xref: /openbmc/linux/drivers/scsi/iscsi_tcp.c (revision e1cd7b80)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * iSCSI Initiator over TCP/IP Data-Path
4  *
5  * Copyright (C) 2004 Dmitry Yusupov
6  * Copyright (C) 2004 Alex Aizman
7  * Copyright (C) 2005 - 2006 Mike Christie
8  * Copyright (C) 2006 Red Hat, Inc.  All rights reserved.
9  * maintained by open-iscsi@googlegroups.com
10  *
11  * See the file COPYING included with this distribution for more details.
12  *
13  * Credits:
14  *	Christoph Hellwig
15  *	FUJITA Tomonori
16  *	Arne Redlich
17  *	Zhenyu Wang
18  */
19 
20 #include <crypto/hash.h>
21 #include <linux/types.h>
22 #include <linux/inet.h>
23 #include <linux/slab.h>
24 #include <linux/sched/mm.h>
25 #include <linux/file.h>
26 #include <linux/blkdev.h>
27 #include <linux/delay.h>
28 #include <linux/kfifo.h>
29 #include <linux/scatterlist.h>
30 #include <linux/module.h>
31 #include <linux/backing-dev.h>
32 #include <net/tcp.h>
33 #include <scsi/scsi_cmnd.h>
34 #include <scsi/scsi_device.h>
35 #include <scsi/scsi_host.h>
36 #include <scsi/scsi.h>
37 #include <scsi/scsi_transport_iscsi.h>
38 #include <trace/events/iscsi.h>
39 
40 #include "iscsi_tcp.h"
41 
42 MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
43 	      "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
44 	      "Alex Aizman <itn780@yahoo.com>");
45 MODULE_DESCRIPTION("iSCSI/TCP data-path");
46 MODULE_LICENSE("GPL");
47 
48 static struct scsi_transport_template *iscsi_sw_tcp_scsi_transport;
49 static struct scsi_host_template iscsi_sw_tcp_sht;
50 static struct iscsi_transport iscsi_sw_tcp_transport;
51 
52 static unsigned int iscsi_max_lun = ~0;
53 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
54 
55 static int iscsi_sw_tcp_dbg;
56 module_param_named(debug_iscsi_tcp, iscsi_sw_tcp_dbg, int,
57 		   S_IRUGO | S_IWUSR);
58 MODULE_PARM_DESC(debug_iscsi_tcp, "Turn on debugging for iscsi_tcp module "
59 		 "Set to 1 to turn on, and zero to turn off. Default is off.");
60 
61 #define ISCSI_SW_TCP_DBG(_conn, dbg_fmt, arg...)		\
62 	do {							\
63 		if (iscsi_sw_tcp_dbg)				\
64 			iscsi_conn_printk(KERN_INFO, _conn,	\
65 					     "%s " dbg_fmt,	\
66 					     __func__, ##arg);	\
67 		iscsi_dbg_trace(trace_iscsi_dbg_sw_tcp,		\
68 				&(_conn)->cls_conn->dev,	\
69 				"%s " dbg_fmt, __func__, ##arg);\
70 	} while (0);
71 
72 
73 /**
74  * iscsi_sw_tcp_recv - TCP receive in sendfile fashion
75  * @rd_desc: read descriptor
76  * @skb: socket buffer
77  * @offset: offset in skb
78  * @len: skb->len - offset
79  */
80 static int iscsi_sw_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
81 			     unsigned int offset, size_t len)
82 {
83 	struct iscsi_conn *conn = rd_desc->arg.data;
84 	unsigned int consumed, total_consumed = 0;
85 	int status;
86 
87 	ISCSI_SW_TCP_DBG(conn, "in %d bytes\n", skb->len - offset);
88 
89 	do {
90 		status = 0;
91 		consumed = iscsi_tcp_recv_skb(conn, skb, offset, 0, &status);
92 		offset += consumed;
93 		total_consumed += consumed;
94 	} while (consumed != 0 && status != ISCSI_TCP_SKB_DONE);
95 
96 	ISCSI_SW_TCP_DBG(conn, "read %d bytes status %d\n",
97 			 skb->len - offset, status);
98 	return total_consumed;
99 }
100 
101 /**
102  * iscsi_sw_sk_state_check - check socket state
103  * @sk: socket
104  *
105  * If the socket is in CLOSE or CLOSE_WAIT we should
106  * not close the connection if there is still some
107  * data pending.
108  *
109  * Must be called with sk_callback_lock.
110  */
111 static inline int iscsi_sw_sk_state_check(struct sock *sk)
112 {
113 	struct iscsi_conn *conn = sk->sk_user_data;
114 
115 	if ((sk->sk_state == TCP_CLOSE_WAIT || sk->sk_state == TCP_CLOSE) &&
116 	    (conn->session->state != ISCSI_STATE_LOGGING_OUT) &&
117 	    !atomic_read(&sk->sk_rmem_alloc)) {
118 		ISCSI_SW_TCP_DBG(conn, "TCP_CLOSE|TCP_CLOSE_WAIT\n");
119 		iscsi_conn_failure(conn, ISCSI_ERR_TCP_CONN_CLOSE);
120 		return -ECONNRESET;
121 	}
122 	return 0;
123 }
124 
125 static void iscsi_sw_tcp_data_ready(struct sock *sk)
126 {
127 	struct iscsi_conn *conn;
128 	struct iscsi_tcp_conn *tcp_conn;
129 	read_descriptor_t rd_desc;
130 
131 	read_lock_bh(&sk->sk_callback_lock);
132 	conn = sk->sk_user_data;
133 	if (!conn) {
134 		read_unlock_bh(&sk->sk_callback_lock);
135 		return;
136 	}
137 	tcp_conn = conn->dd_data;
138 
139 	/*
140 	 * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
141 	 * We set count to 1 because we want the network layer to
142 	 * hand us all the skbs that are available. iscsi_tcp_recv
143 	 * handled pdus that cross buffers or pdus that still need data.
144 	 */
145 	rd_desc.arg.data = conn;
146 	rd_desc.count = 1;
147 	tcp_read_sock(sk, &rd_desc, iscsi_sw_tcp_recv);
148 
149 	iscsi_sw_sk_state_check(sk);
150 
151 	/* If we had to (atomically) map a highmem page,
152 	 * unmap it now. */
153 	iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
154 	read_unlock_bh(&sk->sk_callback_lock);
155 }
156 
157 static void iscsi_sw_tcp_state_change(struct sock *sk)
158 {
159 	struct iscsi_tcp_conn *tcp_conn;
160 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
161 	struct iscsi_conn *conn;
162 	void (*old_state_change)(struct sock *);
163 
164 	read_lock_bh(&sk->sk_callback_lock);
165 	conn = sk->sk_user_data;
166 	if (!conn) {
167 		read_unlock_bh(&sk->sk_callback_lock);
168 		return;
169 	}
170 
171 	iscsi_sw_sk_state_check(sk);
172 
173 	tcp_conn = conn->dd_data;
174 	tcp_sw_conn = tcp_conn->dd_data;
175 	old_state_change = tcp_sw_conn->old_state_change;
176 
177 	read_unlock_bh(&sk->sk_callback_lock);
178 
179 	old_state_change(sk);
180 }
181 
182 /**
183  * iscsi_write_space - Called when more output buffer space is available
184  * @sk: socket space is available for
185  **/
186 static void iscsi_sw_tcp_write_space(struct sock *sk)
187 {
188 	struct iscsi_conn *conn;
189 	struct iscsi_tcp_conn *tcp_conn;
190 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
191 	void (*old_write_space)(struct sock *);
192 
193 	read_lock_bh(&sk->sk_callback_lock);
194 	conn = sk->sk_user_data;
195 	if (!conn) {
196 		read_unlock_bh(&sk->sk_callback_lock);
197 		return;
198 	}
199 
200 	tcp_conn = conn->dd_data;
201 	tcp_sw_conn = tcp_conn->dd_data;
202 	old_write_space = tcp_sw_conn->old_write_space;
203 	read_unlock_bh(&sk->sk_callback_lock);
204 
205 	old_write_space(sk);
206 
207 	ISCSI_SW_TCP_DBG(conn, "iscsi_write_space\n");
208 	iscsi_conn_queue_work(conn);
209 }
210 
211 static void iscsi_sw_tcp_conn_set_callbacks(struct iscsi_conn *conn)
212 {
213 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
214 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
215 	struct sock *sk = tcp_sw_conn->sock->sk;
216 
217 	/* assign new callbacks */
218 	write_lock_bh(&sk->sk_callback_lock);
219 	sk->sk_user_data = conn;
220 	tcp_sw_conn->old_data_ready = sk->sk_data_ready;
221 	tcp_sw_conn->old_state_change = sk->sk_state_change;
222 	tcp_sw_conn->old_write_space = sk->sk_write_space;
223 	sk->sk_data_ready = iscsi_sw_tcp_data_ready;
224 	sk->sk_state_change = iscsi_sw_tcp_state_change;
225 	sk->sk_write_space = iscsi_sw_tcp_write_space;
226 	write_unlock_bh(&sk->sk_callback_lock);
227 }
228 
229 static void
230 iscsi_sw_tcp_conn_restore_callbacks(struct iscsi_conn *conn)
231 {
232 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
233 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
234 	struct sock *sk = tcp_sw_conn->sock->sk;
235 
236 	/* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
237 	write_lock_bh(&sk->sk_callback_lock);
238 	sk->sk_user_data    = NULL;
239 	sk->sk_data_ready   = tcp_sw_conn->old_data_ready;
240 	sk->sk_state_change = tcp_sw_conn->old_state_change;
241 	sk->sk_write_space  = tcp_sw_conn->old_write_space;
242 	sk->sk_no_check_tx = 0;
243 	write_unlock_bh(&sk->sk_callback_lock);
244 }
245 
246 /**
247  * iscsi_sw_tcp_xmit_segment - transmit segment
248  * @tcp_conn: the iSCSI TCP connection
249  * @segment: the buffer to transmnit
250  *
251  * This function transmits as much of the buffer as
252  * the network layer will accept, and returns the number of
253  * bytes transmitted.
254  *
255  * If CRC hashing is enabled, the function will compute the
256  * hash as it goes. When the entire segment has been transmitted,
257  * it will retrieve the hash value and send it as well.
258  */
259 static int iscsi_sw_tcp_xmit_segment(struct iscsi_tcp_conn *tcp_conn,
260 				     struct iscsi_segment *segment)
261 {
262 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
263 	struct socket *sk = tcp_sw_conn->sock;
264 	unsigned int copied = 0;
265 	int r = 0;
266 
267 	while (!iscsi_tcp_segment_done(tcp_conn, segment, 0, r)) {
268 		struct scatterlist *sg;
269 		unsigned int offset, copy;
270 		int flags = 0;
271 
272 		r = 0;
273 		offset = segment->copied;
274 		copy = segment->size - offset;
275 
276 		if (segment->total_copied + segment->size < segment->total_size)
277 			flags |= MSG_MORE;
278 
279 		/* Use sendpage if we can; else fall back to sendmsg */
280 		if (!segment->data) {
281 			sg = segment->sg;
282 			offset += segment->sg_offset + sg->offset;
283 			r = tcp_sw_conn->sendpage(sk, sg_page(sg), offset,
284 						  copy, flags);
285 		} else {
286 			struct msghdr msg = { .msg_flags = flags };
287 			struct kvec iov = {
288 				.iov_base = segment->data + offset,
289 				.iov_len = copy
290 			};
291 
292 			r = kernel_sendmsg(sk, &msg, &iov, 1, copy);
293 		}
294 
295 		if (r < 0) {
296 			iscsi_tcp_segment_unmap(segment);
297 			return r;
298 		}
299 		copied += r;
300 	}
301 	return copied;
302 }
303 
304 /**
305  * iscsi_sw_tcp_xmit - TCP transmit
306  * @conn: iscsi connection
307  **/
308 static int iscsi_sw_tcp_xmit(struct iscsi_conn *conn)
309 {
310 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
311 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
312 	struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
313 	unsigned int consumed = 0;
314 	int rc = 0;
315 
316 	while (1) {
317 		rc = iscsi_sw_tcp_xmit_segment(tcp_conn, segment);
318 		/*
319 		 * We may not have been able to send data because the conn
320 		 * is getting stopped. libiscsi will know so propagate err
321 		 * for it to do the right thing.
322 		 */
323 		if (rc == -EAGAIN)
324 			return rc;
325 		else if (rc < 0) {
326 			rc = ISCSI_ERR_XMIT_FAILED;
327 			goto error;
328 		} else if (rc == 0)
329 			break;
330 
331 		consumed += rc;
332 
333 		if (segment->total_copied >= segment->total_size) {
334 			if (segment->done != NULL) {
335 				rc = segment->done(tcp_conn, segment);
336 				if (rc != 0)
337 					goto error;
338 			}
339 		}
340 	}
341 
342 	ISCSI_SW_TCP_DBG(conn, "xmit %d bytes\n", consumed);
343 
344 	conn->txdata_octets += consumed;
345 	return consumed;
346 
347 error:
348 	/* Transmit error. We could initiate error recovery
349 	 * here. */
350 	ISCSI_SW_TCP_DBG(conn, "Error sending PDU, errno=%d\n", rc);
351 	iscsi_conn_failure(conn, rc);
352 	return -EIO;
353 }
354 
355 /**
356  * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
357  * @conn: iscsi connection
358  */
359 static inline int iscsi_sw_tcp_xmit_qlen(struct iscsi_conn *conn)
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 	struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
364 
365 	return segment->total_copied - segment->total_size;
366 }
367 
368 static int iscsi_sw_tcp_pdu_xmit(struct iscsi_task *task)
369 {
370 	struct iscsi_conn *conn = task->conn;
371 	unsigned int noreclaim_flag;
372 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
373 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
374 	int rc = 0;
375 
376 	if (!tcp_sw_conn->sock) {
377 		iscsi_conn_printk(KERN_ERR, conn,
378 				  "Transport not bound to socket!\n");
379 		return -EINVAL;
380 	}
381 
382 	noreclaim_flag = memalloc_noreclaim_save();
383 
384 	while (iscsi_sw_tcp_xmit_qlen(conn)) {
385 		rc = iscsi_sw_tcp_xmit(conn);
386 		if (rc == 0) {
387 			rc = -EAGAIN;
388 			break;
389 		}
390 		if (rc < 0)
391 			break;
392 		rc = 0;
393 	}
394 
395 	memalloc_noreclaim_restore(noreclaim_flag);
396 	return rc;
397 }
398 
399 /*
400  * This is called when we're done sending the header.
401  * Simply copy the data_segment to the send segment, and return.
402  */
403 static int iscsi_sw_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
404 				      struct iscsi_segment *segment)
405 {
406 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
407 
408 	tcp_sw_conn->out.segment = tcp_sw_conn->out.data_segment;
409 	ISCSI_SW_TCP_DBG(tcp_conn->iscsi_conn,
410 			 "Header done. Next segment size %u total_size %u\n",
411 			 tcp_sw_conn->out.segment.size,
412 			 tcp_sw_conn->out.segment.total_size);
413 	return 0;
414 }
415 
416 static void iscsi_sw_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr,
417 				       size_t hdrlen)
418 {
419 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
420 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
421 
422 	ISCSI_SW_TCP_DBG(conn, "%s\n", conn->hdrdgst_en ?
423 			 "digest enabled" : "digest disabled");
424 
425 	/* Clear the data segment - needs to be filled in by the
426 	 * caller using iscsi_tcp_send_data_prep() */
427 	memset(&tcp_sw_conn->out.data_segment, 0,
428 	       sizeof(struct iscsi_segment));
429 
430 	/* If header digest is enabled, compute the CRC and
431 	 * place the digest into the same buffer. We make
432 	 * sure that both iscsi_tcp_task and mtask have
433 	 * sufficient room.
434 	 */
435 	if (conn->hdrdgst_en) {
436 		iscsi_tcp_dgst_header(tcp_sw_conn->tx_hash, hdr, hdrlen,
437 				      hdr + hdrlen);
438 		hdrlen += ISCSI_DIGEST_SIZE;
439 	}
440 
441 	/* Remember header pointer for later, when we need
442 	 * to decide whether there's a payload to go along
443 	 * with the header. */
444 	tcp_sw_conn->out.hdr = hdr;
445 
446 	iscsi_segment_init_linear(&tcp_sw_conn->out.segment, hdr, hdrlen,
447 				  iscsi_sw_tcp_send_hdr_done, NULL);
448 }
449 
450 /*
451  * Prepare the send buffer for the payload data.
452  * Padding and checksumming will all be taken care
453  * of by the iscsi_segment routines.
454  */
455 static int
456 iscsi_sw_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
457 			    unsigned int count, unsigned int offset,
458 			    unsigned int len)
459 {
460 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
461 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
462 	struct ahash_request *tx_hash = NULL;
463 	unsigned int hdr_spec_len;
464 
465 	ISCSI_SW_TCP_DBG(conn, "offset=%d, datalen=%d %s\n", offset, len,
466 			 conn->datadgst_en ?
467 			 "digest enabled" : "digest disabled");
468 
469 	/* Make sure the datalen matches what the caller
470 	   said he would send. */
471 	hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
472 	WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
473 
474 	if (conn->datadgst_en)
475 		tx_hash = tcp_sw_conn->tx_hash;
476 
477 	return iscsi_segment_seek_sg(&tcp_sw_conn->out.data_segment,
478 				     sg, count, offset, len,
479 				     NULL, tx_hash);
480 }
481 
482 static void
483 iscsi_sw_tcp_send_linear_data_prep(struct iscsi_conn *conn, void *data,
484 				   size_t len)
485 {
486 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
487 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
488 	struct ahash_request *tx_hash = NULL;
489 	unsigned int hdr_spec_len;
490 
491 	ISCSI_SW_TCP_DBG(conn, "datalen=%zd %s\n", len, conn->datadgst_en ?
492 			 "digest enabled" : "digest disabled");
493 
494 	/* Make sure the datalen matches what the caller
495 	   said he would send. */
496 	hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
497 	WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
498 
499 	if (conn->datadgst_en)
500 		tx_hash = tcp_sw_conn->tx_hash;
501 
502 	iscsi_segment_init_linear(&tcp_sw_conn->out.data_segment,
503 				data, len, NULL, tx_hash);
504 }
505 
506 static int iscsi_sw_tcp_pdu_init(struct iscsi_task *task,
507 				 unsigned int offset, unsigned int count)
508 {
509 	struct iscsi_conn *conn = task->conn;
510 	int err = 0;
511 
512 	iscsi_sw_tcp_send_hdr_prep(conn, task->hdr, task->hdr_len);
513 
514 	if (!count)
515 		return 0;
516 
517 	if (!task->sc)
518 		iscsi_sw_tcp_send_linear_data_prep(conn, task->data, count);
519 	else {
520 		struct scsi_data_buffer *sdb = &task->sc->sdb;
521 
522 		err = iscsi_sw_tcp_send_data_prep(conn, sdb->table.sgl,
523 						  sdb->table.nents, offset,
524 						  count);
525 	}
526 
527 	if (err) {
528 		/* got invalid offset/len */
529 		return -EIO;
530 	}
531 	return 0;
532 }
533 
534 static int iscsi_sw_tcp_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
535 {
536 	struct iscsi_tcp_task *tcp_task = task->dd_data;
537 
538 	task->hdr = task->dd_data + sizeof(*tcp_task);
539 	task->hdr_max = sizeof(struct iscsi_sw_tcp_hdrbuf) - ISCSI_DIGEST_SIZE;
540 	return 0;
541 }
542 
543 static struct iscsi_cls_conn *
544 iscsi_sw_tcp_conn_create(struct iscsi_cls_session *cls_session,
545 			 uint32_t conn_idx)
546 {
547 	struct iscsi_conn *conn;
548 	struct iscsi_cls_conn *cls_conn;
549 	struct iscsi_tcp_conn *tcp_conn;
550 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
551 	struct crypto_ahash *tfm;
552 
553 	cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*tcp_sw_conn),
554 					conn_idx);
555 	if (!cls_conn)
556 		return NULL;
557 	conn = cls_conn->dd_data;
558 	tcp_conn = conn->dd_data;
559 	tcp_sw_conn = tcp_conn->dd_data;
560 
561 	tfm = crypto_alloc_ahash("crc32c", 0, CRYPTO_ALG_ASYNC);
562 	if (IS_ERR(tfm))
563 		goto free_conn;
564 
565 	tcp_sw_conn->tx_hash = ahash_request_alloc(tfm, GFP_KERNEL);
566 	if (!tcp_sw_conn->tx_hash)
567 		goto free_tfm;
568 	ahash_request_set_callback(tcp_sw_conn->tx_hash, 0, NULL, NULL);
569 
570 	tcp_sw_conn->rx_hash = ahash_request_alloc(tfm, GFP_KERNEL);
571 	if (!tcp_sw_conn->rx_hash)
572 		goto free_tx_hash;
573 	ahash_request_set_callback(tcp_sw_conn->rx_hash, 0, NULL, NULL);
574 
575 	tcp_conn->rx_hash = tcp_sw_conn->rx_hash;
576 
577 	return cls_conn;
578 
579 free_tx_hash:
580 	ahash_request_free(tcp_sw_conn->tx_hash);
581 free_tfm:
582 	crypto_free_ahash(tfm);
583 free_conn:
584 	iscsi_conn_printk(KERN_ERR, conn,
585 			  "Could not create connection due to crc32c "
586 			  "loading error. Make sure the crc32c "
587 			  "module is built as a module or into the "
588 			  "kernel\n");
589 	iscsi_tcp_conn_teardown(cls_conn);
590 	return NULL;
591 }
592 
593 static void iscsi_sw_tcp_release_conn(struct iscsi_conn *conn)
594 {
595 	struct iscsi_session *session = conn->session;
596 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
597 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
598 	struct socket *sock = tcp_sw_conn->sock;
599 
600 	if (!sock)
601 		return;
602 
603 	sock_hold(sock->sk);
604 	iscsi_sw_tcp_conn_restore_callbacks(conn);
605 	sock_put(sock->sk);
606 
607 	spin_lock_bh(&session->frwd_lock);
608 	tcp_sw_conn->sock = NULL;
609 	spin_unlock_bh(&session->frwd_lock);
610 	sockfd_put(sock);
611 }
612 
613 static void iscsi_sw_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
614 {
615 	struct iscsi_conn *conn = cls_conn->dd_data;
616 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
617 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
618 
619 	iscsi_sw_tcp_release_conn(conn);
620 
621 	ahash_request_free(tcp_sw_conn->rx_hash);
622 	if (tcp_sw_conn->tx_hash) {
623 		struct crypto_ahash *tfm;
624 
625 		tfm = crypto_ahash_reqtfm(tcp_sw_conn->tx_hash);
626 		ahash_request_free(tcp_sw_conn->tx_hash);
627 		crypto_free_ahash(tfm);
628 	}
629 
630 	iscsi_tcp_conn_teardown(cls_conn);
631 }
632 
633 static void iscsi_sw_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
634 {
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 socket *sock = tcp_sw_conn->sock;
639 
640 	/* userspace may have goofed up and not bound us */
641 	if (!sock)
642 		return;
643 
644 	sock->sk->sk_err = EIO;
645 	wake_up_interruptible(sk_sleep(sock->sk));
646 
647 	/* stop xmit side */
648 	iscsi_suspend_tx(conn);
649 
650 	/* stop recv side and release socket */
651 	iscsi_sw_tcp_release_conn(conn);
652 
653 	iscsi_conn_stop(cls_conn, flag);
654 }
655 
656 static int
657 iscsi_sw_tcp_conn_bind(struct iscsi_cls_session *cls_session,
658 		       struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
659 		       int is_leading)
660 {
661 	struct iscsi_session *session = cls_session->dd_data;
662 	struct iscsi_conn *conn = cls_conn->dd_data;
663 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
664 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
665 	struct sock *sk;
666 	struct socket *sock;
667 	int err;
668 
669 	/* lookup for existing socket */
670 	sock = sockfd_lookup((int)transport_eph, &err);
671 	if (!sock) {
672 		iscsi_conn_printk(KERN_ERR, conn,
673 				  "sockfd_lookup failed %d\n", err);
674 		return -EEXIST;
675 	}
676 
677 	err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
678 	if (err)
679 		goto free_socket;
680 
681 	spin_lock_bh(&session->frwd_lock);
682 	/* bind iSCSI connection and socket */
683 	tcp_sw_conn->sock = sock;
684 	spin_unlock_bh(&session->frwd_lock);
685 
686 	/* setup Socket parameters */
687 	sk = sock->sk;
688 	sk->sk_reuse = SK_CAN_REUSE;
689 	sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
690 	sk->sk_allocation = GFP_ATOMIC;
691 	sk_set_memalloc(sk);
692 
693 	iscsi_sw_tcp_conn_set_callbacks(conn);
694 	tcp_sw_conn->sendpage = tcp_sw_conn->sock->ops->sendpage;
695 	/*
696 	 * set receive state machine into initial state
697 	 */
698 	iscsi_tcp_hdr_recv_prep(tcp_conn);
699 	return 0;
700 
701 free_socket:
702 	sockfd_put(sock);
703 	return err;
704 }
705 
706 static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn,
707 				       enum iscsi_param param, char *buf,
708 				       int buflen)
709 {
710 	struct iscsi_conn *conn = cls_conn->dd_data;
711 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
712 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
713 
714 	switch(param) {
715 	case ISCSI_PARAM_HDRDGST_EN:
716 		iscsi_set_param(cls_conn, param, buf, buflen);
717 		break;
718 	case ISCSI_PARAM_DATADGST_EN:
719 		iscsi_set_param(cls_conn, param, buf, buflen);
720 		tcp_sw_conn->sendpage = conn->datadgst_en ?
721 			sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage;
722 		break;
723 	case ISCSI_PARAM_MAX_R2T:
724 		return iscsi_tcp_set_max_r2t(conn, buf);
725 	default:
726 		return iscsi_set_param(cls_conn, param, buf, buflen);
727 	}
728 
729 	return 0;
730 }
731 
732 static int iscsi_sw_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
733 				       enum iscsi_param param, char *buf)
734 {
735 	struct iscsi_conn *conn = cls_conn->dd_data;
736 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
737 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
738 	struct sockaddr_in6 addr;
739 	int rc;
740 
741 	switch(param) {
742 	case ISCSI_PARAM_CONN_PORT:
743 	case ISCSI_PARAM_CONN_ADDRESS:
744 	case ISCSI_PARAM_LOCAL_PORT:
745 		spin_lock_bh(&conn->session->frwd_lock);
746 		if (!tcp_sw_conn || !tcp_sw_conn->sock) {
747 			spin_unlock_bh(&conn->session->frwd_lock);
748 			return -ENOTCONN;
749 		}
750 		if (param == ISCSI_PARAM_LOCAL_PORT)
751 			rc = kernel_getsockname(tcp_sw_conn->sock,
752 						(struct sockaddr *)&addr);
753 		else
754 			rc = kernel_getpeername(tcp_sw_conn->sock,
755 						(struct sockaddr *)&addr);
756 		spin_unlock_bh(&conn->session->frwd_lock);
757 		if (rc < 0)
758 			return rc;
759 
760 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
761 						 &addr, param, buf);
762 	default:
763 		return iscsi_conn_get_param(cls_conn, param, buf);
764 	}
765 
766 	return 0;
767 }
768 
769 static int iscsi_sw_tcp_host_get_param(struct Scsi_Host *shost,
770 				       enum iscsi_host_param param, char *buf)
771 {
772 	struct iscsi_sw_tcp_host *tcp_sw_host = iscsi_host_priv(shost);
773 	struct iscsi_session *session = tcp_sw_host->session;
774 	struct iscsi_conn *conn;
775 	struct iscsi_tcp_conn *tcp_conn;
776 	struct iscsi_sw_tcp_conn *tcp_sw_conn;
777 	struct sockaddr_in6 addr;
778 	int rc;
779 
780 	switch (param) {
781 	case ISCSI_HOST_PARAM_IPADDRESS:
782 		if (!session)
783 			return -ENOTCONN;
784 
785 		spin_lock_bh(&session->frwd_lock);
786 		conn = session->leadconn;
787 		if (!conn) {
788 			spin_unlock_bh(&session->frwd_lock);
789 			return -ENOTCONN;
790 		}
791 		tcp_conn = conn->dd_data;
792 
793 		tcp_sw_conn = tcp_conn->dd_data;
794 		if (!tcp_sw_conn->sock) {
795 			spin_unlock_bh(&session->frwd_lock);
796 			return -ENOTCONN;
797 		}
798 
799 		rc = kernel_getsockname(tcp_sw_conn->sock,
800 					(struct sockaddr *)&addr);
801 		spin_unlock_bh(&session->frwd_lock);
802 		if (rc < 0)
803 			return rc;
804 
805 		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
806 						 &addr,
807 						 (enum iscsi_param)param, buf);
808 	default:
809 		return iscsi_host_get_param(shost, param, buf);
810 	}
811 
812 	return 0;
813 }
814 
815 static void
816 iscsi_sw_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
817 			    struct iscsi_stats *stats)
818 {
819 	struct iscsi_conn *conn = cls_conn->dd_data;
820 	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
821 	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
822 
823 	stats->custom_length = 3;
824 	strcpy(stats->custom[0].desc, "tx_sendpage_failures");
825 	stats->custom[0].value = tcp_sw_conn->sendpage_failures_cnt;
826 	strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
827 	stats->custom[1].value = tcp_sw_conn->discontiguous_hdr_cnt;
828 	strcpy(stats->custom[2].desc, "eh_abort_cnt");
829 	stats->custom[2].value = conn->eh_abort_cnt;
830 
831 	iscsi_tcp_conn_get_stats(cls_conn, stats);
832 }
833 
834 static struct iscsi_cls_session *
835 iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
836 			    uint16_t qdepth, uint32_t initial_cmdsn)
837 {
838 	struct iscsi_cls_session *cls_session;
839 	struct iscsi_session *session;
840 	struct iscsi_sw_tcp_host *tcp_sw_host;
841 	struct Scsi_Host *shost;
842 
843 	if (ep) {
844 		printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
845 		return NULL;
846 	}
847 
848 	shost = iscsi_host_alloc(&iscsi_sw_tcp_sht,
849 				 sizeof(struct iscsi_sw_tcp_host), 1);
850 	if (!shost)
851 		return NULL;
852 	shost->transportt = iscsi_sw_tcp_scsi_transport;
853 	shost->cmd_per_lun = qdepth;
854 	shost->max_lun = iscsi_max_lun;
855 	shost->max_id = 0;
856 	shost->max_channel = 0;
857 	shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE;
858 
859 	if (iscsi_host_add(shost, NULL))
860 		goto free_host;
861 
862 	cls_session = iscsi_session_setup(&iscsi_sw_tcp_transport, shost,
863 					  cmds_max, 0,
864 					  sizeof(struct iscsi_tcp_task) +
865 					  sizeof(struct iscsi_sw_tcp_hdrbuf),
866 					  initial_cmdsn, 0);
867 	if (!cls_session)
868 		goto remove_host;
869 	session = cls_session->dd_data;
870 	tcp_sw_host = iscsi_host_priv(shost);
871 	tcp_sw_host->session = session;
872 
873 	shost->can_queue = session->scsi_cmds_max;
874 	if (iscsi_tcp_r2tpool_alloc(session))
875 		goto remove_session;
876 	return cls_session;
877 
878 remove_session:
879 	iscsi_session_teardown(cls_session);
880 remove_host:
881 	iscsi_host_remove(shost);
882 free_host:
883 	iscsi_host_free(shost);
884 	return NULL;
885 }
886 
887 static void iscsi_sw_tcp_session_destroy(struct iscsi_cls_session *cls_session)
888 {
889 	struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
890 
891 	iscsi_tcp_r2tpool_free(cls_session->dd_data);
892 	iscsi_session_teardown(cls_session);
893 
894 	iscsi_host_remove(shost);
895 	iscsi_host_free(shost);
896 }
897 
898 static umode_t iscsi_sw_tcp_attr_is_visible(int param_type, int param)
899 {
900 	switch (param_type) {
901 	case ISCSI_HOST_PARAM:
902 		switch (param) {
903 		case ISCSI_HOST_PARAM_NETDEV_NAME:
904 		case ISCSI_HOST_PARAM_HWADDRESS:
905 		case ISCSI_HOST_PARAM_IPADDRESS:
906 		case ISCSI_HOST_PARAM_INITIATOR_NAME:
907 			return S_IRUGO;
908 		default:
909 			return 0;
910 		}
911 	case ISCSI_PARAM:
912 		switch (param) {
913 		case ISCSI_PARAM_MAX_RECV_DLENGTH:
914 		case ISCSI_PARAM_MAX_XMIT_DLENGTH:
915 		case ISCSI_PARAM_HDRDGST_EN:
916 		case ISCSI_PARAM_DATADGST_EN:
917 		case ISCSI_PARAM_CONN_ADDRESS:
918 		case ISCSI_PARAM_CONN_PORT:
919 		case ISCSI_PARAM_LOCAL_PORT:
920 		case ISCSI_PARAM_EXP_STATSN:
921 		case ISCSI_PARAM_PERSISTENT_ADDRESS:
922 		case ISCSI_PARAM_PERSISTENT_PORT:
923 		case ISCSI_PARAM_PING_TMO:
924 		case ISCSI_PARAM_RECV_TMO:
925 		case ISCSI_PARAM_INITIAL_R2T_EN:
926 		case ISCSI_PARAM_MAX_R2T:
927 		case ISCSI_PARAM_IMM_DATA_EN:
928 		case ISCSI_PARAM_FIRST_BURST:
929 		case ISCSI_PARAM_MAX_BURST:
930 		case ISCSI_PARAM_PDU_INORDER_EN:
931 		case ISCSI_PARAM_DATASEQ_INORDER_EN:
932 		case ISCSI_PARAM_ERL:
933 		case ISCSI_PARAM_TARGET_NAME:
934 		case ISCSI_PARAM_TPGT:
935 		case ISCSI_PARAM_USERNAME:
936 		case ISCSI_PARAM_PASSWORD:
937 		case ISCSI_PARAM_USERNAME_IN:
938 		case ISCSI_PARAM_PASSWORD_IN:
939 		case ISCSI_PARAM_FAST_ABORT:
940 		case ISCSI_PARAM_ABORT_TMO:
941 		case ISCSI_PARAM_LU_RESET_TMO:
942 		case ISCSI_PARAM_TGT_RESET_TMO:
943 		case ISCSI_PARAM_IFACE_NAME:
944 		case ISCSI_PARAM_INITIATOR_NAME:
945 			return S_IRUGO;
946 		default:
947 			return 0;
948 		}
949 	}
950 
951 	return 0;
952 }
953 
954 static int iscsi_sw_tcp_slave_configure(struct scsi_device *sdev)
955 {
956 	struct iscsi_sw_tcp_host *tcp_sw_host = iscsi_host_priv(sdev->host);
957 	struct iscsi_session *session = tcp_sw_host->session;
958 	struct iscsi_conn *conn = session->leadconn;
959 
960 	if (conn->datadgst_en)
961 		sdev->request_queue->backing_dev_info->capabilities
962 			|= BDI_CAP_STABLE_WRITES;
963 	blk_queue_dma_alignment(sdev->request_queue, 0);
964 	return 0;
965 }
966 
967 static struct scsi_host_template iscsi_sw_tcp_sht = {
968 	.module			= THIS_MODULE,
969 	.name			= "iSCSI Initiator over TCP/IP",
970 	.queuecommand           = iscsi_queuecommand,
971 	.change_queue_depth	= scsi_change_queue_depth,
972 	.can_queue		= ISCSI_DEF_XMIT_CMDS_MAX - 1,
973 	.sg_tablesize		= 4096,
974 	.max_sectors		= 0xFFFF,
975 	.cmd_per_lun		= ISCSI_DEF_CMD_PER_LUN,
976 	.eh_timed_out		= iscsi_eh_cmd_timed_out,
977 	.eh_abort_handler       = iscsi_eh_abort,
978 	.eh_device_reset_handler= iscsi_eh_device_reset,
979 	.eh_target_reset_handler = iscsi_eh_recover_target,
980 	.dma_boundary		= PAGE_SIZE - 1,
981 	.slave_configure        = iscsi_sw_tcp_slave_configure,
982 	.target_alloc		= iscsi_target_alloc,
983 	.proc_name		= "iscsi_tcp",
984 	.this_id		= -1,
985 	.track_queue_depth	= 1,
986 };
987 
988 static struct iscsi_transport iscsi_sw_tcp_transport = {
989 	.owner			= THIS_MODULE,
990 	.name			= "tcp",
991 	.caps			= CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
992 				  | CAP_DATADGST,
993 	/* session management */
994 	.create_session		= iscsi_sw_tcp_session_create,
995 	.destroy_session	= iscsi_sw_tcp_session_destroy,
996 	/* connection management */
997 	.create_conn		= iscsi_sw_tcp_conn_create,
998 	.bind_conn		= iscsi_sw_tcp_conn_bind,
999 	.destroy_conn		= iscsi_sw_tcp_conn_destroy,
1000 	.attr_is_visible	= iscsi_sw_tcp_attr_is_visible,
1001 	.set_param		= iscsi_sw_tcp_conn_set_param,
1002 	.get_conn_param		= iscsi_sw_tcp_conn_get_param,
1003 	.get_session_param	= iscsi_session_get_param,
1004 	.start_conn		= iscsi_conn_start,
1005 	.stop_conn		= iscsi_sw_tcp_conn_stop,
1006 	/* iscsi host params */
1007 	.get_host_param		= iscsi_sw_tcp_host_get_param,
1008 	.set_host_param		= iscsi_host_set_param,
1009 	/* IO */
1010 	.send_pdu		= iscsi_conn_send_pdu,
1011 	.get_stats		= iscsi_sw_tcp_conn_get_stats,
1012 	/* iscsi task/cmd helpers */
1013 	.init_task		= iscsi_tcp_task_init,
1014 	.xmit_task		= iscsi_tcp_task_xmit,
1015 	.cleanup_task		= iscsi_tcp_cleanup_task,
1016 	/* low level pdu helpers */
1017 	.xmit_pdu		= iscsi_sw_tcp_pdu_xmit,
1018 	.init_pdu		= iscsi_sw_tcp_pdu_init,
1019 	.alloc_pdu		= iscsi_sw_tcp_pdu_alloc,
1020 	/* recovery */
1021 	.session_recovery_timedout = iscsi_session_recovery_timedout,
1022 };
1023 
1024 static int __init iscsi_sw_tcp_init(void)
1025 {
1026 	if (iscsi_max_lun < 1) {
1027 		printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
1028 		       iscsi_max_lun);
1029 		return -EINVAL;
1030 	}
1031 
1032 	iscsi_sw_tcp_scsi_transport = iscsi_register_transport(
1033 						&iscsi_sw_tcp_transport);
1034 	if (!iscsi_sw_tcp_scsi_transport)
1035 		return -ENODEV;
1036 
1037 	return 0;
1038 }
1039 
1040 static void __exit iscsi_sw_tcp_exit(void)
1041 {
1042 	iscsi_unregister_transport(&iscsi_sw_tcp_transport);
1043 }
1044 
1045 module_init(iscsi_sw_tcp_init);
1046 module_exit(iscsi_sw_tcp_exit);
1047