xref: /openbmc/linux/drivers/scsi/iscsi_tcp.c (revision 1f9f6a78)
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/slab.h>
32 #include <linux/file.h>
33 #include <linux/blkdev.h>
34 #include <linux/crypto.h>
35 #include <linux/delay.h>
36 #include <linux/kfifo.h>
37 #include <linux/scatterlist.h>
38 #include <linux/module.h>
39 #include <net/tcp.h>
40 #include <scsi/scsi_cmnd.h>
41 #include <scsi/scsi_device.h>
42 #include <scsi/scsi_host.h>
43 #include <scsi/scsi.h>
44 #include <scsi/scsi_transport_iscsi.h>
45 
46 #include "iscsi_tcp.h"
47 
48 MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
49 	      "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
50 	      "Alex Aizman <itn780@yahoo.com>");
51 MODULE_DESCRIPTION("iSCSI/TCP data-path");
52 MODULE_LICENSE("GPL");
53 
54 static struct scsi_transport_template *iscsi_sw_tcp_scsi_transport;
55 static struct scsi_host_template iscsi_sw_tcp_sht;
56 static struct iscsi_transport iscsi_sw_tcp_transport;
57 
58 static unsigned int iscsi_max_lun = ~0;
59 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
60 
61 static int iscsi_sw_tcp_dbg;
62 module_param_named(debug_iscsi_tcp, iscsi_sw_tcp_dbg, int,
63 		   S_IRUGO | S_IWUSR);
64 MODULE_PARM_DESC(debug_iscsi_tcp, "Turn on debugging for iscsi_tcp module "
65 		 "Set to 1 to turn on, and zero to turn off. Default is off.");
66 
67 #define ISCSI_SW_TCP_DBG(_conn, dbg_fmt, arg...)		\
68 	do {							\
69 		if (iscsi_sw_tcp_dbg)				\
70 			iscsi_conn_printk(KERN_INFO, _conn,	\
71 					     "%s " dbg_fmt,	\
72 					     __func__, ##arg);	\
73 	} while (0);
74 
75 
76 /**
77  * iscsi_sw_tcp_recv - TCP receive in sendfile fashion
78  * @rd_desc: read descriptor
79  * @skb: socket buffer
80  * @offset: offset in skb
81  * @len: skb->len - offset
82  */
83 static int iscsi_sw_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
84 			     unsigned int offset, size_t len)
85 {
86 	struct iscsi_conn *conn = rd_desc->arg.data;
87 	unsigned int consumed, total_consumed = 0;
88 	int status;
89 
90 	ISCSI_SW_TCP_DBG(conn, "in %d bytes\n", skb->len - offset);
91 
92 	do {
93 		status = 0;
94 		consumed = iscsi_tcp_recv_skb(conn, skb, offset, 0, &status);
95 		offset += consumed;
96 		total_consumed += consumed;
97 	} while (consumed != 0 && status != ISCSI_TCP_SKB_DONE);
98 
99 	ISCSI_SW_TCP_DBG(conn, "read %d bytes status %d\n",
100 			 skb->len - offset, status);
101 	return total_consumed;
102 }
103 
104 /**
105  * iscsi_sw_sk_state_check - check socket state
106  * @sk: socket
107  *
108  * If the socket is in CLOSE or CLOSE_WAIT we should
109  * not close the connection if there is still some
110  * data pending.
111  *
112  * Must be called with sk_callback_lock.
113  */
114 static inline int iscsi_sw_sk_state_check(struct sock *sk)
115 {
116 	struct iscsi_conn *conn = sk->sk_user_data;
117 
118 	if ((sk->sk_state == TCP_CLOSE_WAIT || sk->sk_state == TCP_CLOSE) &&
119 	    (conn->session->state != ISCSI_STATE_LOGGING_OUT) &&
120 	    !atomic_read(&sk->sk_rmem_alloc)) {
121 		ISCSI_SW_TCP_DBG(conn, "TCP_CLOSE|TCP_CLOSE_WAIT\n");
122 		iscsi_conn_failure(conn, ISCSI_ERR_TCP_CONN_CLOSE);
123 		return -ECONNRESET;
124 	}
125 	return 0;
126 }
127 
128 static void iscsi_sw_tcp_data_ready(struct sock *sk)
129 {
130 	struct iscsi_conn *conn;
131 	struct iscsi_tcp_conn *tcp_conn;
132 	read_descriptor_t rd_desc;
133 
134 	read_lock(&sk->sk_callback_lock);
135 	conn = sk->sk_user_data;
136 	if (!conn) {
137 		read_unlock(&sk->sk_callback_lock);
138 		return;
139 	}
140 	tcp_conn = conn->dd_data;
141 
142 	/*
143 	 * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
144 	 * We set count to 1 because we want the network layer to
145 	 * hand us all the skbs that are available. iscsi_tcp_recv
146 	 * handled pdus that cross buffers or pdus that still need data.
147 	 */
148 	rd_desc.arg.data = conn;
149 	rd_desc.count = 1;
150 	tcp_read_sock(sk, &rd_desc, iscsi_sw_tcp_recv);
151 
152 	iscsi_sw_sk_state_check(sk);
153 
154 	/* If we had to (atomically) map a highmem page,
155 	 * unmap it now. */
156 	iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
157 	read_unlock(&sk->sk_callback_lock);
158 }
159 
160 static void iscsi_sw_tcp_state_change(struct sock *sk)
161 {
162 	struct iscsi_tcp_conn *tcp_conn;
163 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
164 	struct iscsi_conn *conn;
165 	struct iscsi_session *session;
166 	void (*old_state_change)(struct sock *);
167 
168 	read_lock(&sk->sk_callback_lock);
169 	conn = sk->sk_user_data;
170 	if (!conn) {
171 		read_unlock(&sk->sk_callback_lock);
172 		return;
173 	}
174 	session = conn->session;
175 
176 	iscsi_sw_sk_state_check(sk);
177 
178 	tcp_conn = conn->dd_data;
179 	tcp_sw_conn = tcp_conn->dd_data;
180 	old_state_change = tcp_sw_conn->old_state_change;
181 
182 	read_unlock(&sk->sk_callback_lock);
183 
184 	old_state_change(sk);
185 }
186 
187 /**
188  * iscsi_write_space - Called when more output buffer space is available
189  * @sk: socket space is available for
190  **/
191 static void iscsi_sw_tcp_write_space(struct sock *sk)
192 {
193 	struct iscsi_conn *conn;
194 	struct iscsi_tcp_conn *tcp_conn;
195 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
196 	void (*old_write_space)(struct sock *);
197 
198 	read_lock_bh(&sk->sk_callback_lock);
199 	conn = sk->sk_user_data;
200 	if (!conn) {
201 		read_unlock_bh(&sk->sk_callback_lock);
202 		return;
203 	}
204 
205 	tcp_conn = conn->dd_data;
206 	tcp_sw_conn = tcp_conn->dd_data;
207 	old_write_space = tcp_sw_conn->old_write_space;
208 	read_unlock_bh(&sk->sk_callback_lock);
209 
210 	old_write_space(sk);
211 
212 	ISCSI_SW_TCP_DBG(conn, "iscsi_write_space\n");
213 	iscsi_conn_queue_work(conn);
214 }
215 
216 static void iscsi_sw_tcp_conn_set_callbacks(struct iscsi_conn *conn)
217 {
218 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
219 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
220 	struct sock *sk = tcp_sw_conn->sock->sk;
221 
222 	/* assign new callbacks */
223 	write_lock_bh(&sk->sk_callback_lock);
224 	sk->sk_user_data = conn;
225 	tcp_sw_conn->old_data_ready = sk->sk_data_ready;
226 	tcp_sw_conn->old_state_change = sk->sk_state_change;
227 	tcp_sw_conn->old_write_space = sk->sk_write_space;
228 	sk->sk_data_ready = iscsi_sw_tcp_data_ready;
229 	sk->sk_state_change = iscsi_sw_tcp_state_change;
230 	sk->sk_write_space = iscsi_sw_tcp_write_space;
231 	write_unlock_bh(&sk->sk_callback_lock);
232 }
233 
234 static void
235 iscsi_sw_tcp_conn_restore_callbacks(struct iscsi_conn *conn)
236 {
237 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
238 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
239 	struct sock *sk = tcp_sw_conn->sock->sk;
240 
241 	/* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
242 	write_lock_bh(&sk->sk_callback_lock);
243 	sk->sk_user_data    = NULL;
244 	sk->sk_data_ready   = tcp_sw_conn->old_data_ready;
245 	sk->sk_state_change = tcp_sw_conn->old_state_change;
246 	sk->sk_write_space  = tcp_sw_conn->old_write_space;
247 	sk->sk_no_check_tx = 0;
248 	write_unlock_bh(&sk->sk_callback_lock);
249 }
250 
251 /**
252  * iscsi_sw_tcp_xmit_segment - transmit segment
253  * @tcp_conn: the iSCSI TCP connection
254  * @segment: the buffer to transmnit
255  *
256  * This function transmits as much of the buffer as
257  * the network layer will accept, and returns the number of
258  * bytes transmitted.
259  *
260  * If CRC hashing is enabled, the function will compute the
261  * hash as it goes. When the entire segment has been transmitted,
262  * it will retrieve the hash value and send it as well.
263  */
264 static int iscsi_sw_tcp_xmit_segment(struct iscsi_tcp_conn *tcp_conn,
265 				     struct iscsi_segment *segment)
266 {
267 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
268 	struct socket *sk = tcp_sw_conn->sock;
269 	unsigned int copied = 0;
270 	int r = 0;
271 
272 	while (!iscsi_tcp_segment_done(tcp_conn, segment, 0, r)) {
273 		struct scatterlist *sg;
274 		unsigned int offset, copy;
275 		int flags = 0;
276 
277 		r = 0;
278 		offset = segment->copied;
279 		copy = segment->size - offset;
280 
281 		if (segment->total_copied + segment->size < segment->total_size)
282 			flags |= MSG_MORE;
283 
284 		/* Use sendpage if we can; else fall back to sendmsg */
285 		if (!segment->data) {
286 			sg = segment->sg;
287 			offset += segment->sg_offset + sg->offset;
288 			r = tcp_sw_conn->sendpage(sk, sg_page(sg), offset,
289 						  copy, flags);
290 		} else {
291 			struct msghdr msg = { .msg_flags = flags };
292 			struct kvec iov = {
293 				.iov_base = segment->data + offset,
294 				.iov_len = copy
295 			};
296 
297 			r = kernel_sendmsg(sk, &msg, &iov, 1, copy);
298 		}
299 
300 		if (r < 0) {
301 			iscsi_tcp_segment_unmap(segment);
302 			return r;
303 		}
304 		copied += r;
305 	}
306 	return copied;
307 }
308 
309 /**
310  * iscsi_sw_tcp_xmit - TCP transmit
311  **/
312 static int iscsi_sw_tcp_xmit(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 	unsigned int consumed = 0;
318 	int rc = 0;
319 
320 	while (1) {
321 		rc = iscsi_sw_tcp_xmit_segment(tcp_conn, segment);
322 		/*
323 		 * We may not have been able to send data because the conn
324 		 * is getting stopped. libiscsi will know so propagate err
325 		 * for it to do the right thing.
326 		 */
327 		if (rc == -EAGAIN)
328 			return rc;
329 		else if (rc < 0) {
330 			rc = ISCSI_ERR_XMIT_FAILED;
331 			goto error;
332 		} else if (rc == 0)
333 			break;
334 
335 		consumed += rc;
336 
337 		if (segment->total_copied >= segment->total_size) {
338 			if (segment->done != NULL) {
339 				rc = segment->done(tcp_conn, segment);
340 				if (rc != 0)
341 					goto error;
342 			}
343 		}
344 	}
345 
346 	ISCSI_SW_TCP_DBG(conn, "xmit %d bytes\n", consumed);
347 
348 	conn->txdata_octets += consumed;
349 	return consumed;
350 
351 error:
352 	/* Transmit error. We could initiate error recovery
353 	 * here. */
354 	ISCSI_SW_TCP_DBG(conn, "Error sending PDU, errno=%d\n", rc);
355 	iscsi_conn_failure(conn, rc);
356 	return -EIO;
357 }
358 
359 /**
360  * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
361  */
362 static inline int iscsi_sw_tcp_xmit_qlen(struct iscsi_conn *conn)
363 {
364 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
365 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
366 	struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
367 
368 	return segment->total_copied - segment->total_size;
369 }
370 
371 static int iscsi_sw_tcp_pdu_xmit(struct iscsi_task *task)
372 {
373 	struct iscsi_conn *conn = task->conn;
374 	unsigned long pflags = current->flags;
375 	int rc = 0;
376 
377 	current->flags |= PF_MEMALLOC;
378 
379 	while (iscsi_sw_tcp_xmit_qlen(conn)) {
380 		rc = iscsi_sw_tcp_xmit(conn);
381 		if (rc == 0) {
382 			rc = -EAGAIN;
383 			break;
384 		}
385 		if (rc < 0)
386 			break;
387 		rc = 0;
388 	}
389 
390 	tsk_restore_flags(current, pflags, PF_MEMALLOC);
391 	return rc;
392 }
393 
394 /*
395  * This is called when we're done sending the header.
396  * Simply copy the data_segment to the send segment, and return.
397  */
398 static int iscsi_sw_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
399 				      struct iscsi_segment *segment)
400 {
401 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
402 
403 	tcp_sw_conn->out.segment = tcp_sw_conn->out.data_segment;
404 	ISCSI_SW_TCP_DBG(tcp_conn->iscsi_conn,
405 			 "Header done. Next segment size %u total_size %u\n",
406 			 tcp_sw_conn->out.segment.size,
407 			 tcp_sw_conn->out.segment.total_size);
408 	return 0;
409 }
410 
411 static void iscsi_sw_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr,
412 				       size_t hdrlen)
413 {
414 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
415 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
416 
417 	ISCSI_SW_TCP_DBG(conn, "%s\n", conn->hdrdgst_en ?
418 			 "digest enabled" : "digest disabled");
419 
420 	/* Clear the data segment - needs to be filled in by the
421 	 * caller using iscsi_tcp_send_data_prep() */
422 	memset(&tcp_sw_conn->out.data_segment, 0,
423 	       sizeof(struct iscsi_segment));
424 
425 	/* If header digest is enabled, compute the CRC and
426 	 * place the digest into the same buffer. We make
427 	 * sure that both iscsi_tcp_task and mtask have
428 	 * sufficient room.
429 	 */
430 	if (conn->hdrdgst_en) {
431 		iscsi_tcp_dgst_header(&tcp_sw_conn->tx_hash, hdr, hdrlen,
432 				      hdr + hdrlen);
433 		hdrlen += ISCSI_DIGEST_SIZE;
434 	}
435 
436 	/* Remember header pointer for later, when we need
437 	 * to decide whether there's a payload to go along
438 	 * with the header. */
439 	tcp_sw_conn->out.hdr = hdr;
440 
441 	iscsi_segment_init_linear(&tcp_sw_conn->out.segment, hdr, hdrlen,
442 				  iscsi_sw_tcp_send_hdr_done, NULL);
443 }
444 
445 /*
446  * Prepare the send buffer for the payload data.
447  * Padding and checksumming will all be taken care
448  * of by the iscsi_segment routines.
449  */
450 static int
451 iscsi_sw_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
452 			    unsigned int count, unsigned int offset,
453 			    unsigned int len)
454 {
455 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
456 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
457 	struct hash_desc *tx_hash = NULL;
458 	unsigned int hdr_spec_len;
459 
460 	ISCSI_SW_TCP_DBG(conn, "offset=%d, datalen=%d %s\n", offset, len,
461 			 conn->datadgst_en ?
462 			 "digest enabled" : "digest disabled");
463 
464 	/* Make sure the datalen matches what the caller
465 	   said he would send. */
466 	hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
467 	WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
468 
469 	if (conn->datadgst_en)
470 		tx_hash = &tcp_sw_conn->tx_hash;
471 
472 	return iscsi_segment_seek_sg(&tcp_sw_conn->out.data_segment,
473 				     sg, count, offset, len,
474 				     NULL, tx_hash);
475 }
476 
477 static void
478 iscsi_sw_tcp_send_linear_data_prep(struct iscsi_conn *conn, void *data,
479 				   size_t len)
480 {
481 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
482 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
483 	struct hash_desc *tx_hash = NULL;
484 	unsigned int hdr_spec_len;
485 
486 	ISCSI_SW_TCP_DBG(conn, "datalen=%zd %s\n", len, conn->datadgst_en ?
487 			 "digest enabled" : "digest disabled");
488 
489 	/* Make sure the datalen matches what the caller
490 	   said he would send. */
491 	hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
492 	WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
493 
494 	if (conn->datadgst_en)
495 		tx_hash = &tcp_sw_conn->tx_hash;
496 
497 	iscsi_segment_init_linear(&tcp_sw_conn->out.data_segment,
498 				data, len, NULL, tx_hash);
499 }
500 
501 static int iscsi_sw_tcp_pdu_init(struct iscsi_task *task,
502 				 unsigned int offset, unsigned int count)
503 {
504 	struct iscsi_conn *conn = task->conn;
505 	int err = 0;
506 
507 	iscsi_sw_tcp_send_hdr_prep(conn, task->hdr, task->hdr_len);
508 
509 	if (!count)
510 		return 0;
511 
512 	if (!task->sc)
513 		iscsi_sw_tcp_send_linear_data_prep(conn, task->data, count);
514 	else {
515 		struct scsi_data_buffer *sdb = scsi_out(task->sc);
516 
517 		err = iscsi_sw_tcp_send_data_prep(conn, sdb->table.sgl,
518 						  sdb->table.nents, offset,
519 						  count);
520 	}
521 
522 	if (err) {
523 		/* got invalid offset/len */
524 		return -EIO;
525 	}
526 	return 0;
527 }
528 
529 static int iscsi_sw_tcp_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
530 {
531 	struct iscsi_tcp_task *tcp_task = task->dd_data;
532 
533 	task->hdr = task->dd_data + sizeof(*tcp_task);
534 	task->hdr_max = sizeof(struct iscsi_sw_tcp_hdrbuf) - ISCSI_DIGEST_SIZE;
535 	return 0;
536 }
537 
538 static struct iscsi_cls_conn *
539 iscsi_sw_tcp_conn_create(struct iscsi_cls_session *cls_session,
540 			 uint32_t conn_idx)
541 {
542 	struct iscsi_conn *conn;
543 	struct iscsi_cls_conn *cls_conn;
544 	struct iscsi_tcp_conn *tcp_conn;
545 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
546 
547 	cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*tcp_sw_conn),
548 					conn_idx);
549 	if (!cls_conn)
550 		return NULL;
551 	conn = cls_conn->dd_data;
552 	tcp_conn = conn->dd_data;
553 	tcp_sw_conn = tcp_conn->dd_data;
554 
555 	tcp_sw_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
556 						     CRYPTO_ALG_ASYNC);
557 	tcp_sw_conn->tx_hash.flags = 0;
558 	if (IS_ERR(tcp_sw_conn->tx_hash.tfm))
559 		goto free_conn;
560 
561 	tcp_sw_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
562 						     CRYPTO_ALG_ASYNC);
563 	tcp_sw_conn->rx_hash.flags = 0;
564 	if (IS_ERR(tcp_sw_conn->rx_hash.tfm))
565 		goto free_tx_tfm;
566 	tcp_conn->rx_hash = &tcp_sw_conn->rx_hash;
567 
568 	return cls_conn;
569 
570 free_tx_tfm:
571 	crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
572 free_conn:
573 	iscsi_conn_printk(KERN_ERR, conn,
574 			  "Could not create connection due to crc32c "
575 			  "loading error. Make sure the crc32c "
576 			  "module is built as a module or into the "
577 			  "kernel\n");
578 	iscsi_tcp_conn_teardown(cls_conn);
579 	return NULL;
580 }
581 
582 static void iscsi_sw_tcp_release_conn(struct iscsi_conn *conn)
583 {
584 	struct iscsi_session *session = conn->session;
585 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
586 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
587 	struct socket *sock = tcp_sw_conn->sock;
588 
589 	if (!sock)
590 		return;
591 
592 	sock_hold(sock->sk);
593 	iscsi_sw_tcp_conn_restore_callbacks(conn);
594 	sock_put(sock->sk);
595 
596 	spin_lock_bh(&session->frwd_lock);
597 	tcp_sw_conn->sock = NULL;
598 	spin_unlock_bh(&session->frwd_lock);
599 	sockfd_put(sock);
600 }
601 
602 static void iscsi_sw_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
603 {
604 	struct iscsi_conn *conn = cls_conn->dd_data;
605 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
606 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
607 
608 	iscsi_sw_tcp_release_conn(conn);
609 
610 	if (tcp_sw_conn->tx_hash.tfm)
611 		crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
612 	if (tcp_sw_conn->rx_hash.tfm)
613 		crypto_free_hash(tcp_sw_conn->rx_hash.tfm);
614 
615 	iscsi_tcp_conn_teardown(cls_conn);
616 }
617 
618 static void iscsi_sw_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
619 {
620 	struct iscsi_conn *conn = cls_conn->dd_data;
621 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
622 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
623 	struct socket *sock = tcp_sw_conn->sock;
624 
625 	/* userspace may have goofed up and not bound us */
626 	if (!sock)
627 		return;
628 
629 	sock->sk->sk_err = EIO;
630 	wake_up_interruptible(sk_sleep(sock->sk));
631 
632 	/* stop xmit side */
633 	iscsi_suspend_tx(conn);
634 
635 	/* stop recv side and release socket */
636 	iscsi_sw_tcp_release_conn(conn);
637 
638 	iscsi_conn_stop(cls_conn, flag);
639 }
640 
641 static int
642 iscsi_sw_tcp_conn_bind(struct iscsi_cls_session *cls_session,
643 		       struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
644 		       int is_leading)
645 {
646 	struct iscsi_session *session = cls_session->dd_data;
647 	struct iscsi_conn *conn = cls_conn->dd_data;
648 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
649 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
650 	struct sock *sk;
651 	struct socket *sock;
652 	int err;
653 
654 	/* lookup for existing socket */
655 	sock = sockfd_lookup((int)transport_eph, &err);
656 	if (!sock) {
657 		iscsi_conn_printk(KERN_ERR, conn,
658 				  "sockfd_lookup failed %d\n", err);
659 		return -EEXIST;
660 	}
661 
662 	err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
663 	if (err)
664 		goto free_socket;
665 
666 	spin_lock_bh(&session->frwd_lock);
667 	/* bind iSCSI connection and socket */
668 	tcp_sw_conn->sock = sock;
669 	spin_unlock_bh(&session->frwd_lock);
670 
671 	/* setup Socket parameters */
672 	sk = sock->sk;
673 	sk->sk_reuse = SK_CAN_REUSE;
674 	sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
675 	sk->sk_allocation = GFP_ATOMIC;
676 	sk_set_memalloc(sk);
677 
678 	iscsi_sw_tcp_conn_set_callbacks(conn);
679 	tcp_sw_conn->sendpage = tcp_sw_conn->sock->ops->sendpage;
680 	/*
681 	 * set receive state machine into initial state
682 	 */
683 	iscsi_tcp_hdr_recv_prep(tcp_conn);
684 	return 0;
685 
686 free_socket:
687 	sockfd_put(sock);
688 	return err;
689 }
690 
691 static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn,
692 				       enum iscsi_param param, char *buf,
693 				       int buflen)
694 {
695 	struct iscsi_conn *conn = cls_conn->dd_data;
696 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
697 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
698 
699 	switch(param) {
700 	case ISCSI_PARAM_HDRDGST_EN:
701 		iscsi_set_param(cls_conn, param, buf, buflen);
702 		break;
703 	case ISCSI_PARAM_DATADGST_EN:
704 		iscsi_set_param(cls_conn, param, buf, buflen);
705 		tcp_sw_conn->sendpage = conn->datadgst_en ?
706 			sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage;
707 		break;
708 	case ISCSI_PARAM_MAX_R2T:
709 		return iscsi_tcp_set_max_r2t(conn, buf);
710 	default:
711 		return iscsi_set_param(cls_conn, param, buf, buflen);
712 	}
713 
714 	return 0;
715 }
716 
717 static int iscsi_sw_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
718 				       enum iscsi_param param, char *buf)
719 {
720 	struct iscsi_conn *conn = cls_conn->dd_data;
721 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
722 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
723 	struct sockaddr_in6 addr;
724 	int rc, len;
725 
726 	switch(param) {
727 	case ISCSI_PARAM_CONN_PORT:
728 	case ISCSI_PARAM_CONN_ADDRESS:
729 	case ISCSI_PARAM_LOCAL_PORT:
730 		spin_lock_bh(&conn->session->frwd_lock);
731 		if (!tcp_sw_conn || !tcp_sw_conn->sock) {
732 			spin_unlock_bh(&conn->session->frwd_lock);
733 			return -ENOTCONN;
734 		}
735 		if (param == ISCSI_PARAM_LOCAL_PORT)
736 			rc = kernel_getsockname(tcp_sw_conn->sock,
737 						(struct sockaddr *)&addr, &len);
738 		else
739 			rc = kernel_getpeername(tcp_sw_conn->sock,
740 						(struct sockaddr *)&addr, &len);
741 		spin_unlock_bh(&conn->session->frwd_lock);
742 		if (rc)
743 			return rc;
744 
745 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
746 						 &addr, param, buf);
747 	default:
748 		return iscsi_conn_get_param(cls_conn, param, buf);
749 	}
750 
751 	return 0;
752 }
753 
754 static int iscsi_sw_tcp_host_get_param(struct Scsi_Host *shost,
755 				       enum iscsi_host_param param, char *buf)
756 {
757 	struct iscsi_sw_tcp_host *tcp_sw_host = iscsi_host_priv(shost);
758 	struct iscsi_session *session = tcp_sw_host->session;
759 	struct iscsi_conn *conn;
760 	struct iscsi_tcp_conn *tcp_conn;
761 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
762 	struct sockaddr_in6 addr;
763 	int rc, len;
764 
765 	switch (param) {
766 	case ISCSI_HOST_PARAM_IPADDRESS:
767 		if (!session)
768 			return -ENOTCONN;
769 
770 		spin_lock_bh(&session->frwd_lock);
771 		conn = session->leadconn;
772 		if (!conn) {
773 			spin_unlock_bh(&session->frwd_lock);
774 			return -ENOTCONN;
775 		}
776 		tcp_conn = conn->dd_data;
777 
778 		tcp_sw_conn = tcp_conn->dd_data;
779 		if (!tcp_sw_conn->sock) {
780 			spin_unlock_bh(&session->frwd_lock);
781 			return -ENOTCONN;
782 		}
783 
784 		rc = kernel_getsockname(tcp_sw_conn->sock,
785 					(struct sockaddr *)&addr, &len);
786 		spin_unlock_bh(&session->frwd_lock);
787 		if (rc)
788 			return rc;
789 
790 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
791 						 &addr, param, buf);
792 	default:
793 		return iscsi_host_get_param(shost, param, buf);
794 	}
795 
796 	return 0;
797 }
798 
799 static void
800 iscsi_sw_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
801 			    struct iscsi_stats *stats)
802 {
803 	struct iscsi_conn *conn = cls_conn->dd_data;
804 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
805 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
806 
807 	stats->custom_length = 3;
808 	strcpy(stats->custom[0].desc, "tx_sendpage_failures");
809 	stats->custom[0].value = tcp_sw_conn->sendpage_failures_cnt;
810 	strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
811 	stats->custom[1].value = tcp_sw_conn->discontiguous_hdr_cnt;
812 	strcpy(stats->custom[2].desc, "eh_abort_cnt");
813 	stats->custom[2].value = conn->eh_abort_cnt;
814 
815 	iscsi_tcp_conn_get_stats(cls_conn, stats);
816 }
817 
818 static struct iscsi_cls_session *
819 iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
820 			    uint16_t qdepth, uint32_t initial_cmdsn)
821 {
822 	struct iscsi_cls_session *cls_session;
823 	struct iscsi_session *session;
824 	struct iscsi_sw_tcp_host *tcp_sw_host;
825 	struct Scsi_Host *shost;
826 
827 	if (ep) {
828 		printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
829 		return NULL;
830 	}
831 
832 	shost = iscsi_host_alloc(&iscsi_sw_tcp_sht,
833 				 sizeof(struct iscsi_sw_tcp_host), 1);
834 	if (!shost)
835 		return NULL;
836 	shost->transportt = iscsi_sw_tcp_scsi_transport;
837 	shost->cmd_per_lun = qdepth;
838 	shost->max_lun = iscsi_max_lun;
839 	shost->max_id = 0;
840 	shost->max_channel = 0;
841 	shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE;
842 
843 	if (iscsi_host_add(shost, NULL))
844 		goto free_host;
845 
846 	cls_session = iscsi_session_setup(&iscsi_sw_tcp_transport, shost,
847 					  cmds_max, 0,
848 					  sizeof(struct iscsi_tcp_task) +
849 					  sizeof(struct iscsi_sw_tcp_hdrbuf),
850 					  initial_cmdsn, 0);
851 	if (!cls_session)
852 		goto remove_host;
853 	session = cls_session->dd_data;
854 	tcp_sw_host = iscsi_host_priv(shost);
855 	tcp_sw_host->session = session;
856 
857 	shost->can_queue = session->scsi_cmds_max;
858 	if (iscsi_tcp_r2tpool_alloc(session))
859 		goto remove_session;
860 	return cls_session;
861 
862 remove_session:
863 	iscsi_session_teardown(cls_session);
864 remove_host:
865 	iscsi_host_remove(shost);
866 free_host:
867 	iscsi_host_free(shost);
868 	return NULL;
869 }
870 
871 static void iscsi_sw_tcp_session_destroy(struct iscsi_cls_session *cls_session)
872 {
873 	struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
874 
875 	iscsi_tcp_r2tpool_free(cls_session->dd_data);
876 	iscsi_session_teardown(cls_session);
877 
878 	iscsi_host_remove(shost);
879 	iscsi_host_free(shost);
880 }
881 
882 static umode_t iscsi_sw_tcp_attr_is_visible(int param_type, int param)
883 {
884 	switch (param_type) {
885 	case ISCSI_HOST_PARAM:
886 		switch (param) {
887 		case ISCSI_HOST_PARAM_NETDEV_NAME:
888 		case ISCSI_HOST_PARAM_HWADDRESS:
889 		case ISCSI_HOST_PARAM_IPADDRESS:
890 		case ISCSI_HOST_PARAM_INITIATOR_NAME:
891 			return S_IRUGO;
892 		default:
893 			return 0;
894 		}
895 	case ISCSI_PARAM:
896 		switch (param) {
897 		case ISCSI_PARAM_MAX_RECV_DLENGTH:
898 		case ISCSI_PARAM_MAX_XMIT_DLENGTH:
899 		case ISCSI_PARAM_HDRDGST_EN:
900 		case ISCSI_PARAM_DATADGST_EN:
901 		case ISCSI_PARAM_CONN_ADDRESS:
902 		case ISCSI_PARAM_CONN_PORT:
903 		case ISCSI_PARAM_LOCAL_PORT:
904 		case ISCSI_PARAM_EXP_STATSN:
905 		case ISCSI_PARAM_PERSISTENT_ADDRESS:
906 		case ISCSI_PARAM_PERSISTENT_PORT:
907 		case ISCSI_PARAM_PING_TMO:
908 		case ISCSI_PARAM_RECV_TMO:
909 		case ISCSI_PARAM_INITIAL_R2T_EN:
910 		case ISCSI_PARAM_MAX_R2T:
911 		case ISCSI_PARAM_IMM_DATA_EN:
912 		case ISCSI_PARAM_FIRST_BURST:
913 		case ISCSI_PARAM_MAX_BURST:
914 		case ISCSI_PARAM_PDU_INORDER_EN:
915 		case ISCSI_PARAM_DATASEQ_INORDER_EN:
916 		case ISCSI_PARAM_ERL:
917 		case ISCSI_PARAM_TARGET_NAME:
918 		case ISCSI_PARAM_TPGT:
919 		case ISCSI_PARAM_USERNAME:
920 		case ISCSI_PARAM_PASSWORD:
921 		case ISCSI_PARAM_USERNAME_IN:
922 		case ISCSI_PARAM_PASSWORD_IN:
923 		case ISCSI_PARAM_FAST_ABORT:
924 		case ISCSI_PARAM_ABORT_TMO:
925 		case ISCSI_PARAM_LU_RESET_TMO:
926 		case ISCSI_PARAM_TGT_RESET_TMO:
927 		case ISCSI_PARAM_IFACE_NAME:
928 		case ISCSI_PARAM_INITIATOR_NAME:
929 			return S_IRUGO;
930 		default:
931 			return 0;
932 		}
933 	}
934 
935 	return 0;
936 }
937 
938 static int iscsi_sw_tcp_slave_alloc(struct scsi_device *sdev)
939 {
940 	set_bit(QUEUE_FLAG_BIDI, &sdev->request_queue->queue_flags);
941 	return 0;
942 }
943 
944 static int iscsi_sw_tcp_slave_configure(struct scsi_device *sdev)
945 {
946 	blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
947 	blk_queue_dma_alignment(sdev->request_queue, 0);
948 	return 0;
949 }
950 
951 static struct scsi_host_template iscsi_sw_tcp_sht = {
952 	.module			= THIS_MODULE,
953 	.name			= "iSCSI Initiator over TCP/IP",
954 	.queuecommand           = iscsi_queuecommand,
955 	.change_queue_depth	= scsi_change_queue_depth,
956 	.can_queue		= ISCSI_DEF_XMIT_CMDS_MAX - 1,
957 	.sg_tablesize		= 4096,
958 	.max_sectors		= 0xFFFF,
959 	.cmd_per_lun		= ISCSI_DEF_CMD_PER_LUN,
960 	.eh_abort_handler       = iscsi_eh_abort,
961 	.eh_device_reset_handler= iscsi_eh_device_reset,
962 	.eh_target_reset_handler = iscsi_eh_recover_target,
963 	.use_clustering         = DISABLE_CLUSTERING,
964 	.slave_alloc            = iscsi_sw_tcp_slave_alloc,
965 	.slave_configure        = iscsi_sw_tcp_slave_configure,
966 	.target_alloc		= iscsi_target_alloc,
967 	.proc_name		= "iscsi_tcp",
968 	.this_id		= -1,
969 	.track_queue_depth	= 1,
970 };
971 
972 static struct iscsi_transport iscsi_sw_tcp_transport = {
973 	.owner			= THIS_MODULE,
974 	.name			= "tcp",
975 	.caps			= CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
976 				  | CAP_DATADGST,
977 	/* session management */
978 	.create_session		= iscsi_sw_tcp_session_create,
979 	.destroy_session	= iscsi_sw_tcp_session_destroy,
980 	/* connection management */
981 	.create_conn		= iscsi_sw_tcp_conn_create,
982 	.bind_conn		= iscsi_sw_tcp_conn_bind,
983 	.destroy_conn		= iscsi_sw_tcp_conn_destroy,
984 	.attr_is_visible	= iscsi_sw_tcp_attr_is_visible,
985 	.set_param		= iscsi_sw_tcp_conn_set_param,
986 	.get_conn_param		= iscsi_sw_tcp_conn_get_param,
987 	.get_session_param	= iscsi_session_get_param,
988 	.start_conn		= iscsi_conn_start,
989 	.stop_conn		= iscsi_sw_tcp_conn_stop,
990 	/* iscsi host params */
991 	.get_host_param		= iscsi_sw_tcp_host_get_param,
992 	.set_host_param		= iscsi_host_set_param,
993 	/* IO */
994 	.send_pdu		= iscsi_conn_send_pdu,
995 	.get_stats		= iscsi_sw_tcp_conn_get_stats,
996 	/* iscsi task/cmd helpers */
997 	.init_task		= iscsi_tcp_task_init,
998 	.xmit_task		= iscsi_tcp_task_xmit,
999 	.cleanup_task		= iscsi_tcp_cleanup_task,
1000 	/* low level pdu helpers */
1001 	.xmit_pdu		= iscsi_sw_tcp_pdu_xmit,
1002 	.init_pdu		= iscsi_sw_tcp_pdu_init,
1003 	.alloc_pdu		= iscsi_sw_tcp_pdu_alloc,
1004 	/* recovery */
1005 	.session_recovery_timedout = iscsi_session_recovery_timedout,
1006 };
1007 
1008 static int __init iscsi_sw_tcp_init(void)
1009 {
1010 	if (iscsi_max_lun < 1) {
1011 		printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
1012 		       iscsi_max_lun);
1013 		return -EINVAL;
1014 	}
1015 
1016 	iscsi_sw_tcp_scsi_transport = iscsi_register_transport(
1017 						&iscsi_sw_tcp_transport);
1018 	if (!iscsi_sw_tcp_scsi_transport)
1019 		return -ENODEV;
1020 
1021 	return 0;
1022 }
1023 
1024 static void __exit iscsi_sw_tcp_exit(void)
1025 {
1026 	iscsi_unregister_transport(&iscsi_sw_tcp_transport);
1027 }
1028 
1029 module_init(iscsi_sw_tcp_init);
1030 module_exit(iscsi_sw_tcp_exit);
1031