xref: /openbmc/linux/drivers/scsi/iscsi_tcp.c (revision b04b4f78)
1 /*
2  * iSCSI Initiator over TCP/IP Data-Path
3  *
4  * Copyright (C) 2004 Dmitry Yusupov
5  * Copyright (C) 2004 Alex Aizman
6  * Copyright (C) 2005 - 2006 Mike Christie
7  * Copyright (C) 2006 Red Hat, Inc.  All rights reserved.
8  * maintained by open-iscsi@googlegroups.com
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published
12  * by the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful, but
16  * WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18  * General Public License for more details.
19  *
20  * See the file COPYING included with this distribution for more details.
21  *
22  * Credits:
23  *	Christoph Hellwig
24  *	FUJITA Tomonori
25  *	Arne Redlich
26  *	Zhenyu Wang
27  */
28 
29 #include <linux/types.h>
30 #include <linux/inet.h>
31 #include <linux/file.h>
32 #include <linux/blkdev.h>
33 #include <linux/crypto.h>
34 #include <linux/delay.h>
35 #include <linux/kfifo.h>
36 #include <linux/scatterlist.h>
37 #include <net/tcp.h>
38 #include <scsi/scsi_cmnd.h>
39 #include <scsi/scsi_device.h>
40 #include <scsi/scsi_host.h>
41 #include <scsi/scsi.h>
42 #include <scsi/scsi_transport_iscsi.h>
43 
44 #include "iscsi_tcp.h"
45 
46 MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
47 	      "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
48 	      "Alex Aizman <itn780@yahoo.com>");
49 MODULE_DESCRIPTION("iSCSI/TCP data-path");
50 MODULE_LICENSE("GPL");
51 
52 static struct scsi_transport_template *iscsi_sw_tcp_scsi_transport;
53 static struct scsi_host_template iscsi_sw_tcp_sht;
54 static struct iscsi_transport iscsi_sw_tcp_transport;
55 
56 static unsigned int iscsi_max_lun = 512;
57 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
58 
59 static int iscsi_sw_tcp_dbg;
60 module_param_named(debug_iscsi_tcp, iscsi_sw_tcp_dbg, int,
61 		   S_IRUGO | S_IWUSR);
62 MODULE_PARM_DESC(debug_iscsi_tcp, "Turn on debugging for iscsi_tcp module "
63 		 "Set to 1 to turn on, and zero to turn off. Default is off.");
64 
65 #define ISCSI_SW_TCP_DBG(_conn, dbg_fmt, arg...)		\
66 	do {							\
67 		if (iscsi_sw_tcp_dbg)				\
68 			iscsi_conn_printk(KERN_INFO, _conn,	\
69 					     "%s " dbg_fmt,	\
70 					     __func__, ##arg);	\
71 	} while (0);
72 
73 
74 /**
75  * iscsi_sw_tcp_recv - TCP receive in sendfile fashion
76  * @rd_desc: read descriptor
77  * @skb: socket buffer
78  * @offset: offset in skb
79  * @len: skb->len - offset
80  */
81 static int iscsi_sw_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
82 			     unsigned int offset, size_t len)
83 {
84 	struct iscsi_conn *conn = rd_desc->arg.data;
85 	unsigned int consumed, total_consumed = 0;
86 	int status;
87 
88 	ISCSI_SW_TCP_DBG(conn, "in %d bytes\n", skb->len - offset);
89 
90 	do {
91 		status = 0;
92 		consumed = iscsi_tcp_recv_skb(conn, skb, offset, 0, &status);
93 		offset += consumed;
94 		total_consumed += consumed;
95 	} while (consumed != 0 && status != ISCSI_TCP_SKB_DONE);
96 
97 	ISCSI_SW_TCP_DBG(conn, "read %d bytes status %d\n",
98 			 skb->len - offset, status);
99 	return total_consumed;
100 }
101 
102 static void iscsi_sw_tcp_data_ready(struct sock *sk, int flag)
103 {
104 	struct iscsi_conn *conn = sk->sk_user_data;
105 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
106 	read_descriptor_t rd_desc;
107 
108 	read_lock(&sk->sk_callback_lock);
109 
110 	/*
111 	 * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
112 	 * We set count to 1 because we want the network layer to
113 	 * hand us all the skbs that are available. iscsi_tcp_recv
114 	 * handled pdus that cross buffers or pdus that still need data.
115 	 */
116 	rd_desc.arg.data = conn;
117 	rd_desc.count = 1;
118 	tcp_read_sock(sk, &rd_desc, iscsi_sw_tcp_recv);
119 
120 	read_unlock(&sk->sk_callback_lock);
121 
122 	/* If we had to (atomically) map a highmem page,
123 	 * unmap it now. */
124 	iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
125 }
126 
127 static void iscsi_sw_tcp_state_change(struct sock *sk)
128 {
129 	struct iscsi_tcp_conn *tcp_conn;
130 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
131 	struct iscsi_conn *conn;
132 	struct iscsi_session *session;
133 	void (*old_state_change)(struct sock *);
134 
135 	read_lock(&sk->sk_callback_lock);
136 
137 	conn = (struct iscsi_conn*)sk->sk_user_data;
138 	session = conn->session;
139 
140 	if ((sk->sk_state == TCP_CLOSE_WAIT ||
141 	     sk->sk_state == TCP_CLOSE) &&
142 	    !atomic_read(&sk->sk_rmem_alloc)) {
143 		ISCSI_SW_TCP_DBG(conn, "iscsi_tcp_state_change: "
144 				 "TCP_CLOSE|TCP_CLOSE_WAIT\n");
145 		iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
146 	}
147 
148 	tcp_conn = conn->dd_data;
149 	tcp_sw_conn = tcp_conn->dd_data;
150 	old_state_change = tcp_sw_conn->old_state_change;
151 
152 	read_unlock(&sk->sk_callback_lock);
153 
154 	old_state_change(sk);
155 }
156 
157 /**
158  * iscsi_write_space - Called when more output buffer space is available
159  * @sk: socket space is available for
160  **/
161 static void iscsi_sw_tcp_write_space(struct sock *sk)
162 {
163 	struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
164 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
165 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
166 
167 	tcp_sw_conn->old_write_space(sk);
168 	ISCSI_SW_TCP_DBG(conn, "iscsi_write_space\n");
169 	iscsi_conn_queue_work(conn);
170 }
171 
172 static void iscsi_sw_tcp_conn_set_callbacks(struct iscsi_conn *conn)
173 {
174 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
175 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
176 	struct sock *sk = tcp_sw_conn->sock->sk;
177 
178 	/* assign new callbacks */
179 	write_lock_bh(&sk->sk_callback_lock);
180 	sk->sk_user_data = conn;
181 	tcp_sw_conn->old_data_ready = sk->sk_data_ready;
182 	tcp_sw_conn->old_state_change = sk->sk_state_change;
183 	tcp_sw_conn->old_write_space = sk->sk_write_space;
184 	sk->sk_data_ready = iscsi_sw_tcp_data_ready;
185 	sk->sk_state_change = iscsi_sw_tcp_state_change;
186 	sk->sk_write_space = iscsi_sw_tcp_write_space;
187 	write_unlock_bh(&sk->sk_callback_lock);
188 }
189 
190 static void
191 iscsi_sw_tcp_conn_restore_callbacks(struct iscsi_sw_tcp_conn *tcp_sw_conn)
192 {
193 	struct sock *sk = tcp_sw_conn->sock->sk;
194 
195 	/* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
196 	write_lock_bh(&sk->sk_callback_lock);
197 	sk->sk_user_data    = NULL;
198 	sk->sk_data_ready   = tcp_sw_conn->old_data_ready;
199 	sk->sk_state_change = tcp_sw_conn->old_state_change;
200 	sk->sk_write_space  = tcp_sw_conn->old_write_space;
201 	sk->sk_no_check	 = 0;
202 	write_unlock_bh(&sk->sk_callback_lock);
203 }
204 
205 /**
206  * iscsi_sw_tcp_xmit_segment - transmit segment
207  * @tcp_conn: the iSCSI TCP connection
208  * @segment: the buffer to transmnit
209  *
210  * This function transmits as much of the buffer as
211  * the network layer will accept, and returns the number of
212  * bytes transmitted.
213  *
214  * If CRC hashing is enabled, the function will compute the
215  * hash as it goes. When the entire segment has been transmitted,
216  * it will retrieve the hash value and send it as well.
217  */
218 static int iscsi_sw_tcp_xmit_segment(struct iscsi_tcp_conn *tcp_conn,
219 				     struct iscsi_segment *segment)
220 {
221 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
222 	struct socket *sk = tcp_sw_conn->sock;
223 	unsigned int copied = 0;
224 	int r = 0;
225 
226 	while (!iscsi_tcp_segment_done(tcp_conn, segment, 0, r)) {
227 		struct scatterlist *sg;
228 		unsigned int offset, copy;
229 		int flags = 0;
230 
231 		r = 0;
232 		offset = segment->copied;
233 		copy = segment->size - offset;
234 
235 		if (segment->total_copied + segment->size < segment->total_size)
236 			flags |= MSG_MORE;
237 
238 		/* Use sendpage if we can; else fall back to sendmsg */
239 		if (!segment->data) {
240 			sg = segment->sg;
241 			offset += segment->sg_offset + sg->offset;
242 			r = tcp_sw_conn->sendpage(sk, sg_page(sg), offset,
243 						  copy, flags);
244 		} else {
245 			struct msghdr msg = { .msg_flags = flags };
246 			struct kvec iov = {
247 				.iov_base = segment->data + offset,
248 				.iov_len = copy
249 			};
250 
251 			r = kernel_sendmsg(sk, &msg, &iov, 1, copy);
252 		}
253 
254 		if (r < 0) {
255 			iscsi_tcp_segment_unmap(segment);
256 			if (copied || r == -EAGAIN)
257 				break;
258 			return r;
259 		}
260 		copied += r;
261 	}
262 	return copied;
263 }
264 
265 /**
266  * iscsi_sw_tcp_xmit - TCP transmit
267  **/
268 static int iscsi_sw_tcp_xmit(struct iscsi_conn *conn)
269 {
270 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
271 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
272 	struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
273 	unsigned int consumed = 0;
274 	int rc = 0;
275 
276 	while (1) {
277 		rc = iscsi_sw_tcp_xmit_segment(tcp_conn, segment);
278 		if (rc < 0) {
279 			rc = ISCSI_ERR_XMIT_FAILED;
280 			goto error;
281 		}
282 		if (rc == 0)
283 			break;
284 
285 		consumed += rc;
286 
287 		if (segment->total_copied >= segment->total_size) {
288 			if (segment->done != NULL) {
289 				rc = segment->done(tcp_conn, segment);
290 				if (rc != 0)
291 					goto error;
292 			}
293 		}
294 	}
295 
296 	ISCSI_SW_TCP_DBG(conn, "xmit %d bytes\n", consumed);
297 
298 	conn->txdata_octets += consumed;
299 	return consumed;
300 
301 error:
302 	/* Transmit error. We could initiate error recovery
303 	 * here. */
304 	ISCSI_SW_TCP_DBG(conn, "Error sending PDU, errno=%d\n", rc);
305 	iscsi_conn_failure(conn, rc);
306 	return -EIO;
307 }
308 
309 /**
310  * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
311  */
312 static inline int iscsi_sw_tcp_xmit_qlen(struct iscsi_conn *conn)
313 {
314 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
315 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
316 	struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
317 
318 	return segment->total_copied - segment->total_size;
319 }
320 
321 static int iscsi_sw_tcp_pdu_xmit(struct iscsi_task *task)
322 {
323 	struct iscsi_conn *conn = task->conn;
324 	int rc;
325 
326 	while (iscsi_sw_tcp_xmit_qlen(conn)) {
327 		rc = iscsi_sw_tcp_xmit(conn);
328 		if (rc == 0)
329 			return -EAGAIN;
330 		if (rc < 0)
331 			return rc;
332 	}
333 
334 	return 0;
335 }
336 
337 /*
338  * This is called when we're done sending the header.
339  * Simply copy the data_segment to the send segment, and return.
340  */
341 static int iscsi_sw_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
342 				      struct iscsi_segment *segment)
343 {
344 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
345 
346 	tcp_sw_conn->out.segment = tcp_sw_conn->out.data_segment;
347 	ISCSI_SW_TCP_DBG(tcp_conn->iscsi_conn,
348 			 "Header done. Next segment size %u total_size %u\n",
349 			 tcp_sw_conn->out.segment.size,
350 			 tcp_sw_conn->out.segment.total_size);
351 	return 0;
352 }
353 
354 static void iscsi_sw_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr,
355 				       size_t hdrlen)
356 {
357 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
358 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
359 
360 	ISCSI_SW_TCP_DBG(conn, "%s\n", conn->hdrdgst_en ?
361 			 "digest enabled" : "digest disabled");
362 
363 	/* Clear the data segment - needs to be filled in by the
364 	 * caller using iscsi_tcp_send_data_prep() */
365 	memset(&tcp_sw_conn->out.data_segment, 0,
366 	       sizeof(struct iscsi_segment));
367 
368 	/* If header digest is enabled, compute the CRC and
369 	 * place the digest into the same buffer. We make
370 	 * sure that both iscsi_tcp_task and mtask have
371 	 * sufficient room.
372 	 */
373 	if (conn->hdrdgst_en) {
374 		iscsi_tcp_dgst_header(&tcp_sw_conn->tx_hash, hdr, hdrlen,
375 				      hdr + hdrlen);
376 		hdrlen += ISCSI_DIGEST_SIZE;
377 	}
378 
379 	/* Remember header pointer for later, when we need
380 	 * to decide whether there's a payload to go along
381 	 * with the header. */
382 	tcp_sw_conn->out.hdr = hdr;
383 
384 	iscsi_segment_init_linear(&tcp_sw_conn->out.segment, hdr, hdrlen,
385 				  iscsi_sw_tcp_send_hdr_done, NULL);
386 }
387 
388 /*
389  * Prepare the send buffer for the payload data.
390  * Padding and checksumming will all be taken care
391  * of by the iscsi_segment routines.
392  */
393 static int
394 iscsi_sw_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
395 			    unsigned int count, unsigned int offset,
396 			    unsigned int len)
397 {
398 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
399 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
400 	struct hash_desc *tx_hash = NULL;
401 	unsigned int hdr_spec_len;
402 
403 	ISCSI_SW_TCP_DBG(conn, "offset=%d, datalen=%d %s\n", offset, len,
404 			 conn->datadgst_en ?
405 			 "digest enabled" : "digest disabled");
406 
407 	/* Make sure the datalen matches what the caller
408 	   said he would send. */
409 	hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
410 	WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
411 
412 	if (conn->datadgst_en)
413 		tx_hash = &tcp_sw_conn->tx_hash;
414 
415 	return iscsi_segment_seek_sg(&tcp_sw_conn->out.data_segment,
416 				     sg, count, offset, len,
417 				     NULL, tx_hash);
418 }
419 
420 static void
421 iscsi_sw_tcp_send_linear_data_prep(struct iscsi_conn *conn, void *data,
422 				   size_t len)
423 {
424 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
425 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
426 	struct hash_desc *tx_hash = NULL;
427 	unsigned int hdr_spec_len;
428 
429 	ISCSI_SW_TCP_DBG(conn, "datalen=%zd %s\n", len, conn->datadgst_en ?
430 			 "digest enabled" : "digest disabled");
431 
432 	/* Make sure the datalen matches what the caller
433 	   said he would send. */
434 	hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
435 	WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
436 
437 	if (conn->datadgst_en)
438 		tx_hash = &tcp_sw_conn->tx_hash;
439 
440 	iscsi_segment_init_linear(&tcp_sw_conn->out.data_segment,
441 				data, len, NULL, tx_hash);
442 }
443 
444 static int iscsi_sw_tcp_pdu_init(struct iscsi_task *task,
445 				 unsigned int offset, unsigned int count)
446 {
447 	struct iscsi_conn *conn = task->conn;
448 	int err = 0;
449 
450 	iscsi_sw_tcp_send_hdr_prep(conn, task->hdr, task->hdr_len);
451 
452 	if (!count)
453 		return 0;
454 
455 	if (!task->sc)
456 		iscsi_sw_tcp_send_linear_data_prep(conn, task->data, count);
457 	else {
458 		struct scsi_data_buffer *sdb = scsi_out(task->sc);
459 
460 		err = iscsi_sw_tcp_send_data_prep(conn, sdb->table.sgl,
461 						  sdb->table.nents, offset,
462 						  count);
463 	}
464 
465 	if (err) {
466 		/* got invalid offset/len */
467 		return -EIO;
468 	}
469 	return 0;
470 }
471 
472 static int iscsi_sw_tcp_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
473 {
474 	struct iscsi_tcp_task *tcp_task = task->dd_data;
475 
476 	task->hdr = task->dd_data + sizeof(*tcp_task);
477 	task->hdr_max = sizeof(struct iscsi_sw_tcp_hdrbuf) - ISCSI_DIGEST_SIZE;
478 	return 0;
479 }
480 
481 static struct iscsi_cls_conn *
482 iscsi_sw_tcp_conn_create(struct iscsi_cls_session *cls_session,
483 			 uint32_t conn_idx)
484 {
485 	struct iscsi_conn *conn;
486 	struct iscsi_cls_conn *cls_conn;
487 	struct iscsi_tcp_conn *tcp_conn;
488 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
489 
490 	cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*tcp_sw_conn),
491 					conn_idx);
492 	if (!cls_conn)
493 		return NULL;
494 	conn = cls_conn->dd_data;
495 	tcp_conn = conn->dd_data;
496 	tcp_sw_conn = tcp_conn->dd_data;
497 
498 	tcp_sw_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
499 						     CRYPTO_ALG_ASYNC);
500 	tcp_sw_conn->tx_hash.flags = 0;
501 	if (IS_ERR(tcp_sw_conn->tx_hash.tfm))
502 		goto free_conn;
503 
504 	tcp_sw_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
505 						     CRYPTO_ALG_ASYNC);
506 	tcp_sw_conn->rx_hash.flags = 0;
507 	if (IS_ERR(tcp_sw_conn->rx_hash.tfm))
508 		goto free_tx_tfm;
509 	tcp_conn->rx_hash = &tcp_sw_conn->rx_hash;
510 
511 	return cls_conn;
512 
513 free_tx_tfm:
514 	crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
515 free_conn:
516 	iscsi_conn_printk(KERN_ERR, conn,
517 			  "Could not create connection due to crc32c "
518 			  "loading error. Make sure the crc32c "
519 			  "module is built as a module or into the "
520 			  "kernel\n");
521 	iscsi_tcp_conn_teardown(cls_conn);
522 	return NULL;
523 }
524 
525 static void iscsi_sw_tcp_release_conn(struct iscsi_conn *conn)
526 {
527 	struct iscsi_session *session = conn->session;
528 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
529 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
530 	struct socket *sock = tcp_sw_conn->sock;
531 
532 	if (!sock)
533 		return;
534 
535 	sock_hold(sock->sk);
536 	iscsi_sw_tcp_conn_restore_callbacks(tcp_sw_conn);
537 	sock_put(sock->sk);
538 
539 	spin_lock_bh(&session->lock);
540 	tcp_sw_conn->sock = NULL;
541 	spin_unlock_bh(&session->lock);
542 	sockfd_put(sock);
543 }
544 
545 static void iscsi_sw_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
546 {
547 	struct iscsi_conn *conn = cls_conn->dd_data;
548 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
549 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
550 
551 	iscsi_sw_tcp_release_conn(conn);
552 
553 	if (tcp_sw_conn->tx_hash.tfm)
554 		crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
555 	if (tcp_sw_conn->rx_hash.tfm)
556 		crypto_free_hash(tcp_sw_conn->rx_hash.tfm);
557 
558 	iscsi_tcp_conn_teardown(cls_conn);
559 }
560 
561 static void iscsi_sw_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
562 {
563 	struct iscsi_conn *conn = cls_conn->dd_data;
564 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
565 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
566 
567 	/* userspace may have goofed up and not bound us */
568 	if (!tcp_sw_conn->sock)
569 		return;
570 	/*
571 	 * Make sure our recv side is stopped.
572 	 * Older tools called conn stop before ep_disconnect
573 	 * so IO could still be coming in.
574 	 */
575 	write_lock_bh(&tcp_sw_conn->sock->sk->sk_callback_lock);
576 	set_bit(ISCSI_SUSPEND_BIT, &conn->suspend_rx);
577 	write_unlock_bh(&tcp_sw_conn->sock->sk->sk_callback_lock);
578 
579 	iscsi_conn_stop(cls_conn, flag);
580 	iscsi_sw_tcp_release_conn(conn);
581 }
582 
583 static int iscsi_sw_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
584 				 char *buf, int *port,
585 				 int (*getname)(struct socket *,
586 						struct sockaddr *,
587 						int *addrlen))
588 {
589 	struct sockaddr_storage *addr;
590 	struct sockaddr_in6 *sin6;
591 	struct sockaddr_in *sin;
592 	int rc = 0, len;
593 
594 	addr = kmalloc(sizeof(*addr), GFP_KERNEL);
595 	if (!addr)
596 		return -ENOMEM;
597 
598 	if (getname(sock, (struct sockaddr *) addr, &len)) {
599 		rc = -ENODEV;
600 		goto free_addr;
601 	}
602 
603 	switch (addr->ss_family) {
604 	case AF_INET:
605 		sin = (struct sockaddr_in *)addr;
606 		spin_lock_bh(&conn->session->lock);
607 		sprintf(buf, "%pI4", &sin->sin_addr.s_addr);
608 		*port = be16_to_cpu(sin->sin_port);
609 		spin_unlock_bh(&conn->session->lock);
610 		break;
611 	case AF_INET6:
612 		sin6 = (struct sockaddr_in6 *)addr;
613 		spin_lock_bh(&conn->session->lock);
614 		sprintf(buf, "%pI6", &sin6->sin6_addr);
615 		*port = be16_to_cpu(sin6->sin6_port);
616 		spin_unlock_bh(&conn->session->lock);
617 		break;
618 	}
619 free_addr:
620 	kfree(addr);
621 	return rc;
622 }
623 
624 static int
625 iscsi_sw_tcp_conn_bind(struct iscsi_cls_session *cls_session,
626 		       struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
627 		       int is_leading)
628 {
629 	struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
630 	struct iscsi_host *ihost = shost_priv(shost);
631 	struct iscsi_conn *conn = cls_conn->dd_data;
632 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
633 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
634 	struct sock *sk;
635 	struct socket *sock;
636 	int err;
637 
638 	/* lookup for existing socket */
639 	sock = sockfd_lookup((int)transport_eph, &err);
640 	if (!sock) {
641 		iscsi_conn_printk(KERN_ERR, conn,
642 				  "sockfd_lookup failed %d\n", err);
643 		return -EEXIST;
644 	}
645 	/*
646 	 * copy these values now because if we drop the session
647 	 * userspace may still want to query the values since we will
648 	 * be using them for the reconnect
649 	 */
650 	err = iscsi_sw_tcp_get_addr(conn, sock, conn->portal_address,
651 				    &conn->portal_port, kernel_getpeername);
652 	if (err)
653 		goto free_socket;
654 
655 	err = iscsi_sw_tcp_get_addr(conn, sock, ihost->local_address,
656 				    &ihost->local_port, kernel_getsockname);
657 	if (err)
658 		goto free_socket;
659 
660 	err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
661 	if (err)
662 		goto free_socket;
663 
664 	/* bind iSCSI connection and socket */
665 	tcp_sw_conn->sock = sock;
666 
667 	/* setup Socket parameters */
668 	sk = sock->sk;
669 	sk->sk_reuse = 1;
670 	sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
671 	sk->sk_allocation = GFP_ATOMIC;
672 
673 	iscsi_sw_tcp_conn_set_callbacks(conn);
674 	tcp_sw_conn->sendpage = tcp_sw_conn->sock->ops->sendpage;
675 	/*
676 	 * set receive state machine into initial state
677 	 */
678 	iscsi_tcp_hdr_recv_prep(tcp_conn);
679 	return 0;
680 
681 free_socket:
682 	sockfd_put(sock);
683 	return err;
684 }
685 
686 static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn,
687 				       enum iscsi_param param, char *buf,
688 				       int buflen)
689 {
690 	struct iscsi_conn *conn = cls_conn->dd_data;
691 	struct iscsi_session *session = conn->session;
692 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
693 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
694 	int value;
695 
696 	switch(param) {
697 	case ISCSI_PARAM_HDRDGST_EN:
698 		iscsi_set_param(cls_conn, param, buf, buflen);
699 		break;
700 	case ISCSI_PARAM_DATADGST_EN:
701 		iscsi_set_param(cls_conn, param, buf, buflen);
702 		tcp_sw_conn->sendpage = conn->datadgst_en ?
703 			sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage;
704 		break;
705 	case ISCSI_PARAM_MAX_R2T:
706 		sscanf(buf, "%d", &value);
707 		if (value <= 0 || !is_power_of_2(value))
708 			return -EINVAL;
709 		if (session->max_r2t == value)
710 			break;
711 		iscsi_tcp_r2tpool_free(session);
712 		iscsi_set_param(cls_conn, param, buf, buflen);
713 		if (iscsi_tcp_r2tpool_alloc(session))
714 			return -ENOMEM;
715 		break;
716 	default:
717 		return iscsi_set_param(cls_conn, param, buf, buflen);
718 	}
719 
720 	return 0;
721 }
722 
723 static int iscsi_sw_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
724 				       enum iscsi_param param, char *buf)
725 {
726 	struct iscsi_conn *conn = cls_conn->dd_data;
727 	int len;
728 
729 	switch(param) {
730 	case ISCSI_PARAM_CONN_PORT:
731 		spin_lock_bh(&conn->session->lock);
732 		len = sprintf(buf, "%hu\n", conn->portal_port);
733 		spin_unlock_bh(&conn->session->lock);
734 		break;
735 	case ISCSI_PARAM_CONN_ADDRESS:
736 		spin_lock_bh(&conn->session->lock);
737 		len = sprintf(buf, "%s\n", conn->portal_address);
738 		spin_unlock_bh(&conn->session->lock);
739 		break;
740 	default:
741 		return iscsi_conn_get_param(cls_conn, param, buf);
742 	}
743 
744 	return len;
745 }
746 
747 static void
748 iscsi_sw_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
749 			    struct iscsi_stats *stats)
750 {
751 	struct iscsi_conn *conn = cls_conn->dd_data;
752 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
753 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
754 
755 	stats->custom_length = 3;
756 	strcpy(stats->custom[0].desc, "tx_sendpage_failures");
757 	stats->custom[0].value = tcp_sw_conn->sendpage_failures_cnt;
758 	strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
759 	stats->custom[1].value = tcp_sw_conn->discontiguous_hdr_cnt;
760 	strcpy(stats->custom[2].desc, "eh_abort_cnt");
761 	stats->custom[2].value = conn->eh_abort_cnt;
762 
763 	iscsi_tcp_conn_get_stats(cls_conn, stats);
764 }
765 
766 static struct iscsi_cls_session *
767 iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
768 			    uint16_t qdepth, uint32_t initial_cmdsn)
769 {
770 	struct iscsi_cls_session *cls_session;
771 	struct iscsi_session *session;
772 	struct Scsi_Host *shost;
773 
774 	if (ep) {
775 		printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
776 		return NULL;
777 	}
778 
779 	shost = iscsi_host_alloc(&iscsi_sw_tcp_sht, 0, 1);
780 	if (!shost)
781 		return NULL;
782 	shost->transportt = iscsi_sw_tcp_scsi_transport;
783 	shost->cmd_per_lun = qdepth;
784 	shost->max_lun = iscsi_max_lun;
785 	shost->max_id = 0;
786 	shost->max_channel = 0;
787 	shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE;
788 
789 	if (iscsi_host_add(shost, NULL))
790 		goto free_host;
791 
792 	cls_session = iscsi_session_setup(&iscsi_sw_tcp_transport, shost,
793 					  cmds_max,
794 					  sizeof(struct iscsi_tcp_task) +
795 					  sizeof(struct iscsi_sw_tcp_hdrbuf),
796 					  initial_cmdsn, 0);
797 	if (!cls_session)
798 		goto remove_host;
799 	session = cls_session->dd_data;
800 
801 	shost->can_queue = session->scsi_cmds_max;
802 	if (iscsi_tcp_r2tpool_alloc(session))
803 		goto remove_session;
804 	return cls_session;
805 
806 remove_session:
807 	iscsi_session_teardown(cls_session);
808 remove_host:
809 	iscsi_host_remove(shost);
810 free_host:
811 	iscsi_host_free(shost);
812 	return NULL;
813 }
814 
815 static void iscsi_sw_tcp_session_destroy(struct iscsi_cls_session *cls_session)
816 {
817 	struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
818 
819 	iscsi_tcp_r2tpool_free(cls_session->dd_data);
820 	iscsi_session_teardown(cls_session);
821 
822 	iscsi_host_remove(shost);
823 	iscsi_host_free(shost);
824 }
825 
826 static int iscsi_sw_tcp_slave_alloc(struct scsi_device *sdev)
827 {
828 	set_bit(QUEUE_FLAG_BIDI, &sdev->request_queue->queue_flags);
829 	return 0;
830 }
831 
832 static int iscsi_sw_tcp_slave_configure(struct scsi_device *sdev)
833 {
834 	blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
835 	blk_queue_dma_alignment(sdev->request_queue, 0);
836 	return 0;
837 }
838 
839 static struct scsi_host_template iscsi_sw_tcp_sht = {
840 	.module			= THIS_MODULE,
841 	.name			= "iSCSI Initiator over TCP/IP",
842 	.queuecommand           = iscsi_queuecommand,
843 	.change_queue_depth	= iscsi_change_queue_depth,
844 	.can_queue		= ISCSI_DEF_XMIT_CMDS_MAX - 1,
845 	.sg_tablesize		= 4096,
846 	.max_sectors		= 0xFFFF,
847 	.cmd_per_lun		= ISCSI_DEF_CMD_PER_LUN,
848 	.eh_abort_handler       = iscsi_eh_abort,
849 	.eh_device_reset_handler= iscsi_eh_device_reset,
850 	.eh_target_reset_handler= iscsi_eh_target_reset,
851 	.use_clustering         = DISABLE_CLUSTERING,
852 	.slave_alloc            = iscsi_sw_tcp_slave_alloc,
853 	.slave_configure        = iscsi_sw_tcp_slave_configure,
854 	.target_alloc		= iscsi_target_alloc,
855 	.proc_name		= "iscsi_tcp",
856 	.this_id		= -1,
857 };
858 
859 static struct iscsi_transport iscsi_sw_tcp_transport = {
860 	.owner			= THIS_MODULE,
861 	.name			= "tcp",
862 	.caps			= CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
863 				  | CAP_DATADGST,
864 	.param_mask		= ISCSI_MAX_RECV_DLENGTH |
865 				  ISCSI_MAX_XMIT_DLENGTH |
866 				  ISCSI_HDRDGST_EN |
867 				  ISCSI_DATADGST_EN |
868 				  ISCSI_INITIAL_R2T_EN |
869 				  ISCSI_MAX_R2T |
870 				  ISCSI_IMM_DATA_EN |
871 				  ISCSI_FIRST_BURST |
872 				  ISCSI_MAX_BURST |
873 				  ISCSI_PDU_INORDER_EN |
874 				  ISCSI_DATASEQ_INORDER_EN |
875 				  ISCSI_ERL |
876 				  ISCSI_CONN_PORT |
877 				  ISCSI_CONN_ADDRESS |
878 				  ISCSI_EXP_STATSN |
879 				  ISCSI_PERSISTENT_PORT |
880 				  ISCSI_PERSISTENT_ADDRESS |
881 				  ISCSI_TARGET_NAME | ISCSI_TPGT |
882 				  ISCSI_USERNAME | ISCSI_PASSWORD |
883 				  ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
884 				  ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
885 				  ISCSI_LU_RESET_TMO |
886 				  ISCSI_PING_TMO | ISCSI_RECV_TMO |
887 				  ISCSI_IFACE_NAME | ISCSI_INITIATOR_NAME,
888 	.host_param_mask	= ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
889 				  ISCSI_HOST_INITIATOR_NAME |
890 				  ISCSI_HOST_NETDEV_NAME,
891 	/* session management */
892 	.create_session		= iscsi_sw_tcp_session_create,
893 	.destroy_session	= iscsi_sw_tcp_session_destroy,
894 	/* connection management */
895 	.create_conn		= iscsi_sw_tcp_conn_create,
896 	.bind_conn		= iscsi_sw_tcp_conn_bind,
897 	.destroy_conn		= iscsi_sw_tcp_conn_destroy,
898 	.set_param		= iscsi_sw_tcp_conn_set_param,
899 	.get_conn_param		= iscsi_sw_tcp_conn_get_param,
900 	.get_session_param	= iscsi_session_get_param,
901 	.start_conn		= iscsi_conn_start,
902 	.stop_conn		= iscsi_sw_tcp_conn_stop,
903 	/* iscsi host params */
904 	.get_host_param		= iscsi_host_get_param,
905 	.set_host_param		= iscsi_host_set_param,
906 	/* IO */
907 	.send_pdu		= iscsi_conn_send_pdu,
908 	.get_stats		= iscsi_sw_tcp_conn_get_stats,
909 	/* iscsi task/cmd helpers */
910 	.init_task		= iscsi_tcp_task_init,
911 	.xmit_task		= iscsi_tcp_task_xmit,
912 	.cleanup_task		= iscsi_tcp_cleanup_task,
913 	/* low level pdu helpers */
914 	.xmit_pdu		= iscsi_sw_tcp_pdu_xmit,
915 	.init_pdu		= iscsi_sw_tcp_pdu_init,
916 	.alloc_pdu		= iscsi_sw_tcp_pdu_alloc,
917 	/* recovery */
918 	.session_recovery_timedout = iscsi_session_recovery_timedout,
919 };
920 
921 static int __init iscsi_sw_tcp_init(void)
922 {
923 	if (iscsi_max_lun < 1) {
924 		printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
925 		       iscsi_max_lun);
926 		return -EINVAL;
927 	}
928 
929 	iscsi_sw_tcp_scsi_transport = iscsi_register_transport(
930 						&iscsi_sw_tcp_transport);
931 	if (!iscsi_sw_tcp_scsi_transport)
932 		return -ENODEV;
933 
934 	return 0;
935 }
936 
937 static void __exit iscsi_sw_tcp_exit(void)
938 {
939 	iscsi_unregister_transport(&iscsi_sw_tcp_transport);
940 }
941 
942 module_init(iscsi_sw_tcp_init);
943 module_exit(iscsi_sw_tcp_exit);
944