xref: /openbmc/linux/drivers/scsi/cxgbi/libcxgbi.c (revision a20eefae)
1 /*
2  * libcxgbi.c: Chelsio common library for T3/T4 iSCSI driver.
3  *
4  * Copyright (c) 2010-2015 Chelsio Communications, Inc.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License as published by
8  * the Free Software Foundation.
9  *
10  * Written by: Karen Xie (kxie@chelsio.com)
11  * Written by: Rakesh Ranjan (rranjan@chelsio.com)
12  */
13 
14 #define pr_fmt(fmt)	KBUILD_MODNAME ":%s: " fmt, __func__
15 
16 #include <linux/skbuff.h>
17 #include <linux/crypto.h>
18 #include <linux/scatterlist.h>
19 #include <linux/pci.h>
20 #include <scsi/scsi.h>
21 #include <scsi/scsi_cmnd.h>
22 #include <scsi/scsi_host.h>
23 #include <linux/if_vlan.h>
24 #include <linux/inet.h>
25 #include <net/dst.h>
26 #include <net/route.h>
27 #include <net/ipv6.h>
28 #include <net/ip6_route.h>
29 #include <net/addrconf.h>
30 
31 #include <linux/inetdevice.h>	/* ip_dev_find */
32 #include <linux/module.h>
33 #include <net/tcp.h>
34 
35 static unsigned int dbg_level;
36 
37 #include "libcxgbi.h"
38 
39 #define DRV_MODULE_NAME		"libcxgbi"
40 #define DRV_MODULE_DESC		"Chelsio iSCSI driver library"
41 #define DRV_MODULE_VERSION	"0.9.1-ko"
42 #define DRV_MODULE_RELDATE	"Apr. 2015"
43 
44 static char version[] =
45 	DRV_MODULE_DESC " " DRV_MODULE_NAME
46 	" v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
47 
48 MODULE_AUTHOR("Chelsio Communications, Inc.");
49 MODULE_DESCRIPTION(DRV_MODULE_DESC);
50 MODULE_VERSION(DRV_MODULE_VERSION);
51 MODULE_LICENSE("GPL");
52 
53 module_param(dbg_level, uint, 0644);
54 MODULE_PARM_DESC(dbg_level, "libiscsi debug level (default=0)");
55 
56 
57 /*
58  * cxgbi device management
59  * maintains a list of the cxgbi devices
60  */
61 static LIST_HEAD(cdev_list);
62 static DEFINE_MUTEX(cdev_mutex);
63 
64 static LIST_HEAD(cdev_rcu_list);
65 static DEFINE_SPINLOCK(cdev_rcu_lock);
66 
67 static inline void cxgbi_decode_sw_tag(u32 sw_tag, int *idx, int *age)
68 {
69 	if (age)
70 		*age = sw_tag & 0x7FFF;
71 	if (idx)
72 		*idx = (sw_tag >> 16) & 0x7FFF;
73 }
74 
75 int cxgbi_device_portmap_create(struct cxgbi_device *cdev, unsigned int base,
76 				unsigned int max_conn)
77 {
78 	struct cxgbi_ports_map *pmap = &cdev->pmap;
79 
80 	pmap->port_csk = cxgbi_alloc_big_mem(max_conn *
81 					     sizeof(struct cxgbi_sock *),
82 					     GFP_KERNEL);
83 	if (!pmap->port_csk) {
84 		pr_warn("cdev 0x%p, portmap OOM %u.\n", cdev, max_conn);
85 		return -ENOMEM;
86 	}
87 
88 	pmap->max_connect = max_conn;
89 	pmap->sport_base = base;
90 	spin_lock_init(&pmap->lock);
91 	return 0;
92 }
93 EXPORT_SYMBOL_GPL(cxgbi_device_portmap_create);
94 
95 void cxgbi_device_portmap_cleanup(struct cxgbi_device *cdev)
96 {
97 	struct cxgbi_ports_map *pmap = &cdev->pmap;
98 	struct cxgbi_sock *csk;
99 	int i;
100 
101 	for (i = 0; i < pmap->max_connect; i++) {
102 		if (pmap->port_csk[i]) {
103 			csk = pmap->port_csk[i];
104 			pmap->port_csk[i] = NULL;
105 			log_debug(1 << CXGBI_DBG_SOCK,
106 				"csk 0x%p, cdev 0x%p, offload down.\n",
107 				csk, cdev);
108 			spin_lock_bh(&csk->lock);
109 			cxgbi_sock_set_flag(csk, CTPF_OFFLOAD_DOWN);
110 			cxgbi_sock_closed(csk);
111 			spin_unlock_bh(&csk->lock);
112 			cxgbi_sock_put(csk);
113 		}
114 	}
115 }
116 EXPORT_SYMBOL_GPL(cxgbi_device_portmap_cleanup);
117 
118 static inline void cxgbi_device_destroy(struct cxgbi_device *cdev)
119 {
120 	log_debug(1 << CXGBI_DBG_DEV,
121 		"cdev 0x%p, p# %u.\n", cdev, cdev->nports);
122 	cxgbi_hbas_remove(cdev);
123 	cxgbi_device_portmap_cleanup(cdev);
124 	cxgbi_ppm_release(cdev->cdev2ppm(cdev));
125 	if (cdev->pmap.max_connect)
126 		cxgbi_free_big_mem(cdev->pmap.port_csk);
127 	kfree(cdev);
128 }
129 
130 struct cxgbi_device *cxgbi_device_register(unsigned int extra,
131 					   unsigned int nports)
132 {
133 	struct cxgbi_device *cdev;
134 
135 	cdev = kzalloc(sizeof(*cdev) + extra + nports *
136 			(sizeof(struct cxgbi_hba *) +
137 			 sizeof(struct net_device *)),
138 			GFP_KERNEL);
139 	if (!cdev) {
140 		pr_warn("nport %d, OOM.\n", nports);
141 		return NULL;
142 	}
143 	cdev->ports = (struct net_device **)(cdev + 1);
144 	cdev->hbas = (struct cxgbi_hba **)(((char*)cdev->ports) + nports *
145 						sizeof(struct net_device *));
146 	if (extra)
147 		cdev->dd_data = ((char *)cdev->hbas) +
148 				nports * sizeof(struct cxgbi_hba *);
149 	spin_lock_init(&cdev->pmap.lock);
150 
151 	mutex_lock(&cdev_mutex);
152 	list_add_tail(&cdev->list_head, &cdev_list);
153 	mutex_unlock(&cdev_mutex);
154 
155 	spin_lock(&cdev_rcu_lock);
156 	list_add_tail_rcu(&cdev->rcu_node, &cdev_rcu_list);
157 	spin_unlock(&cdev_rcu_lock);
158 
159 	log_debug(1 << CXGBI_DBG_DEV,
160 		"cdev 0x%p, p# %u.\n", cdev, nports);
161 	return cdev;
162 }
163 EXPORT_SYMBOL_GPL(cxgbi_device_register);
164 
165 void cxgbi_device_unregister(struct cxgbi_device *cdev)
166 {
167 	log_debug(1 << CXGBI_DBG_DEV,
168 		"cdev 0x%p, p# %u,%s.\n",
169 		cdev, cdev->nports, cdev->nports ? cdev->ports[0]->name : "");
170 
171 	mutex_lock(&cdev_mutex);
172 	list_del(&cdev->list_head);
173 	mutex_unlock(&cdev_mutex);
174 
175 	spin_lock(&cdev_rcu_lock);
176 	list_del_rcu(&cdev->rcu_node);
177 	spin_unlock(&cdev_rcu_lock);
178 	synchronize_rcu();
179 
180 	cxgbi_device_destroy(cdev);
181 }
182 EXPORT_SYMBOL_GPL(cxgbi_device_unregister);
183 
184 void cxgbi_device_unregister_all(unsigned int flag)
185 {
186 	struct cxgbi_device *cdev, *tmp;
187 
188 	mutex_lock(&cdev_mutex);
189 	list_for_each_entry_safe(cdev, tmp, &cdev_list, list_head) {
190 		if ((cdev->flags & flag) == flag) {
191 			mutex_unlock(&cdev_mutex);
192 			cxgbi_device_unregister(cdev);
193 			mutex_lock(&cdev_mutex);
194 		}
195 	}
196 	mutex_unlock(&cdev_mutex);
197 }
198 EXPORT_SYMBOL_GPL(cxgbi_device_unregister_all);
199 
200 struct cxgbi_device *cxgbi_device_find_by_lldev(void *lldev)
201 {
202 	struct cxgbi_device *cdev, *tmp;
203 
204 	mutex_lock(&cdev_mutex);
205 	list_for_each_entry_safe(cdev, tmp, &cdev_list, list_head) {
206 		if (cdev->lldev == lldev) {
207 			mutex_unlock(&cdev_mutex);
208 			return cdev;
209 		}
210 	}
211 	mutex_unlock(&cdev_mutex);
212 
213 	log_debug(1 << CXGBI_DBG_DEV,
214 		"lldev 0x%p, NO match found.\n", lldev);
215 	return NULL;
216 }
217 EXPORT_SYMBOL_GPL(cxgbi_device_find_by_lldev);
218 
219 struct cxgbi_device *cxgbi_device_find_by_netdev(struct net_device *ndev,
220 						 int *port)
221 {
222 	struct net_device *vdev = NULL;
223 	struct cxgbi_device *cdev, *tmp;
224 	int i;
225 
226 	if (is_vlan_dev(ndev)) {
227 		vdev = ndev;
228 		ndev = vlan_dev_real_dev(ndev);
229 		log_debug(1 << CXGBI_DBG_DEV,
230 			"vlan dev %s -> %s.\n", vdev->name, ndev->name);
231 	}
232 
233 	mutex_lock(&cdev_mutex);
234 	list_for_each_entry_safe(cdev, tmp, &cdev_list, list_head) {
235 		for (i = 0; i < cdev->nports; i++) {
236 			if (ndev == cdev->ports[i]) {
237 				cdev->hbas[i]->vdev = vdev;
238 				mutex_unlock(&cdev_mutex);
239 				if (port)
240 					*port = i;
241 				return cdev;
242 			}
243 		}
244 	}
245 	mutex_unlock(&cdev_mutex);
246 	log_debug(1 << CXGBI_DBG_DEV,
247 		"ndev 0x%p, %s, NO match found.\n", ndev, ndev->name);
248 	return NULL;
249 }
250 EXPORT_SYMBOL_GPL(cxgbi_device_find_by_netdev);
251 
252 struct cxgbi_device *cxgbi_device_find_by_netdev_rcu(struct net_device *ndev,
253 						     int *port)
254 {
255 	struct net_device *vdev = NULL;
256 	struct cxgbi_device *cdev;
257 	int i;
258 
259 	if (is_vlan_dev(ndev)) {
260 		vdev = ndev;
261 		ndev = vlan_dev_real_dev(ndev);
262 		pr_info("vlan dev %s -> %s.\n", vdev->name, ndev->name);
263 	}
264 
265 	rcu_read_lock();
266 	list_for_each_entry_rcu(cdev, &cdev_rcu_list, rcu_node) {
267 		for (i = 0; i < cdev->nports; i++) {
268 			if (ndev == cdev->ports[i]) {
269 				cdev->hbas[i]->vdev = vdev;
270 				rcu_read_unlock();
271 				if (port)
272 					*port = i;
273 				return cdev;
274 			}
275 		}
276 	}
277 	rcu_read_unlock();
278 
279 	log_debug(1 << CXGBI_DBG_DEV,
280 		  "ndev 0x%p, %s, NO match found.\n", ndev, ndev->name);
281 	return NULL;
282 }
283 EXPORT_SYMBOL_GPL(cxgbi_device_find_by_netdev_rcu);
284 
285 static struct cxgbi_device *cxgbi_device_find_by_mac(struct net_device *ndev,
286 						     int *port)
287 {
288 	struct net_device *vdev = NULL;
289 	struct cxgbi_device *cdev, *tmp;
290 	int i;
291 
292 	if (is_vlan_dev(ndev)) {
293 		vdev = ndev;
294 		ndev = vlan_dev_real_dev(ndev);
295 		pr_info("vlan dev %s -> %s.\n", vdev->name, ndev->name);
296 	}
297 
298 	mutex_lock(&cdev_mutex);
299 	list_for_each_entry_safe(cdev, tmp, &cdev_list, list_head) {
300 		for (i = 0; i < cdev->nports; i++) {
301 			if (!memcmp(ndev->dev_addr, cdev->ports[i]->dev_addr,
302 				    MAX_ADDR_LEN)) {
303 				cdev->hbas[i]->vdev = vdev;
304 				mutex_unlock(&cdev_mutex);
305 				if (port)
306 					*port = i;
307 				return cdev;
308 			}
309 		}
310 	}
311 	mutex_unlock(&cdev_mutex);
312 	log_debug(1 << CXGBI_DBG_DEV,
313 		  "ndev 0x%p, %s, NO match mac found.\n",
314 		  ndev, ndev->name);
315 	return NULL;
316 }
317 
318 void cxgbi_hbas_remove(struct cxgbi_device *cdev)
319 {
320 	int i;
321 	struct cxgbi_hba *chba;
322 
323 	log_debug(1 << CXGBI_DBG_DEV,
324 		"cdev 0x%p, p#%u.\n", cdev, cdev->nports);
325 
326 	for (i = 0; i < cdev->nports; i++) {
327 		chba = cdev->hbas[i];
328 		if (chba) {
329 			cdev->hbas[i] = NULL;
330 			iscsi_host_remove(chba->shost);
331 			pci_dev_put(cdev->pdev);
332 			iscsi_host_free(chba->shost);
333 		}
334 	}
335 }
336 EXPORT_SYMBOL_GPL(cxgbi_hbas_remove);
337 
338 int cxgbi_hbas_add(struct cxgbi_device *cdev, u64 max_lun,
339 		unsigned int max_id, struct scsi_host_template *sht,
340 		struct scsi_transport_template *stt)
341 {
342 	struct cxgbi_hba *chba;
343 	struct Scsi_Host *shost;
344 	int i, err;
345 
346 	log_debug(1 << CXGBI_DBG_DEV, "cdev 0x%p, p#%u.\n", cdev, cdev->nports);
347 
348 	for (i = 0; i < cdev->nports; i++) {
349 		shost = iscsi_host_alloc(sht, sizeof(*chba), 1);
350 		if (!shost) {
351 			pr_info("0x%p, p%d, %s, host alloc failed.\n",
352 				cdev, i, cdev->ports[i]->name);
353 			err = -ENOMEM;
354 			goto err_out;
355 		}
356 
357 		shost->transportt = stt;
358 		shost->max_lun = max_lun;
359 		shost->max_id = max_id;
360 		shost->max_channel = 0;
361 		shost->max_cmd_len = 16;
362 
363 		chba = iscsi_host_priv(shost);
364 		chba->cdev = cdev;
365 		chba->ndev = cdev->ports[i];
366 		chba->shost = shost;
367 
368 		log_debug(1 << CXGBI_DBG_DEV,
369 			"cdev 0x%p, p#%d %s: chba 0x%p.\n",
370 			cdev, i, cdev->ports[i]->name, chba);
371 
372 		pci_dev_get(cdev->pdev);
373 		err = iscsi_host_add(shost, &cdev->pdev->dev);
374 		if (err) {
375 			pr_info("cdev 0x%p, p#%d %s, host add failed.\n",
376 				cdev, i, cdev->ports[i]->name);
377 			pci_dev_put(cdev->pdev);
378 			scsi_host_put(shost);
379 			goto  err_out;
380 		}
381 
382 		cdev->hbas[i] = chba;
383 	}
384 
385 	return 0;
386 
387 err_out:
388 	cxgbi_hbas_remove(cdev);
389 	return err;
390 }
391 EXPORT_SYMBOL_GPL(cxgbi_hbas_add);
392 
393 /*
394  * iSCSI offload
395  *
396  * - source port management
397  *   To find a free source port in the port allocation map we use a very simple
398  *   rotor scheme to look for the next free port.
399  *
400  *   If a source port has been specified make sure that it doesn't collide with
401  *   our normal source port allocation map.  If it's outside the range of our
402  *   allocation/deallocation scheme just let them use it.
403  *
404  *   If the source port is outside our allocation range, the caller is
405  *   responsible for keeping track of their port usage.
406  */
407 
408 static struct cxgbi_sock *find_sock_on_port(struct cxgbi_device *cdev,
409 					    unsigned char port_id)
410 {
411 	struct cxgbi_ports_map *pmap = &cdev->pmap;
412 	unsigned int i;
413 	unsigned int used;
414 
415 	if (!pmap->max_connect || !pmap->used)
416 		return NULL;
417 
418 	spin_lock_bh(&pmap->lock);
419 	used = pmap->used;
420 	for (i = 0; used && i < pmap->max_connect; i++) {
421 		struct cxgbi_sock *csk = pmap->port_csk[i];
422 
423 		if (csk) {
424 			if (csk->port_id == port_id) {
425 				spin_unlock_bh(&pmap->lock);
426 				return csk;
427 			}
428 			used--;
429 		}
430 	}
431 	spin_unlock_bh(&pmap->lock);
432 
433 	return NULL;
434 }
435 
436 static int sock_get_port(struct cxgbi_sock *csk)
437 {
438 	struct cxgbi_device *cdev = csk->cdev;
439 	struct cxgbi_ports_map *pmap = &cdev->pmap;
440 	unsigned int start;
441 	int idx;
442 	__be16 *port;
443 
444 	if (!pmap->max_connect) {
445 		pr_err("cdev 0x%p, p#%u %s, NO port map.\n",
446 			   cdev, csk->port_id, cdev->ports[csk->port_id]->name);
447 		return -EADDRNOTAVAIL;
448 	}
449 
450 	if (csk->csk_family == AF_INET)
451 		port = &csk->saddr.sin_port;
452 	else /* ipv6 */
453 		port = &csk->saddr6.sin6_port;
454 
455 	if (*port) {
456 		pr_err("source port NON-ZERO %u.\n",
457 			ntohs(*port));
458 		return -EADDRINUSE;
459 	}
460 
461 	spin_lock_bh(&pmap->lock);
462 	if (pmap->used >= pmap->max_connect) {
463 		spin_unlock_bh(&pmap->lock);
464 		pr_info("cdev 0x%p, p#%u %s, ALL ports used.\n",
465 			cdev, csk->port_id, cdev->ports[csk->port_id]->name);
466 		return -EADDRNOTAVAIL;
467 	}
468 
469 	start = idx = pmap->next;
470 	do {
471 		if (++idx >= pmap->max_connect)
472 			idx = 0;
473 		if (!pmap->port_csk[idx]) {
474 			pmap->used++;
475 			*port = htons(pmap->sport_base + idx);
476 			pmap->next = idx;
477 			pmap->port_csk[idx] = csk;
478 			spin_unlock_bh(&pmap->lock);
479 			cxgbi_sock_get(csk);
480 			log_debug(1 << CXGBI_DBG_SOCK,
481 				"cdev 0x%p, p#%u %s, p %u, %u.\n",
482 				cdev, csk->port_id,
483 				cdev->ports[csk->port_id]->name,
484 				pmap->sport_base + idx, pmap->next);
485 			return 0;
486 		}
487 	} while (idx != start);
488 	spin_unlock_bh(&pmap->lock);
489 
490 	/* should not happen */
491 	pr_warn("cdev 0x%p, p#%u %s, next %u?\n",
492 		cdev, csk->port_id, cdev->ports[csk->port_id]->name,
493 		pmap->next);
494 	return -EADDRNOTAVAIL;
495 }
496 
497 static void sock_put_port(struct cxgbi_sock *csk)
498 {
499 	struct cxgbi_device *cdev = csk->cdev;
500 	struct cxgbi_ports_map *pmap = &cdev->pmap;
501 	__be16 *port;
502 
503 	if (csk->csk_family == AF_INET)
504 		port = &csk->saddr.sin_port;
505 	else /* ipv6 */
506 		port = &csk->saddr6.sin6_port;
507 
508 	if (*port) {
509 		int idx = ntohs(*port) - pmap->sport_base;
510 
511 		*port = 0;
512 		if (idx < 0 || idx >= pmap->max_connect) {
513 			pr_err("cdev 0x%p, p#%u %s, port %u OOR.\n",
514 				cdev, csk->port_id,
515 				cdev->ports[csk->port_id]->name,
516 				ntohs(*port));
517 			return;
518 		}
519 
520 		spin_lock_bh(&pmap->lock);
521 		pmap->port_csk[idx] = NULL;
522 		pmap->used--;
523 		spin_unlock_bh(&pmap->lock);
524 
525 		log_debug(1 << CXGBI_DBG_SOCK,
526 			"cdev 0x%p, p#%u %s, release %u.\n",
527 			cdev, csk->port_id, cdev->ports[csk->port_id]->name,
528 			pmap->sport_base + idx);
529 
530 		cxgbi_sock_put(csk);
531 	}
532 }
533 
534 /*
535  * iscsi tcp connection
536  */
537 void cxgbi_sock_free_cpl_skbs(struct cxgbi_sock *csk)
538 {
539 	if (csk->cpl_close) {
540 		kfree_skb(csk->cpl_close);
541 		csk->cpl_close = NULL;
542 	}
543 	if (csk->cpl_abort_req) {
544 		kfree_skb(csk->cpl_abort_req);
545 		csk->cpl_abort_req = NULL;
546 	}
547 	if (csk->cpl_abort_rpl) {
548 		kfree_skb(csk->cpl_abort_rpl);
549 		csk->cpl_abort_rpl = NULL;
550 	}
551 }
552 EXPORT_SYMBOL_GPL(cxgbi_sock_free_cpl_skbs);
553 
554 static struct cxgbi_sock *cxgbi_sock_create(struct cxgbi_device *cdev)
555 {
556 	struct cxgbi_sock *csk = kzalloc(sizeof(*csk), GFP_NOIO);
557 
558 	if (!csk) {
559 		pr_info("alloc csk %zu failed.\n", sizeof(*csk));
560 		return NULL;
561 	}
562 
563 	if (cdev->csk_alloc_cpls(csk) < 0) {
564 		pr_info("csk 0x%p, alloc cpls failed.\n", csk);
565 		kfree(csk);
566 		return NULL;
567 	}
568 
569 	spin_lock_init(&csk->lock);
570 	kref_init(&csk->refcnt);
571 	skb_queue_head_init(&csk->receive_queue);
572 	skb_queue_head_init(&csk->write_queue);
573 	timer_setup(&csk->retry_timer, NULL, 0);
574 	init_completion(&csk->cmpl);
575 	rwlock_init(&csk->callback_lock);
576 	csk->cdev = cdev;
577 	csk->flags = 0;
578 	cxgbi_sock_set_state(csk, CTP_CLOSED);
579 
580 	log_debug(1 << CXGBI_DBG_SOCK, "cdev 0x%p, new csk 0x%p.\n", cdev, csk);
581 
582 	return csk;
583 }
584 
585 static struct rtable *find_route_ipv4(struct flowi4 *fl4,
586 				      __be32 saddr, __be32 daddr,
587 				      __be16 sport, __be16 dport, u8 tos,
588 				      int ifindex)
589 {
590 	struct rtable *rt;
591 
592 	rt = ip_route_output_ports(&init_net, fl4, NULL, daddr, saddr,
593 				   dport, sport, IPPROTO_TCP, tos, ifindex);
594 	if (IS_ERR(rt))
595 		return NULL;
596 
597 	return rt;
598 }
599 
600 static struct cxgbi_sock *
601 cxgbi_check_route(struct sockaddr *dst_addr, int ifindex)
602 {
603 	struct sockaddr_in *daddr = (struct sockaddr_in *)dst_addr;
604 	struct dst_entry *dst;
605 	struct net_device *ndev;
606 	struct cxgbi_device *cdev;
607 	struct rtable *rt = NULL;
608 	struct neighbour *n;
609 	struct flowi4 fl4;
610 	struct cxgbi_sock *csk = NULL;
611 	unsigned int mtu = 0;
612 	int port = 0xFFFF;
613 	int err = 0;
614 
615 	rt = find_route_ipv4(&fl4, 0, daddr->sin_addr.s_addr, 0,
616 			     daddr->sin_port, 0, ifindex);
617 	if (!rt) {
618 		pr_info("no route to ipv4 0x%x, port %u.\n",
619 			be32_to_cpu(daddr->sin_addr.s_addr),
620 			be16_to_cpu(daddr->sin_port));
621 		err = -ENETUNREACH;
622 		goto err_out;
623 	}
624 	dst = &rt->dst;
625 	n = dst_neigh_lookup(dst, &daddr->sin_addr.s_addr);
626 	if (!n) {
627 		err = -ENODEV;
628 		goto rel_rt;
629 	}
630 	ndev = n->dev;
631 
632 	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
633 		pr_info("multi-cast route %pI4, port %u, dev %s.\n",
634 			&daddr->sin_addr.s_addr, ntohs(daddr->sin_port),
635 			ndev->name);
636 		err = -ENETUNREACH;
637 		goto rel_neigh;
638 	}
639 
640 	if (ndev->flags & IFF_LOOPBACK) {
641 		ndev = ip_dev_find(&init_net, daddr->sin_addr.s_addr);
642 		if (!ndev) {
643 			err = -ENETUNREACH;
644 			goto rel_neigh;
645 		}
646 		mtu = ndev->mtu;
647 		pr_info("rt dev %s, loopback -> %s, mtu %u.\n",
648 			n->dev->name, ndev->name, mtu);
649 	}
650 
651 	if (!(ndev->flags & IFF_UP) || !netif_carrier_ok(ndev)) {
652 		pr_info("%s interface not up.\n", ndev->name);
653 		err = -ENETDOWN;
654 		goto rel_neigh;
655 	}
656 
657 	cdev = cxgbi_device_find_by_netdev(ndev, &port);
658 	if (!cdev)
659 		cdev = cxgbi_device_find_by_mac(ndev, &port);
660 	if (!cdev) {
661 		pr_info("dst %pI4, %s, NOT cxgbi device.\n",
662 			&daddr->sin_addr.s_addr, ndev->name);
663 		err = -ENETUNREACH;
664 		goto rel_neigh;
665 	}
666 	log_debug(1 << CXGBI_DBG_SOCK,
667 		"route to %pI4 :%u, ndev p#%d,%s, cdev 0x%p.\n",
668 		&daddr->sin_addr.s_addr, ntohs(daddr->sin_port),
669 			   port, ndev->name, cdev);
670 
671 	csk = cxgbi_sock_create(cdev);
672 	if (!csk) {
673 		err = -ENOMEM;
674 		goto rel_neigh;
675 	}
676 	csk->cdev = cdev;
677 	csk->port_id = port;
678 	csk->mtu = mtu;
679 	csk->dst = dst;
680 
681 	csk->csk_family = AF_INET;
682 	csk->daddr.sin_addr.s_addr = daddr->sin_addr.s_addr;
683 	csk->daddr.sin_port = daddr->sin_port;
684 	csk->daddr.sin_family = daddr->sin_family;
685 	csk->saddr.sin_family = daddr->sin_family;
686 	csk->saddr.sin_addr.s_addr = fl4.saddr;
687 	neigh_release(n);
688 
689 	return csk;
690 
691 rel_neigh:
692 	neigh_release(n);
693 
694 rel_rt:
695 	ip_rt_put(rt);
696 err_out:
697 	return ERR_PTR(err);
698 }
699 
700 #if IS_ENABLED(CONFIG_IPV6)
701 static struct rt6_info *find_route_ipv6(const struct in6_addr *saddr,
702 					const struct in6_addr *daddr,
703 					int ifindex)
704 {
705 	struct flowi6 fl;
706 
707 	memset(&fl, 0, sizeof(fl));
708 	fl.flowi6_oif = ifindex;
709 	if (saddr)
710 		memcpy(&fl.saddr, saddr, sizeof(struct in6_addr));
711 	if (daddr)
712 		memcpy(&fl.daddr, daddr, sizeof(struct in6_addr));
713 	return (struct rt6_info *)ip6_route_output(&init_net, NULL, &fl);
714 }
715 
716 static struct cxgbi_sock *
717 cxgbi_check_route6(struct sockaddr *dst_addr, int ifindex)
718 {
719 	struct sockaddr_in6 *daddr6 = (struct sockaddr_in6 *)dst_addr;
720 	struct dst_entry *dst;
721 	struct net_device *ndev;
722 	struct cxgbi_device *cdev;
723 	struct rt6_info *rt = NULL;
724 	struct neighbour *n;
725 	struct in6_addr pref_saddr;
726 	struct cxgbi_sock *csk = NULL;
727 	unsigned int mtu = 0;
728 	int port = 0xFFFF;
729 	int err = 0;
730 
731 	rt = find_route_ipv6(NULL, &daddr6->sin6_addr, ifindex);
732 
733 	if (!rt) {
734 		pr_info("no route to ipv6 %pI6 port %u\n",
735 			daddr6->sin6_addr.s6_addr,
736 			be16_to_cpu(daddr6->sin6_port));
737 		err = -ENETUNREACH;
738 		goto err_out;
739 	}
740 
741 	dst = &rt->dst;
742 
743 	n = dst_neigh_lookup(dst, &daddr6->sin6_addr);
744 
745 	if (!n) {
746 		pr_info("%pI6, port %u, dst no neighbour.\n",
747 			daddr6->sin6_addr.s6_addr,
748 			be16_to_cpu(daddr6->sin6_port));
749 		err = -ENETUNREACH;
750 		goto rel_rt;
751 	}
752 	ndev = n->dev;
753 
754 	if (!(ndev->flags & IFF_UP) || !netif_carrier_ok(ndev)) {
755 		pr_info("%s interface not up.\n", ndev->name);
756 		err = -ENETDOWN;
757 		goto rel_rt;
758 	}
759 
760 	if (ipv6_addr_is_multicast(&daddr6->sin6_addr)) {
761 		pr_info("multi-cast route %pI6 port %u, dev %s.\n",
762 			daddr6->sin6_addr.s6_addr,
763 			ntohs(daddr6->sin6_port), ndev->name);
764 		err = -ENETUNREACH;
765 		goto rel_rt;
766 	}
767 
768 	cdev = cxgbi_device_find_by_netdev(ndev, &port);
769 	if (!cdev)
770 		cdev = cxgbi_device_find_by_mac(ndev, &port);
771 	if (!cdev) {
772 		pr_info("dst %pI6 %s, NOT cxgbi device.\n",
773 			daddr6->sin6_addr.s6_addr, ndev->name);
774 		err = -ENETUNREACH;
775 		goto rel_rt;
776 	}
777 	log_debug(1 << CXGBI_DBG_SOCK,
778 		  "route to %pI6 :%u, ndev p#%d,%s, cdev 0x%p.\n",
779 		  daddr6->sin6_addr.s6_addr, ntohs(daddr6->sin6_port), port,
780 		  ndev->name, cdev);
781 
782 	csk = cxgbi_sock_create(cdev);
783 	if (!csk) {
784 		err = -ENOMEM;
785 		goto rel_rt;
786 	}
787 	csk->cdev = cdev;
788 	csk->port_id = port;
789 	csk->mtu = mtu;
790 	csk->dst = dst;
791 
792 	rt6_get_prefsrc(rt, &pref_saddr);
793 	if (ipv6_addr_any(&pref_saddr)) {
794 		struct inet6_dev *idev = ip6_dst_idev((struct dst_entry *)rt);
795 
796 		err = ipv6_dev_get_saddr(&init_net, idev ? idev->dev : NULL,
797 					 &daddr6->sin6_addr, 0, &pref_saddr);
798 		if (err) {
799 			pr_info("failed to get source address to reach %pI6\n",
800 				&daddr6->sin6_addr);
801 			goto rel_rt;
802 		}
803 	}
804 
805 	csk->csk_family = AF_INET6;
806 	csk->daddr6.sin6_addr = daddr6->sin6_addr;
807 	csk->daddr6.sin6_port = daddr6->sin6_port;
808 	csk->daddr6.sin6_family = daddr6->sin6_family;
809 	csk->saddr6.sin6_family = daddr6->sin6_family;
810 	csk->saddr6.sin6_addr = pref_saddr;
811 
812 	neigh_release(n);
813 	return csk;
814 
815 rel_rt:
816 	if (n)
817 		neigh_release(n);
818 
819 	ip6_rt_put(rt);
820 	if (csk)
821 		cxgbi_sock_closed(csk);
822 err_out:
823 	return ERR_PTR(err);
824 }
825 #endif /* IS_ENABLED(CONFIG_IPV6) */
826 
827 void cxgbi_sock_established(struct cxgbi_sock *csk, unsigned int snd_isn,
828 			unsigned int opt)
829 {
830 	csk->write_seq = csk->snd_nxt = csk->snd_una = snd_isn;
831 	dst_confirm(csk->dst);
832 	smp_mb();
833 	cxgbi_sock_set_state(csk, CTP_ESTABLISHED);
834 }
835 EXPORT_SYMBOL_GPL(cxgbi_sock_established);
836 
837 static void cxgbi_inform_iscsi_conn_closing(struct cxgbi_sock *csk)
838 {
839 	log_debug(1 << CXGBI_DBG_SOCK,
840 		"csk 0x%p, state %u, flags 0x%lx, conn 0x%p.\n",
841 		csk, csk->state, csk->flags, csk->user_data);
842 
843 	if (csk->state != CTP_ESTABLISHED) {
844 		read_lock_bh(&csk->callback_lock);
845 		if (csk->user_data)
846 			iscsi_conn_failure(csk->user_data,
847 					ISCSI_ERR_TCP_CONN_CLOSE);
848 		read_unlock_bh(&csk->callback_lock);
849 	}
850 }
851 
852 void cxgbi_sock_closed(struct cxgbi_sock *csk)
853 {
854 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
855 		csk, (csk)->state, (csk)->flags, (csk)->tid);
856 	cxgbi_sock_set_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED);
857 	if (csk->state == CTP_ACTIVE_OPEN || csk->state == CTP_CLOSED)
858 		return;
859 	if (csk->saddr.sin_port)
860 		sock_put_port(csk);
861 	if (csk->dst)
862 		dst_release(csk->dst);
863 	csk->cdev->csk_release_offload_resources(csk);
864 	cxgbi_sock_set_state(csk, CTP_CLOSED);
865 	cxgbi_inform_iscsi_conn_closing(csk);
866 	cxgbi_sock_put(csk);
867 }
868 EXPORT_SYMBOL_GPL(cxgbi_sock_closed);
869 
870 static void need_active_close(struct cxgbi_sock *csk)
871 {
872 	int data_lost;
873 	int close_req = 0;
874 
875 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
876 		csk, (csk)->state, (csk)->flags, (csk)->tid);
877 	spin_lock_bh(&csk->lock);
878 	if (csk->dst)
879 		dst_confirm(csk->dst);
880 	data_lost = skb_queue_len(&csk->receive_queue);
881 	__skb_queue_purge(&csk->receive_queue);
882 
883 	if (csk->state == CTP_ACTIVE_OPEN)
884 		cxgbi_sock_set_flag(csk, CTPF_ACTIVE_CLOSE_NEEDED);
885 	else if (csk->state == CTP_ESTABLISHED) {
886 		close_req = 1;
887 		cxgbi_sock_set_state(csk, CTP_ACTIVE_CLOSE);
888 	} else if (csk->state == CTP_PASSIVE_CLOSE) {
889 		close_req = 1;
890 		cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_2);
891 	}
892 
893 	if (close_req) {
894 		if (!cxgbi_sock_flag(csk, CTPF_LOGOUT_RSP_RCVD) ||
895 		    data_lost)
896 			csk->cdev->csk_send_abort_req(csk);
897 		else
898 			csk->cdev->csk_send_close_req(csk);
899 	}
900 
901 	spin_unlock_bh(&csk->lock);
902 }
903 
904 void cxgbi_sock_fail_act_open(struct cxgbi_sock *csk, int errno)
905 {
906 	pr_info("csk 0x%p,%u,%lx, %pI4:%u-%pI4:%u, err %d.\n",
907 			csk, csk->state, csk->flags,
908 			&csk->saddr.sin_addr.s_addr, csk->saddr.sin_port,
909 			&csk->daddr.sin_addr.s_addr, csk->daddr.sin_port,
910 			errno);
911 
912 	cxgbi_sock_set_state(csk, CTP_CONNECTING);
913 	csk->err = errno;
914 	cxgbi_sock_closed(csk);
915 }
916 EXPORT_SYMBOL_GPL(cxgbi_sock_fail_act_open);
917 
918 void cxgbi_sock_act_open_req_arp_failure(void *handle, struct sk_buff *skb)
919 {
920 	struct cxgbi_sock *csk = (struct cxgbi_sock *)skb->sk;
921 	struct module *owner = csk->cdev->owner;
922 
923 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
924 		csk, (csk)->state, (csk)->flags, (csk)->tid);
925 	cxgbi_sock_get(csk);
926 	spin_lock_bh(&csk->lock);
927 	if (csk->state == CTP_ACTIVE_OPEN)
928 		cxgbi_sock_fail_act_open(csk, -EHOSTUNREACH);
929 	spin_unlock_bh(&csk->lock);
930 	cxgbi_sock_put(csk);
931 	__kfree_skb(skb);
932 
933 	module_put(owner);
934 }
935 EXPORT_SYMBOL_GPL(cxgbi_sock_act_open_req_arp_failure);
936 
937 void cxgbi_sock_rcv_abort_rpl(struct cxgbi_sock *csk)
938 {
939 	cxgbi_sock_get(csk);
940 	spin_lock_bh(&csk->lock);
941 
942 	cxgbi_sock_set_flag(csk, CTPF_ABORT_RPL_RCVD);
943 	if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING)) {
944 		cxgbi_sock_clear_flag(csk, CTPF_ABORT_RPL_PENDING);
945 		if (cxgbi_sock_flag(csk, CTPF_ABORT_REQ_RCVD))
946 			pr_err("csk 0x%p,%u,0x%lx,%u,ABT_RPL_RSS.\n",
947 			       csk, csk->state, csk->flags, csk->tid);
948 		cxgbi_sock_closed(csk);
949 	}
950 
951 	spin_unlock_bh(&csk->lock);
952 	cxgbi_sock_put(csk);
953 }
954 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_abort_rpl);
955 
956 void cxgbi_sock_rcv_peer_close(struct cxgbi_sock *csk)
957 {
958 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
959 		csk, (csk)->state, (csk)->flags, (csk)->tid);
960 	cxgbi_sock_get(csk);
961 	spin_lock_bh(&csk->lock);
962 
963 	if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING))
964 		goto done;
965 
966 	switch (csk->state) {
967 	case CTP_ESTABLISHED:
968 		cxgbi_sock_set_state(csk, CTP_PASSIVE_CLOSE);
969 		break;
970 	case CTP_ACTIVE_CLOSE:
971 		cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_2);
972 		break;
973 	case CTP_CLOSE_WAIT_1:
974 		cxgbi_sock_closed(csk);
975 		break;
976 	case CTP_ABORTING:
977 		break;
978 	default:
979 		pr_err("csk 0x%p,%u,0x%lx,%u, bad state.\n",
980 			csk, csk->state, csk->flags, csk->tid);
981 	}
982 	cxgbi_inform_iscsi_conn_closing(csk);
983 done:
984 	spin_unlock_bh(&csk->lock);
985 	cxgbi_sock_put(csk);
986 }
987 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_peer_close);
988 
989 void cxgbi_sock_rcv_close_conn_rpl(struct cxgbi_sock *csk, u32 snd_nxt)
990 {
991 	log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p,%u,0x%lx,%u.\n",
992 		csk, (csk)->state, (csk)->flags, (csk)->tid);
993 	cxgbi_sock_get(csk);
994 	spin_lock_bh(&csk->lock);
995 
996 	csk->snd_una = snd_nxt - 1;
997 	if (cxgbi_sock_flag(csk, CTPF_ABORT_RPL_PENDING))
998 		goto done;
999 
1000 	switch (csk->state) {
1001 	case CTP_ACTIVE_CLOSE:
1002 		cxgbi_sock_set_state(csk, CTP_CLOSE_WAIT_1);
1003 		break;
1004 	case CTP_CLOSE_WAIT_1:
1005 	case CTP_CLOSE_WAIT_2:
1006 		cxgbi_sock_closed(csk);
1007 		break;
1008 	case CTP_ABORTING:
1009 		break;
1010 	default:
1011 		pr_err("csk 0x%p,%u,0x%lx,%u, bad state.\n",
1012 			csk, csk->state, csk->flags, csk->tid);
1013 	}
1014 done:
1015 	spin_unlock_bh(&csk->lock);
1016 	cxgbi_sock_put(csk);
1017 }
1018 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_close_conn_rpl);
1019 
1020 void cxgbi_sock_rcv_wr_ack(struct cxgbi_sock *csk, unsigned int credits,
1021 			   unsigned int snd_una, int seq_chk)
1022 {
1023 	log_debug(1 << CXGBI_DBG_TOE | 1 << CXGBI_DBG_SOCK,
1024 			"csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, snd_una %u,%d.\n",
1025 			csk, csk->state, csk->flags, csk->tid, credits,
1026 			csk->wr_cred, csk->wr_una_cred, snd_una, seq_chk);
1027 
1028 	spin_lock_bh(&csk->lock);
1029 
1030 	csk->wr_cred += credits;
1031 	if (csk->wr_una_cred > csk->wr_max_cred - csk->wr_cred)
1032 		csk->wr_una_cred = csk->wr_max_cred - csk->wr_cred;
1033 
1034 	while (credits) {
1035 		struct sk_buff *p = cxgbi_sock_peek_wr(csk);
1036 
1037 		if (unlikely(!p)) {
1038 			pr_err("csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, empty.\n",
1039 				csk, csk->state, csk->flags, csk->tid, credits,
1040 				csk->wr_cred, csk->wr_una_cred);
1041 			break;
1042 		}
1043 
1044 		if (unlikely(credits < p->csum)) {
1045 			pr_warn("csk 0x%p,%u,0x%lx,%u, cr %u,%u+%u, < %u.\n",
1046 				csk, csk->state, csk->flags, csk->tid,
1047 				credits, csk->wr_cred, csk->wr_una_cred,
1048 				p->csum);
1049 			p->csum -= credits;
1050 			break;
1051 		} else {
1052 			cxgbi_sock_dequeue_wr(csk);
1053 			credits -= p->csum;
1054 			kfree_skb(p);
1055 		}
1056 	}
1057 
1058 	cxgbi_sock_check_wr_invariants(csk);
1059 
1060 	if (seq_chk) {
1061 		if (unlikely(before(snd_una, csk->snd_una))) {
1062 			pr_warn("csk 0x%p,%u,0x%lx,%u, snd_una %u/%u.",
1063 				csk, csk->state, csk->flags, csk->tid, snd_una,
1064 				csk->snd_una);
1065 			goto done;
1066 		}
1067 
1068 		if (csk->snd_una != snd_una) {
1069 			csk->snd_una = snd_una;
1070 			dst_confirm(csk->dst);
1071 		}
1072 	}
1073 
1074 	if (skb_queue_len(&csk->write_queue)) {
1075 		if (csk->cdev->csk_push_tx_frames(csk, 0))
1076 			cxgbi_conn_tx_open(csk);
1077 	} else
1078 		cxgbi_conn_tx_open(csk);
1079 done:
1080 	spin_unlock_bh(&csk->lock);
1081 }
1082 EXPORT_SYMBOL_GPL(cxgbi_sock_rcv_wr_ack);
1083 
1084 static unsigned int cxgbi_sock_find_best_mtu(struct cxgbi_sock *csk,
1085 					     unsigned short mtu)
1086 {
1087 	int i = 0;
1088 
1089 	while (i < csk->cdev->nmtus - 1 && csk->cdev->mtus[i + 1] <= mtu)
1090 		++i;
1091 
1092 	return i;
1093 }
1094 
1095 unsigned int cxgbi_sock_select_mss(struct cxgbi_sock *csk, unsigned int pmtu)
1096 {
1097 	unsigned int idx;
1098 	struct dst_entry *dst = csk->dst;
1099 
1100 	csk->advmss = dst_metric_advmss(dst);
1101 
1102 	if (csk->advmss > pmtu - 40)
1103 		csk->advmss = pmtu - 40;
1104 	if (csk->advmss < csk->cdev->mtus[0] - 40)
1105 		csk->advmss = csk->cdev->mtus[0] - 40;
1106 	idx = cxgbi_sock_find_best_mtu(csk, csk->advmss + 40);
1107 
1108 	return idx;
1109 }
1110 EXPORT_SYMBOL_GPL(cxgbi_sock_select_mss);
1111 
1112 void cxgbi_sock_skb_entail(struct cxgbi_sock *csk, struct sk_buff *skb)
1113 {
1114 	cxgbi_skcb_tcp_seq(skb) = csk->write_seq;
1115 	__skb_queue_tail(&csk->write_queue, skb);
1116 }
1117 EXPORT_SYMBOL_GPL(cxgbi_sock_skb_entail);
1118 
1119 void cxgbi_sock_purge_wr_queue(struct cxgbi_sock *csk)
1120 {
1121 	struct sk_buff *skb;
1122 
1123 	while ((skb = cxgbi_sock_dequeue_wr(csk)) != NULL)
1124 		kfree_skb(skb);
1125 }
1126 EXPORT_SYMBOL_GPL(cxgbi_sock_purge_wr_queue);
1127 
1128 void cxgbi_sock_check_wr_invariants(const struct cxgbi_sock *csk)
1129 {
1130 	int pending = cxgbi_sock_count_pending_wrs(csk);
1131 
1132 	if (unlikely(csk->wr_cred + pending != csk->wr_max_cred))
1133 		pr_err("csk 0x%p, tid %u, credit %u + %u != %u.\n",
1134 			csk, csk->tid, csk->wr_cred, pending, csk->wr_max_cred);
1135 }
1136 EXPORT_SYMBOL_GPL(cxgbi_sock_check_wr_invariants);
1137 
1138 static int cxgbi_sock_send_pdus(struct cxgbi_sock *csk, struct sk_buff *skb)
1139 {
1140 	struct cxgbi_device *cdev = csk->cdev;
1141 	struct sk_buff *next;
1142 	int err, copied = 0;
1143 
1144 	spin_lock_bh(&csk->lock);
1145 
1146 	if (csk->state != CTP_ESTABLISHED) {
1147 		log_debug(1 << CXGBI_DBG_PDU_TX,
1148 			"csk 0x%p,%u,0x%lx,%u, EAGAIN.\n",
1149 			csk, csk->state, csk->flags, csk->tid);
1150 		err = -EAGAIN;
1151 		goto out_err;
1152 	}
1153 
1154 	if (csk->err) {
1155 		log_debug(1 << CXGBI_DBG_PDU_TX,
1156 			"csk 0x%p,%u,0x%lx,%u, EPIPE %d.\n",
1157 			csk, csk->state, csk->flags, csk->tid, csk->err);
1158 		err = -EPIPE;
1159 		goto out_err;
1160 	}
1161 
1162 	if (csk->write_seq - csk->snd_una >= csk->snd_win) {
1163 		log_debug(1 << CXGBI_DBG_PDU_TX,
1164 			"csk 0x%p,%u,0x%lx,%u, FULL %u-%u >= %u.\n",
1165 			csk, csk->state, csk->flags, csk->tid, csk->write_seq,
1166 			csk->snd_una, csk->snd_win);
1167 		err = -ENOBUFS;
1168 		goto out_err;
1169 	}
1170 
1171 	while (skb) {
1172 		int frags = skb_shinfo(skb)->nr_frags +
1173 				(skb->len != skb->data_len);
1174 
1175 		if (unlikely(skb_headroom(skb) < cdev->skb_tx_rsvd)) {
1176 			pr_err("csk 0x%p, skb head %u < %u.\n",
1177 				csk, skb_headroom(skb), cdev->skb_tx_rsvd);
1178 			err = -EINVAL;
1179 			goto out_err;
1180 		}
1181 
1182 		if (frags >= SKB_WR_LIST_SIZE) {
1183 			pr_err("csk 0x%p, frags %d, %u,%u >%u.\n",
1184 				csk, skb_shinfo(skb)->nr_frags, skb->len,
1185 				skb->data_len, (uint)(SKB_WR_LIST_SIZE));
1186 			err = -EINVAL;
1187 			goto out_err;
1188 		}
1189 
1190 		next = skb->next;
1191 		skb->next = NULL;
1192 		cxgbi_skcb_set_flag(skb, SKCBF_TX_NEED_HDR);
1193 		cxgbi_sock_skb_entail(csk, skb);
1194 		copied += skb->len;
1195 		csk->write_seq += skb->len +
1196 				cxgbi_ulp_extra_len(cxgbi_skcb_ulp_mode(skb));
1197 		skb = next;
1198 	}
1199 
1200 	if (likely(skb_queue_len(&csk->write_queue)))
1201 		cdev->csk_push_tx_frames(csk, 1);
1202 done:
1203 	spin_unlock_bh(&csk->lock);
1204 	return copied;
1205 
1206 out_err:
1207 	if (copied == 0 && err == -EPIPE)
1208 		copied = csk->err ? csk->err : -EPIPE;
1209 	else
1210 		copied = err;
1211 	goto done;
1212 }
1213 
1214 static inline void
1215 scmd_get_params(struct scsi_cmnd *sc, struct scatterlist **sgl,
1216 		unsigned int *sgcnt, unsigned int *dlen,
1217 		unsigned int prot)
1218 {
1219 	struct scsi_data_buffer *sdb = prot ? scsi_prot(sc) : &sc->sdb;
1220 
1221 	*sgl = sdb->table.sgl;
1222 	*sgcnt = sdb->table.nents;
1223 	*dlen = sdb->length;
1224 	/* Caution: for protection sdb, sdb->length is invalid */
1225 }
1226 
1227 void cxgbi_ddp_set_one_ppod(struct cxgbi_pagepod *ppod,
1228 			    struct cxgbi_task_tag_info *ttinfo,
1229 			    struct scatterlist **sg_pp, unsigned int *sg_off)
1230 {
1231 	struct scatterlist *sg = sg_pp ? *sg_pp : NULL;
1232 	unsigned int offset = sg_off ? *sg_off : 0;
1233 	dma_addr_t addr = 0UL;
1234 	unsigned int len = 0;
1235 	int i;
1236 
1237 	memcpy(ppod, &ttinfo->hdr, sizeof(struct cxgbi_pagepod_hdr));
1238 
1239 	if (sg) {
1240 		addr = sg_dma_address(sg);
1241 		len = sg_dma_len(sg);
1242 	}
1243 
1244 	for (i = 0; i < PPOD_PAGES_MAX; i++) {
1245 		if (sg) {
1246 			ppod->addr[i] = cpu_to_be64(addr + offset);
1247 			offset += PAGE_SIZE;
1248 			if (offset == (len + sg->offset)) {
1249 				offset = 0;
1250 				sg = sg_next(sg);
1251 				if (sg) {
1252 					addr = sg_dma_address(sg);
1253 					len = sg_dma_len(sg);
1254 				}
1255 			}
1256 		} else {
1257 			ppod->addr[i] = 0ULL;
1258 		}
1259 	}
1260 
1261 	/*
1262 	 * the fifth address needs to be repeated in the next ppod, so do
1263 	 * not move sg
1264 	 */
1265 	if (sg_pp) {
1266 		*sg_pp = sg;
1267 		*sg_off = offset;
1268 	}
1269 
1270 	if (offset == len) {
1271 		offset = 0;
1272 		sg = sg_next(sg);
1273 		if (sg) {
1274 			addr = sg_dma_address(sg);
1275 			len = sg_dma_len(sg);
1276 		}
1277 	}
1278 	ppod->addr[i] = sg ? cpu_to_be64(addr + offset) : 0ULL;
1279 }
1280 EXPORT_SYMBOL_GPL(cxgbi_ddp_set_one_ppod);
1281 
1282 /*
1283  * APIs interacting with open-iscsi libraries
1284  */
1285 
1286 static unsigned char padding[4];
1287 
1288 int cxgbi_ddp_ppm_setup(void **ppm_pp, struct cxgbi_device *cdev,
1289 			struct cxgbi_tag_format *tformat,
1290 			unsigned int iscsi_size, unsigned int llimit,
1291 			unsigned int start, unsigned int rsvd_factor,
1292 			unsigned int edram_start, unsigned int edram_size)
1293 {
1294 	int err = cxgbi_ppm_init(ppm_pp, cdev->ports[0], cdev->pdev,
1295 				cdev->lldev, tformat, iscsi_size, llimit, start,
1296 				rsvd_factor, edram_start, edram_size);
1297 
1298 	if (err >= 0) {
1299 		struct cxgbi_ppm *ppm = (struct cxgbi_ppm *)(*ppm_pp);
1300 
1301 		if (ppm->ppmax < 1024 ||
1302 		    ppm->tformat.pgsz_idx_dflt >= DDP_PGIDX_MAX)
1303 			cdev->flags |= CXGBI_FLAG_DDP_OFF;
1304 		err = 0;
1305 	} else {
1306 		cdev->flags |= CXGBI_FLAG_DDP_OFF;
1307 	}
1308 
1309 	return err;
1310 }
1311 EXPORT_SYMBOL_GPL(cxgbi_ddp_ppm_setup);
1312 
1313 static int cxgbi_ddp_sgl_check(struct scatterlist *sgl, int nents)
1314 {
1315 	int i;
1316 	int last_sgidx = nents - 1;
1317 	struct scatterlist *sg = sgl;
1318 
1319 	for (i = 0; i < nents; i++, sg = sg_next(sg)) {
1320 		unsigned int len = sg->length + sg->offset;
1321 
1322 		if ((sg->offset & 0x3) || (i && sg->offset) ||
1323 		    ((i != last_sgidx) && len != PAGE_SIZE)) {
1324 			log_debug(1 << CXGBI_DBG_DDP,
1325 				  "sg %u/%u, %u,%u, not aligned.\n",
1326 				  i, nents, sg->offset, sg->length);
1327 			goto err_out;
1328 		}
1329 	}
1330 	return 0;
1331 err_out:
1332 	return -EINVAL;
1333 }
1334 
1335 static int cxgbi_ddp_reserve(struct cxgbi_conn *cconn,
1336 			     struct cxgbi_task_data *tdata, u32 sw_tag,
1337 			     unsigned int xferlen)
1338 {
1339 	struct cxgbi_sock *csk = cconn->cep->csk;
1340 	struct cxgbi_device *cdev = csk->cdev;
1341 	struct cxgbi_ppm *ppm = cdev->cdev2ppm(cdev);
1342 	struct cxgbi_task_tag_info *ttinfo = &tdata->ttinfo;
1343 	struct scatterlist *sgl = ttinfo->sgl;
1344 	unsigned int sgcnt = ttinfo->nents;
1345 	unsigned int sg_offset = sgl->offset;
1346 	int err;
1347 
1348 	if (cdev->flags & CXGBI_FLAG_DDP_OFF) {
1349 		log_debug(1 << CXGBI_DBG_DDP,
1350 			  "cdev 0x%p DDP off.\n", cdev);
1351 		return -EINVAL;
1352 	}
1353 
1354 	if (!ppm || xferlen < DDP_THRESHOLD || !sgcnt ||
1355 	    ppm->tformat.pgsz_idx_dflt >= DDP_PGIDX_MAX) {
1356 		log_debug(1 << CXGBI_DBG_DDP,
1357 			  "ppm 0x%p, pgidx %u, xfer %u, sgcnt %u, NO ddp.\n",
1358 			  ppm, ppm ? ppm->tformat.pgsz_idx_dflt : DDP_PGIDX_MAX,
1359 			  xferlen, ttinfo->nents);
1360 		return -EINVAL;
1361 	}
1362 
1363 	/* make sure the buffer is suitable for ddp */
1364 	if (cxgbi_ddp_sgl_check(sgl, sgcnt) < 0)
1365 		return -EINVAL;
1366 
1367 	ttinfo->nr_pages = (xferlen + sgl->offset + (1 << PAGE_SHIFT) - 1) >>
1368 			    PAGE_SHIFT;
1369 
1370 	/*
1371 	 * the ddp tag will be used for the itt in the outgoing pdu,
1372 	 * the itt genrated by libiscsi is saved in the ppm and can be
1373 	 * retrieved via the ddp tag
1374 	 */
1375 	err = cxgbi_ppm_ppods_reserve(ppm, ttinfo->nr_pages, 0, &ttinfo->idx,
1376 				      &ttinfo->tag, (unsigned long)sw_tag);
1377 	if (err < 0) {
1378 		cconn->ddp_full++;
1379 		return err;
1380 	}
1381 	ttinfo->npods = err;
1382 
1383 	 /* setup dma from scsi command sgl */
1384 	sgl->offset = 0;
1385 	err = dma_map_sg(&ppm->pdev->dev, sgl, sgcnt, DMA_FROM_DEVICE);
1386 	sgl->offset = sg_offset;
1387 	if (err == 0) {
1388 		pr_info("%s: 0x%x, xfer %u, sgl %u dma mapping err.\n",
1389 			__func__, sw_tag, xferlen, sgcnt);
1390 		goto rel_ppods;
1391 	}
1392 	if (err != ttinfo->nr_pages) {
1393 		log_debug(1 << CXGBI_DBG_DDP,
1394 			  "%s: sw tag 0x%x, xfer %u, sgl %u, dma count %d.\n",
1395 			  __func__, sw_tag, xferlen, sgcnt, err);
1396 	}
1397 
1398 	ttinfo->flags |= CXGBI_PPOD_INFO_FLAG_MAPPED;
1399 	ttinfo->cid = csk->port_id;
1400 
1401 	cxgbi_ppm_make_ppod_hdr(ppm, ttinfo->tag, csk->tid, sgl->offset,
1402 				xferlen, &ttinfo->hdr);
1403 
1404 	if (cdev->flags & CXGBI_FLAG_USE_PPOD_OFLDQ) {
1405 		/* write ppod from xmit_pdu (of iscsi_scsi_command pdu) */
1406 		ttinfo->flags |= CXGBI_PPOD_INFO_FLAG_VALID;
1407 	} else {
1408 		/* write ppod from control queue now */
1409 		err = cdev->csk_ddp_set_map(ppm, csk, ttinfo);
1410 		if (err < 0)
1411 			goto rel_ppods;
1412 	}
1413 
1414 	return 0;
1415 
1416 rel_ppods:
1417 	cxgbi_ppm_ppod_release(ppm, ttinfo->idx);
1418 
1419 	if (ttinfo->flags & CXGBI_PPOD_INFO_FLAG_MAPPED) {
1420 		ttinfo->flags &= ~CXGBI_PPOD_INFO_FLAG_MAPPED;
1421 		dma_unmap_sg(&ppm->pdev->dev, sgl, sgcnt, DMA_FROM_DEVICE);
1422 	}
1423 	return -EINVAL;
1424 }
1425 
1426 static void task_release_itt(struct iscsi_task *task, itt_t hdr_itt)
1427 {
1428 	struct scsi_cmnd *sc = task->sc;
1429 	struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data;
1430 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
1431 	struct cxgbi_device *cdev = cconn->chba->cdev;
1432 	struct cxgbi_ppm *ppm = cdev->cdev2ppm(cdev);
1433 	u32 tag = ntohl((__force u32)hdr_itt);
1434 
1435 	log_debug(1 << CXGBI_DBG_DDP,
1436 		  "cdev 0x%p, task 0x%p, release tag 0x%x.\n",
1437 		  cdev, task, tag);
1438 	if (sc && sc->sc_data_direction == DMA_FROM_DEVICE &&
1439 	    cxgbi_ppm_is_ddp_tag(ppm, tag)) {
1440 		struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
1441 		struct cxgbi_task_tag_info *ttinfo = &tdata->ttinfo;
1442 
1443 		if (!(cdev->flags & CXGBI_FLAG_USE_PPOD_OFLDQ))
1444 			cdev->csk_ddp_clear_map(cdev, ppm, ttinfo);
1445 		cxgbi_ppm_ppod_release(ppm, ttinfo->idx);
1446 		dma_unmap_sg(&ppm->pdev->dev, ttinfo->sgl, ttinfo->nents,
1447 			     DMA_FROM_DEVICE);
1448 	}
1449 }
1450 
1451 static inline u32 cxgbi_build_sw_tag(u32 idx, u32 age)
1452 {
1453 	/* assume idx and age both are < 0x7FFF (32767) */
1454 	return (idx << 16) | age;
1455 }
1456 
1457 static int task_reserve_itt(struct iscsi_task *task, itt_t *hdr_itt)
1458 {
1459 	struct scsi_cmnd *sc = task->sc;
1460 	struct iscsi_conn *conn = task->conn;
1461 	struct iscsi_session *sess = conn->session;
1462 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1463 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
1464 	struct cxgbi_device *cdev = cconn->chba->cdev;
1465 	struct cxgbi_ppm *ppm = cdev->cdev2ppm(cdev);
1466 	u32 sw_tag = cxgbi_build_sw_tag(task->itt, sess->age);
1467 	u32 tag = 0;
1468 	int err = -EINVAL;
1469 
1470 	if (sc && sc->sc_data_direction == DMA_FROM_DEVICE) {
1471 		struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
1472 		struct cxgbi_task_tag_info *ttinfo = &tdata->ttinfo;
1473 
1474 		scmd_get_params(sc, &ttinfo->sgl, &ttinfo->nents,
1475 				&tdata->dlen, 0);
1476 		err = cxgbi_ddp_reserve(cconn, tdata, sw_tag, tdata->dlen);
1477 		if (!err)
1478 			tag = ttinfo->tag;
1479 		else
1480 			 log_debug(1 << CXGBI_DBG_DDP,
1481 				   "csk 0x%p, R task 0x%p, %u,%u, no ddp.\n",
1482 				   cconn->cep->csk, task, tdata->dlen,
1483 				   ttinfo->nents);
1484 	}
1485 
1486 	if (err < 0) {
1487 		err = cxgbi_ppm_make_non_ddp_tag(ppm, sw_tag, &tag);
1488 		if (err < 0)
1489 			return err;
1490 	}
1491 	/*  the itt need to sent in big-endian order */
1492 	*hdr_itt = (__force itt_t)htonl(tag);
1493 
1494 	log_debug(1 << CXGBI_DBG_DDP,
1495 		  "cdev 0x%p, task 0x%p, 0x%x(0x%x,0x%x)->0x%x/0x%x.\n",
1496 		  cdev, task, sw_tag, task->itt, sess->age, tag, *hdr_itt);
1497 	return 0;
1498 }
1499 
1500 void cxgbi_parse_pdu_itt(struct iscsi_conn *conn, itt_t itt, int *idx, int *age)
1501 {
1502 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1503 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
1504 	struct cxgbi_device *cdev = cconn->chba->cdev;
1505 	struct cxgbi_ppm *ppm = cdev->cdev2ppm(cdev);
1506 	u32 tag = ntohl((__force u32)itt);
1507 	u32 sw_bits;
1508 
1509 	if (ppm) {
1510 		if (cxgbi_ppm_is_ddp_tag(ppm, tag))
1511 			sw_bits = cxgbi_ppm_get_tag_caller_data(ppm, tag);
1512 		else
1513 			sw_bits = cxgbi_ppm_decode_non_ddp_tag(ppm, tag);
1514 	} else {
1515 		sw_bits = tag;
1516 	}
1517 
1518 	cxgbi_decode_sw_tag(sw_bits, idx, age);
1519 	log_debug(1 << CXGBI_DBG_DDP,
1520 		  "cdev 0x%p, tag 0x%x/0x%x, -> 0x%x(0x%x,0x%x).\n",
1521 		  cdev, tag, itt, sw_bits, idx ? *idx : 0xFFFFF,
1522 		  age ? *age : 0xFF);
1523 }
1524 EXPORT_SYMBOL_GPL(cxgbi_parse_pdu_itt);
1525 
1526 void cxgbi_conn_tx_open(struct cxgbi_sock *csk)
1527 {
1528 	struct iscsi_conn *conn = csk->user_data;
1529 
1530 	if (conn) {
1531 		log_debug(1 << CXGBI_DBG_SOCK,
1532 			"csk 0x%p, cid %d.\n", csk, conn->id);
1533 		iscsi_conn_queue_work(conn);
1534 	}
1535 }
1536 EXPORT_SYMBOL_GPL(cxgbi_conn_tx_open);
1537 
1538 /*
1539  * pdu receive, interact with libiscsi_tcp
1540  */
1541 static inline int read_pdu_skb(struct iscsi_conn *conn,
1542 			       struct sk_buff *skb,
1543 			       unsigned int offset,
1544 			       int offloaded)
1545 {
1546 	int status = 0;
1547 	int bytes_read;
1548 
1549 	bytes_read = iscsi_tcp_recv_skb(conn, skb, offset, offloaded, &status);
1550 	switch (status) {
1551 	case ISCSI_TCP_CONN_ERR:
1552 		pr_info("skb 0x%p, off %u, %d, TCP_ERR.\n",
1553 			  skb, offset, offloaded);
1554 		return -EIO;
1555 	case ISCSI_TCP_SUSPENDED:
1556 		log_debug(1 << CXGBI_DBG_PDU_RX,
1557 			"skb 0x%p, off %u, %d, TCP_SUSPEND, rc %d.\n",
1558 			skb, offset, offloaded, bytes_read);
1559 		/* no transfer - just have caller flush queue */
1560 		return bytes_read;
1561 	case ISCSI_TCP_SKB_DONE:
1562 		pr_info("skb 0x%p, off %u, %d, TCP_SKB_DONE.\n",
1563 			skb, offset, offloaded);
1564 		/*
1565 		 * pdus should always fit in the skb and we should get
1566 		 * segment done notifcation.
1567 		 */
1568 		iscsi_conn_printk(KERN_ERR, conn, "Invalid pdu or skb.");
1569 		return -EFAULT;
1570 	case ISCSI_TCP_SEGMENT_DONE:
1571 		log_debug(1 << CXGBI_DBG_PDU_RX,
1572 			"skb 0x%p, off %u, %d, TCP_SEG_DONE, rc %d.\n",
1573 			skb, offset, offloaded, bytes_read);
1574 		return bytes_read;
1575 	default:
1576 		pr_info("skb 0x%p, off %u, %d, invalid status %d.\n",
1577 			skb, offset, offloaded, status);
1578 		return -EINVAL;
1579 	}
1580 }
1581 
1582 static int
1583 skb_read_pdu_bhs(struct cxgbi_sock *csk, struct iscsi_conn *conn,
1584 		 struct sk_buff *skb)
1585 {
1586 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1587 	int err;
1588 
1589 	log_debug(1 << CXGBI_DBG_PDU_RX,
1590 		"conn 0x%p, skb 0x%p, len %u, flag 0x%lx.\n",
1591 		conn, skb, skb->len, cxgbi_skcb_flags(skb));
1592 
1593 	if (!iscsi_tcp_recv_segment_is_hdr(tcp_conn)) {
1594 		pr_info("conn 0x%p, skb 0x%p, not hdr.\n", conn, skb);
1595 		iscsi_conn_failure(conn, ISCSI_ERR_PROTO);
1596 		return -EIO;
1597 	}
1598 
1599 	if (conn->hdrdgst_en &&
1600 	    cxgbi_skcb_test_flag(skb, SKCBF_RX_HCRC_ERR)) {
1601 		pr_info("conn 0x%p, skb 0x%p, hcrc.\n", conn, skb);
1602 		iscsi_conn_failure(conn, ISCSI_ERR_HDR_DGST);
1603 		return -EIO;
1604 	}
1605 
1606 	if (cxgbi_skcb_test_flag(skb, SKCBF_RX_ISCSI_COMPL) &&
1607 	    cxgbi_skcb_test_flag(skb, SKCBF_RX_DATA_DDPD)) {
1608 		/* If completion flag is set and data is directly
1609 		 * placed in to the host memory then update
1610 		 * task->exp_datasn to the datasn in completion
1611 		 * iSCSI hdr as T6 adapter generates completion only
1612 		 * for the last pdu of a sequence.
1613 		 */
1614 		itt_t itt = ((struct iscsi_data *)skb->data)->itt;
1615 		struct iscsi_task *task = iscsi_itt_to_ctask(conn, itt);
1616 		u32 data_sn = be32_to_cpu(((struct iscsi_data *)
1617 							skb->data)->datasn);
1618 		if (task && task->sc) {
1619 			struct iscsi_tcp_task *tcp_task = task->dd_data;
1620 
1621 			tcp_task->exp_datasn = data_sn;
1622 		}
1623 	}
1624 
1625 	err = read_pdu_skb(conn, skb, 0, 0);
1626 	if (likely(err >= 0)) {
1627 		struct iscsi_hdr *hdr = (struct iscsi_hdr *)skb->data;
1628 		u8 opcode = hdr->opcode & ISCSI_OPCODE_MASK;
1629 
1630 		if (unlikely(opcode == ISCSI_OP_LOGOUT_RSP))
1631 			cxgbi_sock_set_flag(csk, CTPF_LOGOUT_RSP_RCVD);
1632 	}
1633 
1634 	return err;
1635 }
1636 
1637 static int skb_read_pdu_data(struct iscsi_conn *conn, struct sk_buff *lskb,
1638 			     struct sk_buff *skb, unsigned int offset)
1639 {
1640 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1641 	bool offloaded = 0;
1642 	int opcode = tcp_conn->in.hdr->opcode & ISCSI_OPCODE_MASK;
1643 
1644 	log_debug(1 << CXGBI_DBG_PDU_RX,
1645 		"conn 0x%p, skb 0x%p, len %u, flag 0x%lx.\n",
1646 		conn, skb, skb->len, cxgbi_skcb_flags(skb));
1647 
1648 	if (conn->datadgst_en &&
1649 	    cxgbi_skcb_test_flag(lskb, SKCBF_RX_DCRC_ERR)) {
1650 		pr_info("conn 0x%p, skb 0x%p, dcrc 0x%lx.\n",
1651 			conn, lskb, cxgbi_skcb_flags(lskb));
1652 		iscsi_conn_failure(conn, ISCSI_ERR_DATA_DGST);
1653 		return -EIO;
1654 	}
1655 
1656 	if (iscsi_tcp_recv_segment_is_hdr(tcp_conn))
1657 		return 0;
1658 
1659 	/* coalesced, add header digest length */
1660 	if (lskb == skb && conn->hdrdgst_en)
1661 		offset += ISCSI_DIGEST_SIZE;
1662 
1663 	if (cxgbi_skcb_test_flag(lskb, SKCBF_RX_DATA_DDPD))
1664 		offloaded = 1;
1665 
1666 	if (opcode == ISCSI_OP_SCSI_DATA_IN)
1667 		log_debug(1 << CXGBI_DBG_PDU_RX,
1668 			"skb 0x%p, op 0x%x, itt 0x%x, %u %s ddp'ed.\n",
1669 			skb, opcode, ntohl(tcp_conn->in.hdr->itt),
1670 			tcp_conn->in.datalen, offloaded ? "is" : "not");
1671 
1672 	return read_pdu_skb(conn, skb, offset, offloaded);
1673 }
1674 
1675 static void csk_return_rx_credits(struct cxgbi_sock *csk, int copied)
1676 {
1677 	struct cxgbi_device *cdev = csk->cdev;
1678 	int must_send;
1679 	u32 credits;
1680 
1681 	log_debug(1 << CXGBI_DBG_PDU_RX,
1682 		"csk 0x%p,%u,0x%lx,%u, seq %u, wup %u, thre %u, %u.\n",
1683 		csk, csk->state, csk->flags, csk->tid, csk->copied_seq,
1684 		csk->rcv_wup, cdev->rx_credit_thres,
1685 		csk->rcv_win);
1686 
1687 	if (!cdev->rx_credit_thres)
1688 		return;
1689 
1690 	if (csk->state != CTP_ESTABLISHED)
1691 		return;
1692 
1693 	credits = csk->copied_seq - csk->rcv_wup;
1694 	if (unlikely(!credits))
1695 		return;
1696 	must_send = credits + 16384 >= csk->rcv_win;
1697 	if (must_send || credits >= cdev->rx_credit_thres)
1698 		csk->rcv_wup += cdev->csk_send_rx_credits(csk, credits);
1699 }
1700 
1701 void cxgbi_conn_pdu_ready(struct cxgbi_sock *csk)
1702 {
1703 	struct cxgbi_device *cdev = csk->cdev;
1704 	struct iscsi_conn *conn = csk->user_data;
1705 	struct sk_buff *skb;
1706 	unsigned int read = 0;
1707 	int err = 0;
1708 
1709 	log_debug(1 << CXGBI_DBG_PDU_RX,
1710 		"csk 0x%p, conn 0x%p.\n", csk, conn);
1711 
1712 	if (unlikely(!conn || conn->suspend_rx)) {
1713 		log_debug(1 << CXGBI_DBG_PDU_RX,
1714 			"csk 0x%p, conn 0x%p, id %d, suspend_rx %lu!\n",
1715 			csk, conn, conn ? conn->id : 0xFF,
1716 			conn ? conn->suspend_rx : 0xFF);
1717 		return;
1718 	}
1719 
1720 	while (!err) {
1721 		skb = skb_peek(&csk->receive_queue);
1722 		if (!skb ||
1723 		    !(cxgbi_skcb_test_flag(skb, SKCBF_RX_STATUS))) {
1724 			if (skb)
1725 				log_debug(1 << CXGBI_DBG_PDU_RX,
1726 					"skb 0x%p, NOT ready 0x%lx.\n",
1727 					skb, cxgbi_skcb_flags(skb));
1728 			break;
1729 		}
1730 		__skb_unlink(skb, &csk->receive_queue);
1731 
1732 		read += cxgbi_skcb_rx_pdulen(skb);
1733 		log_debug(1 << CXGBI_DBG_PDU_RX,
1734 			"csk 0x%p, skb 0x%p,%u,f 0x%lx, pdu len %u.\n",
1735 			csk, skb, skb->len, cxgbi_skcb_flags(skb),
1736 			cxgbi_skcb_rx_pdulen(skb));
1737 
1738 		if (cxgbi_skcb_test_flag(skb, SKCBF_RX_COALESCED)) {
1739 			err = skb_read_pdu_bhs(csk, conn, skb);
1740 			if (err < 0) {
1741 				pr_err("coalesced bhs, csk 0x%p, skb 0x%p,%u, "
1742 					"f 0x%lx, plen %u.\n",
1743 					csk, skb, skb->len,
1744 					cxgbi_skcb_flags(skb),
1745 					cxgbi_skcb_rx_pdulen(skb));
1746 				goto skb_done;
1747 			}
1748 			err = skb_read_pdu_data(conn, skb, skb,
1749 						err + cdev->skb_rx_extra);
1750 			if (err < 0)
1751 				pr_err("coalesced data, csk 0x%p, skb 0x%p,%u, "
1752 					"f 0x%lx, plen %u.\n",
1753 					csk, skb, skb->len,
1754 					cxgbi_skcb_flags(skb),
1755 					cxgbi_skcb_rx_pdulen(skb));
1756 		} else {
1757 			err = skb_read_pdu_bhs(csk, conn, skb);
1758 			if (err < 0) {
1759 				pr_err("bhs, csk 0x%p, skb 0x%p,%u, "
1760 					"f 0x%lx, plen %u.\n",
1761 					csk, skb, skb->len,
1762 					cxgbi_skcb_flags(skb),
1763 					cxgbi_skcb_rx_pdulen(skb));
1764 				goto skb_done;
1765 			}
1766 
1767 			if (cxgbi_skcb_test_flag(skb, SKCBF_RX_DATA)) {
1768 				struct sk_buff *dskb;
1769 
1770 				dskb = skb_peek(&csk->receive_queue);
1771 				if (!dskb) {
1772 					pr_err("csk 0x%p, skb 0x%p,%u, f 0x%lx,"
1773 						" plen %u, NO data.\n",
1774 						csk, skb, skb->len,
1775 						cxgbi_skcb_flags(skb),
1776 						cxgbi_skcb_rx_pdulen(skb));
1777 					err = -EIO;
1778 					goto skb_done;
1779 				}
1780 				__skb_unlink(dskb, &csk->receive_queue);
1781 
1782 				err = skb_read_pdu_data(conn, skb, dskb, 0);
1783 				if (err < 0)
1784 					pr_err("data, csk 0x%p, skb 0x%p,%u, "
1785 						"f 0x%lx, plen %u, dskb 0x%p,"
1786 						"%u.\n",
1787 						csk, skb, skb->len,
1788 						cxgbi_skcb_flags(skb),
1789 						cxgbi_skcb_rx_pdulen(skb),
1790 						dskb, dskb->len);
1791 				__kfree_skb(dskb);
1792 			} else
1793 				err = skb_read_pdu_data(conn, skb, skb, 0);
1794 		}
1795 skb_done:
1796 		__kfree_skb(skb);
1797 
1798 		if (err < 0)
1799 			break;
1800 	}
1801 
1802 	log_debug(1 << CXGBI_DBG_PDU_RX, "csk 0x%p, read %u.\n", csk, read);
1803 	if (read) {
1804 		csk->copied_seq += read;
1805 		csk_return_rx_credits(csk, read);
1806 		conn->rxdata_octets += read;
1807 	}
1808 
1809 	if (err < 0) {
1810 		pr_info("csk 0x%p, 0x%p, rx failed %d, read %u.\n",
1811 			csk, conn, err, read);
1812 		iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1813 	}
1814 }
1815 EXPORT_SYMBOL_GPL(cxgbi_conn_pdu_ready);
1816 
1817 static int sgl_seek_offset(struct scatterlist *sgl, unsigned int sgcnt,
1818 				unsigned int offset, unsigned int *off,
1819 				struct scatterlist **sgp)
1820 {
1821 	int i;
1822 	struct scatterlist *sg;
1823 
1824 	for_each_sg(sgl, sg, sgcnt, i) {
1825 		if (offset < sg->length) {
1826 			*off = offset;
1827 			*sgp = sg;
1828 			return 0;
1829 		}
1830 		offset -= sg->length;
1831 	}
1832 	return -EFAULT;
1833 }
1834 
1835 static int sgl_read_to_frags(struct scatterlist *sg, unsigned int sgoffset,
1836 				unsigned int dlen, struct page_frag *frags,
1837 				int frag_max)
1838 {
1839 	unsigned int datalen = dlen;
1840 	unsigned int sglen = sg->length - sgoffset;
1841 	struct page *page = sg_page(sg);
1842 	int i;
1843 
1844 	i = 0;
1845 	do {
1846 		unsigned int copy;
1847 
1848 		if (!sglen) {
1849 			sg = sg_next(sg);
1850 			if (!sg) {
1851 				pr_warn("sg %d NULL, len %u/%u.\n",
1852 					i, datalen, dlen);
1853 				return -EINVAL;
1854 			}
1855 			sgoffset = 0;
1856 			sglen = sg->length;
1857 			page = sg_page(sg);
1858 
1859 		}
1860 		copy = min(datalen, sglen);
1861 		if (i && page == frags[i - 1].page &&
1862 		    sgoffset + sg->offset ==
1863 			frags[i - 1].offset + frags[i - 1].size) {
1864 			frags[i - 1].size += copy;
1865 		} else {
1866 			if (i >= frag_max) {
1867 				pr_warn("too many pages %u, dlen %u.\n",
1868 					frag_max, dlen);
1869 				return -EINVAL;
1870 			}
1871 
1872 			frags[i].page = page;
1873 			frags[i].offset = sg->offset + sgoffset;
1874 			frags[i].size = copy;
1875 			i++;
1876 		}
1877 		datalen -= copy;
1878 		sgoffset += copy;
1879 		sglen -= copy;
1880 	} while (datalen);
1881 
1882 	return i;
1883 }
1884 
1885 int cxgbi_conn_alloc_pdu(struct iscsi_task *task, u8 opcode)
1886 {
1887 	struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data;
1888 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
1889 	struct cxgbi_device *cdev = cconn->chba->cdev;
1890 	struct iscsi_conn *conn = task->conn;
1891 	struct iscsi_tcp_task *tcp_task = task->dd_data;
1892 	struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
1893 	struct scsi_cmnd *sc = task->sc;
1894 	struct cxgbi_sock *csk = cconn->cep->csk;
1895 	struct net_device *ndev = cdev->ports[csk->port_id];
1896 	int headroom = SKB_TX_ISCSI_PDU_HEADER_MAX;
1897 
1898 	tcp_task->dd_data = tdata;
1899 	task->hdr = NULL;
1900 
1901 	if (SKB_MAX_HEAD(cdev->skb_tx_rsvd) > (512 * MAX_SKB_FRAGS) &&
1902 	    (opcode == ISCSI_OP_SCSI_DATA_OUT ||
1903 	     (opcode == ISCSI_OP_SCSI_CMD &&
1904 	      sc->sc_data_direction == DMA_TO_DEVICE)))
1905 		/* data could goes into skb head */
1906 		headroom += min_t(unsigned int,
1907 				SKB_MAX_HEAD(cdev->skb_tx_rsvd),
1908 				conn->max_xmit_dlength);
1909 
1910 	tdata->skb = alloc_skb(cdev->skb_tx_rsvd + headroom, GFP_ATOMIC);
1911 	if (!tdata->skb) {
1912 		ndev->stats.tx_dropped++;
1913 		return -ENOMEM;
1914 	}
1915 
1916 	skb_reserve(tdata->skb, cdev->skb_tx_rsvd);
1917 
1918 	if (task->sc) {
1919 		task->hdr = (struct iscsi_hdr *)tdata->skb->data;
1920 	} else {
1921 		task->hdr = kzalloc(SKB_TX_ISCSI_PDU_HEADER_MAX, GFP_ATOMIC);
1922 		if (!task->hdr) {
1923 			__kfree_skb(tdata->skb);
1924 			tdata->skb = NULL;
1925 			ndev->stats.tx_dropped++;
1926 			return -ENOMEM;
1927 		}
1928 	}
1929 	task->hdr_max = SKB_TX_ISCSI_PDU_HEADER_MAX; /* BHS + AHS */
1930 
1931 	/* data_out uses scsi_cmd's itt */
1932 	if (opcode != ISCSI_OP_SCSI_DATA_OUT)
1933 		task_reserve_itt(task, &task->hdr->itt);
1934 
1935 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
1936 		"task 0x%p, op 0x%x, skb 0x%p,%u+%u/%u, itt 0x%x.\n",
1937 		task, opcode, tdata->skb, cdev->skb_tx_rsvd, headroom,
1938 		conn->max_xmit_dlength, ntohl(task->hdr->itt));
1939 
1940 	return 0;
1941 }
1942 EXPORT_SYMBOL_GPL(cxgbi_conn_alloc_pdu);
1943 
1944 static inline void tx_skb_setmode(struct sk_buff *skb, int hcrc, int dcrc)
1945 {
1946 	if (hcrc || dcrc) {
1947 		u8 submode = 0;
1948 
1949 		if (hcrc)
1950 			submode |= 1;
1951 		if (dcrc)
1952 			submode |= 2;
1953 		cxgbi_skcb_ulp_mode(skb) = (ULP2_MODE_ISCSI << 4) | submode;
1954 	} else
1955 		cxgbi_skcb_ulp_mode(skb) = 0;
1956 }
1957 
1958 int cxgbi_conn_init_pdu(struct iscsi_task *task, unsigned int offset,
1959 			      unsigned int count)
1960 {
1961 	struct iscsi_conn *conn = task->conn;
1962 	struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
1963 	struct sk_buff *skb = tdata->skb;
1964 	unsigned int datalen = count;
1965 	int i, padlen = iscsi_padding(count);
1966 	struct page *pg;
1967 
1968 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
1969 		"task 0x%p,0x%p, skb 0x%p, 0x%x,0x%x,0x%x, %u+%u.\n",
1970 		task, task->sc, skb, (*skb->data) & ISCSI_OPCODE_MASK,
1971 		ntohl(task->cmdsn), ntohl(task->hdr->itt), offset, count);
1972 
1973 	skb_put(skb, task->hdr_len);
1974 	tx_skb_setmode(skb, conn->hdrdgst_en, datalen ? conn->datadgst_en : 0);
1975 	if (!count)
1976 		return 0;
1977 
1978 	if (task->sc) {
1979 		struct scsi_data_buffer *sdb = &task->sc->sdb;
1980 		struct scatterlist *sg = NULL;
1981 		int err;
1982 
1983 		tdata->offset = offset;
1984 		tdata->count = count;
1985 		err = sgl_seek_offset(
1986 					sdb->table.sgl, sdb->table.nents,
1987 					tdata->offset, &tdata->sgoffset, &sg);
1988 		if (err < 0) {
1989 			pr_warn("tpdu, sgl %u, bad offset %u/%u.\n",
1990 				sdb->table.nents, tdata->offset, sdb->length);
1991 			return err;
1992 		}
1993 		err = sgl_read_to_frags(sg, tdata->sgoffset, tdata->count,
1994 					tdata->frags, MAX_PDU_FRAGS);
1995 		if (err < 0) {
1996 			pr_warn("tpdu, sgl %u, bad offset %u + %u.\n",
1997 				sdb->table.nents, tdata->offset, tdata->count);
1998 			return err;
1999 		}
2000 		tdata->nr_frags = err;
2001 
2002 		if (tdata->nr_frags > MAX_SKB_FRAGS ||
2003 		    (padlen && tdata->nr_frags == MAX_SKB_FRAGS)) {
2004 			char *dst = skb->data + task->hdr_len;
2005 			struct page_frag *frag = tdata->frags;
2006 
2007 			/* data fits in the skb's headroom */
2008 			for (i = 0; i < tdata->nr_frags; i++, frag++) {
2009 				char *src = kmap_atomic(frag->page);
2010 
2011 				memcpy(dst, src+frag->offset, frag->size);
2012 				dst += frag->size;
2013 				kunmap_atomic(src);
2014 			}
2015 			if (padlen) {
2016 				memset(dst, 0, padlen);
2017 				padlen = 0;
2018 			}
2019 			skb_put(skb, count + padlen);
2020 		} else {
2021 			/* data fit into frag_list */
2022 			for (i = 0; i < tdata->nr_frags; i++) {
2023 				__skb_fill_page_desc(skb, i,
2024 						tdata->frags[i].page,
2025 						tdata->frags[i].offset,
2026 						tdata->frags[i].size);
2027 				skb_frag_ref(skb, i);
2028 			}
2029 			skb_shinfo(skb)->nr_frags = tdata->nr_frags;
2030 			skb->len += count;
2031 			skb->data_len += count;
2032 			skb->truesize += count;
2033 		}
2034 
2035 	} else {
2036 		pg = virt_to_page(task->data);
2037 
2038 		get_page(pg);
2039 		skb_fill_page_desc(skb, 0, pg, offset_in_page(task->data),
2040 					count);
2041 		skb->len += count;
2042 		skb->data_len += count;
2043 		skb->truesize += count;
2044 	}
2045 
2046 	if (padlen) {
2047 		i = skb_shinfo(skb)->nr_frags;
2048 		skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
2049 				virt_to_page(padding), offset_in_page(padding),
2050 				padlen);
2051 
2052 		skb->data_len += padlen;
2053 		skb->truesize += padlen;
2054 		skb->len += padlen;
2055 	}
2056 
2057 	return 0;
2058 }
2059 EXPORT_SYMBOL_GPL(cxgbi_conn_init_pdu);
2060 
2061 int cxgbi_conn_xmit_pdu(struct iscsi_task *task)
2062 {
2063 	struct iscsi_tcp_conn *tcp_conn = task->conn->dd_data;
2064 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2065 	struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
2066 	struct cxgbi_task_tag_info *ttinfo = &tdata->ttinfo;
2067 	struct sk_buff *skb = tdata->skb;
2068 	struct cxgbi_sock *csk = NULL;
2069 	unsigned int datalen;
2070 	int err;
2071 
2072 	if (!skb) {
2073 		log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
2074 			"task 0x%p\n", task);
2075 		return 0;
2076 	}
2077 
2078 	if (cconn && cconn->cep)
2079 		csk = cconn->cep->csk;
2080 	if (!csk) {
2081 		log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
2082 			  "task 0x%p, csk gone.\n", task);
2083 		return -EPIPE;
2084 	}
2085 
2086 	tdata->skb = NULL;
2087 	datalen = skb->data_len;
2088 
2089 	/* write ppod first if using ofldq to write ppod */
2090 	if (ttinfo->flags & CXGBI_PPOD_INFO_FLAG_VALID) {
2091 		struct cxgbi_ppm *ppm = csk->cdev->cdev2ppm(csk->cdev);
2092 
2093 		ttinfo->flags &= ~CXGBI_PPOD_INFO_FLAG_VALID;
2094 		if (csk->cdev->csk_ddp_set_map(ppm, csk, ttinfo) < 0)
2095 			pr_err("task 0x%p, ppod writing using ofldq failed.\n",
2096 			       task);
2097 			/* continue. Let fl get the data */
2098 	}
2099 
2100 	if (!task->sc)
2101 		memcpy(skb->data, task->hdr, SKB_TX_ISCSI_PDU_HEADER_MAX);
2102 
2103 	err = cxgbi_sock_send_pdus(cconn->cep->csk, skb);
2104 	if (err > 0) {
2105 		int pdulen = err;
2106 
2107 		log_debug(1 << CXGBI_DBG_PDU_TX,
2108 			"task 0x%p,0x%p, skb 0x%p, len %u/%u, rv %d.\n",
2109 			task, task->sc, skb, skb->len, skb->data_len, err);
2110 
2111 		if (task->conn->hdrdgst_en)
2112 			pdulen += ISCSI_DIGEST_SIZE;
2113 
2114 		if (datalen && task->conn->datadgst_en)
2115 			pdulen += ISCSI_DIGEST_SIZE;
2116 
2117 		task->conn->txdata_octets += pdulen;
2118 		return 0;
2119 	}
2120 
2121 	if (err == -EAGAIN || err == -ENOBUFS) {
2122 		log_debug(1 << CXGBI_DBG_PDU_TX,
2123 			"task 0x%p, skb 0x%p, len %u/%u, %d EAGAIN.\n",
2124 			task, skb, skb->len, skb->data_len, err);
2125 		/* reset skb to send when we are called again */
2126 		tdata->skb = skb;
2127 		return err;
2128 	}
2129 
2130 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_PDU_TX,
2131 		"itt 0x%x, skb 0x%p, len %u/%u, xmit err %d.\n",
2132 		task->itt, skb, skb->len, skb->data_len, err);
2133 
2134 	__kfree_skb(skb);
2135 
2136 	iscsi_conn_printk(KERN_ERR, task->conn, "xmit err %d.\n", err);
2137 	iscsi_conn_failure(task->conn, ISCSI_ERR_XMIT_FAILED);
2138 	return err;
2139 }
2140 EXPORT_SYMBOL_GPL(cxgbi_conn_xmit_pdu);
2141 
2142 void cxgbi_cleanup_task(struct iscsi_task *task)
2143 {
2144 	struct iscsi_tcp_task *tcp_task = task->dd_data;
2145 	struct cxgbi_task_data *tdata = iscsi_task_cxgbi_data(task);
2146 
2147 	if (!tcp_task || !tdata || (tcp_task->dd_data != tdata)) {
2148 		pr_info("task 0x%p,0x%p, tcp_task 0x%p, tdata 0x%p/0x%p.\n",
2149 			task, task->sc, tcp_task,
2150 			tcp_task ? tcp_task->dd_data : NULL, tdata);
2151 		return;
2152 	}
2153 
2154 	log_debug(1 << CXGBI_DBG_ISCSI,
2155 		"task 0x%p, skb 0x%p, itt 0x%x.\n",
2156 		task, tdata->skb, task->hdr_itt);
2157 
2158 	tcp_task->dd_data = NULL;
2159 
2160 	if (!task->sc)
2161 		kfree(task->hdr);
2162 	task->hdr = NULL;
2163 
2164 	/*  never reached the xmit task callout */
2165 	if (tdata->skb) {
2166 		__kfree_skb(tdata->skb);
2167 		tdata->skb = NULL;
2168 	}
2169 
2170 	task_release_itt(task, task->hdr_itt);
2171 	memset(tdata, 0, sizeof(*tdata));
2172 
2173 	iscsi_tcp_cleanup_task(task);
2174 }
2175 EXPORT_SYMBOL_GPL(cxgbi_cleanup_task);
2176 
2177 void cxgbi_get_conn_stats(struct iscsi_cls_conn *cls_conn,
2178 				struct iscsi_stats *stats)
2179 {
2180 	struct iscsi_conn *conn = cls_conn->dd_data;
2181 
2182 	stats->txdata_octets = conn->txdata_octets;
2183 	stats->rxdata_octets = conn->rxdata_octets;
2184 	stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
2185 	stats->dataout_pdus = conn->dataout_pdus_cnt;
2186 	stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
2187 	stats->datain_pdus = conn->datain_pdus_cnt;
2188 	stats->r2t_pdus = conn->r2t_pdus_cnt;
2189 	stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
2190 	stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
2191 	stats->digest_err = 0;
2192 	stats->timeout_err = 0;
2193 	stats->custom_length = 1;
2194 	strcpy(stats->custom[0].desc, "eh_abort_cnt");
2195 	stats->custom[0].value = conn->eh_abort_cnt;
2196 }
2197 EXPORT_SYMBOL_GPL(cxgbi_get_conn_stats);
2198 
2199 static int cxgbi_conn_max_xmit_dlength(struct iscsi_conn *conn)
2200 {
2201 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2202 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2203 	struct cxgbi_device *cdev = cconn->chba->cdev;
2204 	unsigned int headroom = SKB_MAX_HEAD(cdev->skb_tx_rsvd);
2205 	unsigned int max_def = 512 * MAX_SKB_FRAGS;
2206 	unsigned int max = max(max_def, headroom);
2207 
2208 	max = min(cconn->chba->cdev->tx_max_size, max);
2209 	if (conn->max_xmit_dlength)
2210 		conn->max_xmit_dlength = min(conn->max_xmit_dlength, max);
2211 	else
2212 		conn->max_xmit_dlength = max;
2213 	cxgbi_align_pdu_size(conn->max_xmit_dlength);
2214 
2215 	return 0;
2216 }
2217 
2218 static int cxgbi_conn_max_recv_dlength(struct iscsi_conn *conn)
2219 {
2220 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2221 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2222 	unsigned int max = cconn->chba->cdev->rx_max_size;
2223 
2224 	cxgbi_align_pdu_size(max);
2225 
2226 	if (conn->max_recv_dlength) {
2227 		if (conn->max_recv_dlength > max) {
2228 			pr_err("MaxRecvDataSegmentLength %u > %u.\n",
2229 				conn->max_recv_dlength, max);
2230 			return -EINVAL;
2231 		}
2232 		conn->max_recv_dlength = min(conn->max_recv_dlength, max);
2233 		cxgbi_align_pdu_size(conn->max_recv_dlength);
2234 	} else
2235 		conn->max_recv_dlength = max;
2236 
2237 	return 0;
2238 }
2239 
2240 int cxgbi_set_conn_param(struct iscsi_cls_conn *cls_conn,
2241 			enum iscsi_param param, char *buf, int buflen)
2242 {
2243 	struct iscsi_conn *conn = cls_conn->dd_data;
2244 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2245 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2246 	struct cxgbi_sock *csk = cconn->cep->csk;
2247 	int err;
2248 
2249 	log_debug(1 << CXGBI_DBG_ISCSI,
2250 		"cls_conn 0x%p, param %d, buf(%d) %s.\n",
2251 		cls_conn, param, buflen, buf);
2252 
2253 	switch (param) {
2254 	case ISCSI_PARAM_HDRDGST_EN:
2255 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2256 		if (!err && conn->hdrdgst_en)
2257 			err = csk->cdev->csk_ddp_setup_digest(csk, csk->tid,
2258 							conn->hdrdgst_en,
2259 							conn->datadgst_en);
2260 		break;
2261 	case ISCSI_PARAM_DATADGST_EN:
2262 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2263 		if (!err && conn->datadgst_en)
2264 			err = csk->cdev->csk_ddp_setup_digest(csk, csk->tid,
2265 							conn->hdrdgst_en,
2266 							conn->datadgst_en);
2267 		break;
2268 	case ISCSI_PARAM_MAX_R2T:
2269 		return iscsi_tcp_set_max_r2t(conn, buf);
2270 	case ISCSI_PARAM_MAX_RECV_DLENGTH:
2271 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2272 		if (!err)
2273 			err = cxgbi_conn_max_recv_dlength(conn);
2274 		break;
2275 	case ISCSI_PARAM_MAX_XMIT_DLENGTH:
2276 		err = iscsi_set_param(cls_conn, param, buf, buflen);
2277 		if (!err)
2278 			err = cxgbi_conn_max_xmit_dlength(conn);
2279 		break;
2280 	default:
2281 		return iscsi_set_param(cls_conn, param, buf, buflen);
2282 	}
2283 	return err;
2284 }
2285 EXPORT_SYMBOL_GPL(cxgbi_set_conn_param);
2286 
2287 static inline int csk_print_port(struct cxgbi_sock *csk, char *buf)
2288 {
2289 	int len;
2290 
2291 	cxgbi_sock_get(csk);
2292 	len = sprintf(buf, "%hu\n", ntohs(csk->daddr.sin_port));
2293 	cxgbi_sock_put(csk);
2294 
2295 	return len;
2296 }
2297 
2298 static inline int csk_print_ip(struct cxgbi_sock *csk, char *buf)
2299 {
2300 	int len;
2301 
2302 	cxgbi_sock_get(csk);
2303 	if (csk->csk_family == AF_INET)
2304 		len = sprintf(buf, "%pI4",
2305 			      &csk->daddr.sin_addr.s_addr);
2306 	else
2307 		len = sprintf(buf, "%pI6",
2308 			      &csk->daddr6.sin6_addr);
2309 
2310 	cxgbi_sock_put(csk);
2311 
2312 	return len;
2313 }
2314 
2315 int cxgbi_get_ep_param(struct iscsi_endpoint *ep, enum iscsi_param param,
2316 		       char *buf)
2317 {
2318 	struct cxgbi_endpoint *cep = ep->dd_data;
2319 	struct cxgbi_sock *csk;
2320 
2321 	log_debug(1 << CXGBI_DBG_ISCSI,
2322 		"cls_conn 0x%p, param %d.\n", ep, param);
2323 
2324 	switch (param) {
2325 	case ISCSI_PARAM_CONN_PORT:
2326 	case ISCSI_PARAM_CONN_ADDRESS:
2327 		if (!cep)
2328 			return -ENOTCONN;
2329 
2330 		csk = cep->csk;
2331 		if (!csk)
2332 			return -ENOTCONN;
2333 
2334 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
2335 						 &csk->daddr, param, buf);
2336 	default:
2337 		break;
2338 	}
2339 	return -ENOSYS;
2340 }
2341 EXPORT_SYMBOL_GPL(cxgbi_get_ep_param);
2342 
2343 struct iscsi_cls_conn *
2344 cxgbi_create_conn(struct iscsi_cls_session *cls_session, u32 cid)
2345 {
2346 	struct iscsi_cls_conn *cls_conn;
2347 	struct iscsi_conn *conn;
2348 	struct iscsi_tcp_conn *tcp_conn;
2349 	struct cxgbi_conn *cconn;
2350 
2351 	cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*cconn), cid);
2352 	if (!cls_conn)
2353 		return NULL;
2354 
2355 	conn = cls_conn->dd_data;
2356 	tcp_conn = conn->dd_data;
2357 	cconn = tcp_conn->dd_data;
2358 	cconn->iconn = conn;
2359 
2360 	log_debug(1 << CXGBI_DBG_ISCSI,
2361 		"cid %u(0x%x), cls 0x%p,0x%p, conn 0x%p,0x%p,0x%p.\n",
2362 		cid, cid, cls_session, cls_conn, conn, tcp_conn, cconn);
2363 
2364 	return cls_conn;
2365 }
2366 EXPORT_SYMBOL_GPL(cxgbi_create_conn);
2367 
2368 int cxgbi_bind_conn(struct iscsi_cls_session *cls_session,
2369 				struct iscsi_cls_conn *cls_conn,
2370 				u64 transport_eph, int is_leading)
2371 {
2372 	struct iscsi_conn *conn = cls_conn->dd_data;
2373 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
2374 	struct cxgbi_conn *cconn = tcp_conn->dd_data;
2375 	struct cxgbi_ppm *ppm;
2376 	struct iscsi_endpoint *ep;
2377 	struct cxgbi_endpoint *cep;
2378 	struct cxgbi_sock *csk;
2379 	int err;
2380 
2381 	ep = iscsi_lookup_endpoint(transport_eph);
2382 	if (!ep)
2383 		return -EINVAL;
2384 
2385 	/*  setup ddp pagesize */
2386 	cep = ep->dd_data;
2387 	csk = cep->csk;
2388 
2389 	ppm = csk->cdev->cdev2ppm(csk->cdev);
2390 	err = csk->cdev->csk_ddp_setup_pgidx(csk, csk->tid,
2391 					     ppm->tformat.pgsz_idx_dflt);
2392 	if (err < 0)
2393 		return err;
2394 
2395 	err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
2396 	if (err)
2397 		return -EINVAL;
2398 
2399 	/*  calculate the tag idx bits needed for this conn based on cmds_max */
2400 	cconn->task_idx_bits = (__ilog2_u32(conn->session->cmds_max - 1)) + 1;
2401 
2402 	write_lock_bh(&csk->callback_lock);
2403 	csk->user_data = conn;
2404 	cconn->chba = cep->chba;
2405 	cconn->cep = cep;
2406 	cep->cconn = cconn;
2407 	write_unlock_bh(&csk->callback_lock);
2408 
2409 	cxgbi_conn_max_xmit_dlength(conn);
2410 	cxgbi_conn_max_recv_dlength(conn);
2411 
2412 	log_debug(1 << CXGBI_DBG_ISCSI,
2413 		"cls 0x%p,0x%p, ep 0x%p, cconn 0x%p, csk 0x%p.\n",
2414 		cls_session, cls_conn, ep, cconn, csk);
2415 	/*  init recv engine */
2416 	iscsi_tcp_hdr_recv_prep(tcp_conn);
2417 
2418 	return 0;
2419 }
2420 EXPORT_SYMBOL_GPL(cxgbi_bind_conn);
2421 
2422 struct iscsi_cls_session *cxgbi_create_session(struct iscsi_endpoint *ep,
2423 						u16 cmds_max, u16 qdepth,
2424 						u32 initial_cmdsn)
2425 {
2426 	struct cxgbi_endpoint *cep;
2427 	struct cxgbi_hba *chba;
2428 	struct Scsi_Host *shost;
2429 	struct iscsi_cls_session *cls_session;
2430 	struct iscsi_session *session;
2431 
2432 	if (!ep) {
2433 		pr_err("missing endpoint.\n");
2434 		return NULL;
2435 	}
2436 
2437 	cep = ep->dd_data;
2438 	chba = cep->chba;
2439 	shost = chba->shost;
2440 
2441 	BUG_ON(chba != iscsi_host_priv(shost));
2442 
2443 	cls_session = iscsi_session_setup(chba->cdev->itp, shost,
2444 					cmds_max, 0,
2445 					sizeof(struct iscsi_tcp_task) +
2446 					sizeof(struct cxgbi_task_data),
2447 					initial_cmdsn, ISCSI_MAX_TARGET);
2448 	if (!cls_session)
2449 		return NULL;
2450 
2451 	session = cls_session->dd_data;
2452 	if (iscsi_tcp_r2tpool_alloc(session))
2453 		goto remove_session;
2454 
2455 	log_debug(1 << CXGBI_DBG_ISCSI,
2456 		"ep 0x%p, cls sess 0x%p.\n", ep, cls_session);
2457 	return cls_session;
2458 
2459 remove_session:
2460 	iscsi_session_teardown(cls_session);
2461 	return NULL;
2462 }
2463 EXPORT_SYMBOL_GPL(cxgbi_create_session);
2464 
2465 void cxgbi_destroy_session(struct iscsi_cls_session *cls_session)
2466 {
2467 	log_debug(1 << CXGBI_DBG_ISCSI,
2468 		"cls sess 0x%p.\n", cls_session);
2469 
2470 	iscsi_tcp_r2tpool_free(cls_session->dd_data);
2471 	iscsi_session_teardown(cls_session);
2472 }
2473 EXPORT_SYMBOL_GPL(cxgbi_destroy_session);
2474 
2475 int cxgbi_set_host_param(struct Scsi_Host *shost, enum iscsi_host_param param,
2476 			char *buf, int buflen)
2477 {
2478 	struct cxgbi_hba *chba = iscsi_host_priv(shost);
2479 
2480 	if (!chba->ndev) {
2481 		shost_printk(KERN_ERR, shost, "Could not get host param. "
2482 				"netdev for host not set.\n");
2483 		return -ENODEV;
2484 	}
2485 
2486 	log_debug(1 << CXGBI_DBG_ISCSI,
2487 		"shost 0x%p, hba 0x%p,%s, param %d, buf(%d) %s.\n",
2488 		shost, chba, chba->ndev->name, param, buflen, buf);
2489 
2490 	switch (param) {
2491 	case ISCSI_HOST_PARAM_IPADDRESS:
2492 	{
2493 		__be32 addr = in_aton(buf);
2494 		log_debug(1 << CXGBI_DBG_ISCSI,
2495 			"hba %s, req. ipv4 %pI4.\n", chba->ndev->name, &addr);
2496 		cxgbi_set_iscsi_ipv4(chba, addr);
2497 		return 0;
2498 	}
2499 	case ISCSI_HOST_PARAM_HWADDRESS:
2500 	case ISCSI_HOST_PARAM_NETDEV_NAME:
2501 		return 0;
2502 	default:
2503 		return iscsi_host_set_param(shost, param, buf, buflen);
2504 	}
2505 }
2506 EXPORT_SYMBOL_GPL(cxgbi_set_host_param);
2507 
2508 int cxgbi_get_host_param(struct Scsi_Host *shost, enum iscsi_host_param param,
2509 			char *buf)
2510 {
2511 	struct cxgbi_hba *chba = iscsi_host_priv(shost);
2512 	int len = 0;
2513 
2514 	if (!chba->ndev) {
2515 		shost_printk(KERN_ERR, shost, "Could not get host param. "
2516 				"netdev for host not set.\n");
2517 		return -ENODEV;
2518 	}
2519 
2520 	log_debug(1 << CXGBI_DBG_ISCSI,
2521 		"shost 0x%p, hba 0x%p,%s, param %d.\n",
2522 		shost, chba, chba->ndev->name, param);
2523 
2524 	switch (param) {
2525 	case ISCSI_HOST_PARAM_HWADDRESS:
2526 		len = sysfs_format_mac(buf, chba->ndev->dev_addr, 6);
2527 		break;
2528 	case ISCSI_HOST_PARAM_NETDEV_NAME:
2529 		len = sprintf(buf, "%s\n", chba->ndev->name);
2530 		break;
2531 	case ISCSI_HOST_PARAM_IPADDRESS:
2532 	{
2533 		struct cxgbi_sock *csk = find_sock_on_port(chba->cdev,
2534 							   chba->port_id);
2535 		if (csk) {
2536 			len = sprintf(buf, "%pIS",
2537 				      (struct sockaddr *)&csk->saddr);
2538 		}
2539 		log_debug(1 << CXGBI_DBG_ISCSI,
2540 			  "hba %s, addr %s.\n", chba->ndev->name, buf);
2541 		break;
2542 	}
2543 	default:
2544 		return iscsi_host_get_param(shost, param, buf);
2545 	}
2546 
2547 	return len;
2548 }
2549 EXPORT_SYMBOL_GPL(cxgbi_get_host_param);
2550 
2551 struct iscsi_endpoint *cxgbi_ep_connect(struct Scsi_Host *shost,
2552 					struct sockaddr *dst_addr,
2553 					int non_blocking)
2554 {
2555 	struct iscsi_endpoint *ep;
2556 	struct cxgbi_endpoint *cep;
2557 	struct cxgbi_hba *hba = NULL;
2558 	struct cxgbi_sock *csk;
2559 	int ifindex = 0;
2560 	int err = -EINVAL;
2561 
2562 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_SOCK,
2563 		"shost 0x%p, non_blocking %d, dst_addr 0x%p.\n",
2564 		shost, non_blocking, dst_addr);
2565 
2566 	if (shost) {
2567 		hba = iscsi_host_priv(shost);
2568 		if (!hba) {
2569 			pr_info("shost 0x%p, priv NULL.\n", shost);
2570 			goto err_out;
2571 		}
2572 	}
2573 
2574 check_route:
2575 	if (dst_addr->sa_family == AF_INET) {
2576 		csk = cxgbi_check_route(dst_addr, ifindex);
2577 #if IS_ENABLED(CONFIG_IPV6)
2578 	} else if (dst_addr->sa_family == AF_INET6) {
2579 		csk = cxgbi_check_route6(dst_addr, ifindex);
2580 #endif
2581 	} else {
2582 		pr_info("address family 0x%x NOT supported.\n",
2583 			dst_addr->sa_family);
2584 		err = -EAFNOSUPPORT;
2585 		return (struct iscsi_endpoint *)ERR_PTR(err);
2586 	}
2587 
2588 	if (IS_ERR(csk))
2589 		return (struct iscsi_endpoint *)csk;
2590 	cxgbi_sock_get(csk);
2591 
2592 	if (!hba)
2593 		hba = csk->cdev->hbas[csk->port_id];
2594 	else if (hba != csk->cdev->hbas[csk->port_id]) {
2595 		if (ifindex != hba->ndev->ifindex) {
2596 			cxgbi_sock_put(csk);
2597 			cxgbi_sock_closed(csk);
2598 			ifindex = hba->ndev->ifindex;
2599 			goto check_route;
2600 		}
2601 
2602 		pr_info("Could not connect through requested host %u"
2603 			"hba 0x%p != 0x%p (%u).\n",
2604 			shost->host_no, hba,
2605 			csk->cdev->hbas[csk->port_id], csk->port_id);
2606 		err = -ENOSPC;
2607 		goto release_conn;
2608 	}
2609 
2610 	err = sock_get_port(csk);
2611 	if (err)
2612 		goto release_conn;
2613 
2614 	cxgbi_sock_set_state(csk, CTP_CONNECTING);
2615 	err = csk->cdev->csk_init_act_open(csk);
2616 	if (err)
2617 		goto release_conn;
2618 
2619 	if (cxgbi_sock_is_closing(csk)) {
2620 		err = -ENOSPC;
2621 		pr_info("csk 0x%p is closing.\n", csk);
2622 		goto release_conn;
2623 	}
2624 
2625 	ep = iscsi_create_endpoint(sizeof(*cep));
2626 	if (!ep) {
2627 		err = -ENOMEM;
2628 		pr_info("iscsi alloc ep, OOM.\n");
2629 		goto release_conn;
2630 	}
2631 
2632 	cep = ep->dd_data;
2633 	cep->csk = csk;
2634 	cep->chba = hba;
2635 
2636 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_SOCK,
2637 		"ep 0x%p, cep 0x%p, csk 0x%p, hba 0x%p,%s.\n",
2638 		ep, cep, csk, hba, hba->ndev->name);
2639 	return ep;
2640 
2641 release_conn:
2642 	cxgbi_sock_put(csk);
2643 	cxgbi_sock_closed(csk);
2644 err_out:
2645 	return ERR_PTR(err);
2646 }
2647 EXPORT_SYMBOL_GPL(cxgbi_ep_connect);
2648 
2649 int cxgbi_ep_poll(struct iscsi_endpoint *ep, int timeout_ms)
2650 {
2651 	struct cxgbi_endpoint *cep = ep->dd_data;
2652 	struct cxgbi_sock *csk = cep->csk;
2653 
2654 	if (!cxgbi_sock_is_established(csk))
2655 		return 0;
2656 	return 1;
2657 }
2658 EXPORT_SYMBOL_GPL(cxgbi_ep_poll);
2659 
2660 void cxgbi_ep_disconnect(struct iscsi_endpoint *ep)
2661 {
2662 	struct cxgbi_endpoint *cep = ep->dd_data;
2663 	struct cxgbi_conn *cconn = cep->cconn;
2664 	struct cxgbi_sock *csk = cep->csk;
2665 
2666 	log_debug(1 << CXGBI_DBG_ISCSI | 1 << CXGBI_DBG_SOCK,
2667 		"ep 0x%p, cep 0x%p, cconn 0x%p, csk 0x%p,%u,0x%lx.\n",
2668 		ep, cep, cconn, csk, csk->state, csk->flags);
2669 
2670 	if (cconn && cconn->iconn) {
2671 		iscsi_suspend_tx(cconn->iconn);
2672 		write_lock_bh(&csk->callback_lock);
2673 		cep->csk->user_data = NULL;
2674 		cconn->cep = NULL;
2675 		write_unlock_bh(&csk->callback_lock);
2676 	}
2677 	iscsi_destroy_endpoint(ep);
2678 
2679 	if (likely(csk->state >= CTP_ESTABLISHED))
2680 		need_active_close(csk);
2681 	else
2682 		cxgbi_sock_closed(csk);
2683 
2684 	cxgbi_sock_put(csk);
2685 }
2686 EXPORT_SYMBOL_GPL(cxgbi_ep_disconnect);
2687 
2688 int cxgbi_iscsi_init(struct iscsi_transport *itp,
2689 			struct scsi_transport_template **stt)
2690 {
2691 	*stt = iscsi_register_transport(itp);
2692 	if (*stt == NULL) {
2693 		pr_err("unable to register %s transport 0x%p.\n",
2694 			itp->name, itp);
2695 		return -ENODEV;
2696 	}
2697 	log_debug(1 << CXGBI_DBG_ISCSI,
2698 		"%s, registered iscsi transport 0x%p.\n",
2699 		itp->name, stt);
2700 	return 0;
2701 }
2702 EXPORT_SYMBOL_GPL(cxgbi_iscsi_init);
2703 
2704 void cxgbi_iscsi_cleanup(struct iscsi_transport *itp,
2705 			struct scsi_transport_template **stt)
2706 {
2707 	if (*stt) {
2708 		log_debug(1 << CXGBI_DBG_ISCSI,
2709 			"de-register transport 0x%p, %s, stt 0x%p.\n",
2710 			itp, itp->name, *stt);
2711 		*stt = NULL;
2712 		iscsi_unregister_transport(itp);
2713 	}
2714 }
2715 EXPORT_SYMBOL_GPL(cxgbi_iscsi_cleanup);
2716 
2717 umode_t cxgbi_attr_is_visible(int param_type, int param)
2718 {
2719 	switch (param_type) {
2720 	case ISCSI_HOST_PARAM:
2721 		switch (param) {
2722 		case ISCSI_HOST_PARAM_NETDEV_NAME:
2723 		case ISCSI_HOST_PARAM_HWADDRESS:
2724 		case ISCSI_HOST_PARAM_IPADDRESS:
2725 		case ISCSI_HOST_PARAM_INITIATOR_NAME:
2726 			return S_IRUGO;
2727 		default:
2728 			return 0;
2729 		}
2730 	case ISCSI_PARAM:
2731 		switch (param) {
2732 		case ISCSI_PARAM_MAX_RECV_DLENGTH:
2733 		case ISCSI_PARAM_MAX_XMIT_DLENGTH:
2734 		case ISCSI_PARAM_HDRDGST_EN:
2735 		case ISCSI_PARAM_DATADGST_EN:
2736 		case ISCSI_PARAM_CONN_ADDRESS:
2737 		case ISCSI_PARAM_CONN_PORT:
2738 		case ISCSI_PARAM_EXP_STATSN:
2739 		case ISCSI_PARAM_PERSISTENT_ADDRESS:
2740 		case ISCSI_PARAM_PERSISTENT_PORT:
2741 		case ISCSI_PARAM_PING_TMO:
2742 		case ISCSI_PARAM_RECV_TMO:
2743 		case ISCSI_PARAM_INITIAL_R2T_EN:
2744 		case ISCSI_PARAM_MAX_R2T:
2745 		case ISCSI_PARAM_IMM_DATA_EN:
2746 		case ISCSI_PARAM_FIRST_BURST:
2747 		case ISCSI_PARAM_MAX_BURST:
2748 		case ISCSI_PARAM_PDU_INORDER_EN:
2749 		case ISCSI_PARAM_DATASEQ_INORDER_EN:
2750 		case ISCSI_PARAM_ERL:
2751 		case ISCSI_PARAM_TARGET_NAME:
2752 		case ISCSI_PARAM_TPGT:
2753 		case ISCSI_PARAM_USERNAME:
2754 		case ISCSI_PARAM_PASSWORD:
2755 		case ISCSI_PARAM_USERNAME_IN:
2756 		case ISCSI_PARAM_PASSWORD_IN:
2757 		case ISCSI_PARAM_FAST_ABORT:
2758 		case ISCSI_PARAM_ABORT_TMO:
2759 		case ISCSI_PARAM_LU_RESET_TMO:
2760 		case ISCSI_PARAM_TGT_RESET_TMO:
2761 		case ISCSI_PARAM_IFACE_NAME:
2762 		case ISCSI_PARAM_INITIATOR_NAME:
2763 			return S_IRUGO;
2764 		default:
2765 			return 0;
2766 		}
2767 	}
2768 
2769 	return 0;
2770 }
2771 EXPORT_SYMBOL_GPL(cxgbi_attr_is_visible);
2772 
2773 static int __init libcxgbi_init_module(void)
2774 {
2775 	pr_info("%s", version);
2776 
2777 	BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, cb) <
2778 		     sizeof(struct cxgbi_skb_cb));
2779 	return 0;
2780 }
2781 
2782 static void __exit libcxgbi_exit_module(void)
2783 {
2784 	cxgbi_device_unregister_all(0xFF);
2785 	return;
2786 }
2787 
2788 module_init(libcxgbi_init_module);
2789 module_exit(libcxgbi_exit_module);
2790