?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
libcxgb,iw_cxgb4,cxgbit: add cxgb_get_4tuple()
[linux/fpc-iii.git] / drivers / infiniband / hw / cxgb4 / cm.c
blobe591f61c5601176beca2598ff57d9448785ae90f
1 /*
2 * Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved.
3 *
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
9 *
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
13 *
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
17 *
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
22 *
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
31 */
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38 #include <linux/inetdevice.h>
39 #include <linux/ip.h>
40 #include <linux/tcp.h>
41 #include <linux/if_vlan.h>
42
43 #include <net/neighbour.h>
44 #include <net/netevent.h>
45 #include <net/route.h>
46 #include <net/tcp.h>
47 #include <net/ip6_route.h>
48 #include <net/addrconf.h>
49
50 #include <rdma/ib_addr.h>
51
52 #include <libcxgb_cm.h>
53 #include "iw_cxgb4.h"
54 #include "clip_tbl.h"
55
56 static char *states[] = {
57 "idle",
58 "listen",
59 "connecting",
60 "mpa_wait_req",
61 "mpa_req_sent",
62 "mpa_req_rcvd",
63 "mpa_rep_sent",
64 "fpdu_mode",
65 "aborting",
66 "closing",
67 "moribund",
68 "dead",
69 NULL,
70 };
71
72 static int nocong;
73 module_param(nocong, int, 0644);
74 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
75
76 static int enable_ecn;
77 module_param(enable_ecn, int, 0644);
78 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
79
80 static int dack_mode = 1;
81 module_param(dack_mode, int, 0644);
82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
83
84 uint c4iw_max_read_depth = 32;
85 module_param(c4iw_max_read_depth, int, 0644);
86 MODULE_PARM_DESC(c4iw_max_read_depth,
87 "Per-connection max ORD/IRD (default=32)");
88
89 static int enable_tcp_timestamps;
90 module_param(enable_tcp_timestamps, int, 0644);
91 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
92
93 static int enable_tcp_sack;
94 module_param(enable_tcp_sack, int, 0644);
95 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
96
97 static int enable_tcp_window_scaling = 1;
98 module_param(enable_tcp_window_scaling, int, 0644);
99 MODULE_PARM_DESC(enable_tcp_window_scaling,
100 "Enable tcp window scaling (default=1)");
101
102 int c4iw_debug;
103 module_param(c4iw_debug, int, 0644);
104 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
105
106 static int peer2peer = 1;
107 module_param(peer2peer, int, 0644);
108 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
109
110 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
111 module_param(p2p_type, int, 0644);
112 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
113 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
114
115 static int ep_timeout_secs = 60;
116 module_param(ep_timeout_secs, int, 0644);
117 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
118 "in seconds (default=60)");
119
120 static int mpa_rev = 2;
121 module_param(mpa_rev, int, 0644);
122 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
123 "1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
124 " compliant (default=2)");
125
126 static int markers_enabled;
127 module_param(markers_enabled, int, 0644);
128 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
129
130 static int crc_enabled = 1;
131 module_param(crc_enabled, int, 0644);
132 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
133
134 static int rcv_win = 256 * 1024;
135 module_param(rcv_win, int, 0644);
136 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
137
138 static int snd_win = 128 * 1024;
139 module_param(snd_win, int, 0644);
140 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
141
142 static struct workqueue_struct *workq;
143
144 static struct sk_buff_head rxq;
145
146 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
147 static void ep_timeout(unsigned long arg);
148 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
149 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
150
151 static LIST_HEAD(timeout_list);
152 static spinlock_t timeout_lock;
153
154 static void deref_cm_id(struct c4iw_ep_common *epc)
155 {
156 epc->cm_id->rem_ref(epc->cm_id);
157 epc->cm_id = NULL;
158 set_bit(CM_ID_DEREFED, &epc->history);
159 }
160
161 static void ref_cm_id(struct c4iw_ep_common *epc)
162 {
163 set_bit(CM_ID_REFED, &epc->history);
164 epc->cm_id->add_ref(epc->cm_id);
165 }
166
167 static void deref_qp(struct c4iw_ep *ep)
168 {
169 c4iw_qp_rem_ref(&ep->com.qp->ibqp);
170 clear_bit(QP_REFERENCED, &ep->com.flags);
171 set_bit(QP_DEREFED, &ep->com.history);
172 }
173
174 static void ref_qp(struct c4iw_ep *ep)
175 {
176 set_bit(QP_REFERENCED, &ep->com.flags);
177 set_bit(QP_REFED, &ep->com.history);
178 c4iw_qp_add_ref(&ep->com.qp->ibqp);
179 }
180
181 static void start_ep_timer(struct c4iw_ep *ep)
182 {
183 PDBG("%s ep %p\n", __func__, ep);
184 if (timer_pending(&ep->timer)) {
185 pr_err("%s timer already started! ep %p\n",
186 __func__, ep);
187 return;
188 }
189 clear_bit(TIMEOUT, &ep->com.flags);
190 c4iw_get_ep(&ep->com);
191 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
192 ep->timer.data = (unsigned long)ep;
193 ep->timer.function = ep_timeout;
194 add_timer(&ep->timer);
195 }
196
197 static int stop_ep_timer(struct c4iw_ep *ep)
198 {
199 PDBG("%s ep %p stopping\n", __func__, ep);
200 del_timer_sync(&ep->timer);
201 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
202 c4iw_put_ep(&ep->com);
203 return 0;
204 }
205 return 1;
206 }
207
208 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
209 struct l2t_entry *l2e)
210 {
211 int error = 0;
212
213 if (c4iw_fatal_error(rdev)) {
214 kfree_skb(skb);
215 PDBG("%s - device in error state - dropping\n", __func__);
216 return -EIO;
217 }
218 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
219 if (error < 0)
220 kfree_skb(skb);
221 else if (error == NET_XMIT_DROP)
222 return -ENOMEM;
223 return error < 0 ? error : 0;
224 }
225
226 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
227 {
228 int error = 0;
229
230 if (c4iw_fatal_error(rdev)) {
231 kfree_skb(skb);
232 PDBG("%s - device in error state - dropping\n", __func__);
233 return -EIO;
234 }
235 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
236 if (error < 0)
237 kfree_skb(skb);
238 return error < 0 ? error : 0;
239 }
240
241 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
242 {
243 struct cpl_tid_release *req;
244
245 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
246 if (!skb)
247 return;
248 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
249 INIT_TP_WR(req, hwtid);
250 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
251 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
252 c4iw_ofld_send(rdev, skb);
253 return;
254 }
255
256 static void set_emss(struct c4iw_ep *ep, u16 opt)
257 {
258 ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
259 ((AF_INET == ep->com.remote_addr.ss_family) ?
260 sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
261 sizeof(struct tcphdr);
262 ep->mss = ep->emss;
263 if (TCPOPT_TSTAMP_G(opt))
264 ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
265 if (ep->emss < 128)
266 ep->emss = 128;
267 if (ep->emss & 7)
268 PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
269 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
270 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, TCPOPT_MSS_G(opt),
271 ep->mss, ep->emss);
272 }
273
274 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
275 {
276 enum c4iw_ep_state state;
277
278 mutex_lock(&epc->mutex);
279 state = epc->state;
280 mutex_unlock(&epc->mutex);
281 return state;
282 }
283
284 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
285 {
286 epc->state = new;
287 }
288
289 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
290 {
291 mutex_lock(&epc->mutex);
292 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
293 __state_set(epc, new);
294 mutex_unlock(&epc->mutex);
295 return;
296 }
297
298 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
299 {
300 struct sk_buff *skb;
301 unsigned int i;
302 size_t len;
303
304 len = roundup(sizeof(union cpl_wr_size), 16);
305 for (i = 0; i < size; i++) {
306 skb = alloc_skb(len, GFP_KERNEL);
307 if (!skb)
308 goto fail;
309 skb_queue_tail(ep_skb_list, skb);
310 }
311 return 0;
312 fail:
313 skb_queue_purge(ep_skb_list);
314 return -ENOMEM;
315 }
316
317 static void *alloc_ep(int size, gfp_t gfp)
318 {
319 struct c4iw_ep_common *epc;
320
321 epc = kzalloc(size, gfp);
322 if (epc) {
323 kref_init(&epc->kref);
324 mutex_init(&epc->mutex);
325 c4iw_init_wr_wait(&epc->wr_wait);
326 }
327 PDBG("%s alloc ep %p\n", __func__, epc);
328 return epc;
329 }
330
331 static void remove_ep_tid(struct c4iw_ep *ep)
332 {
333 unsigned long flags;
334
335 spin_lock_irqsave(&ep->com.dev->lock, flags);
336 _remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid, 0);
337 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
338 }
339
340 static void insert_ep_tid(struct c4iw_ep *ep)
341 {
342 unsigned long flags;
343
344 spin_lock_irqsave(&ep->com.dev->lock, flags);
345 _insert_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep, ep->hwtid, 0);
346 spin_unlock_irqrestore(&ep->com.dev->lock, flags);
347 }
348
349 /*
350 * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
351 */
352 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
353 {
354 struct c4iw_ep *ep;
355 unsigned long flags;
356
357 spin_lock_irqsave(&dev->lock, flags);
358 ep = idr_find(&dev->hwtid_idr, tid);
359 if (ep)
360 c4iw_get_ep(&ep->com);
361 spin_unlock_irqrestore(&dev->lock, flags);
362 return ep;
363 }
364
365 /*
366 * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
367 */
368 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
369 unsigned int stid)
370 {
371 struct c4iw_listen_ep *ep;
372 unsigned long flags;
373
374 spin_lock_irqsave(&dev->lock, flags);
375 ep = idr_find(&dev->stid_idr, stid);
376 if (ep)
377 c4iw_get_ep(&ep->com);
378 spin_unlock_irqrestore(&dev->lock, flags);
379 return ep;
380 }
381
382 void _c4iw_free_ep(struct kref *kref)
383 {
384 struct c4iw_ep *ep;
385
386 ep = container_of(kref, struct c4iw_ep, com.kref);
387 PDBG("%s ep %p state %s\n", __func__, ep, states[ep->com.state]);
388 if (test_bit(QP_REFERENCED, &ep->com.flags))
389 deref_qp(ep);
390 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
391 if (ep->com.remote_addr.ss_family == AF_INET6) {
392 struct sockaddr_in6 *sin6 =
393 (struct sockaddr_in6 *)
394 &ep->com.local_addr;
395
396 cxgb4_clip_release(
397 ep->com.dev->rdev.lldi.ports[0],
398 (const u32 *)&sin6->sin6_addr.s6_addr,
399 1);
400 }
401 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
402 dst_release(ep->dst);
403 cxgb4_l2t_release(ep->l2t);
404 if (ep->mpa_skb)
405 kfree_skb(ep->mpa_skb);
406 }
407 if (!skb_queue_empty(&ep->com.ep_skb_list))
408 skb_queue_purge(&ep->com.ep_skb_list);
409 kfree(ep);
410 }
411
412 static void release_ep_resources(struct c4iw_ep *ep)
413 {
414 set_bit(RELEASE_RESOURCES, &ep->com.flags);
415
416 /*
417 * If we have a hwtid, then remove it from the idr table
418 * so lookups will no longer find this endpoint. Otherwise
419 * we have a race where one thread finds the ep ptr just
420 * before the other thread is freeing the ep memory.
421 */
422 if (ep->hwtid != -1)
423 remove_ep_tid(ep);
424 c4iw_put_ep(&ep->com);
425 }
426
427 static int status2errno(int status)
428 {
429 switch (status) {
430 case CPL_ERR_NONE:
431 return 0;
432 case CPL_ERR_CONN_RESET:
433 return -ECONNRESET;
434 case CPL_ERR_ARP_MISS:
435 return -EHOSTUNREACH;
436 case CPL_ERR_CONN_TIMEDOUT:
437 return -ETIMEDOUT;
438 case CPL_ERR_TCAM_FULL:
439 return -ENOMEM;
440 case CPL_ERR_CONN_EXIST:
441 return -EADDRINUSE;
442 default:
443 return -EIO;
444 }
445 }
446
447 /*
448 * Try and reuse skbs already allocated...
449 */
450 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
451 {
452 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
453 skb_trim(skb, 0);
454 skb_get(skb);
455 skb_reset_transport_header(skb);
456 } else {
457 skb = alloc_skb(len, gfp);
458 }
459 t4_set_arp_err_handler(skb, NULL, NULL);
460 return skb;
461 }
462
463 static struct net_device *get_real_dev(struct net_device *egress_dev)
464 {
465 return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
466 }
467
468 static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
469 {
470 int i;
471
472 egress_dev = get_real_dev(egress_dev);
473 for (i = 0; i < dev->rdev.lldi.nports; i++)
474 if (dev->rdev.lldi.ports[i] == egress_dev)
475 return 1;
476 return 0;
477 }
478
479 static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
480 __u8 *peer_ip, __be16 local_port,
481 __be16 peer_port, u8 tos,
482 __u32 sin6_scope_id)
483 {
484 struct dst_entry *dst = NULL;
485
486 if (IS_ENABLED(CONFIG_IPV6)) {
487 struct flowi6 fl6;
488
489 memset(&fl6, 0, sizeof(fl6));
490 memcpy(&fl6.daddr, peer_ip, 16);
491 memcpy(&fl6.saddr, local_ip, 16);
492 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
493 fl6.flowi6_oif = sin6_scope_id;
494 dst = ip6_route_output(&init_net, NULL, &fl6);
495 if (!dst)
496 goto out;
497 if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
498 !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
499 dst_release(dst);
500 dst = NULL;
501 }
502 }
503
504 out:
505 return dst;
506 }
507
508 static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
509 __be32 peer_ip, __be16 local_port,
510 __be16 peer_port, u8 tos)
511 {
512 struct rtable *rt;
513 struct flowi4 fl4;
514 struct neighbour *n;
515
516 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
517 peer_port, local_port, IPPROTO_TCP,
518 tos, 0);
519 if (IS_ERR(rt))
520 return NULL;
521 n = dst_neigh_lookup(&rt->dst, &peer_ip);
522 if (!n)
523 return NULL;
524 if (!our_interface(dev, n->dev) &&
525 !(n->dev->flags & IFF_LOOPBACK)) {
526 neigh_release(n);
527 dst_release(&rt->dst);
528 return NULL;
529 }
530 neigh_release(n);
531 return &rt->dst;
532 }
533
534 static void arp_failure_discard(void *handle, struct sk_buff *skb)
535 {
536 pr_err(MOD "ARP failure\n");
537 kfree_skb(skb);
538 }
539
540 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
541 {
542 pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
543 }
544
545 enum {
546 NUM_FAKE_CPLS = 2,
547 FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
548 FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
549 };
550
551 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
552 {
553 struct c4iw_ep *ep;
554
555 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
556 release_ep_resources(ep);
557 return 0;
558 }
559
560 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
561 {
562 struct c4iw_ep *ep;
563
564 ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
565 c4iw_put_ep(&ep->parent_ep->com);
566 release_ep_resources(ep);
567 return 0;
568 }
569
570 /*
571 * Fake up a special CPL opcode and call sched() so process_work() will call
572 * _put_ep_safe() in a safe context to free the ep resources. This is needed
573 * because ARP error handlers are called in an ATOMIC context, and
574 * _c4iw_free_ep() needs to block.
575 */
576 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
577 int cpl)
578 {
579 struct cpl_act_establish *rpl = cplhdr(skb);
580
581 /* Set our special ARP_FAILURE opcode */
582 rpl->ot.opcode = cpl;
583
584 /*
585 * Save ep in the skb->cb area, after where sched() will save the dev
586 * ptr.
587 */
588 *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
589 sched(ep->com.dev, skb);
590 }
591
592 /* Handle an ARP failure for an accept */
593 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
594 {
595 struct c4iw_ep *ep = handle;
596
597 pr_err(MOD "ARP failure during accept - tid %u -dropping connection\n",
598 ep->hwtid);
599
600 __state_set(&ep->com, DEAD);
601 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
602 }
603
604 /*
605 * Handle an ARP failure for an active open.
606 */
607 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
608 {
609 struct c4iw_ep *ep = handle;
610
611 printk(KERN_ERR MOD "ARP failure during connect\n");
612 connect_reply_upcall(ep, -EHOSTUNREACH);
613 __state_set(&ep->com, DEAD);
614 if (ep->com.remote_addr.ss_family == AF_INET6) {
615 struct sockaddr_in6 *sin6 =
616 (struct sockaddr_in6 *)&ep->com.local_addr;
617 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
618 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
619 }
620 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
621 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
622 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
623 }
624
625 /*
626 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
627 * and send it along.
628 */
629 static void abort_arp_failure(void *handle, struct sk_buff *skb)
630 {
631 int ret;
632 struct c4iw_ep *ep = handle;
633 struct c4iw_rdev *rdev = &ep->com.dev->rdev;
634 struct cpl_abort_req *req = cplhdr(skb);
635
636 PDBG("%s rdev %p\n", __func__, rdev);
637 req->cmd = CPL_ABORT_NO_RST;
638 ret = c4iw_ofld_send(rdev, skb);
639 if (ret) {
640 __state_set(&ep->com, DEAD);
641 queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
642 }
643 }
644
645 static int send_flowc(struct c4iw_ep *ep)
646 {
647 struct fw_flowc_wr *flowc;
648 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
649 int i;
650 u16 vlan = ep->l2t->vlan;
651 int nparams;
652
653 if (WARN_ON(!skb))
654 return -ENOMEM;
655
656 if (vlan == CPL_L2T_VLAN_NONE)
657 nparams = 8;
658 else
659 nparams = 9;
660
661 flowc = (struct fw_flowc_wr *)__skb_put(skb, FLOWC_LEN);
662
663 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
664 FW_FLOWC_WR_NPARAMS_V(nparams));
665 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(DIV_ROUND_UP(FLOWC_LEN,
666 16)) | FW_WR_FLOWID_V(ep->hwtid));
667
668 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
669 flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
670 (ep->com.dev->rdev.lldi.pf));
671 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
672 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
673 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
674 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
675 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
676 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
677 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
678 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
679 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
680 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
681 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
682 flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
683 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
684 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
685 if (nparams == 9) {
686 u16 pri;
687
688 pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
689 flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
690 flowc->mnemval[8].val = cpu_to_be32(pri);
691 } else {
692 /* Pad WR to 16 byte boundary */
693 flowc->mnemval[8].mnemonic = 0;
694 flowc->mnemval[8].val = 0;
695 }
696 for (i = 0; i < 9; i++) {
697 flowc->mnemval[i].r4[0] = 0;
698 flowc->mnemval[i].r4[1] = 0;
699 flowc->mnemval[i].r4[2] = 0;
700 }
701
702 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
703 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
704 }
705
706 static int send_halfclose(struct c4iw_ep *ep)
707 {
708 struct cpl_close_con_req *req;
709 struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
710 int wrlen = roundup(sizeof *req, 16);
711
712 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
713 if (WARN_ON(!skb))
714 return -ENOMEM;
715
716 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
717 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
718 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
719 memset(req, 0, wrlen);
720 INIT_TP_WR(req, ep->hwtid);
721 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
722 ep->hwtid));
723 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
724 }
725
726 static int send_abort(struct c4iw_ep *ep)
727 {
728 struct cpl_abort_req *req;
729 int wrlen = roundup(sizeof *req, 16);
730 struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
731
732 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
733 if (WARN_ON(!req_skb))
734 return -ENOMEM;
735
736 set_wr_txq(req_skb, CPL_PRIORITY_DATA, ep->txq_idx);
737 t4_set_arp_err_handler(req_skb, ep, abort_arp_failure);
738 req = (struct cpl_abort_req *)skb_put(req_skb, wrlen);
739 memset(req, 0, wrlen);
740 INIT_TP_WR(req, ep->hwtid);
741 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
742 req->cmd = CPL_ABORT_SEND_RST;
743 return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
744 }
745
746 static void best_mtu(const unsigned short *mtus, unsigned short mtu,
747 unsigned int *idx, int use_ts, int ipv6)
748 {
749 unsigned short hdr_size = (ipv6 ?
750 sizeof(struct ipv6hdr) :
751 sizeof(struct iphdr)) +
752 sizeof(struct tcphdr) +
753 (use_ts ?
754 round_up(TCPOLEN_TIMESTAMP, 4) : 0);
755 unsigned short data_size = mtu - hdr_size;
756
757 cxgb4_best_aligned_mtu(mtus, hdr_size, data_size, 8, idx);
758 }
759
760 static int send_connect(struct c4iw_ep *ep)
761 {
762 struct cpl_act_open_req *req = NULL;
763 struct cpl_t5_act_open_req *t5req = NULL;
764 struct cpl_t6_act_open_req *t6req = NULL;
765 struct cpl_act_open_req6 *req6 = NULL;
766 struct cpl_t5_act_open_req6 *t5req6 = NULL;
767 struct cpl_t6_act_open_req6 *t6req6 = NULL;
768 struct sk_buff *skb;
769 u64 opt0;
770 u32 opt2;
771 unsigned int mtu_idx;
772 int wscale;
773 int win, sizev4, sizev6, wrlen;
774 struct sockaddr_in *la = (struct sockaddr_in *)
775 &ep->com.local_addr;
776 struct sockaddr_in *ra = (struct sockaddr_in *)
777 &ep->com.remote_addr;
778 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
779 &ep->com.local_addr;
780 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
781 &ep->com.remote_addr;
782 int ret;
783 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
784 u32 isn = (prandom_u32() & ~7UL) - 1;
785
786 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
787 case CHELSIO_T4:
788 sizev4 = sizeof(struct cpl_act_open_req);
789 sizev6 = sizeof(struct cpl_act_open_req6);
790 break;
791 case CHELSIO_T5:
792 sizev4 = sizeof(struct cpl_t5_act_open_req);
793 sizev6 = sizeof(struct cpl_t5_act_open_req6);
794 break;
795 case CHELSIO_T6:
796 sizev4 = sizeof(struct cpl_t6_act_open_req);
797 sizev6 = sizeof(struct cpl_t6_act_open_req6);
798 break;
799 default:
800 pr_err("T%d Chip is not supported\n",
801 CHELSIO_CHIP_VERSION(adapter_type));
802 return -EINVAL;
803 }
804
805 wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
806 roundup(sizev4, 16) :
807 roundup(sizev6, 16);
808
809 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
810
811 skb = get_skb(NULL, wrlen, GFP_KERNEL);
812 if (!skb) {
813 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
814 __func__);
815 return -ENOMEM;
816 }
817 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
818
819 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
820 enable_tcp_timestamps,
821 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
822 wscale = compute_wscale(rcv_win);
823
824 /*
825 * Specify the largest window that will fit in opt0. The
826 * remainder will be specified in the rx_data_ack.
827 */
828 win = ep->rcv_win >> 10;
829 if (win > RCV_BUFSIZ_M)
830 win = RCV_BUFSIZ_M;
831
832 opt0 = (nocong ? NO_CONG_F : 0) |
833 KEEP_ALIVE_F |
834 DELACK_F |
835 WND_SCALE_V(wscale) |
836 MSS_IDX_V(mtu_idx) |
837 L2T_IDX_V(ep->l2t->idx) |
838 TX_CHAN_V(ep->tx_chan) |
839 SMAC_SEL_V(ep->smac_idx) |
840 DSCP_V(ep->tos >> 2) |
841 ULP_MODE_V(ULP_MODE_TCPDDP) |
842 RCV_BUFSIZ_V(win);
843 opt2 = RX_CHANNEL_V(0) |
844 CCTRL_ECN_V(enable_ecn) |
845 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
846 if (enable_tcp_timestamps)
847 opt2 |= TSTAMPS_EN_F;
848 if (enable_tcp_sack)
849 opt2 |= SACK_EN_F;
850 if (wscale && enable_tcp_window_scaling)
851 opt2 |= WND_SCALE_EN_F;
852 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
853 if (peer2peer)
854 isn += 4;
855
856 opt2 |= T5_OPT_2_VALID_F;
857 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
858 opt2 |= T5_ISS_F;
859 }
860
861 if (ep->com.remote_addr.ss_family == AF_INET6)
862 cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
863 (const u32 *)&la6->sin6_addr.s6_addr, 1);
864
865 t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
866
867 if (ep->com.remote_addr.ss_family == AF_INET) {
868 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
869 case CHELSIO_T4:
870 req = (struct cpl_act_open_req *)skb_put(skb, wrlen);
871 INIT_TP_WR(req, 0);
872 break;
873 case CHELSIO_T5:
874 t5req = (struct cpl_t5_act_open_req *)skb_put(skb,
875 wrlen);
876 INIT_TP_WR(t5req, 0);
877 req = (struct cpl_act_open_req *)t5req;
878 break;
879 case CHELSIO_T6:
880 t6req = (struct cpl_t6_act_open_req *)skb_put(skb,
881 wrlen);
882 INIT_TP_WR(t6req, 0);
883 req = (struct cpl_act_open_req *)t6req;
884 t5req = (struct cpl_t5_act_open_req *)t6req;
885 break;
886 default:
887 pr_err("T%d Chip is not supported\n",
888 CHELSIO_CHIP_VERSION(adapter_type));
889 ret = -EINVAL;
890 goto clip_release;
891 }
892
893 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
894 ((ep->rss_qid<<14) | ep->atid)));
895 req->local_port = la->sin_port;
896 req->peer_port = ra->sin_port;
897 req->local_ip = la->sin_addr.s_addr;
898 req->peer_ip = ra->sin_addr.s_addr;
899 req->opt0 = cpu_to_be64(opt0);
900
901 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
902 req->params = cpu_to_be32(cxgb4_select_ntuple(
903 ep->com.dev->rdev.lldi.ports[0],
904 ep->l2t));
905 req->opt2 = cpu_to_be32(opt2);
906 } else {
907 t5req->params = cpu_to_be64(FILTER_TUPLE_V(
908 cxgb4_select_ntuple(
909 ep->com.dev->rdev.lldi.ports[0],
910 ep->l2t)));
911 t5req->rsvd = cpu_to_be32(isn);
912 PDBG("%s snd_isn %u\n", __func__, t5req->rsvd);
913 t5req->opt2 = cpu_to_be32(opt2);
914 }
915 } else {
916 switch (CHELSIO_CHIP_VERSION(adapter_type)) {
917 case CHELSIO_T4:
918 req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
919 INIT_TP_WR(req6, 0);
920 break;
921 case CHELSIO_T5:
922 t5req6 = (struct cpl_t5_act_open_req6 *)skb_put(skb,
923 wrlen);
924 INIT_TP_WR(t5req6, 0);
925 req6 = (struct cpl_act_open_req6 *)t5req6;
926 break;
927 case CHELSIO_T6:
928 t6req6 = (struct cpl_t6_act_open_req6 *)skb_put(skb,
929 wrlen);
930 INIT_TP_WR(t6req6, 0);
931 req6 = (struct cpl_act_open_req6 *)t6req6;
932 t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
933 break;
934 default:
935 pr_err("T%d Chip is not supported\n",
936 CHELSIO_CHIP_VERSION(adapter_type));
937 ret = -EINVAL;
938 goto clip_release;
939 }
940
941 OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
942 ((ep->rss_qid<<14)|ep->atid)));
943 req6->local_port = la6->sin6_port;
944 req6->peer_port = ra6->sin6_port;
945 req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
946 req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
947 req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
948 req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
949 req6->opt0 = cpu_to_be64(opt0);
950
951 if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
952 req6->params = cpu_to_be32(cxgb4_select_ntuple(
953 ep->com.dev->rdev.lldi.ports[0],
954 ep->l2t));
955 req6->opt2 = cpu_to_be32(opt2);
956 } else {
957 t5req6->params = cpu_to_be64(FILTER_TUPLE_V(
958 cxgb4_select_ntuple(
959 ep->com.dev->rdev.lldi.ports[0],
960 ep->l2t)));
961 t5req6->rsvd = cpu_to_be32(isn);
962 PDBG("%s snd_isn %u\n", __func__, t5req6->rsvd);
963 t5req6->opt2 = cpu_to_be32(opt2);
964 }
965 }
966
967 set_bit(ACT_OPEN_REQ, &ep->com.history);
968 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
969 clip_release:
970 if (ret && ep->com.remote_addr.ss_family == AF_INET6)
971 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
972 (const u32 *)&la6->sin6_addr.s6_addr, 1);
973 return ret;
974 }
975
976 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
977 u8 mpa_rev_to_use)
978 {
979 int mpalen, wrlen, ret;
980 struct fw_ofld_tx_data_wr *req;
981 struct mpa_message *mpa;
982 struct mpa_v2_conn_params mpa_v2_params;
983
984 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
985
986 BUG_ON(skb_cloned(skb));
987
988 mpalen = sizeof(*mpa) + ep->plen;
989 if (mpa_rev_to_use == 2)
990 mpalen += sizeof(struct mpa_v2_conn_params);
991 wrlen = roundup(mpalen + sizeof *req, 16);
992 skb = get_skb(skb, wrlen, GFP_KERNEL);
993 if (!skb) {
994 connect_reply_upcall(ep, -ENOMEM);
995 return -ENOMEM;
996 }
997 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
998
999 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1000 memset(req, 0, wrlen);
1001 req->op_to_immdlen = cpu_to_be32(
1002 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1003 FW_WR_COMPL_F |
1004 FW_WR_IMMDLEN_V(mpalen));
1005 req->flowid_len16 = cpu_to_be32(
1006 FW_WR_FLOWID_V(ep->hwtid) |
1007 FW_WR_LEN16_V(wrlen >> 4));
1008 req->plen = cpu_to_be32(mpalen);
1009 req->tunnel_to_proxy = cpu_to_be32(
1010 FW_OFLD_TX_DATA_WR_FLUSH_F |
1011 FW_OFLD_TX_DATA_WR_SHOVE_F);
1012
1013 mpa = (struct mpa_message *)(req + 1);
1014 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
1015
1016 mpa->flags = 0;
1017 if (crc_enabled)
1018 mpa->flags |= MPA_CRC;
1019 if (markers_enabled) {
1020 mpa->flags |= MPA_MARKERS;
1021 ep->mpa_attr.recv_marker_enabled = 1;
1022 } else {
1023 ep->mpa_attr.recv_marker_enabled = 0;
1024 }
1025 if (mpa_rev_to_use == 2)
1026 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1027
1028 mpa->private_data_size = htons(ep->plen);
1029 mpa->revision = mpa_rev_to_use;
1030 if (mpa_rev_to_use == 1) {
1031 ep->tried_with_mpa_v1 = 1;
1032 ep->retry_with_mpa_v1 = 0;
1033 }
1034
1035 if (mpa_rev_to_use == 2) {
1036 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1037 sizeof (struct mpa_v2_conn_params));
1038 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1039 ep->ord);
1040 mpa_v2_params.ird = htons((u16)ep->ird);
1041 mpa_v2_params.ord = htons((u16)ep->ord);
1042
1043 if (peer2peer) {
1044 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1045 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1046 mpa_v2_params.ord |=
1047 htons(MPA_V2_RDMA_WRITE_RTR);
1048 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1049 mpa_v2_params.ord |=
1050 htons(MPA_V2_RDMA_READ_RTR);
1051 }
1052 memcpy(mpa->private_data, &mpa_v2_params,
1053 sizeof(struct mpa_v2_conn_params));
1054
1055 if (ep->plen)
1056 memcpy(mpa->private_data +
1057 sizeof(struct mpa_v2_conn_params),
1058 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1059 } else
1060 if (ep->plen)
1061 memcpy(mpa->private_data,
1062 ep->mpa_pkt + sizeof(*mpa), ep->plen);
1063
1064 /*
1065 * Reference the mpa skb. This ensures the data area
1066 * will remain in memory until the hw acks the tx.
1067 * Function fw4_ack() will deref it.
1068 */
1069 skb_get(skb);
1070 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1071 BUG_ON(ep->mpa_skb);
1072 ep->mpa_skb = skb;
1073 ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1074 if (ret)
1075 return ret;
1076 start_ep_timer(ep);
1077 __state_set(&ep->com, MPA_REQ_SENT);
1078 ep->mpa_attr.initiator = 1;
1079 ep->snd_seq += mpalen;
1080 return ret;
1081 }
1082
1083 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1084 {
1085 int mpalen, wrlen;
1086 struct fw_ofld_tx_data_wr *req;
1087 struct mpa_message *mpa;
1088 struct sk_buff *skb;
1089 struct mpa_v2_conn_params mpa_v2_params;
1090
1091 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1092
1093 mpalen = sizeof(*mpa) + plen;
1094 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1095 mpalen += sizeof(struct mpa_v2_conn_params);
1096 wrlen = roundup(mpalen + sizeof *req, 16);
1097
1098 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1099 if (!skb) {
1100 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1101 return -ENOMEM;
1102 }
1103 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1104
1105 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
1106 memset(req, 0, wrlen);
1107 req->op_to_immdlen = cpu_to_be32(
1108 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1109 FW_WR_COMPL_F |
1110 FW_WR_IMMDLEN_V(mpalen));
1111 req->flowid_len16 = cpu_to_be32(
1112 FW_WR_FLOWID_V(ep->hwtid) |
1113 FW_WR_LEN16_V(wrlen >> 4));
1114 req->plen = cpu_to_be32(mpalen);
1115 req->tunnel_to_proxy = cpu_to_be32(
1116 FW_OFLD_TX_DATA_WR_FLUSH_F |
1117 FW_OFLD_TX_DATA_WR_SHOVE_F);
1118
1119 mpa = (struct mpa_message *)(req + 1);
1120 memset(mpa, 0, sizeof(*mpa));
1121 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1122 mpa->flags = MPA_REJECT;
1123 mpa->revision = ep->mpa_attr.version;
1124 mpa->private_data_size = htons(plen);
1125
1126 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1127 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1128 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1129 sizeof (struct mpa_v2_conn_params));
1130 mpa_v2_params.ird = htons(((u16)ep->ird) |
1131 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1132 0));
1133 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1134 (p2p_type ==
1135 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1136 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1137 FW_RI_INIT_P2PTYPE_READ_REQ ?
1138 MPA_V2_RDMA_READ_RTR : 0) : 0));
1139 memcpy(mpa->private_data, &mpa_v2_params,
1140 sizeof(struct mpa_v2_conn_params));
1141
1142 if (ep->plen)
1143 memcpy(mpa->private_data +
1144 sizeof(struct mpa_v2_conn_params), pdata, plen);
1145 } else
1146 if (plen)
1147 memcpy(mpa->private_data, pdata, plen);
1148
1149 /*
1150 * Reference the mpa skb again. This ensures the data area
1151 * will remain in memory until the hw acks the tx.
1152 * Function fw4_ack() will deref it.
1153 */
1154 skb_get(skb);
1155 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1156 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1157 BUG_ON(ep->mpa_skb);
1158 ep->mpa_skb = skb;
1159 ep->snd_seq += mpalen;
1160 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1161 }
1162
1163 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1164 {
1165 int mpalen, wrlen;
1166 struct fw_ofld_tx_data_wr *req;
1167 struct mpa_message *mpa;
1168 struct sk_buff *skb;
1169 struct mpa_v2_conn_params mpa_v2_params;
1170
1171 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
1172
1173 mpalen = sizeof(*mpa) + plen;
1174 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1175 mpalen += sizeof(struct mpa_v2_conn_params);
1176 wrlen = roundup(mpalen + sizeof *req, 16);
1177
1178 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1179 if (!skb) {
1180 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
1181 return -ENOMEM;
1182 }
1183 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1184
1185 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
1186 memset(req, 0, wrlen);
1187 req->op_to_immdlen = cpu_to_be32(
1188 FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1189 FW_WR_COMPL_F |
1190 FW_WR_IMMDLEN_V(mpalen));
1191 req->flowid_len16 = cpu_to_be32(
1192 FW_WR_FLOWID_V(ep->hwtid) |
1193 FW_WR_LEN16_V(wrlen >> 4));
1194 req->plen = cpu_to_be32(mpalen);
1195 req->tunnel_to_proxy = cpu_to_be32(
1196 FW_OFLD_TX_DATA_WR_FLUSH_F |
1197 FW_OFLD_TX_DATA_WR_SHOVE_F);
1198
1199 mpa = (struct mpa_message *)(req + 1);
1200 memset(mpa, 0, sizeof(*mpa));
1201 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1202 mpa->flags = 0;
1203 if (ep->mpa_attr.crc_enabled)
1204 mpa->flags |= MPA_CRC;
1205 if (ep->mpa_attr.recv_marker_enabled)
1206 mpa->flags |= MPA_MARKERS;
1207 mpa->revision = ep->mpa_attr.version;
1208 mpa->private_data_size = htons(plen);
1209
1210 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1211 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1212 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1213 sizeof (struct mpa_v2_conn_params));
1214 mpa_v2_params.ird = htons((u16)ep->ird);
1215 mpa_v2_params.ord = htons((u16)ep->ord);
1216 if (peer2peer && (ep->mpa_attr.p2p_type !=
1217 FW_RI_INIT_P2PTYPE_DISABLED)) {
1218 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1219
1220 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1221 mpa_v2_params.ord |=
1222 htons(MPA_V2_RDMA_WRITE_RTR);
1223 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1224 mpa_v2_params.ord |=
1225 htons(MPA_V2_RDMA_READ_RTR);
1226 }
1227
1228 memcpy(mpa->private_data, &mpa_v2_params,
1229 sizeof(struct mpa_v2_conn_params));
1230
1231 if (ep->plen)
1232 memcpy(mpa->private_data +
1233 sizeof(struct mpa_v2_conn_params), pdata, plen);
1234 } else
1235 if (plen)
1236 memcpy(mpa->private_data, pdata, plen);
1237
1238 /*
1239 * Reference the mpa skb. This ensures the data area
1240 * will remain in memory until the hw acks the tx.
1241 * Function fw4_ack() will deref it.
1242 */
1243 skb_get(skb);
1244 t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1245 ep->mpa_skb = skb;
1246 __state_set(&ep->com, MPA_REP_SENT);
1247 ep->snd_seq += mpalen;
1248 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1249 }
1250
1251 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1252 {
1253 struct c4iw_ep *ep;
1254 struct cpl_act_establish *req = cplhdr(skb);
1255 unsigned int tid = GET_TID(req);
1256 unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1257 struct tid_info *t = dev->rdev.lldi.tids;
1258 int ret;
1259
1260 ep = lookup_atid(t, atid);
1261
1262 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
1263 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1264
1265 mutex_lock(&ep->com.mutex);
1266 dst_confirm(ep->dst);
1267
1268 /* setup the hwtid for this connection */
1269 ep->hwtid = tid;
1270 cxgb4_insert_tid(t, ep, tid);
1271 insert_ep_tid(ep);
1272
1273 ep->snd_seq = be32_to_cpu(req->snd_isn);
1274 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1275
1276 set_emss(ep, ntohs(req->tcp_opt));
1277
1278 /* dealloc the atid */
1279 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
1280 cxgb4_free_atid(t, atid);
1281 set_bit(ACT_ESTAB, &ep->com.history);
1282
1283 /* start MPA negotiation */
1284 ret = send_flowc(ep);
1285 if (ret)
1286 goto err;
1287 if (ep->retry_with_mpa_v1)
1288 ret = send_mpa_req(ep, skb, 1);
1289 else
1290 ret = send_mpa_req(ep, skb, mpa_rev);
1291 if (ret)
1292 goto err;
1293 mutex_unlock(&ep->com.mutex);
1294 return 0;
1295 err:
1296 mutex_unlock(&ep->com.mutex);
1297 connect_reply_upcall(ep, -ENOMEM);
1298 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1299 return 0;
1300 }
1301
1302 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1303 {
1304 struct iw_cm_event event;
1305
1306 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1307 memset(&event, 0, sizeof(event));
1308 event.event = IW_CM_EVENT_CLOSE;
1309 event.status = status;
1310 if (ep->com.cm_id) {
1311 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
1312 ep, ep->com.cm_id, ep->hwtid);
1313 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1314 deref_cm_id(&ep->com);
1315 set_bit(CLOSE_UPCALL, &ep->com.history);
1316 }
1317 }
1318
1319 static void peer_close_upcall(struct c4iw_ep *ep)
1320 {
1321 struct iw_cm_event event;
1322
1323 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1324 memset(&event, 0, sizeof(event));
1325 event.event = IW_CM_EVENT_DISCONNECT;
1326 if (ep->com.cm_id) {
1327 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
1328 ep, ep->com.cm_id, ep->hwtid);
1329 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1330 set_bit(DISCONN_UPCALL, &ep->com.history);
1331 }
1332 }
1333
1334 static void peer_abort_upcall(struct c4iw_ep *ep)
1335 {
1336 struct iw_cm_event event;
1337
1338 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1339 memset(&event, 0, sizeof(event));
1340 event.event = IW_CM_EVENT_CLOSE;
1341 event.status = -ECONNRESET;
1342 if (ep->com.cm_id) {
1343 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
1344 ep->com.cm_id, ep->hwtid);
1345 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1346 deref_cm_id(&ep->com);
1347 set_bit(ABORT_UPCALL, &ep->com.history);
1348 }
1349 }
1350
1351 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1352 {
1353 struct iw_cm_event event;
1354
1355 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
1356 memset(&event, 0, sizeof(event));
1357 event.event = IW_CM_EVENT_CONNECT_REPLY;
1358 event.status = status;
1359 memcpy(&event.local_addr, &ep->com.local_addr,
1360 sizeof(ep->com.local_addr));
1361 memcpy(&event.remote_addr, &ep->com.remote_addr,
1362 sizeof(ep->com.remote_addr));
1363
1364 if ((status == 0) || (status == -ECONNREFUSED)) {
1365 if (!ep->tried_with_mpa_v1) {
1366 /* this means MPA_v2 is used */
1367 event.ord = ep->ird;
1368 event.ird = ep->ord;
1369 event.private_data_len = ep->plen -
1370 sizeof(struct mpa_v2_conn_params);
1371 event.private_data = ep->mpa_pkt +
1372 sizeof(struct mpa_message) +
1373 sizeof(struct mpa_v2_conn_params);
1374 } else {
1375 /* this means MPA_v1 is used */
1376 event.ord = cur_max_read_depth(ep->com.dev);
1377 event.ird = cur_max_read_depth(ep->com.dev);
1378 event.private_data_len = ep->plen;
1379 event.private_data = ep->mpa_pkt +
1380 sizeof(struct mpa_message);
1381 }
1382 }
1383
1384 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
1385 ep->hwtid, status);
1386 set_bit(CONN_RPL_UPCALL, &ep->com.history);
1387 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1388
1389 if (status < 0)
1390 deref_cm_id(&ep->com);
1391 }
1392
1393 static int connect_request_upcall(struct c4iw_ep *ep)
1394 {
1395 struct iw_cm_event event;
1396 int ret;
1397
1398 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1399 memset(&event, 0, sizeof(event));
1400 event.event = IW_CM_EVENT_CONNECT_REQUEST;
1401 memcpy(&event.local_addr, &ep->com.local_addr,
1402 sizeof(ep->com.local_addr));
1403 memcpy(&event.remote_addr, &ep->com.remote_addr,
1404 sizeof(ep->com.remote_addr));
1405 event.provider_data = ep;
1406 if (!ep->tried_with_mpa_v1) {
1407 /* this means MPA_v2 is used */
1408 event.ord = ep->ord;
1409 event.ird = ep->ird;
1410 event.private_data_len = ep->plen -
1411 sizeof(struct mpa_v2_conn_params);
1412 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1413 sizeof(struct mpa_v2_conn_params);
1414 } else {
1415 /* this means MPA_v1 is used. Send max supported */
1416 event.ord = cur_max_read_depth(ep->com.dev);
1417 event.ird = cur_max_read_depth(ep->com.dev);
1418 event.private_data_len = ep->plen;
1419 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1420 }
1421 c4iw_get_ep(&ep->com);
1422 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1423 &event);
1424 if (ret)
1425 c4iw_put_ep(&ep->com);
1426 set_bit(CONNREQ_UPCALL, &ep->com.history);
1427 c4iw_put_ep(&ep->parent_ep->com);
1428 return ret;
1429 }
1430
1431 static void established_upcall(struct c4iw_ep *ep)
1432 {
1433 struct iw_cm_event event;
1434
1435 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1436 memset(&event, 0, sizeof(event));
1437 event.event = IW_CM_EVENT_ESTABLISHED;
1438 event.ird = ep->ord;
1439 event.ord = ep->ird;
1440 if (ep->com.cm_id) {
1441 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1442 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1443 set_bit(ESTAB_UPCALL, &ep->com.history);
1444 }
1445 }
1446
1447 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1448 {
1449 struct cpl_rx_data_ack *req;
1450 struct sk_buff *skb;
1451 int wrlen = roundup(sizeof *req, 16);
1452
1453 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1454 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1455 if (!skb) {
1456 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1457 return 0;
1458 }
1459
1460 /*
1461 * If we couldn't specify the entire rcv window at connection setup
1462 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1463 * then add the overage in to the credits returned.
1464 */
1465 if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1466 credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1467
1468 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1469 memset(req, 0, wrlen);
1470 INIT_TP_WR(req, ep->hwtid);
1471 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1472 ep->hwtid));
1473 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK_F |
1474 RX_DACK_CHANGE_F |
1475 RX_DACK_MODE_V(dack_mode));
1476 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1477 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1478 return credits;
1479 }
1480
1481 #define RELAXED_IRD_NEGOTIATION 1
1482
1483 /*
1484 * process_mpa_reply - process streaming mode MPA reply
1485 *
1486 * Returns:
1487 *
1488 * 0 upon success indicating a connect request was delivered to the ULP
1489 * or the mpa request is incomplete but valid so far.
1490 *
1491 * 1 if a failure requires the caller to close the connection.
1492 *
1493 * 2 if a failure requires the caller to abort the connection.
1494 */
1495 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1496 {
1497 struct mpa_message *mpa;
1498 struct mpa_v2_conn_params *mpa_v2_params;
1499 u16 plen;
1500 u16 resp_ird, resp_ord;
1501 u8 rtr_mismatch = 0, insuff_ird = 0;
1502 struct c4iw_qp_attributes attrs;
1503 enum c4iw_qp_attr_mask mask;
1504 int err;
1505 int disconnect = 0;
1506
1507 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1508
1509 /*
1510 * If we get more than the supported amount of private data
1511 * then we must fail this connection.
1512 */
1513 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1514 err = -EINVAL;
1515 goto err_stop_timer;
1516 }
1517
1518 /*
1519 * copy the new data into our accumulation buffer.
1520 */
1521 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1522 skb->len);
1523 ep->mpa_pkt_len += skb->len;
1524
1525 /*
1526 * if we don't even have the mpa message, then bail.
1527 */
1528 if (ep->mpa_pkt_len < sizeof(*mpa))
1529 return 0;
1530 mpa = (struct mpa_message *) ep->mpa_pkt;
1531
1532 /* Validate MPA header. */
1533 if (mpa->revision > mpa_rev) {
1534 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1535 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1536 err = -EPROTO;
1537 goto err_stop_timer;
1538 }
1539 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1540 err = -EPROTO;
1541 goto err_stop_timer;
1542 }
1543
1544 plen = ntohs(mpa->private_data_size);
1545
1546 /*
1547 * Fail if there's too much private data.
1548 */
1549 if (plen > MPA_MAX_PRIVATE_DATA) {
1550 err = -EPROTO;
1551 goto err_stop_timer;
1552 }
1553
1554 /*
1555 * If plen does not account for pkt size
1556 */
1557 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1558 err = -EPROTO;
1559 goto err_stop_timer;
1560 }
1561
1562 ep->plen = (u8) plen;
1563
1564 /*
1565 * If we don't have all the pdata yet, then bail.
1566 * We'll continue process when more data arrives.
1567 */
1568 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1569 return 0;
1570
1571 if (mpa->flags & MPA_REJECT) {
1572 err = -ECONNREFUSED;
1573 goto err_stop_timer;
1574 }
1575
1576 /*
1577 * Stop mpa timer. If it expired, then
1578 * we ignore the MPA reply. process_timeout()
1579 * will abort the connection.
1580 */
1581 if (stop_ep_timer(ep))
1582 return 0;
1583
1584 /*
1585 * If we get here we have accumulated the entire mpa
1586 * start reply message including private data. And
1587 * the MPA header is valid.
1588 */
1589 __state_set(&ep->com, FPDU_MODE);
1590 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1591 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1592 ep->mpa_attr.version = mpa->revision;
1593 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1594
1595 if (mpa->revision == 2) {
1596 ep->mpa_attr.enhanced_rdma_conn =
1597 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1598 if (ep->mpa_attr.enhanced_rdma_conn) {
1599 mpa_v2_params = (struct mpa_v2_conn_params *)
1600 (ep->mpa_pkt + sizeof(*mpa));
1601 resp_ird = ntohs(mpa_v2_params->ird) &
1602 MPA_V2_IRD_ORD_MASK;
1603 resp_ord = ntohs(mpa_v2_params->ord) &
1604 MPA_V2_IRD_ORD_MASK;
1605 PDBG("%s responder ird %u ord %u ep ird %u ord %u\n",
1606 __func__, resp_ird, resp_ord, ep->ird, ep->ord);
1607
1608 /*
1609 * This is a double-check. Ideally, below checks are
1610 * not required since ird/ord stuff has been taken
1611 * care of in c4iw_accept_cr
1612 */
1613 if (ep->ird < resp_ord) {
1614 if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1615 ep->com.dev->rdev.lldi.max_ordird_qp)
1616 ep->ird = resp_ord;
1617 else
1618 insuff_ird = 1;
1619 } else if (ep->ird > resp_ord) {
1620 ep->ird = resp_ord;
1621 }
1622 if (ep->ord > resp_ird) {
1623 if (RELAXED_IRD_NEGOTIATION)
1624 ep->ord = resp_ird;
1625 else
1626 insuff_ird = 1;
1627 }
1628 if (insuff_ird) {
1629 err = -ENOMEM;
1630 ep->ird = resp_ord;
1631 ep->ord = resp_ird;
1632 }
1633
1634 if (ntohs(mpa_v2_params->ird) &
1635 MPA_V2_PEER2PEER_MODEL) {
1636 if (ntohs(mpa_v2_params->ord) &
1637 MPA_V2_RDMA_WRITE_RTR)
1638 ep->mpa_attr.p2p_type =
1639 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1640 else if (ntohs(mpa_v2_params->ord) &
1641 MPA_V2_RDMA_READ_RTR)
1642 ep->mpa_attr.p2p_type =
1643 FW_RI_INIT_P2PTYPE_READ_REQ;
1644 }
1645 }
1646 } else if (mpa->revision == 1)
1647 if (peer2peer)
1648 ep->mpa_attr.p2p_type = p2p_type;
1649
1650 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1651 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1652 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1653 ep->mpa_attr.recv_marker_enabled,
1654 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1655 ep->mpa_attr.p2p_type, p2p_type);
1656
1657 /*
1658 * If responder's RTR does not match with that of initiator, assign
1659 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1660 * generated when moving QP to RTS state.
1661 * A TERM message will be sent after QP has moved to RTS state
1662 */
1663 if ((ep->mpa_attr.version == 2) && peer2peer &&
1664 (ep->mpa_attr.p2p_type != p2p_type)) {
1665 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1666 rtr_mismatch = 1;
1667 }
1668
1669 attrs.mpa_attr = ep->mpa_attr;
1670 attrs.max_ird = ep->ird;
1671 attrs.max_ord = ep->ord;
1672 attrs.llp_stream_handle = ep;
1673 attrs.next_state = C4IW_QP_STATE_RTS;
1674
1675 mask = C4IW_QP_ATTR_NEXT_STATE |
1676 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1677 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1678
1679 /* bind QP and TID with INIT_WR */
1680 err = c4iw_modify_qp(ep->com.qp->rhp,
1681 ep->com.qp, mask, &attrs, 1);
1682 if (err)
1683 goto err;
1684
1685 /*
1686 * If responder's RTR requirement did not match with what initiator
1687 * supports, generate TERM message
1688 */
1689 if (rtr_mismatch) {
1690 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1691 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1692 attrs.ecode = MPA_NOMATCH_RTR;
1693 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1694 attrs.send_term = 1;
1695 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1696 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1697 err = -ENOMEM;
1698 disconnect = 1;
1699 goto out;
1700 }
1701
1702 /*
1703 * Generate TERM if initiator IRD is not sufficient for responder
1704 * provided ORD. Currently, we do the same behaviour even when
1705 * responder provided IRD is also not sufficient as regards to
1706 * initiator ORD.
1707 */
1708 if (insuff_ird) {
1709 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1710 __func__);
1711 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1712 attrs.ecode = MPA_INSUFF_IRD;
1713 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1714 attrs.send_term = 1;
1715 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1716 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1717 err = -ENOMEM;
1718 disconnect = 1;
1719 goto out;
1720 }
1721 goto out;
1722 err_stop_timer:
1723 stop_ep_timer(ep);
1724 err:
1725 disconnect = 2;
1726 out:
1727 connect_reply_upcall(ep, err);
1728 return disconnect;
1729 }
1730
1731 /*
1732 * process_mpa_request - process streaming mode MPA request
1733 *
1734 * Returns:
1735 *
1736 * 0 upon success indicating a connect request was delivered to the ULP
1737 * or the mpa request is incomplete but valid so far.
1738 *
1739 * 1 if a failure requires the caller to close the connection.
1740 *
1741 * 2 if a failure requires the caller to abort the connection.
1742 */
1743 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1744 {
1745 struct mpa_message *mpa;
1746 struct mpa_v2_conn_params *mpa_v2_params;
1747 u16 plen;
1748
1749 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1750
1751 /*
1752 * If we get more than the supported amount of private data
1753 * then we must fail this connection.
1754 */
1755 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1756 goto err_stop_timer;
1757
1758 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1759
1760 /*
1761 * Copy the new data into our accumulation buffer.
1762 */
1763 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1764 skb->len);
1765 ep->mpa_pkt_len += skb->len;
1766
1767 /*
1768 * If we don't even have the mpa message, then bail.
1769 * We'll continue process when more data arrives.
1770 */
1771 if (ep->mpa_pkt_len < sizeof(*mpa))
1772 return 0;
1773
1774 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1775 mpa = (struct mpa_message *) ep->mpa_pkt;
1776
1777 /*
1778 * Validate MPA Header.
1779 */
1780 if (mpa->revision > mpa_rev) {
1781 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1782 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1783 goto err_stop_timer;
1784 }
1785
1786 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1787 goto err_stop_timer;
1788
1789 plen = ntohs(mpa->private_data_size);
1790
1791 /*
1792 * Fail if there's too much private data.
1793 */
1794 if (plen > MPA_MAX_PRIVATE_DATA)
1795 goto err_stop_timer;
1796
1797 /*
1798 * If plen does not account for pkt size
1799 */
1800 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1801 goto err_stop_timer;
1802 ep->plen = (u8) plen;
1803
1804 /*
1805 * If we don't have all the pdata yet, then bail.
1806 */
1807 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1808 return 0;
1809
1810 /*
1811 * If we get here we have accumulated the entire mpa
1812 * start reply message including private data.
1813 */
1814 ep->mpa_attr.initiator = 0;
1815 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1816 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1817 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1818 ep->mpa_attr.version = mpa->revision;
1819 if (mpa->revision == 1)
1820 ep->tried_with_mpa_v1 = 1;
1821 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1822
1823 if (mpa->revision == 2) {
1824 ep->mpa_attr.enhanced_rdma_conn =
1825 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1826 if (ep->mpa_attr.enhanced_rdma_conn) {
1827 mpa_v2_params = (struct mpa_v2_conn_params *)
1828 (ep->mpa_pkt + sizeof(*mpa));
1829 ep->ird = ntohs(mpa_v2_params->ird) &
1830 MPA_V2_IRD_ORD_MASK;
1831 ep->ird = min_t(u32, ep->ird,
1832 cur_max_read_depth(ep->com.dev));
1833 ep->ord = ntohs(mpa_v2_params->ord) &
1834 MPA_V2_IRD_ORD_MASK;
1835 ep->ord = min_t(u32, ep->ord,
1836 cur_max_read_depth(ep->com.dev));
1837 PDBG("%s initiator ird %u ord %u\n", __func__, ep->ird,
1838 ep->ord);
1839 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1840 if (peer2peer) {
1841 if (ntohs(mpa_v2_params->ord) &
1842 MPA_V2_RDMA_WRITE_RTR)
1843 ep->mpa_attr.p2p_type =
1844 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1845 else if (ntohs(mpa_v2_params->ord) &
1846 MPA_V2_RDMA_READ_RTR)
1847 ep->mpa_attr.p2p_type =
1848 FW_RI_INIT_P2PTYPE_READ_REQ;
1849 }
1850 }
1851 } else if (mpa->revision == 1)
1852 if (peer2peer)
1853 ep->mpa_attr.p2p_type = p2p_type;
1854
1855 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1856 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1857 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1858 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1859 ep->mpa_attr.p2p_type);
1860
1861 __state_set(&ep->com, MPA_REQ_RCVD);
1862
1863 /* drive upcall */
1864 mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1865 if (ep->parent_ep->com.state != DEAD) {
1866 if (connect_request_upcall(ep))
1867 goto err_unlock_parent;
1868 } else {
1869 goto err_unlock_parent;
1870 }
1871 mutex_unlock(&ep->parent_ep->com.mutex);
1872 return 0;
1873
1874 err_unlock_parent:
1875 mutex_unlock(&ep->parent_ep->com.mutex);
1876 goto err_out;
1877 err_stop_timer:
1878 (void)stop_ep_timer(ep);
1879 err_out:
1880 return 2;
1881 }
1882
1883 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1884 {
1885 struct c4iw_ep *ep;
1886 struct cpl_rx_data *hdr = cplhdr(skb);
1887 unsigned int dlen = ntohs(hdr->len);
1888 unsigned int tid = GET_TID(hdr);
1889 __u8 status = hdr->status;
1890 int disconnect = 0;
1891
1892 ep = get_ep_from_tid(dev, tid);
1893 if (!ep)
1894 return 0;
1895 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1896 skb_pull(skb, sizeof(*hdr));
1897 skb_trim(skb, dlen);
1898 mutex_lock(&ep->com.mutex);
1899
1900 /* update RX credits */
1901 update_rx_credits(ep, dlen);
1902
1903 switch (ep->com.state) {
1904 case MPA_REQ_SENT:
1905 ep->rcv_seq += dlen;
1906 disconnect = process_mpa_reply(ep, skb);
1907 break;
1908 case MPA_REQ_WAIT:
1909 ep->rcv_seq += dlen;
1910 disconnect = process_mpa_request(ep, skb);
1911 break;
1912 case FPDU_MODE: {
1913 struct c4iw_qp_attributes attrs;
1914 BUG_ON(!ep->com.qp);
1915 if (status)
1916 pr_err("%s Unexpected streaming data." \
1917 " qpid %u ep %p state %d tid %u status %d\n",
1918 __func__, ep->com.qp->wq.sq.qid, ep,
1919 ep->com.state, ep->hwtid, status);
1920 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1921 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1922 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1923 disconnect = 1;
1924 break;
1925 }
1926 default:
1927 break;
1928 }
1929 mutex_unlock(&ep->com.mutex);
1930 if (disconnect)
1931 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1932 c4iw_put_ep(&ep->com);
1933 return 0;
1934 }
1935
1936 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1937 {
1938 struct c4iw_ep *ep;
1939 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1940 int release = 0;
1941 unsigned int tid = GET_TID(rpl);
1942
1943 ep = get_ep_from_tid(dev, tid);
1944 if (!ep) {
1945 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1946 return 0;
1947 }
1948 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1949 mutex_lock(&ep->com.mutex);
1950 switch (ep->com.state) {
1951 case ABORTING:
1952 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1953 __state_set(&ep->com, DEAD);
1954 release = 1;
1955 break;
1956 default:
1957 printk(KERN_ERR "%s ep %p state %d\n",
1958 __func__, ep, ep->com.state);
1959 break;
1960 }
1961 mutex_unlock(&ep->com.mutex);
1962
1963 if (release)
1964 release_ep_resources(ep);
1965 c4iw_put_ep(&ep->com);
1966 return 0;
1967 }
1968
1969 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1970 {
1971 struct sk_buff *skb;
1972 struct fw_ofld_connection_wr *req;
1973 unsigned int mtu_idx;
1974 int wscale;
1975 struct sockaddr_in *sin;
1976 int win;
1977
1978 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1979 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1980 memset(req, 0, sizeof(*req));
1981 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1982 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1983 req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1984 ep->com.dev->rdev.lldi.ports[0],
1985 ep->l2t));
1986 sin = (struct sockaddr_in *)&ep->com.local_addr;
1987 req->le.lport = sin->sin_port;
1988 req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1989 sin = (struct sockaddr_in *)&ep->com.remote_addr;
1990 req->le.pport = sin->sin_port;
1991 req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1992 req->tcb.t_state_to_astid =
1993 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1994 FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1995 req->tcb.cplrxdataack_cplpassacceptrpl =
1996 htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1997 req->tcb.tx_max = (__force __be32) jiffies;
1998 req->tcb.rcv_adv = htons(1);
1999 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2000 enable_tcp_timestamps,
2001 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
2002 wscale = compute_wscale(rcv_win);
2003
2004 /*
2005 * Specify the largest window that will fit in opt0. The
2006 * remainder will be specified in the rx_data_ack.
2007 */
2008 win = ep->rcv_win >> 10;
2009 if (win > RCV_BUFSIZ_M)
2010 win = RCV_BUFSIZ_M;
2011
2012 req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
2013 (nocong ? NO_CONG_F : 0) |
2014 KEEP_ALIVE_F |
2015 DELACK_F |
2016 WND_SCALE_V(wscale) |
2017 MSS_IDX_V(mtu_idx) |
2018 L2T_IDX_V(ep->l2t->idx) |
2019 TX_CHAN_V(ep->tx_chan) |
2020 SMAC_SEL_V(ep->smac_idx) |
2021 DSCP_V(ep->tos >> 2) |
2022 ULP_MODE_V(ULP_MODE_TCPDDP) |
2023 RCV_BUFSIZ_V(win));
2024 req->tcb.opt2 = (__force __be32) (PACE_V(1) |
2025 TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
2026 RX_CHANNEL_V(0) |
2027 CCTRL_ECN_V(enable_ecn) |
2028 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
2029 if (enable_tcp_timestamps)
2030 req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
2031 if (enable_tcp_sack)
2032 req->tcb.opt2 |= (__force __be32)SACK_EN_F;
2033 if (wscale && enable_tcp_window_scaling)
2034 req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
2035 req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2036 req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
2037 set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2038 set_bit(ACT_OFLD_CONN, &ep->com.history);
2039 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2040 }
2041
2042 /*
2043 * Some of the error codes above implicitly indicate that there is no TID
2044 * allocated with the result of an ACT_OPEN. We use this predicate to make
2045 * that explicit.
2046 */
2047 static inline int act_open_has_tid(int status)
2048 {
2049 return (status != CPL_ERR_TCAM_PARITY &&
2050 status != CPL_ERR_TCAM_MISS &&
2051 status != CPL_ERR_TCAM_FULL &&
2052 status != CPL_ERR_CONN_EXIST_SYNRECV &&
2053 status != CPL_ERR_CONN_EXIST);
2054 }
2055
2056 /* Returns whether a CPL status conveys negative advice.
2057 */
2058 static int is_neg_adv(unsigned int status)
2059 {
2060 return status == CPL_ERR_RTX_NEG_ADVICE ||
2061 status == CPL_ERR_PERSIST_NEG_ADVICE ||
2062 status == CPL_ERR_KEEPALV_NEG_ADVICE;
2063 }
2064
2065 static char *neg_adv_str(unsigned int status)
2066 {
2067 switch (status) {
2068 case CPL_ERR_RTX_NEG_ADVICE:
2069 return "Retransmit timeout";
2070 case CPL_ERR_PERSIST_NEG_ADVICE:
2071 return "Persist timeout";
2072 case CPL_ERR_KEEPALV_NEG_ADVICE:
2073 return "Keepalive timeout";
2074 default:
2075 return "Unknown";
2076 }
2077 }
2078
2079 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2080 {
2081 ep->snd_win = snd_win;
2082 ep->rcv_win = rcv_win;
2083 PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
2084 }
2085
2086 #define ACT_OPEN_RETRY_COUNT 2
2087
2088 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2089 struct dst_entry *dst, struct c4iw_dev *cdev,
2090 bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2091 {
2092 struct neighbour *n;
2093 int err, step;
2094 struct net_device *pdev;
2095
2096 n = dst_neigh_lookup(dst, peer_ip);
2097 if (!n)
2098 return -ENODEV;
2099
2100 rcu_read_lock();
2101 err = -ENOMEM;
2102 if (n->dev->flags & IFF_LOOPBACK) {
2103 if (iptype == 4)
2104 pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2105 else if (IS_ENABLED(CONFIG_IPV6))
2106 for_each_netdev(&init_net, pdev) {
2107 if (ipv6_chk_addr(&init_net,
2108 (struct in6_addr *)peer_ip,
2109 pdev, 1))
2110 break;
2111 }
2112 else
2113 pdev = NULL;
2114
2115 if (!pdev) {
2116 err = -ENODEV;
2117 goto out;
2118 }
2119 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2120 n, pdev, rt_tos2priority(tos));
2121 if (!ep->l2t)
2122 goto out;
2123 ep->mtu = pdev->mtu;
2124 ep->tx_chan = cxgb4_port_chan(pdev);
2125 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2126 cxgb4_port_viid(pdev));
2127 step = cdev->rdev.lldi.ntxq /
2128 cdev->rdev.lldi.nchan;
2129 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2130 step = cdev->rdev.lldi.nrxq /
2131 cdev->rdev.lldi.nchan;
2132 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2133 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2134 cxgb4_port_idx(pdev) * step];
2135 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2136 dev_put(pdev);
2137 } else {
2138 pdev = get_real_dev(n->dev);
2139 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2140 n, pdev, 0);
2141 if (!ep->l2t)
2142 goto out;
2143 ep->mtu = dst_mtu(dst);
2144 ep->tx_chan = cxgb4_port_chan(pdev);
2145 ep->smac_idx = cxgb4_tp_smt_idx(adapter_type,
2146 cxgb4_port_viid(pdev));
2147 step = cdev->rdev.lldi.ntxq /
2148 cdev->rdev.lldi.nchan;
2149 ep->txq_idx = cxgb4_port_idx(pdev) * step;
2150 ep->ctrlq_idx = cxgb4_port_idx(pdev);
2151 step = cdev->rdev.lldi.nrxq /
2152 cdev->rdev.lldi.nchan;
2153 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2154 cxgb4_port_idx(pdev) * step];
2155 set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2156
2157 if (clear_mpa_v1) {
2158 ep->retry_with_mpa_v1 = 0;
2159 ep->tried_with_mpa_v1 = 0;
2160 }
2161 }
2162 err = 0;
2163 out:
2164 rcu_read_unlock();
2165
2166 neigh_release(n);
2167
2168 return err;
2169 }
2170
2171 static int c4iw_reconnect(struct c4iw_ep *ep)
2172 {
2173 int err = 0;
2174 int size = 0;
2175 struct sockaddr_in *laddr = (struct sockaddr_in *)
2176 &ep->com.cm_id->m_local_addr;
2177 struct sockaddr_in *raddr = (struct sockaddr_in *)
2178 &ep->com.cm_id->m_remote_addr;
2179 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2180 &ep->com.cm_id->m_local_addr;
2181 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2182 &ep->com.cm_id->m_remote_addr;
2183 int iptype;
2184 __u8 *ra;
2185
2186 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
2187 init_timer(&ep->timer);
2188 c4iw_init_wr_wait(&ep->com.wr_wait);
2189
2190 /* When MPA revision is different on nodes, the node with MPA_rev=2
2191 * tries to reconnect with MPA_rev 1 for the same EP through
2192 * c4iw_reconnect(), where the same EP is assigned with new tid for
2193 * further connection establishment. As we are using the same EP pointer
2194 * for reconnect, few skbs are used during the previous c4iw_connect(),
2195 * which leaves the EP with inadequate skbs for further
2196 * c4iw_reconnect(), Further causing an assert BUG_ON() due to empty
2197 * skb_list() during peer_abort(). Allocate skbs which is already used.
2198 */
2199 size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2200 if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2201 err = -ENOMEM;
2202 goto fail1;
2203 }
2204
2205 /*
2206 * Allocate an active TID to initiate a TCP connection.
2207 */
2208 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2209 if (ep->atid == -1) {
2210 pr_err("%s - cannot alloc atid.\n", __func__);
2211 err = -ENOMEM;
2212 goto fail2;
2213 }
2214 insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
2215
2216 /* find a route */
2217 if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2218 ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
2219 raddr->sin_addr.s_addr, laddr->sin_port,
2220 raddr->sin_port, ep->com.cm_id->tos);
2221 iptype = 4;
2222 ra = (__u8 *)&raddr->sin_addr;
2223 } else {
2224 ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
2225 raddr6->sin6_addr.s6_addr,
2226 laddr6->sin6_port, raddr6->sin6_port, 0,
2227 raddr6->sin6_scope_id);
2228 iptype = 6;
2229 ra = (__u8 *)&raddr6->sin6_addr;
2230 }
2231 if (!ep->dst) {
2232 pr_err("%s - cannot find route.\n", __func__);
2233 err = -EHOSTUNREACH;
2234 goto fail3;
2235 }
2236 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2237 ep->com.dev->rdev.lldi.adapter_type,
2238 ep->com.cm_id->tos);
2239 if (err) {
2240 pr_err("%s - cannot alloc l2e.\n", __func__);
2241 goto fail4;
2242 }
2243
2244 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2245 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2246 ep->l2t->idx);
2247
2248 state_set(&ep->com, CONNECTING);
2249 ep->tos = ep->com.cm_id->tos;
2250
2251 /* send connect request to rnic */
2252 err = send_connect(ep);
2253 if (!err)
2254 goto out;
2255
2256 cxgb4_l2t_release(ep->l2t);
2257 fail4:
2258 dst_release(ep->dst);
2259 fail3:
2260 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
2261 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2262 fail2:
2263 /*
2264 * remember to send notification to upper layer.
2265 * We are in here so the upper layer is not aware that this is
2266 * re-connect attempt and so, upper layer is still waiting for
2267 * response of 1st connect request.
2268 */
2269 connect_reply_upcall(ep, -ECONNRESET);
2270 fail1:
2271 c4iw_put_ep(&ep->com);
2272 out:
2273 return err;
2274 }
2275
2276 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2277 {
2278 struct c4iw_ep *ep;
2279 struct cpl_act_open_rpl *rpl = cplhdr(skb);
2280 unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2281 ntohl(rpl->atid_status)));
2282 struct tid_info *t = dev->rdev.lldi.tids;
2283 int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2284 struct sockaddr_in *la;
2285 struct sockaddr_in *ra;
2286 struct sockaddr_in6 *la6;
2287 struct sockaddr_in6 *ra6;
2288 int ret = 0;
2289
2290 ep = lookup_atid(t, atid);
2291 la = (struct sockaddr_in *)&ep->com.local_addr;
2292 ra = (struct sockaddr_in *)&ep->com.remote_addr;
2293 la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2294 ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2295
2296 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
2297 status, status2errno(status));
2298
2299 if (is_neg_adv(status)) {
2300 PDBG("%s Connection problems for atid %u status %u (%s)\n",
2301 __func__, atid, status, neg_adv_str(status));
2302 ep->stats.connect_neg_adv++;
2303 mutex_lock(&dev->rdev.stats.lock);
2304 dev->rdev.stats.neg_adv++;
2305 mutex_unlock(&dev->rdev.stats.lock);
2306 return 0;
2307 }
2308
2309 set_bit(ACT_OPEN_RPL, &ep->com.history);
2310
2311 /*
2312 * Log interesting failures.
2313 */
2314 switch (status) {
2315 case CPL_ERR_CONN_RESET:
2316 case CPL_ERR_CONN_TIMEDOUT:
2317 break;
2318 case CPL_ERR_TCAM_FULL:
2319 mutex_lock(&dev->rdev.stats.lock);
2320 dev->rdev.stats.tcam_full++;
2321 mutex_unlock(&dev->rdev.stats.lock);
2322 if (ep->com.local_addr.ss_family == AF_INET &&
2323 dev->rdev.lldi.enable_fw_ofld_conn) {
2324 ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2325 ntohl(rpl->atid_status))));
2326 if (ret)
2327 goto fail;
2328 return 0;
2329 }
2330 break;
2331 case CPL_ERR_CONN_EXIST:
2332 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2333 set_bit(ACT_RETRY_INUSE, &ep->com.history);
2334 if (ep->com.remote_addr.ss_family == AF_INET6) {
2335 struct sockaddr_in6 *sin6 =
2336 (struct sockaddr_in6 *)
2337 &ep->com.local_addr;
2338 cxgb4_clip_release(
2339 ep->com.dev->rdev.lldi.ports[0],
2340 (const u32 *)
2341 &sin6->sin6_addr.s6_addr, 1);
2342 }
2343 remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
2344 atid);
2345 cxgb4_free_atid(t, atid);
2346 dst_release(ep->dst);
2347 cxgb4_l2t_release(ep->l2t);
2348 c4iw_reconnect(ep);
2349 return 0;
2350 }
2351 break;
2352 default:
2353 if (ep->com.local_addr.ss_family == AF_INET) {
2354 pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2355 atid, status, status2errno(status),
2356 &la->sin_addr.s_addr, ntohs(la->sin_port),
2357 &ra->sin_addr.s_addr, ntohs(ra->sin_port));
2358 } else {
2359 pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2360 atid, status, status2errno(status),
2361 la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2362 ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2363 }
2364 break;
2365 }
2366
2367 fail:
2368 connect_reply_upcall(ep, status2errno(status));
2369 state_set(&ep->com, DEAD);
2370
2371 if (ep->com.remote_addr.ss_family == AF_INET6) {
2372 struct sockaddr_in6 *sin6 =
2373 (struct sockaddr_in6 *)&ep->com.local_addr;
2374 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2375 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2376 }
2377 if (status && act_open_has_tid(status))
2378 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
2379
2380 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
2381 cxgb4_free_atid(t, atid);
2382 dst_release(ep->dst);
2383 cxgb4_l2t_release(ep->l2t);
2384 c4iw_put_ep(&ep->com);
2385
2386 return 0;
2387 }
2388
2389 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2390 {
2391 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2392 unsigned int stid = GET_TID(rpl);
2393 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2394
2395 if (!ep) {
2396 PDBG("%s stid %d lookup failure!\n", __func__, stid);
2397 goto out;
2398 }
2399 PDBG("%s ep %p status %d error %d\n", __func__, ep,
2400 rpl->status, status2errno(rpl->status));
2401 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2402 c4iw_put_ep(&ep->com);
2403 out:
2404 return 0;
2405 }
2406
2407 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2408 {
2409 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2410 unsigned int stid = GET_TID(rpl);
2411 struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2412
2413 PDBG("%s ep %p\n", __func__, ep);
2414 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
2415 c4iw_put_ep(&ep->com);
2416 return 0;
2417 }
2418
2419 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2420 struct cpl_pass_accept_req *req)
2421 {
2422 struct cpl_pass_accept_rpl *rpl;
2423 unsigned int mtu_idx;
2424 u64 opt0;
2425 u32 opt2;
2426 int wscale;
2427 struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2428 int win;
2429 enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2430
2431 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2432 BUG_ON(skb_cloned(skb));
2433
2434 skb_get(skb);
2435 rpl = cplhdr(skb);
2436 if (!is_t4(adapter_type)) {
2437 skb_trim(skb, roundup(sizeof(*rpl5), 16));
2438 rpl5 = (void *)rpl;
2439 INIT_TP_WR(rpl5, ep->hwtid);
2440 } else {
2441 skb_trim(skb, sizeof(*rpl));
2442 INIT_TP_WR(rpl, ep->hwtid);
2443 }
2444 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2445 ep->hwtid));
2446
2447 best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2448 enable_tcp_timestamps && req->tcpopt.tstamp,
2449 (AF_INET == ep->com.remote_addr.ss_family) ? 0 : 1);
2450 wscale = compute_wscale(rcv_win);
2451
2452 /*
2453 * Specify the largest window that will fit in opt0. The
2454 * remainder will be specified in the rx_data_ack.
2455 */
2456 win = ep->rcv_win >> 10;
2457 if (win > RCV_BUFSIZ_M)
2458 win = RCV_BUFSIZ_M;
2459 opt0 = (nocong ? NO_CONG_F : 0) |
2460 KEEP_ALIVE_F |
2461 DELACK_F |
2462 WND_SCALE_V(wscale) |
2463 MSS_IDX_V(mtu_idx) |
2464 L2T_IDX_V(ep->l2t->idx) |
2465 TX_CHAN_V(ep->tx_chan) |
2466 SMAC_SEL_V(ep->smac_idx) |
2467 DSCP_V(ep->tos >> 2) |
2468 ULP_MODE_V(ULP_MODE_TCPDDP) |
2469 RCV_BUFSIZ_V(win);
2470 opt2 = RX_CHANNEL_V(0) |
2471 RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2472
2473 if (enable_tcp_timestamps && req->tcpopt.tstamp)
2474 opt2 |= TSTAMPS_EN_F;
2475 if (enable_tcp_sack && req->tcpopt.sack)
2476 opt2 |= SACK_EN_F;
2477 if (wscale && enable_tcp_window_scaling)
2478 opt2 |= WND_SCALE_EN_F;
2479 if (enable_ecn) {
2480 const struct tcphdr *tcph;
2481 u32 hlen = ntohl(req->hdr_len);
2482
2483 if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2484 tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2485 IP_HDR_LEN_G(hlen);
2486 else
2487 tcph = (const void *)(req + 1) +
2488 T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2489 if (tcph->ece && tcph->cwr)
2490 opt2 |= CCTRL_ECN_V(1);
2491 }
2492 if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2493 u32 isn = (prandom_u32() & ~7UL) - 1;
2494 opt2 |= T5_OPT_2_VALID_F;
2495 opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2496 opt2 |= T5_ISS_F;
2497 rpl5 = (void *)rpl;
2498 memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2499 if (peer2peer)
2500 isn += 4;
2501 rpl5->iss = cpu_to_be32(isn);
2502 PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
2503 }
2504
2505 rpl->opt0 = cpu_to_be64(opt0);
2506 rpl->opt2 = cpu_to_be32(opt2);
2507 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2508 t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2509
2510 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2511 }
2512
2513 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2514 {
2515 PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
2516 BUG_ON(skb_cloned(skb));
2517 skb_trim(skb, sizeof(struct cpl_tid_release));
2518 release_tid(&dev->rdev, hwtid, skb);
2519 return;
2520 }
2521
2522 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2523 {
2524 struct c4iw_ep *child_ep = NULL, *parent_ep;
2525 struct cpl_pass_accept_req *req = cplhdr(skb);
2526 unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2527 struct tid_info *t = dev->rdev.lldi.tids;
2528 unsigned int hwtid = GET_TID(req);
2529 struct dst_entry *dst;
2530 __u8 local_ip[16], peer_ip[16];
2531 __be16 local_port, peer_port;
2532 struct sockaddr_in6 *sin6;
2533 int err;
2534 u16 peer_mss = ntohs(req->tcpopt.mss);
2535 int iptype;
2536 unsigned short hdrs;
2537 u8 tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2538
2539 parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2540 if (!parent_ep) {
2541 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2542 goto reject;
2543 }
2544
2545 if (state_read(&parent_ep->com) != LISTEN) {
2546 PDBG("%s - listening ep not in LISTEN\n", __func__);
2547 goto reject;
2548 }
2549
2550 cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2551 &iptype, local_ip, peer_ip, &local_port, &peer_port);
2552
2553 /* Find output route */
2554 if (iptype == 4) {
2555 PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2556 , __func__, parent_ep, hwtid,
2557 local_ip, peer_ip, ntohs(local_port),
2558 ntohs(peer_port), peer_mss);
2559 dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
2560 local_port, peer_port,
2561 tos);
2562 } else {
2563 PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2564 , __func__, parent_ep, hwtid,
2565 local_ip, peer_ip, ntohs(local_port),
2566 ntohs(peer_port), peer_mss);
2567 dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
2568 PASS_OPEN_TOS_G(ntohl(req->tos_stid)),
2569 ((struct sockaddr_in6 *)
2570 &parent_ep->com.local_addr)->sin6_scope_id);
2571 }
2572 if (!dst) {
2573 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
2574 __func__);
2575 goto reject;
2576 }
2577
2578 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2579 if (!child_ep) {
2580 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
2581 __func__);
2582 dst_release(dst);
2583 goto reject;
2584 }
2585
2586 err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2587 parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2588 if (err) {
2589 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
2590 __func__);
2591 dst_release(dst);
2592 kfree(child_ep);
2593 goto reject;
2594 }
2595
2596 hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
2597 ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2598 if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2599 child_ep->mtu = peer_mss + hdrs;
2600
2601 skb_queue_head_init(&child_ep->com.ep_skb_list);
2602 if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2603 goto fail;
2604
2605 state_set(&child_ep->com, CONNECTING);
2606 child_ep->com.dev = dev;
2607 child_ep->com.cm_id = NULL;
2608
2609 if (iptype == 4) {
2610 struct sockaddr_in *sin = (struct sockaddr_in *)
2611 &child_ep->com.local_addr;
2612
2613 sin->sin_family = PF_INET;
2614 sin->sin_port = local_port;
2615 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2616
2617 sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2618 sin->sin_family = PF_INET;
2619 sin->sin_port = ((struct sockaddr_in *)
2620 &parent_ep->com.local_addr)->sin_port;
2621 sin->sin_addr.s_addr = *(__be32 *)local_ip;
2622
2623 sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2624 sin->sin_family = PF_INET;
2625 sin->sin_port = peer_port;
2626 sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2627 } else {
2628 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2629 sin6->sin6_family = PF_INET6;
2630 sin6->sin6_port = local_port;
2631 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2632
2633 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2634 sin6->sin6_family = PF_INET6;
2635 sin6->sin6_port = ((struct sockaddr_in6 *)
2636 &parent_ep->com.local_addr)->sin6_port;
2637 memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2638
2639 sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2640 sin6->sin6_family = PF_INET6;
2641 sin6->sin6_port = peer_port;
2642 memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2643 }
2644
2645 c4iw_get_ep(&parent_ep->com);
2646 child_ep->parent_ep = parent_ep;
2647 child_ep->tos = tos;
2648 child_ep->dst = dst;
2649 child_ep->hwtid = hwtid;
2650
2651 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
2652 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2653
2654 init_timer(&child_ep->timer);
2655 cxgb4_insert_tid(t, child_ep, hwtid);
2656 insert_ep_tid(child_ep);
2657 if (accept_cr(child_ep, skb, req)) {
2658 c4iw_put_ep(&parent_ep->com);
2659 release_ep_resources(child_ep);
2660 } else {
2661 set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2662 }
2663 if (iptype == 6) {
2664 sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2665 cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2666 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2667 }
2668 goto out;
2669 fail:
2670 c4iw_put_ep(&child_ep->com);
2671 reject:
2672 reject_cr(dev, hwtid, skb);
2673 if (parent_ep)
2674 c4iw_put_ep(&parent_ep->com);
2675 out:
2676 return 0;
2677 }
2678
2679 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2680 {
2681 struct c4iw_ep *ep;
2682 struct cpl_pass_establish *req = cplhdr(skb);
2683 unsigned int tid = GET_TID(req);
2684 int ret;
2685
2686 ep = get_ep_from_tid(dev, tid);
2687 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2688 ep->snd_seq = be32_to_cpu(req->snd_isn);
2689 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2690
2691 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
2692 ntohs(req->tcp_opt));
2693
2694 set_emss(ep, ntohs(req->tcp_opt));
2695
2696 dst_confirm(ep->dst);
2697 mutex_lock(&ep->com.mutex);
2698 ep->com.state = MPA_REQ_WAIT;
2699 start_ep_timer(ep);
2700 set_bit(PASS_ESTAB, &ep->com.history);
2701 ret = send_flowc(ep);
2702 mutex_unlock(&ep->com.mutex);
2703 if (ret)
2704 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2705 c4iw_put_ep(&ep->com);
2706
2707 return 0;
2708 }
2709
2710 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2711 {
2712 struct cpl_peer_close *hdr = cplhdr(skb);
2713 struct c4iw_ep *ep;
2714 struct c4iw_qp_attributes attrs;
2715 int disconnect = 1;
2716 int release = 0;
2717 unsigned int tid = GET_TID(hdr);
2718 int ret;
2719
2720 ep = get_ep_from_tid(dev, tid);
2721 if (!ep)
2722 return 0;
2723
2724 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2725 dst_confirm(ep->dst);
2726
2727 set_bit(PEER_CLOSE, &ep->com.history);
2728 mutex_lock(&ep->com.mutex);
2729 switch (ep->com.state) {
2730 case MPA_REQ_WAIT:
2731 __state_set(&ep->com, CLOSING);
2732 break;
2733 case MPA_REQ_SENT:
2734 __state_set(&ep->com, CLOSING);
2735 connect_reply_upcall(ep, -ECONNRESET);
2736 break;
2737 case MPA_REQ_RCVD:
2738
2739 /*
2740 * We're gonna mark this puppy DEAD, but keep
2741 * the reference on it until the ULP accepts or
2742 * rejects the CR. Also wake up anyone waiting
2743 * in rdma connection migration (see c4iw_accept_cr()).
2744 */
2745 __state_set(&ep->com, CLOSING);
2746 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2747 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2748 break;
2749 case MPA_REP_SENT:
2750 __state_set(&ep->com, CLOSING);
2751 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
2752 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2753 break;
2754 case FPDU_MODE:
2755 start_ep_timer(ep);
2756 __state_set(&ep->com, CLOSING);
2757 attrs.next_state = C4IW_QP_STATE_CLOSING;
2758 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2759 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2760 if (ret != -ECONNRESET) {
2761 peer_close_upcall(ep);
2762 disconnect = 1;
2763 }
2764 break;
2765 case ABORTING:
2766 disconnect = 0;
2767 break;
2768 case CLOSING:
2769 __state_set(&ep->com, MORIBUND);
2770 disconnect = 0;
2771 break;
2772 case MORIBUND:
2773 (void)stop_ep_timer(ep);
2774 if (ep->com.cm_id && ep->com.qp) {
2775 attrs.next_state = C4IW_QP_STATE_IDLE;
2776 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2777 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2778 }
2779 close_complete_upcall(ep, 0);
2780 __state_set(&ep->com, DEAD);
2781 release = 1;
2782 disconnect = 0;
2783 break;
2784 case DEAD:
2785 disconnect = 0;
2786 break;
2787 default:
2788 BUG_ON(1);
2789 }
2790 mutex_unlock(&ep->com.mutex);
2791 if (disconnect)
2792 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2793 if (release)
2794 release_ep_resources(ep);
2795 c4iw_put_ep(&ep->com);
2796 return 0;
2797 }
2798
2799 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2800 {
2801 struct cpl_abort_req_rss *req = cplhdr(skb);
2802 struct c4iw_ep *ep;
2803 struct cpl_abort_rpl *rpl;
2804 struct sk_buff *rpl_skb;
2805 struct c4iw_qp_attributes attrs;
2806 int ret;
2807 int release = 0;
2808 unsigned int tid = GET_TID(req);
2809
2810 ep = get_ep_from_tid(dev, tid);
2811 if (!ep)
2812 return 0;
2813
2814 if (is_neg_adv(req->status)) {
2815 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
2816 __func__, ep->hwtid, req->status,
2817 neg_adv_str(req->status));
2818 ep->stats.abort_neg_adv++;
2819 mutex_lock(&dev->rdev.stats.lock);
2820 dev->rdev.stats.neg_adv++;
2821 mutex_unlock(&dev->rdev.stats.lock);
2822 goto deref_ep;
2823 }
2824 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2825 ep->com.state);
2826 set_bit(PEER_ABORT, &ep->com.history);
2827
2828 /*
2829 * Wake up any threads in rdma_init() or rdma_fini().
2830 * However, this is not needed if com state is just
2831 * MPA_REQ_SENT
2832 */
2833 if (ep->com.state != MPA_REQ_SENT)
2834 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2835
2836 mutex_lock(&ep->com.mutex);
2837 switch (ep->com.state) {
2838 case CONNECTING:
2839 c4iw_put_ep(&ep->parent_ep->com);
2840 break;
2841 case MPA_REQ_WAIT:
2842 (void)stop_ep_timer(ep);
2843 break;
2844 case MPA_REQ_SENT:
2845 (void)stop_ep_timer(ep);
2846 if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
2847 connect_reply_upcall(ep, -ECONNRESET);
2848 else {
2849 /*
2850 * we just don't send notification upwards because we
2851 * want to retry with mpa_v1 without upper layers even
2852 * knowing it.
2853 *
2854 * do some housekeeping so as to re-initiate the
2855 * connection
2856 */
2857 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2858 mpa_rev);
2859 ep->retry_with_mpa_v1 = 1;
2860 }
2861 break;
2862 case MPA_REP_SENT:
2863 break;
2864 case MPA_REQ_RCVD:
2865 break;
2866 case MORIBUND:
2867 case CLOSING:
2868 stop_ep_timer(ep);
2869 /*FALLTHROUGH*/
2870 case FPDU_MODE:
2871 if (ep->com.cm_id && ep->com.qp) {
2872 attrs.next_state = C4IW_QP_STATE_ERROR;
2873 ret = c4iw_modify_qp(ep->com.qp->rhp,
2874 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2875 &attrs, 1);
2876 if (ret)
2877 printk(KERN_ERR MOD
2878 "%s - qp <- error failed!\n",
2879 __func__);
2880 }
2881 peer_abort_upcall(ep);
2882 break;
2883 case ABORTING:
2884 break;
2885 case DEAD:
2886 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2887 mutex_unlock(&ep->com.mutex);
2888 goto deref_ep;
2889 default:
2890 BUG_ON(1);
2891 break;
2892 }
2893 dst_confirm(ep->dst);
2894 if (ep->com.state != ABORTING) {
2895 __state_set(&ep->com, DEAD);
2896 /* we don't release if we want to retry with mpa_v1 */
2897 if (!ep->retry_with_mpa_v1)
2898 release = 1;
2899 }
2900 mutex_unlock(&ep->com.mutex);
2901
2902 rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2903 if (WARN_ON(!rpl_skb)) {
2904 release = 1;
2905 goto out;
2906 }
2907 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2908 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2909 INIT_TP_WR(rpl, ep->hwtid);
2910 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2911 rpl->cmd = CPL_ABORT_NO_RST;
2912 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2913 out:
2914 if (release)
2915 release_ep_resources(ep);
2916 else if (ep->retry_with_mpa_v1) {
2917 if (ep->com.remote_addr.ss_family == AF_INET6) {
2918 struct sockaddr_in6 *sin6 =
2919 (struct sockaddr_in6 *)
2920 &ep->com.local_addr;
2921 cxgb4_clip_release(
2922 ep->com.dev->rdev.lldi.ports[0],
2923 (const u32 *)&sin6->sin6_addr.s6_addr,
2924 1);
2925 }
2926 remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
2927 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2928 dst_release(ep->dst);
2929 cxgb4_l2t_release(ep->l2t);
2930 c4iw_reconnect(ep);
2931 }
2932
2933 deref_ep:
2934 c4iw_put_ep(&ep->com);
2935 /* Dereferencing ep, referenced in peer_abort_intr() */
2936 c4iw_put_ep(&ep->com);
2937 return 0;
2938 }
2939
2940 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2941 {
2942 struct c4iw_ep *ep;
2943 struct c4iw_qp_attributes attrs;
2944 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2945 int release = 0;
2946 unsigned int tid = GET_TID(rpl);
2947
2948 ep = get_ep_from_tid(dev, tid);
2949 if (!ep)
2950 return 0;
2951
2952 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2953 BUG_ON(!ep);
2954
2955 /* The cm_id may be null if we failed to connect */
2956 mutex_lock(&ep->com.mutex);
2957 set_bit(CLOSE_CON_RPL, &ep->com.history);
2958 switch (ep->com.state) {
2959 case CLOSING:
2960 __state_set(&ep->com, MORIBUND);
2961 break;
2962 case MORIBUND:
2963 (void)stop_ep_timer(ep);
2964 if ((ep->com.cm_id) && (ep->com.qp)) {
2965 attrs.next_state = C4IW_QP_STATE_IDLE;
2966 c4iw_modify_qp(ep->com.qp->rhp,
2967 ep->com.qp,
2968 C4IW_QP_ATTR_NEXT_STATE,
2969 &attrs, 1);
2970 }
2971 close_complete_upcall(ep, 0);
2972 __state_set(&ep->com, DEAD);
2973 release = 1;
2974 break;
2975 case ABORTING:
2976 case DEAD:
2977 break;
2978 default:
2979 BUG_ON(1);
2980 break;
2981 }
2982 mutex_unlock(&ep->com.mutex);
2983 if (release)
2984 release_ep_resources(ep);
2985 c4iw_put_ep(&ep->com);
2986 return 0;
2987 }
2988
2989 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2990 {
2991 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2992 unsigned int tid = GET_TID(rpl);
2993 struct c4iw_ep *ep;
2994 struct c4iw_qp_attributes attrs;
2995
2996 ep = get_ep_from_tid(dev, tid);
2997 BUG_ON(!ep);
2998
2999 if (ep && ep->com.qp) {
3000 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
3001 ep->com.qp->wq.sq.qid);
3002 attrs.next_state = C4IW_QP_STATE_TERMINATE;
3003 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
3004 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
3005 } else
3006 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
3007 c4iw_put_ep(&ep->com);
3008
3009 return 0;
3010 }
3011
3012 /*
3013 * Upcall from the adapter indicating data has been transmitted.
3014 * For us its just the single MPA request or reply. We can now free
3015 * the skb holding the mpa message.
3016 */
3017 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
3018 {
3019 struct c4iw_ep *ep;
3020 struct cpl_fw4_ack *hdr = cplhdr(skb);
3021 u8 credits = hdr->credits;
3022 unsigned int tid = GET_TID(hdr);
3023
3024
3025 ep = get_ep_from_tid(dev, tid);
3026 if (!ep)
3027 return 0;
3028 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
3029 if (credits == 0) {
3030 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
3031 __func__, ep, ep->hwtid, state_read(&ep->com));
3032 goto out;
3033 }
3034
3035 dst_confirm(ep->dst);
3036 if (ep->mpa_skb) {
3037 PDBG("%s last streaming msg ack ep %p tid %u state %u "
3038 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
3039 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
3040 mutex_lock(&ep->com.mutex);
3041 kfree_skb(ep->mpa_skb);
3042 ep->mpa_skb = NULL;
3043 if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3044 stop_ep_timer(ep);
3045 mutex_unlock(&ep->com.mutex);
3046 }
3047 out:
3048 c4iw_put_ep(&ep->com);
3049 return 0;
3050 }
3051
3052 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3053 {
3054 int abort;
3055 struct c4iw_ep *ep = to_ep(cm_id);
3056
3057 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3058
3059 mutex_lock(&ep->com.mutex);
3060 if (ep->com.state != MPA_REQ_RCVD) {
3061 mutex_unlock(&ep->com.mutex);
3062 c4iw_put_ep(&ep->com);
3063 return -ECONNRESET;
3064 }
3065 set_bit(ULP_REJECT, &ep->com.history);
3066 if (mpa_rev == 0)
3067 abort = 1;
3068 else
3069 abort = send_mpa_reject(ep, pdata, pdata_len);
3070 mutex_unlock(&ep->com.mutex);
3071
3072 stop_ep_timer(ep);
3073 c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3074 c4iw_put_ep(&ep->com);
3075 return 0;
3076 }
3077
3078 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3079 {
3080 int err;
3081 struct c4iw_qp_attributes attrs;
3082 enum c4iw_qp_attr_mask mask;
3083 struct c4iw_ep *ep = to_ep(cm_id);
3084 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3085 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3086 int abort = 0;
3087
3088 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
3089
3090 mutex_lock(&ep->com.mutex);
3091 if (ep->com.state != MPA_REQ_RCVD) {
3092 err = -ECONNRESET;
3093 goto err_out;
3094 }
3095
3096 BUG_ON(!qp);
3097
3098 set_bit(ULP_ACCEPT, &ep->com.history);
3099 if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3100 (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3101 err = -EINVAL;
3102 goto err_abort;
3103 }
3104
3105 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3106 if (conn_param->ord > ep->ird) {
3107 if (RELAXED_IRD_NEGOTIATION) {
3108 conn_param->ord = ep->ird;
3109 } else {
3110 ep->ird = conn_param->ird;
3111 ep->ord = conn_param->ord;
3112 send_mpa_reject(ep, conn_param->private_data,
3113 conn_param->private_data_len);
3114 err = -ENOMEM;
3115 goto err_abort;
3116 }
3117 }
3118 if (conn_param->ird < ep->ord) {
3119 if (RELAXED_IRD_NEGOTIATION &&
3120 ep->ord <= h->rdev.lldi.max_ordird_qp) {
3121 conn_param->ird = ep->ord;
3122 } else {
3123 err = -ENOMEM;
3124 goto err_abort;
3125 }
3126 }
3127 }
3128 ep->ird = conn_param->ird;
3129 ep->ord = conn_param->ord;
3130
3131 if (ep->mpa_attr.version == 1) {
3132 if (peer2peer && ep->ird == 0)
3133 ep->ird = 1;
3134 } else {
3135 if (peer2peer &&
3136 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3137 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3138 ep->ird = 1;
3139 }
3140
3141 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
3142
3143 ep->com.cm_id = cm_id;
3144 ref_cm_id(&ep->com);
3145 ep->com.qp = qp;
3146 ref_qp(ep);
3147
3148 /* bind QP to EP and move to RTS */
3149 attrs.mpa_attr = ep->mpa_attr;
3150 attrs.max_ird = ep->ird;
3151 attrs.max_ord = ep->ord;
3152 attrs.llp_stream_handle = ep;
3153 attrs.next_state = C4IW_QP_STATE_RTS;
3154
3155 /* bind QP and TID with INIT_WR */
3156 mask = C4IW_QP_ATTR_NEXT_STATE |
3157 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3158 C4IW_QP_ATTR_MPA_ATTR |
3159 C4IW_QP_ATTR_MAX_IRD |
3160 C4IW_QP_ATTR_MAX_ORD;
3161
3162 err = c4iw_modify_qp(ep->com.qp->rhp,
3163 ep->com.qp, mask, &attrs, 1);
3164 if (err)
3165 goto err_deref_cm_id;
3166
3167 set_bit(STOP_MPA_TIMER, &ep->com.flags);
3168 err = send_mpa_reply(ep, conn_param->private_data,
3169 conn_param->private_data_len);
3170 if (err)
3171 goto err_deref_cm_id;
3172
3173 __state_set(&ep->com, FPDU_MODE);
3174 established_upcall(ep);
3175 mutex_unlock(&ep->com.mutex);
3176 c4iw_put_ep(&ep->com);
3177 return 0;
3178 err_deref_cm_id:
3179 deref_cm_id(&ep->com);
3180 err_abort:
3181 abort = 1;
3182 err_out:
3183 mutex_unlock(&ep->com.mutex);
3184 if (abort)
3185 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3186 c4iw_put_ep(&ep->com);
3187 return err;
3188 }
3189
3190 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3191 {
3192 struct in_device *ind;
3193 int found = 0;
3194 struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3195 struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3196
3197 ind = in_dev_get(dev->rdev.lldi.ports[0]);
3198 if (!ind)
3199 return -EADDRNOTAVAIL;
3200 for_primary_ifa(ind) {
3201 laddr->sin_addr.s_addr = ifa->ifa_address;
3202 raddr->sin_addr.s_addr = ifa->ifa_address;
3203 found = 1;
3204 break;
3205 }
3206 endfor_ifa(ind);
3207 in_dev_put(ind);
3208 return found ? 0 : -EADDRNOTAVAIL;
3209 }
3210
3211 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3212 unsigned char banned_flags)
3213 {
3214 struct inet6_dev *idev;
3215 int err = -EADDRNOTAVAIL;
3216
3217 rcu_read_lock();
3218 idev = __in6_dev_get(dev);
3219 if (idev != NULL) {
3220 struct inet6_ifaddr *ifp;
3221
3222 read_lock_bh(&idev->lock);
3223 list_for_each_entry(ifp, &idev->addr_list, if_list) {
3224 if (ifp->scope == IFA_LINK &&
3225 !(ifp->flags & banned_flags)) {
3226 memcpy(addr, &ifp->addr, 16);
3227 err = 0;
3228 break;
3229 }
3230 }
3231 read_unlock_bh(&idev->lock);
3232 }
3233 rcu_read_unlock();
3234 return err;
3235 }
3236
3237 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3238 {
3239 struct in6_addr uninitialized_var(addr);
3240 struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3241 struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3242
3243 if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3244 memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3245 memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3246 return 0;
3247 }
3248 return -EADDRNOTAVAIL;
3249 }
3250
3251 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3252 {
3253 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3254 struct c4iw_ep *ep;
3255 int err = 0;
3256 struct sockaddr_in *laddr;
3257 struct sockaddr_in *raddr;
3258 struct sockaddr_in6 *laddr6;
3259 struct sockaddr_in6 *raddr6;
3260 __u8 *ra;
3261 int iptype;
3262
3263 if ((conn_param->ord > cur_max_read_depth(dev)) ||
3264 (conn_param->ird > cur_max_read_depth(dev))) {
3265 err = -EINVAL;
3266 goto out;
3267 }
3268 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3269 if (!ep) {
3270 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3271 err = -ENOMEM;
3272 goto out;
3273 }
3274
3275 skb_queue_head_init(&ep->com.ep_skb_list);
3276 if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3277 err = -ENOMEM;
3278 goto fail1;
3279 }
3280
3281 init_timer(&ep->timer);
3282 ep->plen = conn_param->private_data_len;
3283 if (ep->plen)
3284 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3285 conn_param->private_data, ep->plen);
3286 ep->ird = conn_param->ird;
3287 ep->ord = conn_param->ord;
3288
3289 if (peer2peer && ep->ord == 0)
3290 ep->ord = 1;
3291
3292 ep->com.cm_id = cm_id;
3293 ref_cm_id(&ep->com);
3294 ep->com.dev = dev;
3295 ep->com.qp = get_qhp(dev, conn_param->qpn);
3296 if (!ep->com.qp) {
3297 PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3298 err = -EINVAL;
3299 goto fail2;
3300 }
3301 ref_qp(ep);
3302 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
3303 ep->com.qp, cm_id);
3304
3305 /*
3306 * Allocate an active TID to initiate a TCP connection.
3307 */
3308 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3309 if (ep->atid == -1) {
3310 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
3311 err = -ENOMEM;
3312 goto fail2;
3313 }
3314 insert_handle(dev, &dev->atid_idr, ep, ep->atid);
3315
3316 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3317 sizeof(ep->com.local_addr));
3318 memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3319 sizeof(ep->com.remote_addr));
3320
3321 laddr = (struct sockaddr_in *)&ep->com.local_addr;
3322 raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3323 laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3324 raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3325
3326 if (cm_id->m_remote_addr.ss_family == AF_INET) {
3327 iptype = 4;
3328 ra = (__u8 *)&raddr->sin_addr;
3329
3330 /*
3331 * Handle loopback requests to INADDR_ANY.
3332 */
3333 if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3334 err = pick_local_ipaddrs(dev, cm_id);
3335 if (err)
3336 goto fail2;
3337 }
3338
3339 /* find a route */
3340 PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3341 __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
3342 ra, ntohs(raddr->sin_port));
3343 ep->dst = find_route(dev, laddr->sin_addr.s_addr,
3344 raddr->sin_addr.s_addr, laddr->sin_port,
3345 raddr->sin_port, cm_id->tos);
3346 } else {
3347 iptype = 6;
3348 ra = (__u8 *)&raddr6->sin6_addr;
3349
3350 /*
3351 * Handle loopback requests to INADDR_ANY.
3352 */
3353 if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3354 err = pick_local_ip6addrs(dev, cm_id);
3355 if (err)
3356 goto fail2;
3357 }
3358
3359 /* find a route */
3360 PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3361 __func__, laddr6->sin6_addr.s6_addr,
3362 ntohs(laddr6->sin6_port),
3363 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3364 ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
3365 raddr6->sin6_addr.s6_addr,
3366 laddr6->sin6_port, raddr6->sin6_port, 0,
3367 raddr6->sin6_scope_id);
3368 }
3369 if (!ep->dst) {
3370 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
3371 err = -EHOSTUNREACH;
3372 goto fail3;
3373 }
3374
3375 err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3376 ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3377 if (err) {
3378 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
3379 goto fail4;
3380 }
3381
3382 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3383 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3384 ep->l2t->idx);
3385
3386 state_set(&ep->com, CONNECTING);
3387 ep->tos = cm_id->tos;
3388
3389 /* send connect request to rnic */
3390 err = send_connect(ep);
3391 if (!err)
3392 goto out;
3393
3394 cxgb4_l2t_release(ep->l2t);
3395 fail4:
3396 dst_release(ep->dst);
3397 fail3:
3398 remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
3399 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3400 fail2:
3401 skb_queue_purge(&ep->com.ep_skb_list);
3402 deref_cm_id(&ep->com);
3403 fail1:
3404 c4iw_put_ep(&ep->com);
3405 out:
3406 return err;
3407 }
3408
3409 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3410 {
3411 int err;
3412 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3413 &ep->com.local_addr;
3414
3415 if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3416 err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3417 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3418 if (err)
3419 return err;
3420 }
3421 c4iw_init_wr_wait(&ep->com.wr_wait);
3422 err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3423 ep->stid, &sin6->sin6_addr,
3424 sin6->sin6_port,
3425 ep->com.dev->rdev.lldi.rxq_ids[0]);
3426 if (!err)
3427 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3428 &ep->com.wr_wait,
3429 0, 0, __func__);
3430 else if (err > 0)
3431 err = net_xmit_errno(err);
3432 if (err) {
3433 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3434 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3435 pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3436 err, ep->stid,
3437 sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3438 }
3439 return err;
3440 }
3441
3442 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3443 {
3444 int err;
3445 struct sockaddr_in *sin = (struct sockaddr_in *)
3446 &ep->com.local_addr;
3447
3448 if (dev->rdev.lldi.enable_fw_ofld_conn) {
3449 do {
3450 err = cxgb4_create_server_filter(
3451 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3452 sin->sin_addr.s_addr, sin->sin_port, 0,
3453 ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3454 if (err == -EBUSY) {
3455 if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3456 err = -EIO;
3457 break;
3458 }
3459 set_current_state(TASK_UNINTERRUPTIBLE);
3460 schedule_timeout(usecs_to_jiffies(100));
3461 }
3462 } while (err == -EBUSY);
3463 } else {
3464 c4iw_init_wr_wait(&ep->com.wr_wait);
3465 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3466 ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3467 0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3468 if (!err)
3469 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3470 &ep->com.wr_wait,
3471 0, 0, __func__);
3472 else if (err > 0)
3473 err = net_xmit_errno(err);
3474 }
3475 if (err)
3476 pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3477 , err, ep->stid,
3478 &sin->sin_addr, ntohs(sin->sin_port));
3479 return err;
3480 }
3481
3482 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3483 {
3484 int err = 0;
3485 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3486 struct c4iw_listen_ep *ep;
3487
3488 might_sleep();
3489
3490 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3491 if (!ep) {
3492 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
3493 err = -ENOMEM;
3494 goto fail1;
3495 }
3496 skb_queue_head_init(&ep->com.ep_skb_list);
3497 PDBG("%s ep %p\n", __func__, ep);
3498 ep->com.cm_id = cm_id;
3499 ref_cm_id(&ep->com);
3500 ep->com.dev = dev;
3501 ep->backlog = backlog;
3502 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3503 sizeof(ep->com.local_addr));
3504
3505 /*
3506 * Allocate a server TID.
3507 */
3508 if (dev->rdev.lldi.enable_fw_ofld_conn &&
3509 ep->com.local_addr.ss_family == AF_INET)
3510 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3511 cm_id->m_local_addr.ss_family, ep);
3512 else
3513 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3514 cm_id->m_local_addr.ss_family, ep);
3515
3516 if (ep->stid == -1) {
3517 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
3518 err = -ENOMEM;
3519 goto fail2;
3520 }
3521 insert_handle(dev, &dev->stid_idr, ep, ep->stid);
3522
3523 memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3524 sizeof(ep->com.local_addr));
3525
3526 state_set(&ep->com, LISTEN);
3527 if (ep->com.local_addr.ss_family == AF_INET)
3528 err = create_server4(dev, ep);
3529 else
3530 err = create_server6(dev, ep);
3531 if (!err) {
3532 cm_id->provider_data = ep;
3533 goto out;
3534 }
3535
3536 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3537 ep->com.local_addr.ss_family);
3538 fail2:
3539 deref_cm_id(&ep->com);
3540 c4iw_put_ep(&ep->com);
3541 fail1:
3542 out:
3543 return err;
3544 }
3545
3546 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3547 {
3548 int err;
3549 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3550
3551 PDBG("%s ep %p\n", __func__, ep);
3552
3553 might_sleep();
3554 state_set(&ep->com, DEAD);
3555 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3556 ep->com.local_addr.ss_family == AF_INET) {
3557 err = cxgb4_remove_server_filter(
3558 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3559 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3560 } else {
3561 struct sockaddr_in6 *sin6;
3562 c4iw_init_wr_wait(&ep->com.wr_wait);
3563 err = cxgb4_remove_server(
3564 ep->com.dev->rdev.lldi.ports[0], ep->stid,
3565 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
3566 if (err)
3567 goto done;
3568 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
3569 0, 0, __func__);
3570 sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3571 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3572 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3573 }
3574 remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
3575 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3576 ep->com.local_addr.ss_family);
3577 done:
3578 deref_cm_id(&ep->com);
3579 c4iw_put_ep(&ep->com);
3580 return err;
3581 }
3582
3583 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3584 {
3585 int ret = 0;
3586 int close = 0;
3587 int fatal = 0;
3588 struct c4iw_rdev *rdev;
3589
3590 mutex_lock(&ep->com.mutex);
3591
3592 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
3593 states[ep->com.state], abrupt);
3594
3595 /*
3596 * Ref the ep here in case we have fatal errors causing the
3597 * ep to be released and freed.
3598 */
3599 c4iw_get_ep(&ep->com);
3600
3601 rdev = &ep->com.dev->rdev;
3602 if (c4iw_fatal_error(rdev)) {
3603 fatal = 1;
3604 close_complete_upcall(ep, -EIO);
3605 ep->com.state = DEAD;
3606 }
3607 switch (ep->com.state) {
3608 case MPA_REQ_WAIT:
3609 case MPA_REQ_SENT:
3610 case MPA_REQ_RCVD:
3611 case MPA_REP_SENT:
3612 case FPDU_MODE:
3613 case CONNECTING:
3614 close = 1;
3615 if (abrupt)
3616 ep->com.state = ABORTING;
3617 else {
3618 ep->com.state = CLOSING;
3619
3620 /*
3621 * if we close before we see the fw4_ack() then we fix
3622 * up the timer state since we're reusing it.
3623 */
3624 if (ep->mpa_skb &&
3625 test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3626 clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3627 stop_ep_timer(ep);
3628 }
3629 start_ep_timer(ep);
3630 }
3631 set_bit(CLOSE_SENT, &ep->com.flags);
3632 break;
3633 case CLOSING:
3634 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3635 close = 1;
3636 if (abrupt) {
3637 (void)stop_ep_timer(ep);
3638 ep->com.state = ABORTING;
3639 } else
3640 ep->com.state = MORIBUND;
3641 }
3642 break;
3643 case MORIBUND:
3644 case ABORTING:
3645 case DEAD:
3646 PDBG("%s ignoring disconnect ep %p state %u\n",
3647 __func__, ep, ep->com.state);
3648 break;
3649 default:
3650 BUG();
3651 break;
3652 }
3653
3654 if (close) {
3655 if (abrupt) {
3656 set_bit(EP_DISC_ABORT, &ep->com.history);
3657 close_complete_upcall(ep, -ECONNRESET);
3658 ret = send_abort(ep);
3659 } else {
3660 set_bit(EP_DISC_CLOSE, &ep->com.history);
3661 ret = send_halfclose(ep);
3662 }
3663 if (ret) {
3664 set_bit(EP_DISC_FAIL, &ep->com.history);
3665 if (!abrupt) {
3666 stop_ep_timer(ep);
3667 close_complete_upcall(ep, -EIO);
3668 }
3669 if (ep->com.qp) {
3670 struct c4iw_qp_attributes attrs;
3671
3672 attrs.next_state = C4IW_QP_STATE_ERROR;
3673 ret = c4iw_modify_qp(ep->com.qp->rhp,
3674 ep->com.qp,
3675 C4IW_QP_ATTR_NEXT_STATE,
3676 &attrs, 1);
3677 if (ret)
3678 pr_err(MOD
3679 "%s - qp <- error failed!\n",
3680 __func__);
3681 }
3682 fatal = 1;
3683 }
3684 }
3685 mutex_unlock(&ep->com.mutex);
3686 c4iw_put_ep(&ep->com);
3687 if (fatal)
3688 release_ep_resources(ep);
3689 return ret;
3690 }
3691
3692 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3693 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3694 {
3695 struct c4iw_ep *ep;
3696 int atid = be32_to_cpu(req->tid);
3697
3698 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3699 (__force u32) req->tid);
3700 if (!ep)
3701 return;
3702
3703 switch (req->retval) {
3704 case FW_ENOMEM:
3705 set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3706 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3707 send_fw_act_open_req(ep, atid);
3708 return;
3709 }
3710 case FW_EADDRINUSE:
3711 set_bit(ACT_RETRY_INUSE, &ep->com.history);
3712 if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3713 send_fw_act_open_req(ep, atid);
3714 return;
3715 }
3716 break;
3717 default:
3718 pr_info("%s unexpected ofld conn wr retval %d\n",
3719 __func__, req->retval);
3720 break;
3721 }
3722 pr_err("active ofld_connect_wr failure %d atid %d\n",
3723 req->retval, atid);
3724 mutex_lock(&dev->rdev.stats.lock);
3725 dev->rdev.stats.act_ofld_conn_fails++;
3726 mutex_unlock(&dev->rdev.stats.lock);
3727 connect_reply_upcall(ep, status2errno(req->retval));
3728 state_set(&ep->com, DEAD);
3729 if (ep->com.remote_addr.ss_family == AF_INET6) {
3730 struct sockaddr_in6 *sin6 =
3731 (struct sockaddr_in6 *)&ep->com.local_addr;
3732 cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3733 (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3734 }
3735 remove_handle(dev, &dev->atid_idr, atid);
3736 cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3737 dst_release(ep->dst);
3738 cxgb4_l2t_release(ep->l2t);
3739 c4iw_put_ep(&ep->com);
3740 }
3741
3742 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3743 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3744 {
3745 struct sk_buff *rpl_skb;
3746 struct cpl_pass_accept_req *cpl;
3747 int ret;
3748
3749 rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3750 BUG_ON(!rpl_skb);
3751 if (req->retval) {
3752 PDBG("%s passive open failure %d\n", __func__, req->retval);
3753 mutex_lock(&dev->rdev.stats.lock);
3754 dev->rdev.stats.pas_ofld_conn_fails++;
3755 mutex_unlock(&dev->rdev.stats.lock);
3756 kfree_skb(rpl_skb);
3757 } else {
3758 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3759 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3760 (__force u32) htonl(
3761 (__force u32) req->tid)));
3762 ret = pass_accept_req(dev, rpl_skb);
3763 if (!ret)
3764 kfree_skb(rpl_skb);
3765 }
3766 return;
3767 }
3768
3769 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3770 {
3771 struct cpl_fw6_msg *rpl = cplhdr(skb);
3772 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3773
3774 switch (rpl->type) {
3775 case FW6_TYPE_CQE:
3776 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3777 break;
3778 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3779 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3780 switch (req->t_state) {
3781 case TCP_SYN_SENT:
3782 active_ofld_conn_reply(dev, skb, req);
3783 break;
3784 case TCP_SYN_RECV:
3785 passive_ofld_conn_reply(dev, skb, req);
3786 break;
3787 default:
3788 pr_err("%s unexpected ofld conn wr state %d\n",
3789 __func__, req->t_state);
3790 break;
3791 }
3792 break;
3793 }
3794 return 0;
3795 }
3796
3797 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3798 {
3799 __be32 l2info;
3800 __be16 hdr_len, vlantag, len;
3801 u16 eth_hdr_len;
3802 int tcp_hdr_len, ip_hdr_len;
3803 u8 intf;
3804 struct cpl_rx_pkt *cpl = cplhdr(skb);
3805 struct cpl_pass_accept_req *req;
3806 struct tcp_options_received tmp_opt;
3807 struct c4iw_dev *dev;
3808 enum chip_type type;
3809
3810 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3811 /* Store values from cpl_rx_pkt in temporary location. */
3812 vlantag = cpl->vlan;
3813 len = cpl->len;
3814 l2info = cpl->l2info;
3815 hdr_len = cpl->hdr_len;
3816 intf = cpl->iff;
3817
3818 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3819
3820 /*
3821 * We need to parse the TCP options from SYN packet.
3822 * to generate cpl_pass_accept_req.
3823 */
3824 memset(&tmp_opt, 0, sizeof(tmp_opt));
3825 tcp_clear_options(&tmp_opt);
3826 tcp_parse_options(skb, &tmp_opt, 0, NULL);
3827
3828 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
3829 memset(req, 0, sizeof(*req));
3830 req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3831 SYN_MAC_IDX_V(RX_MACIDX_G(
3832 be32_to_cpu(l2info))) |
3833 SYN_XACT_MATCH_F);
3834 type = dev->rdev.lldi.adapter_type;
3835 tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3836 ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3837 req->hdr_len =
3838 cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3839 if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3840 eth_hdr_len = is_t4(type) ?
3841 RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3842 RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3843 req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3844 IP_HDR_LEN_V(ip_hdr_len) |
3845 ETH_HDR_LEN_V(eth_hdr_len));
3846 } else { /* T6 and later */
3847 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3848 req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3849 T6_IP_HDR_LEN_V(ip_hdr_len) |
3850 T6_ETH_HDR_LEN_V(eth_hdr_len));
3851 }
3852 req->vlan = vlantag;
3853 req->len = len;
3854 req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3855 PASS_OPEN_TOS_V(tos));
3856 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3857 if (tmp_opt.wscale_ok)
3858 req->tcpopt.wsf = tmp_opt.snd_wscale;
3859 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3860 if (tmp_opt.sack_ok)
3861 req->tcpopt.sack = 1;
3862 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3863 return;
3864 }
3865
3866 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
3867 __be32 laddr, __be16 lport,
3868 __be32 raddr, __be16 rport,
3869 u32 rcv_isn, u32 filter, u16 window,
3870 u32 rss_qid, u8 port_id)
3871 {
3872 struct sk_buff *req_skb;
3873 struct fw_ofld_connection_wr *req;
3874 struct cpl_pass_accept_req *cpl = cplhdr(skb);
3875 int ret;
3876
3877 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
3878 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
3879 memset(req, 0, sizeof(*req));
3880 req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
3881 req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
3882 req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
3883 req->le.filter = (__force __be32) filter;
3884 req->le.lport = lport;
3885 req->le.pport = rport;
3886 req->le.u.ipv4.lip = laddr;
3887 req->le.u.ipv4.pip = raddr;
3888 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
3889 req->tcb.rcv_adv = htons(window);
3890 req->tcb.t_state_to_astid =
3891 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
3892 FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
3893 FW_OFLD_CONNECTION_WR_ASTID_V(
3894 PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
3895
3896 /*
3897 * We store the qid in opt2 which will be used by the firmware
3898 * to send us the wr response.
3899 */
3900 req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
3901
3902 /*
3903 * We initialize the MSS index in TCB to 0xF.
3904 * So that when driver sends cpl_pass_accept_rpl
3905 * TCB picks up the correct value. If this was 0
3906 * TP will ignore any value > 0 for MSS index.
3907 */
3908 req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
3909 req->cookie = (uintptr_t)skb;
3910
3911 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
3912 ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
3913 if (ret < 0) {
3914 pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
3915 ret);
3916 kfree_skb(skb);
3917 kfree_skb(req_skb);
3918 }
3919 }
3920
3921 /*
3922 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
3923 * messages when a filter is being used instead of server to
3924 * redirect a syn packet. When packets hit filter they are redirected
3925 * to the offload queue and driver tries to establish the connection
3926 * using firmware work request.
3927 */
3928 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
3929 {
3930 int stid;
3931 unsigned int filter;
3932 struct ethhdr *eh = NULL;
3933 struct vlan_ethhdr *vlan_eh = NULL;
3934 struct iphdr *iph;
3935 struct tcphdr *tcph;
3936 struct rss_header *rss = (void *)skb->data;
3937 struct cpl_rx_pkt *cpl = (void *)skb->data;
3938 struct cpl_pass_accept_req *req = (void *)(rss + 1);
3939 struct l2t_entry *e;
3940 struct dst_entry *dst;
3941 struct c4iw_ep *lep = NULL;
3942 u16 window;
3943 struct port_info *pi;
3944 struct net_device *pdev;
3945 u16 rss_qid, eth_hdr_len;
3946 int step;
3947 u32 tx_chan;
3948 struct neighbour *neigh;
3949
3950 /* Drop all non-SYN packets */
3951 if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
3952 goto reject;
3953
3954 /*
3955 * Drop all packets which did not hit the filter.
3956 * Unlikely to happen.
3957 */
3958 if (!(rss->filter_hit && rss->filter_tid))
3959 goto reject;
3960
3961 /*
3962 * Calculate the server tid from filter hit index from cpl_rx_pkt.
3963 */
3964 stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
3965
3966 lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
3967 if (!lep) {
3968 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
3969 goto reject;
3970 }
3971
3972 switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
3973 case CHELSIO_T4:
3974 eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3975 break;
3976 case CHELSIO_T5:
3977 eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3978 break;
3979 case CHELSIO_T6:
3980 eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
3981 break;
3982 default:
3983 pr_err("T%d Chip is not supported\n",
3984 CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
3985 goto reject;
3986 }
3987
3988 if (eth_hdr_len == ETH_HLEN) {
3989 eh = (struct ethhdr *)(req + 1);
3990 iph = (struct iphdr *)(eh + 1);
3991 } else {
3992 vlan_eh = (struct vlan_ethhdr *)(req + 1);
3993 iph = (struct iphdr *)(vlan_eh + 1);
3994 skb->vlan_tci = ntohs(cpl->vlan);
3995 }
3996
3997 if (iph->version != 0x4)
3998 goto reject;
3999
4000 tcph = (struct tcphdr *)(iph + 1);
4001 skb_set_network_header(skb, (void *)iph - (void *)rss);
4002 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
4003 skb_get(skb);
4004
4005 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
4006 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
4007 ntohs(tcph->source), iph->tos);
4008
4009 dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
4010 iph->tos);
4011 if (!dst) {
4012 pr_err("%s - failed to find dst entry!\n",
4013 __func__);
4014 goto reject;
4015 }
4016 neigh = dst_neigh_lookup_skb(dst, skb);
4017
4018 if (!neigh) {
4019 pr_err("%s - failed to allocate neigh!\n",
4020 __func__);
4021 goto free_dst;
4022 }
4023
4024 if (neigh->dev->flags & IFF_LOOPBACK) {
4025 pdev = ip_dev_find(&init_net, iph->daddr);
4026 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4027 pdev, 0);
4028 pi = (struct port_info *)netdev_priv(pdev);
4029 tx_chan = cxgb4_port_chan(pdev);
4030 dev_put(pdev);
4031 } else {
4032 pdev = get_real_dev(neigh->dev);
4033 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4034 pdev, 0);
4035 pi = (struct port_info *)netdev_priv(pdev);
4036 tx_chan = cxgb4_port_chan(pdev);
4037 }
4038 neigh_release(neigh);
4039 if (!e) {
4040 pr_err("%s - failed to allocate l2t entry!\n",
4041 __func__);
4042 goto free_dst;
4043 }
4044
4045 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4046 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
4047 window = (__force u16) htons((__force u16)tcph->window);
4048
4049 /* Calcuate filter portion for LE region. */
4050 filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4051 dev->rdev.lldi.ports[0],
4052 e));
4053
4054 /*
4055 * Synthesize the cpl_pass_accept_req. We have everything except the
4056 * TID. Once firmware sends a reply with TID we update the TID field
4057 * in cpl and pass it through the regular cpl_pass_accept_req path.
4058 */
4059 build_cpl_pass_accept_req(skb, stid, iph->tos);
4060 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4061 tcph->source, ntohl(tcph->seq), filter, window,
4062 rss_qid, pi->port_id);
4063 cxgb4_l2t_release(e);
4064 free_dst:
4065 dst_release(dst);
4066 reject:
4067 if (lep)
4068 c4iw_put_ep(&lep->com);
4069 return 0;
4070 }
4071
4072 /*
4073 * These are the real handlers that are called from a
4074 * work queue.
4075 */
4076 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4077 [CPL_ACT_ESTABLISH] = act_establish,
4078 [CPL_ACT_OPEN_RPL] = act_open_rpl,
4079 [CPL_RX_DATA] = rx_data,
4080 [CPL_ABORT_RPL_RSS] = abort_rpl,
4081 [CPL_ABORT_RPL] = abort_rpl,
4082 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
4083 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4084 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4085 [CPL_PASS_ESTABLISH] = pass_establish,
4086 [CPL_PEER_CLOSE] = peer_close,
4087 [CPL_ABORT_REQ_RSS] = peer_abort,
4088 [CPL_CLOSE_CON_RPL] = close_con_rpl,
4089 [CPL_RDMA_TERMINATE] = terminate,
4090 [CPL_FW4_ACK] = fw4_ack,
4091 [CPL_FW6_MSG] = deferred_fw6_msg,
4092 [CPL_RX_PKT] = rx_pkt,
4093 [FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4094 [FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4095 };
4096
4097 static void process_timeout(struct c4iw_ep *ep)
4098 {
4099 struct c4iw_qp_attributes attrs;
4100 int abort = 1;
4101
4102 mutex_lock(&ep->com.mutex);
4103 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
4104 ep->com.state);
4105 set_bit(TIMEDOUT, &ep->com.history);
4106 switch (ep->com.state) {
4107 case MPA_REQ_SENT:
4108 connect_reply_upcall(ep, -ETIMEDOUT);
4109 break;
4110 case MPA_REQ_WAIT:
4111 case MPA_REQ_RCVD:
4112 case MPA_REP_SENT:
4113 case FPDU_MODE:
4114 break;
4115 case CLOSING:
4116 case MORIBUND:
4117 if (ep->com.cm_id && ep->com.qp) {
4118 attrs.next_state = C4IW_QP_STATE_ERROR;
4119 c4iw_modify_qp(ep->com.qp->rhp,
4120 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4121 &attrs, 1);
4122 }
4123 close_complete_upcall(ep, -ETIMEDOUT);
4124 break;
4125 case ABORTING:
4126 case DEAD:
4127
4128 /*
4129 * These states are expected if the ep timed out at the same
4130 * time as another thread was calling stop_ep_timer().
4131 * So we silently do nothing for these states.
4132 */
4133 abort = 0;
4134 break;
4135 default:
4136 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4137 __func__, ep, ep->hwtid, ep->com.state);
4138 abort = 0;
4139 }
4140 mutex_unlock(&ep->com.mutex);
4141 if (abort)
4142 c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4143 c4iw_put_ep(&ep->com);
4144 }
4145
4146 static void process_timedout_eps(void)
4147 {
4148 struct c4iw_ep *ep;
4149
4150 spin_lock_irq(&timeout_lock);
4151 while (!list_empty(&timeout_list)) {
4152 struct list_head *tmp;
4153
4154 tmp = timeout_list.next;
4155 list_del(tmp);
4156 tmp->next = NULL;
4157 tmp->prev = NULL;
4158 spin_unlock_irq(&timeout_lock);
4159 ep = list_entry(tmp, struct c4iw_ep, entry);
4160 process_timeout(ep);
4161 spin_lock_irq(&timeout_lock);
4162 }
4163 spin_unlock_irq(&timeout_lock);
4164 }
4165
4166 static void process_work(struct work_struct *work)
4167 {
4168 struct sk_buff *skb = NULL;
4169 struct c4iw_dev *dev;
4170 struct cpl_act_establish *rpl;
4171 unsigned int opcode;
4172 int ret;
4173
4174 process_timedout_eps();
4175 while ((skb = skb_dequeue(&rxq))) {
4176 rpl = cplhdr(skb);
4177 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4178 opcode = rpl->ot.opcode;
4179
4180 BUG_ON(!work_handlers[opcode]);
4181 ret = work_handlers[opcode](dev, skb);
4182 if (!ret)
4183 kfree_skb(skb);
4184 process_timedout_eps();
4185 }
4186 }
4187
4188 static DECLARE_WORK(skb_work, process_work);
4189
4190 static void ep_timeout(unsigned long arg)
4191 {
4192 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
4193 int kickit = 0;
4194
4195 spin_lock(&timeout_lock);
4196 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4197 /*
4198 * Only insert if it is not already on the list.
4199 */
4200 if (!ep->entry.next) {
4201 list_add_tail(&ep->entry, &timeout_list);
4202 kickit = 1;
4203 }
4204 }
4205 spin_unlock(&timeout_lock);
4206 if (kickit)
4207 queue_work(workq, &skb_work);
4208 }
4209
4210 /*
4211 * All the CM events are handled on a work queue to have a safe context.
4212 */
4213 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4214 {
4215
4216 /*
4217 * Save dev in the skb->cb area.
4218 */
4219 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4220
4221 /*
4222 * Queue the skb and schedule the worker thread.
4223 */
4224 skb_queue_tail(&rxq, skb);
4225 queue_work(workq, &skb_work);
4226 return 0;
4227 }
4228
4229 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4230 {
4231 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4232
4233 if (rpl->status != CPL_ERR_NONE) {
4234 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
4235 "for tid %u\n", rpl->status, GET_TID(rpl));
4236 }
4237 kfree_skb(skb);
4238 return 0;
4239 }
4240
4241 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4242 {
4243 struct cpl_fw6_msg *rpl = cplhdr(skb);
4244 struct c4iw_wr_wait *wr_waitp;
4245 int ret;
4246
4247 PDBG("%s type %u\n", __func__, rpl->type);
4248
4249 switch (rpl->type) {
4250 case FW6_TYPE_WR_RPL:
4251 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4252 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4253 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
4254 if (wr_waitp)
4255 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
4256 kfree_skb(skb);
4257 break;
4258 case FW6_TYPE_CQE:
4259 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4260 sched(dev, skb);
4261 break;
4262 default:
4263 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
4264 rpl->type);
4265 kfree_skb(skb);
4266 break;
4267 }
4268 return 0;
4269 }
4270
4271 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4272 {
4273 struct cpl_abort_req_rss *req = cplhdr(skb);
4274 struct c4iw_ep *ep;
4275 unsigned int tid = GET_TID(req);
4276
4277 ep = get_ep_from_tid(dev, tid);
4278 /* This EP will be dereferenced in peer_abort() */
4279 if (!ep) {
4280 printk(KERN_WARNING MOD
4281 "Abort on non-existent endpoint, tid %d\n", tid);
4282 kfree_skb(skb);
4283 return 0;
4284 }
4285 if (is_neg_adv(req->status)) {
4286 PDBG("%s Negative advice on abort- tid %u status %d (%s)\n",
4287 __func__, ep->hwtid, req->status,
4288 neg_adv_str(req->status));
4289 goto out;
4290 }
4291 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
4292 ep->com.state);
4293
4294 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
4295 out:
4296 sched(dev, skb);
4297 return 0;
4298 }
4299
4300 /*
4301 * Most upcalls from the T4 Core go to sched() to
4302 * schedule the processing on a work queue.
4303 */
4304 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4305 [CPL_ACT_ESTABLISH] = sched,
4306 [CPL_ACT_OPEN_RPL] = sched,
4307 [CPL_RX_DATA] = sched,
4308 [CPL_ABORT_RPL_RSS] = sched,
4309 [CPL_ABORT_RPL] = sched,
4310 [CPL_PASS_OPEN_RPL] = sched,
4311 [CPL_CLOSE_LISTSRV_RPL] = sched,
4312 [CPL_PASS_ACCEPT_REQ] = sched,
4313 [CPL_PASS_ESTABLISH] = sched,
4314 [CPL_PEER_CLOSE] = sched,
4315 [CPL_CLOSE_CON_RPL] = sched,
4316 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
4317 [CPL_RDMA_TERMINATE] = sched,
4318 [CPL_FW4_ACK] = sched,
4319 [CPL_SET_TCB_RPL] = set_tcb_rpl,
4320 [CPL_FW6_MSG] = fw6_msg,
4321 [CPL_RX_PKT] = sched
4322 };
4323
4324 int __init c4iw_cm_init(void)
4325 {
4326 spin_lock_init(&timeout_lock);
4327 skb_queue_head_init(&rxq);
4328
4329 workq = create_singlethread_workqueue("iw_cxgb4");
4330 if (!workq)
4331 return -ENOMEM;
4332
4333 return 0;
4334 }
4335
4336 void c4iw_cm_term(void)
4337 {
4338 WARN_ON(!list_empty(&timeout_list));
4339 flush_workqueue(workq);
4340 destroy_workqueue(workq);
4341 }
Linux with added FPC-III support