xref: /openbmc/linux/drivers/scsi/iscsi_tcp.c (revision 42bc47b3)
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 <crypto/hash.h>
30 #include <linux/types.h>
31 #include <linux/inet.h>
32 #include <linux/slab.h>
33 #include <linux/sched/mm.h>
34 #include <linux/file.h>
35 #include <linux/blkdev.h>
36 #include <linux/delay.h>
37 #include <linux/kfifo.h>
38 #include <linux/scatterlist.h>
39 #include <linux/module.h>
40 #include <linux/backing-dev.h>
41 #include <net/tcp.h>
42 #include <scsi/scsi_cmnd.h>
43 #include <scsi/scsi_device.h>
44 #include <scsi/scsi_host.h>
45 #include <scsi/scsi.h>
46 #include <scsi/scsi_transport_iscsi.h>
47 
48 #include "iscsi_tcp.h"
49 
50 MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
51 	      "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
52 	      "Alex Aizman <itn780@yahoo.com>");
53 MODULE_DESCRIPTION("iSCSI/TCP data-path");
54 MODULE_LICENSE("GPL");
55 
56 static struct scsi_transport_template *iscsi_sw_tcp_scsi_transport;
57 static struct scsi_host_template iscsi_sw_tcp_sht;
58 static struct iscsi_transport iscsi_sw_tcp_transport;
59 
60 static unsigned int iscsi_max_lun = ~0;
61 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
62 
63 static int iscsi_sw_tcp_dbg;
64 module_param_named(debug_iscsi_tcp, iscsi_sw_tcp_dbg, int,
65 		   S_IRUGO | S_IWUSR);
66 MODULE_PARM_DESC(debug_iscsi_tcp, "Turn on debugging for iscsi_tcp module "
67 		 "Set to 1 to turn on, and zero to turn off. Default is off.");
68 
69 #define ISCSI_SW_TCP_DBG(_conn, dbg_fmt, arg...)		\
70 	do {							\
71 		if (iscsi_sw_tcp_dbg)				\
72 			iscsi_conn_printk(KERN_INFO, _conn,	\
73 					     "%s " dbg_fmt,	\
74 					     __func__, ##arg);	\
75 	} while (0);
76 
77 
78 /**
79  * iscsi_sw_tcp_recv - TCP receive in sendfile fashion
80  * @rd_desc: read descriptor
81  * @skb: socket buffer
82  * @offset: offset in skb
83  * @len: skb->len - offset
84  */
85 static int iscsi_sw_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
86 			     unsigned int offset, size_t len)
87 {
88 	struct iscsi_conn *conn = rd_desc->arg.data;
89 	unsigned int consumed, total_consumed = 0;
90 	int status;
91 
92 	ISCSI_SW_TCP_DBG(conn, "in %d bytes\n", skb->len - offset);
93 
94 	do {
95 		status = 0;
96 		consumed = iscsi_tcp_recv_skb(conn, skb, offset, 0, &status);
97 		offset += consumed;
98 		total_consumed += consumed;
99 	} while (consumed != 0 && status != ISCSI_TCP_SKB_DONE);
100 
101 	ISCSI_SW_TCP_DBG(conn, "read %d bytes status %d\n",
102 			 skb->len - offset, status);
103 	return total_consumed;
104 }
105 
106 /**
107  * iscsi_sw_sk_state_check - check socket state
108  * @sk: socket
109  *
110  * If the socket is in CLOSE or CLOSE_WAIT we should
111  * not close the connection if there is still some
112  * data pending.
113  *
114  * Must be called with sk_callback_lock.
115  */
116 static inline int iscsi_sw_sk_state_check(struct sock *sk)
117 {
118 	struct iscsi_conn *conn = sk->sk_user_data;
119 
120 	if ((sk->sk_state == TCP_CLOSE_WAIT || sk->sk_state == TCP_CLOSE) &&
121 	    (conn->session->state != ISCSI_STATE_LOGGING_OUT) &&
122 	    !atomic_read(&sk->sk_rmem_alloc)) {
123 		ISCSI_SW_TCP_DBG(conn, "TCP_CLOSE|TCP_CLOSE_WAIT\n");
124 		iscsi_conn_failure(conn, ISCSI_ERR_TCP_CONN_CLOSE);
125 		return -ECONNRESET;
126 	}
127 	return 0;
128 }
129 
130 static void iscsi_sw_tcp_data_ready(struct sock *sk)
131 {
132 	struct iscsi_conn *conn;
133 	struct iscsi_tcp_conn *tcp_conn;
134 	read_descriptor_t rd_desc;
135 
136 	read_lock_bh(&sk->sk_callback_lock);
137 	conn = sk->sk_user_data;
138 	if (!conn) {
139 		read_unlock_bh(&sk->sk_callback_lock);
140 		return;
141 	}
142 	tcp_conn = conn->dd_data;
143 
144 	/*
145 	 * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
146 	 * We set count to 1 because we want the network layer to
147 	 * hand us all the skbs that are available. iscsi_tcp_recv
148 	 * handled pdus that cross buffers or pdus that still need data.
149 	 */
150 	rd_desc.arg.data = conn;
151 	rd_desc.count = 1;
152 	tcp_read_sock(sk, &rd_desc, iscsi_sw_tcp_recv);
153 
154 	iscsi_sw_sk_state_check(sk);
155 
156 	/* If we had to (atomically) map a highmem page,
157 	 * unmap it now. */
158 	iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
159 	read_unlock_bh(&sk->sk_callback_lock);
160 }
161 
162 static void iscsi_sw_tcp_state_change(struct sock *sk)
163 {
164 	struct iscsi_tcp_conn *tcp_conn;
165 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
166 	struct iscsi_conn *conn;
167 	void (*old_state_change)(struct sock *);
168 
169 	read_lock_bh(&sk->sk_callback_lock);
170 	conn = sk->sk_user_data;
171 	if (!conn) {
172 		read_unlock_bh(&sk->sk_callback_lock);
173 		return;
174 	}
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_bh(&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  * @conn: iscsi connection
312  **/
313 static int iscsi_sw_tcp_xmit(struct iscsi_conn *conn)
314 {
315 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
316 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
317 	struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
318 	unsigned int consumed = 0;
319 	int rc = 0;
320 
321 	while (1) {
322 		rc = iscsi_sw_tcp_xmit_segment(tcp_conn, segment);
323 		/*
324 		 * We may not have been able to send data because the conn
325 		 * is getting stopped. libiscsi will know so propagate err
326 		 * for it to do the right thing.
327 		 */
328 		if (rc == -EAGAIN)
329 			return rc;
330 		else if (rc < 0) {
331 			rc = ISCSI_ERR_XMIT_FAILED;
332 			goto error;
333 		} else if (rc == 0)
334 			break;
335 
336 		consumed += rc;
337 
338 		if (segment->total_copied >= segment->total_size) {
339 			if (segment->done != NULL) {
340 				rc = segment->done(tcp_conn, segment);
341 				if (rc != 0)
342 					goto error;
343 			}
344 		}
345 	}
346 
347 	ISCSI_SW_TCP_DBG(conn, "xmit %d bytes\n", consumed);
348 
349 	conn->txdata_octets += consumed;
350 	return consumed;
351 
352 error:
353 	/* Transmit error. We could initiate error recovery
354 	 * here. */
355 	ISCSI_SW_TCP_DBG(conn, "Error sending PDU, errno=%d\n", rc);
356 	iscsi_conn_failure(conn, rc);
357 	return -EIO;
358 }
359 
360 /**
361  * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
362  * @conn: iscsi connection
363  */
364 static inline int iscsi_sw_tcp_xmit_qlen(struct iscsi_conn *conn)
365 {
366 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
367 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
368 	struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
369 
370 	return segment->total_copied - segment->total_size;
371 }
372 
373 static int iscsi_sw_tcp_pdu_xmit(struct iscsi_task *task)
374 {
375 	struct iscsi_conn *conn = task->conn;
376 	unsigned int noreclaim_flag;
377 	int rc = 0;
378 
379 	noreclaim_flag = memalloc_noreclaim_save();
380 
381 	while (iscsi_sw_tcp_xmit_qlen(conn)) {
382 		rc = iscsi_sw_tcp_xmit(conn);
383 		if (rc == 0) {
384 			rc = -EAGAIN;
385 			break;
386 		}
387 		if (rc < 0)
388 			break;
389 		rc = 0;
390 	}
391 
392 	memalloc_noreclaim_restore(noreclaim_flag);
393 	return rc;
394 }
395 
396 /*
397  * This is called when we're done sending the header.
398  * Simply copy the data_segment to the send segment, and return.
399  */
400 static int iscsi_sw_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
401 				      struct iscsi_segment *segment)
402 {
403 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
404 
405 	tcp_sw_conn->out.segment = tcp_sw_conn->out.data_segment;
406 	ISCSI_SW_TCP_DBG(tcp_conn->iscsi_conn,
407 			 "Header done. Next segment size %u total_size %u\n",
408 			 tcp_sw_conn->out.segment.size,
409 			 tcp_sw_conn->out.segment.total_size);
410 	return 0;
411 }
412 
413 static void iscsi_sw_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr,
414 				       size_t hdrlen)
415 {
416 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
417 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
418 
419 	ISCSI_SW_TCP_DBG(conn, "%s\n", conn->hdrdgst_en ?
420 			 "digest enabled" : "digest disabled");
421 
422 	/* Clear the data segment - needs to be filled in by the
423 	 * caller using iscsi_tcp_send_data_prep() */
424 	memset(&tcp_sw_conn->out.data_segment, 0,
425 	       sizeof(struct iscsi_segment));
426 
427 	/* If header digest is enabled, compute the CRC and
428 	 * place the digest into the same buffer. We make
429 	 * sure that both iscsi_tcp_task and mtask have
430 	 * sufficient room.
431 	 */
432 	if (conn->hdrdgst_en) {
433 		iscsi_tcp_dgst_header(tcp_sw_conn->tx_hash, hdr, hdrlen,
434 				      hdr + hdrlen);
435 		hdrlen += ISCSI_DIGEST_SIZE;
436 	}
437 
438 	/* Remember header pointer for later, when we need
439 	 * to decide whether there's a payload to go along
440 	 * with the header. */
441 	tcp_sw_conn->out.hdr = hdr;
442 
443 	iscsi_segment_init_linear(&tcp_sw_conn->out.segment, hdr, hdrlen,
444 				  iscsi_sw_tcp_send_hdr_done, NULL);
445 }
446 
447 /*
448  * Prepare the send buffer for the payload data.
449  * Padding and checksumming will all be taken care
450  * of by the iscsi_segment routines.
451  */
452 static int
453 iscsi_sw_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
454 			    unsigned int count, unsigned int offset,
455 			    unsigned int len)
456 {
457 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
458 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
459 	struct ahash_request *tx_hash = NULL;
460 	unsigned int hdr_spec_len;
461 
462 	ISCSI_SW_TCP_DBG(conn, "offset=%d, datalen=%d %s\n", offset, len,
463 			 conn->datadgst_en ?
464 			 "digest enabled" : "digest disabled");
465 
466 	/* Make sure the datalen matches what the caller
467 	   said he would send. */
468 	hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
469 	WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
470 
471 	if (conn->datadgst_en)
472 		tx_hash = tcp_sw_conn->tx_hash;
473 
474 	return iscsi_segment_seek_sg(&tcp_sw_conn->out.data_segment,
475 				     sg, count, offset, len,
476 				     NULL, tx_hash);
477 }
478 
479 static void
480 iscsi_sw_tcp_send_linear_data_prep(struct iscsi_conn *conn, void *data,
481 				   size_t len)
482 {
483 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
484 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
485 	struct ahash_request *tx_hash = NULL;
486 	unsigned int hdr_spec_len;
487 
488 	ISCSI_SW_TCP_DBG(conn, "datalen=%zd %s\n", len, conn->datadgst_en ?
489 			 "digest enabled" : "digest disabled");
490 
491 	/* Make sure the datalen matches what the caller
492 	   said he would send. */
493 	hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
494 	WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
495 
496 	if (conn->datadgst_en)
497 		tx_hash = tcp_sw_conn->tx_hash;
498 
499 	iscsi_segment_init_linear(&tcp_sw_conn->out.data_segment,
500 				data, len, NULL, tx_hash);
501 }
502 
503 static int iscsi_sw_tcp_pdu_init(struct iscsi_task *task,
504 				 unsigned int offset, unsigned int count)
505 {
506 	struct iscsi_conn *conn = task->conn;
507 	int err = 0;
508 
509 	iscsi_sw_tcp_send_hdr_prep(conn, task->hdr, task->hdr_len);
510 
511 	if (!count)
512 		return 0;
513 
514 	if (!task->sc)
515 		iscsi_sw_tcp_send_linear_data_prep(conn, task->data, count);
516 	else {
517 		struct scsi_data_buffer *sdb = scsi_out(task->sc);
518 
519 		err = iscsi_sw_tcp_send_data_prep(conn, sdb->table.sgl,
520 						  sdb->table.nents, offset,
521 						  count);
522 	}
523 
524 	if (err) {
525 		/* got invalid offset/len */
526 		return -EIO;
527 	}
528 	return 0;
529 }
530 
531 static int iscsi_sw_tcp_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
532 {
533 	struct iscsi_tcp_task *tcp_task = task->dd_data;
534 
535 	task->hdr = task->dd_data + sizeof(*tcp_task);
536 	task->hdr_max = sizeof(struct iscsi_sw_tcp_hdrbuf) - ISCSI_DIGEST_SIZE;
537 	return 0;
538 }
539 
540 static struct iscsi_cls_conn *
541 iscsi_sw_tcp_conn_create(struct iscsi_cls_session *cls_session,
542 			 uint32_t conn_idx)
543 {
544 	struct iscsi_conn *conn;
545 	struct iscsi_cls_conn *cls_conn;
546 	struct iscsi_tcp_conn *tcp_conn;
547 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
548 	struct crypto_ahash *tfm;
549 
550 	cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*tcp_sw_conn),
551 					conn_idx);
552 	if (!cls_conn)
553 		return NULL;
554 	conn = cls_conn->dd_data;
555 	tcp_conn = conn->dd_data;
556 	tcp_sw_conn = tcp_conn->dd_data;
557 
558 	tfm = crypto_alloc_ahash("crc32c", 0, CRYPTO_ALG_ASYNC);
559 	if (IS_ERR(tfm))
560 		goto free_conn;
561 
562 	tcp_sw_conn->tx_hash = ahash_request_alloc(tfm, GFP_KERNEL);
563 	if (!tcp_sw_conn->tx_hash)
564 		goto free_tfm;
565 	ahash_request_set_callback(tcp_sw_conn->tx_hash, 0, NULL, NULL);
566 
567 	tcp_sw_conn->rx_hash = ahash_request_alloc(tfm, GFP_KERNEL);
568 	if (!tcp_sw_conn->rx_hash)
569 		goto free_tx_hash;
570 	ahash_request_set_callback(tcp_sw_conn->rx_hash, 0, NULL, NULL);
571 
572 	tcp_conn->rx_hash = tcp_sw_conn->rx_hash;
573 
574 	return cls_conn;
575 
576 free_tx_hash:
577 	ahash_request_free(tcp_sw_conn->tx_hash);
578 free_tfm:
579 	crypto_free_ahash(tfm);
580 free_conn:
581 	iscsi_conn_printk(KERN_ERR, conn,
582 			  "Could not create connection due to crc32c "
583 			  "loading error. Make sure the crc32c "
584 			  "module is built as a module or into the "
585 			  "kernel\n");
586 	iscsi_tcp_conn_teardown(cls_conn);
587 	return NULL;
588 }
589 
590 static void iscsi_sw_tcp_release_conn(struct iscsi_conn *conn)
591 {
592 	struct iscsi_session *session = conn->session;
593 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
594 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
595 	struct socket *sock = tcp_sw_conn->sock;
596 
597 	if (!sock)
598 		return;
599 
600 	sock_hold(sock->sk);
601 	iscsi_sw_tcp_conn_restore_callbacks(conn);
602 	sock_put(sock->sk);
603 
604 	spin_lock_bh(&session->frwd_lock);
605 	tcp_sw_conn->sock = NULL;
606 	spin_unlock_bh(&session->frwd_lock);
607 	sockfd_put(sock);
608 }
609 
610 static void iscsi_sw_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
611 {
612 	struct iscsi_conn *conn = cls_conn->dd_data;
613 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
614 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
615 
616 	iscsi_sw_tcp_release_conn(conn);
617 
618 	ahash_request_free(tcp_sw_conn->rx_hash);
619 	if (tcp_sw_conn->tx_hash) {
620 		struct crypto_ahash *tfm;
621 
622 		tfm = crypto_ahash_reqtfm(tcp_sw_conn->tx_hash);
623 		ahash_request_free(tcp_sw_conn->tx_hash);
624 		crypto_free_ahash(tfm);
625 	}
626 
627 	iscsi_tcp_conn_teardown(cls_conn);
628 }
629 
630 static void iscsi_sw_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
631 {
632 	struct iscsi_conn *conn = cls_conn->dd_data;
633 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
634 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
635 	struct socket *sock = tcp_sw_conn->sock;
636 
637 	/* userspace may have goofed up and not bound us */
638 	if (!sock)
639 		return;
640 
641 	sock->sk->sk_err = EIO;
642 	wake_up_interruptible(sk_sleep(sock->sk));
643 
644 	/* stop xmit side */
645 	iscsi_suspend_tx(conn);
646 
647 	/* stop recv side and release socket */
648 	iscsi_sw_tcp_release_conn(conn);
649 
650 	iscsi_conn_stop(cls_conn, flag);
651 }
652 
653 static int
654 iscsi_sw_tcp_conn_bind(struct iscsi_cls_session *cls_session,
655 		       struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
656 		       int is_leading)
657 {
658 	struct iscsi_session *session = cls_session->dd_data;
659 	struct iscsi_conn *conn = cls_conn->dd_data;
660 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
661 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
662 	struct sock *sk;
663 	struct socket *sock;
664 	int err;
665 
666 	/* lookup for existing socket */
667 	sock = sockfd_lookup((int)transport_eph, &err);
668 	if (!sock) {
669 		iscsi_conn_printk(KERN_ERR, conn,
670 				  "sockfd_lookup failed %d\n", err);
671 		return -EEXIST;
672 	}
673 
674 	err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
675 	if (err)
676 		goto free_socket;
677 
678 	spin_lock_bh(&session->frwd_lock);
679 	/* bind iSCSI connection and socket */
680 	tcp_sw_conn->sock = sock;
681 	spin_unlock_bh(&session->frwd_lock);
682 
683 	/* setup Socket parameters */
684 	sk = sock->sk;
685 	sk->sk_reuse = SK_CAN_REUSE;
686 	sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
687 	sk->sk_allocation = GFP_ATOMIC;
688 	sk_set_memalloc(sk);
689 
690 	iscsi_sw_tcp_conn_set_callbacks(conn);
691 	tcp_sw_conn->sendpage = tcp_sw_conn->sock->ops->sendpage;
692 	/*
693 	 * set receive state machine into initial state
694 	 */
695 	iscsi_tcp_hdr_recv_prep(tcp_conn);
696 	return 0;
697 
698 free_socket:
699 	sockfd_put(sock);
700 	return err;
701 }
702 
703 static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn,
704 				       enum iscsi_param param, char *buf,
705 				       int buflen)
706 {
707 	struct iscsi_conn *conn = cls_conn->dd_data;
708 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
709 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
710 
711 	switch(param) {
712 	case ISCSI_PARAM_HDRDGST_EN:
713 		iscsi_set_param(cls_conn, param, buf, buflen);
714 		break;
715 	case ISCSI_PARAM_DATADGST_EN:
716 		iscsi_set_param(cls_conn, param, buf, buflen);
717 		tcp_sw_conn->sendpage = conn->datadgst_en ?
718 			sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage;
719 		break;
720 	case ISCSI_PARAM_MAX_R2T:
721 		return iscsi_tcp_set_max_r2t(conn, buf);
722 	default:
723 		return iscsi_set_param(cls_conn, param, buf, buflen);
724 	}
725 
726 	return 0;
727 }
728 
729 static int iscsi_sw_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
730 				       enum iscsi_param param, char *buf)
731 {
732 	struct iscsi_conn *conn = cls_conn->dd_data;
733 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
734 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
735 	struct sockaddr_in6 addr;
736 	int rc;
737 
738 	switch(param) {
739 	case ISCSI_PARAM_CONN_PORT:
740 	case ISCSI_PARAM_CONN_ADDRESS:
741 	case ISCSI_PARAM_LOCAL_PORT:
742 		spin_lock_bh(&conn->session->frwd_lock);
743 		if (!tcp_sw_conn || !tcp_sw_conn->sock) {
744 			spin_unlock_bh(&conn->session->frwd_lock);
745 			return -ENOTCONN;
746 		}
747 		if (param == ISCSI_PARAM_LOCAL_PORT)
748 			rc = kernel_getsockname(tcp_sw_conn->sock,
749 						(struct sockaddr *)&addr);
750 		else
751 			rc = kernel_getpeername(tcp_sw_conn->sock,
752 						(struct sockaddr *)&addr);
753 		spin_unlock_bh(&conn->session->frwd_lock);
754 		if (rc < 0)
755 			return rc;
756 
757 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
758 						 &addr, param, buf);
759 	default:
760 		return iscsi_conn_get_param(cls_conn, param, buf);
761 	}
762 
763 	return 0;
764 }
765 
766 static int iscsi_sw_tcp_host_get_param(struct Scsi_Host *shost,
767 				       enum iscsi_host_param param, char *buf)
768 {
769 	struct iscsi_sw_tcp_host *tcp_sw_host = iscsi_host_priv(shost);
770 	struct iscsi_session *session = tcp_sw_host->session;
771 	struct iscsi_conn *conn;
772 	struct iscsi_tcp_conn *tcp_conn;
773 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
774 	struct sockaddr_in6 addr;
775 	int rc;
776 
777 	switch (param) {
778 	case ISCSI_HOST_PARAM_IPADDRESS:
779 		if (!session)
780 			return -ENOTCONN;
781 
782 		spin_lock_bh(&session->frwd_lock);
783 		conn = session->leadconn;
784 		if (!conn) {
785 			spin_unlock_bh(&session->frwd_lock);
786 			return -ENOTCONN;
787 		}
788 		tcp_conn = conn->dd_data;
789 
790 		tcp_sw_conn = tcp_conn->dd_data;
791 		if (!tcp_sw_conn->sock) {
792 			spin_unlock_bh(&session->frwd_lock);
793 			return -ENOTCONN;
794 		}
795 
796 		rc = kernel_getsockname(tcp_sw_conn->sock,
797 					(struct sockaddr *)&addr);
798 		spin_unlock_bh(&session->frwd_lock);
799 		if (rc < 0)
800 			return rc;
801 
802 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
803 						 &addr, param, buf);
804 	default:
805 		return iscsi_host_get_param(shost, param, buf);
806 	}
807 
808 	return 0;
809 }
810 
811 static void
812 iscsi_sw_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
813 			    struct iscsi_stats *stats)
814 {
815 	struct iscsi_conn *conn = cls_conn->dd_data;
816 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
817 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
818 
819 	stats->custom_length = 3;
820 	strcpy(stats->custom[0].desc, "tx_sendpage_failures");
821 	stats->custom[0].value = tcp_sw_conn->sendpage_failures_cnt;
822 	strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
823 	stats->custom[1].value = tcp_sw_conn->discontiguous_hdr_cnt;
824 	strcpy(stats->custom[2].desc, "eh_abort_cnt");
825 	stats->custom[2].value = conn->eh_abort_cnt;
826 
827 	iscsi_tcp_conn_get_stats(cls_conn, stats);
828 }
829 
830 static struct iscsi_cls_session *
831 iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
832 			    uint16_t qdepth, uint32_t initial_cmdsn)
833 {
834 	struct iscsi_cls_session *cls_session;
835 	struct iscsi_session *session;
836 	struct iscsi_sw_tcp_host *tcp_sw_host;
837 	struct Scsi_Host *shost;
838 
839 	if (ep) {
840 		printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
841 		return NULL;
842 	}
843 
844 	shost = iscsi_host_alloc(&iscsi_sw_tcp_sht,
845 				 sizeof(struct iscsi_sw_tcp_host), 1);
846 	if (!shost)
847 		return NULL;
848 	shost->transportt = iscsi_sw_tcp_scsi_transport;
849 	shost->cmd_per_lun = qdepth;
850 	shost->max_lun = iscsi_max_lun;
851 	shost->max_id = 0;
852 	shost->max_channel = 0;
853 	shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE;
854 
855 	if (iscsi_host_add(shost, NULL))
856 		goto free_host;
857 
858 	cls_session = iscsi_session_setup(&iscsi_sw_tcp_transport, shost,
859 					  cmds_max, 0,
860 					  sizeof(struct iscsi_tcp_task) +
861 					  sizeof(struct iscsi_sw_tcp_hdrbuf),
862 					  initial_cmdsn, 0);
863 	if (!cls_session)
864 		goto remove_host;
865 	session = cls_session->dd_data;
866 	tcp_sw_host = iscsi_host_priv(shost);
867 	tcp_sw_host->session = session;
868 
869 	shost->can_queue = session->scsi_cmds_max;
870 	if (iscsi_tcp_r2tpool_alloc(session))
871 		goto remove_session;
872 	return cls_session;
873 
874 remove_session:
875 	iscsi_session_teardown(cls_session);
876 remove_host:
877 	iscsi_host_remove(shost);
878 free_host:
879 	iscsi_host_free(shost);
880 	return NULL;
881 }
882 
883 static void iscsi_sw_tcp_session_destroy(struct iscsi_cls_session *cls_session)
884 {
885 	struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
886 
887 	iscsi_tcp_r2tpool_free(cls_session->dd_data);
888 	iscsi_session_teardown(cls_session);
889 
890 	iscsi_host_remove(shost);
891 	iscsi_host_free(shost);
892 }
893 
894 static umode_t iscsi_sw_tcp_attr_is_visible(int param_type, int param)
895 {
896 	switch (param_type) {
897 	case ISCSI_HOST_PARAM:
898 		switch (param) {
899 		case ISCSI_HOST_PARAM_NETDEV_NAME:
900 		case ISCSI_HOST_PARAM_HWADDRESS:
901 		case ISCSI_HOST_PARAM_IPADDRESS:
902 		case ISCSI_HOST_PARAM_INITIATOR_NAME:
903 			return S_IRUGO;
904 		default:
905 			return 0;
906 		}
907 	case ISCSI_PARAM:
908 		switch (param) {
909 		case ISCSI_PARAM_MAX_RECV_DLENGTH:
910 		case ISCSI_PARAM_MAX_XMIT_DLENGTH:
911 		case ISCSI_PARAM_HDRDGST_EN:
912 		case ISCSI_PARAM_DATADGST_EN:
913 		case ISCSI_PARAM_CONN_ADDRESS:
914 		case ISCSI_PARAM_CONN_PORT:
915 		case ISCSI_PARAM_LOCAL_PORT:
916 		case ISCSI_PARAM_EXP_STATSN:
917 		case ISCSI_PARAM_PERSISTENT_ADDRESS:
918 		case ISCSI_PARAM_PERSISTENT_PORT:
919 		case ISCSI_PARAM_PING_TMO:
920 		case ISCSI_PARAM_RECV_TMO:
921 		case ISCSI_PARAM_INITIAL_R2T_EN:
922 		case ISCSI_PARAM_MAX_R2T:
923 		case ISCSI_PARAM_IMM_DATA_EN:
924 		case ISCSI_PARAM_FIRST_BURST:
925 		case ISCSI_PARAM_MAX_BURST:
926 		case ISCSI_PARAM_PDU_INORDER_EN:
927 		case ISCSI_PARAM_DATASEQ_INORDER_EN:
928 		case ISCSI_PARAM_ERL:
929 		case ISCSI_PARAM_TARGET_NAME:
930 		case ISCSI_PARAM_TPGT:
931 		case ISCSI_PARAM_USERNAME:
932 		case ISCSI_PARAM_PASSWORD:
933 		case ISCSI_PARAM_USERNAME_IN:
934 		case ISCSI_PARAM_PASSWORD_IN:
935 		case ISCSI_PARAM_FAST_ABORT:
936 		case ISCSI_PARAM_ABORT_TMO:
937 		case ISCSI_PARAM_LU_RESET_TMO:
938 		case ISCSI_PARAM_TGT_RESET_TMO:
939 		case ISCSI_PARAM_IFACE_NAME:
940 		case ISCSI_PARAM_INITIATOR_NAME:
941 			return S_IRUGO;
942 		default:
943 			return 0;
944 		}
945 	}
946 
947 	return 0;
948 }
949 
950 static int iscsi_sw_tcp_slave_alloc(struct scsi_device *sdev)
951 {
952 	blk_queue_flag_set(QUEUE_FLAG_BIDI, sdev->request_queue);
953 	return 0;
954 }
955 
956 static int iscsi_sw_tcp_slave_configure(struct scsi_device *sdev)
957 {
958 	struct iscsi_sw_tcp_host *tcp_sw_host = iscsi_host_priv(sdev->host);
959 	struct iscsi_session *session = tcp_sw_host->session;
960 	struct iscsi_conn *conn = session->leadconn;
961 
962 	if (conn->datadgst_en)
963 		sdev->request_queue->backing_dev_info->capabilities
964 			|= BDI_CAP_STABLE_WRITES;
965 	blk_queue_dma_alignment(sdev->request_queue, 0);
966 	return 0;
967 }
968 
969 static struct scsi_host_template iscsi_sw_tcp_sht = {
970 	.module			= THIS_MODULE,
971 	.name			= "iSCSI Initiator over TCP/IP",
972 	.queuecommand           = iscsi_queuecommand,
973 	.change_queue_depth	= scsi_change_queue_depth,
974 	.can_queue		= ISCSI_DEF_XMIT_CMDS_MAX - 1,
975 	.sg_tablesize		= 4096,
976 	.max_sectors		= 0xFFFF,
977 	.cmd_per_lun		= ISCSI_DEF_CMD_PER_LUN,
978 	.eh_timed_out		= iscsi_eh_cmd_timed_out,
979 	.eh_abort_handler       = iscsi_eh_abort,
980 	.eh_device_reset_handler= iscsi_eh_device_reset,
981 	.eh_target_reset_handler = iscsi_eh_recover_target,
982 	.use_clustering         = DISABLE_CLUSTERING,
983 	.slave_alloc            = iscsi_sw_tcp_slave_alloc,
984 	.slave_configure        = iscsi_sw_tcp_slave_configure,
985 	.target_alloc		= iscsi_target_alloc,
986 	.proc_name		= "iscsi_tcp",
987 	.this_id		= -1,
988 	.track_queue_depth	= 1,
989 };
990 
991 static struct iscsi_transport iscsi_sw_tcp_transport = {
992 	.owner			= THIS_MODULE,
993 	.name			= "tcp",
994 	.caps			= CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
995 				  | CAP_DATADGST,
996 	/* session management */
997 	.create_session		= iscsi_sw_tcp_session_create,
998 	.destroy_session	= iscsi_sw_tcp_session_destroy,
999 	/* connection management */
1000 	.create_conn		= iscsi_sw_tcp_conn_create,
1001 	.bind_conn		= iscsi_sw_tcp_conn_bind,
1002 	.destroy_conn		= iscsi_sw_tcp_conn_destroy,
1003 	.attr_is_visible	= iscsi_sw_tcp_attr_is_visible,
1004 	.set_param		= iscsi_sw_tcp_conn_set_param,
1005 	.get_conn_param		= iscsi_sw_tcp_conn_get_param,
1006 	.get_session_param	= iscsi_session_get_param,
1007 	.start_conn		= iscsi_conn_start,
1008 	.stop_conn		= iscsi_sw_tcp_conn_stop,
1009 	/* iscsi host params */
1010 	.get_host_param		= iscsi_sw_tcp_host_get_param,
1011 	.set_host_param		= iscsi_host_set_param,
1012 	/* IO */
1013 	.send_pdu		= iscsi_conn_send_pdu,
1014 	.get_stats		= iscsi_sw_tcp_conn_get_stats,
1015 	/* iscsi task/cmd helpers */
1016 	.init_task		= iscsi_tcp_task_init,
1017 	.xmit_task		= iscsi_tcp_task_xmit,
1018 	.cleanup_task		= iscsi_tcp_cleanup_task,
1019 	/* low level pdu helpers */
1020 	.xmit_pdu		= iscsi_sw_tcp_pdu_xmit,
1021 	.init_pdu		= iscsi_sw_tcp_pdu_init,
1022 	.alloc_pdu		= iscsi_sw_tcp_pdu_alloc,
1023 	/* recovery */
1024 	.session_recovery_timedout = iscsi_session_recovery_timedout,
1025 };
1026 
1027 static int __init iscsi_sw_tcp_init(void)
1028 {
1029 	if (iscsi_max_lun < 1) {
1030 		printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
1031 		       iscsi_max_lun);
1032 		return -EINVAL;
1033 	}
1034 
1035 	iscsi_sw_tcp_scsi_transport = iscsi_register_transport(
1036 						&iscsi_sw_tcp_transport);
1037 	if (!iscsi_sw_tcp_scsi_transport)
1038 		return -ENODEV;
1039 
1040 	return 0;
1041 }
1042 
1043 static void __exit iscsi_sw_tcp_exit(void)
1044 {
1045 	iscsi_unregister_transport(&iscsi_sw_tcp_transport);
1046 }
1047 
1048 module_init(iscsi_sw_tcp_init);
1049 module_exit(iscsi_sw_tcp_exit);
1050