persist timeout"; 1883 case CPL_ERR_KEEPALV_NEG_ADVICE: 1884 return "Keepalive timeout"; 1885 default: 1886 return "Unknown"; 1887 } 1888 } 1889 1890 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi) 1891 { 1892 ep->snd_win = snd_win; 1893 ep->rcv_win = rcv_win; 1894 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win); 1895 } 1896 1897 #define ACT_OPEN_RETRY_COUNT 2 1898 1899 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip, 1900 struct dst_entry *dst, struct c4iw_dev *cdev, 1901 bool clear_mpa_v1) 1902 { 1903 struct neighbour *n; 1904 int err, step; 1905 struct net_device *pdev; 1906 1907 n = dst_neigh_lookup(dst, peer_ip); 1908 if (!n) 1909 return -ENODEV; 1910 1911 rcu_read_lock(); 1912 err = -ENOMEM; 1913 if (n->dev->flags & IFF_LOOPBACK) { 1914 if (iptype == 4) 1915 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip); 1916 else if (IS_ENABLED(CONFIG_IPV6)) 1917 for_each_netdev(&init_net, pdev) { 1918 if (ipv6_chk_addr(&init_net, 1919 (struct in6_addr *)peer_ip, 1920 pdev, 1)) 1921 break; 1922 } 1923 else 1924 pdev = NULL; 1925 1926 if (!pdev) { 1927 err = -ENODEV; 1928 goto out; 1929 } 1930 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t, 1931 n, pdev, 0); 1932 if (!ep->l2t) 1933 goto out; 1934 ep->mtu = pdev->mtu; 1935 ep->tx_chan = cxgb4_port_chan(pdev); 1936 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1; 1937 step = cdev->rdev.lldi.ntxq / 1938 cdev->rdev.lldi.nchan; 1939 ep->txq_idx = cxgb4_port_idx(pdev) * step; 1940 step = cdev->rdev.lldi.nrxq / 1941 cdev->rdev.lldi.nchan; 1942 ep->ctrlq_idx = cxgb4_port_idx(pdev); 1943 ep->rss_qid = cdev->rdev.lldi.rxq_ids[ 1944 cxgb4_port_idx(pdev) * step]; 1945 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev)); 1946 dev_put(pdev); 1947 } else { 1948 pdev = get_real_dev(n->dev); 1949 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t, 1950 n, pdev, 0); 1951 if (!ep->l2t) 1952 goto out; 1953 ep->mtu = dst_mtu(dst); 1954 ep->tx_chan = cxgb4_port_chan(pdev); 1955 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1; 1956 step = cdev->rdev.lldi.ntxq / 1957 cdev->rdev.lldi.nchan; 1958 ep->txq_idx = cxgb4_port_idx(pdev) * step; 1959 ep->ctrlq_idx = cxgb4_port_idx(pdev); 1960 step = cdev->rdev.lldi.nrxq / 1961 cdev->rdev.lldi.nchan; 1962 ep->rss_qid = cdev->rdev.lldi.rxq_ids[ 1963 cxgb4_port_idx(pdev) * step]; 1964 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev)); 1965 1966 if (clear_mpa_v1) { 1967 ep->retry_with_mpa_v1 = 0; 1968 ep->tried_with_mpa_v1 = 0; 1969 } 1970 } 1971 err = 0; 1972 out: 1973 rcu_read_unlock(); 1974 1975 neigh_release(n); 1976 1977 return err; 1978 } 1979 1980 static int c4iw_reconnect(struct c4iw_ep *ep) 1981 { 1982 int err = 0; 1983 struct sockaddr_in *laddr = (struct sockaddr_in *) 1984 &ep->com.cm_id->local_addr; 1985 struct sockaddr_in *raddr = (struct sockaddr_in *) 1986 &ep->com.cm_id->remote_addr; 1987 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *) 1988 &ep->com.cm_id->local_addr; 1989 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *) 1990 &ep->com.cm_id->remote_addr; 1991 int iptype; 1992 __u8 *ra; 1993 1994 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id); 1995 init_timer(&ep->timer); 1996 1997 /* 1998 * Allocate an active TID to initiate a TCP connection. 1999 */ 2000 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep); 2001 if (ep->atid == -1) { 2002 pr_err("%s - cannot alloc atid.\n", __func__); 2003 err = -ENOMEM; 2004 goto fail2; 2005 } 2006 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid); 2007 2008 /* find a route */ 2009 if (ep->com.cm_id->local_addr.ss_family == AF_INET) { 2010 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr, 2011 raddr->sin_addr.s_addr, laddr->sin_port, 2012 raddr->sin_port, 0); 2013 iptype = 4; 2014 ra = (__u8 *)&raddr->sin_addr; 2015 } else { 2016 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr, 2017 raddr6->sin6_addr.s6_addr, 2018 laddr6->sin6_port, raddr6->sin6_port, 0, 2019 raddr6->sin6_scope_id); 2020 iptype = 6; 2021 ra = (__u8 *)&raddr6->sin6_addr; 2022 } 2023 if (!ep->dst) { 2024 pr_err("%s - cannot find route.\n", __func__); 2025 err = -EHOSTUNREACH; 2026 goto fail3; 2027 } 2028 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false); 2029 if (err) { 2030 pr_err("%s - cannot alloc l2e.\n", __func__); 2031 goto fail4; 2032 } 2033 2034 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n", 2035 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid, 2036 ep->l2t->idx); 2037 2038 state_set(&ep->com, CONNECTING); 2039 ep->tos = 0; 2040 2041 /* send connect request to rnic */ 2042 err = send_connect(ep); 2043 if (!err) 2044 goto out; 2045 2046 cxgb4_l2t_release(ep->l2t); 2047 fail4: 2048 dst_release(ep->dst); 2049 fail3: 2050 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid); 2051 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid); 2052 fail2: 2053 /* 2054 * remember to send notification to upper layer. 2055 * We are in here so the upper layer is not aware that this is 2056 * re-connect attempt and so, upper layer is still waiting for 2057 * response of 1st connect request. 2058 */ 2059 connect_reply_upcall(ep, -ECONNRESET); 2060 c4iw_put_ep(&ep->com); 2061 out: 2062 return err; 2063 } 2064 2065 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2066 { 2067 struct c4iw_ep *ep; 2068 struct cpl_act_open_rpl *rpl = cplhdr(skb); 2069 unsigned int atid = TID_TID_G(AOPEN_ATID_G( 2070 ntohl(rpl->atid_status))); 2071 struct tid_info *t = dev->rdev.lldi.tids; 2072 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status)); 2073 struct sockaddr_in *la; 2074 struct sockaddr_in *ra; 2075 struct sockaddr_in6 *la6; 2076 struct sockaddr_in6 *ra6; 2077 2078 ep = lookup_atid(t, atid); 2079 la = (struct sockaddr_in *)&ep->com.mapped_local_addr; 2080 ra = (struct sockaddr_in *)&ep->com.mapped_remote_addr; 2081 la6 = (struct sockaddr_in6 *)&ep->com.mapped_local_addr; 2082 ra6 = (struct sockaddr_in6 *)&ep->com.mapped_remote_addr; 2083 2084 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid, 2085 status, status2errno(status)); 2086 2087 if (is_neg_adv(status)) { 2088 PDBG("%s Connection problems for atid %u status %u (%s)\n", 2089 __func__, atid, status, neg_adv_str(status)); 2090 ep->stats.connect_neg_adv++; 2091 mutex_lock(&dev->rdev.stats.lock); 2092 dev->rdev.stats.neg_adv++; 2093 mutex_unlock(&dev->rdev.stats.lock); 2094 return 0; 2095 } 2096 2097 set_bit(ACT_OPEN_RPL, &ep->com.history); 2098 2099 /* 2100 * Log interesting failures. 2101 */ 2102 switch (status) { 2103 case CPL_ERR_CONN_RESET: 2104 case CPL_ERR_CONN_TIMEDOUT: 2105 break; 2106 case CPL_ERR_TCAM_FULL: 2107 mutex_lock(&dev->rdev.stats.lock); 2108 dev->rdev.stats.tcam_full++; 2109 mutex_unlock(&dev->rdev.stats.lock); 2110 if (ep->com.local_addr.ss_family == AF_INET && 2111 dev->rdev.lldi.enable_fw_ofld_conn) { 2112 send_fw_act_open_req(ep, 2113 TID_TID_G(AOPEN_ATID_G( 2114 ntohl(rpl->atid_status)))); 2115 return 0; 2116 } 2117 break; 2118 case CPL_ERR_CONN_EXIST: 2119 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) { 2120 set_bit(ACT_RETRY_INUSE, &ep->com.history); 2121 if (ep->com.remote_addr.ss_family == AF_INET6) { 2122 struct sockaddr_in6 *sin6 = 2123 (struct sockaddr_in6 *) 2124 &ep->com.mapped_local_addr; 2125 cxgb4_clip_release( 2126 ep->com.dev->rdev.lldi.ports[0], 2127 (const u32 *) 2128 &sin6->sin6_addr.s6_addr, 1); 2129 } 2130 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, 2131 atid); 2132 cxgb4_free_atid(t, atid); 2133 dst_release(ep->dst); 2134 cxgb4_l2t_release(ep->l2t); 2135 c4iw_reconnect(ep); 2136 return 0; 2137 } 2138 break; 2139 default: 2140 if (ep->com.local_addr.ss_family == AF_INET) { 2141 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n", 2142 atid, status, status2errno(status), 2143 &la->sin_addr.s_addr, ntohs(la->sin_port), 2144 &ra->sin_addr.s_addr, ntohs(ra->sin_port)); 2145 } else { 2146 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n", 2147 atid, status, status2errno(status), 2148 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port), 2149 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port)); 2150 } 2151 break; 2152 } 2153 2154 connect_reply_upcall(ep, status2errno(status)); 2155 state_set(&ep->com, DEAD); 2156 2157 if (ep->com.remote_addr.ss_family == AF_INET6) { 2158 struct sockaddr_in6 *sin6 = 2159 (struct sockaddr_in6 *)&ep->com.mapped_local_addr; 2160 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 2161 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 2162 } 2163 if (status && act_open_has_tid(status)) 2164 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl)); 2165 2166 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid); 2167 cxgb4_free_atid(t, atid); 2168 dst_release(ep->dst); 2169 cxgb4_l2t_release(ep->l2t); 2170 c4iw_put_ep(&ep->com); 2171 2172 return 0; 2173 } 2174 2175 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2176 { 2177 struct cpl_pass_open_rpl *rpl = cplhdr(skb); 2178 struct tid_info *t = dev->rdev.lldi.tids; 2179 unsigned int stid = GET_TID(rpl); 2180 struct c4iw_listen_ep *ep = lookup_stid(t, stid); 2181 2182 if (!ep) { 2183 PDBG("%s stid %d lookup failure!\n", __func__, stid); 2184 goto out; 2185 } 2186 PDBG("%s ep %p status %d error %d\n", __func__, ep, 2187 rpl->status, status2errno(rpl->status)); 2188 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status)); 2189 2190 out: 2191 return 0; 2192 } 2193 2194 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2195 { 2196 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb); 2197 struct tid_info *t = dev->rdev.lldi.tids; 2198 unsigned int stid = GET_TID(rpl); 2199 struct c4iw_listen_ep *ep = lookup_stid(t, stid); 2200 2201 PDBG("%s ep %p\n", __func__, ep); 2202 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status)); 2203 return 0; 2204 } 2205 2206 static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb, 2207 struct cpl_pass_accept_req *req) 2208 { 2209 struct cpl_pass_accept_rpl *rpl; 2210 unsigned int mtu_idx; 2211 u64 opt0; 2212 u32 opt2; 2213 int wscale; 2214 struct cpl_t5_pass_accept_rpl *rpl5 = NULL; 2215 int win; 2216 2217 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2218 BUG_ON(skb_cloned(skb)); 2219 2220 skb_get(skb); 2221 rpl = cplhdr(skb); 2222 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) { 2223 skb_trim(skb, roundup(sizeof(*rpl5), 16)); 2224 rpl5 = (void *)rpl; 2225 INIT_TP_WR(rpl5, ep->hwtid); 2226 } else { 2227 skb_trim(skb, sizeof(*rpl)); 2228 INIT_TP_WR(rpl, ep->hwtid); 2229 } 2230 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL, 2231 ep->hwtid)); 2232 2233 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx, 2234 enable_tcp_timestamps && req->tcpopt.tstamp, 2235 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1); 2236 wscale = compute_wscale(rcv_win); 2237 2238 /* 2239 * Specify the largest window that will fit in opt0. The 2240 * remainder will be specified in the rx_data_ack. 2241 */ 2242 win = ep->rcv_win >> 10; 2243 if (win > RCV_BUFSIZ_M) 2244 win = RCV_BUFSIZ_M; 2245 opt0 = (nocong ? NO_CONG_F : 0) | 2246 KEEP_ALIVE_F | 2247 DELACK_F | 2248 WND_SCALE_V(wscale) | 2249 MSS_IDX_V(mtu_idx) | 2250 L2T_IDX_V(ep->l2t->idx) | 2251 TX_CHAN_V(ep->tx_chan) | 2252 SMAC_SEL_V(ep->smac_idx) | 2253 DSCP_V(ep->tos >> 2) | 2254 ULP_MODE_V(ULP_MODE_TCPDDP) | 2255 RCV_BUFSIZ_V(win); 2256 opt2 = RX_CHANNEL_V(0) | 2257 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid); 2258 2259 if (enable_tcp_timestamps && req->tcpopt.tstamp) 2260 opt2 |= TSTAMPS_EN_F; 2261 if (enable_tcp_sack && req->tcpopt.sack) 2262 opt2 |= SACK_EN_F; 2263 if (wscale && enable_tcp_window_scaling) 2264 opt2 |= WND_SCALE_EN_F; 2265 if (enable_ecn) { 2266 const struct tcphdr *tcph; 2267 u32 hlen = ntohl(req->hdr_len); 2268 2269 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) + 2270 IP_HDR_LEN_G(hlen); 2271 if (tcph->ece && tcph->cwr) 2272 opt2 |= CCTRL_ECN_V(1); 2273 } 2274 if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) { 2275 u32 isn = (prandom_u32() & ~7UL) - 1; 2276 opt2 |= T5_OPT_2_VALID_F; 2277 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE); 2278 opt2 |= T5_ISS_F; 2279 rpl5 = (void *)rpl; 2280 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16)); 2281 if (peer2peer) 2282 isn += 4; 2283 rpl5->iss = cpu_to_be32(isn); 2284 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss)); 2285 } 2286 2287 rpl->opt0 = cpu_to_be64(opt0); 2288 rpl->opt2 = cpu_to_be32(opt2); 2289 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx); 2290 t4_set_arp_err_handler(skb, NULL, arp_failure_discard); 2291 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t); 2292 2293 return; 2294 } 2295 2296 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb) 2297 { 2298 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid); 2299 BUG_ON(skb_cloned(skb)); 2300 skb_trim(skb, sizeof(struct cpl_tid_release)); 2301 release_tid(&dev->rdev, hwtid, skb); 2302 return; 2303 } 2304 2305 static void get_4tuple(struct cpl_pass_accept_req *req, int *iptype, 2306 __u8 *local_ip, __u8 *peer_ip, 2307 __be16 *local_port, __be16 *peer_port) 2308 { 2309 int eth_len = ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len)); 2310 int ip_len = IP_HDR_LEN_G(be32_to_cpu(req->hdr_len)); 2311 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len); 2312 struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len); 2313 struct tcphdr *tcp = (struct tcphdr *) 2314 ((u8 *)(req + 1) + eth_len + ip_len); 2315 2316 if (ip->version == 4) { 2317 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__, 2318 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source), 2319 ntohs(tcp->dest)); 2320 *iptype = 4; 2321 memcpy(peer_ip, &ip->saddr, 4); 2322 memcpy(local_ip, &ip->daddr, 4); 2323 } else { 2324 PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__, 2325 ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source), 2326 ntohs(tcp->dest)); 2327 *iptype = 6; 2328 memcpy(peer_ip, ip6->saddr.s6_addr, 16); 2329 memcpy(local_ip, ip6->daddr.s6_addr, 16); 2330 } 2331 *peer_port = tcp->source; 2332 *local_port = tcp->dest; 2333 2334 return; 2335 } 2336 2337 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb) 2338 { 2339 struct c4iw_ep *child_ep = NULL, *parent_ep; 2340 struct cpl_pass_accept_req *req = cplhdr(skb); 2341 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid)); 2342 struct tid_info *t = dev->rdev.lldi.tids; 2343 unsigned int hwtid = GET_TID(req); 2344 struct dst_entry *dst; 2345 __u8 local_ip[16], peer_ip[16]; 2346 __be16 local_port, peer_port; 2347 struct sockaddr_in6 *sin6; 2348 int err; 2349 u16 peer_mss = ntohs(req->tcpopt.mss); 2350 int iptype; 2351 unsigned short hdrs; 2352 2353 parent_ep = lookup_stid(t, stid); 2354 if (!parent_ep) { 2355 PDBG("%s connect request on invalid stid %d\n", __func__, stid); 2356 goto reject; 2357 } 2358 2359 if (state_read(&parent_ep->com) != LISTEN) { 2360 printk(KERN_ERR "%s - listening ep not in LISTEN\n", 2361 __func__); 2362 goto reject; 2363 } 2364 2365 get_4tuple(req, &iptype, local_ip, peer_ip, &local_port, &peer_port); 2366 2367 /* Find output route */ 2368 if (iptype == 4) { 2369 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n" 2370 , __func__, parent_ep, hwtid, 2371 local_ip, peer_ip, ntohs(local_port), 2372 ntohs(peer_port), peer_mss); 2373 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip, 2374 local_port, peer_port, 2375 PASS_OPEN_TOS_G(ntohl(req->tos_stid))); 2376 } else { 2377 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n" 2378 , __func__, parent_ep, hwtid, 2379 local_ip, peer_ip, ntohs(local_port), 2380 ntohs(peer_port), peer_mss); 2381 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port, 2382 PASS_OPEN_TOS_G(ntohl(req->tos_stid)), 2383 ((struct sockaddr_in6 *) 2384 &parent_ep->com.local_addr)->sin6_scope_id); 2385 } 2386 if (!dst) { 2387 printk(KERN_ERR MOD "%s - failed to find dst entry!\n", 2388 __func__); 2389 goto reject; 2390 } 2391 2392 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL); 2393 if (!child_ep) { 2394 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n", 2395 __func__); 2396 dst_release(dst); 2397 goto reject; 2398 } 2399 2400 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false); 2401 if (err) { 2402 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n", 2403 __func__); 2404 dst_release(dst); 2405 kfree(child_ep); 2406 goto reject; 2407 } 2408 2409 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) + 2410 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0); 2411 if (peer_mss && child_ep->mtu > (peer_mss + hdrs)) 2412 child_ep->mtu = peer_mss + hdrs; 2413 2414 state_set(&child_ep->com, CONNECTING); 2415 child_ep->com.dev = dev; 2416 child_ep->com.cm_id = NULL; 2417 2418 /* 2419 * The mapped_local and mapped_remote addresses get setup with 2420 * the actual 4-tuple. The local address will be based on the 2421 * actual local address of the connection, but on the port number 2422 * of the parent listening endpoint. The remote address is 2423 * setup based on a query to the IWPM since we don't know what it 2424 * originally was before mapping. If no mapping was done, then 2425 * mapped_remote == remote, and mapped_local == local. 2426 */ 2427 if (iptype == 4) { 2428 struct sockaddr_in *sin = (struct sockaddr_in *) 2429 &child_ep->com.mapped_local_addr; 2430 2431 sin->sin_family = PF_INET; 2432 sin->sin_port = local_port; 2433 sin->sin_addr.s_addr = *(__be32 *)local_ip; 2434 2435 sin = (struct sockaddr_in *)&child_ep->com.local_addr; 2436 sin->sin_family = PF_INET; 2437 sin->sin_port = ((struct sockaddr_in *) 2438 &parent_ep->com.local_addr)->sin_port; 2439 sin->sin_addr.s_addr = *(__be32 *)local_ip; 2440 2441 sin = (struct sockaddr_in *)&child_ep->com.mapped_remote_addr; 2442 sin->sin_family = PF_INET; 2443 sin->sin_port = peer_port; 2444 sin->sin_addr.s_addr = *(__be32 *)peer_ip; 2445 } else { 2446 sin6 = (struct sockaddr_in6 *)&child_ep->com.mapped_local_addr; 2447 sin6->sin6_family = PF_INET6; 2448 sin6->sin6_port = local_port; 2449 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16); 2450 2451 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr; 2452 sin6->sin6_family = PF_INET6; 2453 sin6->sin6_port = ((struct sockaddr_in6 *) 2454 &parent_ep->com.local_addr)->sin6_port; 2455 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16); 2456 2457 sin6 = (struct sockaddr_in6 *)&child_ep->com.mapped_remote_addr; 2458 sin6->sin6_family = PF_INET6; 2459 sin6->sin6_port = peer_port; 2460 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16); 2461 } 2462 memcpy(&child_ep->com.remote_addr, &child_ep->com.mapped_remote_addr, 2463 sizeof(child_ep->com.remote_addr)); 2464 get_remote_addr(parent_ep, child_ep); 2465 2466 c4iw_get_ep(&parent_ep->com); 2467 child_ep->parent_ep = parent_ep; 2468 child_ep->tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid)); 2469 child_ep->dst = dst; 2470 child_ep->hwtid = hwtid; 2471 2472 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__, 2473 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid); 2474 2475 init_timer(&child_ep->timer); 2476 cxgb4_insert_tid(t, child_ep, hwtid); 2477 insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid); 2478 accept_cr(child_ep, skb, req); 2479 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history); 2480 if (iptype == 6) { 2481 sin6 = (struct sockaddr_in6 *)&child_ep->com.mapped_local_addr; 2482 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0], 2483 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 2484 } 2485 goto out; 2486 reject: 2487 reject_cr(dev, hwtid, skb); 2488 out: 2489 return 0; 2490 } 2491 2492 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb) 2493 { 2494 struct c4iw_ep *ep; 2495 struct cpl_pass_establish *req = cplhdr(skb); 2496 struct tid_info *t = dev->rdev.lldi.tids; 2497 unsigned int tid = GET_TID(req); 2498 2499 ep = lookup_tid(t, tid); 2500 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2501 ep->snd_seq = be32_to_cpu(req->snd_isn); 2502 ep->rcv_seq = be32_to_cpu(req->rcv_isn); 2503 2504 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid, 2505 ntohs(req->tcp_opt)); 2506 2507 set_emss(ep, ntohs(req->tcp_opt)); 2508 2509 dst_confirm(ep->dst); 2510 state_set(&ep->com, MPA_REQ_WAIT); 2511 start_ep_timer(ep); 2512 send_flowc(ep, skb); 2513 set_bit(PASS_ESTAB, &ep->com.history); 2514 2515 return 0; 2516 } 2517 2518 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb) 2519 { 2520 struct cpl_peer_close *hdr = cplhdr(skb); 2521 struct c4iw_ep *ep; 2522 struct c4iw_qp_attributes attrs; 2523 int disconnect = 1; 2524 int release = 0; 2525 struct tid_info *t = dev->rdev.lldi.tids; 2526 unsigned int tid = GET_TID(hdr); 2527 int ret; 2528 2529 ep = lookup_tid(t, tid); 2530 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2531 dst_confirm(ep->dst); 2532 2533 set_bit(PEER_CLOSE, &ep->com.history); 2534 mutex_lock(&ep->com.mutex); 2535 switch (ep->com.state) { 2536 case MPA_REQ_WAIT: 2537 __state_set(&ep->com, CLOSING); 2538 break; 2539 case MPA_REQ_SENT: 2540 __state_set(&ep->com, CLOSING); 2541 connect_reply_upcall(ep, -ECONNRESET); 2542 break; 2543 case MPA_REQ_RCVD: 2544 2545 /* 2546 * We're gonna mark this puppy DEAD, but keep 2547 * the reference on it until the ULP accepts or 2548 * rejects the CR. Also wake up anyone waiting 2549 * in rdma connection migration (see c4iw_accept_cr()). 2550 */ 2551 __state_set(&ep->com, CLOSING); 2552 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid); 2553 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 2554 break; 2555 case MPA_REP_SENT: 2556 __state_set(&ep->com, CLOSING); 2557 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid); 2558 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 2559 break; 2560 case FPDU_MODE: 2561 start_ep_timer(ep); 2562 __state_set(&ep->com, CLOSING); 2563 attrs.next_state = C4IW_QP_STATE_CLOSING; 2564 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2565 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2566 if (ret != -ECONNRESET) { 2567 peer_close_upcall(ep); 2568 disconnect = 1; 2569 } 2570 break; 2571 case ABORTING: 2572 disconnect = 0; 2573 break; 2574 case CLOSING: 2575 __state_set(&ep->com, MORIBUND); 2576 disconnect = 0; 2577 break; 2578 case MORIBUND: 2579 (void)stop_ep_timer(ep); 2580 if (ep->com.cm_id && ep->com.qp) { 2581 attrs.next_state = C4IW_QP_STATE_IDLE; 2582 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2583 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2584 } 2585 close_complete_upcall(ep, 0); 2586 __state_set(&ep->com, DEAD); 2587 release = 1; 2588 disconnect = 0; 2589 break; 2590 case DEAD: 2591 disconnect = 0; 2592 break; 2593 default: 2594 BUG_ON(1); 2595 } 2596 mutex_unlock(&ep->com.mutex); 2597 if (disconnect) 2598 c4iw_ep_disconnect(ep, 0, GFP_KERNEL); 2599 if (release) 2600 release_ep_resources(ep); 2601 return 0; 2602 } 2603 2604 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb) 2605 { 2606 struct cpl_abort_req_rss *req = cplhdr(skb); 2607 struct c4iw_ep *ep; 2608 struct cpl_abort_rpl *rpl; 2609 struct sk_buff *rpl_skb; 2610 struct c4iw_qp_attributes attrs; 2611 int ret; 2612 int release = 0; 2613 struct tid_info *t = dev->rdev.lldi.tids; 2614 unsigned int tid = GET_TID(req); 2615 2616 ep = lookup_tid(t, tid); 2617 if (is_neg_adv(req->status)) { 2618 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n", 2619 __func__, ep->hwtid, req->status, 2620 neg_adv_str(req->status)); 2621 ep->stats.abort_neg_adv++; 2622 mutex_lock(&dev->rdev.stats.lock); 2623 dev->rdev.stats.neg_adv++; 2624 mutex_unlock(&dev->rdev.stats.lock); 2625 return 0; 2626 } 2627 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid, 2628 ep->com.state); 2629 set_bit(PEER_ABORT, &ep->com.history); 2630 2631 /* 2632 * Wake up any threads in rdma_init() or rdma_fini(). 2633 * However, this is not needed if com state is just 2634 * MPA_REQ_SENT 2635 */ 2636 if (ep->com.state != MPA_REQ_SENT) 2637 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 2638 2639 mutex_lock(&ep->com.mutex); 2640 switch (ep->com.state) { 2641 case CONNECTING: 2642 break; 2643 case MPA_REQ_WAIT: 2644 (void)stop_ep_timer(ep); 2645 break; 2646 case MPA_REQ_SENT: 2647 (void)stop_ep_timer(ep); 2648 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1)) 2649 connect_reply_upcall(ep, -ECONNRESET); 2650 else { 2651 /* 2652 * we just don't send notification upwards because we 2653 * want to retry with mpa_v1 without upper layers even 2654 * knowing it. 2655 * 2656 * do some housekeeping so as to re-initiate the 2657 * connection 2658 */ 2659 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__, 2660 mpa_rev); 2661 ep->retry_with_mpa_v1 = 1; 2662 } 2663 break; 2664 case MPA_REP_SENT: 2665 break; 2666 case MPA_REQ_RCVD: 2667 break; 2668 case MORIBUND: 2669 case CLOSING: 2670 stop_ep_timer(ep); 2671 /*FALLTHROUGH*/ 2672 case FPDU_MODE: 2673 if (ep->com.cm_id && ep->com.qp) { 2674 attrs.next_state = C4IW_QP_STATE_ERROR; 2675 ret = c4iw_modify_qp(ep->com.qp->rhp, 2676 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, 2677 &attrs, 1); 2678 if (ret) 2679 printk(KERN_ERR MOD 2680 "%s - qp <- error failed!\n", 2681 __func__); 2682 } 2683 peer_abort_upcall(ep); 2684 break; 2685 case ABORTING: 2686 break; 2687 case DEAD: 2688 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__); 2689 mutex_unlock(&ep->com.mutex); 2690 return 0; 2691 default: 2692 BUG_ON(1); 2693 break; 2694 } 2695 dst_confirm(ep->dst); 2696 if (ep->com.state != ABORTING) { 2697 __state_set(&ep->com, DEAD); 2698 /* we don't release if we want to retry with mpa_v1 */ 2699 if (!ep->retry_with_mpa_v1) 2700 release = 1; 2701 } 2702 mutex_unlock(&ep->com.mutex); 2703 2704 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL); 2705 if (!rpl_skb) { 2706 printk(KERN_ERR MOD "%s - cannot allocate skb!\n", 2707 __func__); 2708 release = 1; 2709 goto out; 2710 } 2711 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx); 2712 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl)); 2713 INIT_TP_WR(rpl, ep->hwtid); 2714 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid)); 2715 rpl->cmd = CPL_ABORT_NO_RST; 2716 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb); 2717 out: 2718 if (release) 2719 release_ep_resources(ep); 2720 else if (ep->retry_with_mpa_v1) { 2721 if (ep->com.remote_addr.ss_family == AF_INET6) { 2722 struct sockaddr_in6 *sin6 = 2723 (struct sockaddr_in6 *) 2724 &ep->com.mapped_local_addr; 2725 cxgb4_clip_release( 2726 ep->com.dev->rdev.lldi.ports[0], 2727 (const u32 *)&sin6->sin6_addr.s6_addr, 2728 1); 2729 } 2730 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid); 2731 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid); 2732 dst_release(ep->dst); 2733 cxgb4_l2t_release(ep->l2t); 2734 c4iw_reconnect(ep); 2735 } 2736 2737 return 0; 2738 } 2739 2740 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 2741 { 2742 struct c4iw_ep *ep; 2743 struct c4iw_qp_attributes attrs; 2744 struct cpl_close_con_rpl *rpl = cplhdr(skb); 2745 int release = 0; 2746 struct tid_info *t = dev->rdev.lldi.tids; 2747 unsigned int tid = GET_TID(rpl); 2748 2749 ep = lookup_tid(t, tid); 2750 2751 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2752 BUG_ON(!ep); 2753 2754 /* The cm_id may be null if we failed to connect */ 2755 mutex_lock(&ep->com.mutex); 2756 switch (ep->com.state) { 2757 case CLOSING: 2758 __state_set(&ep->com, MORIBUND); 2759 break; 2760 case MORIBUND: 2761 (void)stop_ep_timer(ep); 2762 if ((ep->com.cm_id) && (ep->com.qp)) { 2763 attrs.next_state = C4IW_QP_STATE_IDLE; 2764 c4iw_modify_qp(ep->com.qp->rhp, 2765 ep->com.qp, 2766 C4IW_QP_ATTR_NEXT_STATE, 2767 &attrs, 1); 2768 } 2769 close_complete_upcall(ep, 0); 2770 __state_set(&ep->com, DEAD); 2771 release = 1; 2772 break; 2773 case ABORTING: 2774 case DEAD: 2775 break; 2776 default: 2777 BUG_ON(1); 2778 break; 2779 } 2780 mutex_unlock(&ep->com.mutex); 2781 if (release) 2782 release_ep_resources(ep); 2783 return 0; 2784 } 2785 2786 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb) 2787 { 2788 struct cpl_rdma_terminate *rpl = cplhdr(skb); 2789 struct tid_info *t = dev->rdev.lldi.tids; 2790 unsigned int tid = GET_TID(rpl); 2791 struct c4iw_ep *ep; 2792 struct c4iw_qp_attributes attrs; 2793 2794 ep = lookup_tid(t, tid); 2795 BUG_ON(!ep); 2796 2797 if (ep && ep->com.qp) { 2798 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid, 2799 ep->com.qp->wq.sq.qid); 2800 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2801 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2802 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2803 } else 2804 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid); 2805 2806 return 0; 2807 } 2808 2809 /* 2810 * Upcall from the adapter indicating data has been transmitted. 2811 * For us its just the single MPA request or reply. We can now free 2812 * the skb holding the mpa message. 2813 */ 2814 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb) 2815 { 2816 struct c4iw_ep *ep; 2817 struct cpl_fw4_ack *hdr = cplhdr(skb); 2818 u8 credits = hdr->credits; 2819 unsigned int tid = GET_TID(hdr); 2820 struct tid_info *t = dev->rdev.lldi.tids; 2821 2822 2823 ep = lookup_tid(t, tid); 2824 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits); 2825 if (credits == 0) { 2826 PDBG("%s 0 credit ack ep %p tid %u state %u\n", 2827 __func__, ep, ep->hwtid, state_read(&ep->com)); 2828 return 0; 2829 } 2830 2831 dst_confirm(ep->dst); 2832 if (ep->mpa_skb) { 2833 PDBG("%s last streaming msg ack ep %p tid %u state %u " 2834 "initiator %u freeing skb\n", __func__, ep, ep->hwtid, 2835 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0); 2836 kfree_skb(ep->mpa_skb); 2837 ep->mpa_skb = NULL; 2838 } 2839 return 0; 2840 } 2841 2842 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len) 2843 { 2844 int err = 0; 2845 int disconnect = 0; 2846 struct c4iw_ep *ep = to_ep(cm_id); 2847 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2848 2849 mutex_lock(&ep->com.mutex); 2850 if (ep->com.state == DEAD) { 2851 mutex_unlock(&ep->com.mutex); 2852 c4iw_put_ep(&ep->com); 2853 return -ECONNRESET; 2854 } 2855 set_bit(ULP_REJECT, &ep->com.history); 2856 BUG_ON(ep->com.state != MPA_REQ_RCVD); 2857 if (mpa_rev == 0) 2858 abort_connection(ep, NULL, GFP_KERNEL); 2859 else { 2860 err = send_mpa_reject(ep, pdata, pdata_len); 2861 disconnect = 1; 2862 } 2863 mutex_unlock(&ep->com.mutex); 2864 if (disconnect) 2865 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL); 2866 c4iw_put_ep(&ep->com); 2867 return 0; 2868 } 2869 2870 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 2871 { 2872 int err; 2873 struct c4iw_qp_attributes attrs; 2874 enum c4iw_qp_attr_mask mask; 2875 struct c4iw_ep *ep = to_ep(cm_id); 2876 struct c4iw_dev *h = to_c4iw_dev(cm_id->device); 2877 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn); 2878 2879 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid); 2880 2881 mutex_lock(&ep->com.mutex); 2882 if (ep->com.state == DEAD) { 2883 err = -ECONNRESET; 2884 goto err; 2885 } 2886 2887 BUG_ON(ep->com.state != MPA_REQ_RCVD); 2888 BUG_ON(!qp); 2889 2890 set_bit(ULP_ACCEPT, &ep->com.history); 2891 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) || 2892 (conn_param->ird > cur_max_read_depth(ep->com.dev))) { 2893 abort_connection(ep, NULL, GFP_KERNEL); 2894 err = -EINVAL; 2895 goto err; 2896 } 2897 2898 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 2899 if (conn_param->ord > ep->ird) { 2900 if (RELAXED_IRD_NEGOTIATION) { 2901 ep->ord = ep->ird; 2902 } else { 2903 ep->ird = conn_param->ird; 2904 ep->ord = conn_param->ord; 2905 send_mpa_reject(ep, conn_param->private_data, 2906 conn_param->private_data_len); 2907 abort_connection(ep, NULL, GFP_KERNEL); 2908 err = -ENOMEM; 2909 goto err; 2910 } 2911 } 2912 if (conn_param->ird < ep->ord) { 2913 if (RELAXED_IRD_NEGOTIATION && 2914 ep->ord <= h->rdev.lldi.max_ordird_qp) { 2915 conn_param->ird = ep->ord; 2916 } else { 2917 abort_connection(ep, NULL, GFP_KERNEL); 2918 err = -ENOMEM; 2919 goto err; 2920 } 2921 } 2922 } 2923 ep->ird = conn_param->ird; 2924 ep->ord = conn_param->ord; 2925 2926 if (ep->mpa_attr.version == 1) { 2927 if (peer2peer && ep->ird == 0) 2928 ep->ird = 1; 2929 } else { 2930 if (peer2peer && 2931 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) && 2932 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ord == 0) 2933 ep->ird = 1; 2934 } 2935 2936 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord); 2937 2938 cm_id->add_ref(cm_id); 2939 ep->com.cm_id = cm_id; 2940 ep->com.qp = qp; 2941 ref_qp(ep); 2942 2943 /* bind QP to EP and move to RTS */ 2944 attrs.mpa_attr = ep->mpa_attr; 2945 attrs.max_ird = ep->ird; 2946 attrs.max_ord = ep->ord; 2947 attrs.llp_stream_handle = ep; 2948 attrs.next_state = C4IW_QP_STATE_RTS; 2949 2950 /* bind QP and TID with INIT_WR */ 2951 mask = C4IW_QP_ATTR_NEXT_STATE | 2952 C4IW_QP_ATTR_LLP_STREAM_HANDLE | 2953 C4IW_QP_ATTR_MPA_ATTR | 2954 C4IW_QP_ATTR_MAX_IRD | 2955 C4IW_QP_ATTR_MAX_ORD; 2956 2957 err = c4iw_modify_qp(ep->com.qp->rhp, 2958 ep->com.qp, mask, &attrs, 1); 2959 if (err) 2960 goto err1; 2961 err = send_mpa_reply(ep, conn_param->private_data, 2962 conn_param->private_data_len); 2963 if (err) 2964 goto err1; 2965 2966 __state_set(&ep->com, FPDU_MODE); 2967 established_upcall(ep); 2968 mutex_unlock(&ep->com.mutex); 2969 c4iw_put_ep(&ep->com); 2970 return 0; 2971 err1: 2972 ep->com.cm_id = NULL; 2973 abort_connection(ep, NULL, GFP_KERNEL); 2974 cm_id->rem_ref(cm_id); 2975 err: 2976 mutex_unlock(&ep->com.mutex); 2977 c4iw_put_ep(&ep->com); 2978 return err; 2979 } 2980 2981 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id) 2982 { 2983 struct in_device *ind; 2984 int found = 0; 2985 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr; 2986 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr; 2987 2988 ind = in_dev_get(dev->rdev.lldi.ports[0]); 2989 if (!ind) 2990 return -EADDRNOTAVAIL; 2991 for_primary_ifa(ind) { 2992 laddr->sin_addr.s_addr = ifa->ifa_address; 2993 raddr->sin_addr.s_addr = ifa->ifa_address; 2994 found = 1; 2995 break; 2996 } 2997 endfor_ifa(ind); 2998 in_dev_put(ind); 2999 return found ? 0 : -EADDRNOTAVAIL; 3000 } 3001 3002 static int get_lladdr(struct net_device *dev, struct in6_addr *addr, 3003 unsigned char banned_flags) 3004 { 3005 struct inet6_dev *idev; 3006 int err = -EADDRNOTAVAIL; 3007 3008 rcu_read_lock(); 3009 idev = __in6_dev_get(dev); 3010 if (idev != NULL) { 3011 struct inet6_ifaddr *ifp; 3012 3013 read_lock_bh(&idev->lock); 3014 list_for_each_entry(ifp, &idev->addr_list, if_list) { 3015 if (ifp->scope == IFA_LINK && 3016 !(ifp->flags & banned_flags)) { 3017 memcpy(addr, &ifp->addr, 16); 3018 err = 0; 3019 break; 3020 } 3021 } 3022 read_unlock_bh(&idev->lock); 3023 } 3024 rcu_read_unlock(); 3025 return err; 3026 } 3027 3028 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id) 3029 { 3030 struct in6_addr uninitialized_var(addr); 3031 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr; 3032 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr; 3033 3034 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) { 3035 memcpy(la6->sin6_addr.s6_addr, &addr, 16); 3036 memcpy(ra6->sin6_addr.s6_addr, &addr, 16); 3037 return 0; 3038 } 3039 return -EADDRNOTAVAIL; 3040 } 3041 3042 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 3043 { 3044 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 3045 struct c4iw_ep *ep; 3046 int err = 0; 3047 struct sockaddr_in *laddr; 3048 struct sockaddr_in *raddr; 3049 struct sockaddr_in6 *laddr6; 3050 struct sockaddr_in6 *raddr6; 3051 struct iwpm_dev_data pm_reg_msg; 3052 struct iwpm_sa_data pm_msg; 3053 __u8 *ra; 3054 int iptype; 3055 int iwpm_err = 0; 3056 3057 if ((conn_param->ord > cur_max_read_depth(dev)) || 3058 (conn_param->ird > cur_max_read_depth(dev))) { 3059 err = -EINVAL; 3060 goto out; 3061 } 3062 ep = alloc_ep(sizeof(*ep), GFP_KERNEL); 3063 if (!ep) { 3064 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__); 3065 err = -ENOMEM; 3066 goto out; 3067 } 3068 init_timer(&ep->timer); 3069 ep->plen = conn_param->private_data_len; 3070 if (ep->plen) 3071 memcpy(ep->mpa_pkt + sizeof(struct mpa_message), 3072 conn_param->private_data, ep->plen); 3073 ep->ird = conn_param->ird; 3074 ep->ord = conn_param->ord; 3075 3076 if (peer2peer && ep->ord == 0) 3077 ep->ord = 1; 3078 3079 cm_id->add_ref(cm_id); 3080 ep->com.dev = dev; 3081 ep->com.cm_id = cm_id; 3082 ep->com.qp = get_qhp(dev, conn_param->qpn); 3083 if (!ep->com.qp) { 3084 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn); 3085 err = -EINVAL; 3086 goto fail1; 3087 } 3088 ref_qp(ep); 3089 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn, 3090 ep->com.qp, cm_id); 3091 3092 /* 3093 * Allocate an active TID to initiate a TCP connection. 3094 */ 3095 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep); 3096 if (ep->atid == -1) { 3097 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__); 3098 err = -ENOMEM; 3099 goto fail1; 3100 } 3101 insert_handle(dev, &dev->atid_idr, ep, ep->atid); 3102 3103 memcpy(&ep->com.local_addr, &cm_id->local_addr, 3104 sizeof(ep->com.local_addr)); 3105 memcpy(&ep->com.remote_addr, &cm_id->remote_addr, 3106 sizeof(ep->com.remote_addr)); 3107 3108 /* No port mapper available, go with the specified peer information */ 3109 memcpy(&ep->com.mapped_local_addr, &cm_id->local_addr, 3110 sizeof(ep->com.mapped_local_addr)); 3111 memcpy(&ep->com.mapped_remote_addr, &cm_id->remote_addr, 3112 sizeof(ep->com.mapped_remote_addr)); 3113 3114 c4iw_form_reg_msg(dev, &pm_reg_msg); 3115 iwpm_err = iwpm_register_pid(&pm_reg_msg, RDMA_NL_C4IW); 3116 if (iwpm_err) { 3117 PDBG("%s: Port Mapper reg pid fail (err = %d).\n", 3118 __func__, iwpm_err); 3119 } 3120 if (iwpm_valid_pid() && !iwpm_err) { 3121 c4iw_form_pm_msg(ep, &pm_msg); 3122 iwpm_err = iwpm_add_and_query_mapping(&pm_msg, RDMA_NL_C4IW); 3123 if (iwpm_err) 3124 PDBG("%s: Port Mapper query fail (err = %d).\n", 3125 __func__, iwpm_err); 3126 else 3127 c4iw_record_pm_msg(ep, &pm_msg); 3128 } 3129 if (iwpm_create_mapinfo(&ep->com.local_addr, 3130 &ep->com.mapped_local_addr, RDMA_NL_C4IW)) { 3131 iwpm_remove_mapping(&ep->com.local_addr, RDMA_NL_C4IW); 3132 err = -ENOMEM; 3133 goto fail1; 3134 } 3135 print_addr(&ep->com, __func__, "add_query/create_mapinfo"); 3136 set_bit(RELEASE_MAPINFO, &ep->com.flags); 3137 3138 laddr = (struct sockaddr_in *)&ep->com.mapped_local_addr; 3139 raddr = (struct sockaddr_in *)&ep->com.mapped_remote_addr; 3140 laddr6 = (struct sockaddr_in6 *)&ep->com.mapped_local_addr; 3141 raddr6 = (struct sockaddr_in6 *) &ep->com.mapped_remote_addr; 3142 3143 if (cm_id->remote_addr.ss_family == AF_INET) { 3144 iptype = 4; 3145 ra = (__u8 *)&raddr->sin_addr; 3146 3147 /* 3148 * Handle loopback requests to INADDR_ANY. 3149 */ 3150 if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) { 3151 err = pick_local_ipaddrs(dev, cm_id); 3152 if (err) 3153 goto fail1; 3154 } 3155 3156 /* find a route */ 3157 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n", 3158 __func__, &laddr->sin_addr, ntohs(laddr->sin_port), 3159 ra, ntohs(raddr->sin_port)); 3160 ep->dst = find_route(dev, laddr->sin_addr.s_addr, 3161 raddr->sin_addr.s_addr, laddr->sin_port, 3162 raddr->sin_port, 0); 3163 } else { 3164 iptype = 6; 3165 ra = (__u8 *)&raddr6->sin6_addr; 3166 3167 /* 3168 * Handle loopback requests to INADDR_ANY. 3169 */ 3170 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) { 3171 err = pick_local_ip6addrs(dev, cm_id); 3172 if (err) 3173 goto fail1; 3174 } 3175 3176 /* find a route */ 3177 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n", 3178 __func__, laddr6->sin6_addr.s6_addr, 3179 ntohs(laddr6->sin6_port), 3180 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port)); 3181 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr, 3182 raddr6->sin6_addr.s6_addr, 3183 laddr6->sin6_port, raddr6->sin6_port, 0, 3184 raddr6->sin6_scope_id); 3185 } 3186 if (!ep->dst) { 3187 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__); 3188 err = -EHOSTUNREACH; 3189 goto fail2; 3190 } 3191 3192 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true); 3193 if (err) { 3194 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__); 3195 goto fail3; 3196 } 3197 3198 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n", 3199 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid, 3200 ep->l2t->idx); 3201 3202 state_set(&ep->com, CONNECTING); 3203 ep->tos = 0; 3204 3205 /* send connect request to rnic */ 3206 err = send_connect(ep); 3207 if (!err) 3208 goto out; 3209 3210 cxgb4_l2t_release(ep->l2t); 3211 fail3: 3212 dst_release(ep->dst); 3213 fail2: 3214 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid); 3215 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid); 3216 fail1: 3217 cm_id->rem_ref(cm_id); 3218 c4iw_put_ep(&ep->com); 3219 out: 3220 return err; 3221 } 3222 3223 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep) 3224 { 3225 int err; 3226 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) 3227 &ep->com.mapped_local_addr; 3228 3229 c4iw_init_wr_wait(&ep->com.wr_wait); 3230 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0], 3231 ep->stid, &sin6->sin6_addr, 3232 sin6->sin6_port, 3233 ep->com.dev->rdev.lldi.rxq_ids[0]); 3234 if (!err) 3235 err = c4iw_wait_for_reply(&ep->com.dev->rdev, 3236 &ep->com.wr_wait, 3237 0, 0, __func__); 3238 else if (err > 0) 3239 err = net_xmit_errno(err); 3240 if (err) 3241 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n", 3242 err, ep->stid, 3243 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port)); 3244 else 3245 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0], 3246 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3247 return err; 3248 } 3249 3250 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep) 3251 { 3252 int err; 3253 struct sockaddr_in *sin = (struct sockaddr_in *) 3254 &ep->com.mapped_local_addr; 3255 3256 if (dev->rdev.lldi.enable_fw_ofld_conn) { 3257 do { 3258 err = cxgb4_create_server_filter( 3259 ep->com.dev->rdev.lldi.ports[0], ep->stid, 3260 sin->sin_addr.s_addr, sin->sin_port, 0, 3261 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0); 3262 if (err == -EBUSY) { 3263 set_current_state(TASK_UNINTERRUPTIBLE); 3264 schedule_timeout(usecs_to_jiffies(100)); 3265 } 3266 } while (err == -EBUSY); 3267 } else { 3268 c4iw_init_wr_wait(&ep->com.wr_wait); 3269 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0], 3270 ep->stid, sin->sin_addr.s_addr, sin->sin_port, 3271 0, ep->com.dev->rdev.lldi.rxq_ids[0]); 3272 if (!err) 3273 err = c4iw_wait_for_reply(&ep->com.dev->rdev, 3274 &ep->com.wr_wait, 3275 0, 0, __func__); 3276 else if (err > 0) 3277 err = net_xmit_errno(err); 3278 } 3279 if (err) 3280 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n" 3281 , err, ep->stid, 3282 &sin->sin_addr, ntohs(sin->sin_port)); 3283 return err; 3284 } 3285 3286 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog) 3287 { 3288 int err = 0; 3289 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 3290 struct c4iw_listen_ep *ep; 3291 struct iwpm_dev_data pm_reg_msg; 3292 struct iwpm_sa_data pm_msg; 3293 int iwpm_err = 0; 3294 3295 might_sleep(); 3296 3297 ep = alloc_ep(sizeof(*ep), GFP_KERNEL); 3298 if (!ep) { 3299 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__); 3300 err = -ENOMEM; 3301 goto fail1; 3302 } 3303 PDBG("%s ep %p\n", __func__, ep); 3304 cm_id->add_ref(cm_id); 3305 ep->com.cm_id = cm_id; 3306 ep->com.dev = dev; 3307 ep->backlog = backlog; 3308 memcpy(&ep->com.local_addr, &cm_id->local_addr, 3309 sizeof(ep->com.local_addr)); 3310 3311 /* 3312 * Allocate a server TID. 3313 */ 3314 if (dev->rdev.lldi.enable_fw_ofld_conn && 3315 ep->com.local_addr.ss_family == AF_INET) 3316 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids, 3317 cm_id->local_addr.ss_family, ep); 3318 else 3319 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids, 3320 cm_id->local_addr.ss_family, ep); 3321 3322 if (ep->stid == -1) { 3323 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__); 3324 err = -ENOMEM; 3325 goto fail2; 3326 } 3327 insert_handle(dev, &dev->stid_idr, ep, ep->stid); 3328 3329 /* No port mapper available, go with the specified info */ 3330 memcpy(&ep->com.mapped_local_addr, &cm_id->local_addr, 3331 sizeof(ep->com.mapped_local_addr)); 3332 3333 c4iw_form_reg_msg(dev, &pm_reg_msg); 3334 iwpm_err = iwpm_register_pid(&pm_reg_msg, RDMA_NL_C4IW); 3335 if (iwpm_err) { 3336 PDBG("%s: Port Mapper reg pid fail (err = %d).\n", 3337 __func__, iwpm_err); 3338 } 3339 if (iwpm_valid_pid() && !iwpm_err) { 3340 memcpy(&pm_msg.loc_addr, &ep->com.local_addr, 3341 sizeof(ep->com.local_addr)); 3342 iwpm_err = iwpm_add_mapping(&pm_msg, RDMA_NL_C4IW); 3343 if (iwpm_err) 3344 PDBG("%s: Port Mapper query fail (err = %d).\n", 3345 __func__, iwpm_err); 3346 else 3347 memcpy(&ep->com.mapped_local_addr, 3348 &pm_msg.mapped_loc_addr, 3349 sizeof(ep->com.mapped_local_addr)); 3350 } 3351 if (iwpm_create_mapinfo(&ep->com.local_addr, 3352 &ep->com.mapped_local_addr, RDMA_NL_C4IW)) { 3353 err = -ENOMEM; 3354 goto fail3; 3355 } 3356 print_addr(&ep->com, __func__, "add_mapping/create_mapinfo"); 3357 3358 set_bit(RELEASE_MAPINFO, &ep->com.flags); 3359 state_set(&ep->com, LISTEN); 3360 if (ep->com.local_addr.ss_family == AF_INET) 3361 err = create_server4(dev, ep); 3362 else 3363 err = create_server6(dev, ep); 3364 if (!err) { 3365 cm_id->provider_data = ep; 3366 goto out; 3367 } 3368 3369 fail3: 3370 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, 3371 ep->com.local_addr.ss_family); 3372 fail2: 3373 cm_id->rem_ref(cm_id); 3374 c4iw_put_ep(&ep->com); 3375 fail1: 3376 out: 3377 return err; 3378 } 3379 3380 int c4iw_destroy_listen(struct iw_cm_id *cm_id) 3381 { 3382 int err; 3383 struct c4iw_listen_ep *ep = to_listen_ep(cm_id); 3384 3385 PDBG("%s ep %p\n", __func__, ep); 3386 3387 might_sleep(); 3388 state_set(&ep->com, DEAD); 3389 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn && 3390 ep->com.local_addr.ss_family == AF_INET) { 3391 err = cxgb4_remove_server_filter( 3392 ep->com.dev->rdev.lldi.ports[0], ep->stid, 3393 ep->com.dev->rdev.lldi.rxq_ids[0], 0); 3394 } else { 3395 struct sockaddr_in6 *sin6; 3396 c4iw_init_wr_wait(&ep->com.wr_wait); 3397 err = cxgb4_remove_server( 3398 ep->com.dev->rdev.lldi.ports[0], ep->stid, 3399 ep->com.dev->rdev.lldi.rxq_ids[0], 0); 3400 if (err) 3401 goto done; 3402 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait, 3403 0, 0, __func__); 3404 sin6 = (struct sockaddr_in6 *)&ep->com.mapped_local_addr; 3405 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 3406 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3407 } 3408 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid); 3409 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, 3410 ep->com.local_addr.ss_family); 3411 done: 3412 cm_id->rem_ref(cm_id); 3413 c4iw_put_ep(&ep->com); 3414 return err; 3415 } 3416 3417 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) 3418 { 3419 int ret = 0; 3420 int close = 0; 3421 int fatal = 0; 3422 struct c4iw_rdev *rdev; 3423 3424 mutex_lock(&ep->com.mutex); 3425 3426 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep, 3427 states[ep->com.state], abrupt); 3428 3429 rdev = &ep->com.dev->rdev; 3430 if (c4iw_fatal_error(rdev)) { 3431 fatal = 1; 3432 close_complete_upcall(ep, -EIO); 3433 ep->com.state = DEAD; 3434 } 3435 switch (ep->com.state) { 3436 case MPA_REQ_WAIT: 3437 case MPA_REQ_SENT: 3438 case MPA_REQ_RCVD: 3439 case MPA_REP_SENT: 3440 case FPDU_MODE: 3441 close = 1; 3442 if (abrupt) 3443 ep->com.state = ABORTING; 3444 else { 3445 ep->com.state = CLOSING; 3446 start_ep_timer(ep); 3447 } 3448 set_bit(CLOSE_SENT, &ep->com.flags); 3449 break; 3450 case CLOSING: 3451 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) { 3452 close = 1; 3453 if (abrupt) { 3454 (void)stop_ep_timer(ep); 3455 ep->com.state = ABORTING; 3456 } else 3457 ep->com.state = MORIBUND; 3458 } 3459 break; 3460 case MORIBUND: 3461 case ABORTING: 3462 case DEAD: 3463 PDBG("%s ignoring disconnect ep %p state %u\n", 3464 __func__, ep, ep->com.state); 3465 break; 3466 default: 3467 BUG(); 3468 break; 3469 } 3470 3471 if (close) { 3472 if (abrupt) { 3473 set_bit(EP_DISC_ABORT, &ep->com.history); 3474 close_complete_upcall(ep, -ECONNRESET); 3475 ret = send_abort(ep, NULL, gfp); 3476 } else { 3477 set_bit(EP_DISC_CLOSE, &ep->com.history); 3478 ret = send_halfclose(ep, gfp); 3479 } 3480 if (ret) 3481 fatal = 1; 3482 } 3483 mutex_unlock(&ep->com.mutex); 3484 if (fatal) 3485 release_ep_resources(ep); 3486 return ret; 3487 } 3488 3489 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb, 3490 struct cpl_fw6_msg_ofld_connection_wr_rpl *req) 3491 { 3492 struct c4iw_ep *ep; 3493 int atid = be32_to_cpu(req->tid); 3494 3495 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids, 3496 (__force u32) req->tid); 3497 if (!ep) 3498 return; 3499 3500 switch (req->retval) { 3501 case FW_ENOMEM: 3502 set_bit(ACT_RETRY_NOMEM, &ep->com.history); 3503 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) { 3504 send_fw_act_open_req(ep, atid); 3505 return; 3506 } 3507 case FW_EADDRINUSE: 3508 set_bit(ACT_RETRY_INUSE, &ep->com.history); 3509 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) { 3510 send_fw_act_open_req(ep, atid); 3511 return; 3512 } 3513 break; 3514 default: 3515 pr_info("%s unexpected ofld conn wr retval %d\n", 3516 __func__, req->retval); 3517 break; 3518 } 3519 pr_err("active ofld_connect_wr failure %d atid %d\n", 3520 req->retval, atid); 3521 mutex_lock(&dev->rdev.stats.lock); 3522 dev->rdev.stats.act_ofld_conn_fails++; 3523 mutex_unlock(&dev->rdev.stats.lock); 3524 connect_reply_upcall(ep, status2errno(req->retval)); 3525 state_set(&ep->com, DEAD); 3526 if (ep->com.remote_addr.ss_family == AF_INET6) { 3527 struct sockaddr_in6 *sin6 = 3528 (struct sockaddr_in6 *)&ep->com.mapped_local_addr; 3529 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0], 3530 (const u32 *)&sin6->sin6_addr.s6_addr, 1); 3531 } 3532 remove_handle(dev, &dev->atid_idr, atid); 3533 cxgb4_free_atid(dev->rdev.lldi.tids, atid); 3534 dst_release(ep->dst); 3535 cxgb4_l2t_release(ep->l2t); 3536 c4iw_put_ep(&ep->com); 3537 } 3538 3539 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb, 3540 struct cpl_fw6_msg_ofld_connection_wr_rpl *req) 3541 { 3542 struct sk_buff *rpl_skb; 3543 struct cpl_pass_accept_req *cpl; 3544 int ret; 3545 3546 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie; 3547 BUG_ON(!rpl_skb); 3548 if (req->retval) { 3549 PDBG("%s passive open failure %d\n", __func__, req->retval); 3550 mutex_lock(&dev->rdev.stats.lock); 3551 dev->rdev.stats.pas_ofld_conn_fails++; 3552 mutex_unlock(&dev->rdev.stats.lock); 3553 kfree_skb(rpl_skb); 3554 } else { 3555 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb); 3556 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 3557 (__force u32) htonl( 3558 (__force u32) req->tid))); 3559 ret = pass_accept_req(dev, rpl_skb); 3560 if (!ret) 3561 kfree_skb(rpl_skb); 3562 } 3563 return; 3564 } 3565 3566 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb) 3567 { 3568 struct cpl_fw6_msg *rpl = cplhdr(skb); 3569 struct cpl_fw6_msg_ofld_connection_wr_rpl *req; 3570 3571 switch (rpl->type) { 3572 case FW6_TYPE_CQE: 3573 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]); 3574 break; 3575 case FW6_TYPE_OFLD_CONNECTION_WR_RPL: 3576 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data; 3577 switch (req->t_state) { 3578 case TCP_SYN_SENT: 3579 active_ofld_conn_reply(dev, skb, req); 3580 break; 3581 case TCP_SYN_RECV: 3582 passive_ofld_conn_reply(dev, skb, req); 3583 break; 3584 default: 3585 pr_err("%s unexpected ofld conn wr state %d\n", 3586 __func__, req->t_state); 3587 break; 3588 } 3589 break; 3590 } 3591 return 0; 3592 } 3593 3594 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos) 3595 { 3596 u32 l2info; 3597 u16 vlantag, len, hdr_len, eth_hdr_len; 3598 u8 intf; 3599 struct cpl_rx_pkt *cpl = cplhdr(skb); 3600 struct cpl_pass_accept_req *req; 3601 struct tcp_options_received tmp_opt; 3602 struct c4iw_dev *dev; 3603 3604 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *))); 3605 /* Store values from cpl_rx_pkt in temporary location. */ 3606 vlantag = (__force u16) cpl->vlan; 3607 len = (__force u16) cpl->len; 3608 l2info = (__force u32) cpl->l2info; 3609 hdr_len = (__force u16) cpl->hdr_len; 3610 intf = cpl->iff; 3611 3612 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header)); 3613 3614 /* 3615 * We need to parse the TCP options from SYN packet. 3616 * to generate cpl_pass_accept_req. 3617 */ 3618 memset(&tmp_opt, 0, sizeof(tmp_opt)); 3619 tcp_clear_options(&tmp_opt); 3620 tcp_parse_options(skb, &tmp_opt, 0, NULL); 3621 3622 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req)); 3623 memset(req, 0, sizeof(*req)); 3624 req->l2info = cpu_to_be16(SYN_INTF_V(intf) | 3625 SYN_MAC_IDX_V(RX_MACIDX_G( 3626 (__force int) htonl(l2info))) | 3627 SYN_XACT_MATCH_F); 3628 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ? 3629 RX_ETHHDR_LEN_G((__force int)htonl(l2info)) : 3630 RX_T5_ETHHDR_LEN_G((__force int)htonl(l2info)); 3631 req->hdr_len = cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G( 3632 (__force int) htonl(l2info))) | 3633 TCP_HDR_LEN_V(RX_TCPHDR_LEN_G( 3634 (__force int) htons(hdr_len))) | 3635 IP_HDR_LEN_V(RX_IPHDR_LEN_G( 3636 (__force int) htons(hdr_len))) | 3637 ETH_HDR_LEN_V(RX_ETHHDR_LEN_G(eth_hdr_len))); 3638 req->vlan = (__force __be16) vlantag; 3639 req->len = (__force __be16) len; 3640 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) | 3641 PASS_OPEN_TOS_V(tos)); 3642 req->tcpopt.mss = htons(tmp_opt.mss_clamp); 3643 if (tmp_opt.wscale_ok) 3644 req->tcpopt.wsf = tmp_opt.snd_wscale; 3645 req->tcpopt.tstamp = tmp_opt.saw_tstamp; 3646 if (tmp_opt.sack_ok) 3647 req->tcpopt.sack = 1; 3648 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0)); 3649 return; 3650 } 3651 3652 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb, 3653 __be32 laddr, __be16 lport, 3654 __be32 raddr, __be16 rport, 3655 u32 rcv_isn, u32 filter, u16 window, 3656 u32 rss_qid, u8 port_id) 3657 { 3658 struct sk_buff *req_skb; 3659 struct fw_ofld_connection_wr *req; 3660 struct cpl_pass_accept_req *cpl = cplhdr(skb); 3661 int ret; 3662 3663 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL); 3664 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req)); 3665 memset(req, 0, sizeof(*req)); 3666 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F); 3667 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16))); 3668 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F); 3669 req->le.filter = (__force __be32) filter; 3670 req->le.lport = lport; 3671 req->le.pport = rport; 3672 req->le.u.ipv4.lip = laddr; 3673 req->le.u.ipv4.pip = raddr; 3674 req->tcb.rcv_nxt = htonl(rcv_isn + 1); 3675 req->tcb.rcv_adv = htons(window); 3676 req->tcb.t_state_to_astid = 3677 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) | 3678 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) | 3679 FW_OFLD_CONNECTION_WR_ASTID_V( 3680 PASS_OPEN_TID_G(ntohl(cpl->tos_stid)))); 3681 3682 /* 3683 * We store the qid in opt2 which will be used by the firmware 3684 * to send us the wr response. 3685 */ 3686 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid)); 3687 3688 /* 3689 * We initialize the MSS index in TCB to 0xF. 3690 * So that when driver sends cpl_pass_accept_rpl 3691 * TCB picks up the correct value. If this was 0 3692 * TP will ignore any value > 0 for MSS index. 3693 */ 3694 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF)); 3695 req->cookie = (uintptr_t)skb; 3696 3697 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id); 3698 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb); 3699 if (ret < 0) { 3700 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__, 3701 ret); 3702 kfree_skb(skb); 3703 kfree_skb(req_skb); 3704 } 3705 } 3706 3707 /* 3708 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt 3709 * messages when a filter is being used instead of server to 3710 * redirect a syn packet. When packets hit filter they are redirected 3711 * to the offload queue and driver tries to establish the connection 3712 * using firmware work request. 3713 */ 3714 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb) 3715 { 3716 int stid; 3717 unsigned int filter; 3718 struct ethhdr *eh = NULL; 3719 struct vlan_ethhdr *vlan_eh = NULL; 3720 struct iphdr *iph; 3721 struct tcphdr *tcph; 3722 struct rss_header *rss = (void *)skb->data; 3723 struct cpl_rx_pkt *cpl = (void *)skb->data; 3724 struct cpl_pass_accept_req *req = (void *)(rss + 1); 3725 struct l2t_entry *e; 3726 struct dst_entry *dst; 3727 struct c4iw_ep *lep; 3728 u16 window; 3729 struct port_info *pi; 3730 struct net_device *pdev; 3731 u16 rss_qid, eth_hdr_len; 3732 int step; 3733 u32 tx_chan; 3734 struct neighbour *neigh; 3735 3736 /* Drop all non-SYN packets */ 3737 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F))) 3738 goto reject; 3739 3740 /* 3741 * Drop all packets which did not hit the filter. 3742 * Unlikely to happen. 3743 */ 3744 if (!(rss->filter_hit && rss->filter_tid)) 3745 goto reject; 3746 3747 /* 3748 * Calculate the server tid from filter hit index from cpl_rx_pkt. 3749 */ 3750 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val); 3751 3752 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid); 3753 if (!lep) { 3754 PDBG("%s connect request on invalid stid %d\n", __func__, stid); 3755 goto reject; 3756 } 3757 3758 eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ? 3759 RX_ETHHDR_LEN_G(htonl(cpl->l2info)) : 3760 RX_T5_ETHHDR_LEN_G(htonl(cpl->l2info)); 3761 if (eth_hdr_len == ETH_HLEN) { 3762 eh = (struct ethhdr *)(req + 1); 3763 iph = (struct iphdr *)(eh + 1); 3764 } else { 3765 vlan_eh = (struct vlan_ethhdr *)(req + 1); 3766 iph = (struct iphdr *)(vlan_eh + 1); 3767 skb->vlan_tci = ntohs(cpl->vlan); 3768 } 3769 3770 if (iph->version != 0x4) 3771 goto reject; 3772 3773 tcph = (struct tcphdr *)(iph + 1); 3774 skb_set_network_header(skb, (void *)iph - (void *)rss); 3775 skb_set_transport_header(skb, (void *)tcph - (void *)rss); 3776 skb_get(skb); 3777 3778 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__, 3779 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr), 3780 ntohs(tcph->source), iph->tos); 3781 3782 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source, 3783 iph->tos); 3784 if (!dst) { 3785 pr_err("%s - failed to find dst entry!\n", 3786 __func__); 3787 goto reject; 3788 } 3789 neigh = dst_neigh_lookup_skb(dst, skb); 3790 3791 if (!neigh) { 3792 pr_err("%s - failed to allocate neigh!\n", 3793 __func__); 3794 goto free_dst; 3795 } 3796 3797 if (neigh->dev->flags & IFF_LOOPBACK) { 3798 pdev = ip_dev_find(&init_net, iph->daddr); 3799 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh, 3800 pdev, 0); 3801 pi = (struct port_info *)netdev_priv(pdev); 3802 tx_chan = cxgb4_port_chan(pdev); 3803 dev_put(pdev); 3804 } else { 3805 pdev = get_real_dev(neigh->dev); 3806 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh, 3807 pdev, 0); 3808 pi = (struct port_info *)netdev_priv(pdev); 3809 tx_chan = cxgb4_port_chan(pdev); 3810 } 3811 neigh_release(neigh); 3812 if (!e) { 3813 pr_err("%s - failed to allocate l2t entry!\n", 3814 __func__); 3815 goto free_dst; 3816 } 3817 3818 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan; 3819 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step]; 3820 window = (__force u16) htons((__force u16)tcph->window); 3821 3822 /* Calcuate filter portion for LE region. */ 3823 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple( 3824 dev->rdev.lldi.ports[0], 3825 e)); 3826 3827 /* 3828 * Synthesize the cpl_pass_accept_req. We have everything except the 3829 * TID. Once firmware sends a reply with TID we update the TID field 3830 * in cpl and pass it through the regular cpl_pass_accept_req path. 3831 */ 3832 build_cpl_pass_accept_req(skb, stid, iph->tos); 3833 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr, 3834 tcph->source, ntohl(tcph->seq), filter, window, 3835 rss_qid, pi->port_id); 3836 cxgb4_l2t_release(e); 3837 free_dst: 3838 dst_release(dst); 3839 reject: 3840 return 0; 3841 } 3842 3843 /* 3844 * These are the real handlers that are called from a 3845 * work queue. 3846 */ 3847 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = { 3848 [CPL_ACT_ESTABLISH] = act_establish, 3849 [CPL_ACT_OPEN_RPL] = act_open_rpl, 3850 [CPL_RX_DATA] = rx_data, 3851 [CPL_ABORT_RPL_RSS] = abort_rpl, 3852 [CPL_ABORT_RPL] = abort_rpl, 3853 [CPL_PASS_OPEN_RPL] = pass_open_rpl, 3854 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl, 3855 [CPL_PASS_ACCEPT_REQ] = pass_accept_req, 3856 [CPL_PASS_ESTABLISH] = pass_establish, 3857 [CPL_PEER_CLOSE] = peer_close, 3858 [CPL_ABORT_REQ_RSS] = peer_abort, 3859 [CPL_CLOSE_CON_RPL] = close_con_rpl, 3860 [CPL_RDMA_TERMINATE] = terminate, 3861 [CPL_FW4_ACK] = fw4_ack, 3862 [CPL_FW6_MSG] = deferred_fw6_msg, 3863 [CPL_RX_PKT] = rx_pkt 3864 }; 3865 3866 static void process_timeout(struct c4iw_ep *ep) 3867 { 3868 struct c4iw_qp_attributes attrs; 3869 int abort = 1; 3870 3871 mutex_lock(&ep->com.mutex); 3872 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid, 3873 ep->com.state); 3874 set_bit(TIMEDOUT, &ep->com.history); 3875 switch (ep->com.state) { 3876 case MPA_REQ_SENT: 3877 __state_set(&ep->com, ABORTING); 3878 connect_reply_upcall(ep, -ETIMEDOUT); 3879 break; 3880 case MPA_REQ_WAIT: 3881 __state_set(&ep->com, ABORTING); 3882 break; 3883 case CLOSING: 3884 case MORIBUND: 3885 if (ep->com.cm_id && ep->com.qp) { 3886 attrs.next_state = C4IW_QP_STATE_ERROR; 3887 c4iw_modify_qp(ep->com.qp->rhp, 3888 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, 3889 &attrs, 1); 3890 } 3891 __state_set(&ep->com, ABORTING); 3892 close_complete_upcall(ep, -ETIMEDOUT); 3893 break; 3894 case ABORTING: 3895 case DEAD: 3896 3897 /* 3898 * These states are expected if the ep timed out at the same 3899 * time as another thread was calling stop_ep_timer(). 3900 * So we silently do nothing for these states. 3901 */ 3902 abort = 0; 3903 break; 3904 default: 3905 WARN(1, "%s unexpected state ep %p tid %u state %u\n", 3906 __func__, ep, ep->hwtid, ep->com.state); 3907 abort = 0; 3908 } 3909 if (abort) 3910 abort_connection(ep, NULL, GFP_KERNEL); 3911 mutex_unlock(&ep->com.mutex); 3912 c4iw_put_ep(&ep->com); 3913 } 3914 3915 static void process_timedout_eps(void) 3916 { 3917 struct c4iw_ep *ep; 3918 3919 spin_lock_irq(&timeout_lock); 3920 while (!list_empty(&timeout_list)) { 3921 struct list_head *tmp; 3922 3923 tmp = timeout_list.next; 3924 list_del(tmp); 3925 tmp->next = NULL; 3926 tmp->prev = NULL; 3927 spin_unlock_irq(&timeout_lock); 3928 ep = list_entry(tmp, struct c4iw_ep, entry); 3929 process_timeout(ep); 3930 spin_lock_irq(&timeout_lock); 3931 } 3932 spin_unlock_irq(&timeout_lock); 3933 } 3934 3935 static void process_work(struct work_struct *work) 3936 { 3937 struct sk_buff *skb = NULL; 3938 struct c4iw_dev *dev; 3939 struct cpl_act_establish *rpl; 3940 unsigned int opcode; 3941 int ret; 3942 3943 process_timedout_eps(); 3944 while ((skb = skb_dequeue(&rxq))) { 3945 rpl = cplhdr(skb); 3946 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *))); 3947 opcode = rpl->ot.opcode; 3948 3949 BUG_ON(!work_handlers[opcode]); 3950 ret = work_handlers[opcode](dev, skb); 3951 if (!ret) 3952 kfree_skb(skb); 3953 process_timedout_eps(); 3954 } 3955 } 3956 3957 static DECLARE_WORK(skb_work, process_work); 3958 3959 static void ep_timeout(unsigned long arg) 3960 { 3961 struct c4iw_ep *ep = (struct c4iw_ep *)arg; 3962 int kickit = 0; 3963 3964 spin_lock(&timeout_lock); 3965 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 3966 /* 3967 * Only insert if it is not already on the list. 3968 */ 3969 if (!ep->entry.next) { 3970 list_add_tail(&ep->entry, &timeout_list); 3971 kickit = 1; 3972 } 3973 } 3974 spin_unlock(&timeout_lock); 3975 if (kickit) 3976 queue_work(workq, &skb_work); 3977 } 3978 3979 /* 3980 * All the CM events are handled on a work queue to have a safe context. 3981 */ 3982 static int sched(struct c4iw_dev *dev, struct sk_buff *skb) 3983 { 3984 3985 /* 3986 * Save dev in the skb->cb area. 3987 */ 3988 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev; 3989 3990 /* 3991 * Queue the skb and schedule the worker thread. 3992 */ 3993 skb_queue_tail(&rxq, skb); 3994 queue_work(workq, &skb_work); 3995 return 0; 3996 } 3997 3998 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb) 3999 { 4000 struct cpl_set_tcb_rpl *rpl = cplhdr(skb); 4001 4002 if (rpl->status != CPL_ERR_NONE) { 4003 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u " 4004 "for tid %u\n", rpl->status, GET_TID(rpl)); 4005 } 4006 kfree_skb(skb); 4007 return 0; 4008 } 4009 4010 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb) 4011 { 4012 struct cpl_fw6_msg *rpl = cplhdr(skb); 4013 struct c4iw_wr_wait *wr_waitp; 4014 int ret; 4015 4016 PDBG("%s type %u\n", __func__, rpl->type); 4017 4018 switch (rpl->type) { 4019 case FW6_TYPE_WR_RPL: 4020 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff); 4021 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1]; 4022 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret); 4023 if (wr_waitp) 4024 c4iw_wake_up(wr_waitp, ret ? -ret : 0); 4025 kfree_skb(skb); 4026 break; 4027 case FW6_TYPE_CQE: 4028 case FW6_TYPE_OFLD_CONNECTION_WR_RPL: 4029 sched(dev, skb); 4030 break; 4031 default: 4032 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__, 4033 rpl->type); 4034 kfree_skb(skb); 4035 break; 4036 } 4037 return 0; 4038 } 4039 4040 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb) 4041 { 4042 struct cpl_abort_req_rss *req = cplhdr(skb); 4043 struct c4iw_ep *ep; 4044 struct tid_info *t = dev->rdev.lldi.tids; 4045 unsigned int tid = GET_TID(req); 4046 4047 ep = lookup_tid(t, tid); 4048 if (!ep) { 4049 printk(KERN_WARNING MOD 4050 "Abort on non-existent endpoint, tid %d\n", tid); 4051 kfree_skb(skb); 4052 return 0; 4053 } 4054 if (is_neg_adv(req->status)) { 4055 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n", 4056 __func__, ep->hwtid, req->status, 4057 neg_adv_str(req->status)); 4058 ep->stats.abort_neg_adv++; 4059 dev->rdev.stats.neg_adv++; 4060 kfree_skb(skb); 4061 return 0; 4062 } 4063 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid, 4064 ep->com.state); 4065 4066 /* 4067 * Wake up any threads in rdma_init() or rdma_fini(). 4068 * However, if we are on MPAv2 and want to retry with MPAv1 4069 * then, don't wake up yet. 4070 */ 4071 if (mpa_rev == 2 && !ep->tried_with_mpa_v1) { 4072 if (ep->com.state != MPA_REQ_SENT) 4073 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 4074 } else 4075 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 4076 sched(dev, skb); 4077 return 0; 4078 } 4079 4080 /* 4081 * Most upcalls from the T4 Core go to sched() to 4082 * schedule the processing on a work queue. 4083 */ 4084 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = { 4085 [CPL_ACT_ESTABLISH] = sched, 4086 [CPL_ACT_OPEN_RPL] = sched, 4087 [CPL_RX_DATA] = sched, 4088 [CPL_ABORT_RPL_RSS] = sched, 4089 [CPL_ABORT_RPL] = sched, 4090 [CPL_PASS_OPEN_RPL] = sched, 4091 [CPL_CLOSE_LISTSRV_RPL] = sched, 4092 [CPL_PASS_ACCEPT_REQ] = sched, 4093 [CPL_PASS_ESTABLISH] = sched, 4094 [CPL_PEER_CLOSE] = sched, 4095 [CPL_CLOSE_CON_RPL] = sched, 4096 [CPL_ABORT_REQ_RSS] = peer_abort_intr, 4097 [CPL_RDMA_TERMINATE] = sched, 4098 [CPL_FW4_ACK] = sched, 4099 [CPL_SET_TCB_RPL] = set_tcb_rpl, 4100 [CPL_FW6_MSG] = fw6_msg, 4101 [CPL_RX_PKT] = sched 4102 }; 4103 4104 int __init c4iw_cm_init(void) 4105 { 4106 spin_lock_init(&timeout_lock); 4107 skb_queue_head_init(&rxq); 4108 4109 workq = create_singlethread_workqueue("iw_cxgb4"); 4110 if (!workq) 4111 return -ENOMEM; 4112 4113 return 0; 4114 } 4115 4116 void c4iw_cm_term(void) 4117 { 4118 WARN_ON(!list_empty(&timeout_list)); 4119 flush_workqueue(workq); 4120 destroy_workqueue(workq); 4121 } 4122