projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
RDMA/iw_cxgb4: remove abort_connection() usage from accept/reject
[deliverable/linux.git] / drivers / infiniband / hw / cxgb4 / cm.c
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 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
149
150 static LIST_HEAD(timeout_list);
151 static spinlock_t timeout_lock;
152
153 static void deref_qp(struct c4iw_ep *ep)
154 {
155 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
156 clear_bit(QP_REFERENCED, &ep->com.flags);
157 }
158
159 static void ref_qp(struct c4iw_ep *ep)
160 {
161 set_bit(QP_REFERENCED, &ep->com.flags);
162 c4iw_qp_add_ref(&ep->com.qp->ibqp);
163 }
164
165 static void start_ep_timer(struct c4iw_ep *ep)
166 {
167 PDBG("%s ep %p\n", __func__, ep);
168 if (timer_pending(&ep->timer)) {
169 pr_err("%s timer already started! ep %p\n",
170 __func__, ep);
171 return;
172 }
173 clear_bit(TIMEOUT, &ep->com.flags);
174 c4iw_get_ep(&ep->com);
175 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
176 ep->timer.data = (unsigned long)ep;
177 ep->timer.function = ep_timeout;
178 add_timer(&ep->timer);
179 }
180
181 static int stop_ep_timer(struct c4iw_ep *ep)
182 {
183 PDBG("%s ep %p stopping\n", __func__, ep);
184 del_timer_sync(&ep->timer);
185 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
186 c4iw_put_ep(&ep->com);
187 return 0;
188 }
189 return 1;
190 }
191
192 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
193 struct l2t_entry *l2e)
194 {
195 int error = 0;
196
197 if (c4iw_fatal_error(rdev)) {
198 kfree_skb(skb);
199 PDBG("%s - device in error state - dropping\n", __func__);
200 return -EIO;
201 }
202 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
203 if (error < 0)
204 kfree_skb(skb);
205 return error < 0 ? error : 0;
206 }
207
208 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
209 {
210 int error = 0;
211
212 if (c4iw_fatal_error(rdev)) {
213 kfree_skb(skb);
214 PDBG("%s - device in error state - dropping\n", __func__);
215 return -EIO;
216 }
217 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
218 if (error < 0)
219 kfree_skb(skb);
220 return error < 0 ? error : 0;
221 }
222
223 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
224 {
225 struct cpl_tid_release *req;
226
227 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
228 if (!skb)
229 return;
230 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
231 INIT_TP_WR(req, hwtid);
232 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
233 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
234 c4iw_ofld_send(rdev, skb);
235 return;
236 }
237
238 static void set_emss(struct c4iw_ep *ep, u16 opt)
239 {
240 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
241 ((AF_INET == ep->com.remote_addr.ss_family) ?
242 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
243 sizeof(struct tcphdr);
244 ep->mss = ep->emss;
245 if (TCPOPT_TSTAMP_G(opt))
246 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
247 if (ep->emss < 128)
248 ep->emss = 128;
249 if (ep->emss & 7)
250 PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
251 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
252 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
253 ep->mss, ep->emss);
254 }
255
256 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
257 {
258 enum c4iw_ep_state state;
259
260 mutex_lock(&epc->mutex);
261 state = epc->state;
262 mutex_unlock(&epc->mutex);
263 return state;
264 }
265
266 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
267 {
268 epc->state = new;
269 }
270
271 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
272 {
273 mutex_lock(&epc->mutex);
274 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
275 __state_set(epc, new);
276 mutex_unlock(&epc->mutex);
277 return;
278 }
279
280 static void *alloc_ep(int size, gfp_t gfp)
281 {
282 struct c4iw_ep_common *epc;
283
284 epc = kzalloc(size, gfp);
285 if (epc) {
286 kref_init(&epc->kref);
287 mutex_init(&epc->mutex);
288 c4iw_init_wr_wait(&epc->wr_wait);
289 }
290 PDBG("%s alloc ep %p\n", __func__, epc);
291 return epc;
292 }
293
294 void _c4iw_free_ep(struct kref *kref)
295 {
296 struct c4iw_ep *ep;
297
298 ep = container_of(kref, struct c4iw_ep, com.kref);
299 PDBG("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
300 if (test_bit(QP_REFERENCED, &ep->com.flags))
301 deref_qp(ep);
302 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
303 if (ep->com.remote_addr.ss_family == AF_INET6) {
304 struct sockaddr_in6 *sin6 =
305 (struct sockaddr_in6 *)
306 &ep->com.local_addr;
307
308 cxgb4_clip_release(
309 ep->com.dev->rdev.lldi.ports[0],
310 (const u32 *)&sin6->sin6_addr.s6_addr,
311 1);
312 }
313 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
314 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
315 dst_release(ep->dst);
316 cxgb4_l2t_release(ep->l2t);
317 }
318 kfree(ep);
319 }
320
321 static void release_ep_resources(struct c4iw_ep *ep)
322 {
323 set_bit(RELEASE_RESOURCES, &ep->com.flags);
324 c4iw_put_ep(&ep->com);
325 }
326
327 static int status2errno(int status)
328 {
329 switch (status) {
330 case CPL_ERR_NONE:
331 return 0;
332 case CPL_ERR_CONN_RESET:
333 return -ECONNRESET;
334 case CPL_ERR_ARP_MISS:
335 return -EHOSTUNREACH;
336 case CPL_ERR_CONN_TIMEDOUT:
337 return -ETIMEDOUT;
338 case CPL_ERR_TCAM_FULL:
339 return -ENOMEM;
340 case CPL_ERR_CONN_EXIST:
341 return -EADDRINUSE;
342 default:
343 return -EIO;
344 }
345 }
346
347 /*
348 * Try and reuse skbs already allocated...
349 */
350 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
351 {
352 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
353 skb_trim(skb, 0);
354 skb_get(skb);
355 skb_reset_transport_header(skb);
356 } else {
357 skb = alloc_skb(len, gfp);
358 }
359 t4_set_arp_err_handler(skb, NULL, NULL);
360 return skb;
361 }
362
363 static struct net_device *get_real_dev(struct net_device *egress_dev)
364 {
365 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
366 }
367
368 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
369 {
370 int i;
371
372 egress_dev = get_real_dev(egress_dev);
373 for (i = 0; i < dev->rdev.lldi.nports; i++)
374 if (dev->rdev.lldi.ports[i] == egress_dev)
375 return 1;
376 return 0;
377 }
378
379 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
380 __u8 *peer_ip, __be16 local_port,
381 __be16 peer_port, u8 tos,
382 __u32 sin6_scope_id)
383 {
384 struct dst_entry *dst = NULL;
385
386 if (IS_ENABLED(CONFIG_IPV6)) {
387 struct flowi6 fl6;
388
389 memset(&fl6, 0, sizeof(fl6));
390 memcpy(&fl6.daddr, peer_ip, 16);
391 memcpy(&fl6.saddr, local_ip, 16);
392 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
393 fl6.flowi6_oif = sin6_scope_id;
394 dst = ip6_route_output(&init_net, NULL, &fl6);
395 if (!dst)
396 goto out;
397 if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
398 !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
399 dst_release(dst);
400 dst = NULL;
401 }
402 }
403
404 out:
405 return dst;
406 }
407
408 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
409 __be32 peer_ip, __be16 local_port,
410 __be16 peer_port, u8 tos)
411 {
412 struct rtable *rt;
413 struct flowi4 fl4;
414 struct neighbour *n;
415
416 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
417 peer_port, local_port, IPPROTO_TCP,
418 tos, 0);
419 if (IS_ERR(rt))
420 return NULL;
421 n = dst_neigh_lookup(&rt->dst, &peer_ip);
422 if (!n)
423 return NULL;
424 if (!our_interface(dev, n->dev) &&
425 !(n->dev->flags & IFF_LOOPBACK)) {
426 neigh_release(n);
427 dst_release(&rt->dst);
428 return NULL;
429 }
430 neigh_release(n);
431 return &rt->dst;
432 }
433
434 static void arp_failure_discard(void *handle, struct sk_buff *skb)
435 {
436 pr_err(MOD "ARP failure\n");
437 kfree_skb(skb);
438 }
439
440 enum {
441 NUM_FAKE_CPLS = 1,
442 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
443 };
444
445 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
446 {
447 struct c4iw_ep *ep;
448
449 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
450 release_ep_resources(ep);
451 return 0;
452 }
453
454 /*
455 * Fake up a special CPL opcode and call sched() so process_work() will call
456 * _put_ep_safe() in a safe context to free the ep resources. This is needed
457 * because ARP error handlers are called in an ATOMIC context, and
458 * _c4iw_free_ep() needs to block.
459 */
460 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb)
461 {
462 struct cpl_act_establish *rpl = cplhdr(skb);
463
464 /* Set our special ARP_FAILURE opcode */
465 rpl->ot.opcode = FAKE_CPL_PUT_EP_SAFE;
466
467 /*
468 * Save ep in the skb->cb area, after where sched() will save the dev
469 * ptr.
470 */
471 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
472 sched(ep->com.dev, skb);
473 }
474
475 /* Handle an ARP failure for an accept */
476 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
477 {
478 struct c4iw_ep *ep = handle;
479
480 pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n",
481 ep->hwtid);
482
483 __state_set(&ep->com, DEAD);
484 queue_arp_failure_cpl(ep, skb);
485 }
486
487 /*
488 * Handle an ARP failure for an active open.
489 */
490 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
491 {
492 struct c4iw_ep *ep = handle;
493
494 printk(KERN_ERR MOD "ARP failure during connect\n");
495 connect_reply_upcall(ep, -EHOSTUNREACH);
496 __state_set(&ep->com, DEAD);
497 if (ep->com.remote_addr.ss_family == AF_INET6) {
498 struct sockaddr_in6 *sin6 =
499 (struct sockaddr_in6 *)&ep->com.local_addr;
500 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
501 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
502 }
503 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
504 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
505 queue_arp_failure_cpl(ep, skb);
506 }
507
508 /*
509 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
510 * and send it along.
511 */
512 static void abort_arp_failure(void *handle, struct sk_buff *skb)
513 {
514 struct c4iw_rdev *rdev = handle;
515 struct cpl_abort_req *req = cplhdr(skb);
516
517 PDBG("%s rdev %p\n", __func__, rdev);
518 req->cmd = CPL_ABORT_NO_RST;
519 c4iw_ofld_send(rdev, skb);
520 }
521
522 static int send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
523 {
524 unsigned int flowclen = 80;
525 struct fw_flowc_wr *flowc;
526 int i;
527 u16 vlan = ep->l2t->vlan;
528 int nparams;
529
530 if (vlan == CPL_L2T_VLAN_NONE)
531 nparams = 8;
532 else
533 nparams = 9;
534
535 skb = get_skb(skb, flowclen, GFP_KERNEL);
536 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
537
538 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
539 FW_FLOWC_WR_NPARAMS_V(nparams));
540 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(flowclen,
541 16)) | FW_WR_FLOWID_V(ep->hwtid));
542
543 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
544 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
545 (ep->com.dev->rdev.lldi.pf));
546 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
547 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
548 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
549 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
550 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
551 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
552 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
553 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
554 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
555 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
556 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
557 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
558 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
559 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
560 if (nparams == 9) {
561 u16 pri;
562
563 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
564 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
565 flowc->mnemval[8].val = cpu_to_be32(pri);
566 } else {
567 /* Pad WR to 16 byte boundary */
568 flowc->mnemval[8].mnemonic = 0;
569 flowc->mnemval[8].val = 0;
570 }
571 for (i = 0; i < 9; i++) {
572 flowc->mnemval[i].r4[0] = 0;
573 flowc->mnemval[i].r4[1] = 0;
574 flowc->mnemval[i].r4[2] = 0;
575 }
576
577 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
578 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
579 }
580
581 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
582 {
583 struct cpl_close_con_req *req;
584 struct sk_buff *skb;
585 int wrlen = roundup(sizeof *req, 16);
586
587 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
588 skb = get_skb(NULL, wrlen, gfp);
589 if (!skb) {
590 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
591 return -ENOMEM;
592 }
593 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
594 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
595 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
596 memset(req, 0, wrlen);
597 INIT_TP_WR(req, ep->hwtid);
598 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
599 ep->hwtid));
600 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
601 }
602
603 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
604 {
605 struct cpl_abort_req *req;
606 int wrlen = roundup(sizeof *req, 16);
607
608 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
609 skb = get_skb(skb, wrlen, gfp);
610 if (!skb) {
611 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
612 __func__);
613 return -ENOMEM;
614 }
615 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
616 t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
617 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
618 memset(req, 0, wrlen);
619 INIT_TP_WR(req, ep->hwtid);
620 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
621 req->cmd = CPL_ABORT_SEND_RST;
622 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
623 }
624
625 static void best_mtu(const unsigned short *mtus, unsigned short mtu,
626 unsigned int *idx, int use_ts, int ipv6)
627 {
628 unsigned short hdr_size = (ipv6 ?
629 sizeof(struct ipv6hdr) :
630 sizeof(struct iphdr)) +
631 sizeof(struct tcphdr) +
632 (use_ts ?
633 round_up(TCPOLEN_TIMESTAMP, 4) : 0);
634 unsigned short data_size = mtu - hdr_size;
635
636 cxgb4_best_aligned_mtu(mtus, hdr_size, data_size, 8, idx);
637 }
638
639 static int send_connect(struct c4iw_ep *ep)
640 {
641 struct cpl_act_open_req *req = NULL;
642 struct cpl_t5_act_open_req *t5req = NULL;
643 struct cpl_t6_act_open_req *t6req = NULL;
644 struct cpl_act_open_req6 *req6 = NULL;
645 struct cpl_t5_act_open_req6 *t5req6 = NULL;
646 struct cpl_t6_act_open_req6 *t6req6 = NULL;
647 struct sk_buff *skb;
648 u64 opt0;
649 u32 opt2;
650 unsigned int mtu_idx;
651 int wscale;
652 int win, sizev4, sizev6, wrlen;
653 struct sockaddr_in *la = (struct sockaddr_in *)
654 &ep->com.local_addr;
655 struct sockaddr_in *ra = (struct sockaddr_in *)
656 &ep->com.remote_addr;
657 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
658 &ep->com.local_addr;
659 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
660 &ep->com.remote_addr;
661 int ret;
662 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
663 u32 isn = (prandom_u32() & ~7UL) - 1;
664
665 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
666 case CHELSIO_T4:
667 sizev4 = sizeof(struct cpl_act_open_req);
668 sizev6 = sizeof(struct cpl_act_open_req6);
669 break;
670 case CHELSIO_T5:
671 sizev4 = sizeof(struct cpl_t5_act_open_req);
672 sizev6 = sizeof(struct cpl_t5_act_open_req6);
673 break;
674 case CHELSIO_T6:
675 sizev4 = sizeof(struct cpl_t6_act_open_req);
676 sizev6 = sizeof(struct cpl_t6_act_open_req6);
677 break;
678 default:
679 pr_err("T%d Chip is not supported\n",
680 CHELSIO_CHIP_VERSION(adapter_type));
681 return -EINVAL;
682 }
683
684 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
685 roundup(sizev4, 16) :
686 roundup(sizev6, 16);
687
688 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
689
690 skb = get_skb(NULL, wrlen, GFP_KERNEL);
691 if (!skb) {
692 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
693 __func__);
694 return -ENOMEM;
695 }
696 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
697
698 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
699 enable_tcp_timestamps,
700 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
701 wscale = compute_wscale(rcv_win);
702
703 /*
704 * Specify the largest window that will fit in opt0. The
705 * remainder will be specified in the rx_data_ack.
706 */
707 win = ep->rcv_win >> 10;
708 if (win > RCV_BUFSIZ_M)
709 win = RCV_BUFSIZ_M;
710
711 opt0 = (nocong ? NO_CONG_F : 0) |
712 KEEP_ALIVE_F |
713 DELACK_F |
714 WND_SCALE_V(wscale) |
715 MSS_IDX_V(mtu_idx) |
716 L2T_IDX_V(ep->l2t->idx) |
717 TX_CHAN_V(ep->tx_chan) |
718 SMAC_SEL_V(ep->smac_idx) |
719 DSCP_V(ep->tos >> 2) |
720 ULP_MODE_V(ULP_MODE_TCPDDP) |
721 RCV_BUFSIZ_V(win);
722 opt2 = RX_CHANNEL_V(0) |
723 CCTRL_ECN_V(enable_ecn) |
724 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
725 if (enable_tcp_timestamps)
726 opt2 |= TSTAMPS_EN_F;
727 if (enable_tcp_sack)
728 opt2 |= SACK_EN_F;
729 if (wscale && enable_tcp_window_scaling)
730 opt2 |= WND_SCALE_EN_F;
731 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
732 if (peer2peer)
733 isn += 4;
734
735 opt2 |= T5_OPT_2_VALID_F;
736 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
737 opt2 |= T5_ISS_F;
738 }
739
740 if (ep->com.remote_addr.ss_family == AF_INET6)
741 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
742 (const u32 *)&la6->sin6_addr.s6_addr, 1);
743
744 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
745
746 if (ep->com.remote_addr.ss_family == AF_INET) {
747 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
748 case CHELSIO_T4:
749 req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
750 INIT_TP_WR(req, 0);
751 break;
752 case CHELSIO_T5:
753 t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
754 wrlen);
755 INIT_TP_WR(t5req, 0);
756 req = (struct cpl_act_open_req *)t5req;
757 break;
758 case CHELSIO_T6:
759 t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
760 wrlen);
761 INIT_TP_WR(t6req, 0);
762 req = (struct cpl_act_open_req *)t6req;
763 t5req = (struct cpl_t5_act_open_req *)t6req;
764 break;
765 default:
766 pr_err("T%d Chip is not supported\n",
767 CHELSIO_CHIP_VERSION(adapter_type));
768 ret = -EINVAL;
769 goto clip_release;
770 }
771
772 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
773 ((ep->rss_qid<<14) | ep->atid)));
774 req->local_port = la->sin_port;
775 req->peer_port = ra->sin_port;
776 req->local_ip = la->sin_addr.s_addr;
777 req->peer_ip = ra->sin_addr.s_addr;
778 req->opt0 = cpu_to_be64(opt0);
779
780 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
781 req->params = cpu_to_be32(cxgb4_select_ntuple(
782 ep->com.dev->rdev.lldi.ports[0],
783 ep->l2t));
784 req->opt2 = cpu_to_be32(opt2);
785 } else {
786 t5req->params = cpu_to_be64(FILTER_TUPLE_V(
787 cxgb4_select_ntuple(
788 ep->com.dev->rdev.lldi.ports[0],
789 ep->l2t)));
790 t5req->rsvd = cpu_to_be32(isn);
791 PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
792 t5req->opt2 = cpu_to_be32(opt2);
793 }
794 } else {
795 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
796 case CHELSIO_T4:
797 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
798 INIT_TP_WR(req6, 0);
799 break;
800 case CHELSIO_T5:
801 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
802 wrlen);
803 INIT_TP_WR(t5req6, 0);
804 req6 = (struct cpl_act_open_req6 *)t5req6;
805 break;
806 case CHELSIO_T6:
807 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
808 wrlen);
809 INIT_TP_WR(t6req6, 0);
810 req6 = (struct cpl_act_open_req6 *)t6req6;
811 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
812 break;
813 default:
814 pr_err("T%d Chip is not supported\n",
815 CHELSIO_CHIP_VERSION(adapter_type));
816 ret = -EINVAL;
817 goto clip_release;
818 }
819
820 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
821 ((ep->rss_qid<<14)|ep->atid)));
822 req6->local_port = la6->sin6_port;
823 req6->peer_port = ra6->sin6_port;
824 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
825 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
826 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
827 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
828 req6->opt0 = cpu_to_be64(opt0);
829
830 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
831 req6->params = cpu_to_be32(cxgb4_select_ntuple(
832 ep->com.dev->rdev.lldi.ports[0],
833 ep->l2t));
834 req6->opt2 = cpu_to_be32(opt2);
835 } else {
836 t5req6->params = cpu_to_be64(FILTER_TUPLE_V(
837 cxgb4_select_ntuple(
838 ep->com.dev->rdev.lldi.ports[0],
839 ep->l2t)));
840 t5req6->rsvd = cpu_to_be32(isn);
841 PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
842 t5req6->opt2 = cpu_to_be32(opt2);
843 }
844 }
845
846 set_bit(ACT_OPEN_REQ, &ep->com.history);
847 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
848 clip_release:
849 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
850 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
851 (const u32 *)&la6->sin6_addr.s6_addr, 1);
852 return ret;
853 }
854
855 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
856 u8 mpa_rev_to_use)
857 {
858 int mpalen, wrlen;
859 struct fw_ofld_tx_data_wr *req;
860 struct mpa_message *mpa;
861 struct mpa_v2_conn_params mpa_v2_params;
862
863 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
864
865 BUG_ON(skb_cloned(skb));
866
867 mpalen = sizeof(*mpa) + ep->plen;
868 if (mpa_rev_to_use == 2)
869 mpalen += sizeof(struct mpa_v2_conn_params);
870 wrlen = roundup(mpalen + sizeof *req, 16);
871 skb = get_skb(skb, wrlen, GFP_KERNEL);
872 if (!skb) {
873 connect_reply_upcall(ep, -ENOMEM);
874 return;
875 }
876 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
877
878 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
879 memset(req, 0, wrlen);
880 req->op_to_immdlen = cpu_to_be32(
881 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
882 FW_WR_COMPL_F |
883 FW_WR_IMMDLEN_V(mpalen));
884 req->flowid_len16 = cpu_to_be32(
885 FW_WR_FLOWID_V(ep->hwtid) |
886 FW_WR_LEN16_V(wrlen >> 4));
887 req->plen = cpu_to_be32(mpalen);
888 req->tunnel_to_proxy = cpu_to_be32(
889 FW_OFLD_TX_DATA_WR_FLUSH_F |
890 FW_OFLD_TX_DATA_WR_SHOVE_F);
891
892 mpa = (struct mpa_message *)(req + 1);
893 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
894 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
895 (markers_enabled ? MPA_MARKERS : 0) |
896 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
897 mpa->private_data_size = htons(ep->plen);
898 mpa->revision = mpa_rev_to_use;
899 if (mpa_rev_to_use == 1) {
900 ep->tried_with_mpa_v1 = 1;
901 ep->retry_with_mpa_v1 = 0;
902 }
903
904 if (mpa_rev_to_use == 2) {
905 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
906 sizeof (struct mpa_v2_conn_params));
907 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
908 ep->ord);
909 mpa_v2_params.ird = htons((u16)ep->ird);
910 mpa_v2_params.ord = htons((u16)ep->ord);
911
912 if (peer2peer) {
913 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
914 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
915 mpa_v2_params.ord |=
916 htons(MPA_V2_RDMA_WRITE_RTR);
917 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
918 mpa_v2_params.ord |=
919 htons(MPA_V2_RDMA_READ_RTR);
920 }
921 memcpy(mpa->private_data, &mpa_v2_params,
922 sizeof(struct mpa_v2_conn_params));
923
924 if (ep->plen)
925 memcpy(mpa->private_data +
926 sizeof(struct mpa_v2_conn_params),
927 ep->mpa_pkt + sizeof(*mpa), ep->plen);
928 } else
929 if (ep->plen)
930 memcpy(mpa->private_data,
931 ep->mpa_pkt + sizeof(*mpa), ep->plen);
932
933 /*
934 * Reference the mpa skb. This ensures the data area
935 * will remain in memory until the hw acks the tx.
936 * Function fw4_ack() will deref it.
937 */
938 skb_get(skb);
939 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
940 BUG_ON(ep->mpa_skb);
941 ep->mpa_skb = skb;
942 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
943 start_ep_timer(ep);
944 __state_set(&ep->com, MPA_REQ_SENT);
945 ep->mpa_attr.initiator = 1;
946 ep->snd_seq += mpalen;
947 return;
948 }
949
950 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
951 {
952 int mpalen, wrlen;
953 struct fw_ofld_tx_data_wr *req;
954 struct mpa_message *mpa;
955 struct sk_buff *skb;
956 struct mpa_v2_conn_params mpa_v2_params;
957
958 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
959
960 mpalen = sizeof(*mpa) + plen;
961 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
962 mpalen += sizeof(struct mpa_v2_conn_params);
963 wrlen = roundup(mpalen + sizeof *req, 16);
964
965 skb = get_skb(NULL, wrlen, GFP_KERNEL);
966 if (!skb) {
967 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
968 return -ENOMEM;
969 }
970 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
971
972 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
973 memset(req, 0, wrlen);
974 req->op_to_immdlen = cpu_to_be32(
975 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
976 FW_WR_COMPL_F |
977 FW_WR_IMMDLEN_V(mpalen));
978 req->flowid_len16 = cpu_to_be32(
979 FW_WR_FLOWID_V(ep->hwtid) |
980 FW_WR_LEN16_V(wrlen >> 4));
981 req->plen = cpu_to_be32(mpalen);
982 req->tunnel_to_proxy = cpu_to_be32(
983 FW_OFLD_TX_DATA_WR_FLUSH_F |
984 FW_OFLD_TX_DATA_WR_SHOVE_F);
985
986 mpa = (struct mpa_message *)(req + 1);
987 memset(mpa, 0, sizeof(*mpa));
988 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
989 mpa->flags = MPA_REJECT;
990 mpa->revision = ep->mpa_attr.version;
991 mpa->private_data_size = htons(plen);
992
993 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
994 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
995 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
996 sizeof (struct mpa_v2_conn_params));
997 mpa_v2_params.ird = htons(((u16)ep->ird) |
998 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
999 0));
1000 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1001 (p2p_type ==
1002 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1003 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1004 FW_RI_INIT_P2PTYPE_READ_REQ ?
1005 MPA_V2_RDMA_READ_RTR : 0) : 0));
1006 memcpy(mpa->private_data, &mpa_v2_params,
1007 sizeof(struct mpa_v2_conn_params));
1008
1009 if (ep->plen)
1010 memcpy(mpa->private_data +
1011 sizeof(struct mpa_v2_conn_params), pdata, plen);
1012 } else
1013 if (plen)
1014 memcpy(mpa->private_data, pdata, plen);
1015
1016 /*
1017 * Reference the mpa skb again. This ensures the data area
1018 * will remain in memory until the hw acks the tx.
1019 * Function fw4_ack() will deref it.
1020 */
1021 skb_get(skb);
1022 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1023 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1024 BUG_ON(ep->mpa_skb);
1025 ep->mpa_skb = skb;
1026 ep->snd_seq += mpalen;
1027 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1028 }
1029
1030 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1031 {
1032 int mpalen, wrlen;
1033 struct fw_ofld_tx_data_wr *req;
1034 struct mpa_message *mpa;
1035 struct sk_buff *skb;
1036 struct mpa_v2_conn_params mpa_v2_params;
1037
1038 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1039
1040 mpalen = sizeof(*mpa) + plen;
1041 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1042 mpalen += sizeof(struct mpa_v2_conn_params);
1043 wrlen = roundup(mpalen + sizeof *req, 16);
1044
1045 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1046 if (!skb) {
1047 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1048 return -ENOMEM;
1049 }
1050 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1051
1052 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1053 memset(req, 0, wrlen);
1054 req->op_to_immdlen = cpu_to_be32(
1055 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1056 FW_WR_COMPL_F |
1057 FW_WR_IMMDLEN_V(mpalen));
1058 req->flowid_len16 = cpu_to_be32(
1059 FW_WR_FLOWID_V(ep->hwtid) |
1060 FW_WR_LEN16_V(wrlen >> 4));
1061 req->plen = cpu_to_be32(mpalen);
1062 req->tunnel_to_proxy = cpu_to_be32(
1063 FW_OFLD_TX_DATA_WR_FLUSH_F |
1064 FW_OFLD_TX_DATA_WR_SHOVE_F);
1065
1066 mpa = (struct mpa_message *)(req + 1);
1067 memset(mpa, 0, sizeof(*mpa));
1068 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1069 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
1070 (markers_enabled ? MPA_MARKERS : 0);
1071 mpa->revision = ep->mpa_attr.version;
1072 mpa->private_data_size = htons(plen);
1073
1074 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1075 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1076 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1077 sizeof (struct mpa_v2_conn_params));
1078 mpa_v2_params.ird = htons((u16)ep->ird);
1079 mpa_v2_params.ord = htons((u16)ep->ord);
1080 if (peer2peer && (ep->mpa_attr.p2p_type !=
1081 FW_RI_INIT_P2PTYPE_DISABLED)) {
1082 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1083
1084 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1085 mpa_v2_params.ord |=
1086 htons(MPA_V2_RDMA_WRITE_RTR);
1087 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1088 mpa_v2_params.ord |=
1089 htons(MPA_V2_RDMA_READ_RTR);
1090 }
1091
1092 memcpy(mpa->private_data, &mpa_v2_params,
1093 sizeof(struct mpa_v2_conn_params));
1094
1095 if (ep->plen)
1096 memcpy(mpa->private_data +
1097 sizeof(struct mpa_v2_conn_params), pdata, plen);
1098 } else
1099 if (plen)
1100 memcpy(mpa->private_data, pdata, plen);
1101
1102 /*
1103 * Reference the mpa skb. This ensures the data area
1104 * will remain in memory until the hw acks the tx.
1105 * Function fw4_ack() will deref it.
1106 */
1107 skb_get(skb);
1108 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1109 ep->mpa_skb = skb;
1110 __state_set(&ep->com, MPA_REP_SENT);
1111 ep->snd_seq += mpalen;
1112 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1113 }
1114
1115 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1116 {
1117 struct c4iw_ep *ep;
1118 struct cpl_act_establish *req = cplhdr(skb);
1119 unsigned int tid = GET_TID(req);
1120 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1121 struct tid_info *t = dev->rdev.lldi.tids;
1122 int ret;
1123
1124 ep = lookup_atid(t, atid);
1125
1126 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1127 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1128
1129 mutex_lock(&ep->com.mutex);
1130 dst_confirm(ep->dst);
1131
1132 /* setup the hwtid for this connection */
1133 ep->hwtid = tid;
1134 cxgb4_insert_tid(t, ep, tid);
1135 insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
1136
1137 ep->snd_seq = be32_to_cpu(req->snd_isn);
1138 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1139
1140 set_emss(ep, ntohs(req->tcp_opt));
1141
1142 /* dealloc the atid */
1143 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1144 cxgb4_free_atid(t, atid);
1145 set_bit(ACT_ESTAB, &ep->com.history);
1146
1147 /* start MPA negotiation */
1148 ret = send_flowc(ep, NULL);
1149 if (ret)
1150 goto err;
1151 if (ep->retry_with_mpa_v1)
1152 send_mpa_req(ep, skb, 1);
1153 else
1154 send_mpa_req(ep, skb, mpa_rev);
1155 mutex_unlock(&ep->com.mutex);
1156 return 0;
1157 err:
1158 mutex_unlock(&ep->com.mutex);
1159 connect_reply_upcall(ep, -ENOMEM);
1160 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1161 return 0;
1162 }
1163
1164 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1165 {
1166 struct iw_cm_event event;
1167
1168 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1169 memset(&event, 0, sizeof(event));
1170 event.event = IW_CM_EVENT_CLOSE;
1171 event.status = status;
1172 if (ep->com.cm_id) {
1173 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1174 ep, ep->com.cm_id, ep->hwtid);
1175 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1176 ep->com.cm_id->rem_ref(ep->com.cm_id);
1177 ep->com.cm_id = NULL;
1178 set_bit(CLOSE_UPCALL, &ep->com.history);
1179 }
1180 }
1181
1182 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
1183 {
1184 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1185 __state_set(&ep->com, ABORTING);
1186 set_bit(ABORT_CONN, &ep->com.history);
1187 return send_abort(ep, skb, gfp);
1188 }
1189
1190 static void peer_close_upcall(struct c4iw_ep *ep)
1191 {
1192 struct iw_cm_event event;
1193
1194 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1195 memset(&event, 0, sizeof(event));
1196 event.event = IW_CM_EVENT_DISCONNECT;
1197 if (ep->com.cm_id) {
1198 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1199 ep, ep->com.cm_id, ep->hwtid);
1200 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1201 set_bit(DISCONN_UPCALL, &ep->com.history);
1202 }
1203 }
1204
1205 static void peer_abort_upcall(struct c4iw_ep *ep)
1206 {
1207 struct iw_cm_event event;
1208
1209 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1210 memset(&event, 0, sizeof(event));
1211 event.event = IW_CM_EVENT_CLOSE;
1212 event.status = -ECONNRESET;
1213 if (ep->com.cm_id) {
1214 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1215 ep->com.cm_id, ep->hwtid);
1216 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1217 ep->com.cm_id->rem_ref(ep->com.cm_id);
1218 ep->com.cm_id = NULL;
1219 set_bit(ABORT_UPCALL, &ep->com.history);
1220 }
1221 }
1222
1223 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1224 {
1225 struct iw_cm_event event;
1226
1227 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1228 memset(&event, 0, sizeof(event));
1229 event.event = IW_CM_EVENT_CONNECT_REPLY;
1230 event.status = status;
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
1236 if ((status == 0) || (status == -ECONNREFUSED)) {
1237 if (!ep->tried_with_mpa_v1) {
1238 /* this means MPA_v2 is used */
1239 event.ord = ep->ird;
1240 event.ird = ep->ord;
1241 event.private_data_len = ep->plen -
1242 sizeof(struct mpa_v2_conn_params);
1243 event.private_data = ep->mpa_pkt +
1244 sizeof(struct mpa_message) +
1245 sizeof(struct mpa_v2_conn_params);
1246 } else {
1247 /* this means MPA_v1 is used */
1248 event.ord = cur_max_read_depth(ep->com.dev);
1249 event.ird = cur_max_read_depth(ep->com.dev);
1250 event.private_data_len = ep->plen;
1251 event.private_data = ep->mpa_pkt +
1252 sizeof(struct mpa_message);
1253 }
1254 }
1255
1256 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1257 ep->hwtid, status);
1258 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1259 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1260
1261 if (status < 0) {
1262 ep->com.cm_id->rem_ref(ep->com.cm_id);
1263 ep->com.cm_id = NULL;
1264 }
1265 }
1266
1267 static int connect_request_upcall(struct c4iw_ep *ep)
1268 {
1269 struct iw_cm_event event;
1270 int ret;
1271
1272 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1273 memset(&event, 0, sizeof(event));
1274 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1275 memcpy(&event.local_addr, &ep->com.local_addr,
1276 sizeof(ep->com.local_addr));
1277 memcpy(&event.remote_addr, &ep->com.remote_addr,
1278 sizeof(ep->com.remote_addr));
1279 event.provider_data = ep;
1280 if (!ep->tried_with_mpa_v1) {
1281 /* this means MPA_v2 is used */
1282 event.ord = ep->ord;
1283 event.ird = ep->ird;
1284 event.private_data_len = ep->plen -
1285 sizeof(struct mpa_v2_conn_params);
1286 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1287 sizeof(struct mpa_v2_conn_params);
1288 } else {
1289 /* this means MPA_v1 is used. Send max supported */
1290 event.ord = cur_max_read_depth(ep->com.dev);
1291 event.ird = cur_max_read_depth(ep->com.dev);
1292 event.private_data_len = ep->plen;
1293 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1294 }
1295 c4iw_get_ep(&ep->com);
1296 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1297 &event);
1298 if (ret)
1299 c4iw_put_ep(&ep->com);
1300 set_bit(CONNREQ_UPCALL, &ep->com.history);
1301 c4iw_put_ep(&ep->parent_ep->com);
1302 return ret;
1303 }
1304
1305 static void established_upcall(struct c4iw_ep *ep)
1306 {
1307 struct iw_cm_event event;
1308
1309 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1310 memset(&event, 0, sizeof(event));
1311 event.event = IW_CM_EVENT_ESTABLISHED;
1312 event.ird = ep->ord;
1313 event.ord = ep->ird;
1314 if (ep->com.cm_id) {
1315 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1316 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1317 set_bit(ESTAB_UPCALL, &ep->com.history);
1318 }
1319 }
1320
1321 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1322 {
1323 struct cpl_rx_data_ack *req;
1324 struct sk_buff *skb;
1325 int wrlen = roundup(sizeof *req, 16);
1326
1327 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1328 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1329 if (!skb) {
1330 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1331 return 0;
1332 }
1333
1334 /*
1335 * If we couldn't specify the entire rcv window at connection setup
1336 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1337 * then add the overage in to the credits returned.
1338 */
1339 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1340 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1341
1342 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1343 memset(req, 0, wrlen);
1344 INIT_TP_WR(req, ep->hwtid);
1345 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1346 ep->hwtid));
1347 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK_F |
1348 RX_DACK_CHANGE_F |
1349 RX_DACK_MODE_V(dack_mode));
1350 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1351 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1352 return credits;
1353 }
1354
1355 #define RELAXED_IRD_NEGOTIATION 1
1356
1357 /*
1358 * process_mpa_reply - process streaming mode MPA reply
1359 *
1360 * Returns:
1361 *
1362 * 0 upon success indicating a connect request was delivered to the ULP
1363 * or the mpa request is incomplete but valid so far.
1364 *
1365 * 1 if a failure requires the caller to close the connection.
1366 *
1367 * 2 if a failure requires the caller to abort the connection.
1368 */
1369 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1370 {
1371 struct mpa_message *mpa;
1372 struct mpa_v2_conn_params *mpa_v2_params;
1373 u16 plen;
1374 u16 resp_ird, resp_ord;
1375 u8 rtr_mismatch = 0, insuff_ird = 0;
1376 struct c4iw_qp_attributes attrs;
1377 enum c4iw_qp_attr_mask mask;
1378 int err;
1379 int disconnect = 0;
1380
1381 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1382
1383 /*
1384 * Stop mpa timer. If it expired, then
1385 * we ignore the MPA reply. process_timeout()
1386 * will abort the connection.
1387 */
1388 if (stop_ep_timer(ep))
1389 return 0;
1390
1391 /*
1392 * If we get more than the supported amount of private data
1393 * then we must fail this connection.
1394 */
1395 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1396 err = -EINVAL;
1397 goto err;
1398 }
1399
1400 /*
1401 * copy the new data into our accumulation buffer.
1402 */
1403 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1404 skb->len);
1405 ep->mpa_pkt_len += skb->len;
1406
1407 /*
1408 * if we don't even have the mpa message, then bail.
1409 */
1410 if (ep->mpa_pkt_len < sizeof(*mpa))
1411 return 0;
1412 mpa = (struct mpa_message *) ep->mpa_pkt;
1413
1414 /* Validate MPA header. */
1415 if (mpa->revision > mpa_rev) {
1416 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1417 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1418 err = -EPROTO;
1419 goto err;
1420 }
1421 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1422 err = -EPROTO;
1423 goto err;
1424 }
1425
1426 plen = ntohs(mpa->private_data_size);
1427
1428 /*
1429 * Fail if there's too much private data.
1430 */
1431 if (plen > MPA_MAX_PRIVATE_DATA) {
1432 err = -EPROTO;
1433 goto err;
1434 }
1435
1436 /*
1437 * If plen does not account for pkt size
1438 */
1439 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1440 err = -EPROTO;
1441 goto err;
1442 }
1443
1444 ep->plen = (u8) plen;
1445
1446 /*
1447 * If we don't have all the pdata yet, then bail.
1448 * We'll continue process when more data arrives.
1449 */
1450 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1451 return 0;
1452
1453 if (mpa->flags & MPA_REJECT) {
1454 err = -ECONNREFUSED;
1455 goto err;
1456 }
1457
1458 /*
1459 * If we get here we have accumulated the entire mpa
1460 * start reply message including private data. And
1461 * the MPA header is valid.
1462 */
1463 __state_set(&ep->com, FPDU_MODE);
1464 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1465 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1466 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1467 ep->mpa_attr.version = mpa->revision;
1468 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1469
1470 if (mpa->revision == 2) {
1471 ep->mpa_attr.enhanced_rdma_conn =
1472 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1473 if (ep->mpa_attr.enhanced_rdma_conn) {
1474 mpa_v2_params = (struct mpa_v2_conn_params *)
1475 (ep->mpa_pkt + sizeof(*mpa));
1476 resp_ird = ntohs(mpa_v2_params->ird) &
1477 MPA_V2_IRD_ORD_MASK;
1478 resp_ord = ntohs(mpa_v2_params->ord) &
1479 MPA_V2_IRD_ORD_MASK;
1480 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1481 __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1482
1483 /*
1484 * This is a double-check. Ideally, below checks are
1485 * not required since ird/ord stuff has been taken
1486 * care of in c4iw_accept_cr
1487 */
1488 if (ep->ird < resp_ord) {
1489 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1490 ep->com.dev->rdev.lldi.max_ordird_qp)
1491 ep->ird = resp_ord;
1492 else
1493 insuff_ird = 1;
1494 } else if (ep->ird > resp_ord) {
1495 ep->ird = resp_ord;
1496 }
1497 if (ep->ord > resp_ird) {
1498 if (RELAXED_IRD_NEGOTIATION)
1499 ep->ord = resp_ird;
1500 else
1501 insuff_ird = 1;
1502 }
1503 if (insuff_ird) {
1504 err = -ENOMEM;
1505 ep->ird = resp_ord;
1506 ep->ord = resp_ird;
1507 }
1508
1509 if (ntohs(mpa_v2_params->ird) &
1510 MPA_V2_PEER2PEER_MODEL) {
1511 if (ntohs(mpa_v2_params->ord) &
1512 MPA_V2_RDMA_WRITE_RTR)
1513 ep->mpa_attr.p2p_type =
1514 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1515 else if (ntohs(mpa_v2_params->ord) &
1516 MPA_V2_RDMA_READ_RTR)
1517 ep->mpa_attr.p2p_type =
1518 FW_RI_INIT_P2PTYPE_READ_REQ;
1519 }
1520 }
1521 } else if (mpa->revision == 1)
1522 if (peer2peer)
1523 ep->mpa_attr.p2p_type = p2p_type;
1524
1525 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1526 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1527 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1528 ep->mpa_attr.recv_marker_enabled,
1529 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1530 ep->mpa_attr.p2p_type, p2p_type);
1531
1532 /*
1533 * If responder's RTR does not match with that of initiator, assign
1534 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1535 * generated when moving QP to RTS state.
1536 * A TERM message will be sent after QP has moved to RTS state
1537 */
1538 if ((ep->mpa_attr.version == 2) && peer2peer &&
1539 (ep->mpa_attr.p2p_type != p2p_type)) {
1540 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1541 rtr_mismatch = 1;
1542 }
1543
1544 attrs.mpa_attr = ep->mpa_attr;
1545 attrs.max_ird = ep->ird;
1546 attrs.max_ord = ep->ord;
1547 attrs.llp_stream_handle = ep;
1548 attrs.next_state = C4IW_QP_STATE_RTS;
1549
1550 mask = C4IW_QP_ATTR_NEXT_STATE |
1551 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1552 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1553
1554 /* bind QP and TID with INIT_WR */
1555 err = c4iw_modify_qp(ep->com.qp->rhp,
1556 ep->com.qp, mask, &attrs, 1);
1557 if (err)
1558 goto err;
1559
1560 /*
1561 * If responder's RTR requirement did not match with what initiator
1562 * supports, generate TERM message
1563 */
1564 if (rtr_mismatch) {
1565 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1566 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1567 attrs.ecode = MPA_NOMATCH_RTR;
1568 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1569 attrs.send_term = 1;
1570 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1571 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1572 err = -ENOMEM;
1573 disconnect = 1;
1574 goto out;
1575 }
1576
1577 /*
1578 * Generate TERM if initiator IRD is not sufficient for responder
1579 * provided ORD. Currently, we do the same behaviour even when
1580 * responder provided IRD is also not sufficient as regards to
1581 * initiator ORD.
1582 */
1583 if (insuff_ird) {
1584 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1585 __func__);
1586 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1587 attrs.ecode = MPA_INSUFF_IRD;
1588 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1589 attrs.send_term = 1;
1590 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1591 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1592 err = -ENOMEM;
1593 disconnect = 1;
1594 goto out;
1595 }
1596 goto out;
1597 err:
1598 disconnect = 2;
1599 out:
1600 connect_reply_upcall(ep, err);
1601 return disconnect;
1602 }
1603
1604 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1605 {
1606 struct mpa_message *mpa;
1607 struct mpa_v2_conn_params *mpa_v2_params;
1608 u16 plen;
1609
1610 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1611
1612 /*
1613 * If we get more than the supported amount of private data
1614 * then we must fail this connection.
1615 */
1616 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1617 (void)stop_ep_timer(ep);
1618 abort_connection(ep, skb, GFP_KERNEL);
1619 return;
1620 }
1621
1622 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1623
1624 /*
1625 * Copy the new data into our accumulation buffer.
1626 */
1627 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1628 skb->len);
1629 ep->mpa_pkt_len += skb->len;
1630
1631 /*
1632 * If we don't even have the mpa message, then bail.
1633 * We'll continue process when more data arrives.
1634 */
1635 if (ep->mpa_pkt_len < sizeof(*mpa))
1636 return;
1637
1638 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1639 mpa = (struct mpa_message *) ep->mpa_pkt;
1640
1641 /*
1642 * Validate MPA Header.
1643 */
1644 if (mpa->revision > mpa_rev) {
1645 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1646 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1647 (void)stop_ep_timer(ep);
1648 abort_connection(ep, skb, GFP_KERNEL);
1649 return;
1650 }
1651
1652 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1653 (void)stop_ep_timer(ep);
1654 abort_connection(ep, skb, GFP_KERNEL);
1655 return;
1656 }
1657
1658 plen = ntohs(mpa->private_data_size);
1659
1660 /*
1661 * Fail if there's too much private data.
1662 */
1663 if (plen > MPA_MAX_PRIVATE_DATA) {
1664 (void)stop_ep_timer(ep);
1665 abort_connection(ep, skb, GFP_KERNEL);
1666 return;
1667 }
1668
1669 /*
1670 * If plen does not account for pkt size
1671 */
1672 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1673 (void)stop_ep_timer(ep);
1674 abort_connection(ep, skb, GFP_KERNEL);
1675 return;
1676 }
1677 ep->plen = (u8) plen;
1678
1679 /*
1680 * If we don't have all the pdata yet, then bail.
1681 */
1682 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1683 return;
1684
1685 /*
1686 * If we get here we have accumulated the entire mpa
1687 * start reply message including private data.
1688 */
1689 ep->mpa_attr.initiator = 0;
1690 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1691 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1692 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1693 ep->mpa_attr.version = mpa->revision;
1694 if (mpa->revision == 1)
1695 ep->tried_with_mpa_v1 = 1;
1696 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1697
1698 if (mpa->revision == 2) {
1699 ep->mpa_attr.enhanced_rdma_conn =
1700 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1701 if (ep->mpa_attr.enhanced_rdma_conn) {
1702 mpa_v2_params = (struct mpa_v2_conn_params *)
1703 (ep->mpa_pkt + sizeof(*mpa));
1704 ep->ird = ntohs(mpa_v2_params->ird) &
1705 MPA_V2_IRD_ORD_MASK;
1706 ep->ord = ntohs(mpa_v2_params->ord) &
1707 MPA_V2_IRD_ORD_MASK;
1708 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1709 ep->ord);
1710 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1711 if (peer2peer) {
1712 if (ntohs(mpa_v2_params->ord) &
1713 MPA_V2_RDMA_WRITE_RTR)
1714 ep->mpa_attr.p2p_type =
1715 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1716 else if (ntohs(mpa_v2_params->ord) &
1717 MPA_V2_RDMA_READ_RTR)
1718 ep->mpa_attr.p2p_type =
1719 FW_RI_INIT_P2PTYPE_READ_REQ;
1720 }
1721 }
1722 } else if (mpa->revision == 1)
1723 if (peer2peer)
1724 ep->mpa_attr.p2p_type = p2p_type;
1725
1726 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1727 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1728 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1729 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1730 ep->mpa_attr.p2p_type);
1731
1732 /*
1733 * If the endpoint timer already expired, then we ignore
1734 * the start request. process_timeout() will abort
1735 * the connection.
1736 */
1737 if (!stop_ep_timer(ep)) {
1738 __state_set(&ep->com, MPA_REQ_RCVD);
1739
1740 /* drive upcall */
1741 mutex_lock_nested(&ep->parent_ep->com.mutex,
1742 SINGLE_DEPTH_NESTING);
1743 if (ep->parent_ep->com.state != DEAD) {
1744 if (connect_request_upcall(ep))
1745 abort_connection(ep, skb, GFP_KERNEL);
1746 } else {
1747 abort_connection(ep, skb, GFP_KERNEL);
1748 }
1749 mutex_unlock(&ep->parent_ep->com.mutex);
1750 }
1751 return;
1752 }
1753
1754 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1755 {
1756 struct c4iw_ep *ep;
1757 struct cpl_rx_data *hdr = cplhdr(skb);
1758 unsigned int dlen = ntohs(hdr->len);
1759 unsigned int tid = GET_TID(hdr);
1760 struct tid_info *t = dev->rdev.lldi.tids;
1761 __u8 status = hdr->status;
1762 int disconnect = 0;
1763
1764 ep = lookup_tid(t, tid);
1765 if (!ep)
1766 return 0;
1767 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1768 skb_pull(skb, sizeof(*hdr));
1769 skb_trim(skb, dlen);
1770 mutex_lock(&ep->com.mutex);
1771
1772 /* update RX credits */
1773 update_rx_credits(ep, dlen);
1774
1775 switch (ep->com.state) {
1776 case MPA_REQ_SENT:
1777 ep->rcv_seq += dlen;
1778 disconnect = process_mpa_reply(ep, skb);
1779 break;
1780 case MPA_REQ_WAIT:
1781 ep->rcv_seq += dlen;
1782 process_mpa_request(ep, skb);
1783 break;
1784 case FPDU_MODE: {
1785 struct c4iw_qp_attributes attrs;
1786 BUG_ON(!ep->com.qp);
1787 if (status)
1788 pr_err("%s Unexpected streaming data." \
1789 " qpid %u ep %p state %d tid %u status %d\n",
1790 __func__, ep->com.qp->wq.sq.qid, ep,
1791 ep->com.state, ep->hwtid, status);
1792 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1793 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1794 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1795 disconnect = 1;
1796 break;
1797 }
1798 default:
1799 break;
1800 }
1801 mutex_unlock(&ep->com.mutex);
1802 if (disconnect)
1803 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1804 return 0;
1805 }
1806
1807 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1808 {
1809 struct c4iw_ep *ep;
1810 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1811 int release = 0;
1812 unsigned int tid = GET_TID(rpl);
1813 struct tid_info *t = dev->rdev.lldi.tids;
1814
1815 ep = lookup_tid(t, tid);
1816 if (!ep) {
1817 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1818 return 0;
1819 }
1820 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1821 mutex_lock(&ep->com.mutex);
1822 switch (ep->com.state) {
1823 case ABORTING:
1824 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1825 __state_set(&ep->com, DEAD);
1826 release = 1;
1827 break;
1828 default:
1829 printk(KERN_ERR "%s ep %p state %d\n",
1830 __func__, ep, ep->com.state);
1831 break;
1832 }
1833 mutex_unlock(&ep->com.mutex);
1834
1835 if (release)
1836 release_ep_resources(ep);
1837 return 0;
1838 }
1839
1840 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1841 {
1842 struct sk_buff *skb;
1843 struct fw_ofld_connection_wr *req;
1844 unsigned int mtu_idx;
1845 int wscale;
1846 struct sockaddr_in *sin;
1847 int win;
1848
1849 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1850 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1851 memset(req, 0, sizeof(*req));
1852 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1853 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1854 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1855 ep->com.dev->rdev.lldi.ports[0],
1856 ep->l2t));
1857 sin = (struct sockaddr_in *)&ep->com.local_addr;
1858 req->le.lport = sin->sin_port;
1859 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1860 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1861 req->le.pport = sin->sin_port;
1862 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1863 req->tcb.t_state_to_astid =
1864 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1865 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1866 req->tcb.cplrxdataack_cplpassacceptrpl =
1867 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1868 req->tcb.tx_max = (__force __be32) jiffies;
1869 req->tcb.rcv_adv = htons(1);
1870 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1871 enable_tcp_timestamps,
1872 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
1873 wscale = compute_wscale(rcv_win);
1874
1875 /*
1876 * Specify the largest window that will fit in opt0. The
1877 * remainder will be specified in the rx_data_ack.
1878 */
1879 win = ep->rcv_win >> 10;
1880 if (win > RCV_BUFSIZ_M)
1881 win = RCV_BUFSIZ_M;
1882
1883 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
1884 (nocong ? NO_CONG_F : 0) |
1885 KEEP_ALIVE_F |
1886 DELACK_F |
1887 WND_SCALE_V(wscale) |
1888 MSS_IDX_V(mtu_idx) |
1889 L2T_IDX_V(ep->l2t->idx) |
1890 TX_CHAN_V(ep->tx_chan) |
1891 SMAC_SEL_V(ep->smac_idx) |
1892 DSCP_V(ep->tos >> 2) |
1893 ULP_MODE_V(ULP_MODE_TCPDDP) |
1894 RCV_BUFSIZ_V(win));
1895 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
1896 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1897 RX_CHANNEL_V(0) |
1898 CCTRL_ECN_V(enable_ecn) |
1899 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
1900 if (enable_tcp_timestamps)
1901 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
1902 if (enable_tcp_sack)
1903 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
1904 if (wscale && enable_tcp_window_scaling)
1905 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
1906 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
1907 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
1908 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
1909 set_bit(ACT_OFLD_CONN, &ep->com.history);
1910 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1911 }
1912
1913 /*
1914 * Return whether a failed active open has allocated a TID
1915 */
1916 static inline int act_open_has_tid(int status)
1917 {
1918 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1919 status != CPL_ERR_ARP_MISS;
1920 }
1921
1922 /* Returns whether a CPL status conveys negative advice.
1923 */
1924 static int is_neg_adv(unsigned int status)
1925 {
1926 return status == CPL_ERR_RTX_NEG_ADVICE ||
1927 status == CPL_ERR_PERSIST_NEG_ADVICE ||
1928 status == CPL_ERR_KEEPALV_NEG_ADVICE;
1929 }
1930
1931 static char *neg_adv_str(unsigned int status)
1932 {
1933 switch (status) {
1934 case CPL_ERR_RTX_NEG_ADVICE:
1935 return "Retransmit timeout";
1936 case CPL_ERR_PERSIST_NEG_ADVICE:
1937 return "Persist timeout";
1938 case CPL_ERR_KEEPALV_NEG_ADVICE:
1939 return "Keepalive timeout";
1940 default:
1941 return "Unknown";
1942 }
1943 }
1944
1945 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
1946 {
1947 ep->snd_win = snd_win;
1948 ep->rcv_win = rcv_win;
1949 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
1950 }
1951
1952 #define ACT_OPEN_RETRY_COUNT 2
1953
1954 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
1955 struct dst_entry *dst, struct c4iw_dev *cdev,
1956 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
1957 {
1958 struct neighbour *n;
1959 int err, step;
1960 struct net_device *pdev;
1961
1962 n = dst_neigh_lookup(dst, peer_ip);
1963 if (!n)
1964 return -ENODEV;
1965
1966 rcu_read_lock();
1967 err = -ENOMEM;
1968 if (n->dev->flags & IFF_LOOPBACK) {
1969 if (iptype == 4)
1970 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
1971 else if (IS_ENABLED(CONFIG_IPV6))
1972 for_each_netdev(&init_net, pdev) {
1973 if (ipv6_chk_addr(&init_net,
1974 (struct in6_addr *)peer_ip,
1975 pdev, 1))
1976 break;
1977 }
1978 else
1979 pdev = NULL;
1980
1981 if (!pdev) {
1982 err = -ENODEV;
1983 goto out;
1984 }
1985 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1986 n, pdev, rt_tos2priority(tos));
1987 if (!ep->l2t)
1988 goto out;
1989 ep->mtu = pdev->mtu;
1990 ep->tx_chan = cxgb4_port_chan(pdev);
1991 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
1992 cxgb4_port_viid(pdev));
1993 step = cdev->rdev.lldi.ntxq /
1994 cdev->rdev.lldi.nchan;
1995 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1996 step = cdev->rdev.lldi.nrxq /
1997 cdev->rdev.lldi.nchan;
1998 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1999 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2000 cxgb4_port_idx(pdev) * step];
2001 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2002 dev_put(pdev);
2003 } else {
2004 pdev = get_real_dev(n->dev);
2005 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2006 n, pdev, 0);
2007 if (!ep->l2t)
2008 goto out;
2009 ep->mtu = dst_mtu(dst);
2010 ep->tx_chan = cxgb4_port_chan(pdev);
2011 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2012 cxgb4_port_viid(pdev));
2013 step = cdev->rdev.lldi.ntxq /
2014 cdev->rdev.lldi.nchan;
2015 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2016 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2017 step = cdev->rdev.lldi.nrxq /
2018 cdev->rdev.lldi.nchan;
2019 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2020 cxgb4_port_idx(pdev) * step];
2021 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2022
2023 if (clear_mpa_v1) {
2024 ep->retry_with_mpa_v1 = 0;
2025 ep->tried_with_mpa_v1 = 0;
2026 }
2027 }
2028 err = 0;
2029 out:
2030 rcu_read_unlock();
2031
2032 neigh_release(n);
2033
2034 return err;
2035 }
2036
2037 static int c4iw_reconnect(struct c4iw_ep *ep)
2038 {
2039 int err = 0;
2040 struct sockaddr_in *laddr = (struct sockaddr_in *)
2041 &ep->com.cm_id->m_local_addr;
2042 struct sockaddr_in *raddr = (struct sockaddr_in *)
2043 &ep->com.cm_id->m_remote_addr;
2044 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2045 &ep->com.cm_id->m_local_addr;
2046 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2047 &ep->com.cm_id->m_remote_addr;
2048 int iptype;
2049 __u8 *ra;
2050
2051 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2052 init_timer(&ep->timer);
2053
2054 /*
2055 * Allocate an active TID to initiate a TCP connection.
2056 */
2057 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2058 if (ep->atid == -1) {
2059 pr_err("%s - cannot alloc atid.\n", __func__);
2060 err = -ENOMEM;
2061 goto fail2;
2062 }
2063 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2064
2065 /* find a route */
2066 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2067 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
2068 raddr->sin_addr.s_addr, laddr->sin_port,
2069 raddr->sin_port, ep->com.cm_id->tos);
2070 iptype = 4;
2071 ra = (__u8 *)&raddr->sin_addr;
2072 } else {
2073 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
2074 raddr6->sin6_addr.s6_addr,
2075 laddr6->sin6_port, raddr6->sin6_port, 0,
2076 raddr6->sin6_scope_id);
2077 iptype = 6;
2078 ra = (__u8 *)&raddr6->sin6_addr;
2079 }
2080 if (!ep->dst) {
2081 pr_err("%s - cannot find route.\n", __func__);
2082 err = -EHOSTUNREACH;
2083 goto fail3;
2084 }
2085 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2086 ep->com.dev->rdev.lldi.adapter_type,
2087 ep->com.cm_id->tos);
2088 if (err) {
2089 pr_err("%s - cannot alloc l2e.\n", __func__);
2090 goto fail4;
2091 }
2092
2093 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2094 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2095 ep->l2t->idx);
2096
2097 state_set(&ep->com, CONNECTING);
2098 ep->tos = ep->com.cm_id->tos;
2099
2100 /* send connect request to rnic */
2101 err = send_connect(ep);
2102 if (!err)
2103 goto out;
2104
2105 cxgb4_l2t_release(ep->l2t);
2106 fail4:
2107 dst_release(ep->dst);
2108 fail3:
2109 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2110 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2111 fail2:
2112 /*
2113 * remember to send notification to upper layer.
2114 * We are in here so the upper layer is not aware that this is
2115 * re-connect attempt and so, upper layer is still waiting for
2116 * response of 1st connect request.
2117 */
2118 connect_reply_upcall(ep, -ECONNRESET);
2119 c4iw_put_ep(&ep->com);
2120 out:
2121 return err;
2122 }
2123
2124 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2125 {
2126 struct c4iw_ep *ep;
2127 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2128 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2129 ntohl(rpl->atid_status)));
2130 struct tid_info *t = dev->rdev.lldi.tids;
2131 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2132 struct sockaddr_in *la;
2133 struct sockaddr_in *ra;
2134 struct sockaddr_in6 *la6;
2135 struct sockaddr_in6 *ra6;
2136
2137 ep = lookup_atid(t, atid);
2138 la = (struct sockaddr_in *)&ep->com.local_addr;
2139 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2140 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2141 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2142
2143 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2144 status, status2errno(status));
2145
2146 if (is_neg_adv(status)) {
2147 PDBG("%s Connection problems for atid %u status %u (%s)\n",
2148 __func__, atid, status, neg_adv_str(status));
2149 ep->stats.connect_neg_adv++;
2150 mutex_lock(&dev->rdev.stats.lock);
2151 dev->rdev.stats.neg_adv++;
2152 mutex_unlock(&dev->rdev.stats.lock);
2153 return 0;
2154 }
2155
2156 set_bit(ACT_OPEN_RPL, &ep->com.history);
2157
2158 /*
2159 * Log interesting failures.
2160 */
2161 switch (status) {
2162 case CPL_ERR_CONN_RESET:
2163 case CPL_ERR_CONN_TIMEDOUT:
2164 break;
2165 case CPL_ERR_TCAM_FULL:
2166 mutex_lock(&dev->rdev.stats.lock);
2167 dev->rdev.stats.tcam_full++;
2168 mutex_unlock(&dev->rdev.stats.lock);
2169 if (ep->com.local_addr.ss_family == AF_INET &&
2170 dev->rdev.lldi.enable_fw_ofld_conn) {
2171 send_fw_act_open_req(ep,
2172 TID_TID_G(AOPEN_ATID_G(
2173 ntohl(rpl->atid_status))));
2174 return 0;
2175 }
2176 break;
2177 case CPL_ERR_CONN_EXIST:
2178 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2179 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2180 if (ep->com.remote_addr.ss_family == AF_INET6) {
2181 struct sockaddr_in6 *sin6 =
2182 (struct sockaddr_in6 *)
2183 &ep->com.local_addr;
2184 cxgb4_clip_release(
2185 ep->com.dev->rdev.lldi.ports[0],
2186 (const u32 *)
2187 &sin6->sin6_addr.s6_addr, 1);
2188 }
2189 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2190 atid);
2191 cxgb4_free_atid(t, atid);
2192 dst_release(ep->dst);
2193 cxgb4_l2t_release(ep->l2t);
2194 c4iw_reconnect(ep);
2195 return 0;
2196 }
2197 break;
2198 default:
2199 if (ep->com.local_addr.ss_family == AF_INET) {
2200 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2201 atid, status, status2errno(status),
2202 &la->sin_addr.s_addr, ntohs(la->sin_port),
2203 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2204 } else {
2205 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2206 atid, status, status2errno(status),
2207 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2208 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2209 }
2210 break;
2211 }
2212
2213 connect_reply_upcall(ep, status2errno(status));
2214 state_set(&ep->com, DEAD);
2215
2216 if (ep->com.remote_addr.ss_family == AF_INET6) {
2217 struct sockaddr_in6 *sin6 =
2218 (struct sockaddr_in6 *)&ep->com.local_addr;
2219 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2220 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2221 }
2222 if (status && act_open_has_tid(status))
2223 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2224
2225 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2226 cxgb4_free_atid(t, atid);
2227 dst_release(ep->dst);
2228 cxgb4_l2t_release(ep->l2t);
2229 c4iw_put_ep(&ep->com);
2230
2231 return 0;
2232 }
2233
2234 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2235 {
2236 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2237 struct tid_info *t = dev->rdev.lldi.tids;
2238 unsigned int stid = GET_TID(rpl);
2239 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
2240
2241 if (!ep) {
2242 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2243 goto out;
2244 }
2245 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2246 rpl->status, status2errno(rpl->status));
2247 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2248
2249 out:
2250 return 0;
2251 }
2252
2253 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2254 {
2255 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2256 struct tid_info *t = dev->rdev.lldi.tids;
2257 unsigned int stid = GET_TID(rpl);
2258 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
2259
2260 PDBG("%s ep %p\n", __func__, ep);
2261 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2262 return 0;
2263 }
2264
2265 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2266 struct cpl_pass_accept_req *req)
2267 {
2268 struct cpl_pass_accept_rpl *rpl;
2269 unsigned int mtu_idx;
2270 u64 opt0;
2271 u32 opt2;
2272 int wscale;
2273 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2274 int win;
2275 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2276
2277 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2278 BUG_ON(skb_cloned(skb));
2279
2280 skb_get(skb);
2281 rpl = cplhdr(skb);
2282 if (!is_t4(adapter_type)) {
2283 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2284 rpl5 = (void *)rpl;
2285 INIT_TP_WR(rpl5, ep->hwtid);
2286 } else {
2287 skb_trim(skb, sizeof(*rpl));
2288 INIT_TP_WR(rpl, ep->hwtid);
2289 }
2290 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2291 ep->hwtid));
2292
2293 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2294 enable_tcp_timestamps && req->tcpopt.tstamp,
2295 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
2296 wscale = compute_wscale(rcv_win);
2297
2298 /*
2299 * Specify the largest window that will fit in opt0. The
2300 * remainder will be specified in the rx_data_ack.
2301 */
2302 win = ep->rcv_win >> 10;
2303 if (win > RCV_BUFSIZ_M)
2304 win = RCV_BUFSIZ_M;
2305 opt0 = (nocong ? NO_CONG_F : 0) |
2306 KEEP_ALIVE_F |
2307 DELACK_F |
2308 WND_SCALE_V(wscale) |
2309 MSS_IDX_V(mtu_idx) |
2310 L2T_IDX_V(ep->l2t->idx) |
2311 TX_CHAN_V(ep->tx_chan) |
2312 SMAC_SEL_V(ep->smac_idx) |
2313 DSCP_V(ep->tos >> 2) |
2314 ULP_MODE_V(ULP_MODE_TCPDDP) |
2315 RCV_BUFSIZ_V(win);
2316 opt2 = RX_CHANNEL_V(0) |
2317 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2318
2319 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2320 opt2 |= TSTAMPS_EN_F;
2321 if (enable_tcp_sack && req->tcpopt.sack)
2322 opt2 |= SACK_EN_F;
2323 if (wscale && enable_tcp_window_scaling)
2324 opt2 |= WND_SCALE_EN_F;
2325 if (enable_ecn) {
2326 const struct tcphdr *tcph;
2327 u32 hlen = ntohl(req->hdr_len);
2328
2329 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2330 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2331 IP_HDR_LEN_G(hlen);
2332 else
2333 tcph = (const void *)(req + 1) +
2334 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2335 if (tcph->ece && tcph->cwr)
2336 opt2 |= CCTRL_ECN_V(1);
2337 }
2338 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2339 u32 isn = (prandom_u32() & ~7UL) - 1;
2340 opt2 |= T5_OPT_2_VALID_F;
2341 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2342 opt2 |= T5_ISS_F;
2343 rpl5 = (void *)rpl;
2344 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2345 if (peer2peer)
2346 isn += 4;
2347 rpl5->iss = cpu_to_be32(isn);
2348 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2349 }
2350
2351 rpl->opt0 = cpu_to_be64(opt0);
2352 rpl->opt2 = cpu_to_be32(opt2);
2353 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2354 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2355
2356 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2357 }
2358
2359 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2360 {
2361 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2362 BUG_ON(skb_cloned(skb));
2363 skb_trim(skb, sizeof(struct cpl_tid_release));
2364 release_tid(&dev->rdev, hwtid, skb);
2365 return;
2366 }
2367
2368 static void get_4tuple(struct cpl_pass_accept_req *req, enum chip_type type,
2369 int *iptype, __u8 *local_ip, __u8 *peer_ip,
2370 __be16 *local_port, __be16 *peer_port)
2371 {
2372 int eth_len = (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) ?
2373 ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len)) :
2374 T6_ETH_HDR_LEN_G(be32_to_cpu(req->hdr_len));
2375 int ip_len = (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) ?
2376 IP_HDR_LEN_G(be32_to_cpu(req->hdr_len)) :
2377 T6_IP_HDR_LEN_G(be32_to_cpu(req->hdr_len));
2378 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
2379 struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
2380 struct tcphdr *tcp = (struct tcphdr *)
2381 ((u8 *)(req + 1) + eth_len + ip_len);
2382
2383 if (ip->version == 4) {
2384 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
2385 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
2386 ntohs(tcp->dest));
2387 *iptype = 4;
2388 memcpy(peer_ip, &ip->saddr, 4);
2389 memcpy(local_ip, &ip->daddr, 4);
2390 } else {
2391 PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
2392 ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
2393 ntohs(tcp->dest));
2394 *iptype = 6;
2395 memcpy(peer_ip, ip6->saddr.s6_addr, 16);
2396 memcpy(local_ip, ip6->daddr.s6_addr, 16);
2397 }
2398 *peer_port = tcp->source;
2399 *local_port = tcp->dest;
2400
2401 return;
2402 }
2403
2404 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2405 {
2406 struct c4iw_ep *child_ep = NULL, *parent_ep;
2407 struct cpl_pass_accept_req *req = cplhdr(skb);
2408 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2409 struct tid_info *t = dev->rdev.lldi.tids;
2410 unsigned int hwtid = GET_TID(req);
2411 struct dst_entry *dst;
2412 __u8 local_ip[16], peer_ip[16];
2413 __be16 local_port, peer_port;
2414 struct sockaddr_in6 *sin6;
2415 int err;
2416 u16 peer_mss = ntohs(req->tcpopt.mss);
2417 int iptype;
2418 unsigned short hdrs;
2419 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2420
2421 parent_ep = lookup_stid(t, stid);
2422 if (!parent_ep) {
2423 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2424 goto reject;
2425 }
2426
2427 if (state_read(&parent_ep->com) != LISTEN) {
2428 PDBG("%s - listening ep not in LISTEN\n", __func__);
2429 goto reject;
2430 }
2431
2432 get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type, &iptype,
2433 local_ip, peer_ip, &local_port, &peer_port);
2434
2435 /* Find output route */
2436 if (iptype == 4) {
2437 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2438 , __func__, parent_ep, hwtid,
2439 local_ip, peer_ip, ntohs(local_port),
2440 ntohs(peer_port), peer_mss);
2441 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2442 local_port, peer_port,
2443 tos);
2444 } else {
2445 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2446 , __func__, parent_ep, hwtid,
2447 local_ip, peer_ip, ntohs(local_port),
2448 ntohs(peer_port), peer_mss);
2449 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2450 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2451 ((struct sockaddr_in6 *)
2452 &parent_ep->com.local_addr)->sin6_scope_id);
2453 }
2454 if (!dst) {
2455 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2456 __func__);
2457 goto reject;
2458 }
2459
2460 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2461 if (!child_ep) {
2462 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2463 __func__);
2464 dst_release(dst);
2465 goto reject;
2466 }
2467
2468 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2469 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2470 if (err) {
2471 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2472 __func__);
2473 dst_release(dst);
2474 kfree(child_ep);
2475 goto reject;
2476 }
2477
2478 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2479 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2480 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2481 child_ep->mtu = peer_mss + hdrs;
2482
2483 state_set(&child_ep->com, CONNECTING);
2484 child_ep->com.dev = dev;
2485 child_ep->com.cm_id = NULL;
2486
2487 if (iptype == 4) {
2488 struct sockaddr_in *sin = (struct sockaddr_in *)
2489 &child_ep->com.local_addr;
2490
2491 sin->sin_family = PF_INET;
2492 sin->sin_port = local_port;
2493 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2494
2495 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2496 sin->sin_family = PF_INET;
2497 sin->sin_port = ((struct sockaddr_in *)
2498 &parent_ep->com.local_addr)->sin_port;
2499 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2500
2501 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2502 sin->sin_family = PF_INET;
2503 sin->sin_port = peer_port;
2504 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2505 } else {
2506 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2507 sin6->sin6_family = PF_INET6;
2508 sin6->sin6_port = local_port;
2509 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2510
2511 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2512 sin6->sin6_family = PF_INET6;
2513 sin6->sin6_port = ((struct sockaddr_in6 *)
2514 &parent_ep->com.local_addr)->sin6_port;
2515 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2516
2517 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2518 sin6->sin6_family = PF_INET6;
2519 sin6->sin6_port = peer_port;
2520 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2521 }
2522
2523 c4iw_get_ep(&parent_ep->com);
2524 child_ep->parent_ep = parent_ep;
2525 child_ep->tos = tos;
2526 child_ep->dst = dst;
2527 child_ep->hwtid = hwtid;
2528
2529 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2530 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2531
2532 init_timer(&child_ep->timer);
2533 cxgb4_insert_tid(t, child_ep, hwtid);
2534 insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
2535 if (accept_cr(child_ep, skb, req)) {
2536 c4iw_put_ep(&parent_ep->com);
2537 release_ep_resources(child_ep);
2538 } else {
2539 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2540 }
2541 if (iptype == 6) {
2542 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2543 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2544 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2545 }
2546 goto out;
2547 reject:
2548 reject_cr(dev, hwtid, skb);
2549 out:
2550 return 0;
2551 }
2552
2553 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2554 {
2555 struct c4iw_ep *ep;
2556 struct cpl_pass_establish *req = cplhdr(skb);
2557 struct tid_info *t = dev->rdev.lldi.tids;
2558 unsigned int tid = GET_TID(req);
2559 int ret;
2560
2561 ep = lookup_tid(t, tid);
2562 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2563 ep->snd_seq = be32_to_cpu(req->snd_isn);
2564 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2565
2566 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2567 ntohs(req->tcp_opt));
2568
2569 set_emss(ep, ntohs(req->tcp_opt));
2570
2571 dst_confirm(ep->dst);
2572 mutex_lock(&ep->com.mutex);
2573 ep->com.state = MPA_REQ_WAIT;
2574 start_ep_timer(ep);
2575 set_bit(PASS_ESTAB, &ep->com.history);
2576 ret = send_flowc(ep, skb);
2577 mutex_unlock(&ep->com.mutex);
2578 if (ret)
2579 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2580
2581 return 0;
2582 }
2583
2584 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2585 {
2586 struct cpl_peer_close *hdr = cplhdr(skb);
2587 struct c4iw_ep *ep;
2588 struct c4iw_qp_attributes attrs;
2589 int disconnect = 1;
2590 int release = 0;
2591 struct tid_info *t = dev->rdev.lldi.tids;
2592 unsigned int tid = GET_TID(hdr);
2593 int ret;
2594
2595 ep = lookup_tid(t, tid);
2596 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2597 dst_confirm(ep->dst);
2598
2599 set_bit(PEER_CLOSE, &ep->com.history);
2600 mutex_lock(&ep->com.mutex);
2601 switch (ep->com.state) {
2602 case MPA_REQ_WAIT:
2603 __state_set(&ep->com, CLOSING);
2604 break;
2605 case MPA_REQ_SENT:
2606 __state_set(&ep->com, CLOSING);
2607 connect_reply_upcall(ep, -ECONNRESET);
2608 break;
2609 case MPA_REQ_RCVD:
2610
2611 /*
2612 * We're gonna mark this puppy DEAD, but keep
2613 * the reference on it until the ULP accepts or
2614 * rejects the CR. Also wake up anyone waiting
2615 * in rdma connection migration (see c4iw_accept_cr()).
2616 */
2617 __state_set(&ep->com, CLOSING);
2618 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2619 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2620 break;
2621 case MPA_REP_SENT:
2622 __state_set(&ep->com, CLOSING);
2623 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2624 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2625 break;
2626 case FPDU_MODE:
2627 start_ep_timer(ep);
2628 __state_set(&ep->com, CLOSING);
2629 attrs.next_state = C4IW_QP_STATE_CLOSING;
2630 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2631 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2632 if (ret != -ECONNRESET) {
2633 peer_close_upcall(ep);
2634 disconnect = 1;
2635 }
2636 break;
2637 case ABORTING:
2638 disconnect = 0;
2639 break;
2640 case CLOSING:
2641 __state_set(&ep->com, MORIBUND);
2642 disconnect = 0;
2643 break;
2644 case MORIBUND:
2645 (void)stop_ep_timer(ep);
2646 if (ep->com.cm_id && ep->com.qp) {
2647 attrs.next_state = C4IW_QP_STATE_IDLE;
2648 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2649 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2650 }
2651 close_complete_upcall(ep, 0);
2652 __state_set(&ep->com, DEAD);
2653 release = 1;
2654 disconnect = 0;
2655 break;
2656 case DEAD:
2657 disconnect = 0;
2658 break;
2659 default:
2660 BUG_ON(1);
2661 }
2662 mutex_unlock(&ep->com.mutex);
2663 if (disconnect)
2664 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2665 if (release)
2666 release_ep_resources(ep);
2667 return 0;
2668 }
2669
2670 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2671 {
2672 struct cpl_abort_req_rss *req = cplhdr(skb);
2673 struct c4iw_ep *ep;
2674 struct cpl_abort_rpl *rpl;
2675 struct sk_buff *rpl_skb;
2676 struct c4iw_qp_attributes attrs;
2677 int ret;
2678 int release = 0;
2679 struct tid_info *t = dev->rdev.lldi.tids;
2680 unsigned int tid = GET_TID(req);
2681
2682 ep = lookup_tid(t, tid);
2683 if (is_neg_adv(req->status)) {
2684 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2685 __func__, ep->hwtid, req->status,
2686 neg_adv_str(req->status));
2687 ep->stats.abort_neg_adv++;
2688 mutex_lock(&dev->rdev.stats.lock);
2689 dev->rdev.stats.neg_adv++;
2690 mutex_unlock(&dev->rdev.stats.lock);
2691 return 0;
2692 }
2693 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2694 ep->com.state);
2695 set_bit(PEER_ABORT, &ep->com.history);
2696
2697 /*
2698 * Wake up any threads in rdma_init() or rdma_fini().
2699 * However, this is not needed if com state is just
2700 * MPA_REQ_SENT
2701 */
2702 if (ep->com.state != MPA_REQ_SENT)
2703 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2704
2705 mutex_lock(&ep->com.mutex);
2706 switch (ep->com.state) {
2707 case CONNECTING:
2708 c4iw_put_ep(&ep->parent_ep->com);
2709 break;
2710 case MPA_REQ_WAIT:
2711 (void)stop_ep_timer(ep);
2712 break;
2713 case MPA_REQ_SENT:
2714 (void)stop_ep_timer(ep);
2715 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2716 connect_reply_upcall(ep, -ECONNRESET);
2717 else {
2718 /*
2719 * we just don't send notification upwards because we
2720 * want to retry with mpa_v1 without upper layers even
2721 * knowing it.
2722 *
2723 * do some housekeeping so as to re-initiate the
2724 * connection
2725 */
2726 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2727 mpa_rev);
2728 ep->retry_with_mpa_v1 = 1;
2729 }
2730 break;
2731 case MPA_REP_SENT:
2732 break;
2733 case MPA_REQ_RCVD:
2734 break;
2735 case MORIBUND:
2736 case CLOSING:
2737 stop_ep_timer(ep);
2738 /*FALLTHROUGH*/
2739 case FPDU_MODE:
2740 if (ep->com.cm_id && ep->com.qp) {
2741 attrs.next_state = C4IW_QP_STATE_ERROR;
2742 ret = c4iw_modify_qp(ep->com.qp->rhp,
2743 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2744 &attrs, 1);
2745 if (ret)
2746 printk(KERN_ERR MOD
2747 "%s - qp <- error failed!\n",
2748 __func__);
2749 }
2750 peer_abort_upcall(ep);
2751 break;
2752 case ABORTING:
2753 break;
2754 case DEAD:
2755 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2756 mutex_unlock(&ep->com.mutex);
2757 return 0;
2758 default:
2759 BUG_ON(1);
2760 break;
2761 }
2762 dst_confirm(ep->dst);
2763 if (ep->com.state != ABORTING) {
2764 __state_set(&ep->com, DEAD);
2765 /* we don't release if we want to retry with mpa_v1 */
2766 if (!ep->retry_with_mpa_v1)
2767 release = 1;
2768 }
2769 mutex_unlock(&ep->com.mutex);
2770
2771 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2772 if (!rpl_skb) {
2773 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2774 __func__);
2775 release = 1;
2776 goto out;
2777 }
2778 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2779 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2780 INIT_TP_WR(rpl, ep->hwtid);
2781 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2782 rpl->cmd = CPL_ABORT_NO_RST;
2783 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2784 out:
2785 if (release)
2786 release_ep_resources(ep);
2787 else if (ep->retry_with_mpa_v1) {
2788 if (ep->com.remote_addr.ss_family == AF_INET6) {
2789 struct sockaddr_in6 *sin6 =
2790 (struct sockaddr_in6 *)
2791 &ep->com.local_addr;
2792 cxgb4_clip_release(
2793 ep->com.dev->rdev.lldi.ports[0],
2794 (const u32 *)&sin6->sin6_addr.s6_addr,
2795 1);
2796 }
2797 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2798 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2799 dst_release(ep->dst);
2800 cxgb4_l2t_release(ep->l2t);
2801 c4iw_reconnect(ep);
2802 }
2803
2804 return 0;
2805 }
2806
2807 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2808 {
2809 struct c4iw_ep *ep;
2810 struct c4iw_qp_attributes attrs;
2811 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2812 int release = 0;
2813 struct tid_info *t = dev->rdev.lldi.tids;
2814 unsigned int tid = GET_TID(rpl);
2815
2816 ep = lookup_tid(t, tid);
2817
2818 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2819 BUG_ON(!ep);
2820
2821 /* The cm_id may be null if we failed to connect */
2822 mutex_lock(&ep->com.mutex);
2823 switch (ep->com.state) {
2824 case CLOSING:
2825 __state_set(&ep->com, MORIBUND);
2826 break;
2827 case MORIBUND:
2828 (void)stop_ep_timer(ep);
2829 if ((ep->com.cm_id) && (ep->com.qp)) {
2830 attrs.next_state = C4IW_QP_STATE_IDLE;
2831 c4iw_modify_qp(ep->com.qp->rhp,
2832 ep->com.qp,
2833 C4IW_QP_ATTR_NEXT_STATE,
2834 &attrs, 1);
2835 }
2836 close_complete_upcall(ep, 0);
2837 __state_set(&ep->com, DEAD);
2838 release = 1;
2839 break;
2840 case ABORTING:
2841 case DEAD:
2842 break;
2843 default:
2844 BUG_ON(1);
2845 break;
2846 }
2847 mutex_unlock(&ep->com.mutex);
2848 if (release)
2849 release_ep_resources(ep);
2850 return 0;
2851 }
2852
2853 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2854 {
2855 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2856 struct tid_info *t = dev->rdev.lldi.tids;
2857 unsigned int tid = GET_TID(rpl);
2858 struct c4iw_ep *ep;
2859 struct c4iw_qp_attributes attrs;
2860
2861 ep = lookup_tid(t, tid);
2862 BUG_ON(!ep);
2863
2864 if (ep && ep->com.qp) {
2865 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2866 ep->com.qp->wq.sq.qid);
2867 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2868 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2869 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2870 } else
2871 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2872
2873 return 0;
2874 }
2875
2876 /*
2877 * Upcall from the adapter indicating data has been transmitted.
2878 * For us its just the single MPA request or reply. We can now free
2879 * the skb holding the mpa message.
2880 */
2881 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2882 {
2883 struct c4iw_ep *ep;
2884 struct cpl_fw4_ack *hdr = cplhdr(skb);
2885 u8 credits = hdr->credits;
2886 unsigned int tid = GET_TID(hdr);
2887 struct tid_info *t = dev->rdev.lldi.tids;
2888
2889
2890 ep = lookup_tid(t, tid);
2891 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2892 if (credits == 0) {
2893 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2894 __func__, ep, ep->hwtid, state_read(&ep->com));
2895 return 0;
2896 }
2897
2898 dst_confirm(ep->dst);
2899 if (ep->mpa_skb) {
2900 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2901 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2902 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2903 kfree_skb(ep->mpa_skb);
2904 ep->mpa_skb = NULL;
2905 }
2906 return 0;
2907 }
2908
2909 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2910 {
2911 int err = 0;
2912 int disconnect = 0;
2913 struct c4iw_ep *ep = to_ep(cm_id);
2914 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2915
2916 mutex_lock(&ep->com.mutex);
2917 if (ep->com.state == DEAD) {
2918 mutex_unlock(&ep->com.mutex);
2919 c4iw_put_ep(&ep->com);
2920 return -ECONNRESET;
2921 }
2922 set_bit(ULP_REJECT, &ep->com.history);
2923 BUG_ON(ep->com.state != MPA_REQ_RCVD);
2924 if (mpa_rev == 0)
2925 disconnect = 2;
2926 else {
2927 err = send_mpa_reject(ep, pdata, pdata_len);
2928 disconnect = 1;
2929 }
2930 mutex_unlock(&ep->com.mutex);
2931 if (disconnect)
2932 err = c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
2933 c4iw_put_ep(&ep->com);
2934 return 0;
2935 }
2936
2937 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2938 {
2939 int err;
2940 struct c4iw_qp_attributes attrs;
2941 enum c4iw_qp_attr_mask mask;
2942 struct c4iw_ep *ep = to_ep(cm_id);
2943 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2944 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2945 int abort = 0;
2946
2947 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2948
2949 mutex_lock(&ep->com.mutex);
2950 if (ep->com.state == DEAD) {
2951 err = -ECONNRESET;
2952 goto err_out;
2953 }
2954
2955 BUG_ON(ep->com.state != MPA_REQ_RCVD);
2956 BUG_ON(!qp);
2957
2958 set_bit(ULP_ACCEPT, &ep->com.history);
2959 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
2960 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
2961 err = -EINVAL;
2962 goto err_abort;
2963 }
2964
2965 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2966 if (conn_param->ord > ep->ird) {
2967 if (RELAXED_IRD_NEGOTIATION) {
2968 ep->ord = ep->ird;
2969 } else {
2970 ep->ird = conn_param->ird;
2971 ep->ord = conn_param->ord;
2972 send_mpa_reject(ep, conn_param->private_data,
2973 conn_param->private_data_len);
2974 err = -ENOMEM;
2975 goto err_abort;
2976 }
2977 }
2978 if (conn_param->ird < ep->ord) {
2979 if (RELAXED_IRD_NEGOTIATION &&
2980 ep->ord <= h->rdev.lldi.max_ordird_qp) {
2981 conn_param->ird = ep->ord;
2982 } else {
2983 err = -ENOMEM;
2984 goto err_abort;
2985 }
2986 }
2987 }
2988 ep->ird = conn_param->ird;
2989 ep->ord = conn_param->ord;
2990
2991 if (ep->mpa_attr.version == 1) {
2992 if (peer2peer && ep->ird == 0)
2993 ep->ird = 1;
2994 } else {
2995 if (peer2peer &&
2996 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
2997 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
2998 ep->ird = 1;
2999 }
3000
3001 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3002
3003 cm_id->add_ref(cm_id);
3004 ep->com.cm_id = cm_id;
3005 ep->com.qp = qp;
3006 ref_qp(ep);
3007
3008 /* bind QP to EP and move to RTS */
3009 attrs.mpa_attr = ep->mpa_attr;
3010 attrs.max_ird = ep->ird;
3011 attrs.max_ord = ep->ord;
3012 attrs.llp_stream_handle = ep;
3013 attrs.next_state = C4IW_QP_STATE_RTS;
3014
3015 /* bind QP and TID with INIT_WR */
3016 mask = C4IW_QP_ATTR_NEXT_STATE |
3017 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3018 C4IW_QP_ATTR_MPA_ATTR |
3019 C4IW_QP_ATTR_MAX_IRD |
3020 C4IW_QP_ATTR_MAX_ORD;
3021
3022 err = c4iw_modify_qp(ep->com.qp->rhp,
3023 ep->com.qp, mask, &attrs, 1);
3024 if (err)
3025 goto err_deref_cm_id;
3026 err = send_mpa_reply(ep, conn_param->private_data,
3027 conn_param->private_data_len);
3028 if (err)
3029 goto err_deref_cm_id;
3030
3031 __state_set(&ep->com, FPDU_MODE);
3032 established_upcall(ep);
3033 mutex_unlock(&ep->com.mutex);
3034 c4iw_put_ep(&ep->com);
3035 return 0;
3036 err_deref_cm_id:
3037 ep->com.cm_id = NULL;
3038 cm_id->rem_ref(cm_id);
3039 err_abort:
3040 abort = 1;
3041 err_out:
3042 mutex_unlock(&ep->com.mutex);
3043 if (abort)
3044 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3045 c4iw_put_ep(&ep->com);
3046 return err;
3047 }
3048
3049 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3050 {
3051 struct in_device *ind;
3052 int found = 0;
3053 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3054 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3055
3056 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3057 if (!ind)
3058 return -EADDRNOTAVAIL;
3059 for_primary_ifa(ind) {
3060 laddr->sin_addr.s_addr = ifa->ifa_address;
3061 raddr->sin_addr.s_addr = ifa->ifa_address;
3062 found = 1;
3063 break;
3064 }
3065 endfor_ifa(ind);
3066 in_dev_put(ind);
3067 return found ? 0 : -EADDRNOTAVAIL;
3068 }
3069
3070 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3071 unsigned char banned_flags)
3072 {
3073 struct inet6_dev *idev;
3074 int err = -EADDRNOTAVAIL;
3075
3076 rcu_read_lock();
3077 idev = __in6_dev_get(dev);
3078 if (idev != NULL) {
3079 struct inet6_ifaddr *ifp;
3080
3081 read_lock_bh(&idev->lock);
3082 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3083 if (ifp->scope == IFA_LINK &&
3084 !(ifp->flags & banned_flags)) {
3085 memcpy(addr, &ifp->addr, 16);
3086 err = 0;
3087 break;
3088 }
3089 }
3090 read_unlock_bh(&idev->lock);
3091 }
3092 rcu_read_unlock();
3093 return err;
3094 }
3095
3096 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3097 {
3098 struct in6_addr uninitialized_var(addr);
3099 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3100 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3101
3102 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3103 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3104 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3105 return 0;
3106 }
3107 return -EADDRNOTAVAIL;
3108 }
3109
3110 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3111 {
3112 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3113 struct c4iw_ep *ep;
3114 int err = 0;
3115 struct sockaddr_in *laddr;
3116 struct sockaddr_in *raddr;
3117 struct sockaddr_in6 *laddr6;
3118 struct sockaddr_in6 *raddr6;
3119 __u8 *ra;
3120 int iptype;
3121
3122 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3123 (conn_param->ird > cur_max_read_depth(dev))) {
3124 err = -EINVAL;
3125 goto out;
3126 }
3127 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3128 if (!ep) {
3129 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3130 err = -ENOMEM;
3131 goto out;
3132 }
3133 init_timer(&ep->timer);
3134 ep->plen = conn_param->private_data_len;
3135 if (ep->plen)
3136 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3137 conn_param->private_data, ep->plen);
3138 ep->ird = conn_param->ird;
3139 ep->ord = conn_param->ord;
3140
3141 if (peer2peer && ep->ord == 0)
3142 ep->ord = 1;
3143
3144 cm_id->add_ref(cm_id);
3145 ep->com.dev = dev;
3146 ep->com.cm_id = cm_id;
3147 ep->com.qp = get_qhp(dev, conn_param->qpn);
3148 if (!ep->com.qp) {
3149 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3150 err = -EINVAL;
3151 goto fail1;
3152 }
3153 ref_qp(ep);
3154 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3155 ep->com.qp, cm_id);
3156
3157 /*
3158 * Allocate an active TID to initiate a TCP connection.
3159 */
3160 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3161 if (ep->atid == -1) {
3162 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3163 err = -ENOMEM;
3164 goto fail1;
3165 }
3166 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3167
3168 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3169 sizeof(ep->com.local_addr));
3170 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3171 sizeof(ep->com.remote_addr));
3172
3173 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3174 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3175 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3176 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3177
3178 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3179 iptype = 4;
3180 ra = (__u8 *)&raddr->sin_addr;
3181
3182 /*
3183 * Handle loopback requests to INADDR_ANY.
3184 */
3185 if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
3186 err = pick_local_ipaddrs(dev, cm_id);
3187 if (err)
3188 goto fail1;
3189 }
3190
3191 /* find a route */
3192 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3193 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3194 ra, ntohs(raddr->sin_port));
3195 ep->dst = find_route(dev, laddr->sin_addr.s_addr,
3196 raddr->sin_addr.s_addr, laddr->sin_port,
3197 raddr->sin_port, cm_id->tos);
3198 } else {
3199 iptype = 6;
3200 ra = (__u8 *)&raddr6->sin6_addr;
3201
3202 /*
3203 * Handle loopback requests to INADDR_ANY.
3204 */
3205 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3206 err = pick_local_ip6addrs(dev, cm_id);
3207 if (err)
3208 goto fail1;
3209 }
3210
3211 /* find a route */
3212 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3213 __func__, laddr6->sin6_addr.s6_addr,
3214 ntohs(laddr6->sin6_port),
3215 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3216 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
3217 raddr6->sin6_addr.s6_addr,
3218 laddr6->sin6_port, raddr6->sin6_port, 0,
3219 raddr6->sin6_scope_id);
3220 }
3221 if (!ep->dst) {
3222 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3223 err = -EHOSTUNREACH;
3224 goto fail2;
3225 }
3226
3227 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3228 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3229 if (err) {
3230 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3231 goto fail3;
3232 }
3233
3234 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3235 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3236 ep->l2t->idx);
3237
3238 state_set(&ep->com, CONNECTING);
3239 ep->tos = cm_id->tos;
3240
3241 /* send connect request to rnic */
3242 err = send_connect(ep);
3243 if (!err)
3244 goto out;
3245
3246 cxgb4_l2t_release(ep->l2t);
3247 fail3:
3248 dst_release(ep->dst);
3249 fail2:
3250 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3251 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3252 fail1:
3253 cm_id->rem_ref(cm_id);
3254 c4iw_put_ep(&ep->com);
3255 out:
3256 return err;
3257 }
3258
3259 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3260 {
3261 int err;
3262 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3263 &ep->com.local_addr;
3264
3265 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3266 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3267 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3268 if (err)
3269 return err;
3270 }
3271 c4iw_init_wr_wait(&ep->com.wr_wait);
3272 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3273 ep->stid, &sin6->sin6_addr,
3274 sin6->sin6_port,
3275 ep->com.dev->rdev.lldi.rxq_ids[0]);
3276 if (!err)
3277 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3278 &ep->com.wr_wait,
3279 0, 0, __func__);
3280 else if (err > 0)
3281 err = net_xmit_errno(err);
3282 if (err) {
3283 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3284 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3285 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3286 err, ep->stid,
3287 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3288 }
3289 return err;
3290 }
3291
3292 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3293 {
3294 int err;
3295 struct sockaddr_in *sin = (struct sockaddr_in *)
3296 &ep->com.local_addr;
3297
3298 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3299 do {
3300 err = cxgb4_create_server_filter(
3301 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3302 sin->sin_addr.s_addr, sin->sin_port, 0,
3303 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3304 if (err == -EBUSY) {
3305 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3306 err = -EIO;
3307 break;
3308 }
3309 set_current_state(TASK_UNINTERRUPTIBLE);
3310 schedule_timeout(usecs_to_jiffies(100));
3311 }
3312 } while (err == -EBUSY);
3313 } else {
3314 c4iw_init_wr_wait(&ep->com.wr_wait);
3315 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3316 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3317 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3318 if (!err)
3319 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3320 &ep->com.wr_wait,
3321 0, 0, __func__);
3322 else if (err > 0)
3323 err = net_xmit_errno(err);
3324 }
3325 if (err)
3326 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3327 , err, ep->stid,
3328 &sin->sin_addr, ntohs(sin->sin_port));
3329 return err;
3330 }
3331
3332 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3333 {
3334 int err = 0;
3335 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3336 struct c4iw_listen_ep *ep;
3337
3338 might_sleep();
3339
3340 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3341 if (!ep) {
3342 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3343 err = -ENOMEM;
3344 goto fail1;
3345 }
3346 PDBG("%s ep %p\n", __func__, ep);
3347 cm_id->add_ref(cm_id);
3348 ep->com.cm_id = cm_id;
3349 ep->com.dev = dev;
3350 ep->backlog = backlog;
3351 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3352 sizeof(ep->com.local_addr));
3353
3354 /*
3355 * Allocate a server TID.
3356 */
3357 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3358 ep->com.local_addr.ss_family == AF_INET)
3359 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3360 cm_id->m_local_addr.ss_family, ep);
3361 else
3362 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3363 cm_id->m_local_addr.ss_family, ep);
3364
3365 if (ep->stid == -1) {
3366 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3367 err = -ENOMEM;
3368 goto fail2;
3369 }
3370 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3371
3372 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3373 sizeof(ep->com.local_addr));
3374
3375 state_set(&ep->com, LISTEN);
3376 if (ep->com.local_addr.ss_family == AF_INET)
3377 err = create_server4(dev, ep);
3378 else
3379 err = create_server6(dev, ep);
3380 if (!err) {
3381 cm_id->provider_data = ep;
3382 goto out;
3383 }
3384
3385 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3386 ep->com.local_addr.ss_family);
3387 fail2:
3388 cm_id->rem_ref(cm_id);
3389 c4iw_put_ep(&ep->com);
3390 fail1:
3391 out:
3392 return err;
3393 }
3394
3395 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3396 {
3397 int err;
3398 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3399
3400 PDBG("%s ep %p\n", __func__, ep);
3401
3402 might_sleep();
3403 state_set(&ep->com, DEAD);
3404 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3405 ep->com.local_addr.ss_family == AF_INET) {
3406 err = cxgb4_remove_server_filter(
3407 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3408 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3409 } else {
3410 struct sockaddr_in6 *sin6;
3411 c4iw_init_wr_wait(&ep->com.wr_wait);
3412 err = cxgb4_remove_server(
3413 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3414 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3415 if (err)
3416 goto done;
3417 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3418 0, 0, __func__);
3419 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3420 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3421 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3422 }
3423 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3424 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3425 ep->com.local_addr.ss_family);
3426 done:
3427 cm_id->rem_ref(cm_id);
3428 c4iw_put_ep(&ep->com);
3429 return err;
3430 }
3431
3432 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3433 {
3434 int ret = 0;
3435 int close = 0;
3436 int fatal = 0;
3437 struct c4iw_rdev *rdev;
3438
3439 mutex_lock(&ep->com.mutex);
3440
3441 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3442 states[ep->com.state], abrupt);
3443
3444 /*
3445 * Ref the ep here in case we have fatal errors causing the
3446 * ep to be released and freed.
3447 */
3448 c4iw_get_ep(&ep->com);
3449
3450 rdev = &ep->com.dev->rdev;
3451 if (c4iw_fatal_error(rdev)) {
3452 fatal = 1;
3453 close_complete_upcall(ep, -EIO);
3454 ep->com.state = DEAD;
3455 }
3456 switch (ep->com.state) {
3457 case MPA_REQ_WAIT:
3458 case MPA_REQ_SENT:
3459 case MPA_REQ_RCVD:
3460 case MPA_REP_SENT:
3461 case FPDU_MODE:
3462 close = 1;
3463 if (abrupt)
3464 ep->com.state = ABORTING;
3465 else {
3466 ep->com.state = CLOSING;
3467 start_ep_timer(ep);
3468 }
3469 set_bit(CLOSE_SENT, &ep->com.flags);
3470 break;
3471 case CLOSING:
3472 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3473 close = 1;
3474 if (abrupt) {
3475 (void)stop_ep_timer(ep);
3476 ep->com.state = ABORTING;
3477 } else
3478 ep->com.state = MORIBUND;
3479 }
3480 break;
3481 case MORIBUND:
3482 case ABORTING:
3483 case DEAD:
3484 PDBG("%s ignoring disconnect ep %p state %u\n",
3485 __func__, ep, ep->com.state);
3486 break;
3487 default:
3488 BUG();
3489 break;
3490 }
3491
3492 if (close) {
3493 if (abrupt) {
3494 set_bit(EP_DISC_ABORT, &ep->com.history);
3495 close_complete_upcall(ep, -ECONNRESET);
3496 ret = send_abort(ep, NULL, gfp);
3497 } else {
3498 set_bit(EP_DISC_CLOSE, &ep->com.history);
3499 ret = send_halfclose(ep, gfp);
3500 }
3501 if (ret) {
3502 if (!abrupt) {
3503 stop_ep_timer(ep);
3504 close_complete_upcall(ep, -EIO);
3505 }
3506 fatal = 1;
3507 }
3508 }
3509 mutex_unlock(&ep->com.mutex);
3510 c4iw_put_ep(&ep->com);
3511 if (fatal)
3512 release_ep_resources(ep);
3513 return ret;
3514 }
3515
3516 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3517 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3518 {
3519 struct c4iw_ep *ep;
3520 int atid = be32_to_cpu(req->tid);
3521
3522 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3523 (__force u32) req->tid);
3524 if (!ep)
3525 return;
3526
3527 switch (req->retval) {
3528 case FW_ENOMEM:
3529 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3530 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3531 send_fw_act_open_req(ep, atid);
3532 return;
3533 }
3534 case FW_EADDRINUSE:
3535 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3536 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3537 send_fw_act_open_req(ep, atid);
3538 return;
3539 }
3540 break;
3541 default:
3542 pr_info("%s unexpected ofld conn wr retval %d\n",
3543 __func__, req->retval);
3544 break;
3545 }
3546 pr_err("active ofld_connect_wr failure %d atid %d\n",
3547 req->retval, atid);
3548 mutex_lock(&dev->rdev.stats.lock);
3549 dev->rdev.stats.act_ofld_conn_fails++;
3550 mutex_unlock(&dev->rdev.stats.lock);
3551 connect_reply_upcall(ep, status2errno(req->retval));
3552 state_set(&ep->com, DEAD);
3553 if (ep->com.remote_addr.ss_family == AF_INET6) {
3554 struct sockaddr_in6 *sin6 =
3555 (struct sockaddr_in6 *)&ep->com.local_addr;
3556 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3557 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3558 }
3559 remove_handle(dev, &dev->atid_idr, atid);
3560 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3561 dst_release(ep->dst);
3562 cxgb4_l2t_release(ep->l2t);
3563 c4iw_put_ep(&ep->com);
3564 }
3565
3566 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3567 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3568 {
3569 struct sk_buff *rpl_skb;
3570 struct cpl_pass_accept_req *cpl;
3571 int ret;
3572
3573 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3574 BUG_ON(!rpl_skb);
3575 if (req->retval) {
3576 PDBG("%s passive open failure %d\n", __func__, req->retval);
3577 mutex_lock(&dev->rdev.stats.lock);
3578 dev->rdev.stats.pas_ofld_conn_fails++;
3579 mutex_unlock(&dev->rdev.stats.lock);
3580 kfree_skb(rpl_skb);
3581 } else {
3582 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3583 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3584 (__force u32) htonl(
3585 (__force u32) req->tid)));
3586 ret = pass_accept_req(dev, rpl_skb);
3587 if (!ret)
3588 kfree_skb(rpl_skb);
3589 }
3590 return;
3591 }
3592
3593 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3594 {
3595 struct cpl_fw6_msg *rpl = cplhdr(skb);
3596 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3597
3598 switch (rpl->type) {
3599 case FW6_TYPE_CQE:
3600 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3601 break;
3602 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3603 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3604 switch (req->t_state) {
3605 case TCP_SYN_SENT:
3606 active_ofld_conn_reply(dev, skb, req);
3607 break;
3608 case TCP_SYN_RECV:
3609 passive_ofld_conn_reply(dev, skb, req);
3610 break;
3611 default:
3612 pr_err("%s unexpected ofld conn wr state %d\n",
3613 __func__, req->t_state);
3614 break;
3615 }
3616 break;
3617 }
3618 return 0;
3619 }
3620
3621 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3622 {
3623 __be32 l2info;
3624 __be16 hdr_len, vlantag, len;
3625 u16 eth_hdr_len;
3626 int tcp_hdr_len, ip_hdr_len;
3627 u8 intf;
3628 struct cpl_rx_pkt *cpl = cplhdr(skb);
3629 struct cpl_pass_accept_req *req;
3630 struct tcp_options_received tmp_opt;
3631 struct c4iw_dev *dev;
3632 enum chip_type type;
3633
3634 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3635 /* Store values from cpl_rx_pkt in temporary location. */
3636 vlantag = cpl->vlan;
3637 len = cpl->len;
3638 l2info = cpl->l2info;
3639 hdr_len = cpl->hdr_len;
3640 intf = cpl->iff;
3641
3642 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3643
3644 /*
3645 * We need to parse the TCP options from SYN packet.
3646 * to generate cpl_pass_accept_req.
3647 */
3648 memset(&tmp_opt, 0, sizeof(tmp_opt));
3649 tcp_clear_options(&tmp_opt);
3650 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3651
3652 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3653 memset(req, 0, sizeof(*req));
3654 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3655 SYN_MAC_IDX_V(RX_MACIDX_G(
3656 be32_to_cpu(l2info))) |
3657 SYN_XACT_MATCH_F);
3658 type = dev->rdev.lldi.adapter_type;
3659 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3660 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3661 req->hdr_len =
3662 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3663 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3664 eth_hdr_len = is_t4(type) ?
3665 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3666 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3667 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3668 IP_HDR_LEN_V(ip_hdr_len) |
3669 ETH_HDR_LEN_V(eth_hdr_len));
3670 } else { /* T6 and later */
3671 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3672 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3673 T6_IP_HDR_LEN_V(ip_hdr_len) |
3674 T6_ETH_HDR_LEN_V(eth_hdr_len));
3675 }
3676 req->vlan = vlantag;
3677 req->len = len;
3678 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3679 PASS_OPEN_TOS_V(tos));
3680 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3681 if (tmp_opt.wscale_ok)
3682 req->tcpopt.wsf = tmp_opt.snd_wscale;
3683 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3684 if (tmp_opt.sack_ok)
3685 req->tcpopt.sack = 1;
3686 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3687 return;
3688 }
3689
3690 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3691 __be32 laddr, __be16 lport,
3692 __be32 raddr, __be16 rport,
3693 u32 rcv_isn, u32 filter, u16 window,
3694 u32 rss_qid, u8 port_id)
3695 {
3696 struct sk_buff *req_skb;
3697 struct fw_ofld_connection_wr *req;
3698 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3699 int ret;
3700
3701 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3702 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3703 memset(req, 0, sizeof(*req));
3704 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3705 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3706 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3707 req->le.filter = (__force __be32) filter;
3708 req->le.lport = lport;
3709 req->le.pport = rport;
3710 req->le.u.ipv4.lip = laddr;
3711 req->le.u.ipv4.pip = raddr;
3712 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3713 req->tcb.rcv_adv = htons(window);
3714 req->tcb.t_state_to_astid =
3715 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3716 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3717 FW_OFLD_CONNECTION_WR_ASTID_V(
3718 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3719
3720 /*
3721 * We store the qid in opt2 which will be used by the firmware
3722 * to send us the wr response.
3723 */
3724 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3725
3726 /*
3727 * We initialize the MSS index in TCB to 0xF.
3728 * So that when driver sends cpl_pass_accept_rpl
3729 * TCB picks up the correct value. If this was 0
3730 * TP will ignore any value > 0 for MSS index.
3731 */
3732 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3733 req->cookie = (uintptr_t)skb;
3734
3735 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3736 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3737 if (ret < 0) {
3738 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3739 ret);
3740 kfree_skb(skb);
3741 kfree_skb(req_skb);
3742 }
3743 }
3744
3745 /*
3746 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3747 * messages when a filter is being used instead of server to
3748 * redirect a syn packet. When packets hit filter they are redirected
3749 * to the offload queue and driver tries to establish the connection
3750 * using firmware work request.
3751 */
3752 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3753 {
3754 int stid;
3755 unsigned int filter;
3756 struct ethhdr *eh = NULL;
3757 struct vlan_ethhdr *vlan_eh = NULL;
3758 struct iphdr *iph;
3759 struct tcphdr *tcph;
3760 struct rss_header *rss = (void *)skb->data;
3761 struct cpl_rx_pkt *cpl = (void *)skb->data;
3762 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3763 struct l2t_entry *e;
3764 struct dst_entry *dst;
3765 struct c4iw_ep *lep;
3766 u16 window;
3767 struct port_info *pi;
3768 struct net_device *pdev;
3769 u16 rss_qid, eth_hdr_len;
3770 int step;
3771 u32 tx_chan;
3772 struct neighbour *neigh;
3773
3774 /* Drop all non-SYN packets */
3775 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3776 goto reject;
3777
3778 /*
3779 * Drop all packets which did not hit the filter.
3780 * Unlikely to happen.
3781 */
3782 if (!(rss->filter_hit && rss->filter_tid))
3783 goto reject;
3784
3785 /*
3786 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3787 */
3788 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3789
3790 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
3791 if (!lep) {
3792 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3793 goto reject;
3794 }
3795
3796 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3797 case CHELSIO_T4:
3798 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3799 break;
3800 case CHELSIO_T5:
3801 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3802 break;
3803 case CHELSIO_T6:
3804 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3805 break;
3806 default:
3807 pr_err("T%d Chip is not supported\n",
3808 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3809 goto reject;
3810 }
3811
3812 if (eth_hdr_len == ETH_HLEN) {
3813 eh = (struct ethhdr *)(req + 1);
3814 iph = (struct iphdr *)(eh + 1);
3815 } else {
3816 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3817 iph = (struct iphdr *)(vlan_eh + 1);
3818 skb->vlan_tci = ntohs(cpl->vlan);
3819 }
3820
3821 if (iph->version != 0x4)
3822 goto reject;
3823
3824 tcph = (struct tcphdr *)(iph + 1);
3825 skb_set_network_header(skb, (void *)iph - (void *)rss);
3826 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
3827 skb_get(skb);
3828
3829 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
3830 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
3831 ntohs(tcph->source), iph->tos);
3832
3833 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
3834 iph->tos);
3835 if (!dst) {
3836 pr_err("%s - failed to find dst entry!\n",
3837 __func__);
3838 goto reject;
3839 }
3840 neigh = dst_neigh_lookup_skb(dst, skb);
3841
3842 if (!neigh) {
3843 pr_err("%s - failed to allocate neigh!\n",
3844 __func__);
3845 goto free_dst;
3846 }
3847
3848 if (neigh->dev->flags & IFF_LOOPBACK) {
3849 pdev = ip_dev_find(&init_net, iph->daddr);
3850 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3851 pdev, 0);
3852 pi = (struct port_info *)netdev_priv(pdev);
3853 tx_chan = cxgb4_port_chan(pdev);
3854 dev_put(pdev);
3855 } else {
3856 pdev = get_real_dev(neigh->dev);
3857 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
3858 pdev, 0);
3859 pi = (struct port_info *)netdev_priv(pdev);
3860 tx_chan = cxgb4_port_chan(pdev);
3861 }
3862 neigh_release(neigh);
3863 if (!e) {
3864 pr_err("%s - failed to allocate l2t entry!\n",
3865 __func__);
3866 goto free_dst;
3867 }
3868
3869 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
3870 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
3871 window = (__force u16) htons((__force u16)tcph->window);
3872
3873 /* Calcuate filter portion for LE region. */
3874 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
3875 dev->rdev.lldi.ports[0],
3876 e));
3877
3878 /*
3879 * Synthesize the cpl_pass_accept_req. We have everything except the
3880 * TID. Once firmware sends a reply with TID we update the TID field
3881 * in cpl and pass it through the regular cpl_pass_accept_req path.
3882 */
3883 build_cpl_pass_accept_req(skb, stid, iph->tos);
3884 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
3885 tcph->source, ntohl(tcph->seq), filter, window,
3886 rss_qid, pi->port_id);
3887 cxgb4_l2t_release(e);
3888 free_dst:
3889 dst_release(dst);
3890 reject:
3891 return 0;
3892 }
3893
3894 /*
3895 * These are the real handlers that are called from a
3896 * work queue.
3897 */
3898 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
3899 [CPL_ACT_ESTABLISH] = act_establish,
3900 [CPL_ACT_OPEN_RPL] = act_open_rpl,
3901 [CPL_RX_DATA] = rx_data,
3902 [CPL_ABORT_RPL_RSS] = abort_rpl,
3903 [CPL_ABORT_RPL] = abort_rpl,
3904 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
3905 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
3906 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
3907 [CPL_PASS_ESTABLISH] = pass_establish,
3908 [CPL_PEER_CLOSE] = peer_close,
3909 [CPL_ABORT_REQ_RSS] = peer_abort,
3910 [CPL_CLOSE_CON_RPL] = close_con_rpl,
3911 [CPL_RDMA_TERMINATE] = terminate,
3912 [CPL_FW4_ACK] = fw4_ack,
3913 [CPL_FW6_MSG] = deferred_fw6_msg,
3914 [CPL_RX_PKT] = rx_pkt,
3915 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe
3916 };
3917
3918 static void process_timeout(struct c4iw_ep *ep)
3919 {
3920 struct c4iw_qp_attributes attrs;
3921 int abort = 1;
3922
3923 mutex_lock(&ep->com.mutex);
3924 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
3925 ep->com.state);
3926 set_bit(TIMEDOUT, &ep->com.history);
3927 switch (ep->com.state) {
3928 case MPA_REQ_SENT:
3929 __state_set(&ep->com, ABORTING);
3930 connect_reply_upcall(ep, -ETIMEDOUT);
3931 break;
3932 case MPA_REQ_WAIT:
3933 __state_set(&ep->com, ABORTING);
3934 break;
3935 case CLOSING:
3936 case MORIBUND:
3937 if (ep->com.cm_id && ep->com.qp) {
3938 attrs.next_state = C4IW_QP_STATE_ERROR;
3939 c4iw_modify_qp(ep->com.qp->rhp,
3940 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3941 &attrs, 1);
3942 }
3943 __state_set(&ep->com, ABORTING);
3944 close_complete_upcall(ep, -ETIMEDOUT);
3945 break;
3946 case ABORTING:
3947 case DEAD:
3948
3949 /*
3950 * These states are expected if the ep timed out at the same
3951 * time as another thread was calling stop_ep_timer().
3952 * So we silently do nothing for these states.
3953 */
3954 abort = 0;
3955 break;
3956 default:
3957 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3958 __func__, ep, ep->hwtid, ep->com.state);
3959 abort = 0;
3960 }
3961 if (abort)
3962 abort_connection(ep, NULL, GFP_KERNEL);
3963 mutex_unlock(&ep->com.mutex);
3964 c4iw_put_ep(&ep->com);
3965 }
3966
3967 static void process_timedout_eps(void)
3968 {
3969 struct c4iw_ep *ep;
3970
3971 spin_lock_irq(&timeout_lock);
3972 while (!list_empty(&timeout_list)) {
3973 struct list_head *tmp;
3974
3975 tmp = timeout_list.next;
3976 list_del(tmp);
3977 tmp->next = NULL;
3978 tmp->prev = NULL;
3979 spin_unlock_irq(&timeout_lock);
3980 ep = list_entry(tmp, struct c4iw_ep, entry);
3981 process_timeout(ep);
3982 spin_lock_irq(&timeout_lock);
3983 }
3984 spin_unlock_irq(&timeout_lock);
3985 }
3986
3987 static void process_work(struct work_struct *work)
3988 {
3989 struct sk_buff *skb = NULL;
3990 struct c4iw_dev *dev;
3991 struct cpl_act_establish *rpl;
3992 unsigned int opcode;
3993 int ret;
3994
3995 process_timedout_eps();
3996 while ((skb = skb_dequeue(&rxq))) {
3997 rpl = cplhdr(skb);
3998 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3999 opcode = rpl->ot.opcode;
4000
4001 BUG_ON(!work_handlers[opcode]);
4002 ret = work_handlers[opcode](dev, skb);
4003 if (!ret)
4004 kfree_skb(skb);
4005 process_timedout_eps();
4006 }
4007 }
4008
4009 static DECLARE_WORK(skb_work, process_work);
4010
4011 static void ep_timeout(unsigned long arg)
4012 {
4013 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4014 int kickit = 0;
4015
4016 spin_lock(&timeout_lock);
4017 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4018 /*
4019 * Only insert if it is not already on the list.
4020 */
4021 if (!ep->entry.next) {
4022 list_add_tail(&ep->entry, &timeout_list);
4023 kickit = 1;
4024 }
4025 }
4026 spin_unlock(&timeout_lock);
4027 if (kickit)
4028 queue_work(workq, &skb_work);
4029 }
4030
4031 /*
4032 * All the CM events are handled on a work queue to have a safe context.
4033 */
4034 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4035 {
4036
4037 /*
4038 * Save dev in the skb->cb area.
4039 */
4040 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4041
4042 /*
4043 * Queue the skb and schedule the worker thread.
4044 */
4045 skb_queue_tail(&rxq, skb);
4046 queue_work(workq, &skb_work);
4047 return 0;
4048 }
4049
4050 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4051 {
4052 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4053
4054 if (rpl->status != CPL_ERR_NONE) {
4055 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4056 "for tid %u\n", rpl->status, GET_TID(rpl));
4057 }
4058 kfree_skb(skb);
4059 return 0;
4060 }
4061
4062 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4063 {
4064 struct cpl_fw6_msg *rpl = cplhdr(skb);
4065 struct c4iw_wr_wait *wr_waitp;
4066 int ret;
4067
4068 PDBG("%s type %u\n", __func__, rpl->type);
4069
4070 switch (rpl->type) {
4071 case FW6_TYPE_WR_RPL:
4072 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4073 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4074 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4075 if (wr_waitp)
4076 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4077 kfree_skb(skb);
4078 break;
4079 case FW6_TYPE_CQE:
4080 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4081 sched(dev, skb);
4082 break;
4083 default:
4084 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4085 rpl->type);
4086 kfree_skb(skb);
4087 break;
4088 }
4089 return 0;
4090 }
4091
4092 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4093 {
4094 struct cpl_abort_req_rss *req = cplhdr(skb);
4095 struct c4iw_ep *ep;
4096 struct tid_info *t = dev->rdev.lldi.tids;
4097 unsigned int tid = GET_TID(req);
4098
4099 ep = lookup_tid(t, tid);
4100 if (!ep) {
4101 printk(KERN_WARNING MOD
4102 "Abort on non-existent endpoint, tid %d\n", tid);
4103 kfree_skb(skb);
4104 return 0;
4105 }
4106 if (is_neg_adv(req->status)) {
4107 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4108 __func__, ep->hwtid, req->status,
4109 neg_adv_str(req->status));
4110 ep->stats.abort_neg_adv++;
4111 dev->rdev.stats.neg_adv++;
4112 kfree_skb(skb);
4113 return 0;
4114 }
4115 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4116 ep->com.state);
4117
4118 /*
4119 * Wake up any threads in rdma_init() or rdma_fini().
4120 * However, if we are on MPAv2 and want to retry with MPAv1
4121 * then, don't wake up yet.
4122 */
4123 if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
4124 if (ep->com.state != MPA_REQ_SENT)
4125 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4126 } else
4127 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4128 sched(dev, skb);
4129 return 0;
4130 }
4131
4132 /*
4133 * Most upcalls from the T4 Core go to sched() to
4134 * schedule the processing on a work queue.
4135 */
4136 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4137 [CPL_ACT_ESTABLISH] = sched,
4138 [CPL_ACT_OPEN_RPL] = sched,
4139 [CPL_RX_DATA] = sched,
4140 [CPL_ABORT_RPL_RSS] = sched,
4141 [CPL_ABORT_RPL] = sched,
4142 [CPL_PASS_OPEN_RPL] = sched,
4143 [CPL_CLOSE_LISTSRV_RPL] = sched,
4144 [CPL_PASS_ACCEPT_REQ] = sched,
4145 [CPL_PASS_ESTABLISH] = sched,
4146 [CPL_PEER_CLOSE] = sched,
4147 [CPL_CLOSE_CON_RPL] = sched,
4148 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4149 [CPL_RDMA_TERMINATE] = sched,
4150 [CPL_FW4_ACK] = sched,
4151 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4152 [CPL_FW6_MSG] = fw6_msg,
4153 [CPL_RX_PKT] = sched
4154 };
4155
4156 int __init c4iw_cm_init(void)
4157 {
4158 spin_lock_init(&timeout_lock);
4159 skb_queue_head_init(&rxq);
4160
4161 workq = create_singlethread_workqueue("iw_cxgb4");
4162 if (!workq)
4163 return -ENOMEM;
4164
4165 return 0;
4166 }
4167
4168 void c4iw_cm_term(void)
4169 {
4170 WARN_ON(!list_empty(&timeout_list));
4171 flush_workqueue(workq);
4172 destroy_workqueue(workq);
4173 }
Linux kernel - Deliverables
This page took 0.123653 seconds and 5 git commands to generate.