thp: transparent hugepage vmstat
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / scsi / iscsi_tcp.c
blobfec47de72535984f1e36b83048f763e7cd0a15a8
1 /*
2 * iSCSI Initiator over TCP/IP Data-Path
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
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.
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.
20 * See the file COPYING included with this distribution for more details.
22 * Credits:
23 * Christoph Hellwig
24 * FUJITA Tomonori
25 * Arne Redlich
26 * Zhenyu Wang
29 #include <linux/types.h>
30 #include <linux/inet.h>
31 #include <linux/slab.h>
32 #include <linux/file.h>
33 #include <linux/blkdev.h>
34 #include <linux/crypto.h>
35 #include <linux/delay.h>
36 #include <linux/kfifo.h>
37 #include <linux/scatterlist.h>
38 #include <net/tcp.h>
39 #include <scsi/scsi_cmnd.h>
40 #include <scsi/scsi_device.h>
41 #include <scsi/scsi_host.h>
42 #include <scsi/scsi.h>
43 #include <scsi/scsi_transport_iscsi.h>
45 #include "iscsi_tcp.h"
47 MODULE_AUTHOR("Mike Christie <michaelc@cs.wisc.edu>, "
48 "Dmitry Yusupov <dmitry_yus@yahoo.com>, "
49 "Alex Aizman <itn780@yahoo.com>");
50 MODULE_DESCRIPTION("iSCSI/TCP data-path");
51 MODULE_LICENSE("GPL");
53 static struct scsi_transport_template *iscsi_sw_tcp_scsi_transport;
54 static struct scsi_host_template iscsi_sw_tcp_sht;
55 static struct iscsi_transport iscsi_sw_tcp_transport;
57 static unsigned int iscsi_max_lun = 512;
58 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
60 static int iscsi_sw_tcp_dbg;
61 module_param_named(debug_iscsi_tcp, iscsi_sw_tcp_dbg, int,
62 S_IRUGO | S_IWUSR);
63 MODULE_PARM_DESC(debug_iscsi_tcp, "Turn on debugging for iscsi_tcp module "
64 "Set to 1 to turn on, and zero to turn off. Default is off.");
66 #define ISCSI_SW_TCP_DBG(_conn, dbg_fmt, arg...) \
67 do { \
68 if (iscsi_sw_tcp_dbg) \
69 iscsi_conn_printk(KERN_INFO, _conn, \
70 "%s " dbg_fmt, \
71 __func__, ##arg); \
72 } while (0);
75 /**
76 * iscsi_sw_tcp_recv - TCP receive in sendfile fashion
77 * @rd_desc: read descriptor
78 * @skb: socket buffer
79 * @offset: offset in skb
80 * @len: skb->len - offset
82 static int iscsi_sw_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
83 unsigned int offset, size_t len)
85 struct iscsi_conn *conn = rd_desc->arg.data;
86 unsigned int consumed, total_consumed = 0;
87 int status;
89 ISCSI_SW_TCP_DBG(conn, "in %d bytes\n", skb->len - offset);
91 do {
92 status = 0;
93 consumed = iscsi_tcp_recv_skb(conn, skb, offset, 0, &status);
94 offset += consumed;
95 total_consumed += consumed;
96 } while (consumed != 0 && status != ISCSI_TCP_SKB_DONE);
98 ISCSI_SW_TCP_DBG(conn, "read %d bytes status %d\n",
99 skb->len - offset, status);
100 return total_consumed;
104 * iscsi_sw_sk_state_check - check socket state
105 * @sk: socket
107 * If the socket is in CLOSE or CLOSE_WAIT we should
108 * not close the connection if there is still some
109 * data pending.
111 static inline int iscsi_sw_sk_state_check(struct sock *sk)
113 struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
115 if ((sk->sk_state == TCP_CLOSE_WAIT || sk->sk_state == TCP_CLOSE) &&
116 !atomic_read(&sk->sk_rmem_alloc)) {
117 ISCSI_SW_TCP_DBG(conn, "TCP_CLOSE|TCP_CLOSE_WAIT\n");
118 iscsi_conn_failure(conn, ISCSI_ERR_TCP_CONN_CLOSE);
119 return -ECONNRESET;
121 return 0;
124 static void iscsi_sw_tcp_data_ready(struct sock *sk, int flag)
126 struct iscsi_conn *conn = sk->sk_user_data;
127 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
128 read_descriptor_t rd_desc;
130 read_lock(&sk->sk_callback_lock);
133 * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
134 * We set count to 1 because we want the network layer to
135 * hand us all the skbs that are available. iscsi_tcp_recv
136 * handled pdus that cross buffers or pdus that still need data.
138 rd_desc.arg.data = conn;
139 rd_desc.count = 1;
140 tcp_read_sock(sk, &rd_desc, iscsi_sw_tcp_recv);
142 iscsi_sw_sk_state_check(sk);
144 read_unlock(&sk->sk_callback_lock);
146 /* If we had to (atomically) map a highmem page,
147 * unmap it now. */
148 iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
151 static void iscsi_sw_tcp_state_change(struct sock *sk)
153 struct iscsi_tcp_conn *tcp_conn;
154 struct iscsi_sw_tcp_conn *tcp_sw_conn;
155 struct iscsi_conn *conn;
156 struct iscsi_session *session;
157 void (*old_state_change)(struct sock *);
159 read_lock(&sk->sk_callback_lock);
161 conn = (struct iscsi_conn*)sk->sk_user_data;
162 session = conn->session;
164 iscsi_sw_sk_state_check(sk);
166 tcp_conn = conn->dd_data;
167 tcp_sw_conn = tcp_conn->dd_data;
168 old_state_change = tcp_sw_conn->old_state_change;
170 read_unlock(&sk->sk_callback_lock);
172 old_state_change(sk);
176 * iscsi_write_space - Called when more output buffer space is available
177 * @sk: socket space is available for
179 static void iscsi_sw_tcp_write_space(struct sock *sk)
181 struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
182 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
183 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
185 tcp_sw_conn->old_write_space(sk);
186 ISCSI_SW_TCP_DBG(conn, "iscsi_write_space\n");
187 iscsi_conn_queue_work(conn);
190 static void iscsi_sw_tcp_conn_set_callbacks(struct iscsi_conn *conn)
192 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
193 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
194 struct sock *sk = tcp_sw_conn->sock->sk;
196 /* assign new callbacks */
197 write_lock_bh(&sk->sk_callback_lock);
198 sk->sk_user_data = conn;
199 tcp_sw_conn->old_data_ready = sk->sk_data_ready;
200 tcp_sw_conn->old_state_change = sk->sk_state_change;
201 tcp_sw_conn->old_write_space = sk->sk_write_space;
202 sk->sk_data_ready = iscsi_sw_tcp_data_ready;
203 sk->sk_state_change = iscsi_sw_tcp_state_change;
204 sk->sk_write_space = iscsi_sw_tcp_write_space;
205 write_unlock_bh(&sk->sk_callback_lock);
208 static void
209 iscsi_sw_tcp_conn_restore_callbacks(struct iscsi_conn *conn)
211 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
212 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
213 struct sock *sk = tcp_sw_conn->sock->sk;
215 /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
216 write_lock_bh(&sk->sk_callback_lock);
217 sk->sk_user_data = NULL;
218 sk->sk_data_ready = tcp_sw_conn->old_data_ready;
219 sk->sk_state_change = tcp_sw_conn->old_state_change;
220 sk->sk_write_space = tcp_sw_conn->old_write_space;
221 sk->sk_no_check = 0;
222 write_unlock_bh(&sk->sk_callback_lock);
226 * iscsi_sw_tcp_xmit_segment - transmit segment
227 * @tcp_conn: the iSCSI TCP connection
228 * @segment: the buffer to transmnit
230 * This function transmits as much of the buffer as
231 * the network layer will accept, and returns the number of
232 * bytes transmitted.
234 * If CRC hashing is enabled, the function will compute the
235 * hash as it goes. When the entire segment has been transmitted,
236 * it will retrieve the hash value and send it as well.
238 static int iscsi_sw_tcp_xmit_segment(struct iscsi_tcp_conn *tcp_conn,
239 struct iscsi_segment *segment)
241 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
242 struct socket *sk = tcp_sw_conn->sock;
243 unsigned int copied = 0;
244 int r = 0;
246 while (!iscsi_tcp_segment_done(tcp_conn, segment, 0, r)) {
247 struct scatterlist *sg;
248 unsigned int offset, copy;
249 int flags = 0;
251 r = 0;
252 offset = segment->copied;
253 copy = segment->size - offset;
255 if (segment->total_copied + segment->size < segment->total_size)
256 flags |= MSG_MORE;
258 /* Use sendpage if we can; else fall back to sendmsg */
259 if (!segment->data) {
260 sg = segment->sg;
261 offset += segment->sg_offset + sg->offset;
262 r = tcp_sw_conn->sendpage(sk, sg_page(sg), offset,
263 copy, flags);
264 } else {
265 struct msghdr msg = { .msg_flags = flags };
266 struct kvec iov = {
267 .iov_base = segment->data + offset,
268 .iov_len = copy
271 r = kernel_sendmsg(sk, &msg, &iov, 1, copy);
274 if (r < 0) {
275 iscsi_tcp_segment_unmap(segment);
276 return r;
278 copied += r;
280 return copied;
284 * iscsi_sw_tcp_xmit - TCP transmit
286 static int iscsi_sw_tcp_xmit(struct iscsi_conn *conn)
288 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
289 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
290 struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
291 unsigned int consumed = 0;
292 int rc = 0;
294 while (1) {
295 rc = iscsi_sw_tcp_xmit_segment(tcp_conn, segment);
297 * We may not have been able to send data because the conn
298 * is getting stopped. libiscsi will know so propogate err
299 * for it to do the right thing.
301 if (rc == -EAGAIN)
302 return rc;
303 else if (rc < 0) {
304 rc = ISCSI_ERR_XMIT_FAILED;
305 goto error;
306 } else if (rc == 0)
307 break;
309 consumed += rc;
311 if (segment->total_copied >= segment->total_size) {
312 if (segment->done != NULL) {
313 rc = segment->done(tcp_conn, segment);
314 if (rc != 0)
315 goto error;
320 ISCSI_SW_TCP_DBG(conn, "xmit %d bytes\n", consumed);
322 conn->txdata_octets += consumed;
323 return consumed;
325 error:
326 /* Transmit error. We could initiate error recovery
327 * here. */
328 ISCSI_SW_TCP_DBG(conn, "Error sending PDU, errno=%d\n", rc);
329 iscsi_conn_failure(conn, rc);
330 return -EIO;
334 * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
336 static inline int iscsi_sw_tcp_xmit_qlen(struct iscsi_conn *conn)
338 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
339 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
340 struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
342 return segment->total_copied - segment->total_size;
345 static int iscsi_sw_tcp_pdu_xmit(struct iscsi_task *task)
347 struct iscsi_conn *conn = task->conn;
348 int rc;
350 while (iscsi_sw_tcp_xmit_qlen(conn)) {
351 rc = iscsi_sw_tcp_xmit(conn);
352 if (rc == 0)
353 return -EAGAIN;
354 if (rc < 0)
355 return rc;
358 return 0;
362 * This is called when we're done sending the header.
363 * Simply copy the data_segment to the send segment, and return.
365 static int iscsi_sw_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
366 struct iscsi_segment *segment)
368 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
370 tcp_sw_conn->out.segment = tcp_sw_conn->out.data_segment;
371 ISCSI_SW_TCP_DBG(tcp_conn->iscsi_conn,
372 "Header done. Next segment size %u total_size %u\n",
373 tcp_sw_conn->out.segment.size,
374 tcp_sw_conn->out.segment.total_size);
375 return 0;
378 static void iscsi_sw_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr,
379 size_t hdrlen)
381 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
382 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
384 ISCSI_SW_TCP_DBG(conn, "%s\n", conn->hdrdgst_en ?
385 "digest enabled" : "digest disabled");
387 /* Clear the data segment - needs to be filled in by the
388 * caller using iscsi_tcp_send_data_prep() */
389 memset(&tcp_sw_conn->out.data_segment, 0,
390 sizeof(struct iscsi_segment));
392 /* If header digest is enabled, compute the CRC and
393 * place the digest into the same buffer. We make
394 * sure that both iscsi_tcp_task and mtask have
395 * sufficient room.
397 if (conn->hdrdgst_en) {
398 iscsi_tcp_dgst_header(&tcp_sw_conn->tx_hash, hdr, hdrlen,
399 hdr + hdrlen);
400 hdrlen += ISCSI_DIGEST_SIZE;
403 /* Remember header pointer for later, when we need
404 * to decide whether there's a payload to go along
405 * with the header. */
406 tcp_sw_conn->out.hdr = hdr;
408 iscsi_segment_init_linear(&tcp_sw_conn->out.segment, hdr, hdrlen,
409 iscsi_sw_tcp_send_hdr_done, NULL);
413 * Prepare the send buffer for the payload data.
414 * Padding and checksumming will all be taken care
415 * of by the iscsi_segment routines.
417 static int
418 iscsi_sw_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
419 unsigned int count, unsigned int offset,
420 unsigned int len)
422 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
423 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
424 struct hash_desc *tx_hash = NULL;
425 unsigned int hdr_spec_len;
427 ISCSI_SW_TCP_DBG(conn, "offset=%d, datalen=%d %s\n", offset, len,
428 conn->datadgst_en ?
429 "digest enabled" : "digest disabled");
431 /* Make sure the datalen matches what the caller
432 said he would send. */
433 hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
434 WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
436 if (conn->datadgst_en)
437 tx_hash = &tcp_sw_conn->tx_hash;
439 return iscsi_segment_seek_sg(&tcp_sw_conn->out.data_segment,
440 sg, count, offset, len,
441 NULL, tx_hash);
444 static void
445 iscsi_sw_tcp_send_linear_data_prep(struct iscsi_conn *conn, void *data,
446 size_t len)
448 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
449 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
450 struct hash_desc *tx_hash = NULL;
451 unsigned int hdr_spec_len;
453 ISCSI_SW_TCP_DBG(conn, "datalen=%zd %s\n", len, conn->datadgst_en ?
454 "digest enabled" : "digest disabled");
456 /* Make sure the datalen matches what the caller
457 said he would send. */
458 hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
459 WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
461 if (conn->datadgst_en)
462 tx_hash = &tcp_sw_conn->tx_hash;
464 iscsi_segment_init_linear(&tcp_sw_conn->out.data_segment,
465 data, len, NULL, tx_hash);
468 static int iscsi_sw_tcp_pdu_init(struct iscsi_task *task,
469 unsigned int offset, unsigned int count)
471 struct iscsi_conn *conn = task->conn;
472 int err = 0;
474 iscsi_sw_tcp_send_hdr_prep(conn, task->hdr, task->hdr_len);
476 if (!count)
477 return 0;
479 if (!task->sc)
480 iscsi_sw_tcp_send_linear_data_prep(conn, task->data, count);
481 else {
482 struct scsi_data_buffer *sdb = scsi_out(task->sc);
484 err = iscsi_sw_tcp_send_data_prep(conn, sdb->table.sgl,
485 sdb->table.nents, offset,
486 count);
489 if (err) {
490 /* got invalid offset/len */
491 return -EIO;
493 return 0;
496 static int iscsi_sw_tcp_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
498 struct iscsi_tcp_task *tcp_task = task->dd_data;
500 task->hdr = task->dd_data + sizeof(*tcp_task);
501 task->hdr_max = sizeof(struct iscsi_sw_tcp_hdrbuf) - ISCSI_DIGEST_SIZE;
502 return 0;
505 static struct iscsi_cls_conn *
506 iscsi_sw_tcp_conn_create(struct iscsi_cls_session *cls_session,
507 uint32_t conn_idx)
509 struct iscsi_conn *conn;
510 struct iscsi_cls_conn *cls_conn;
511 struct iscsi_tcp_conn *tcp_conn;
512 struct iscsi_sw_tcp_conn *tcp_sw_conn;
514 cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*tcp_sw_conn),
515 conn_idx);
516 if (!cls_conn)
517 return NULL;
518 conn = cls_conn->dd_data;
519 tcp_conn = conn->dd_data;
520 tcp_sw_conn = tcp_conn->dd_data;
522 tcp_sw_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
523 CRYPTO_ALG_ASYNC);
524 tcp_sw_conn->tx_hash.flags = 0;
525 if (IS_ERR(tcp_sw_conn->tx_hash.tfm))
526 goto free_conn;
528 tcp_sw_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
529 CRYPTO_ALG_ASYNC);
530 tcp_sw_conn->rx_hash.flags = 0;
531 if (IS_ERR(tcp_sw_conn->rx_hash.tfm))
532 goto free_tx_tfm;
533 tcp_conn->rx_hash = &tcp_sw_conn->rx_hash;
535 return cls_conn;
537 free_tx_tfm:
538 crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
539 free_conn:
540 iscsi_conn_printk(KERN_ERR, conn,
541 "Could not create connection due to crc32c "
542 "loading error. Make sure the crc32c "
543 "module is built as a module or into the "
544 "kernel\n");
545 iscsi_tcp_conn_teardown(cls_conn);
546 return NULL;
549 static void iscsi_sw_tcp_release_conn(struct iscsi_conn *conn)
551 struct iscsi_session *session = conn->session;
552 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
553 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
554 struct socket *sock = tcp_sw_conn->sock;
556 if (!sock)
557 return;
559 sock_hold(sock->sk);
560 iscsi_sw_tcp_conn_restore_callbacks(conn);
561 sock_put(sock->sk);
563 spin_lock_bh(&session->lock);
564 tcp_sw_conn->sock = NULL;
565 spin_unlock_bh(&session->lock);
566 sockfd_put(sock);
569 static void iscsi_sw_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
571 struct iscsi_conn *conn = cls_conn->dd_data;
572 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
573 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
575 iscsi_sw_tcp_release_conn(conn);
577 if (tcp_sw_conn->tx_hash.tfm)
578 crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
579 if (tcp_sw_conn->rx_hash.tfm)
580 crypto_free_hash(tcp_sw_conn->rx_hash.tfm);
582 iscsi_tcp_conn_teardown(cls_conn);
585 static void iscsi_sw_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
587 struct iscsi_conn *conn = cls_conn->dd_data;
588 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
589 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
590 struct socket *sock = tcp_sw_conn->sock;
592 /* userspace may have goofed up and not bound us */
593 if (!sock)
594 return;
596 * Make sure our recv side is stopped.
597 * Older tools called conn stop before ep_disconnect
598 * so IO could still be coming in.
600 write_lock_bh(&tcp_sw_conn->sock->sk->sk_callback_lock);
601 set_bit(ISCSI_SUSPEND_BIT, &conn->suspend_rx);
602 write_unlock_bh(&tcp_sw_conn->sock->sk->sk_callback_lock);
604 sock->sk->sk_err = EIO;
605 wake_up_interruptible(sk_sleep(sock->sk));
607 iscsi_conn_stop(cls_conn, flag);
608 iscsi_sw_tcp_release_conn(conn);
611 static int iscsi_sw_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
612 char *buf, int *port,
613 int (*getname)(struct socket *,
614 struct sockaddr *,
615 int *addrlen))
617 struct sockaddr_storage *addr;
618 struct sockaddr_in6 *sin6;
619 struct sockaddr_in *sin;
620 int rc = 0, len;
622 addr = kmalloc(sizeof(*addr), GFP_KERNEL);
623 if (!addr)
624 return -ENOMEM;
626 if (getname(sock, (struct sockaddr *) addr, &len)) {
627 rc = -ENODEV;
628 goto free_addr;
631 switch (addr->ss_family) {
632 case AF_INET:
633 sin = (struct sockaddr_in *)addr;
634 spin_lock_bh(&conn->session->lock);
635 sprintf(buf, "%pI4", &sin->sin_addr.s_addr);
636 *port = be16_to_cpu(sin->sin_port);
637 spin_unlock_bh(&conn->session->lock);
638 break;
639 case AF_INET6:
640 sin6 = (struct sockaddr_in6 *)addr;
641 spin_lock_bh(&conn->session->lock);
642 sprintf(buf, "%pI6", &sin6->sin6_addr);
643 *port = be16_to_cpu(sin6->sin6_port);
644 spin_unlock_bh(&conn->session->lock);
645 break;
647 free_addr:
648 kfree(addr);
649 return rc;
652 static int
653 iscsi_sw_tcp_conn_bind(struct iscsi_cls_session *cls_session,
654 struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
655 int is_leading)
657 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
658 struct iscsi_host *ihost = shost_priv(shost);
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;
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;
674 * copy these values now because if we drop the session
675 * userspace may still want to query the values since we will
676 * be using them for the reconnect
678 err = iscsi_sw_tcp_get_addr(conn, sock, conn->portal_address,
679 &conn->portal_port, kernel_getpeername);
680 if (err)
681 goto free_socket;
683 err = iscsi_sw_tcp_get_addr(conn, sock, ihost->local_address,
684 &ihost->local_port, kernel_getsockname);
685 if (err)
686 goto free_socket;
688 err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
689 if (err)
690 goto free_socket;
692 /* bind iSCSI connection and socket */
693 tcp_sw_conn->sock = sock;
695 /* setup Socket parameters */
696 sk = sock->sk;
697 sk->sk_reuse = 1;
698 sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
699 sk->sk_allocation = GFP_ATOMIC;
701 iscsi_sw_tcp_conn_set_callbacks(conn);
702 tcp_sw_conn->sendpage = tcp_sw_conn->sock->ops->sendpage;
704 * set receive state machine into initial state
706 iscsi_tcp_hdr_recv_prep(tcp_conn);
707 return 0;
709 free_socket:
710 sockfd_put(sock);
711 return err;
714 static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn,
715 enum iscsi_param param, char *buf,
716 int buflen)
718 struct iscsi_conn *conn = cls_conn->dd_data;
719 struct iscsi_session *session = conn->session;
720 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
721 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
722 int value;
724 switch(param) {
725 case ISCSI_PARAM_HDRDGST_EN:
726 iscsi_set_param(cls_conn, param, buf, buflen);
727 break;
728 case ISCSI_PARAM_DATADGST_EN:
729 iscsi_set_param(cls_conn, param, buf, buflen);
730 tcp_sw_conn->sendpage = conn->datadgst_en ?
731 sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage;
732 break;
733 case ISCSI_PARAM_MAX_R2T:
734 sscanf(buf, "%d", &value);
735 if (value <= 0 || !is_power_of_2(value))
736 return -EINVAL;
737 if (session->max_r2t == value)
738 break;
739 iscsi_tcp_r2tpool_free(session);
740 iscsi_set_param(cls_conn, param, buf, buflen);
741 if (iscsi_tcp_r2tpool_alloc(session))
742 return -ENOMEM;
743 break;
744 default:
745 return iscsi_set_param(cls_conn, param, buf, buflen);
748 return 0;
751 static int iscsi_sw_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
752 enum iscsi_param param, char *buf)
754 struct iscsi_conn *conn = cls_conn->dd_data;
755 int len;
757 switch(param) {
758 case ISCSI_PARAM_CONN_PORT:
759 spin_lock_bh(&conn->session->lock);
760 len = sprintf(buf, "%hu\n", conn->portal_port);
761 spin_unlock_bh(&conn->session->lock);
762 break;
763 case ISCSI_PARAM_CONN_ADDRESS:
764 spin_lock_bh(&conn->session->lock);
765 len = sprintf(buf, "%s\n", conn->portal_address);
766 spin_unlock_bh(&conn->session->lock);
767 break;
768 default:
769 return iscsi_conn_get_param(cls_conn, param, buf);
772 return len;
775 static void
776 iscsi_sw_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
777 struct iscsi_stats *stats)
779 struct iscsi_conn *conn = cls_conn->dd_data;
780 struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
781 struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
783 stats->custom_length = 3;
784 strcpy(stats->custom[0].desc, "tx_sendpage_failures");
785 stats->custom[0].value = tcp_sw_conn->sendpage_failures_cnt;
786 strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
787 stats->custom[1].value = tcp_sw_conn->discontiguous_hdr_cnt;
788 strcpy(stats->custom[2].desc, "eh_abort_cnt");
789 stats->custom[2].value = conn->eh_abort_cnt;
791 iscsi_tcp_conn_get_stats(cls_conn, stats);
794 static struct iscsi_cls_session *
795 iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
796 uint16_t qdepth, uint32_t initial_cmdsn)
798 struct iscsi_cls_session *cls_session;
799 struct iscsi_session *session;
800 struct Scsi_Host *shost;
802 if (ep) {
803 printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
804 return NULL;
807 shost = iscsi_host_alloc(&iscsi_sw_tcp_sht, 0, 1);
808 if (!shost)
809 return NULL;
810 shost->transportt = iscsi_sw_tcp_scsi_transport;
811 shost->cmd_per_lun = qdepth;
812 shost->max_lun = iscsi_max_lun;
813 shost->max_id = 0;
814 shost->max_channel = 0;
815 shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE;
817 if (iscsi_host_add(shost, NULL))
818 goto free_host;
820 cls_session = iscsi_session_setup(&iscsi_sw_tcp_transport, shost,
821 cmds_max, 0,
822 sizeof(struct iscsi_tcp_task) +
823 sizeof(struct iscsi_sw_tcp_hdrbuf),
824 initial_cmdsn, 0);
825 if (!cls_session)
826 goto remove_host;
827 session = cls_session->dd_data;
829 shost->can_queue = session->scsi_cmds_max;
830 if (iscsi_tcp_r2tpool_alloc(session))
831 goto remove_session;
832 return cls_session;
834 remove_session:
835 iscsi_session_teardown(cls_session);
836 remove_host:
837 iscsi_host_remove(shost);
838 free_host:
839 iscsi_host_free(shost);
840 return NULL;
843 static void iscsi_sw_tcp_session_destroy(struct iscsi_cls_session *cls_session)
845 struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
847 iscsi_tcp_r2tpool_free(cls_session->dd_data);
848 iscsi_session_teardown(cls_session);
850 iscsi_host_remove(shost);
851 iscsi_host_free(shost);
854 static int iscsi_sw_tcp_slave_alloc(struct scsi_device *sdev)
856 set_bit(QUEUE_FLAG_BIDI, &sdev->request_queue->queue_flags);
857 return 0;
860 static int iscsi_sw_tcp_slave_configure(struct scsi_device *sdev)
862 blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
863 blk_queue_dma_alignment(sdev->request_queue, 0);
864 return 0;
867 static struct scsi_host_template iscsi_sw_tcp_sht = {
868 .module = THIS_MODULE,
869 .name = "iSCSI Initiator over TCP/IP",
870 .queuecommand = iscsi_queuecommand,
871 .change_queue_depth = iscsi_change_queue_depth,
872 .can_queue = ISCSI_DEF_XMIT_CMDS_MAX - 1,
873 .sg_tablesize = 4096,
874 .max_sectors = 0xFFFF,
875 .cmd_per_lun = ISCSI_DEF_CMD_PER_LUN,
876 .eh_abort_handler = iscsi_eh_abort,
877 .eh_device_reset_handler= iscsi_eh_device_reset,
878 .eh_target_reset_handler = iscsi_eh_recover_target,
879 .use_clustering = DISABLE_CLUSTERING,
880 .slave_alloc = iscsi_sw_tcp_slave_alloc,
881 .slave_configure = iscsi_sw_tcp_slave_configure,
882 .target_alloc = iscsi_target_alloc,
883 .proc_name = "iscsi_tcp",
884 .this_id = -1,
887 static struct iscsi_transport iscsi_sw_tcp_transport = {
888 .owner = THIS_MODULE,
889 .name = "tcp",
890 .caps = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
891 | CAP_DATADGST,
892 .param_mask = ISCSI_MAX_RECV_DLENGTH |
893 ISCSI_MAX_XMIT_DLENGTH |
894 ISCSI_HDRDGST_EN |
895 ISCSI_DATADGST_EN |
896 ISCSI_INITIAL_R2T_EN |
897 ISCSI_MAX_R2T |
898 ISCSI_IMM_DATA_EN |
899 ISCSI_FIRST_BURST |
900 ISCSI_MAX_BURST |
901 ISCSI_PDU_INORDER_EN |
902 ISCSI_DATASEQ_INORDER_EN |
903 ISCSI_ERL |
904 ISCSI_CONN_PORT |
905 ISCSI_CONN_ADDRESS |
906 ISCSI_EXP_STATSN |
907 ISCSI_PERSISTENT_PORT |
908 ISCSI_PERSISTENT_ADDRESS |
909 ISCSI_TARGET_NAME | ISCSI_TPGT |
910 ISCSI_USERNAME | ISCSI_PASSWORD |
911 ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
912 ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
913 ISCSI_LU_RESET_TMO | ISCSI_TGT_RESET_TMO |
914 ISCSI_PING_TMO | ISCSI_RECV_TMO |
915 ISCSI_IFACE_NAME | ISCSI_INITIATOR_NAME,
916 .host_param_mask = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
917 ISCSI_HOST_INITIATOR_NAME |
918 ISCSI_HOST_NETDEV_NAME,
919 /* session management */
920 .create_session = iscsi_sw_tcp_session_create,
921 .destroy_session = iscsi_sw_tcp_session_destroy,
922 /* connection management */
923 .create_conn = iscsi_sw_tcp_conn_create,
924 .bind_conn = iscsi_sw_tcp_conn_bind,
925 .destroy_conn = iscsi_sw_tcp_conn_destroy,
926 .set_param = iscsi_sw_tcp_conn_set_param,
927 .get_conn_param = iscsi_sw_tcp_conn_get_param,
928 .get_session_param = iscsi_session_get_param,
929 .start_conn = iscsi_conn_start,
930 .stop_conn = iscsi_sw_tcp_conn_stop,
931 /* iscsi host params */
932 .get_host_param = iscsi_host_get_param,
933 .set_host_param = iscsi_host_set_param,
934 /* IO */
935 .send_pdu = iscsi_conn_send_pdu,
936 .get_stats = iscsi_sw_tcp_conn_get_stats,
937 /* iscsi task/cmd helpers */
938 .init_task = iscsi_tcp_task_init,
939 .xmit_task = iscsi_tcp_task_xmit,
940 .cleanup_task = iscsi_tcp_cleanup_task,
941 /* low level pdu helpers */
942 .xmit_pdu = iscsi_sw_tcp_pdu_xmit,
943 .init_pdu = iscsi_sw_tcp_pdu_init,
944 .alloc_pdu = iscsi_sw_tcp_pdu_alloc,
945 /* recovery */
946 .session_recovery_timedout = iscsi_session_recovery_timedout,
949 static int __init iscsi_sw_tcp_init(void)
951 if (iscsi_max_lun < 1) {
952 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
953 iscsi_max_lun);
954 return -EINVAL;
957 iscsi_sw_tcp_scsi_transport = iscsi_register_transport(
958 &iscsi_sw_tcp_transport);
959 if (!iscsi_sw_tcp_scsi_transport)
960 return -ENODEV;
962 return 0;
965 static void __exit iscsi_sw_tcp_exit(void)
967 iscsi_unregister_transport(&iscsi_sw_tcp_transport);
970 module_init(iscsi_sw_tcp_init);
971 module_exit(iscsi_sw_tcp_exit);