projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
RDMA/cxgb4: Fix LE hash collision bug for passive open connection
[deliverable/linux.git] / drivers / infiniband / hw / cxgb4 / cm.c
1 /*
2 * Copyright (c) 2009-2010 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
48 #include "iw_cxgb4.h"
49
50 static char *states[] = {
51 "idle",
52 "listen",
53 "connecting",
54 "mpa_wait_req",
55 "mpa_req_sent",
56 "mpa_req_rcvd",
57 "mpa_rep_sent",
58 "fpdu_mode",
59 "aborting",
60 "closing",
61 "moribund",
62 "dead",
63 NULL,
64 };
65
66 static int nocong;
67 module_param(nocong, int, 0644);
68 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
69
70 static int enable_ecn;
71 module_param(enable_ecn, int, 0644);
72 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
73
74 static int dack_mode = 1;
75 module_param(dack_mode, int, 0644);
76 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
77
78 int c4iw_max_read_depth = 8;
79 module_param(c4iw_max_read_depth, int, 0644);
80 MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
81
82 static int enable_tcp_timestamps;
83 module_param(enable_tcp_timestamps, int, 0644);
84 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
85
86 static int enable_tcp_sack;
87 module_param(enable_tcp_sack, int, 0644);
88 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
89
90 static int enable_tcp_window_scaling = 1;
91 module_param(enable_tcp_window_scaling, int, 0644);
92 MODULE_PARM_DESC(enable_tcp_window_scaling,
93 "Enable tcp window scaling (default=1)");
94
95 int c4iw_debug;
96 module_param(c4iw_debug, int, 0644);
97 MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
98
99 static int peer2peer;
100 module_param(peer2peer, int, 0644);
101 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=0)");
102
103 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
104 module_param(p2p_type, int, 0644);
105 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
106 "1=RDMA_READ 0=RDMA_WRITE (default 1)");
107
108 static int ep_timeout_secs = 60;
109 module_param(ep_timeout_secs, int, 0644);
110 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
111 "in seconds (default=60)");
112
113 static int mpa_rev = 1;
114 module_param(mpa_rev, int, 0644);
115 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
116 "1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
117 " compliant (default=1)");
118
119 static int markers_enabled;
120 module_param(markers_enabled, int, 0644);
121 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
122
123 static int crc_enabled = 1;
124 module_param(crc_enabled, int, 0644);
125 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
126
127 static int rcv_win = 256 * 1024;
128 module_param(rcv_win, int, 0644);
129 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
130
131 static int snd_win = 128 * 1024;
132 module_param(snd_win, int, 0644);
133 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
134
135 static struct workqueue_struct *workq;
136
137 static struct sk_buff_head rxq;
138
139 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
140 static void ep_timeout(unsigned long arg);
141 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
142
143 static LIST_HEAD(timeout_list);
144 static spinlock_t timeout_lock;
145
146 static void start_ep_timer(struct c4iw_ep *ep)
147 {
148 PDBG("%s ep %p\n", __func__, ep);
149 if (timer_pending(&ep->timer)) {
150 PDBG("%s stopped / restarted timer ep %p\n", __func__, ep);
151 del_timer_sync(&ep->timer);
152 } else
153 c4iw_get_ep(&ep->com);
154 ep->timer.expires = jiffies + ep_timeout_secs * HZ;
155 ep->timer.data = (unsigned long)ep;
156 ep->timer.function = ep_timeout;
157 add_timer(&ep->timer);
158 }
159
160 static void stop_ep_timer(struct c4iw_ep *ep)
161 {
162 PDBG("%s ep %p\n", __func__, ep);
163 if (!timer_pending(&ep->timer)) {
164 WARN(1, "%s timer stopped when its not running! "
165 "ep %p state %u\n", __func__, ep, ep->com.state);
166 return;
167 }
168 del_timer_sync(&ep->timer);
169 c4iw_put_ep(&ep->com);
170 }
171
172 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
173 struct l2t_entry *l2e)
174 {
175 int error = 0;
176
177 if (c4iw_fatal_error(rdev)) {
178 kfree_skb(skb);
179 PDBG("%s - device in error state - dropping\n", __func__);
180 return -EIO;
181 }
182 error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
183 if (error < 0)
184 kfree_skb(skb);
185 return error < 0 ? error : 0;
186 }
187
188 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
189 {
190 int error = 0;
191
192 if (c4iw_fatal_error(rdev)) {
193 kfree_skb(skb);
194 PDBG("%s - device in error state - dropping\n", __func__);
195 return -EIO;
196 }
197 error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
198 if (error < 0)
199 kfree_skb(skb);
200 return error < 0 ? error : 0;
201 }
202
203 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
204 {
205 struct cpl_tid_release *req;
206
207 skb = get_skb(skb, sizeof *req, GFP_KERNEL);
208 if (!skb)
209 return;
210 req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
211 INIT_TP_WR(req, hwtid);
212 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
213 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
214 c4iw_ofld_send(rdev, skb);
215 return;
216 }
217
218 static void set_emss(struct c4iw_ep *ep, u16 opt)
219 {
220 ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] - 40;
221 ep->mss = ep->emss;
222 if (GET_TCPOPT_TSTAMP(opt))
223 ep->emss -= 12;
224 if (ep->emss < 128)
225 ep->emss = 128;
226 PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
227 ep->mss, ep->emss);
228 }
229
230 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
231 {
232 enum c4iw_ep_state state;
233
234 mutex_lock(&epc->mutex);
235 state = epc->state;
236 mutex_unlock(&epc->mutex);
237 return state;
238 }
239
240 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
241 {
242 epc->state = new;
243 }
244
245 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
246 {
247 mutex_lock(&epc->mutex);
248 PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
249 __state_set(epc, new);
250 mutex_unlock(&epc->mutex);
251 return;
252 }
253
254 static void *alloc_ep(int size, gfp_t gfp)
255 {
256 struct c4iw_ep_common *epc;
257
258 epc = kzalloc(size, gfp);
259 if (epc) {
260 kref_init(&epc->kref);
261 mutex_init(&epc->mutex);
262 c4iw_init_wr_wait(&epc->wr_wait);
263 }
264 PDBG("%s alloc ep %p\n", __func__, epc);
265 return epc;
266 }
267
268 void _c4iw_free_ep(struct kref *kref)
269 {
270 struct c4iw_ep *ep;
271
272 ep = container_of(kref, struct c4iw_ep, com.kref);
273 PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
274 if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
275 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
276 dst_release(ep->dst);
277 cxgb4_l2t_release(ep->l2t);
278 }
279 kfree(ep);
280 }
281
282 static void release_ep_resources(struct c4iw_ep *ep)
283 {
284 set_bit(RELEASE_RESOURCES, &ep->com.flags);
285 c4iw_put_ep(&ep->com);
286 }
287
288 static int status2errno(int status)
289 {
290 switch (status) {
291 case CPL_ERR_NONE:
292 return 0;
293 case CPL_ERR_CONN_RESET:
294 return -ECONNRESET;
295 case CPL_ERR_ARP_MISS:
296 return -EHOSTUNREACH;
297 case CPL_ERR_CONN_TIMEDOUT:
298 return -ETIMEDOUT;
299 case CPL_ERR_TCAM_FULL:
300 return -ENOMEM;
301 case CPL_ERR_CONN_EXIST:
302 return -EADDRINUSE;
303 default:
304 return -EIO;
305 }
306 }
307
308 /*
309 * Try and reuse skbs already allocated...
310 */
311 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
312 {
313 if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
314 skb_trim(skb, 0);
315 skb_get(skb);
316 skb_reset_transport_header(skb);
317 } else {
318 skb = alloc_skb(len, gfp);
319 }
320 return skb;
321 }
322
323 static struct rtable *find_route(struct c4iw_dev *dev, __be32 local_ip,
324 __be32 peer_ip, __be16 local_port,
325 __be16 peer_port, u8 tos)
326 {
327 struct rtable *rt;
328 struct flowi4 fl4;
329
330 rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
331 peer_port, local_port, IPPROTO_TCP,
332 tos, 0);
333 if (IS_ERR(rt))
334 return NULL;
335 return rt;
336 }
337
338 static void arp_failure_discard(void *handle, struct sk_buff *skb)
339 {
340 PDBG("%s c4iw_dev %p\n", __func__, handle);
341 kfree_skb(skb);
342 }
343
344 /*
345 * Handle an ARP failure for an active open.
346 */
347 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
348 {
349 printk(KERN_ERR MOD "ARP failure duing connect\n");
350 kfree_skb(skb);
351 }
352
353 /*
354 * Handle an ARP failure for a CPL_ABORT_REQ. Change it into a no RST variant
355 * and send it along.
356 */
357 static void abort_arp_failure(void *handle, struct sk_buff *skb)
358 {
359 struct c4iw_rdev *rdev = handle;
360 struct cpl_abort_req *req = cplhdr(skb);
361
362 PDBG("%s rdev %p\n", __func__, rdev);
363 req->cmd = CPL_ABORT_NO_RST;
364 c4iw_ofld_send(rdev, skb);
365 }
366
367 static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
368 {
369 unsigned int flowclen = 80;
370 struct fw_flowc_wr *flowc;
371 int i;
372
373 skb = get_skb(skb, flowclen, GFP_KERNEL);
374 flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
375
376 flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
377 FW_FLOWC_WR_NPARAMS(8));
378 flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
379 16)) | FW_WR_FLOWID(ep->hwtid));
380
381 flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
382 flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
383 flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
384 flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
385 flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
386 flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
387 flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
388 flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
389 flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
390 flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
391 flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
392 flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
393 flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
394 flowc->mnemval[6].val = cpu_to_be32(snd_win);
395 flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
396 flowc->mnemval[7].val = cpu_to_be32(ep->emss);
397 /* Pad WR to 16 byte boundary */
398 flowc->mnemval[8].mnemonic = 0;
399 flowc->mnemval[8].val = 0;
400 for (i = 0; i < 9; i++) {
401 flowc->mnemval[i].r4[0] = 0;
402 flowc->mnemval[i].r4[1] = 0;
403 flowc->mnemval[i].r4[2] = 0;
404 }
405
406 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
407 c4iw_ofld_send(&ep->com.dev->rdev, skb);
408 }
409
410 static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
411 {
412 struct cpl_close_con_req *req;
413 struct sk_buff *skb;
414 int wrlen = roundup(sizeof *req, 16);
415
416 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
417 skb = get_skb(NULL, wrlen, gfp);
418 if (!skb) {
419 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
420 return -ENOMEM;
421 }
422 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
423 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
424 req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
425 memset(req, 0, wrlen);
426 INIT_TP_WR(req, ep->hwtid);
427 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
428 ep->hwtid));
429 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
430 }
431
432 static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
433 {
434 struct cpl_abort_req *req;
435 int wrlen = roundup(sizeof *req, 16);
436
437 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
438 skb = get_skb(skb, wrlen, gfp);
439 if (!skb) {
440 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
441 __func__);
442 return -ENOMEM;
443 }
444 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
445 t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
446 req = (struct cpl_abort_req *) skb_put(skb, wrlen);
447 memset(req, 0, wrlen);
448 INIT_TP_WR(req, ep->hwtid);
449 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
450 req->cmd = CPL_ABORT_SEND_RST;
451 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
452 }
453
454 #define VLAN_NONE 0xfff
455 #define FILTER_SEL_VLAN_NONE 0xffff
456 #define FILTER_SEL_WIDTH_P_FC (3+1) /* port uses 3 bits, FCoE one bit */
457 #define FILTER_SEL_WIDTH_VIN_P_FC \
458 (6 + 7 + FILTER_SEL_WIDTH_P_FC) /* 6 bits are unused, VF uses 7 bits*/
459 #define FILTER_SEL_WIDTH_TAG_P_FC \
460 (3 + FILTER_SEL_WIDTH_VIN_P_FC) /* PF uses 3 bits */
461 #define FILTER_SEL_WIDTH_VLD_TAG_P_FC (1 + FILTER_SEL_WIDTH_TAG_P_FC)
462
463 static unsigned int select_ntuple(struct c4iw_dev *dev, struct dst_entry *dst,
464 struct l2t_entry *l2t)
465 {
466 unsigned int ntuple = 0;
467 u32 viid;
468
469 switch (dev->rdev.lldi.filt_mode) {
470
471 /* default filter mode */
472 case HW_TPL_FR_MT_PR_IV_P_FC:
473 if (l2t->vlan == VLAN_NONE)
474 ntuple |= FILTER_SEL_VLAN_NONE << FILTER_SEL_WIDTH_P_FC;
475 else {
476 ntuple |= l2t->vlan << FILTER_SEL_WIDTH_P_FC;
477 ntuple |= 1 << FILTER_SEL_WIDTH_VLD_TAG_P_FC;
478 }
479 ntuple |= l2t->lport << S_PORT | IPPROTO_TCP <<
480 FILTER_SEL_WIDTH_VLD_TAG_P_FC;
481 break;
482 case HW_TPL_FR_MT_PR_OV_P_FC: {
483 viid = cxgb4_port_viid(l2t->neigh->dev);
484
485 ntuple |= FW_VIID_VIN_GET(viid) << FILTER_SEL_WIDTH_P_FC;
486 ntuple |= FW_VIID_PFN_GET(viid) << FILTER_SEL_WIDTH_VIN_P_FC;
487 ntuple |= FW_VIID_VIVLD_GET(viid) << FILTER_SEL_WIDTH_TAG_P_FC;
488 ntuple |= l2t->lport << S_PORT | IPPROTO_TCP <<
489 FILTER_SEL_WIDTH_VLD_TAG_P_FC;
490 break;
491 }
492 default:
493 break;
494 }
495 return ntuple;
496 }
497
498 static int send_connect(struct c4iw_ep *ep)
499 {
500 struct cpl_act_open_req *req;
501 struct sk_buff *skb;
502 u64 opt0;
503 u32 opt2;
504 unsigned int mtu_idx;
505 int wscale;
506 int wrlen = roundup(sizeof *req, 16);
507
508 PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
509
510 skb = get_skb(NULL, wrlen, GFP_KERNEL);
511 if (!skb) {
512 printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
513 __func__);
514 return -ENOMEM;
515 }
516 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
517
518 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
519 wscale = compute_wscale(rcv_win);
520 opt0 = (nocong ? NO_CONG(1) : 0) |
521 KEEP_ALIVE(1) |
522 DELACK(1) |
523 WND_SCALE(wscale) |
524 MSS_IDX(mtu_idx) |
525 L2T_IDX(ep->l2t->idx) |
526 TX_CHAN(ep->tx_chan) |
527 SMAC_SEL(ep->smac_idx) |
528 DSCP(ep->tos) |
529 ULP_MODE(ULP_MODE_TCPDDP) |
530 RCV_BUFSIZ(rcv_win>>10);
531 opt2 = RX_CHANNEL(0) |
532 CCTRL_ECN(enable_ecn) |
533 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
534 if (enable_tcp_timestamps)
535 opt2 |= TSTAMPS_EN(1);
536 if (enable_tcp_sack)
537 opt2 |= SACK_EN(1);
538 if (wscale && enable_tcp_window_scaling)
539 opt2 |= WND_SCALE_EN(1);
540 t4_set_arp_err_handler(skb, NULL, act_open_req_arp_failure);
541
542 req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
543 INIT_TP_WR(req, 0);
544 OPCODE_TID(req) = cpu_to_be32(
545 MK_OPCODE_TID(CPL_ACT_OPEN_REQ, ((ep->rss_qid<<14)|ep->atid)));
546 req->local_port = ep->com.local_addr.sin_port;
547 req->peer_port = ep->com.remote_addr.sin_port;
548 req->local_ip = ep->com.local_addr.sin_addr.s_addr;
549 req->peer_ip = ep->com.remote_addr.sin_addr.s_addr;
550 req->opt0 = cpu_to_be64(opt0);
551 req->params = cpu_to_be32(select_ntuple(ep->com.dev, ep->dst, ep->l2t));
552 req->opt2 = cpu_to_be32(opt2);
553 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
554 }
555
556 static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
557 u8 mpa_rev_to_use)
558 {
559 int mpalen, wrlen;
560 struct fw_ofld_tx_data_wr *req;
561 struct mpa_message *mpa;
562 struct mpa_v2_conn_params mpa_v2_params;
563
564 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
565
566 BUG_ON(skb_cloned(skb));
567
568 mpalen = sizeof(*mpa) + ep->plen;
569 if (mpa_rev_to_use == 2)
570 mpalen += sizeof(struct mpa_v2_conn_params);
571 wrlen = roundup(mpalen + sizeof *req, 16);
572 skb = get_skb(skb, wrlen, GFP_KERNEL);
573 if (!skb) {
574 connect_reply_upcall(ep, -ENOMEM);
575 return;
576 }
577 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
578
579 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
580 memset(req, 0, wrlen);
581 req->op_to_immdlen = cpu_to_be32(
582 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
583 FW_WR_COMPL(1) |
584 FW_WR_IMMDLEN(mpalen));
585 req->flowid_len16 = cpu_to_be32(
586 FW_WR_FLOWID(ep->hwtid) |
587 FW_WR_LEN16(wrlen >> 4));
588 req->plen = cpu_to_be32(mpalen);
589 req->tunnel_to_proxy = cpu_to_be32(
590 FW_OFLD_TX_DATA_WR_FLUSH(1) |
591 FW_OFLD_TX_DATA_WR_SHOVE(1));
592
593 mpa = (struct mpa_message *)(req + 1);
594 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
595 mpa->flags = (crc_enabled ? MPA_CRC : 0) |
596 (markers_enabled ? MPA_MARKERS : 0) |
597 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
598 mpa->private_data_size = htons(ep->plen);
599 mpa->revision = mpa_rev_to_use;
600 if (mpa_rev_to_use == 1) {
601 ep->tried_with_mpa_v1 = 1;
602 ep->retry_with_mpa_v1 = 0;
603 }
604
605 if (mpa_rev_to_use == 2) {
606 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
607 sizeof (struct mpa_v2_conn_params));
608 mpa_v2_params.ird = htons((u16)ep->ird);
609 mpa_v2_params.ord = htons((u16)ep->ord);
610
611 if (peer2peer) {
612 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
613 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
614 mpa_v2_params.ord |=
615 htons(MPA_V2_RDMA_WRITE_RTR);
616 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
617 mpa_v2_params.ord |=
618 htons(MPA_V2_RDMA_READ_RTR);
619 }
620 memcpy(mpa->private_data, &mpa_v2_params,
621 sizeof(struct mpa_v2_conn_params));
622
623 if (ep->plen)
624 memcpy(mpa->private_data +
625 sizeof(struct mpa_v2_conn_params),
626 ep->mpa_pkt + sizeof(*mpa), ep->plen);
627 } else
628 if (ep->plen)
629 memcpy(mpa->private_data,
630 ep->mpa_pkt + sizeof(*mpa), ep->plen);
631
632 /*
633 * Reference the mpa skb. This ensures the data area
634 * will remain in memory until the hw acks the tx.
635 * Function fw4_ack() will deref it.
636 */
637 skb_get(skb);
638 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
639 BUG_ON(ep->mpa_skb);
640 ep->mpa_skb = skb;
641 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
642 start_ep_timer(ep);
643 state_set(&ep->com, MPA_REQ_SENT);
644 ep->mpa_attr.initiator = 1;
645 return;
646 }
647
648 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
649 {
650 int mpalen, wrlen;
651 struct fw_ofld_tx_data_wr *req;
652 struct mpa_message *mpa;
653 struct sk_buff *skb;
654 struct mpa_v2_conn_params mpa_v2_params;
655
656 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
657
658 mpalen = sizeof(*mpa) + plen;
659 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
660 mpalen += sizeof(struct mpa_v2_conn_params);
661 wrlen = roundup(mpalen + sizeof *req, 16);
662
663 skb = get_skb(NULL, wrlen, GFP_KERNEL);
664 if (!skb) {
665 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
666 return -ENOMEM;
667 }
668 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
669
670 req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
671 memset(req, 0, wrlen);
672 req->op_to_immdlen = cpu_to_be32(
673 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
674 FW_WR_COMPL(1) |
675 FW_WR_IMMDLEN(mpalen));
676 req->flowid_len16 = cpu_to_be32(
677 FW_WR_FLOWID(ep->hwtid) |
678 FW_WR_LEN16(wrlen >> 4));
679 req->plen = cpu_to_be32(mpalen);
680 req->tunnel_to_proxy = cpu_to_be32(
681 FW_OFLD_TX_DATA_WR_FLUSH(1) |
682 FW_OFLD_TX_DATA_WR_SHOVE(1));
683
684 mpa = (struct mpa_message *)(req + 1);
685 memset(mpa, 0, sizeof(*mpa));
686 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
687 mpa->flags = MPA_REJECT;
688 mpa->revision = mpa_rev;
689 mpa->private_data_size = htons(plen);
690
691 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
692 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
693 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
694 sizeof (struct mpa_v2_conn_params));
695 mpa_v2_params.ird = htons(((u16)ep->ird) |
696 (peer2peer ? MPA_V2_PEER2PEER_MODEL :
697 0));
698 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
699 (p2p_type ==
700 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
701 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
702 FW_RI_INIT_P2PTYPE_READ_REQ ?
703 MPA_V2_RDMA_READ_RTR : 0) : 0));
704 memcpy(mpa->private_data, &mpa_v2_params,
705 sizeof(struct mpa_v2_conn_params));
706
707 if (ep->plen)
708 memcpy(mpa->private_data +
709 sizeof(struct mpa_v2_conn_params), pdata, plen);
710 } else
711 if (plen)
712 memcpy(mpa->private_data, pdata, plen);
713
714 /*
715 * Reference the mpa skb again. This ensures the data area
716 * will remain in memory until the hw acks the tx.
717 * Function fw4_ack() will deref it.
718 */
719 skb_get(skb);
720 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
721 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
722 BUG_ON(ep->mpa_skb);
723 ep->mpa_skb = skb;
724 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
725 }
726
727 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
728 {
729 int mpalen, wrlen;
730 struct fw_ofld_tx_data_wr *req;
731 struct mpa_message *mpa;
732 struct sk_buff *skb;
733 struct mpa_v2_conn_params mpa_v2_params;
734
735 PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
736
737 mpalen = sizeof(*mpa) + plen;
738 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
739 mpalen += sizeof(struct mpa_v2_conn_params);
740 wrlen = roundup(mpalen + sizeof *req, 16);
741
742 skb = get_skb(NULL, wrlen, GFP_KERNEL);
743 if (!skb) {
744 printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
745 return -ENOMEM;
746 }
747 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
748
749 req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
750 memset(req, 0, wrlen);
751 req->op_to_immdlen = cpu_to_be32(
752 FW_WR_OP(FW_OFLD_TX_DATA_WR) |
753 FW_WR_COMPL(1) |
754 FW_WR_IMMDLEN(mpalen));
755 req->flowid_len16 = cpu_to_be32(
756 FW_WR_FLOWID(ep->hwtid) |
757 FW_WR_LEN16(wrlen >> 4));
758 req->plen = cpu_to_be32(mpalen);
759 req->tunnel_to_proxy = cpu_to_be32(
760 FW_OFLD_TX_DATA_WR_FLUSH(1) |
761 FW_OFLD_TX_DATA_WR_SHOVE(1));
762
763 mpa = (struct mpa_message *)(req + 1);
764 memset(mpa, 0, sizeof(*mpa));
765 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
766 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
767 (markers_enabled ? MPA_MARKERS : 0);
768 mpa->revision = ep->mpa_attr.version;
769 mpa->private_data_size = htons(plen);
770
771 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
772 mpa->flags |= MPA_ENHANCED_RDMA_CONN;
773 mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
774 sizeof (struct mpa_v2_conn_params));
775 mpa_v2_params.ird = htons((u16)ep->ird);
776 mpa_v2_params.ord = htons((u16)ep->ord);
777 if (peer2peer && (ep->mpa_attr.p2p_type !=
778 FW_RI_INIT_P2PTYPE_DISABLED)) {
779 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
780
781 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
782 mpa_v2_params.ord |=
783 htons(MPA_V2_RDMA_WRITE_RTR);
784 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
785 mpa_v2_params.ord |=
786 htons(MPA_V2_RDMA_READ_RTR);
787 }
788
789 memcpy(mpa->private_data, &mpa_v2_params,
790 sizeof(struct mpa_v2_conn_params));
791
792 if (ep->plen)
793 memcpy(mpa->private_data +
794 sizeof(struct mpa_v2_conn_params), pdata, plen);
795 } else
796 if (plen)
797 memcpy(mpa->private_data, pdata, plen);
798
799 /*
800 * Reference the mpa skb. This ensures the data area
801 * will remain in memory until the hw acks the tx.
802 * Function fw4_ack() will deref it.
803 */
804 skb_get(skb);
805 t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
806 ep->mpa_skb = skb;
807 state_set(&ep->com, MPA_REP_SENT);
808 return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
809 }
810
811 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
812 {
813 struct c4iw_ep *ep;
814 struct cpl_act_establish *req = cplhdr(skb);
815 unsigned int tid = GET_TID(req);
816 unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
817 struct tid_info *t = dev->rdev.lldi.tids;
818
819 ep = lookup_atid(t, atid);
820
821 PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
822 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
823
824 dst_confirm(ep->dst);
825
826 /* setup the hwtid for this connection */
827 ep->hwtid = tid;
828 cxgb4_insert_tid(t, ep, tid);
829
830 ep->snd_seq = be32_to_cpu(req->snd_isn);
831 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
832
833 set_emss(ep, ntohs(req->tcp_opt));
834
835 /* dealloc the atid */
836 cxgb4_free_atid(t, atid);
837
838 /* start MPA negotiation */
839 send_flowc(ep, NULL);
840 if (ep->retry_with_mpa_v1)
841 send_mpa_req(ep, skb, 1);
842 else
843 send_mpa_req(ep, skb, mpa_rev);
844
845 return 0;
846 }
847
848 static void close_complete_upcall(struct c4iw_ep *ep)
849 {
850 struct iw_cm_event event;
851
852 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
853 memset(&event, 0, sizeof(event));
854 event.event = IW_CM_EVENT_CLOSE;
855 if (ep->com.cm_id) {
856 PDBG("close complete delivered ep %p cm_id %p tid %u\n",
857 ep, ep->com.cm_id, ep->hwtid);
858 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
859 ep->com.cm_id->rem_ref(ep->com.cm_id);
860 ep->com.cm_id = NULL;
861 ep->com.qp = NULL;
862 }
863 }
864
865 static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
866 {
867 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
868 close_complete_upcall(ep);
869 state_set(&ep->com, ABORTING);
870 return send_abort(ep, skb, gfp);
871 }
872
873 static void peer_close_upcall(struct c4iw_ep *ep)
874 {
875 struct iw_cm_event event;
876
877 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
878 memset(&event, 0, sizeof(event));
879 event.event = IW_CM_EVENT_DISCONNECT;
880 if (ep->com.cm_id) {
881 PDBG("peer close delivered ep %p cm_id %p tid %u\n",
882 ep, ep->com.cm_id, ep->hwtid);
883 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
884 }
885 }
886
887 static void peer_abort_upcall(struct c4iw_ep *ep)
888 {
889 struct iw_cm_event event;
890
891 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
892 memset(&event, 0, sizeof(event));
893 event.event = IW_CM_EVENT_CLOSE;
894 event.status = -ECONNRESET;
895 if (ep->com.cm_id) {
896 PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
897 ep->com.cm_id, ep->hwtid);
898 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
899 ep->com.cm_id->rem_ref(ep->com.cm_id);
900 ep->com.cm_id = NULL;
901 ep->com.qp = NULL;
902 }
903 }
904
905 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
906 {
907 struct iw_cm_event event;
908
909 PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
910 memset(&event, 0, sizeof(event));
911 event.event = IW_CM_EVENT_CONNECT_REPLY;
912 event.status = status;
913 event.local_addr = ep->com.local_addr;
914 event.remote_addr = ep->com.remote_addr;
915
916 if ((status == 0) || (status == -ECONNREFUSED)) {
917 if (!ep->tried_with_mpa_v1) {
918 /* this means MPA_v2 is used */
919 event.private_data_len = ep->plen -
920 sizeof(struct mpa_v2_conn_params);
921 event.private_data = ep->mpa_pkt +
922 sizeof(struct mpa_message) +
923 sizeof(struct mpa_v2_conn_params);
924 } else {
925 /* this means MPA_v1 is used */
926 event.private_data_len = ep->plen;
927 event.private_data = ep->mpa_pkt +
928 sizeof(struct mpa_message);
929 }
930 }
931
932 PDBG("%s ep %p tid %u status %d\n", __func__, ep,
933 ep->hwtid, status);
934 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
935
936 if (status < 0) {
937 ep->com.cm_id->rem_ref(ep->com.cm_id);
938 ep->com.cm_id = NULL;
939 ep->com.qp = NULL;
940 }
941 }
942
943 static void connect_request_upcall(struct c4iw_ep *ep)
944 {
945 struct iw_cm_event event;
946
947 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
948 memset(&event, 0, sizeof(event));
949 event.event = IW_CM_EVENT_CONNECT_REQUEST;
950 event.local_addr = ep->com.local_addr;
951 event.remote_addr = ep->com.remote_addr;
952 event.provider_data = ep;
953 if (!ep->tried_with_mpa_v1) {
954 /* this means MPA_v2 is used */
955 event.ord = ep->ord;
956 event.ird = ep->ird;
957 event.private_data_len = ep->plen -
958 sizeof(struct mpa_v2_conn_params);
959 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
960 sizeof(struct mpa_v2_conn_params);
961 } else {
962 /* this means MPA_v1 is used. Send max supported */
963 event.ord = c4iw_max_read_depth;
964 event.ird = c4iw_max_read_depth;
965 event.private_data_len = ep->plen;
966 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
967 }
968 if (state_read(&ep->parent_ep->com) != DEAD) {
969 c4iw_get_ep(&ep->com);
970 ep->parent_ep->com.cm_id->event_handler(
971 ep->parent_ep->com.cm_id,
972 &event);
973 }
974 c4iw_put_ep(&ep->parent_ep->com);
975 ep->parent_ep = NULL;
976 }
977
978 static void established_upcall(struct c4iw_ep *ep)
979 {
980 struct iw_cm_event event;
981
982 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
983 memset(&event, 0, sizeof(event));
984 event.event = IW_CM_EVENT_ESTABLISHED;
985 event.ird = ep->ird;
986 event.ord = ep->ord;
987 if (ep->com.cm_id) {
988 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
989 ep->com.cm_id->event_handler(ep->com.cm_id, &event);
990 }
991 }
992
993 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
994 {
995 struct cpl_rx_data_ack *req;
996 struct sk_buff *skb;
997 int wrlen = roundup(sizeof *req, 16);
998
999 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
1000 skb = get_skb(NULL, wrlen, GFP_KERNEL);
1001 if (!skb) {
1002 printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
1003 return 0;
1004 }
1005
1006 req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
1007 memset(req, 0, wrlen);
1008 INIT_TP_WR(req, ep->hwtid);
1009 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
1010 ep->hwtid));
1011 req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
1012 F_RX_DACK_CHANGE |
1013 V_RX_DACK_MODE(dack_mode));
1014 set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
1015 c4iw_ofld_send(&ep->com.dev->rdev, skb);
1016 return credits;
1017 }
1018
1019 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1020 {
1021 struct mpa_message *mpa;
1022 struct mpa_v2_conn_params *mpa_v2_params;
1023 u16 plen;
1024 u16 resp_ird, resp_ord;
1025 u8 rtr_mismatch = 0, insuff_ird = 0;
1026 struct c4iw_qp_attributes attrs;
1027 enum c4iw_qp_attr_mask mask;
1028 int err;
1029
1030 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1031
1032 /*
1033 * Stop mpa timer. If it expired, then the state has
1034 * changed and we bail since ep_timeout already aborted
1035 * the connection.
1036 */
1037 stop_ep_timer(ep);
1038 if (state_read(&ep->com) != MPA_REQ_SENT)
1039 return;
1040
1041 /*
1042 * If we get more than the supported amount of private data
1043 * then we must fail this connection.
1044 */
1045 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1046 err = -EINVAL;
1047 goto err;
1048 }
1049
1050 /*
1051 * copy the new data into our accumulation buffer.
1052 */
1053 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1054 skb->len);
1055 ep->mpa_pkt_len += skb->len;
1056
1057 /*
1058 * if we don't even have the mpa message, then bail.
1059 */
1060 if (ep->mpa_pkt_len < sizeof(*mpa))
1061 return;
1062 mpa = (struct mpa_message *) ep->mpa_pkt;
1063
1064 /* Validate MPA header. */
1065 if (mpa->revision > mpa_rev) {
1066 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1067 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1068 err = -EPROTO;
1069 goto err;
1070 }
1071 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1072 err = -EPROTO;
1073 goto err;
1074 }
1075
1076 plen = ntohs(mpa->private_data_size);
1077
1078 /*
1079 * Fail if there's too much private data.
1080 */
1081 if (plen > MPA_MAX_PRIVATE_DATA) {
1082 err = -EPROTO;
1083 goto err;
1084 }
1085
1086 /*
1087 * If plen does not account for pkt size
1088 */
1089 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1090 err = -EPROTO;
1091 goto err;
1092 }
1093
1094 ep->plen = (u8) plen;
1095
1096 /*
1097 * If we don't have all the pdata yet, then bail.
1098 * We'll continue process when more data arrives.
1099 */
1100 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1101 return;
1102
1103 if (mpa->flags & MPA_REJECT) {
1104 err = -ECONNREFUSED;
1105 goto err;
1106 }
1107
1108 /*
1109 * If we get here we have accumulated the entire mpa
1110 * start reply message including private data. And
1111 * the MPA header is valid.
1112 */
1113 state_set(&ep->com, FPDU_MODE);
1114 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1115 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1116 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1117 ep->mpa_attr.version = mpa->revision;
1118 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1119
1120 if (mpa->revision == 2) {
1121 ep->mpa_attr.enhanced_rdma_conn =
1122 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1123 if (ep->mpa_attr.enhanced_rdma_conn) {
1124 mpa_v2_params = (struct mpa_v2_conn_params *)
1125 (ep->mpa_pkt + sizeof(*mpa));
1126 resp_ird = ntohs(mpa_v2_params->ird) &
1127 MPA_V2_IRD_ORD_MASK;
1128 resp_ord = ntohs(mpa_v2_params->ord) &
1129 MPA_V2_IRD_ORD_MASK;
1130
1131 /*
1132 * This is a double-check. Ideally, below checks are
1133 * not required since ird/ord stuff has been taken
1134 * care of in c4iw_accept_cr
1135 */
1136 if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
1137 err = -ENOMEM;
1138 ep->ird = resp_ord;
1139 ep->ord = resp_ird;
1140 insuff_ird = 1;
1141 }
1142
1143 if (ntohs(mpa_v2_params->ird) &
1144 MPA_V2_PEER2PEER_MODEL) {
1145 if (ntohs(mpa_v2_params->ord) &
1146 MPA_V2_RDMA_WRITE_RTR)
1147 ep->mpa_attr.p2p_type =
1148 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1149 else if (ntohs(mpa_v2_params->ord) &
1150 MPA_V2_RDMA_READ_RTR)
1151 ep->mpa_attr.p2p_type =
1152 FW_RI_INIT_P2PTYPE_READ_REQ;
1153 }
1154 }
1155 } else if (mpa->revision == 1)
1156 if (peer2peer)
1157 ep->mpa_attr.p2p_type = p2p_type;
1158
1159 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1160 "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
1161 "%d\n", __func__, ep->mpa_attr.crc_enabled,
1162 ep->mpa_attr.recv_marker_enabled,
1163 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1164 ep->mpa_attr.p2p_type, p2p_type);
1165
1166 /*
1167 * If responder's RTR does not match with that of initiator, assign
1168 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1169 * generated when moving QP to RTS state.
1170 * A TERM message will be sent after QP has moved to RTS state
1171 */
1172 if ((ep->mpa_attr.version == 2) && peer2peer &&
1173 (ep->mpa_attr.p2p_type != p2p_type)) {
1174 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1175 rtr_mismatch = 1;
1176 }
1177
1178 attrs.mpa_attr = ep->mpa_attr;
1179 attrs.max_ird = ep->ird;
1180 attrs.max_ord = ep->ord;
1181 attrs.llp_stream_handle = ep;
1182 attrs.next_state = C4IW_QP_STATE_RTS;
1183
1184 mask = C4IW_QP_ATTR_NEXT_STATE |
1185 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1186 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1187
1188 /* bind QP and TID with INIT_WR */
1189 err = c4iw_modify_qp(ep->com.qp->rhp,
1190 ep->com.qp, mask, &attrs, 1);
1191 if (err)
1192 goto err;
1193
1194 /*
1195 * If responder's RTR requirement did not match with what initiator
1196 * supports, generate TERM message
1197 */
1198 if (rtr_mismatch) {
1199 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
1200 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1201 attrs.ecode = MPA_NOMATCH_RTR;
1202 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1203 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1204 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1205 err = -ENOMEM;
1206 goto out;
1207 }
1208
1209 /*
1210 * Generate TERM if initiator IRD is not sufficient for responder
1211 * provided ORD. Currently, we do the same behaviour even when
1212 * responder provided IRD is also not sufficient as regards to
1213 * initiator ORD.
1214 */
1215 if (insuff_ird) {
1216 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
1217 __func__);
1218 attrs.layer_etype = LAYER_MPA | DDP_LLP;
1219 attrs.ecode = MPA_INSUFF_IRD;
1220 attrs.next_state = C4IW_QP_STATE_TERMINATE;
1221 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1222 C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
1223 err = -ENOMEM;
1224 goto out;
1225 }
1226 goto out;
1227 err:
1228 state_set(&ep->com, ABORTING);
1229 send_abort(ep, skb, GFP_KERNEL);
1230 out:
1231 connect_reply_upcall(ep, err);
1232 return;
1233 }
1234
1235 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1236 {
1237 struct mpa_message *mpa;
1238 struct mpa_v2_conn_params *mpa_v2_params;
1239 u16 plen;
1240
1241 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1242
1243 if (state_read(&ep->com) != MPA_REQ_WAIT)
1244 return;
1245
1246 /*
1247 * If we get more than the supported amount of private data
1248 * then we must fail this connection.
1249 */
1250 if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1251 stop_ep_timer(ep);
1252 abort_connection(ep, skb, GFP_KERNEL);
1253 return;
1254 }
1255
1256 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1257
1258 /*
1259 * Copy the new data into our accumulation buffer.
1260 */
1261 skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1262 skb->len);
1263 ep->mpa_pkt_len += skb->len;
1264
1265 /*
1266 * If we don't even have the mpa message, then bail.
1267 * We'll continue process when more data arrives.
1268 */
1269 if (ep->mpa_pkt_len < sizeof(*mpa))
1270 return;
1271
1272 PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
1273 stop_ep_timer(ep);
1274 mpa = (struct mpa_message *) ep->mpa_pkt;
1275
1276 /*
1277 * Validate MPA Header.
1278 */
1279 if (mpa->revision > mpa_rev) {
1280 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
1281 " Received = %d\n", __func__, mpa_rev, mpa->revision);
1282 abort_connection(ep, skb, GFP_KERNEL);
1283 return;
1284 }
1285
1286 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
1287 abort_connection(ep, skb, GFP_KERNEL);
1288 return;
1289 }
1290
1291 plen = ntohs(mpa->private_data_size);
1292
1293 /*
1294 * Fail if there's too much private data.
1295 */
1296 if (plen > MPA_MAX_PRIVATE_DATA) {
1297 abort_connection(ep, skb, GFP_KERNEL);
1298 return;
1299 }
1300
1301 /*
1302 * If plen does not account for pkt size
1303 */
1304 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1305 abort_connection(ep, skb, GFP_KERNEL);
1306 return;
1307 }
1308 ep->plen = (u8) plen;
1309
1310 /*
1311 * If we don't have all the pdata yet, then bail.
1312 */
1313 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1314 return;
1315
1316 /*
1317 * If we get here we have accumulated the entire mpa
1318 * start reply message including private data.
1319 */
1320 ep->mpa_attr.initiator = 0;
1321 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1322 ep->mpa_attr.recv_marker_enabled = markers_enabled;
1323 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1324 ep->mpa_attr.version = mpa->revision;
1325 if (mpa->revision == 1)
1326 ep->tried_with_mpa_v1 = 1;
1327 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1328
1329 if (mpa->revision == 2) {
1330 ep->mpa_attr.enhanced_rdma_conn =
1331 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1332 if (ep->mpa_attr.enhanced_rdma_conn) {
1333 mpa_v2_params = (struct mpa_v2_conn_params *)
1334 (ep->mpa_pkt + sizeof(*mpa));
1335 ep->ird = ntohs(mpa_v2_params->ird) &
1336 MPA_V2_IRD_ORD_MASK;
1337 ep->ord = ntohs(mpa_v2_params->ord) &
1338 MPA_V2_IRD_ORD_MASK;
1339 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1340 if (peer2peer) {
1341 if (ntohs(mpa_v2_params->ord) &
1342 MPA_V2_RDMA_WRITE_RTR)
1343 ep->mpa_attr.p2p_type =
1344 FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1345 else if (ntohs(mpa_v2_params->ord) &
1346 MPA_V2_RDMA_READ_RTR)
1347 ep->mpa_attr.p2p_type =
1348 FW_RI_INIT_P2PTYPE_READ_REQ;
1349 }
1350 }
1351 } else if (mpa->revision == 1)
1352 if (peer2peer)
1353 ep->mpa_attr.p2p_type = p2p_type;
1354
1355 PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
1356 "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
1357 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1358 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1359 ep->mpa_attr.p2p_type);
1360
1361 state_set(&ep->com, MPA_REQ_RCVD);
1362
1363 /* drive upcall */
1364 connect_request_upcall(ep);
1365 return;
1366 }
1367
1368 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1369 {
1370 struct c4iw_ep *ep;
1371 struct cpl_rx_data *hdr = cplhdr(skb);
1372 unsigned int dlen = ntohs(hdr->len);
1373 unsigned int tid = GET_TID(hdr);
1374 struct tid_info *t = dev->rdev.lldi.tids;
1375
1376 ep = lookup_tid(t, tid);
1377 PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
1378 skb_pull(skb, sizeof(*hdr));
1379 skb_trim(skb, dlen);
1380
1381 ep->rcv_seq += dlen;
1382 BUG_ON(ep->rcv_seq != (ntohl(hdr->seq) + dlen));
1383
1384 /* update RX credits */
1385 update_rx_credits(ep, dlen);
1386
1387 switch (state_read(&ep->com)) {
1388 case MPA_REQ_SENT:
1389 process_mpa_reply(ep, skb);
1390 break;
1391 case MPA_REQ_WAIT:
1392 process_mpa_request(ep, skb);
1393 break;
1394 case MPA_REP_SENT:
1395 break;
1396 default:
1397 printk(KERN_ERR MOD "%s Unexpected streaming data."
1398 " ep %p state %d tid %u\n",
1399 __func__, ep, state_read(&ep->com), ep->hwtid);
1400
1401 /*
1402 * The ep will timeout and inform the ULP of the failure.
1403 * See ep_timeout().
1404 */
1405 break;
1406 }
1407 return 0;
1408 }
1409
1410 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1411 {
1412 struct c4iw_ep *ep;
1413 struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
1414 int release = 0;
1415 unsigned int tid = GET_TID(rpl);
1416 struct tid_info *t = dev->rdev.lldi.tids;
1417
1418 ep = lookup_tid(t, tid);
1419 if (!ep) {
1420 printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
1421 return 0;
1422 }
1423 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1424 mutex_lock(&ep->com.mutex);
1425 switch (ep->com.state) {
1426 case ABORTING:
1427 __state_set(&ep->com, DEAD);
1428 release = 1;
1429 break;
1430 default:
1431 printk(KERN_ERR "%s ep %p state %d\n",
1432 __func__, ep, ep->com.state);
1433 break;
1434 }
1435 mutex_unlock(&ep->com.mutex);
1436
1437 if (release)
1438 release_ep_resources(ep);
1439 return 0;
1440 }
1441
1442 static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1443 {
1444 struct sk_buff *skb;
1445 struct fw_ofld_connection_wr *req;
1446 unsigned int mtu_idx;
1447 int wscale;
1448
1449 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1450 req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
1451 memset(req, 0, sizeof(*req));
1452 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
1453 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
1454 req->le.filter = cpu_to_be32(select_ntuple(ep->com.dev, ep->dst,
1455 ep->l2t));
1456 req->le.lport = ep->com.local_addr.sin_port;
1457 req->le.pport = ep->com.remote_addr.sin_port;
1458 req->le.u.ipv4.lip = ep->com.local_addr.sin_addr.s_addr;
1459 req->le.u.ipv4.pip = ep->com.remote_addr.sin_addr.s_addr;
1460 req->tcb.t_state_to_astid =
1461 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
1462 V_FW_OFLD_CONNECTION_WR_ASTID(atid));
1463 req->tcb.cplrxdataack_cplpassacceptrpl =
1464 htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
1465 req->tcb.tx_max = jiffies;
1466 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1467 wscale = compute_wscale(rcv_win);
1468 req->tcb.opt0 = TCAM_BYPASS(1) |
1469 (nocong ? NO_CONG(1) : 0) |
1470 KEEP_ALIVE(1) |
1471 DELACK(1) |
1472 WND_SCALE(wscale) |
1473 MSS_IDX(mtu_idx) |
1474 L2T_IDX(ep->l2t->idx) |
1475 TX_CHAN(ep->tx_chan) |
1476 SMAC_SEL(ep->smac_idx) |
1477 DSCP(ep->tos) |
1478 ULP_MODE(ULP_MODE_TCPDDP) |
1479 RCV_BUFSIZ(rcv_win >> 10);
1480 req->tcb.opt2 = PACE(1) |
1481 TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
1482 RX_CHANNEL(0) |
1483 CCTRL_ECN(enable_ecn) |
1484 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
1485 if (enable_tcp_timestamps)
1486 req->tcb.opt2 |= TSTAMPS_EN(1);
1487 if (enable_tcp_sack)
1488 req->tcb.opt2 |= SACK_EN(1);
1489 if (wscale && enable_tcp_window_scaling)
1490 req->tcb.opt2 |= WND_SCALE_EN(1);
1491 req->tcb.opt0 = cpu_to_be64(req->tcb.opt0);
1492 req->tcb.opt2 = cpu_to_be32(req->tcb.opt2);
1493 set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
1494 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1495 }
1496
1497 /*
1498 * Return whether a failed active open has allocated a TID
1499 */
1500 static inline int act_open_has_tid(int status)
1501 {
1502 return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
1503 status != CPL_ERR_ARP_MISS;
1504 }
1505
1506 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1507 {
1508 struct c4iw_ep *ep;
1509 struct cpl_act_open_rpl *rpl = cplhdr(skb);
1510 unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
1511 ntohl(rpl->atid_status)));
1512 struct tid_info *t = dev->rdev.lldi.tids;
1513 int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
1514
1515 ep = lookup_atid(t, atid);
1516
1517 PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
1518 status, status2errno(status));
1519
1520 if (status == CPL_ERR_RTX_NEG_ADVICE) {
1521 printk(KERN_WARNING MOD "Connection problems for atid %u\n",
1522 atid);
1523 return 0;
1524 }
1525
1526 /*
1527 * Log interesting failures.
1528 */
1529 switch (status) {
1530 case CPL_ERR_CONN_RESET:
1531 case CPL_ERR_CONN_TIMEDOUT:
1532 break;
1533 case CPL_ERR_TCAM_FULL:
1534 mutex_lock(&dev->rdev.stats.lock);
1535 dev->rdev.stats.tcam_full++;
1536 mutex_unlock(&dev->rdev.stats.lock);
1537 send_fw_act_open_req(ep,
1538 GET_TID_TID(GET_AOPEN_ATID(ntohl(rpl->atid_status))));
1539 return 0;
1540 break;
1541 default:
1542 printk(KERN_INFO MOD "Active open failure - "
1543 "atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
1544 atid, status, status2errno(status),
1545 &ep->com.local_addr.sin_addr.s_addr,
1546 ntohs(ep->com.local_addr.sin_port),
1547 &ep->com.remote_addr.sin_addr.s_addr,
1548 ntohs(ep->com.remote_addr.sin_port));
1549 break;
1550 }
1551
1552 connect_reply_upcall(ep, status2errno(status));
1553 state_set(&ep->com, DEAD);
1554
1555 if (status && act_open_has_tid(status))
1556 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
1557
1558 cxgb4_free_atid(t, atid);
1559 dst_release(ep->dst);
1560 cxgb4_l2t_release(ep->l2t);
1561 c4iw_put_ep(&ep->com);
1562
1563 return 0;
1564 }
1565
1566 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1567 {
1568 struct cpl_pass_open_rpl *rpl = cplhdr(skb);
1569 struct tid_info *t = dev->rdev.lldi.tids;
1570 unsigned int stid = GET_TID(rpl);
1571 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1572
1573 if (!ep) {
1574 PDBG("%s stid %d lookup failure!\n", __func__, stid);
1575 goto out;
1576 }
1577 PDBG("%s ep %p status %d error %d\n", __func__, ep,
1578 rpl->status, status2errno(rpl->status));
1579 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1580
1581 out:
1582 return 0;
1583 }
1584
1585 static int listen_stop(struct c4iw_listen_ep *ep)
1586 {
1587 struct sk_buff *skb;
1588 struct cpl_close_listsvr_req *req;
1589
1590 PDBG("%s ep %p\n", __func__, ep);
1591 skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1592 if (!skb) {
1593 printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
1594 return -ENOMEM;
1595 }
1596 req = (struct cpl_close_listsvr_req *) skb_put(skb, sizeof(*req));
1597 INIT_TP_WR(req, 0);
1598 OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ,
1599 ep->stid));
1600 req->reply_ctrl = cpu_to_be16(
1601 QUEUENO(ep->com.dev->rdev.lldi.rxq_ids[0]));
1602 set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
1603 return c4iw_ofld_send(&ep->com.dev->rdev, skb);
1604 }
1605
1606 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1607 {
1608 struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
1609 struct tid_info *t = dev->rdev.lldi.tids;
1610 unsigned int stid = GET_TID(rpl);
1611 struct c4iw_listen_ep *ep = lookup_stid(t, stid);
1612
1613 PDBG("%s ep %p\n", __func__, ep);
1614 c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
1615 return 0;
1616 }
1617
1618 static void accept_cr(struct c4iw_ep *ep, __be32 peer_ip, struct sk_buff *skb,
1619 struct cpl_pass_accept_req *req)
1620 {
1621 struct cpl_pass_accept_rpl *rpl;
1622 unsigned int mtu_idx;
1623 u64 opt0;
1624 u32 opt2;
1625 int wscale;
1626
1627 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1628 BUG_ON(skb_cloned(skb));
1629 skb_trim(skb, sizeof(*rpl));
1630 skb_get(skb);
1631 cxgb4_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx);
1632 wscale = compute_wscale(rcv_win);
1633 opt0 = (nocong ? NO_CONG(1) : 0) |
1634 KEEP_ALIVE(1) |
1635 DELACK(1) |
1636 WND_SCALE(wscale) |
1637 MSS_IDX(mtu_idx) |
1638 L2T_IDX(ep->l2t->idx) |
1639 TX_CHAN(ep->tx_chan) |
1640 SMAC_SEL(ep->smac_idx) |
1641 DSCP(ep->tos >> 2) |
1642 ULP_MODE(ULP_MODE_TCPDDP) |
1643 RCV_BUFSIZ(rcv_win>>10);
1644 opt2 = RX_CHANNEL(0) |
1645 RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
1646
1647 if (enable_tcp_timestamps && req->tcpopt.tstamp)
1648 opt2 |= TSTAMPS_EN(1);
1649 if (enable_tcp_sack && req->tcpopt.sack)
1650 opt2 |= SACK_EN(1);
1651 if (wscale && enable_tcp_window_scaling)
1652 opt2 |= WND_SCALE_EN(1);
1653 if (enable_ecn) {
1654 const struct tcphdr *tcph;
1655 u32 hlen = ntohl(req->hdr_len);
1656
1657 tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
1658 G_IP_HDR_LEN(hlen);
1659 if (tcph->ece && tcph->cwr)
1660 opt2 |= CCTRL_ECN(1);
1661 }
1662
1663 rpl = cplhdr(skb);
1664 INIT_TP_WR(rpl, ep->hwtid);
1665 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1666 ep->hwtid));
1667 rpl->opt0 = cpu_to_be64(opt0);
1668 rpl->opt2 = cpu_to_be32(opt2);
1669 set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
1670 c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1671
1672 return;
1673 }
1674
1675 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, __be32 peer_ip,
1676 struct sk_buff *skb)
1677 {
1678 PDBG("%s c4iw_dev %p tid %u peer_ip %x\n", __func__, dev, hwtid,
1679 peer_ip);
1680 BUG_ON(skb_cloned(skb));
1681 skb_trim(skb, sizeof(struct cpl_tid_release));
1682 skb_get(skb);
1683 release_tid(&dev->rdev, hwtid, skb);
1684 return;
1685 }
1686
1687 static void get_4tuple(struct cpl_pass_accept_req *req,
1688 __be32 *local_ip, __be32 *peer_ip,
1689 __be16 *local_port, __be16 *peer_port)
1690 {
1691 int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
1692 int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
1693 struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
1694 struct tcphdr *tcp = (struct tcphdr *)
1695 ((u8 *)(req + 1) + eth_len + ip_len);
1696
1697 PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
1698 ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
1699 ntohs(tcp->dest));
1700
1701 *peer_ip = ip->saddr;
1702 *local_ip = ip->daddr;
1703 *peer_port = tcp->source;
1704 *local_port = tcp->dest;
1705
1706 return;
1707 }
1708
1709 static int import_ep(struct c4iw_ep *ep, __be32 peer_ip, struct dst_entry *dst,
1710 struct c4iw_dev *cdev, bool clear_mpa_v1)
1711 {
1712 struct neighbour *n;
1713 int err, step;
1714
1715 n = dst_neigh_lookup(dst, &peer_ip);
1716 if (!n)
1717 return -ENODEV;
1718
1719 rcu_read_lock();
1720 err = -ENOMEM;
1721 if (n->dev->flags & IFF_LOOPBACK) {
1722 struct net_device *pdev;
1723
1724 pdev = ip_dev_find(&init_net, peer_ip);
1725 if (!pdev) {
1726 err = -ENODEV;
1727 goto out;
1728 }
1729 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1730 n, pdev, 0);
1731 if (!ep->l2t)
1732 goto out;
1733 ep->mtu = pdev->mtu;
1734 ep->tx_chan = cxgb4_port_chan(pdev);
1735 ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
1736 step = cdev->rdev.lldi.ntxq /
1737 cdev->rdev.lldi.nchan;
1738 ep->txq_idx = cxgb4_port_idx(pdev) * step;
1739 step = cdev->rdev.lldi.nrxq /
1740 cdev->rdev.lldi.nchan;
1741 ep->ctrlq_idx = cxgb4_port_idx(pdev);
1742 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1743 cxgb4_port_idx(pdev) * step];
1744 dev_put(pdev);
1745 } else {
1746 ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
1747 n, n->dev, 0);
1748 if (!ep->l2t)
1749 goto out;
1750 ep->mtu = dst_mtu(dst);
1751 ep->tx_chan = cxgb4_port_chan(n->dev);
1752 ep->smac_idx = (cxgb4_port_viid(n->dev) & 0x7F) << 1;
1753 step = cdev->rdev.lldi.ntxq /
1754 cdev->rdev.lldi.nchan;
1755 ep->txq_idx = cxgb4_port_idx(n->dev) * step;
1756 ep->ctrlq_idx = cxgb4_port_idx(n->dev);
1757 step = cdev->rdev.lldi.nrxq /
1758 cdev->rdev.lldi.nchan;
1759 ep->rss_qid = cdev->rdev.lldi.rxq_ids[
1760 cxgb4_port_idx(n->dev) * step];
1761
1762 if (clear_mpa_v1) {
1763 ep->retry_with_mpa_v1 = 0;
1764 ep->tried_with_mpa_v1 = 0;
1765 }
1766 }
1767 err = 0;
1768 out:
1769 rcu_read_unlock();
1770
1771 neigh_release(n);
1772
1773 return err;
1774 }
1775
1776 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
1777 {
1778 struct c4iw_ep *child_ep, *parent_ep;
1779 struct cpl_pass_accept_req *req = cplhdr(skb);
1780 unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
1781 struct tid_info *t = dev->rdev.lldi.tids;
1782 unsigned int hwtid = GET_TID(req);
1783 struct dst_entry *dst;
1784 struct rtable *rt;
1785 __be32 local_ip, peer_ip = 0;
1786 __be16 local_port, peer_port;
1787 int err;
1788 u16 peer_mss = ntohs(req->tcpopt.mss);
1789
1790 parent_ep = lookup_stid(t, stid);
1791 if (!parent_ep) {
1792 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
1793 goto reject;
1794 }
1795 get_4tuple(req, &local_ip, &peer_ip, &local_port, &peer_port);
1796
1797 PDBG("%s parent ep %p hwtid %u laddr 0x%x raddr 0x%x lport %d " \
1798 "rport %d peer_mss %d\n", __func__, parent_ep, hwtid,
1799 ntohl(local_ip), ntohl(peer_ip), ntohs(local_port),
1800 ntohs(peer_port), peer_mss);
1801
1802 if (state_read(&parent_ep->com) != LISTEN) {
1803 printk(KERN_ERR "%s - listening ep not in LISTEN\n",
1804 __func__);
1805 goto reject;
1806 }
1807
1808 /* Find output route */
1809 rt = find_route(dev, local_ip, peer_ip, local_port, peer_port,
1810 GET_POPEN_TOS(ntohl(req->tos_stid)));
1811 if (!rt) {
1812 printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
1813 __func__);
1814 goto reject;
1815 }
1816 dst = &rt->dst;
1817
1818 child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
1819 if (!child_ep) {
1820 printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
1821 __func__);
1822 dst_release(dst);
1823 goto reject;
1824 }
1825
1826 err = import_ep(child_ep, peer_ip, dst, dev, false);
1827 if (err) {
1828 printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
1829 __func__);
1830 dst_release(dst);
1831 kfree(child_ep);
1832 goto reject;
1833 }
1834
1835 if (peer_mss && child_ep->mtu > (peer_mss + 40))
1836 child_ep->mtu = peer_mss + 40;
1837
1838 state_set(&child_ep->com, CONNECTING);
1839 child_ep->com.dev = dev;
1840 child_ep->com.cm_id = NULL;
1841 child_ep->com.local_addr.sin_family = PF_INET;
1842 child_ep->com.local_addr.sin_port = local_port;
1843 child_ep->com.local_addr.sin_addr.s_addr = local_ip;
1844 child_ep->com.remote_addr.sin_family = PF_INET;
1845 child_ep->com.remote_addr.sin_port = peer_port;
1846 child_ep->com.remote_addr.sin_addr.s_addr = peer_ip;
1847 c4iw_get_ep(&parent_ep->com);
1848 child_ep->parent_ep = parent_ep;
1849 child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
1850 child_ep->dst = dst;
1851 child_ep->hwtid = hwtid;
1852
1853 PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
1854 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
1855
1856 init_timer(&child_ep->timer);
1857 cxgb4_insert_tid(t, child_ep, hwtid);
1858 accept_cr(child_ep, peer_ip, skb, req);
1859 goto out;
1860 reject:
1861 reject_cr(dev, hwtid, peer_ip, skb);
1862 out:
1863 return 0;
1864 }
1865
1866 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1867 {
1868 struct c4iw_ep *ep;
1869 struct cpl_pass_establish *req = cplhdr(skb);
1870 struct tid_info *t = dev->rdev.lldi.tids;
1871 unsigned int tid = GET_TID(req);
1872
1873 ep = lookup_tid(t, tid);
1874 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1875 ep->snd_seq = be32_to_cpu(req->snd_isn);
1876 ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1877
1878 PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
1879 ntohs(req->tcp_opt));
1880
1881 set_emss(ep, ntohs(req->tcp_opt));
1882
1883 dst_confirm(ep->dst);
1884 state_set(&ep->com, MPA_REQ_WAIT);
1885 start_ep_timer(ep);
1886 send_flowc(ep, skb);
1887
1888 return 0;
1889 }
1890
1891 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
1892 {
1893 struct cpl_peer_close *hdr = cplhdr(skb);
1894 struct c4iw_ep *ep;
1895 struct c4iw_qp_attributes attrs;
1896 int disconnect = 1;
1897 int release = 0;
1898 struct tid_info *t = dev->rdev.lldi.tids;
1899 unsigned int tid = GET_TID(hdr);
1900 int ret;
1901
1902 ep = lookup_tid(t, tid);
1903 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
1904 dst_confirm(ep->dst);
1905
1906 mutex_lock(&ep->com.mutex);
1907 switch (ep->com.state) {
1908 case MPA_REQ_WAIT:
1909 __state_set(&ep->com, CLOSING);
1910 break;
1911 case MPA_REQ_SENT:
1912 __state_set(&ep->com, CLOSING);
1913 connect_reply_upcall(ep, -ECONNRESET);
1914 break;
1915 case MPA_REQ_RCVD:
1916
1917 /*
1918 * We're gonna mark this puppy DEAD, but keep
1919 * the reference on it until the ULP accepts or
1920 * rejects the CR. Also wake up anyone waiting
1921 * in rdma connection migration (see c4iw_accept_cr()).
1922 */
1923 __state_set(&ep->com, CLOSING);
1924 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
1925 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1926 break;
1927 case MPA_REP_SENT:
1928 __state_set(&ep->com, CLOSING);
1929 PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
1930 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1931 break;
1932 case FPDU_MODE:
1933 start_ep_timer(ep);
1934 __state_set(&ep->com, CLOSING);
1935 attrs.next_state = C4IW_QP_STATE_CLOSING;
1936 ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1937 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1938 if (ret != -ECONNRESET) {
1939 peer_close_upcall(ep);
1940 disconnect = 1;
1941 }
1942 break;
1943 case ABORTING:
1944 disconnect = 0;
1945 break;
1946 case CLOSING:
1947 __state_set(&ep->com, MORIBUND);
1948 disconnect = 0;
1949 break;
1950 case MORIBUND:
1951 stop_ep_timer(ep);
1952 if (ep->com.cm_id && ep->com.qp) {
1953 attrs.next_state = C4IW_QP_STATE_IDLE;
1954 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1955 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1956 }
1957 close_complete_upcall(ep);
1958 __state_set(&ep->com, DEAD);
1959 release = 1;
1960 disconnect = 0;
1961 break;
1962 case DEAD:
1963 disconnect = 0;
1964 break;
1965 default:
1966 BUG_ON(1);
1967 }
1968 mutex_unlock(&ep->com.mutex);
1969 if (disconnect)
1970 c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1971 if (release)
1972 release_ep_resources(ep);
1973 return 0;
1974 }
1975
1976 /*
1977 * Returns whether an ABORT_REQ_RSS message is a negative advice.
1978 */
1979 static int is_neg_adv_abort(unsigned int status)
1980 {
1981 return status == CPL_ERR_RTX_NEG_ADVICE ||
1982 status == CPL_ERR_PERSIST_NEG_ADVICE;
1983 }
1984
1985 static int c4iw_reconnect(struct c4iw_ep *ep)
1986 {
1987 struct rtable *rt;
1988 int err = 0;
1989
1990 PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
1991 init_timer(&ep->timer);
1992
1993 /*
1994 * Allocate an active TID to initiate a TCP connection.
1995 */
1996 ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
1997 if (ep->atid == -1) {
1998 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
1999 err = -ENOMEM;
2000 goto fail2;
2001 }
2002
2003 /* find a route */
2004 rt = find_route(ep->com.dev,
2005 ep->com.cm_id->local_addr.sin_addr.s_addr,
2006 ep->com.cm_id->remote_addr.sin_addr.s_addr,
2007 ep->com.cm_id->local_addr.sin_port,
2008 ep->com.cm_id->remote_addr.sin_port, 0);
2009 if (!rt) {
2010 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2011 err = -EHOSTUNREACH;
2012 goto fail3;
2013 }
2014 ep->dst = &rt->dst;
2015
2016 err = import_ep(ep, ep->com.cm_id->remote_addr.sin_addr.s_addr,
2017 ep->dst, ep->com.dev, false);
2018 if (err) {
2019 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
2020 goto fail4;
2021 }
2022
2023 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2024 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2025 ep->l2t->idx);
2026
2027 state_set(&ep->com, CONNECTING);
2028 ep->tos = 0;
2029
2030 /* send connect request to rnic */
2031 err = send_connect(ep);
2032 if (!err)
2033 goto out;
2034
2035 cxgb4_l2t_release(ep->l2t);
2036 fail4:
2037 dst_release(ep->dst);
2038 fail3:
2039 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2040 fail2:
2041 /*
2042 * remember to send notification to upper layer.
2043 * We are in here so the upper layer is not aware that this is
2044 * re-connect attempt and so, upper layer is still waiting for
2045 * response of 1st connect request.
2046 */
2047 connect_reply_upcall(ep, -ECONNRESET);
2048 c4iw_put_ep(&ep->com);
2049 out:
2050 return err;
2051 }
2052
2053 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2054 {
2055 struct cpl_abort_req_rss *req = cplhdr(skb);
2056 struct c4iw_ep *ep;
2057 struct cpl_abort_rpl *rpl;
2058 struct sk_buff *rpl_skb;
2059 struct c4iw_qp_attributes attrs;
2060 int ret;
2061 int release = 0;
2062 struct tid_info *t = dev->rdev.lldi.tids;
2063 unsigned int tid = GET_TID(req);
2064
2065 ep = lookup_tid(t, tid);
2066 if (is_neg_adv_abort(req->status)) {
2067 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
2068 ep->hwtid);
2069 return 0;
2070 }
2071 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
2072 ep->com.state);
2073
2074 /*
2075 * Wake up any threads in rdma_init() or rdma_fini().
2076 * However, this is not needed if com state is just
2077 * MPA_REQ_SENT
2078 */
2079 if (ep->com.state != MPA_REQ_SENT)
2080 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
2081
2082 mutex_lock(&ep->com.mutex);
2083 switch (ep->com.state) {
2084 case CONNECTING:
2085 break;
2086 case MPA_REQ_WAIT:
2087 stop_ep_timer(ep);
2088 break;
2089 case MPA_REQ_SENT:
2090 stop_ep_timer(ep);
2091 if (mpa_rev == 2 && ep->tried_with_mpa_v1)
2092 connect_reply_upcall(ep, -ECONNRESET);
2093 else {
2094 /*
2095 * we just don't send notification upwards because we
2096 * want to retry with mpa_v1 without upper layers even
2097 * knowing it.
2098 *
2099 * do some housekeeping so as to re-initiate the
2100 * connection
2101 */
2102 PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
2103 mpa_rev);
2104 ep->retry_with_mpa_v1 = 1;
2105 }
2106 break;
2107 case MPA_REP_SENT:
2108 break;
2109 case MPA_REQ_RCVD:
2110 break;
2111 case MORIBUND:
2112 case CLOSING:
2113 stop_ep_timer(ep);
2114 /*FALLTHROUGH*/
2115 case FPDU_MODE:
2116 if (ep->com.cm_id && ep->com.qp) {
2117 attrs.next_state = C4IW_QP_STATE_ERROR;
2118 ret = c4iw_modify_qp(ep->com.qp->rhp,
2119 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2120 &attrs, 1);
2121 if (ret)
2122 printk(KERN_ERR MOD
2123 "%s - qp <- error failed!\n",
2124 __func__);
2125 }
2126 peer_abort_upcall(ep);
2127 break;
2128 case ABORTING:
2129 break;
2130 case DEAD:
2131 PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2132 mutex_unlock(&ep->com.mutex);
2133 return 0;
2134 default:
2135 BUG_ON(1);
2136 break;
2137 }
2138 dst_confirm(ep->dst);
2139 if (ep->com.state != ABORTING) {
2140 __state_set(&ep->com, DEAD);
2141 /* we don't release if we want to retry with mpa_v1 */
2142 if (!ep->retry_with_mpa_v1)
2143 release = 1;
2144 }
2145 mutex_unlock(&ep->com.mutex);
2146
2147 rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
2148 if (!rpl_skb) {
2149 printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
2150 __func__);
2151 release = 1;
2152 goto out;
2153 }
2154 set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
2155 rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
2156 INIT_TP_WR(rpl, ep->hwtid);
2157 OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
2158 rpl->cmd = CPL_ABORT_NO_RST;
2159 c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2160 out:
2161 if (release)
2162 release_ep_resources(ep);
2163
2164 /* retry with mpa-v1 */
2165 if (ep && ep->retry_with_mpa_v1) {
2166 cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
2167 dst_release(ep->dst);
2168 cxgb4_l2t_release(ep->l2t);
2169 c4iw_reconnect(ep);
2170 }
2171
2172 return 0;
2173 }
2174
2175 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2176 {
2177 struct c4iw_ep *ep;
2178 struct c4iw_qp_attributes attrs;
2179 struct cpl_close_con_rpl *rpl = cplhdr(skb);
2180 int release = 0;
2181 struct tid_info *t = dev->rdev.lldi.tids;
2182 unsigned int tid = GET_TID(rpl);
2183
2184 ep = lookup_tid(t, tid);
2185
2186 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2187 BUG_ON(!ep);
2188
2189 /* The cm_id may be null if we failed to connect */
2190 mutex_lock(&ep->com.mutex);
2191 switch (ep->com.state) {
2192 case CLOSING:
2193 __state_set(&ep->com, MORIBUND);
2194 break;
2195 case MORIBUND:
2196 stop_ep_timer(ep);
2197 if ((ep->com.cm_id) && (ep->com.qp)) {
2198 attrs.next_state = C4IW_QP_STATE_IDLE;
2199 c4iw_modify_qp(ep->com.qp->rhp,
2200 ep->com.qp,
2201 C4IW_QP_ATTR_NEXT_STATE,
2202 &attrs, 1);
2203 }
2204 close_complete_upcall(ep);
2205 __state_set(&ep->com, DEAD);
2206 release = 1;
2207 break;
2208 case ABORTING:
2209 case DEAD:
2210 break;
2211 default:
2212 BUG_ON(1);
2213 break;
2214 }
2215 mutex_unlock(&ep->com.mutex);
2216 if (release)
2217 release_ep_resources(ep);
2218 return 0;
2219 }
2220
2221 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
2222 {
2223 struct cpl_rdma_terminate *rpl = cplhdr(skb);
2224 struct tid_info *t = dev->rdev.lldi.tids;
2225 unsigned int tid = GET_TID(rpl);
2226 struct c4iw_ep *ep;
2227 struct c4iw_qp_attributes attrs;
2228
2229 ep = lookup_tid(t, tid);
2230 BUG_ON(!ep);
2231
2232 if (ep && ep->com.qp) {
2233 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
2234 ep->com.qp->wq.sq.qid);
2235 attrs.next_state = C4IW_QP_STATE_TERMINATE;
2236 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2237 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2238 } else
2239 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
2240
2241 return 0;
2242 }
2243
2244 /*
2245 * Upcall from the adapter indicating data has been transmitted.
2246 * For us its just the single MPA request or reply. We can now free
2247 * the skb holding the mpa message.
2248 */
2249 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
2250 {
2251 struct c4iw_ep *ep;
2252 struct cpl_fw4_ack *hdr = cplhdr(skb);
2253 u8 credits = hdr->credits;
2254 unsigned int tid = GET_TID(hdr);
2255 struct tid_info *t = dev->rdev.lldi.tids;
2256
2257
2258 ep = lookup_tid(t, tid);
2259 PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
2260 if (credits == 0) {
2261 PDBG("%s 0 credit ack ep %p tid %u state %u\n",
2262 __func__, ep, ep->hwtid, state_read(&ep->com));
2263 return 0;
2264 }
2265
2266 dst_confirm(ep->dst);
2267 if (ep->mpa_skb) {
2268 PDBG("%s last streaming msg ack ep %p tid %u state %u "
2269 "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
2270 state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
2271 kfree_skb(ep->mpa_skb);
2272 ep->mpa_skb = NULL;
2273 }
2274 return 0;
2275 }
2276
2277 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2278 {
2279 int err;
2280 struct c4iw_ep *ep = to_ep(cm_id);
2281 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2282
2283 if (state_read(&ep->com) == DEAD) {
2284 c4iw_put_ep(&ep->com);
2285 return -ECONNRESET;
2286 }
2287 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2288 if (mpa_rev == 0)
2289 abort_connection(ep, NULL, GFP_KERNEL);
2290 else {
2291 err = send_mpa_reject(ep, pdata, pdata_len);
2292 err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2293 }
2294 c4iw_put_ep(&ep->com);
2295 return 0;
2296 }
2297
2298 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2299 {
2300 int err;
2301 struct c4iw_qp_attributes attrs;
2302 enum c4iw_qp_attr_mask mask;
2303 struct c4iw_ep *ep = to_ep(cm_id);
2304 struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2305 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2306
2307 PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
2308 if (state_read(&ep->com) == DEAD) {
2309 err = -ECONNRESET;
2310 goto err;
2311 }
2312
2313 BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
2314 BUG_ON(!qp);
2315
2316 if ((conn_param->ord > c4iw_max_read_depth) ||
2317 (conn_param->ird > c4iw_max_read_depth)) {
2318 abort_connection(ep, NULL, GFP_KERNEL);
2319 err = -EINVAL;
2320 goto err;
2321 }
2322
2323 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2324 if (conn_param->ord > ep->ird) {
2325 ep->ird = conn_param->ird;
2326 ep->ord = conn_param->ord;
2327 send_mpa_reject(ep, conn_param->private_data,
2328 conn_param->private_data_len);
2329 abort_connection(ep, NULL, GFP_KERNEL);
2330 err = -ENOMEM;
2331 goto err;
2332 }
2333 if (conn_param->ird > ep->ord) {
2334 if (!ep->ord)
2335 conn_param->ird = 1;
2336 else {
2337 abort_connection(ep, NULL, GFP_KERNEL);
2338 err = -ENOMEM;
2339 goto err;
2340 }
2341 }
2342
2343 }
2344 ep->ird = conn_param->ird;
2345 ep->ord = conn_param->ord;
2346
2347 if (ep->mpa_attr.version != 2)
2348 if (peer2peer && ep->ird == 0)
2349 ep->ird = 1;
2350
2351 PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
2352
2353 cm_id->add_ref(cm_id);
2354 ep->com.cm_id = cm_id;
2355 ep->com.qp = qp;
2356
2357 /* bind QP to EP and move to RTS */
2358 attrs.mpa_attr = ep->mpa_attr;
2359 attrs.max_ird = ep->ird;
2360 attrs.max_ord = ep->ord;
2361 attrs.llp_stream_handle = ep;
2362 attrs.next_state = C4IW_QP_STATE_RTS;
2363
2364 /* bind QP and TID with INIT_WR */
2365 mask = C4IW_QP_ATTR_NEXT_STATE |
2366 C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2367 C4IW_QP_ATTR_MPA_ATTR |
2368 C4IW_QP_ATTR_MAX_IRD |
2369 C4IW_QP_ATTR_MAX_ORD;
2370
2371 err = c4iw_modify_qp(ep->com.qp->rhp,
2372 ep->com.qp, mask, &attrs, 1);
2373 if (err)
2374 goto err1;
2375 err = send_mpa_reply(ep, conn_param->private_data,
2376 conn_param->private_data_len);
2377 if (err)
2378 goto err1;
2379
2380 state_set(&ep->com, FPDU_MODE);
2381 established_upcall(ep);
2382 c4iw_put_ep(&ep->com);
2383 return 0;
2384 err1:
2385 ep->com.cm_id = NULL;
2386 ep->com.qp = NULL;
2387 cm_id->rem_ref(cm_id);
2388 err:
2389 c4iw_put_ep(&ep->com);
2390 return err;
2391 }
2392
2393 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2394 {
2395 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2396 struct c4iw_ep *ep;
2397 struct rtable *rt;
2398 int err = 0;
2399
2400 if ((conn_param->ord > c4iw_max_read_depth) ||
2401 (conn_param->ird > c4iw_max_read_depth)) {
2402 err = -EINVAL;
2403 goto out;
2404 }
2405 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2406 if (!ep) {
2407 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2408 err = -ENOMEM;
2409 goto out;
2410 }
2411 init_timer(&ep->timer);
2412 ep->plen = conn_param->private_data_len;
2413 if (ep->plen)
2414 memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2415 conn_param->private_data, ep->plen);
2416 ep->ird = conn_param->ird;
2417 ep->ord = conn_param->ord;
2418
2419 if (peer2peer && ep->ord == 0)
2420 ep->ord = 1;
2421
2422 cm_id->add_ref(cm_id);
2423 ep->com.dev = dev;
2424 ep->com.cm_id = cm_id;
2425 ep->com.qp = get_qhp(dev, conn_param->qpn);
2426 BUG_ON(!ep->com.qp);
2427 PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
2428 ep->com.qp, cm_id);
2429
2430 /*
2431 * Allocate an active TID to initiate a TCP connection.
2432 */
2433 ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
2434 if (ep->atid == -1) {
2435 printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
2436 err = -ENOMEM;
2437 goto fail2;
2438 }
2439
2440 PDBG("%s saddr 0x%x sport 0x%x raddr 0x%x rport 0x%x\n", __func__,
2441 ntohl(cm_id->local_addr.sin_addr.s_addr),
2442 ntohs(cm_id->local_addr.sin_port),
2443 ntohl(cm_id->remote_addr.sin_addr.s_addr),
2444 ntohs(cm_id->remote_addr.sin_port));
2445
2446 /* find a route */
2447 rt = find_route(dev,
2448 cm_id->local_addr.sin_addr.s_addr,
2449 cm_id->remote_addr.sin_addr.s_addr,
2450 cm_id->local_addr.sin_port,
2451 cm_id->remote_addr.sin_port, 0);
2452 if (!rt) {
2453 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2454 err = -EHOSTUNREACH;
2455 goto fail3;
2456 }
2457 ep->dst = &rt->dst;
2458
2459 err = import_ep(ep, cm_id->remote_addr.sin_addr.s_addr,
2460 ep->dst, ep->com.dev, true);
2461 if (err) {
2462 printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
2463 goto fail4;
2464 }
2465
2466 PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2467 __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2468 ep->l2t->idx);
2469
2470 state_set(&ep->com, CONNECTING);
2471 ep->tos = 0;
2472 ep->com.local_addr = cm_id->local_addr;
2473 ep->com.remote_addr = cm_id->remote_addr;
2474
2475 /* send connect request to rnic */
2476 err = send_connect(ep);
2477 if (!err)
2478 goto out;
2479
2480 cxgb4_l2t_release(ep->l2t);
2481 fail4:
2482 dst_release(ep->dst);
2483 fail3:
2484 cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2485 fail2:
2486 cm_id->rem_ref(cm_id);
2487 c4iw_put_ep(&ep->com);
2488 out:
2489 return err;
2490 }
2491
2492 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2493 {
2494 int err = 0;
2495 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2496 struct c4iw_listen_ep *ep;
2497
2498 might_sleep();
2499
2500 ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2501 if (!ep) {
2502 printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
2503 err = -ENOMEM;
2504 goto fail1;
2505 }
2506 PDBG("%s ep %p\n", __func__, ep);
2507 cm_id->add_ref(cm_id);
2508 ep->com.cm_id = cm_id;
2509 ep->com.dev = dev;
2510 ep->backlog = backlog;
2511 ep->com.local_addr = cm_id->local_addr;
2512
2513 /*
2514 * Allocate a server TID.
2515 */
2516 if (dev->rdev.lldi.enable_fw_ofld_conn)
2517 ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids, PF_INET, ep);
2518 else
2519 ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids, PF_INET, ep);
2520
2521 if (ep->stid == -1) {
2522 printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
2523 err = -ENOMEM;
2524 goto fail2;
2525 }
2526 state_set(&ep->com, LISTEN);
2527 if (dev->rdev.lldi.enable_fw_ofld_conn) {
2528 do {
2529 err = cxgb4_create_server_filter(
2530 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2531 ep->com.local_addr.sin_addr.s_addr,
2532 ep->com.local_addr.sin_port,
2533 ep->com.dev->rdev.lldi.rxq_ids[0]);
2534 if (err == -EBUSY) {
2535 set_current_state(TASK_UNINTERRUPTIBLE);
2536 schedule_timeout(usecs_to_jiffies(100));
2537 }
2538 } while (err == -EBUSY);
2539 } else {
2540 c4iw_init_wr_wait(&ep->com.wr_wait);
2541 err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
2542 ep->stid, ep->com.local_addr.sin_addr.s_addr,
2543 ep->com.local_addr.sin_port,
2544 ep->com.dev->rdev.lldi.rxq_ids[0]);
2545 if (!err)
2546 err = c4iw_wait_for_reply(&ep->com.dev->rdev,
2547 &ep->com.wr_wait,
2548 0, 0, __func__);
2549 }
2550 if (!err) {
2551 cm_id->provider_data = ep;
2552 goto out;
2553 }
2554 pr_err("%s cxgb4_create_server/filter failed err %d " \
2555 "stid %d laddr %08x lport %d\n", \
2556 __func__, err, ep->stid,
2557 ntohl(ep->com.local_addr.sin_addr.s_addr),
2558 ntohs(ep->com.local_addr.sin_port));
2559 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2560 fail2:
2561 cm_id->rem_ref(cm_id);
2562 c4iw_put_ep(&ep->com);
2563 fail1:
2564 out:
2565 return err;
2566 }
2567
2568 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
2569 {
2570 int err;
2571 struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
2572
2573 PDBG("%s ep %p\n", __func__, ep);
2574
2575 might_sleep();
2576 state_set(&ep->com, DEAD);
2577 if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn) {
2578 err = cxgb4_remove_server_filter(
2579 ep->com.dev->rdev.lldi.ports[0], ep->stid,
2580 ep->com.dev->rdev.lldi.rxq_ids[0], 0);
2581 } else {
2582 c4iw_init_wr_wait(&ep->com.wr_wait);
2583 err = listen_stop(ep);
2584 if (err)
2585 goto done;
2586 err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
2587 0, 0, __func__);
2588 }
2589 cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid, PF_INET);
2590 done:
2591 cm_id->rem_ref(cm_id);
2592 c4iw_put_ep(&ep->com);
2593 return err;
2594 }
2595
2596 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2597 {
2598 int ret = 0;
2599 int close = 0;
2600 int fatal = 0;
2601 struct c4iw_rdev *rdev;
2602
2603 mutex_lock(&ep->com.mutex);
2604
2605 PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
2606 states[ep->com.state], abrupt);
2607
2608 rdev = &ep->com.dev->rdev;
2609 if (c4iw_fatal_error(rdev)) {
2610 fatal = 1;
2611 close_complete_upcall(ep);
2612 ep->com.state = DEAD;
2613 }
2614 switch (ep->com.state) {
2615 case MPA_REQ_WAIT:
2616 case MPA_REQ_SENT:
2617 case MPA_REQ_RCVD:
2618 case MPA_REP_SENT:
2619 case FPDU_MODE:
2620 close = 1;
2621 if (abrupt)
2622 ep->com.state = ABORTING;
2623 else {
2624 ep->com.state = CLOSING;
2625 start_ep_timer(ep);
2626 }
2627 set_bit(CLOSE_SENT, &ep->com.flags);
2628 break;
2629 case CLOSING:
2630 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2631 close = 1;
2632 if (abrupt) {
2633 stop_ep_timer(ep);
2634 ep->com.state = ABORTING;
2635 } else
2636 ep->com.state = MORIBUND;
2637 }
2638 break;
2639 case MORIBUND:
2640 case ABORTING:
2641 case DEAD:
2642 PDBG("%s ignoring disconnect ep %p state %u\n",
2643 __func__, ep, ep->com.state);
2644 break;
2645 default:
2646 BUG();
2647 break;
2648 }
2649
2650 if (close) {
2651 if (abrupt) {
2652 close_complete_upcall(ep);
2653 ret = send_abort(ep, NULL, gfp);
2654 } else
2655 ret = send_halfclose(ep, gfp);
2656 if (ret)
2657 fatal = 1;
2658 }
2659 mutex_unlock(&ep->com.mutex);
2660 if (fatal)
2661 release_ep_resources(ep);
2662 return ret;
2663 }
2664
2665 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
2666 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
2667 {
2668 struct c4iw_ep *ep;
2669
2670 ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids, req->tid);
2671 if (!ep)
2672 return;
2673
2674 switch (req->retval) {
2675 case FW_ENOMEM:
2676 case FW_EADDRINUSE:
2677 PDBG("%s ofld conn wr ret %d\n", __func__, req->retval);
2678 break;
2679 default:
2680 pr_info("%s unexpected ofld conn wr retval %d\n",
2681 __func__, req->retval);
2682 break;
2683 }
2684 connect_reply_upcall(ep, status2errno(req->retval));
2685 }
2686
2687 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
2688 struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
2689 {
2690 struct sk_buff *rpl_skb;
2691 struct cpl_pass_accept_req *cpl;
2692 int ret;
2693
2694 rpl_skb = (struct sk_buff *)cpu_to_be64(req->cookie);
2695 BUG_ON(!rpl_skb);
2696 if (req->retval) {
2697 PDBG("%s passive open failure %d\n", __func__, req->retval);
2698 kfree_skb(rpl_skb);
2699 } else {
2700 cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
2701 OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
2702 htonl(req->tid)));
2703 ret = pass_accept_req(dev, rpl_skb);
2704 if (!ret)
2705 kfree_skb(rpl_skb);
2706 }
2707 return;
2708 }
2709
2710 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
2711 {
2712 struct cpl_fw6_msg *rpl = cplhdr(skb);
2713 struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
2714
2715 switch (rpl->type) {
2716 case FW6_TYPE_CQE:
2717 c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
2718 break;
2719 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
2720 req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
2721 switch (req->t_state) {
2722 case TCP_SYN_SENT:
2723 active_ofld_conn_reply(dev, skb, req);
2724 break;
2725 case TCP_SYN_RECV:
2726 passive_ofld_conn_reply(dev, skb, req);
2727 break;
2728 default:
2729 pr_err("%s unexpected ofld conn wr state %d\n",
2730 __func__, req->t_state);
2731 break;
2732 }
2733 break;
2734 }
2735 return 0;
2736 }
2737
2738 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
2739 {
2740 u32 l2info;
2741 u16 vlantag, len, hdr_len;
2742 u8 intf;
2743 struct cpl_rx_pkt *cpl = cplhdr(skb);
2744 struct cpl_pass_accept_req *req;
2745 struct tcp_options_received tmp_opt;
2746
2747 /* Store values from cpl_rx_pkt in temporary location. */
2748 vlantag = cpl->vlan;
2749 len = cpl->len;
2750 l2info = cpl->l2info;
2751 hdr_len = cpl->hdr_len;
2752 intf = cpl->iff;
2753
2754 __skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
2755
2756 /*
2757 * We need to parse the TCP options from SYN packet.
2758 * to generate cpl_pass_accept_req.
2759 */
2760 memset(&tmp_opt, 0, sizeof(tmp_opt));
2761 tcp_clear_options(&tmp_opt);
2762 tcp_parse_options(skb, &tmp_opt, 0, 0, NULL);
2763
2764 req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
2765 memset(req, 0, sizeof(*req));
2766 req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
2767 V_SYN_MAC_IDX(G_RX_MACIDX(htonl(l2info))) |
2768 F_SYN_XACT_MATCH);
2769 req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(htonl(l2info))) |
2770 V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(htons(hdr_len))) |
2771 V_IP_HDR_LEN(G_RX_IPHDR_LEN(htons(hdr_len))) |
2772 V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(htonl(l2info))));
2773 req->vlan = vlantag;
2774 req->len = len;
2775 req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
2776 PASS_OPEN_TOS(tos));
2777 req->tcpopt.mss = htons(tmp_opt.mss_clamp);
2778 if (tmp_opt.wscale_ok)
2779 req->tcpopt.wsf = tmp_opt.snd_wscale;
2780 req->tcpopt.tstamp = tmp_opt.saw_tstamp;
2781 if (tmp_opt.sack_ok)
2782 req->tcpopt.sack = 1;
2783 OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
2784 return;
2785 }
2786
2787 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
2788 __be32 laddr, __be16 lport,
2789 __be32 raddr, __be16 rport,
2790 u32 rcv_isn, u32 filter, u16 window,
2791 u32 rss_qid, u8 port_id)
2792 {
2793 struct sk_buff *req_skb;
2794 struct fw_ofld_connection_wr *req;
2795 struct cpl_pass_accept_req *cpl = cplhdr(skb);
2796
2797 req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
2798 req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
2799 memset(req, 0, sizeof(*req));
2800 req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
2801 req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
2802 req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
2803 req->le.filter = filter;
2804 req->le.lport = lport;
2805 req->le.pport = rport;
2806 req->le.u.ipv4.lip = laddr;
2807 req->le.u.ipv4.pip = raddr;
2808 req->tcb.rcv_nxt = htonl(rcv_isn + 1);
2809 req->tcb.rcv_adv = htons(window);
2810 req->tcb.t_state_to_astid =
2811 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
2812 V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
2813 V_FW_OFLD_CONNECTION_WR_ASTID(
2814 GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
2815
2816 /*
2817 * We store the qid in opt2 which will be used by the firmware
2818 * to send us the wr response.
2819 */
2820 req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
2821
2822 /*
2823 * We initialize the MSS index in TCB to 0xF.
2824 * So that when driver sends cpl_pass_accept_rpl
2825 * TCB picks up the correct value. If this was 0
2826 * TP will ignore any value > 0 for MSS index.
2827 */
2828 req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
2829 req->cookie = cpu_to_be64((u64)skb);
2830
2831 set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
2832 cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
2833 }
2834
2835 /*
2836 * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
2837 * messages when a filter is being used instead of server to
2838 * redirect a syn packet. When packets hit filter they are redirected
2839 * to the offload queue and driver tries to establish the connection
2840 * using firmware work request.
2841 */
2842 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
2843 {
2844 int stid;
2845 unsigned int filter;
2846 struct ethhdr *eh = NULL;
2847 struct vlan_ethhdr *vlan_eh = NULL;
2848 struct iphdr *iph;
2849 struct tcphdr *tcph;
2850 struct rss_header *rss = (void *)skb->data;
2851 struct cpl_rx_pkt *cpl = (void *)skb->data;
2852 struct cpl_pass_accept_req *req = (void *)(rss + 1);
2853 struct l2t_entry *e;
2854 struct dst_entry *dst;
2855 struct rtable *rt;
2856 struct c4iw_ep *lep;
2857 u16 window;
2858 struct port_info *pi;
2859 struct net_device *pdev;
2860 u16 rss_qid;
2861 int step;
2862 u32 tx_chan;
2863 struct neighbour *neigh;
2864
2865 /* Drop all non-SYN packets */
2866 if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
2867 goto reject;
2868
2869 /*
2870 * Drop all packets which did not hit the filter.
2871 * Unlikely to happen.
2872 */
2873 if (!(rss->filter_hit && rss->filter_tid))
2874 goto reject;
2875
2876 /*
2877 * Calculate the server tid from filter hit index from cpl_rx_pkt.
2878 */
2879 stid = cpu_to_be32(rss->hash_val) - dev->rdev.lldi.tids->sftid_base
2880 + dev->rdev.lldi.tids->nstids;
2881
2882 lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
2883 if (!lep) {
2884 PDBG("%s connect request on invalid stid %d\n", __func__, stid);
2885 goto reject;
2886 }
2887
2888 if (G_RX_ETHHDR_LEN(ntohl(cpl->l2info)) == ETH_HLEN) {
2889 eh = (struct ethhdr *)(req + 1);
2890 iph = (struct iphdr *)(eh + 1);
2891 } else {
2892 vlan_eh = (struct vlan_ethhdr *)(req + 1);
2893 iph = (struct iphdr *)(vlan_eh + 1);
2894 skb->vlan_tci = ntohs(cpl->vlan);
2895 }
2896
2897 if (iph->version != 0x4)
2898 goto reject;
2899
2900 tcph = (struct tcphdr *)(iph + 1);
2901 skb_set_network_header(skb, (void *)iph - (void *)rss);
2902 skb_set_transport_header(skb, (void *)tcph - (void *)rss);
2903 skb_get(skb);
2904
2905 PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
2906 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
2907 ntohs(tcph->source), iph->tos);
2908
2909 rt = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
2910 iph->tos);
2911 if (!rt) {
2912 pr_err("%s - failed to find dst entry!\n",
2913 __func__);
2914 goto reject;
2915 }
2916 dst = &rt->dst;
2917 neigh = dst_neigh_lookup_skb(dst, skb);
2918
2919 if (neigh->dev->flags & IFF_LOOPBACK) {
2920 pdev = ip_dev_find(&init_net, iph->daddr);
2921 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
2922 pdev, 0);
2923 pi = (struct port_info *)netdev_priv(pdev);
2924 tx_chan = cxgb4_port_chan(pdev);
2925 dev_put(pdev);
2926 } else {
2927 e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
2928 neigh->dev, 0);
2929 pi = (struct port_info *)netdev_priv(neigh->dev);
2930 tx_chan = cxgb4_port_chan(neigh->dev);
2931 }
2932 if (!e) {
2933 pr_err("%s - failed to allocate l2t entry!\n",
2934 __func__);
2935 goto free_dst;
2936 }
2937
2938 step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
2939 rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
2940 window = htons(tcph->window);
2941
2942 /* Calcuate filter portion for LE region. */
2943 filter = cpu_to_be32(select_ntuple(dev, dst, e));
2944
2945 /*
2946 * Synthesize the cpl_pass_accept_req. We have everything except the
2947 * TID. Once firmware sends a reply with TID we update the TID field
2948 * in cpl and pass it through the regular cpl_pass_accept_req path.
2949 */
2950 build_cpl_pass_accept_req(skb, stid, iph->tos);
2951 send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
2952 tcph->source, ntohl(tcph->seq), filter, window,
2953 rss_qid, pi->port_id);
2954 cxgb4_l2t_release(e);
2955 free_dst:
2956 dst_release(dst);
2957 reject:
2958 return 0;
2959 }
2960
2961 /*
2962 * These are the real handlers that are called from a
2963 * work queue.
2964 */
2965 static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
2966 [CPL_ACT_ESTABLISH] = act_establish,
2967 [CPL_ACT_OPEN_RPL] = act_open_rpl,
2968 [CPL_RX_DATA] = rx_data,
2969 [CPL_ABORT_RPL_RSS] = abort_rpl,
2970 [CPL_ABORT_RPL] = abort_rpl,
2971 [CPL_PASS_OPEN_RPL] = pass_open_rpl,
2972 [CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
2973 [CPL_PASS_ACCEPT_REQ] = pass_accept_req,
2974 [CPL_PASS_ESTABLISH] = pass_establish,
2975 [CPL_PEER_CLOSE] = peer_close,
2976 [CPL_ABORT_REQ_RSS] = peer_abort,
2977 [CPL_CLOSE_CON_RPL] = close_con_rpl,
2978 [CPL_RDMA_TERMINATE] = terminate,
2979 [CPL_FW4_ACK] = fw4_ack,
2980 [CPL_FW6_MSG] = deferred_fw6_msg,
2981 [CPL_RX_PKT] = rx_pkt
2982 };
2983
2984 static void process_timeout(struct c4iw_ep *ep)
2985 {
2986 struct c4iw_qp_attributes attrs;
2987 int abort = 1;
2988
2989 mutex_lock(&ep->com.mutex);
2990 PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
2991 ep->com.state);
2992 switch (ep->com.state) {
2993 case MPA_REQ_SENT:
2994 __state_set(&ep->com, ABORTING);
2995 connect_reply_upcall(ep, -ETIMEDOUT);
2996 break;
2997 case MPA_REQ_WAIT:
2998 __state_set(&ep->com, ABORTING);
2999 break;
3000 case CLOSING:
3001 case MORIBUND:
3002 if (ep->com.cm_id && ep->com.qp) {
3003 attrs.next_state = C4IW_QP_STATE_ERROR;
3004 c4iw_modify_qp(ep->com.qp->rhp,
3005 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
3006 &attrs, 1);
3007 }
3008 __state_set(&ep->com, ABORTING);
3009 break;
3010 default:
3011 WARN(1, "%s unexpected state ep %p tid %u state %u\n",
3012 __func__, ep, ep->hwtid, ep->com.state);
3013 abort = 0;
3014 }
3015 mutex_unlock(&ep->com.mutex);
3016 if (abort)
3017 abort_connection(ep, NULL, GFP_KERNEL);
3018 c4iw_put_ep(&ep->com);
3019 }
3020
3021 static void process_timedout_eps(void)
3022 {
3023 struct c4iw_ep *ep;
3024
3025 spin_lock_irq(&timeout_lock);
3026 while (!list_empty(&timeout_list)) {
3027 struct list_head *tmp;
3028
3029 tmp = timeout_list.next;
3030 list_del(tmp);
3031 spin_unlock_irq(&timeout_lock);
3032 ep = list_entry(tmp, struct c4iw_ep, entry);
3033 process_timeout(ep);
3034 spin_lock_irq(&timeout_lock);
3035 }
3036 spin_unlock_irq(&timeout_lock);
3037 }
3038
3039 static void process_work(struct work_struct *work)
3040 {
3041 struct sk_buff *skb = NULL;
3042 struct c4iw_dev *dev;
3043 struct cpl_act_establish *rpl;
3044 unsigned int opcode;
3045 int ret;
3046
3047 while ((skb = skb_dequeue(&rxq))) {
3048 rpl = cplhdr(skb);
3049 dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3050 opcode = rpl->ot.opcode;
3051
3052 BUG_ON(!work_handlers[opcode]);
3053 ret = work_handlers[opcode](dev, skb);
3054 if (!ret)
3055 kfree_skb(skb);
3056 }
3057 process_timedout_eps();
3058 }
3059
3060 static DECLARE_WORK(skb_work, process_work);
3061
3062 static void ep_timeout(unsigned long arg)
3063 {
3064 struct c4iw_ep *ep = (struct c4iw_ep *)arg;
3065
3066 spin_lock(&timeout_lock);
3067 list_add_tail(&ep->entry, &timeout_list);
3068 spin_unlock(&timeout_lock);
3069 queue_work(workq, &skb_work);
3070 }
3071
3072 /*
3073 * All the CM events are handled on a work queue to have a safe context.
3074 */
3075 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
3076 {
3077
3078 /*
3079 * Save dev in the skb->cb area.
3080 */
3081 *((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
3082
3083 /*
3084 * Queue the skb and schedule the worker thread.
3085 */
3086 skb_queue_tail(&rxq, skb);
3087 queue_work(workq, &skb_work);
3088 return 0;
3089 }
3090
3091 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3092 {
3093 struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
3094
3095 if (rpl->status != CPL_ERR_NONE) {
3096 printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
3097 "for tid %u\n", rpl->status, GET_TID(rpl));
3098 }
3099 kfree_skb(skb);
3100 return 0;
3101 }
3102
3103 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3104 {
3105 struct cpl_fw6_msg *rpl = cplhdr(skb);
3106 struct c4iw_wr_wait *wr_waitp;
3107 int ret;
3108
3109 PDBG("%s type %u\n", __func__, rpl->type);
3110
3111 switch (rpl->type) {
3112 case FW6_TYPE_WR_RPL:
3113 ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
3114 wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
3115 PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
3116 if (wr_waitp)
3117 c4iw_wake_up(wr_waitp, ret ? -ret : 0);
3118 kfree_skb(skb);
3119 break;
3120 case FW6_TYPE_CQE:
3121 case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3122 sched(dev, skb);
3123 break;
3124 default:
3125 printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
3126 rpl->type);
3127 kfree_skb(skb);
3128 break;
3129 }
3130 return 0;
3131 }
3132
3133 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
3134 {
3135 struct cpl_abort_req_rss *req = cplhdr(skb);
3136 struct c4iw_ep *ep;
3137 struct tid_info *t = dev->rdev.lldi.tids;
3138 unsigned int tid = GET_TID(req);
3139
3140 ep = lookup_tid(t, tid);
3141 if (!ep) {
3142 printk(KERN_WARNING MOD
3143 "Abort on non-existent endpoint, tid %d\n", tid);
3144 kfree_skb(skb);
3145 return 0;
3146 }
3147 if (is_neg_adv_abort(req->status)) {
3148 PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
3149 ep->hwtid);
3150 kfree_skb(skb);
3151 return 0;
3152 }
3153 PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
3154 ep->com.state);
3155
3156 /*
3157 * Wake up any threads in rdma_init() or rdma_fini().
3158 */
3159 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
3160 sched(dev, skb);
3161 return 0;
3162 }
3163
3164 /*
3165 * Most upcalls from the T4 Core go to sched() to
3166 * schedule the processing on a work queue.
3167 */
3168 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
3169 [CPL_ACT_ESTABLISH] = sched,
3170 [CPL_ACT_OPEN_RPL] = sched,
3171 [CPL_RX_DATA] = sched,
3172 [CPL_ABORT_RPL_RSS] = sched,
3173 [CPL_ABORT_RPL] = sched,
3174 [CPL_PASS_OPEN_RPL] = sched,
3175 [CPL_CLOSE_LISTSRV_RPL] = sched,
3176 [CPL_PASS_ACCEPT_REQ] = sched,
3177 [CPL_PASS_ESTABLISH] = sched,
3178 [CPL_PEER_CLOSE] = sched,
3179 [CPL_CLOSE_CON_RPL] = sched,
3180 [CPL_ABORT_REQ_RSS] = peer_abort_intr,
3181 [CPL_RDMA_TERMINATE] = sched,
3182 [CPL_FW4_ACK] = sched,
3183 [CPL_SET_TCB_RPL] = set_tcb_rpl,
3184 [CPL_FW6_MSG] = fw6_msg,
3185 [CPL_RX_PKT] = sched
3186 };
3187
3188 int __init c4iw_cm_init(void)
3189 {
3190 spin_lock_init(&timeout_lock);
3191 skb_queue_head_init(&rxq);
3192
3193 workq = create_singlethread_workqueue("iw_cxgb4");
3194 if (!workq)
3195 return -ENOMEM;
3196
3197 return 0;
3198 }
3199
3200 void __exit c4iw_cm_term(void)
3201 {
3202 WARN_ON(!list_empty(&timeout_list));
3203 flush_workqueue(workq);
3204 destroy_workqueue(workq);
3205 }
Linux kernel - Deliverables
This page took 0.448762 seconds and 5 git commands to generate.