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/file.h>
  34#include <linux/blkdev.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
  48MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
  49	      "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
  50	      "Alex Aizman <itn780@yahoo.com>");
  51MODULE_DESCRIPTION("iSCSI/TCP data-path");
  52MODULE_LICENSE("GPL");
  53
  54static struct scsi_transport_template *iscsi_sw_tcp_scsi_transport;
  55static struct scsi_host_template iscsi_sw_tcp_sht;
  56static struct iscsi_transport iscsi_sw_tcp_transport;
  57
  58static unsigned int iscsi_max_lun = ~0;
  59module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
  60
  61static int iscsi_sw_tcp_dbg;
  62module_param_named(debug_iscsi_tcp, iscsi_sw_tcp_dbg, int,
  63		   S_IRUGO | S_IWUSR);
  64MODULE_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 */
  83static 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 */
 114static 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
 128static 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_bh(&sk->sk_callback_lock);
 135	conn = sk->sk_user_data;
 136	if (!conn) {
 137		read_unlock_bh(&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_bh(&sk->sk_callback_lock);
 158}
 159
 160static 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_bh(&sk->sk_callback_lock);
 169	conn = sk->sk_user_data;
 170	if (!conn) {
 171		read_unlock_bh(&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_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 **/
 191static 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
 216static 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
 234static void
 235iscsi_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 */
 264static 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 **/
 312static 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
 351error:
 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 */
 362static 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
 371static 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 */
 398static 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
 411static 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 */
 450static int
 451iscsi_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 ahash_request *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
 477static void
 478iscsi_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 ahash_request *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
 501static 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
 529static 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
 538static struct iscsi_cls_conn *
 539iscsi_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	struct crypto_ahash *tfm;
 547
 548	cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*tcp_sw_conn),
 549					conn_idx);
 550	if (!cls_conn)
 551		return NULL;
 552	conn = cls_conn->dd_data;
 553	tcp_conn = conn->dd_data;
 554	tcp_sw_conn = tcp_conn->dd_data;
 555
 556	tfm = crypto_alloc_ahash("crc32c", 0, CRYPTO_ALG_ASYNC);
 557	if (IS_ERR(tfm))
 
 
 558		goto free_conn;
 559
 560	tcp_sw_conn->tx_hash = ahash_request_alloc(tfm, GFP_KERNEL);
 561	if (!tcp_sw_conn->tx_hash)
 562		goto free_tfm;
 563	ahash_request_set_callback(tcp_sw_conn->tx_hash, 0, NULL, NULL);
 564
 565	tcp_sw_conn->rx_hash = ahash_request_alloc(tfm, GFP_KERNEL);
 566	if (!tcp_sw_conn->rx_hash)
 567		goto free_tx_hash;
 568	ahash_request_set_callback(tcp_sw_conn->rx_hash, 0, NULL, NULL);
 569
 570	tcp_conn->rx_hash = tcp_sw_conn->rx_hash;
 571
 572	return cls_conn;
 573
 574free_tx_hash:
 575	ahash_request_free(tcp_sw_conn->tx_hash);
 576free_tfm:
 577	crypto_free_ahash(tfm);
 578free_conn:
 579	iscsi_conn_printk(KERN_ERR, conn,
 580			  "Could not create connection due to crc32c "
 581			  "loading error. Make sure the crc32c "
 582			  "module is built as a module or into the "
 583			  "kernel\n");
 584	iscsi_tcp_conn_teardown(cls_conn);
 585	return NULL;
 586}
 587
 588static void iscsi_sw_tcp_release_conn(struct iscsi_conn *conn)
 589{
 590	struct iscsi_session *session = conn->session;
 591	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 592	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 593	struct socket *sock = tcp_sw_conn->sock;
 594
 595	if (!sock)
 596		return;
 597
 598	sock_hold(sock->sk);
 599	iscsi_sw_tcp_conn_restore_callbacks(conn);
 600	sock_put(sock->sk);
 601
 602	spin_lock_bh(&session->frwd_lock);
 603	tcp_sw_conn->sock = NULL;
 604	spin_unlock_bh(&session->frwd_lock);
 605	sockfd_put(sock);
 606}
 607
 608static void iscsi_sw_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
 609{
 610	struct iscsi_conn *conn = cls_conn->dd_data;
 611	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 612	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 613
 614	iscsi_sw_tcp_release_conn(conn);
 615
 616	ahash_request_free(tcp_sw_conn->rx_hash);
 617	if (tcp_sw_conn->tx_hash) {
 618		struct crypto_ahash *tfm;
 619
 620		tfm = crypto_ahash_reqtfm(tcp_sw_conn->tx_hash);
 621		ahash_request_free(tcp_sw_conn->tx_hash);
 622		crypto_free_ahash(tfm);
 623	}
 624
 625	iscsi_tcp_conn_teardown(cls_conn);
 626}
 627
 628static void iscsi_sw_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
 629{
 630	struct iscsi_conn *conn = cls_conn->dd_data;
 631	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 632	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 633	struct socket *sock = tcp_sw_conn->sock;
 634
 635	/* userspace may have goofed up and not bound us */
 636	if (!sock)
 637		return;
 638
 639	sock->sk->sk_err = EIO;
 640	wake_up_interruptible(sk_sleep(sock->sk));
 641
 642	/* stop xmit side */
 643	iscsi_suspend_tx(conn);
 644
 645	/* stop recv side and release socket */
 646	iscsi_sw_tcp_release_conn(conn);
 647
 648	iscsi_conn_stop(cls_conn, flag);
 649}
 650
 651static int
 652iscsi_sw_tcp_conn_bind(struct iscsi_cls_session *cls_session,
 653		       struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
 654		       int is_leading)
 655{
 656	struct iscsi_session *session = cls_session->dd_data;
 657	struct iscsi_conn *conn = cls_conn->dd_data;
 658	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 659	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 660	struct sock *sk;
 661	struct socket *sock;
 662	int err;
 663
 664	/* lookup for existing socket */
 665	sock = sockfd_lookup((int)transport_eph, &err);
 666	if (!sock) {
 667		iscsi_conn_printk(KERN_ERR, conn,
 668				  "sockfd_lookup failed %d\n", err);
 669		return -EEXIST;
 670	}
 671
 672	err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
 673	if (err)
 674		goto free_socket;
 675
 676	spin_lock_bh(&session->frwd_lock);
 677	/* bind iSCSI connection and socket */
 678	tcp_sw_conn->sock = sock;
 679	spin_unlock_bh(&session->frwd_lock);
 680
 681	/* setup Socket parameters */
 682	sk = sock->sk;
 683	sk->sk_reuse = SK_CAN_REUSE;
 684	sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
 685	sk->sk_allocation = GFP_ATOMIC;
 686	sk_set_memalloc(sk);
 687
 688	iscsi_sw_tcp_conn_set_callbacks(conn);
 689	tcp_sw_conn->sendpage = tcp_sw_conn->sock->ops->sendpage;
 690	/*
 691	 * set receive state machine into initial state
 692	 */
 693	iscsi_tcp_hdr_recv_prep(tcp_conn);
 694	return 0;
 695
 696free_socket:
 697	sockfd_put(sock);
 698	return err;
 699}
 700
 701static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn,
 702				       enum iscsi_param param, char *buf,
 703				       int buflen)
 704{
 705	struct iscsi_conn *conn = cls_conn->dd_data;
 706	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 707	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 708
 709	switch(param) {
 710	case ISCSI_PARAM_HDRDGST_EN:
 711		iscsi_set_param(cls_conn, param, buf, buflen);
 712		break;
 713	case ISCSI_PARAM_DATADGST_EN:
 714		iscsi_set_param(cls_conn, param, buf, buflen);
 715		tcp_sw_conn->sendpage = conn->datadgst_en ?
 716			sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage;
 717		break;
 718	case ISCSI_PARAM_MAX_R2T:
 719		return iscsi_tcp_set_max_r2t(conn, buf);
 720	default:
 721		return iscsi_set_param(cls_conn, param, buf, buflen);
 722	}
 723
 724	return 0;
 725}
 726
 727static int iscsi_sw_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
 728				       enum iscsi_param param, char *buf)
 729{
 730	struct iscsi_conn *conn = cls_conn->dd_data;
 731	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 732	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 733	struct sockaddr_in6 addr;
 734	int rc, len;
 735
 736	switch(param) {
 737	case ISCSI_PARAM_CONN_PORT:
 738	case ISCSI_PARAM_CONN_ADDRESS:
 739	case ISCSI_PARAM_LOCAL_PORT:
 740		spin_lock_bh(&conn->session->frwd_lock);
 741		if (!tcp_sw_conn || !tcp_sw_conn->sock) {
 742			spin_unlock_bh(&conn->session->frwd_lock);
 743			return -ENOTCONN;
 744		}
 745		if (param == ISCSI_PARAM_LOCAL_PORT)
 746			rc = kernel_getsockname(tcp_sw_conn->sock,
 747						(struct sockaddr *)&addr, &len);
 748		else
 749			rc = kernel_getpeername(tcp_sw_conn->sock,
 750						(struct sockaddr *)&addr, &len);
 751		spin_unlock_bh(&conn->session->frwd_lock);
 752		if (rc)
 753			return rc;
 754
 755		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
 756						 &addr, param, buf);
 757	default:
 758		return iscsi_conn_get_param(cls_conn, param, buf);
 759	}
 760
 761	return 0;
 762}
 763
 764static int iscsi_sw_tcp_host_get_param(struct Scsi_Host *shost,
 765				       enum iscsi_host_param param, char *buf)
 766{
 767	struct iscsi_sw_tcp_host *tcp_sw_host = iscsi_host_priv(shost);
 768	struct iscsi_session *session = tcp_sw_host->session;
 769	struct iscsi_conn *conn;
 770	struct iscsi_tcp_conn *tcp_conn;
 771	struct iscsi_sw_tcp_conn *tcp_sw_conn;
 772	struct sockaddr_in6 addr;
 773	int rc, len;
 774
 775	switch (param) {
 776	case ISCSI_HOST_PARAM_IPADDRESS:
 777		if (!session)
 778			return -ENOTCONN;
 779
 780		spin_lock_bh(&session->frwd_lock);
 781		conn = session->leadconn;
 782		if (!conn) {
 783			spin_unlock_bh(&session->frwd_lock);
 784			return -ENOTCONN;
 785		}
 786		tcp_conn = conn->dd_data;
 787
 788		tcp_sw_conn = tcp_conn->dd_data;
 789		if (!tcp_sw_conn->sock) {
 790			spin_unlock_bh(&session->frwd_lock);
 791			return -ENOTCONN;
 792		}
 793
 794		rc = kernel_getsockname(tcp_sw_conn->sock,
 795					(struct sockaddr *)&addr, &len);
 796		spin_unlock_bh(&session->frwd_lock);
 797		if (rc)
 798			return rc;
 799
 800		return iscsi_conn_get_addr_param((struct sockaddr_storage *)
 801						 &addr, param, buf);
 802	default:
 803		return iscsi_host_get_param(shost, param, buf);
 804	}
 805
 806	return 0;
 807}
 808
 809static void
 810iscsi_sw_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
 811			    struct iscsi_stats *stats)
 812{
 813	struct iscsi_conn *conn = cls_conn->dd_data;
 814	struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
 815	struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
 816
 817	stats->custom_length = 3;
 818	strcpy(stats->custom[0].desc, "tx_sendpage_failures");
 819	stats->custom[0].value = tcp_sw_conn->sendpage_failures_cnt;
 820	strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
 821	stats->custom[1].value = tcp_sw_conn->discontiguous_hdr_cnt;
 822	strcpy(stats->custom[2].desc, "eh_abort_cnt");
 823	stats->custom[2].value = conn->eh_abort_cnt;
 824
 825	iscsi_tcp_conn_get_stats(cls_conn, stats);
 826}
 827
 828static struct iscsi_cls_session *
 829iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
 830			    uint16_t qdepth, uint32_t initial_cmdsn)
 831{
 832	struct iscsi_cls_session *cls_session;
 833	struct iscsi_session *session;
 834	struct iscsi_sw_tcp_host *tcp_sw_host;
 835	struct Scsi_Host *shost;
 836
 837	if (ep) {
 838		printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
 839		return NULL;
 840	}
 841
 842	shost = iscsi_host_alloc(&iscsi_sw_tcp_sht,
 843				 sizeof(struct iscsi_sw_tcp_host), 1);
 844	if (!shost)
 845		return NULL;
 846	shost->transportt = iscsi_sw_tcp_scsi_transport;
 847	shost->cmd_per_lun = qdepth;
 848	shost->max_lun = iscsi_max_lun;
 849	shost->max_id = 0;
 850	shost->max_channel = 0;
 851	shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE;
 852
 853	if (iscsi_host_add(shost, NULL))
 854		goto free_host;
 855
 856	cls_session = iscsi_session_setup(&iscsi_sw_tcp_transport, shost,
 857					  cmds_max, 0,
 858					  sizeof(struct iscsi_tcp_task) +
 859					  sizeof(struct iscsi_sw_tcp_hdrbuf),
 860					  initial_cmdsn, 0);
 861	if (!cls_session)
 862		goto remove_host;
 863	session = cls_session->dd_data;
 864	tcp_sw_host = iscsi_host_priv(shost);
 865	tcp_sw_host->session = session;
 866
 867	shost->can_queue = session->scsi_cmds_max;
 868	if (iscsi_tcp_r2tpool_alloc(session))
 869		goto remove_session;
 870	return cls_session;
 871
 872remove_session:
 873	iscsi_session_teardown(cls_session);
 874remove_host:
 875	iscsi_host_remove(shost);
 876free_host:
 877	iscsi_host_free(shost);
 878	return NULL;
 879}
 880
 881static void iscsi_sw_tcp_session_destroy(struct iscsi_cls_session *cls_session)
 882{
 883	struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
 884
 885	iscsi_tcp_r2tpool_free(cls_session->dd_data);
 886	iscsi_session_teardown(cls_session);
 887
 888	iscsi_host_remove(shost);
 889	iscsi_host_free(shost);
 890}
 891
 892static umode_t iscsi_sw_tcp_attr_is_visible(int param_type, int param)
 893{
 894	switch (param_type) {
 895	case ISCSI_HOST_PARAM:
 896		switch (param) {
 897		case ISCSI_HOST_PARAM_NETDEV_NAME:
 898		case ISCSI_HOST_PARAM_HWADDRESS:
 899		case ISCSI_HOST_PARAM_IPADDRESS:
 900		case ISCSI_HOST_PARAM_INITIATOR_NAME:
 901			return S_IRUGO;
 902		default:
 903			return 0;
 904		}
 905	case ISCSI_PARAM:
 906		switch (param) {
 907		case ISCSI_PARAM_MAX_RECV_DLENGTH:
 908		case ISCSI_PARAM_MAX_XMIT_DLENGTH:
 909		case ISCSI_PARAM_HDRDGST_EN:
 910		case ISCSI_PARAM_DATADGST_EN:
 911		case ISCSI_PARAM_CONN_ADDRESS:
 912		case ISCSI_PARAM_CONN_PORT:
 913		case ISCSI_PARAM_LOCAL_PORT:
 914		case ISCSI_PARAM_EXP_STATSN:
 915		case ISCSI_PARAM_PERSISTENT_ADDRESS:
 916		case ISCSI_PARAM_PERSISTENT_PORT:
 917		case ISCSI_PARAM_PING_TMO:
 918		case ISCSI_PARAM_RECV_TMO:
 919		case ISCSI_PARAM_INITIAL_R2T_EN:
 920		case ISCSI_PARAM_MAX_R2T:
 921		case ISCSI_PARAM_IMM_DATA_EN:
 922		case ISCSI_PARAM_FIRST_BURST:
 923		case ISCSI_PARAM_MAX_BURST:
 924		case ISCSI_PARAM_PDU_INORDER_EN:
 925		case ISCSI_PARAM_DATASEQ_INORDER_EN:
 926		case ISCSI_PARAM_ERL:
 927		case ISCSI_PARAM_TARGET_NAME:
 928		case ISCSI_PARAM_TPGT:
 929		case ISCSI_PARAM_USERNAME:
 930		case ISCSI_PARAM_PASSWORD:
 931		case ISCSI_PARAM_USERNAME_IN:
 932		case ISCSI_PARAM_PASSWORD_IN:
 933		case ISCSI_PARAM_FAST_ABORT:
 934		case ISCSI_PARAM_ABORT_TMO:
 935		case ISCSI_PARAM_LU_RESET_TMO:
 936		case ISCSI_PARAM_TGT_RESET_TMO:
 937		case ISCSI_PARAM_IFACE_NAME:
 938		case ISCSI_PARAM_INITIATOR_NAME:
 939			return S_IRUGO;
 940		default:
 941			return 0;
 942		}
 943	}
 944
 945	return 0;
 946}
 947
 948static int iscsi_sw_tcp_slave_alloc(struct scsi_device *sdev)
 949{
 950	set_bit(QUEUE_FLAG_BIDI, &sdev->request_queue->queue_flags);
 951	return 0;
 952}
 953
 954static int iscsi_sw_tcp_slave_configure(struct scsi_device *sdev)
 955{
 956	blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
 957	blk_queue_dma_alignment(sdev->request_queue, 0);
 958	return 0;
 959}
 960
 961static struct scsi_host_template iscsi_sw_tcp_sht = {
 962	.module			= THIS_MODULE,
 963	.name			= "iSCSI Initiator over TCP/IP",
 964	.queuecommand           = iscsi_queuecommand,
 965	.change_queue_depth	= scsi_change_queue_depth,
 966	.can_queue		= ISCSI_DEF_XMIT_CMDS_MAX - 1,
 967	.sg_tablesize		= 4096,
 968	.max_sectors		= 0xFFFF,
 969	.cmd_per_lun		= ISCSI_DEF_CMD_PER_LUN,
 970	.eh_abort_handler       = iscsi_eh_abort,
 971	.eh_device_reset_handler= iscsi_eh_device_reset,
 972	.eh_target_reset_handler = iscsi_eh_recover_target,
 973	.use_clustering         = DISABLE_CLUSTERING,
 974	.slave_alloc            = iscsi_sw_tcp_slave_alloc,
 975	.slave_configure        = iscsi_sw_tcp_slave_configure,
 976	.target_alloc		= iscsi_target_alloc,
 977	.proc_name		= "iscsi_tcp",
 978	.this_id		= -1,
 979	.track_queue_depth	= 1,
 980};
 981
 982static struct iscsi_transport iscsi_sw_tcp_transport = {
 983	.owner			= THIS_MODULE,
 984	.name			= "tcp",
 985	.caps			= CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
 986				  | CAP_DATADGST,
 987	/* session management */
 988	.create_session		= iscsi_sw_tcp_session_create,
 989	.destroy_session	= iscsi_sw_tcp_session_destroy,
 990	/* connection management */
 991	.create_conn		= iscsi_sw_tcp_conn_create,
 992	.bind_conn		= iscsi_sw_tcp_conn_bind,
 993	.destroy_conn		= iscsi_sw_tcp_conn_destroy,
 994	.attr_is_visible	= iscsi_sw_tcp_attr_is_visible,
 995	.set_param		= iscsi_sw_tcp_conn_set_param,
 996	.get_conn_param		= iscsi_sw_tcp_conn_get_param,
 997	.get_session_param	= iscsi_session_get_param,
 998	.start_conn		= iscsi_conn_start,
 999	.stop_conn		= iscsi_sw_tcp_conn_stop,
1000	/* iscsi host params */
1001	.get_host_param		= iscsi_sw_tcp_host_get_param,
1002	.set_host_param		= iscsi_host_set_param,
1003	/* IO */
1004	.send_pdu		= iscsi_conn_send_pdu,
1005	.get_stats		= iscsi_sw_tcp_conn_get_stats,
1006	/* iscsi task/cmd helpers */
1007	.init_task		= iscsi_tcp_task_init,
1008	.xmit_task		= iscsi_tcp_task_xmit,
1009	.cleanup_task		= iscsi_tcp_cleanup_task,
1010	/* low level pdu helpers */
1011	.xmit_pdu		= iscsi_sw_tcp_pdu_xmit,
1012	.init_pdu		= iscsi_sw_tcp_pdu_init,
1013	.alloc_pdu		= iscsi_sw_tcp_pdu_alloc,
1014	/* recovery */
1015	.session_recovery_timedout = iscsi_session_recovery_timedout,
1016};
1017
1018static int __init iscsi_sw_tcp_init(void)
1019{
1020	if (iscsi_max_lun < 1) {
1021		printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
1022		       iscsi_max_lun);
1023		return -EINVAL;
1024	}
1025
1026	iscsi_sw_tcp_scsi_transport = iscsi_register_transport(
1027						&iscsi_sw_tcp_transport);
1028	if (!iscsi_sw_tcp_scsi_transport)
1029		return -ENODEV;
1030
1031	return 0;
1032}
1033
1034static void __exit iscsi_sw_tcp_exit(void)
1035{
1036	iscsi_unregister_transport(&iscsi_sw_tcp_transport);
1037}
1038
1039module_init(iscsi_sw_tcp_init);
1040module_exit(iscsi_sw_tcp_exit);