xref: /linux/drivers/scsi/iscsi_tcp.c (revision 2277ab4a1df50e05bc732fe9488d4e902bb8399a)
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 			return r;
257 		}
258 		copied += r;
259 	}
260 	return copied;
261 }
262 
263 /**
264  * iscsi_sw_tcp_xmit - TCP transmit
265  **/
266 static int iscsi_sw_tcp_xmit(struct iscsi_conn *conn)
267 {
268 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
269 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
270 	struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
271 	unsigned int consumed = 0;
272 	int rc = 0;
273 
274 	while (1) {
275 		rc = iscsi_sw_tcp_xmit_segment(tcp_conn, segment);
276 		/*
277 		 * We may not have been able to send data because the conn
278 		 * is getting stopped. libiscsi will know so propogate err
279 		 * for it to do the right thing.
280 		 */
281 		if (rc == -EAGAIN)
282 			return rc;
283 		else if (rc < 0) {
284 			rc = ISCSI_ERR_XMIT_FAILED;
285 			goto error;
286 		} else if (rc == 0)
287 			break;
288 
289 		consumed += rc;
290 
291 		if (segment->total_copied >= segment->total_size) {
292 			if (segment->done != NULL) {
293 				rc = segment->done(tcp_conn, segment);
294 				if (rc != 0)
295 					goto error;
296 			}
297 		}
298 	}
299 
300 	ISCSI_SW_TCP_DBG(conn, "xmit %d bytes\n", consumed);
301 
302 	conn->txdata_octets += consumed;
303 	return consumed;
304 
305 error:
306 	/* Transmit error. We could initiate error recovery
307 	 * here. */
308 	ISCSI_SW_TCP_DBG(conn, "Error sending PDU, errno=%d\n", rc);
309 	iscsi_conn_failure(conn, rc);
310 	return -EIO;
311 }
312 
313 /**
314  * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
315  */
316 static inline int iscsi_sw_tcp_xmit_qlen(struct iscsi_conn *conn)
317 {
318 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
319 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
320 	struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
321 
322 	return segment->total_copied - segment->total_size;
323 }
324 
325 static int iscsi_sw_tcp_pdu_xmit(struct iscsi_task *task)
326 {
327 	struct iscsi_conn *conn = task->conn;
328 	int rc;
329 
330 	while (iscsi_sw_tcp_xmit_qlen(conn)) {
331 		rc = iscsi_sw_tcp_xmit(conn);
332 		if (rc == 0)
333 			return -EAGAIN;
334 		if (rc < 0)
335 			return rc;
336 	}
337 
338 	return 0;
339 }
340 
341 /*
342  * This is called when we're done sending the header.
343  * Simply copy the data_segment to the send segment, and return.
344  */
345 static int iscsi_sw_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
346 				      struct iscsi_segment *segment)
347 {
348 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
349 
350 	tcp_sw_conn->out.segment = tcp_sw_conn->out.data_segment;
351 	ISCSI_SW_TCP_DBG(tcp_conn->iscsi_conn,
352 			 "Header done. Next segment size %u total_size %u\n",
353 			 tcp_sw_conn->out.segment.size,
354 			 tcp_sw_conn->out.segment.total_size);
355 	return 0;
356 }
357 
358 static void iscsi_sw_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr,
359 				       size_t hdrlen)
360 {
361 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
362 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
363 
364 	ISCSI_SW_TCP_DBG(conn, "%s\n", conn->hdrdgst_en ?
365 			 "digest enabled" : "digest disabled");
366 
367 	/* Clear the data segment - needs to be filled in by the
368 	 * caller using iscsi_tcp_send_data_prep() */
369 	memset(&tcp_sw_conn->out.data_segment, 0,
370 	       sizeof(struct iscsi_segment));
371 
372 	/* If header digest is enabled, compute the CRC and
373 	 * place the digest into the same buffer. We make
374 	 * sure that both iscsi_tcp_task and mtask have
375 	 * sufficient room.
376 	 */
377 	if (conn->hdrdgst_en) {
378 		iscsi_tcp_dgst_header(&tcp_sw_conn->tx_hash, hdr, hdrlen,
379 				      hdr + hdrlen);
380 		hdrlen += ISCSI_DIGEST_SIZE;
381 	}
382 
383 	/* Remember header pointer for later, when we need
384 	 * to decide whether there's a payload to go along
385 	 * with the header. */
386 	tcp_sw_conn->out.hdr = hdr;
387 
388 	iscsi_segment_init_linear(&tcp_sw_conn->out.segment, hdr, hdrlen,
389 				  iscsi_sw_tcp_send_hdr_done, NULL);
390 }
391 
392 /*
393  * Prepare the send buffer for the payload data.
394  * Padding and checksumming will all be taken care
395  * of by the iscsi_segment routines.
396  */
397 static int
398 iscsi_sw_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
399 			    unsigned int count, unsigned int offset,
400 			    unsigned int len)
401 {
402 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
403 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
404 	struct hash_desc *tx_hash = NULL;
405 	unsigned int hdr_spec_len;
406 
407 	ISCSI_SW_TCP_DBG(conn, "offset=%d, datalen=%d %s\n", offset, len,
408 			 conn->datadgst_en ?
409 			 "digest enabled" : "digest disabled");
410 
411 	/* Make sure the datalen matches what the caller
412 	   said he would send. */
413 	hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
414 	WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
415 
416 	if (conn->datadgst_en)
417 		tx_hash = &tcp_sw_conn->tx_hash;
418 
419 	return iscsi_segment_seek_sg(&tcp_sw_conn->out.data_segment,
420 				     sg, count, offset, len,
421 				     NULL, tx_hash);
422 }
423 
424 static void
425 iscsi_sw_tcp_send_linear_data_prep(struct iscsi_conn *conn, void *data,
426 				   size_t len)
427 {
428 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
429 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
430 	struct hash_desc *tx_hash = NULL;
431 	unsigned int hdr_spec_len;
432 
433 	ISCSI_SW_TCP_DBG(conn, "datalen=%zd %s\n", len, conn->datadgst_en ?
434 			 "digest enabled" : "digest disabled");
435 
436 	/* Make sure the datalen matches what the caller
437 	   said he would send. */
438 	hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
439 	WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
440 
441 	if (conn->datadgst_en)
442 		tx_hash = &tcp_sw_conn->tx_hash;
443 
444 	iscsi_segment_init_linear(&tcp_sw_conn->out.data_segment,
445 				data, len, NULL, tx_hash);
446 }
447 
448 static int iscsi_sw_tcp_pdu_init(struct iscsi_task *task,
449 				 unsigned int offset, unsigned int count)
450 {
451 	struct iscsi_conn *conn = task->conn;
452 	int err = 0;
453 
454 	iscsi_sw_tcp_send_hdr_prep(conn, task->hdr, task->hdr_len);
455 
456 	if (!count)
457 		return 0;
458 
459 	if (!task->sc)
460 		iscsi_sw_tcp_send_linear_data_prep(conn, task->data, count);
461 	else {
462 		struct scsi_data_buffer *sdb = scsi_out(task->sc);
463 
464 		err = iscsi_sw_tcp_send_data_prep(conn, sdb->table.sgl,
465 						  sdb->table.nents, offset,
466 						  count);
467 	}
468 
469 	if (err) {
470 		/* got invalid offset/len */
471 		return -EIO;
472 	}
473 	return 0;
474 }
475 
476 static int iscsi_sw_tcp_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
477 {
478 	struct iscsi_tcp_task *tcp_task = task->dd_data;
479 
480 	task->hdr = task->dd_data + sizeof(*tcp_task);
481 	task->hdr_max = sizeof(struct iscsi_sw_tcp_hdrbuf) - ISCSI_DIGEST_SIZE;
482 	return 0;
483 }
484 
485 static struct iscsi_cls_conn *
486 iscsi_sw_tcp_conn_create(struct iscsi_cls_session *cls_session,
487 			 uint32_t conn_idx)
488 {
489 	struct iscsi_conn *conn;
490 	struct iscsi_cls_conn *cls_conn;
491 	struct iscsi_tcp_conn *tcp_conn;
492 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
493 
494 	cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*tcp_sw_conn),
495 					conn_idx);
496 	if (!cls_conn)
497 		return NULL;
498 	conn = cls_conn->dd_data;
499 	tcp_conn = conn->dd_data;
500 	tcp_sw_conn = tcp_conn->dd_data;
501 
502 	tcp_sw_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
503 						     CRYPTO_ALG_ASYNC);
504 	tcp_sw_conn->tx_hash.flags = 0;
505 	if (IS_ERR(tcp_sw_conn->tx_hash.tfm))
506 		goto free_conn;
507 
508 	tcp_sw_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
509 						     CRYPTO_ALG_ASYNC);
510 	tcp_sw_conn->rx_hash.flags = 0;
511 	if (IS_ERR(tcp_sw_conn->rx_hash.tfm))
512 		goto free_tx_tfm;
513 	tcp_conn->rx_hash = &tcp_sw_conn->rx_hash;
514 
515 	return cls_conn;
516 
517 free_tx_tfm:
518 	crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
519 free_conn:
520 	iscsi_conn_printk(KERN_ERR, conn,
521 			  "Could not create connection due to crc32c "
522 			  "loading error. Make sure the crc32c "
523 			  "module is built as a module or into the "
524 			  "kernel\n");
525 	iscsi_tcp_conn_teardown(cls_conn);
526 	return NULL;
527 }
528 
529 static void iscsi_sw_tcp_release_conn(struct iscsi_conn *conn)
530 {
531 	struct iscsi_session *session = conn->session;
532 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
533 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
534 	struct socket *sock = tcp_sw_conn->sock;
535 
536 	if (!sock)
537 		return;
538 
539 	sock_hold(sock->sk);
540 	iscsi_sw_tcp_conn_restore_callbacks(tcp_sw_conn);
541 	sock_put(sock->sk);
542 
543 	spin_lock_bh(&session->lock);
544 	tcp_sw_conn->sock = NULL;
545 	spin_unlock_bh(&session->lock);
546 	sockfd_put(sock);
547 }
548 
549 static void iscsi_sw_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
550 {
551 	struct iscsi_conn *conn = cls_conn->dd_data;
552 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
553 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
554 
555 	iscsi_sw_tcp_release_conn(conn);
556 
557 	if (tcp_sw_conn->tx_hash.tfm)
558 		crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
559 	if (tcp_sw_conn->rx_hash.tfm)
560 		crypto_free_hash(tcp_sw_conn->rx_hash.tfm);
561 
562 	iscsi_tcp_conn_teardown(cls_conn);
563 }
564 
565 static void iscsi_sw_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
566 {
567 	struct iscsi_conn *conn = cls_conn->dd_data;
568 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
569 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
570 
571 	/* userspace may have goofed up and not bound us */
572 	if (!tcp_sw_conn->sock)
573 		return;
574 	/*
575 	 * Make sure our recv side is stopped.
576 	 * Older tools called conn stop before ep_disconnect
577 	 * so IO could still be coming in.
578 	 */
579 	write_lock_bh(&tcp_sw_conn->sock->sk->sk_callback_lock);
580 	set_bit(ISCSI_SUSPEND_BIT, &conn->suspend_rx);
581 	write_unlock_bh(&tcp_sw_conn->sock->sk->sk_callback_lock);
582 
583 	iscsi_conn_stop(cls_conn, flag);
584 	iscsi_sw_tcp_release_conn(conn);
585 }
586 
587 static int iscsi_sw_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
588 				 char *buf, int *port,
589 				 int (*getname)(struct socket *,
590 						struct sockaddr *,
591 						int *addrlen))
592 {
593 	struct sockaddr_storage *addr;
594 	struct sockaddr_in6 *sin6;
595 	struct sockaddr_in *sin;
596 	int rc = 0, len;
597 
598 	addr = kmalloc(sizeof(*addr), GFP_KERNEL);
599 	if (!addr)
600 		return -ENOMEM;
601 
602 	if (getname(sock, (struct sockaddr *) addr, &len)) {
603 		rc = -ENODEV;
604 		goto free_addr;
605 	}
606 
607 	switch (addr->ss_family) {
608 	case AF_INET:
609 		sin = (struct sockaddr_in *)addr;
610 		spin_lock_bh(&conn->session->lock);
611 		sprintf(buf, "%pI4", &sin->sin_addr.s_addr);
612 		*port = be16_to_cpu(sin->sin_port);
613 		spin_unlock_bh(&conn->session->lock);
614 		break;
615 	case AF_INET6:
616 		sin6 = (struct sockaddr_in6 *)addr;
617 		spin_lock_bh(&conn->session->lock);
618 		sprintf(buf, "%pI6", &sin6->sin6_addr);
619 		*port = be16_to_cpu(sin6->sin6_port);
620 		spin_unlock_bh(&conn->session->lock);
621 		break;
622 	}
623 free_addr:
624 	kfree(addr);
625 	return rc;
626 }
627 
628 static int
629 iscsi_sw_tcp_conn_bind(struct iscsi_cls_session *cls_session,
630 		       struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
631 		       int is_leading)
632 {
633 	struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
634 	struct iscsi_host *ihost = shost_priv(shost);
635 	struct iscsi_conn *conn = cls_conn->dd_data;
636 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
637 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
638 	struct sock *sk;
639 	struct socket *sock;
640 	int err;
641 
642 	/* lookup for existing socket */
643 	sock = sockfd_lookup((int)transport_eph, &err);
644 	if (!sock) {
645 		iscsi_conn_printk(KERN_ERR, conn,
646 				  "sockfd_lookup failed %d\n", err);
647 		return -EEXIST;
648 	}
649 	/*
650 	 * copy these values now because if we drop the session
651 	 * userspace may still want to query the values since we will
652 	 * be using them for the reconnect
653 	 */
654 	err = iscsi_sw_tcp_get_addr(conn, sock, conn->portal_address,
655 				    &conn->portal_port, kernel_getpeername);
656 	if (err)
657 		goto free_socket;
658 
659 	err = iscsi_sw_tcp_get_addr(conn, sock, ihost->local_address,
660 				    &ihost->local_port, kernel_getsockname);
661 	if (err)
662 		goto free_socket;
663 
664 	err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
665 	if (err)
666 		goto free_socket;
667 
668 	/* bind iSCSI connection and socket */
669 	tcp_sw_conn->sock = sock;
670 
671 	/* setup Socket parameters */
672 	sk = sock->sk;
673 	sk->sk_reuse = 1;
674 	sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
675 	sk->sk_allocation = GFP_ATOMIC;
676 
677 	iscsi_sw_tcp_conn_set_callbacks(conn);
678 	tcp_sw_conn->sendpage = tcp_sw_conn->sock->ops->sendpage;
679 	/*
680 	 * set receive state machine into initial state
681 	 */
682 	iscsi_tcp_hdr_recv_prep(tcp_conn);
683 	return 0;
684 
685 free_socket:
686 	sockfd_put(sock);
687 	return err;
688 }
689 
690 static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn,
691 				       enum iscsi_param param, char *buf,
692 				       int buflen)
693 {
694 	struct iscsi_conn *conn = cls_conn->dd_data;
695 	struct iscsi_session *session = conn->session;
696 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
697 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
698 	int value;
699 
700 	switch(param) {
701 	case ISCSI_PARAM_HDRDGST_EN:
702 		iscsi_set_param(cls_conn, param, buf, buflen);
703 		break;
704 	case ISCSI_PARAM_DATADGST_EN:
705 		iscsi_set_param(cls_conn, param, buf, buflen);
706 		tcp_sw_conn->sendpage = conn->datadgst_en ?
707 			sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage;
708 		break;
709 	case ISCSI_PARAM_MAX_R2T:
710 		sscanf(buf, "%d", &value);
711 		if (value <= 0 || !is_power_of_2(value))
712 			return -EINVAL;
713 		if (session->max_r2t == value)
714 			break;
715 		iscsi_tcp_r2tpool_free(session);
716 		iscsi_set_param(cls_conn, param, buf, buflen);
717 		if (iscsi_tcp_r2tpool_alloc(session))
718 			return -ENOMEM;
719 		break;
720 	default:
721 		return iscsi_set_param(cls_conn, param, buf, buflen);
722 	}
723 
724 	return 0;
725 }
726 
727 static int iscsi_sw_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
728 				       enum iscsi_param param, char *buf)
729 {
730 	struct iscsi_conn *conn = cls_conn->dd_data;
731 	int len;
732 
733 	switch(param) {
734 	case ISCSI_PARAM_CONN_PORT:
735 		spin_lock_bh(&conn->session->lock);
736 		len = sprintf(buf, "%hu\n", conn->portal_port);
737 		spin_unlock_bh(&conn->session->lock);
738 		break;
739 	case ISCSI_PARAM_CONN_ADDRESS:
740 		spin_lock_bh(&conn->session->lock);
741 		len = sprintf(buf, "%s\n", conn->portal_address);
742 		spin_unlock_bh(&conn->session->lock);
743 		break;
744 	default:
745 		return iscsi_conn_get_param(cls_conn, param, buf);
746 	}
747 
748 	return len;
749 }
750 
751 static void
752 iscsi_sw_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
753 			    struct iscsi_stats *stats)
754 {
755 	struct iscsi_conn *conn = cls_conn->dd_data;
756 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
757 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
758 
759 	stats->custom_length = 3;
760 	strcpy(stats->custom[0].desc, "tx_sendpage_failures");
761 	stats->custom[0].value = tcp_sw_conn->sendpage_failures_cnt;
762 	strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
763 	stats->custom[1].value = tcp_sw_conn->discontiguous_hdr_cnt;
764 	strcpy(stats->custom[2].desc, "eh_abort_cnt");
765 	stats->custom[2].value = conn->eh_abort_cnt;
766 
767 	iscsi_tcp_conn_get_stats(cls_conn, stats);
768 }
769 
770 static struct iscsi_cls_session *
771 iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
772 			    uint16_t qdepth, uint32_t initial_cmdsn)
773 {
774 	struct iscsi_cls_session *cls_session;
775 	struct iscsi_session *session;
776 	struct Scsi_Host *shost;
777 
778 	if (ep) {
779 		printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
780 		return NULL;
781 	}
782 
783 	shost = iscsi_host_alloc(&iscsi_sw_tcp_sht, 0, 1);
784 	if (!shost)
785 		return NULL;
786 	shost->transportt = iscsi_sw_tcp_scsi_transport;
787 	shost->cmd_per_lun = qdepth;
788 	shost->max_lun = iscsi_max_lun;
789 	shost->max_id = 0;
790 	shost->max_channel = 0;
791 	shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE;
792 
793 	if (iscsi_host_add(shost, NULL))
794 		goto free_host;
795 
796 	cls_session = iscsi_session_setup(&iscsi_sw_tcp_transport, shost,
797 					  cmds_max,
798 					  sizeof(struct iscsi_tcp_task) +
799 					  sizeof(struct iscsi_sw_tcp_hdrbuf),
800 					  initial_cmdsn, 0);
801 	if (!cls_session)
802 		goto remove_host;
803 	session = cls_session->dd_data;
804 
805 	shost->can_queue = session->scsi_cmds_max;
806 	if (iscsi_tcp_r2tpool_alloc(session))
807 		goto remove_session;
808 	return cls_session;
809 
810 remove_session:
811 	iscsi_session_teardown(cls_session);
812 remove_host:
813 	iscsi_host_remove(shost);
814 free_host:
815 	iscsi_host_free(shost);
816 	return NULL;
817 }
818 
819 static void iscsi_sw_tcp_session_destroy(struct iscsi_cls_session *cls_session)
820 {
821 	struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
822 
823 	iscsi_tcp_r2tpool_free(cls_session->dd_data);
824 	iscsi_session_teardown(cls_session);
825 
826 	iscsi_host_remove(shost);
827 	iscsi_host_free(shost);
828 }
829 
830 static int iscsi_sw_tcp_slave_alloc(struct scsi_device *sdev)
831 {
832 	set_bit(QUEUE_FLAG_BIDI, &sdev->request_queue->queue_flags);
833 	return 0;
834 }
835 
836 static int iscsi_sw_tcp_slave_configure(struct scsi_device *sdev)
837 {
838 	blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
839 	blk_queue_dma_alignment(sdev->request_queue, 0);
840 	return 0;
841 }
842 
843 static struct scsi_host_template iscsi_sw_tcp_sht = {
844 	.module			= THIS_MODULE,
845 	.name			= "iSCSI Initiator over TCP/IP",
846 	.queuecommand           = iscsi_queuecommand,
847 	.change_queue_depth	= iscsi_change_queue_depth,
848 	.can_queue		= ISCSI_DEF_XMIT_CMDS_MAX - 1,
849 	.sg_tablesize		= 4096,
850 	.max_sectors		= 0xFFFF,
851 	.cmd_per_lun		= ISCSI_DEF_CMD_PER_LUN,
852 	.eh_abort_handler       = iscsi_eh_abort,
853 	.eh_device_reset_handler= iscsi_eh_device_reset,
854 	.eh_target_reset_handler= iscsi_eh_target_reset,
855 	.use_clustering         = DISABLE_CLUSTERING,
856 	.slave_alloc            = iscsi_sw_tcp_slave_alloc,
857 	.slave_configure        = iscsi_sw_tcp_slave_configure,
858 	.target_alloc		= iscsi_target_alloc,
859 	.proc_name		= "iscsi_tcp",
860 	.this_id		= -1,
861 };
862 
863 static struct iscsi_transport iscsi_sw_tcp_transport = {
864 	.owner			= THIS_MODULE,
865 	.name			= "tcp",
866 	.caps			= CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
867 				  | CAP_DATADGST,
868 	.param_mask		= ISCSI_MAX_RECV_DLENGTH |
869 				  ISCSI_MAX_XMIT_DLENGTH |
870 				  ISCSI_HDRDGST_EN |
871 				  ISCSI_DATADGST_EN |
872 				  ISCSI_INITIAL_R2T_EN |
873 				  ISCSI_MAX_R2T |
874 				  ISCSI_IMM_DATA_EN |
875 				  ISCSI_FIRST_BURST |
876 				  ISCSI_MAX_BURST |
877 				  ISCSI_PDU_INORDER_EN |
878 				  ISCSI_DATASEQ_INORDER_EN |
879 				  ISCSI_ERL |
880 				  ISCSI_CONN_PORT |
881 				  ISCSI_CONN_ADDRESS |
882 				  ISCSI_EXP_STATSN |
883 				  ISCSI_PERSISTENT_PORT |
884 				  ISCSI_PERSISTENT_ADDRESS |
885 				  ISCSI_TARGET_NAME | ISCSI_TPGT |
886 				  ISCSI_USERNAME | ISCSI_PASSWORD |
887 				  ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
888 				  ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
889 				  ISCSI_LU_RESET_TMO |
890 				  ISCSI_PING_TMO | ISCSI_RECV_TMO |
891 				  ISCSI_IFACE_NAME | ISCSI_INITIATOR_NAME,
892 	.host_param_mask	= ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
893 				  ISCSI_HOST_INITIATOR_NAME |
894 				  ISCSI_HOST_NETDEV_NAME,
895 	/* session management */
896 	.create_session		= iscsi_sw_tcp_session_create,
897 	.destroy_session	= iscsi_sw_tcp_session_destroy,
898 	/* connection management */
899 	.create_conn		= iscsi_sw_tcp_conn_create,
900 	.bind_conn		= iscsi_sw_tcp_conn_bind,
901 	.destroy_conn		= iscsi_sw_tcp_conn_destroy,
902 	.set_param		= iscsi_sw_tcp_conn_set_param,
903 	.get_conn_param		= iscsi_sw_tcp_conn_get_param,
904 	.get_session_param	= iscsi_session_get_param,
905 	.start_conn		= iscsi_conn_start,
906 	.stop_conn		= iscsi_sw_tcp_conn_stop,
907 	/* iscsi host params */
908 	.get_host_param		= iscsi_host_get_param,
909 	.set_host_param		= iscsi_host_set_param,
910 	/* IO */
911 	.send_pdu		= iscsi_conn_send_pdu,
912 	.get_stats		= iscsi_sw_tcp_conn_get_stats,
913 	/* iscsi task/cmd helpers */
914 	.init_task		= iscsi_tcp_task_init,
915 	.xmit_task		= iscsi_tcp_task_xmit,
916 	.cleanup_task		= iscsi_tcp_cleanup_task,
917 	/* low level pdu helpers */
918 	.xmit_pdu		= iscsi_sw_tcp_pdu_xmit,
919 	.init_pdu		= iscsi_sw_tcp_pdu_init,
920 	.alloc_pdu		= iscsi_sw_tcp_pdu_alloc,
921 	/* recovery */
922 	.session_recovery_timedout = iscsi_session_recovery_timedout,
923 };
924 
925 static int __init iscsi_sw_tcp_init(void)
926 {
927 	if (iscsi_max_lun < 1) {
928 		printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
929 		       iscsi_max_lun);
930 		return -EINVAL;
931 	}
932 
933 	iscsi_sw_tcp_scsi_transport = iscsi_register_transport(
934 						&iscsi_sw_tcp_transport);
935 	if (!iscsi_sw_tcp_scsi_transport)
936 		return -ENODEV;
937 
938 	return 0;
939 }
940 
941 static void __exit iscsi_sw_tcp_exit(void)
942 {
943 	iscsi_unregister_transport(&iscsi_sw_tcp_transport);
944 }
945 
946 module_init(iscsi_sw_tcp_init);
947 module_exit(iscsi_sw_tcp_exit);
948