config: tegra3: enable /dev mount with ACL
[linux-2.6.git] / drivers / scsi / iscsi_tcp.c
1 /*
2  * iSCSI Initiator over TCP/IP Data-Path
3  *
4  * Copyright (C) 2004 Dmitry Yusupov
5  * Copyright (C) 2004 Alex Aizman
6  * Copyright (C) 2005 - 2006 Mike Christie
7  * Copyright (C) 2006 Red Hat, Inc.  All rights reserved.
8  * maintained by open-iscsi@googlegroups.com
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published
12  * by the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful, but
16  * WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18  * General Public License for more details.
19  *
20  * See the file COPYING included with this distribution for more details.
21  *
22  * Credits:
23  *      Christoph Hellwig
24  *      FUJITA Tomonori
25  *      Arne Redlich
26  *      Zhenyu Wang
27  */
28
29 #include <linux/types.h>
30 #include <linux/inet.h>
31 #include <linux/slab.h>
32 #include <linux/file.h>
33 #include <linux/blkdev.h>
34 #include <linux/crypto.h>
35 #include <linux/delay.h>
36 #include <linux/kfifo.h>
37 #include <linux/scatterlist.h>
38 #include <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>
44
45 #include "iscsi_tcp.h"
46
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");
52
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;
56
57 static unsigned int iscsi_max_lun = 512;
58 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
59
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.");
65
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);
73
74
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
81  */
82 static int iscsi_sw_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
83                              unsigned int offset, size_t len)
84 {
85         struct iscsi_conn *conn = rd_desc->arg.data;
86         unsigned int consumed, total_consumed = 0;
87         int status;
88
89         ISCSI_SW_TCP_DBG(conn, "in %d bytes\n", skb->len - offset);
90
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);
97
98         ISCSI_SW_TCP_DBG(conn, "read %d bytes status %d\n",
99                          skb->len - offset, status);
100         return total_consumed;
101 }
102
103 /**
104  * iscsi_sw_sk_state_check - check socket state
105  * @sk: socket
106  *
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.
110  *
111  * Must be called with sk_callback_lock.
112  */
113 static inline int iscsi_sw_sk_state_check(struct sock *sk)
114 {
115         struct iscsi_conn *conn = sk->sk_user_data;
116
117         if ((sk->sk_state == TCP_CLOSE_WAIT || sk->sk_state == TCP_CLOSE) &&
118             !atomic_read(&sk->sk_rmem_alloc)) {
119                 ISCSI_SW_TCP_DBG(conn, "TCP_CLOSE|TCP_CLOSE_WAIT\n");
120                 iscsi_conn_failure(conn, ISCSI_ERR_TCP_CONN_CLOSE);
121                 return -ECONNRESET;
122         }
123         return 0;
124 }
125
126 static void iscsi_sw_tcp_data_ready(struct sock *sk, int flag)
127 {
128         struct iscsi_conn *conn;
129         struct iscsi_tcp_conn *tcp_conn;
130         read_descriptor_t rd_desc;
131
132         read_lock(&sk->sk_callback_lock);
133         conn = sk->sk_user_data;
134         if (!conn) {
135                 read_unlock(&sk->sk_callback_lock);
136                 return;
137         }
138         tcp_conn = conn->dd_data;
139
140         /*
141          * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
142          * We set count to 1 because we want the network layer to
143          * hand us all the skbs that are available. iscsi_tcp_recv
144          * handled pdus that cross buffers or pdus that still need data.
145          */
146         rd_desc.arg.data = conn;
147         rd_desc.count = 1;
148         tcp_read_sock(sk, &rd_desc, iscsi_sw_tcp_recv);
149
150         iscsi_sw_sk_state_check(sk);
151
152         /* If we had to (atomically) map a highmem page,
153          * unmap it now. */
154         iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
155         read_unlock(&sk->sk_callback_lock);
156 }
157
158 static void iscsi_sw_tcp_state_change(struct sock *sk)
159 {
160         struct iscsi_tcp_conn *tcp_conn;
161         struct iscsi_sw_tcp_conn *tcp_sw_conn;
162         struct iscsi_conn *conn;
163         struct iscsi_session *session;
164         void (*old_state_change)(struct sock *);
165
166         read_lock(&sk->sk_callback_lock);
167         conn = sk->sk_user_data;
168         if (!conn) {
169                 read_unlock(&sk->sk_callback_lock);
170                 return;
171         }
172         session = conn->session;
173
174         iscsi_sw_sk_state_check(sk);
175
176         tcp_conn = conn->dd_data;
177         tcp_sw_conn = tcp_conn->dd_data;
178         old_state_change = tcp_sw_conn->old_state_change;
179
180         read_unlock(&sk->sk_callback_lock);
181
182         old_state_change(sk);
183 }
184
185 /**
186  * iscsi_write_space - Called when more output buffer space is available
187  * @sk: socket space is available for
188  **/
189 static void iscsi_sw_tcp_write_space(struct sock *sk)
190 {
191         struct iscsi_conn *conn;
192         struct iscsi_tcp_conn *tcp_conn;
193         struct iscsi_sw_tcp_conn *tcp_sw_conn;
194         void (*old_write_space)(struct sock *);
195
196         read_lock_bh(&sk->sk_callback_lock);
197         conn = sk->sk_user_data;
198         if (!conn) {
199                 read_unlock_bh(&sk->sk_callback_lock);
200                 return;
201         }
202
203         tcp_conn = conn->dd_data;
204         tcp_sw_conn = tcp_conn->dd_data;
205         old_write_space = tcp_sw_conn->old_write_space;
206         read_unlock_bh(&sk->sk_callback_lock);
207
208         old_write_space(sk);
209
210         ISCSI_SW_TCP_DBG(conn, "iscsi_write_space\n");
211         iscsi_conn_queue_work(conn);
212 }
213
214 static void iscsi_sw_tcp_conn_set_callbacks(struct iscsi_conn *conn)
215 {
216         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
217         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
218         struct sock *sk = tcp_sw_conn->sock->sk;
219
220         /* assign new callbacks */
221         write_lock_bh(&sk->sk_callback_lock);
222         sk->sk_user_data = conn;
223         tcp_sw_conn->old_data_ready = sk->sk_data_ready;
224         tcp_sw_conn->old_state_change = sk->sk_state_change;
225         tcp_sw_conn->old_write_space = sk->sk_write_space;
226         sk->sk_data_ready = iscsi_sw_tcp_data_ready;
227         sk->sk_state_change = iscsi_sw_tcp_state_change;
228         sk->sk_write_space = iscsi_sw_tcp_write_space;
229         write_unlock_bh(&sk->sk_callback_lock);
230 }
231
232 static void
233 iscsi_sw_tcp_conn_restore_callbacks(struct iscsi_conn *conn)
234 {
235         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
236         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
237         struct sock *sk = tcp_sw_conn->sock->sk;
238
239         /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
240         write_lock_bh(&sk->sk_callback_lock);
241         sk->sk_user_data    = NULL;
242         sk->sk_data_ready   = tcp_sw_conn->old_data_ready;
243         sk->sk_state_change = tcp_sw_conn->old_state_change;
244         sk->sk_write_space  = tcp_sw_conn->old_write_space;
245         sk->sk_no_check  = 0;
246         write_unlock_bh(&sk->sk_callback_lock);
247 }
248
249 /**
250  * iscsi_sw_tcp_xmit_segment - transmit segment
251  * @tcp_conn: the iSCSI TCP connection
252  * @segment: the buffer to transmnit
253  *
254  * This function transmits as much of the buffer as
255  * the network layer will accept, and returns the number of
256  * bytes transmitted.
257  *
258  * If CRC hashing is enabled, the function will compute the
259  * hash as it goes. When the entire segment has been transmitted,
260  * it will retrieve the hash value and send it as well.
261  */
262 static int iscsi_sw_tcp_xmit_segment(struct iscsi_tcp_conn *tcp_conn,
263                                      struct iscsi_segment *segment)
264 {
265         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
266         struct socket *sk = tcp_sw_conn->sock;
267         unsigned int copied = 0;
268         int r = 0;
269
270         while (!iscsi_tcp_segment_done(tcp_conn, segment, 0, r)) {
271                 struct scatterlist *sg;
272                 unsigned int offset, copy;
273                 int flags = 0;
274
275                 r = 0;
276                 offset = segment->copied;
277                 copy = segment->size - offset;
278
279                 if (segment->total_copied + segment->size < segment->total_size)
280                         flags |= MSG_MORE;
281
282                 /* Use sendpage if we can; else fall back to sendmsg */
283                 if (!segment->data) {
284                         sg = segment->sg;
285                         offset += segment->sg_offset + sg->offset;
286                         r = tcp_sw_conn->sendpage(sk, sg_page(sg), offset,
287                                                   copy, flags);
288                 } else {
289                         struct msghdr msg = { .msg_flags = flags };
290                         struct kvec iov = {
291                                 .iov_base = segment->data + offset,
292                                 .iov_len = copy
293                         };
294
295                         r = kernel_sendmsg(sk, &msg, &iov, 1, copy);
296                 }
297
298                 if (r < 0) {
299                         iscsi_tcp_segment_unmap(segment);
300                         return r;
301                 }
302                 copied += r;
303         }
304         return copied;
305 }
306
307 /**
308  * iscsi_sw_tcp_xmit - TCP transmit
309  **/
310 static int iscsi_sw_tcp_xmit(struct iscsi_conn *conn)
311 {
312         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
313         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
314         struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
315         unsigned int consumed = 0;
316         int rc = 0;
317
318         while (1) {
319                 rc = iscsi_sw_tcp_xmit_segment(tcp_conn, segment);
320                 /*
321                  * We may not have been able to send data because the conn
322                  * is getting stopped. libiscsi will know so propagate err
323                  * for it to do the right thing.
324                  */
325                 if (rc == -EAGAIN)
326                         return rc;
327                 else if (rc < 0) {
328                         rc = ISCSI_ERR_XMIT_FAILED;
329                         goto error;
330                 } else if (rc == 0)
331                         break;
332
333                 consumed += rc;
334
335                 if (segment->total_copied >= segment->total_size) {
336                         if (segment->done != NULL) {
337                                 rc = segment->done(tcp_conn, segment);
338                                 if (rc != 0)
339                                         goto error;
340                         }
341                 }
342         }
343
344         ISCSI_SW_TCP_DBG(conn, "xmit %d bytes\n", consumed);
345
346         conn->txdata_octets += consumed;
347         return consumed;
348
349 error:
350         /* Transmit error. We could initiate error recovery
351          * here. */
352         ISCSI_SW_TCP_DBG(conn, "Error sending PDU, errno=%d\n", rc);
353         iscsi_conn_failure(conn, rc);
354         return -EIO;
355 }
356
357 /**
358  * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
359  */
360 static inline int iscsi_sw_tcp_xmit_qlen(struct iscsi_conn *conn)
361 {
362         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
363         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
364         struct iscsi_segment *segment = &tcp_sw_conn->out.segment;
365
366         return segment->total_copied - segment->total_size;
367 }
368
369 static int iscsi_sw_tcp_pdu_xmit(struct iscsi_task *task)
370 {
371         struct iscsi_conn *conn = task->conn;
372         int rc;
373
374         while (iscsi_sw_tcp_xmit_qlen(conn)) {
375                 rc = iscsi_sw_tcp_xmit(conn);
376                 if (rc == 0)
377                         return -EAGAIN;
378                 if (rc < 0)
379                         return rc;
380         }
381
382         return 0;
383 }
384
385 /*
386  * This is called when we're done sending the header.
387  * Simply copy the data_segment to the send segment, and return.
388  */
389 static int iscsi_sw_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
390                                       struct iscsi_segment *segment)
391 {
392         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
393
394         tcp_sw_conn->out.segment = tcp_sw_conn->out.data_segment;
395         ISCSI_SW_TCP_DBG(tcp_conn->iscsi_conn,
396                          "Header done. Next segment size %u total_size %u\n",
397                          tcp_sw_conn->out.segment.size,
398                          tcp_sw_conn->out.segment.total_size);
399         return 0;
400 }
401
402 static void iscsi_sw_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr,
403                                        size_t hdrlen)
404 {
405         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
406         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
407
408         ISCSI_SW_TCP_DBG(conn, "%s\n", conn->hdrdgst_en ?
409                          "digest enabled" : "digest disabled");
410
411         /* Clear the data segment - needs to be filled in by the
412          * caller using iscsi_tcp_send_data_prep() */
413         memset(&tcp_sw_conn->out.data_segment, 0,
414                sizeof(struct iscsi_segment));
415
416         /* If header digest is enabled, compute the CRC and
417          * place the digest into the same buffer. We make
418          * sure that both iscsi_tcp_task and mtask have
419          * sufficient room.
420          */
421         if (conn->hdrdgst_en) {
422                 iscsi_tcp_dgst_header(&tcp_sw_conn->tx_hash, hdr, hdrlen,
423                                       hdr + hdrlen);
424                 hdrlen += ISCSI_DIGEST_SIZE;
425         }
426
427         /* Remember header pointer for later, when we need
428          * to decide whether there's a payload to go along
429          * with the header. */
430         tcp_sw_conn->out.hdr = hdr;
431
432         iscsi_segment_init_linear(&tcp_sw_conn->out.segment, hdr, hdrlen,
433                                   iscsi_sw_tcp_send_hdr_done, NULL);
434 }
435
436 /*
437  * Prepare the send buffer for the payload data.
438  * Padding and checksumming will all be taken care
439  * of by the iscsi_segment routines.
440  */
441 static int
442 iscsi_sw_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
443                             unsigned int count, unsigned int offset,
444                             unsigned int len)
445 {
446         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
447         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
448         struct hash_desc *tx_hash = NULL;
449         unsigned int hdr_spec_len;
450
451         ISCSI_SW_TCP_DBG(conn, "offset=%d, datalen=%d %s\n", offset, len,
452                          conn->datadgst_en ?
453                          "digest enabled" : "digest disabled");
454
455         /* Make sure the datalen matches what the caller
456            said he would send. */
457         hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
458         WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
459
460         if (conn->datadgst_en)
461                 tx_hash = &tcp_sw_conn->tx_hash;
462
463         return iscsi_segment_seek_sg(&tcp_sw_conn->out.data_segment,
464                                      sg, count, offset, len,
465                                      NULL, tx_hash);
466 }
467
468 static void
469 iscsi_sw_tcp_send_linear_data_prep(struct iscsi_conn *conn, void *data,
470                                    size_t len)
471 {
472         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
473         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
474         struct hash_desc *tx_hash = NULL;
475         unsigned int hdr_spec_len;
476
477         ISCSI_SW_TCP_DBG(conn, "datalen=%zd %s\n", len, conn->datadgst_en ?
478                          "digest enabled" : "digest disabled");
479
480         /* Make sure the datalen matches what the caller
481            said he would send. */
482         hdr_spec_len = ntoh24(tcp_sw_conn->out.hdr->dlength);
483         WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
484
485         if (conn->datadgst_en)
486                 tx_hash = &tcp_sw_conn->tx_hash;
487
488         iscsi_segment_init_linear(&tcp_sw_conn->out.data_segment,
489                                 data, len, NULL, tx_hash);
490 }
491
492 static int iscsi_sw_tcp_pdu_init(struct iscsi_task *task,
493                                  unsigned int offset, unsigned int count)
494 {
495         struct iscsi_conn *conn = task->conn;
496         int err = 0;
497
498         iscsi_sw_tcp_send_hdr_prep(conn, task->hdr, task->hdr_len);
499
500         if (!count)
501                 return 0;
502
503         if (!task->sc)
504                 iscsi_sw_tcp_send_linear_data_prep(conn, task->data, count);
505         else {
506                 struct scsi_data_buffer *sdb = scsi_out(task->sc);
507
508                 err = iscsi_sw_tcp_send_data_prep(conn, sdb->table.sgl,
509                                                   sdb->table.nents, offset,
510                                                   count);
511         }
512
513         if (err) {
514                 /* got invalid offset/len */
515                 return -EIO;
516         }
517         return 0;
518 }
519
520 static int iscsi_sw_tcp_pdu_alloc(struct iscsi_task *task, uint8_t opcode)
521 {
522         struct iscsi_tcp_task *tcp_task = task->dd_data;
523
524         task->hdr = task->dd_data + sizeof(*tcp_task);
525         task->hdr_max = sizeof(struct iscsi_sw_tcp_hdrbuf) - ISCSI_DIGEST_SIZE;
526         return 0;
527 }
528
529 static struct iscsi_cls_conn *
530 iscsi_sw_tcp_conn_create(struct iscsi_cls_session *cls_session,
531                          uint32_t conn_idx)
532 {
533         struct iscsi_conn *conn;
534         struct iscsi_cls_conn *cls_conn;
535         struct iscsi_tcp_conn *tcp_conn;
536         struct iscsi_sw_tcp_conn *tcp_sw_conn;
537
538         cls_conn = iscsi_tcp_conn_setup(cls_session, sizeof(*tcp_sw_conn),
539                                         conn_idx);
540         if (!cls_conn)
541                 return NULL;
542         conn = cls_conn->dd_data;
543         tcp_conn = conn->dd_data;
544         tcp_sw_conn = tcp_conn->dd_data;
545
546         tcp_sw_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
547                                                      CRYPTO_ALG_ASYNC);
548         tcp_sw_conn->tx_hash.flags = 0;
549         if (IS_ERR(tcp_sw_conn->tx_hash.tfm))
550                 goto free_conn;
551
552         tcp_sw_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
553                                                      CRYPTO_ALG_ASYNC);
554         tcp_sw_conn->rx_hash.flags = 0;
555         if (IS_ERR(tcp_sw_conn->rx_hash.tfm))
556                 goto free_tx_tfm;
557         tcp_conn->rx_hash = &tcp_sw_conn->rx_hash;
558
559         return cls_conn;
560
561 free_tx_tfm:
562         crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
563 free_conn:
564         iscsi_conn_printk(KERN_ERR, conn,
565                           "Could not create connection due to crc32c "
566                           "loading error. Make sure the crc32c "
567                           "module is built as a module or into the "
568                           "kernel\n");
569         iscsi_tcp_conn_teardown(cls_conn);
570         return NULL;
571 }
572
573 static void iscsi_sw_tcp_release_conn(struct iscsi_conn *conn)
574 {
575         struct iscsi_session *session = conn->session;
576         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
577         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
578         struct socket *sock = tcp_sw_conn->sock;
579
580         if (!sock)
581                 return;
582
583         sock_hold(sock->sk);
584         iscsi_sw_tcp_conn_restore_callbacks(conn);
585         sock_put(sock->sk);
586
587         spin_lock_bh(&session->lock);
588         tcp_sw_conn->sock = NULL;
589         spin_unlock_bh(&session->lock);
590         sockfd_put(sock);
591 }
592
593 static void iscsi_sw_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
594 {
595         struct iscsi_conn *conn = cls_conn->dd_data;
596         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
597         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
598
599         iscsi_sw_tcp_release_conn(conn);
600
601         if (tcp_sw_conn->tx_hash.tfm)
602                 crypto_free_hash(tcp_sw_conn->tx_hash.tfm);
603         if (tcp_sw_conn->rx_hash.tfm)
604                 crypto_free_hash(tcp_sw_conn->rx_hash.tfm);
605
606         iscsi_tcp_conn_teardown(cls_conn);
607 }
608
609 static void iscsi_sw_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
610 {
611         struct iscsi_conn *conn = cls_conn->dd_data;
612         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
613         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
614         struct socket *sock = tcp_sw_conn->sock;
615
616         /* userspace may have goofed up and not bound us */
617         if (!sock)
618                 return;
619
620         sock->sk->sk_err = EIO;
621         wake_up_interruptible(sk_sleep(sock->sk));
622
623         /* stop xmit side */
624         iscsi_suspend_tx(conn);
625
626         /* stop recv side and release socket */
627         iscsi_sw_tcp_release_conn(conn);
628
629         iscsi_conn_stop(cls_conn, flag);
630 }
631
632 static int
633 iscsi_sw_tcp_conn_bind(struct iscsi_cls_session *cls_session,
634                        struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
635                        int is_leading)
636 {
637         struct iscsi_session *session = cls_session->dd_data;
638         struct iscsi_conn *conn = cls_conn->dd_data;
639         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
640         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
641         struct sock *sk;
642         struct socket *sock;
643         int err;
644
645         /* lookup for existing socket */
646         sock = sockfd_lookup((int)transport_eph, &err);
647         if (!sock) {
648                 iscsi_conn_printk(KERN_ERR, conn,
649                                   "sockfd_lookup failed %d\n", err);
650                 return -EEXIST;
651         }
652
653         err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
654         if (err)
655                 goto free_socket;
656
657         spin_lock_bh(&session->lock);
658         /* bind iSCSI connection and socket */
659         tcp_sw_conn->sock = sock;
660         spin_unlock_bh(&session->lock);
661
662         /* setup Socket parameters */
663         sk = sock->sk;
664         sk->sk_reuse = 1;
665         sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
666         sk->sk_allocation = GFP_ATOMIC;
667
668         iscsi_sw_tcp_conn_set_callbacks(conn);
669         tcp_sw_conn->sendpage = tcp_sw_conn->sock->ops->sendpage;
670         /*
671          * set receive state machine into initial state
672          */
673         iscsi_tcp_hdr_recv_prep(tcp_conn);
674         return 0;
675
676 free_socket:
677         sockfd_put(sock);
678         return err;
679 }
680
681 static int iscsi_sw_tcp_conn_set_param(struct iscsi_cls_conn *cls_conn,
682                                        enum iscsi_param param, char *buf,
683                                        int buflen)
684 {
685         struct iscsi_conn *conn = cls_conn->dd_data;
686         struct iscsi_session *session = conn->session;
687         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
688         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
689         int value;
690
691         switch(param) {
692         case ISCSI_PARAM_HDRDGST_EN:
693                 iscsi_set_param(cls_conn, param, buf, buflen);
694                 break;
695         case ISCSI_PARAM_DATADGST_EN:
696                 iscsi_set_param(cls_conn, param, buf, buflen);
697                 tcp_sw_conn->sendpage = conn->datadgst_en ?
698                         sock_no_sendpage : tcp_sw_conn->sock->ops->sendpage;
699                 break;
700         case ISCSI_PARAM_MAX_R2T:
701                 sscanf(buf, "%d", &value);
702                 if (value <= 0 || !is_power_of_2(value))
703                         return -EINVAL;
704                 if (session->max_r2t == value)
705                         break;
706                 iscsi_tcp_r2tpool_free(session);
707                 iscsi_set_param(cls_conn, param, buf, buflen);
708                 if (iscsi_tcp_r2tpool_alloc(session))
709                         return -ENOMEM;
710                 break;
711         default:
712                 return iscsi_set_param(cls_conn, param, buf, buflen);
713         }
714
715         return 0;
716 }
717
718 static int iscsi_sw_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
719                                        enum iscsi_param param, char *buf)
720 {
721         struct iscsi_conn *conn = cls_conn->dd_data;
722         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
723         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
724         struct sockaddr_in6 addr;
725         int rc, len;
726
727         switch(param) {
728         case ISCSI_PARAM_CONN_PORT:
729         case ISCSI_PARAM_CONN_ADDRESS:
730                 spin_lock_bh(&conn->session->lock);
731                 if (!tcp_sw_conn || !tcp_sw_conn->sock) {
732                         spin_unlock_bh(&conn->session->lock);
733                         return -ENOTCONN;
734                 }
735                 rc = kernel_getpeername(tcp_sw_conn->sock,
736                                         (struct sockaddr *)&addr, &len);
737                 spin_unlock_bh(&conn->session->lock);
738                 if (rc)
739                         return rc;
740
741                 return iscsi_conn_get_addr_param((struct sockaddr_storage *)
742                                                  &addr, param, buf);
743         default:
744                 return iscsi_conn_get_param(cls_conn, param, buf);
745         }
746
747         return 0;
748 }
749
750 static int iscsi_sw_tcp_host_get_param(struct Scsi_Host *shost,
751                                        enum iscsi_host_param param, char *buf)
752 {
753         struct iscsi_sw_tcp_host *tcp_sw_host = iscsi_host_priv(shost);
754         struct iscsi_session *session = tcp_sw_host->session;
755         struct iscsi_conn *conn;
756         struct iscsi_tcp_conn *tcp_conn;
757         struct iscsi_sw_tcp_conn *tcp_sw_conn;
758         struct sockaddr_in6 addr;
759         int rc, len;
760
761         switch (param) {
762         case ISCSI_HOST_PARAM_IPADDRESS:
763                 spin_lock_bh(&session->lock);
764                 conn = session->leadconn;
765                 if (!conn) {
766                         spin_unlock_bh(&session->lock);
767                         return -ENOTCONN;
768                 }
769                 tcp_conn = conn->dd_data;
770
771                 tcp_sw_conn = tcp_conn->dd_data;
772                 if (!tcp_sw_conn->sock) {
773                         spin_unlock_bh(&session->lock);
774                         return -ENOTCONN;
775                 }
776
777                 rc = kernel_getsockname(tcp_sw_conn->sock,
778                                         (struct sockaddr *)&addr, &len);
779                 spin_unlock_bh(&session->lock);
780                 if (rc)
781                         return rc;
782
783                 return iscsi_conn_get_addr_param((struct sockaddr_storage *)
784                                                  &addr, param, buf);
785         default:
786                 return iscsi_host_get_param(shost, param, buf);
787         }
788
789         return 0;
790 }
791
792 static void
793 iscsi_sw_tcp_conn_get_stats(struct iscsi_cls_conn *cls_conn,
794                             struct iscsi_stats *stats)
795 {
796         struct iscsi_conn *conn = cls_conn->dd_data;
797         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
798         struct iscsi_sw_tcp_conn *tcp_sw_conn = tcp_conn->dd_data;
799
800         stats->custom_length = 3;
801         strcpy(stats->custom[0].desc, "tx_sendpage_failures");
802         stats->custom[0].value = tcp_sw_conn->sendpage_failures_cnt;
803         strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
804         stats->custom[1].value = tcp_sw_conn->discontiguous_hdr_cnt;
805         strcpy(stats->custom[2].desc, "eh_abort_cnt");
806         stats->custom[2].value = conn->eh_abort_cnt;
807
808         iscsi_tcp_conn_get_stats(cls_conn, stats);
809 }
810
811 static struct iscsi_cls_session *
812 iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,
813                             uint16_t qdepth, uint32_t initial_cmdsn)
814 {
815         struct iscsi_cls_session *cls_session;
816         struct iscsi_session *session;
817         struct iscsi_sw_tcp_host *tcp_sw_host;
818         struct Scsi_Host *shost;
819
820         if (ep) {
821                 printk(KERN_ERR "iscsi_tcp: invalid ep %p.\n", ep);
822                 return NULL;
823         }
824
825         shost = iscsi_host_alloc(&iscsi_sw_tcp_sht,
826                                  sizeof(struct iscsi_sw_tcp_host), 1);
827         if (!shost)
828                 return NULL;
829         shost->transportt = iscsi_sw_tcp_scsi_transport;
830         shost->cmd_per_lun = qdepth;
831         shost->max_lun = iscsi_max_lun;
832         shost->max_id = 0;
833         shost->max_channel = 0;
834         shost->max_cmd_len = SCSI_MAX_VARLEN_CDB_SIZE;
835
836         if (iscsi_host_add(shost, NULL))
837                 goto free_host;
838
839         cls_session = iscsi_session_setup(&iscsi_sw_tcp_transport, shost,
840                                           cmds_max, 0,
841                                           sizeof(struct iscsi_tcp_task) +
842                                           sizeof(struct iscsi_sw_tcp_hdrbuf),
843                                           initial_cmdsn, 0);
844         if (!cls_session)
845                 goto remove_host;
846         session = cls_session->dd_data;
847         tcp_sw_host = iscsi_host_priv(shost);
848         tcp_sw_host->session = session;
849
850         shost->can_queue = session->scsi_cmds_max;
851         if (iscsi_tcp_r2tpool_alloc(session))
852                 goto remove_session;
853         return cls_session;
854
855 remove_session:
856         iscsi_session_teardown(cls_session);
857 remove_host:
858         iscsi_host_remove(shost);
859 free_host:
860         iscsi_host_free(shost);
861         return NULL;
862 }
863
864 static void iscsi_sw_tcp_session_destroy(struct iscsi_cls_session *cls_session)
865 {
866         struct Scsi_Host *shost = iscsi_session_to_shost(cls_session);
867
868         iscsi_tcp_r2tpool_free(cls_session->dd_data);
869         iscsi_session_teardown(cls_session);
870
871         iscsi_host_remove(shost);
872         iscsi_host_free(shost);
873 }
874
875 static int iscsi_sw_tcp_slave_alloc(struct scsi_device *sdev)
876 {
877         set_bit(QUEUE_FLAG_BIDI, &sdev->request_queue->queue_flags);
878         return 0;
879 }
880
881 static int iscsi_sw_tcp_slave_configure(struct scsi_device *sdev)
882 {
883         blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
884         blk_queue_dma_alignment(sdev->request_queue, 0);
885         return 0;
886 }
887
888 static struct scsi_host_template iscsi_sw_tcp_sht = {
889         .module                 = THIS_MODULE,
890         .name                   = "iSCSI Initiator over TCP/IP",
891         .queuecommand           = iscsi_queuecommand,
892         .change_queue_depth     = iscsi_change_queue_depth,
893         .can_queue              = ISCSI_DEF_XMIT_CMDS_MAX - 1,
894         .sg_tablesize           = 4096,
895         .max_sectors            = 0xFFFF,
896         .cmd_per_lun            = ISCSI_DEF_CMD_PER_LUN,
897         .eh_abort_handler       = iscsi_eh_abort,
898         .eh_device_reset_handler= iscsi_eh_device_reset,
899         .eh_target_reset_handler = iscsi_eh_recover_target,
900         .use_clustering         = DISABLE_CLUSTERING,
901         .slave_alloc            = iscsi_sw_tcp_slave_alloc,
902         .slave_configure        = iscsi_sw_tcp_slave_configure,
903         .target_alloc           = iscsi_target_alloc,
904         .proc_name              = "iscsi_tcp",
905         .this_id                = -1,
906 };
907
908 static struct iscsi_transport iscsi_sw_tcp_transport = {
909         .owner                  = THIS_MODULE,
910         .name                   = "tcp",
911         .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
912                                   | CAP_DATADGST,
913         .param_mask             = ISCSI_MAX_RECV_DLENGTH |
914                                   ISCSI_MAX_XMIT_DLENGTH |
915                                   ISCSI_HDRDGST_EN |
916                                   ISCSI_DATADGST_EN |
917                                   ISCSI_INITIAL_R2T_EN |
918                                   ISCSI_MAX_R2T |
919                                   ISCSI_IMM_DATA_EN |
920                                   ISCSI_FIRST_BURST |
921                                   ISCSI_MAX_BURST |
922                                   ISCSI_PDU_INORDER_EN |
923                                   ISCSI_DATASEQ_INORDER_EN |
924                                   ISCSI_ERL |
925                                   ISCSI_CONN_PORT |
926                                   ISCSI_CONN_ADDRESS |
927                                   ISCSI_EXP_STATSN |
928                                   ISCSI_PERSISTENT_PORT |
929                                   ISCSI_PERSISTENT_ADDRESS |
930                                   ISCSI_TARGET_NAME | ISCSI_TPGT |
931                                   ISCSI_USERNAME | ISCSI_PASSWORD |
932                                   ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
933                                   ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
934                                   ISCSI_LU_RESET_TMO | ISCSI_TGT_RESET_TMO |
935                                   ISCSI_PING_TMO | ISCSI_RECV_TMO |
936                                   ISCSI_IFACE_NAME | ISCSI_INITIATOR_NAME,
937         .host_param_mask        = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
938                                   ISCSI_HOST_INITIATOR_NAME |
939                                   ISCSI_HOST_NETDEV_NAME,
940         /* session management */
941         .create_session         = iscsi_sw_tcp_session_create,
942         .destroy_session        = iscsi_sw_tcp_session_destroy,
943         /* connection management */
944         .create_conn            = iscsi_sw_tcp_conn_create,
945         .bind_conn              = iscsi_sw_tcp_conn_bind,
946         .destroy_conn           = iscsi_sw_tcp_conn_destroy,
947         .set_param              = iscsi_sw_tcp_conn_set_param,
948         .get_conn_param         = iscsi_sw_tcp_conn_get_param,
949         .get_session_param      = iscsi_session_get_param,
950         .start_conn             = iscsi_conn_start,
951         .stop_conn              = iscsi_sw_tcp_conn_stop,
952         /* iscsi host params */
953         .get_host_param         = iscsi_sw_tcp_host_get_param,
954         .set_host_param         = iscsi_host_set_param,
955         /* IO */
956         .send_pdu               = iscsi_conn_send_pdu,
957         .get_stats              = iscsi_sw_tcp_conn_get_stats,
958         /* iscsi task/cmd helpers */
959         .init_task              = iscsi_tcp_task_init,
960         .xmit_task              = iscsi_tcp_task_xmit,
961         .cleanup_task           = iscsi_tcp_cleanup_task,
962         /* low level pdu helpers */
963         .xmit_pdu               = iscsi_sw_tcp_pdu_xmit,
964         .init_pdu               = iscsi_sw_tcp_pdu_init,
965         .alloc_pdu              = iscsi_sw_tcp_pdu_alloc,
966         /* recovery */
967         .session_recovery_timedout = iscsi_session_recovery_timedout,
968 };
969
970 static int __init iscsi_sw_tcp_init(void)
971 {
972         if (iscsi_max_lun < 1) {
973                 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
974                        iscsi_max_lun);
975                 return -EINVAL;
976         }
977
978         iscsi_sw_tcp_scsi_transport = iscsi_register_transport(
979                                                 &iscsi_sw_tcp_transport);
980         if (!iscsi_sw_tcp_scsi_transport)
981                 return -ENODEV;
982
983         return 0;
984 }
985
986 static void __exit iscsi_sw_tcp_exit(void)
987 {
988         iscsi_unregister_transport(&iscsi_sw_tcp_transport);
989 }
990
991 module_init(iscsi_sw_tcp_init);
992 module_exit(iscsi_sw_tcp_exit);