[SCSI] iscsi_tcp: hold lock during data rsp processing
[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/list.h>
31 #include <linux/inet.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("Dmitry Yusupov <dmitry_yus@yahoo.com>, "
48               "Alex Aizman <itn780@yahoo.com>");
49 MODULE_DESCRIPTION("iSCSI/TCP data-path");
50 MODULE_LICENSE("GPL");
51 #undef DEBUG_TCP
52 #define DEBUG_ASSERT
53
54 #ifdef DEBUG_TCP
55 #define debug_tcp(fmt...) printk(KERN_INFO "tcp: " fmt)
56 #else
57 #define debug_tcp(fmt...)
58 #endif
59
60 #ifndef DEBUG_ASSERT
61 #ifdef BUG_ON
62 #undef BUG_ON
63 #endif
64 #define BUG_ON(expr)
65 #endif
66
67 static unsigned int iscsi_max_lun = 512;
68 module_param_named(max_lun, iscsi_max_lun, uint, S_IRUGO);
69
70 static int iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
71                                    struct iscsi_segment *segment);
72
73 /*
74  * Scatterlist handling: inside the iscsi_segment, we
75  * remember an index into the scatterlist, and set data/size
76  * to the current scatterlist entry. For highmem pages, we
77  * kmap as needed.
78  *
79  * Note that the page is unmapped when we return from
80  * TCP's data_ready handler, so we may end up mapping and
81  * unmapping the same page repeatedly. The whole reason
82  * for this is that we shouldn't keep the page mapped
83  * outside the softirq.
84  */
85
86 /**
87  * iscsi_tcp_segment_init_sg - init indicated scatterlist entry
88  * @segment: the buffer object
89  * @sg: scatterlist
90  * @offset: byte offset into that sg entry
91  *
92  * This function sets up the segment so that subsequent
93  * data is copied to the indicated sg entry, at the given
94  * offset.
95  */
96 static inline void
97 iscsi_tcp_segment_init_sg(struct iscsi_segment *segment,
98                           struct scatterlist *sg, unsigned int offset)
99 {
100         segment->sg = sg;
101         segment->sg_offset = offset;
102         segment->size = min(sg->length - offset,
103                             segment->total_size - segment->total_copied);
104         segment->data = NULL;
105 }
106
107 /**
108  * iscsi_tcp_segment_map - map the current S/G page
109  * @segment: iscsi_segment
110  * @recv: 1 if called from recv path
111  *
112  * We only need to possibly kmap data if scatter lists are being used,
113  * because the iscsi passthrough and internal IO paths will never use high
114  * mem pages.
115  */
116 static inline void
117 iscsi_tcp_segment_map(struct iscsi_segment *segment, int recv)
118 {
119         struct scatterlist *sg;
120
121         if (segment->data != NULL || !segment->sg)
122                 return;
123
124         sg = segment->sg;
125         BUG_ON(segment->sg_mapped);
126         BUG_ON(sg->length == 0);
127
128         /*
129          * If the page count is greater than one it is ok to send
130          * to the network layer's zero copy send path. If not we
131          * have to go the slow sendmsg path. We always map for the
132          * recv path.
133          */
134         if (page_count(sg_page(sg)) >= 1 && !recv)
135                 return;
136
137         debug_tcp("iscsi_tcp_segment_map %s %p\n", recv ? "recv" : "xmit",
138                   segment);
139         segment->sg_mapped = kmap_atomic(sg_page(sg), KM_SOFTIRQ0);
140         segment->data = segment->sg_mapped + sg->offset + segment->sg_offset;
141 }
142
143 static inline void
144 iscsi_tcp_segment_unmap(struct iscsi_segment *segment)
145 {
146         debug_tcp("iscsi_tcp_segment_unmap %p\n", segment);
147
148         if (segment->sg_mapped) {
149                 debug_tcp("iscsi_tcp_segment_unmap valid\n");
150                 kunmap_atomic(segment->sg_mapped, KM_SOFTIRQ0);
151                 segment->sg_mapped = NULL;
152                 segment->data = NULL;
153         }
154 }
155
156 /*
157  * Splice the digest buffer into the buffer
158  */
159 static inline void
160 iscsi_tcp_segment_splice_digest(struct iscsi_segment *segment, void *digest)
161 {
162         segment->data = digest;
163         segment->digest_len = ISCSI_DIGEST_SIZE;
164         segment->total_size += ISCSI_DIGEST_SIZE;
165         segment->size = ISCSI_DIGEST_SIZE;
166         segment->copied = 0;
167         segment->sg = NULL;
168         segment->hash = NULL;
169 }
170
171 /**
172  * iscsi_tcp_segment_done - check whether the segment is complete
173  * @segment: iscsi segment to check
174  * @recv: set to one of this is called from the recv path
175  * @copied: number of bytes copied
176  *
177  * Check if we're done receiving this segment. If the receive
178  * buffer is full but we expect more data, move on to the
179  * next entry in the scatterlist.
180  *
181  * If the amount of data we received isn't a multiple of 4,
182  * we will transparently receive the pad bytes, too.
183  *
184  * This function must be re-entrant.
185  */
186 static inline int
187 iscsi_tcp_segment_done(struct iscsi_segment *segment, int recv, unsigned copied)
188 {
189         static unsigned char padbuf[ISCSI_PAD_LEN];
190         struct scatterlist sg;
191         unsigned int pad;
192
193         debug_tcp("copied %u %u size %u %s\n", segment->copied, copied,
194                   segment->size, recv ? "recv" : "xmit");
195         if (segment->hash && copied) {
196                 /*
197                  * If a segment is kmapd we must unmap it before sending
198                  * to the crypto layer since that will try to kmap it again.
199                  */
200                 iscsi_tcp_segment_unmap(segment);
201
202                 if (!segment->data) {
203                         sg_init_table(&sg, 1);
204                         sg_set_page(&sg, sg_page(segment->sg), copied,
205                                     segment->copied + segment->sg_offset +
206                                                         segment->sg->offset);
207                 } else
208                         sg_init_one(&sg, segment->data + segment->copied,
209                                     copied);
210                 crypto_hash_update(segment->hash, &sg, copied);
211         }
212
213         segment->copied += copied;
214         if (segment->copied < segment->size) {
215                 iscsi_tcp_segment_map(segment, recv);
216                 return 0;
217         }
218
219         segment->total_copied += segment->copied;
220         segment->copied = 0;
221         segment->size = 0;
222
223         /* Unmap the current scatterlist page, if there is one. */
224         iscsi_tcp_segment_unmap(segment);
225
226         /* Do we have more scatterlist entries? */
227         debug_tcp("total copied %u total size %u\n", segment->total_copied,
228                    segment->total_size);
229         if (segment->total_copied < segment->total_size) {
230                 /* Proceed to the next entry in the scatterlist. */
231                 iscsi_tcp_segment_init_sg(segment, sg_next(segment->sg),
232                                           0);
233                 iscsi_tcp_segment_map(segment, recv);
234                 BUG_ON(segment->size == 0);
235                 return 0;
236         }
237
238         /* Do we need to handle padding? */
239         pad = iscsi_padding(segment->total_copied);
240         if (pad != 0) {
241                 debug_tcp("consume %d pad bytes\n", pad);
242                 segment->total_size += pad;
243                 segment->size = pad;
244                 segment->data = padbuf;
245                 return 0;
246         }
247
248         /*
249          * Set us up for transferring the data digest. hdr digest
250          * is completely handled in hdr done function.
251          */
252         if (segment->hash) {
253                 crypto_hash_final(segment->hash, segment->digest);
254                 iscsi_tcp_segment_splice_digest(segment,
255                                  recv ? segment->recv_digest : segment->digest);
256                 return 0;
257         }
258
259         return 1;
260 }
261
262 /**
263  * iscsi_tcp_xmit_segment - transmit segment
264  * @tcp_conn: the iSCSI TCP connection
265  * @segment: the buffer to transmnit
266  *
267  * This function transmits as much of the buffer as
268  * the network layer will accept, and returns the number of
269  * bytes transmitted.
270  *
271  * If CRC hashing is enabled, the function will compute the
272  * hash as it goes. When the entire segment has been transmitted,
273  * it will retrieve the hash value and send it as well.
274  */
275 static int
276 iscsi_tcp_xmit_segment(struct iscsi_tcp_conn *tcp_conn,
277                        struct iscsi_segment *segment)
278 {
279         struct socket *sk = tcp_conn->sock;
280         unsigned int copied = 0;
281         int r = 0;
282
283         while (!iscsi_tcp_segment_done(segment, 0, r)) {
284                 struct scatterlist *sg;
285                 unsigned int offset, copy;
286                 int flags = 0;
287
288                 r = 0;
289                 offset = segment->copied;
290                 copy = segment->size - offset;
291
292                 if (segment->total_copied + segment->size < segment->total_size)
293                         flags |= MSG_MORE;
294
295                 /* Use sendpage if we can; else fall back to sendmsg */
296                 if (!segment->data) {
297                         sg = segment->sg;
298                         offset += segment->sg_offset + sg->offset;
299                         r = tcp_conn->sendpage(sk, sg_page(sg), offset, copy,
300                                                flags);
301                 } else {
302                         struct msghdr msg = { .msg_flags = flags };
303                         struct kvec iov = {
304                                 .iov_base = segment->data + offset,
305                                 .iov_len = copy
306                         };
307
308                         r = kernel_sendmsg(sk, &msg, &iov, 1, copy);
309                 }
310
311                 if (r < 0) {
312                         iscsi_tcp_segment_unmap(segment);
313                         if (copied || r == -EAGAIN)
314                                 break;
315                         return r;
316                 }
317                 copied += r;
318         }
319         return copied;
320 }
321
322 /**
323  * iscsi_tcp_segment_recv - copy data to segment
324  * @tcp_conn: the iSCSI TCP connection
325  * @segment: the buffer to copy to
326  * @ptr: data pointer
327  * @len: amount of data available
328  *
329  * This function copies up to @len bytes to the
330  * given buffer, and returns the number of bytes
331  * consumed, which can actually be less than @len.
332  *
333  * If hash digest is enabled, the function will update the
334  * hash while copying.
335  * Combining these two operations doesn't buy us a lot (yet),
336  * but in the future we could implement combined copy+crc,
337  * just way we do for network layer checksums.
338  */
339 static int
340 iscsi_tcp_segment_recv(struct iscsi_tcp_conn *tcp_conn,
341                        struct iscsi_segment *segment, const void *ptr,
342                        unsigned int len)
343 {
344         unsigned int copy = 0, copied = 0;
345
346         while (!iscsi_tcp_segment_done(segment, 1, copy)) {
347                 if (copied == len) {
348                         debug_tcp("iscsi_tcp_segment_recv copied %d bytes\n",
349                                   len);
350                         break;
351                 }
352
353                 copy = min(len - copied, segment->size - segment->copied);
354                 debug_tcp("iscsi_tcp_segment_recv copying %d\n", copy);
355                 memcpy(segment->data + segment->copied, ptr + copied, copy);
356                 copied += copy;
357         }
358         return copied;
359 }
360
361 static inline void
362 iscsi_tcp_dgst_header(struct hash_desc *hash, const void *hdr, size_t hdrlen,
363                       unsigned char digest[ISCSI_DIGEST_SIZE])
364 {
365         struct scatterlist sg;
366
367         sg_init_one(&sg, hdr, hdrlen);
368         crypto_hash_digest(hash, &sg, hdrlen, digest);
369 }
370
371 static inline int
372 iscsi_tcp_dgst_verify(struct iscsi_tcp_conn *tcp_conn,
373                       struct iscsi_segment *segment)
374 {
375         if (!segment->digest_len)
376                 return 1;
377
378         if (memcmp(segment->recv_digest, segment->digest,
379                    segment->digest_len)) {
380                 debug_scsi("digest mismatch\n");
381                 return 0;
382         }
383
384         return 1;
385 }
386
387 /*
388  * Helper function to set up segment buffer
389  */
390 static inline void
391 __iscsi_segment_init(struct iscsi_segment *segment, size_t size,
392                      iscsi_segment_done_fn_t *done, struct hash_desc *hash)
393 {
394         memset(segment, 0, sizeof(*segment));
395         segment->total_size = size;
396         segment->done = done;
397
398         if (hash) {
399                 segment->hash = hash;
400                 crypto_hash_init(hash);
401         }
402 }
403
404 static inline void
405 iscsi_segment_init_linear(struct iscsi_segment *segment, void *data,
406                           size_t size, iscsi_segment_done_fn_t *done,
407                           struct hash_desc *hash)
408 {
409         __iscsi_segment_init(segment, size, done, hash);
410         segment->data = data;
411         segment->size = size;
412 }
413
414 static inline int
415 iscsi_segment_seek_sg(struct iscsi_segment *segment,
416                       struct scatterlist *sg_list, unsigned int sg_count,
417                       unsigned int offset, size_t size,
418                       iscsi_segment_done_fn_t *done, struct hash_desc *hash)
419 {
420         struct scatterlist *sg;
421         unsigned int i;
422
423         debug_scsi("iscsi_segment_seek_sg offset %u size %llu\n",
424                   offset, size);
425         __iscsi_segment_init(segment, size, done, hash);
426         for_each_sg(sg_list, sg, sg_count, i) {
427                 debug_scsi("sg %d, len %u offset %u\n", i, sg->length,
428                            sg->offset);
429                 if (offset < sg->length) {
430                         iscsi_tcp_segment_init_sg(segment, sg, offset);
431                         return 0;
432                 }
433                 offset -= sg->length;
434         }
435
436         return ISCSI_ERR_DATA_OFFSET;
437 }
438
439 /**
440  * iscsi_tcp_hdr_recv_prep - prep segment for hdr reception
441  * @tcp_conn: iscsi connection to prep for
442  *
443  * This function always passes NULL for the hash argument, because when this
444  * function is called we do not yet know the final size of the header and want
445  * to delay the digest processing until we know that.
446  */
447 static void
448 iscsi_tcp_hdr_recv_prep(struct iscsi_tcp_conn *tcp_conn)
449 {
450         debug_tcp("iscsi_tcp_hdr_recv_prep(%p%s)\n", tcp_conn,
451                   tcp_conn->iscsi_conn->hdrdgst_en ? ", digest enabled" : "");
452         iscsi_segment_init_linear(&tcp_conn->in.segment,
453                                 tcp_conn->in.hdr_buf, sizeof(struct iscsi_hdr),
454                                 iscsi_tcp_hdr_recv_done, NULL);
455 }
456
457 /*
458  * Handle incoming reply to any other type of command
459  */
460 static int
461 iscsi_tcp_data_recv_done(struct iscsi_tcp_conn *tcp_conn,
462                          struct iscsi_segment *segment)
463 {
464         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
465         int rc = 0;
466
467         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
468                 return ISCSI_ERR_DATA_DGST;
469
470         rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr,
471                         conn->data, tcp_conn->in.datalen);
472         if (rc)
473                 return rc;
474
475         iscsi_tcp_hdr_recv_prep(tcp_conn);
476         return 0;
477 }
478
479 static void
480 iscsi_tcp_data_recv_prep(struct iscsi_tcp_conn *tcp_conn)
481 {
482         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
483         struct hash_desc *rx_hash = NULL;
484
485         if (conn->datadgst_en)
486                 rx_hash = &tcp_conn->rx_hash;
487
488         iscsi_segment_init_linear(&tcp_conn->in.segment,
489                                 conn->data, tcp_conn->in.datalen,
490                                 iscsi_tcp_data_recv_done, rx_hash);
491 }
492
493 /*
494  * must be called with session lock
495  */
496 static void
497 iscsi_tcp_cleanup_ctask(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
498 {
499         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
500         struct iscsi_r2t_info *r2t;
501
502         /* flush ctask's r2t queues */
503         while (__kfifo_get(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*))) {
504                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
505                             sizeof(void*));
506                 debug_scsi("iscsi_tcp_cleanup_ctask pending r2t dropped\n");
507         }
508
509         r2t = tcp_ctask->r2t;
510         if (r2t != NULL) {
511                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
512                             sizeof(void*));
513                 tcp_ctask->r2t = NULL;
514         }
515 }
516
517 /**
518  * iscsi_data_rsp - SCSI Data-In Response processing
519  * @conn: iscsi connection
520  * @ctask: scsi command task
521  **/
522 static int
523 iscsi_data_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
524 {
525         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
526         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
527         struct iscsi_data_rsp *rhdr = (struct iscsi_data_rsp *)tcp_conn->in.hdr;
528         struct iscsi_session *session = conn->session;
529         struct scsi_cmnd *sc = ctask->sc;
530         int datasn = be32_to_cpu(rhdr->datasn);
531
532         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
533         if (tcp_conn->in.datalen == 0)
534                 return 0;
535
536         if (tcp_ctask->exp_datasn != datasn) {
537                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->datasn(%d)\n",
538                           __FUNCTION__, tcp_ctask->exp_datasn, datasn);
539                 return ISCSI_ERR_DATASN;
540         }
541
542         tcp_ctask->exp_datasn++;
543
544         tcp_ctask->data_offset = be32_to_cpu(rhdr->offset);
545         if (tcp_ctask->data_offset + tcp_conn->in.datalen > scsi_bufflen(sc)) {
546                 debug_tcp("%s: data_offset(%d) + data_len(%d) > total_length_in(%d)\n",
547                           __FUNCTION__, tcp_ctask->data_offset,
548                           tcp_conn->in.datalen, scsi_bufflen(sc));
549                 return ISCSI_ERR_DATA_OFFSET;
550         }
551
552         if (rhdr->flags & ISCSI_FLAG_DATA_STATUS) {
553                 sc->result = (DID_OK << 16) | rhdr->cmd_status;
554                 conn->exp_statsn = be32_to_cpu(rhdr->statsn) + 1;
555                 if (rhdr->flags & (ISCSI_FLAG_DATA_UNDERFLOW |
556                                    ISCSI_FLAG_DATA_OVERFLOW)) {
557                         int res_count = be32_to_cpu(rhdr->residual_count);
558
559                         if (res_count > 0 &&
560                             (rhdr->flags & ISCSI_FLAG_CMD_OVERFLOW ||
561                              res_count <= scsi_bufflen(sc)))
562                                 scsi_set_resid(sc, res_count);
563                         else
564                                 sc->result = (DID_BAD_TARGET << 16) |
565                                         rhdr->cmd_status;
566                 }
567         }
568
569         conn->datain_pdus_cnt++;
570         return 0;
571 }
572
573 /**
574  * iscsi_solicit_data_init - initialize first Data-Out
575  * @conn: iscsi connection
576  * @ctask: scsi command task
577  * @r2t: R2T info
578  *
579  * Notes:
580  *      Initialize first Data-Out within this R2T sequence and finds
581  *      proper data_offset within this SCSI command.
582  *
583  *      This function is called with connection lock taken.
584  **/
585 static void
586 iscsi_solicit_data_init(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
587                         struct iscsi_r2t_info *r2t)
588 {
589         struct iscsi_data *hdr;
590
591         hdr = &r2t->dtask.hdr;
592         memset(hdr, 0, sizeof(struct iscsi_data));
593         hdr->ttt = r2t->ttt;
594         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
595         r2t->solicit_datasn++;
596         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
597         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
598         hdr->itt = ctask->hdr->itt;
599         hdr->exp_statsn = r2t->exp_statsn;
600         hdr->offset = cpu_to_be32(r2t->data_offset);
601         if (r2t->data_length > conn->max_xmit_dlength) {
602                 hton24(hdr->dlength, conn->max_xmit_dlength);
603                 r2t->data_count = conn->max_xmit_dlength;
604                 hdr->flags = 0;
605         } else {
606                 hton24(hdr->dlength, r2t->data_length);
607                 r2t->data_count = r2t->data_length;
608                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
609         }
610         conn->dataout_pdus_cnt++;
611
612         r2t->sent = 0;
613 }
614
615 /**
616  * iscsi_r2t_rsp - iSCSI R2T Response processing
617  * @conn: iscsi connection
618  * @ctask: scsi command task
619  **/
620 static int
621 iscsi_r2t_rsp(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
622 {
623         struct iscsi_r2t_info *r2t;
624         struct iscsi_session *session = conn->session;
625         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
626         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
627         struct iscsi_r2t_rsp *rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr;
628         int r2tsn = be32_to_cpu(rhdr->r2tsn);
629         int rc;
630
631         if (tcp_conn->in.datalen) {
632                 printk(KERN_ERR "iscsi_tcp: invalid R2t with datalen %d\n",
633                        tcp_conn->in.datalen);
634                 return ISCSI_ERR_DATALEN;
635         }
636
637         if (tcp_ctask->exp_datasn != r2tsn){
638                 debug_tcp("%s: ctask->exp_datasn(%d) != rhdr->r2tsn(%d)\n",
639                           __FUNCTION__, tcp_ctask->exp_datasn, r2tsn);
640                 return ISCSI_ERR_R2TSN;
641         }
642
643         /* fill-in new R2T associated with the task */
644         iscsi_update_cmdsn(session, (struct iscsi_nopin*)rhdr);
645
646         if (!ctask->sc || session->state != ISCSI_STATE_LOGGED_IN) {
647                 printk(KERN_INFO "iscsi_tcp: dropping R2T itt %d in "
648                        "recovery...\n", ctask->itt);
649                 return 0;
650         }
651
652         rc = __kfifo_get(tcp_ctask->r2tpool.queue, (void*)&r2t, sizeof(void*));
653         BUG_ON(!rc);
654
655         r2t->exp_statsn = rhdr->statsn;
656         r2t->data_length = be32_to_cpu(rhdr->data_length);
657         if (r2t->data_length == 0) {
658                 printk(KERN_ERR "iscsi_tcp: invalid R2T with zero data len\n");
659                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
660                             sizeof(void*));
661                 return ISCSI_ERR_DATALEN;
662         }
663
664         if (r2t->data_length > session->max_burst)
665                 debug_scsi("invalid R2T with data len %u and max burst %u."
666                            "Attempting to execute request.\n",
667                             r2t->data_length, session->max_burst);
668
669         r2t->data_offset = be32_to_cpu(rhdr->data_offset);
670         if (r2t->data_offset + r2t->data_length > scsi_bufflen(ctask->sc)) {
671                 printk(KERN_ERR "iscsi_tcp: invalid R2T with data len %u at "
672                        "offset %u and total length %d\n", r2t->data_length,
673                        r2t->data_offset, scsi_bufflen(ctask->sc));
674                 __kfifo_put(tcp_ctask->r2tpool.queue, (void*)&r2t,
675                             sizeof(void*));
676                 return ISCSI_ERR_DATALEN;
677         }
678
679         r2t->ttt = rhdr->ttt; /* no flip */
680         r2t->solicit_datasn = 0;
681
682         iscsi_solicit_data_init(conn, ctask, r2t);
683
684         tcp_ctask->exp_datasn = r2tsn + 1;
685         __kfifo_put(tcp_ctask->r2tqueue, (void*)&r2t, sizeof(void*));
686         conn->r2t_pdus_cnt++;
687
688         iscsi_requeue_ctask(ctask);
689         return 0;
690 }
691
692 /*
693  * Handle incoming reply to DataIn command
694  */
695 static int
696 iscsi_tcp_process_data_in(struct iscsi_tcp_conn *tcp_conn,
697                           struct iscsi_segment *segment)
698 {
699         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
700         struct iscsi_hdr *hdr = tcp_conn->in.hdr;
701         int rc;
702
703         if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
704                 return ISCSI_ERR_DATA_DGST;
705
706         /* check for non-exceptional status */
707         if (hdr->flags & ISCSI_FLAG_DATA_STATUS) {
708                 rc = iscsi_complete_pdu(conn, tcp_conn->in.hdr, NULL, 0);
709                 if (rc)
710                         return rc;
711         }
712
713         iscsi_tcp_hdr_recv_prep(tcp_conn);
714         return 0;
715 }
716
717 /**
718  * iscsi_tcp_hdr_dissect - process PDU header
719  * @conn: iSCSI connection
720  * @hdr: PDU header
721  *
722  * This function analyzes the header of the PDU received,
723  * and performs several sanity checks. If the PDU is accompanied
724  * by data, the receive buffer is set up to copy the incoming data
725  * to the correct location.
726  */
727 static int
728 iscsi_tcp_hdr_dissect(struct iscsi_conn *conn, struct iscsi_hdr *hdr)
729 {
730         int rc = 0, opcode, ahslen;
731         struct iscsi_session *session = conn->session;
732         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
733         struct iscsi_cmd_task *ctask;
734         uint32_t itt;
735
736         /* verify PDU length */
737         tcp_conn->in.datalen = ntoh24(hdr->dlength);
738         if (tcp_conn->in.datalen > conn->max_recv_dlength) {
739                 printk(KERN_ERR "iscsi_tcp: datalen %d > %d\n",
740                        tcp_conn->in.datalen, conn->max_recv_dlength);
741                 return ISCSI_ERR_DATALEN;
742         }
743
744         /* Additional header segments. So far, we don't
745          * process additional headers.
746          */
747         ahslen = hdr->hlength << 2;
748
749         opcode = hdr->opcode & ISCSI_OPCODE_MASK;
750         /* verify itt (itt encoding: age+cid+itt) */
751         rc = iscsi_verify_itt(conn, hdr, &itt);
752         if (rc)
753                 return rc;
754
755         debug_tcp("opcode 0x%x ahslen %d datalen %d\n",
756                   opcode, ahslen, tcp_conn->in.datalen);
757
758         switch(opcode) {
759         case ISCSI_OP_SCSI_DATA_IN:
760                 ctask = session->cmds[itt];
761                 spin_lock(&conn->session->lock);
762                 rc = iscsi_data_rsp(conn, ctask);
763                 spin_unlock(&conn->session->lock);
764                 if (rc)
765                         return rc;
766                 if (tcp_conn->in.datalen) {
767                         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
768                         struct hash_desc *rx_hash = NULL;
769
770                         /*
771                          * Setup copy of Data-In into the Scsi_Cmnd
772                          * Scatterlist case:
773                          * We set up the iscsi_segment to point to the next
774                          * scatterlist entry to copy to. As we go along,
775                          * we move on to the next scatterlist entry and
776                          * update the digest per-entry.
777                          */
778                         if (conn->datadgst_en)
779                                 rx_hash = &tcp_conn->rx_hash;
780
781                         debug_tcp("iscsi_tcp_begin_data_in(%p, offset=%d, "
782                                   "datalen=%d)\n", tcp_conn,
783                                   tcp_ctask->data_offset,
784                                   tcp_conn->in.datalen);
785                         return iscsi_segment_seek_sg(&tcp_conn->in.segment,
786                                                      scsi_sglist(ctask->sc),
787                                                      scsi_sg_count(ctask->sc),
788                                                      tcp_ctask->data_offset,
789                                                      tcp_conn->in.datalen,
790                                                      iscsi_tcp_process_data_in,
791                                                      rx_hash);
792                 }
793                 /* fall through */
794         case ISCSI_OP_SCSI_CMD_RSP:
795                 if (tcp_conn->in.datalen) {
796                         iscsi_tcp_data_recv_prep(tcp_conn);
797                         return 0;
798                 }
799                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
800                 break;
801         case ISCSI_OP_R2T:
802                 ctask = session->cmds[itt];
803                 if (ahslen)
804                         rc = ISCSI_ERR_AHSLEN;
805                 else if (ctask->sc->sc_data_direction == DMA_TO_DEVICE) {
806                         spin_lock(&session->lock);
807                         rc = iscsi_r2t_rsp(conn, ctask);
808                         spin_unlock(&session->lock);
809                 } else
810                         rc = ISCSI_ERR_PROTO;
811                 break;
812         case ISCSI_OP_LOGIN_RSP:
813         case ISCSI_OP_TEXT_RSP:
814         case ISCSI_OP_REJECT:
815         case ISCSI_OP_ASYNC_EVENT:
816                 /*
817                  * It is possible that we could get a PDU with a buffer larger
818                  * than 8K, but there are no targets that currently do this.
819                  * For now we fail until we find a vendor that needs it
820                  */
821                 if (ISCSI_DEF_MAX_RECV_SEG_LEN < tcp_conn->in.datalen) {
822                         printk(KERN_ERR "iscsi_tcp: received buffer of len %u "
823                               "but conn buffer is only %u (opcode %0x)\n",
824                               tcp_conn->in.datalen,
825                               ISCSI_DEF_MAX_RECV_SEG_LEN, opcode);
826                         rc = ISCSI_ERR_PROTO;
827                         break;
828                 }
829
830                 /* If there's data coming in with the response,
831                  * receive it to the connection's buffer.
832                  */
833                 if (tcp_conn->in.datalen) {
834                         iscsi_tcp_data_recv_prep(tcp_conn);
835                         return 0;
836                 }
837         /* fall through */
838         case ISCSI_OP_LOGOUT_RSP:
839         case ISCSI_OP_NOOP_IN:
840         case ISCSI_OP_SCSI_TMFUNC_RSP:
841                 rc = iscsi_complete_pdu(conn, hdr, NULL, 0);
842                 break;
843         default:
844                 rc = ISCSI_ERR_BAD_OPCODE;
845                 break;
846         }
847
848         if (rc == 0) {
849                 /* Anything that comes with data should have
850                  * been handled above. */
851                 if (tcp_conn->in.datalen)
852                         return ISCSI_ERR_PROTO;
853                 iscsi_tcp_hdr_recv_prep(tcp_conn);
854         }
855
856         return rc;
857 }
858
859 /**
860  * iscsi_tcp_hdr_recv_done - process PDU header
861  *
862  * This is the callback invoked when the PDU header has
863  * been received. If the header is followed by additional
864  * header segments, we go back for more data.
865  */
866 static int
867 iscsi_tcp_hdr_recv_done(struct iscsi_tcp_conn *tcp_conn,
868                         struct iscsi_segment *segment)
869 {
870         struct iscsi_conn *conn = tcp_conn->iscsi_conn;
871         struct iscsi_hdr *hdr;
872
873         /* Check if there are additional header segments
874          * *prior* to computing the digest, because we
875          * may need to go back to the caller for more.
876          */
877         hdr = (struct iscsi_hdr *) tcp_conn->in.hdr_buf;
878         if (segment->copied == sizeof(struct iscsi_hdr) && hdr->hlength) {
879                 /* Bump the header length - the caller will
880                  * just loop around and get the AHS for us, and
881                  * call again. */
882                 unsigned int ahslen = hdr->hlength << 2;
883
884                 /* Make sure we don't overflow */
885                 if (sizeof(*hdr) + ahslen > sizeof(tcp_conn->in.hdr_buf))
886                         return ISCSI_ERR_AHSLEN;
887
888                 segment->total_size += ahslen;
889                 segment->size += ahslen;
890                 return 0;
891         }
892
893         /* We're done processing the header. See if we're doing
894          * header digests; if so, set up the recv_digest buffer
895          * and go back for more. */
896         if (conn->hdrdgst_en) {
897                 if (segment->digest_len == 0) {
898                         iscsi_tcp_segment_splice_digest(segment,
899                                                         segment->recv_digest);
900                         return 0;
901                 }
902                 iscsi_tcp_dgst_header(&tcp_conn->rx_hash, hdr,
903                                       segment->total_copied - ISCSI_DIGEST_SIZE,
904                                       segment->digest);
905
906                 if (!iscsi_tcp_dgst_verify(tcp_conn, segment))
907                         return ISCSI_ERR_HDR_DGST;
908         }
909
910         tcp_conn->in.hdr = hdr;
911         return iscsi_tcp_hdr_dissect(conn, hdr);
912 }
913
914 /**
915  * iscsi_tcp_recv - TCP receive in sendfile fashion
916  * @rd_desc: read descriptor
917  * @skb: socket buffer
918  * @offset: offset in skb
919  * @len: skb->len - offset
920  **/
921 static int
922 iscsi_tcp_recv(read_descriptor_t *rd_desc, struct sk_buff *skb,
923                unsigned int offset, size_t len)
924 {
925         struct iscsi_conn *conn = rd_desc->arg.data;
926         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
927         struct iscsi_segment *segment = &tcp_conn->in.segment;
928         struct skb_seq_state seq;
929         unsigned int consumed = 0;
930         int rc = 0;
931
932         debug_tcp("in %d bytes\n", skb->len - offset);
933
934         if (unlikely(conn->suspend_rx)) {
935                 debug_tcp("conn %d Rx suspended!\n", conn->id);
936                 return 0;
937         }
938
939         skb_prepare_seq_read(skb, offset, skb->len, &seq);
940         while (1) {
941                 unsigned int avail;
942                 const u8 *ptr;
943
944                 avail = skb_seq_read(consumed, &ptr, &seq);
945                 if (avail == 0) {
946                         debug_tcp("no more data avail. Consumed %d\n",
947                                   consumed);
948                         break;
949                 }
950                 BUG_ON(segment->copied >= segment->size);
951
952                 debug_tcp("skb %p ptr=%p avail=%u\n", skb, ptr, avail);
953                 rc = iscsi_tcp_segment_recv(tcp_conn, segment, ptr, avail);
954                 BUG_ON(rc == 0);
955                 consumed += rc;
956
957                 if (segment->total_copied >= segment->total_size) {
958                         debug_tcp("segment done\n");
959                         rc = segment->done(tcp_conn, segment);
960                         if (rc != 0) {
961                                 skb_abort_seq_read(&seq);
962                                 goto error;
963                         }
964
965                         /* The done() functions sets up the
966                          * next segment. */
967                 }
968         }
969         skb_abort_seq_read(&seq);
970         conn->rxdata_octets += consumed;
971         return consumed;
972
973 error:
974         debug_tcp("Error receiving PDU, errno=%d\n", rc);
975         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
976         return 0;
977 }
978
979 static void
980 iscsi_tcp_data_ready(struct sock *sk, int flag)
981 {
982         struct iscsi_conn *conn = sk->sk_user_data;
983         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
984         read_descriptor_t rd_desc;
985
986         read_lock(&sk->sk_callback_lock);
987
988         /*
989          * Use rd_desc to pass 'conn' to iscsi_tcp_recv.
990          * We set count to 1 because we want the network layer to
991          * hand us all the skbs that are available. iscsi_tcp_recv
992          * handled pdus that cross buffers or pdus that still need data.
993          */
994         rd_desc.arg.data = conn;
995         rd_desc.count = 1;
996         tcp_read_sock(sk, &rd_desc, iscsi_tcp_recv);
997
998         read_unlock(&sk->sk_callback_lock);
999
1000         /* If we had to (atomically) map a highmem page,
1001          * unmap it now. */
1002         iscsi_tcp_segment_unmap(&tcp_conn->in.segment);
1003 }
1004
1005 static void
1006 iscsi_tcp_state_change(struct sock *sk)
1007 {
1008         struct iscsi_tcp_conn *tcp_conn;
1009         struct iscsi_conn *conn;
1010         struct iscsi_session *session;
1011         void (*old_state_change)(struct sock *);
1012
1013         read_lock(&sk->sk_callback_lock);
1014
1015         conn = (struct iscsi_conn*)sk->sk_user_data;
1016         session = conn->session;
1017
1018         if ((sk->sk_state == TCP_CLOSE_WAIT ||
1019              sk->sk_state == TCP_CLOSE) &&
1020             !atomic_read(&sk->sk_rmem_alloc)) {
1021                 debug_tcp("iscsi_tcp_state_change: TCP_CLOSE|TCP_CLOSE_WAIT\n");
1022                 iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1023         }
1024
1025         tcp_conn = conn->dd_data;
1026         old_state_change = tcp_conn->old_state_change;
1027
1028         read_unlock(&sk->sk_callback_lock);
1029
1030         old_state_change(sk);
1031 }
1032
1033 /**
1034  * iscsi_write_space - Called when more output buffer space is available
1035  * @sk: socket space is available for
1036  **/
1037 static void
1038 iscsi_write_space(struct sock *sk)
1039 {
1040         struct iscsi_conn *conn = (struct iscsi_conn*)sk->sk_user_data;
1041         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1042
1043         tcp_conn->old_write_space(sk);
1044         debug_tcp("iscsi_write_space: cid %d\n", conn->id);
1045         scsi_queue_work(conn->session->host, &conn->xmitwork);
1046 }
1047
1048 static void
1049 iscsi_conn_set_callbacks(struct iscsi_conn *conn)
1050 {
1051         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1052         struct sock *sk = tcp_conn->sock->sk;
1053
1054         /* assign new callbacks */
1055         write_lock_bh(&sk->sk_callback_lock);
1056         sk->sk_user_data = conn;
1057         tcp_conn->old_data_ready = sk->sk_data_ready;
1058         tcp_conn->old_state_change = sk->sk_state_change;
1059         tcp_conn->old_write_space = sk->sk_write_space;
1060         sk->sk_data_ready = iscsi_tcp_data_ready;
1061         sk->sk_state_change = iscsi_tcp_state_change;
1062         sk->sk_write_space = iscsi_write_space;
1063         write_unlock_bh(&sk->sk_callback_lock);
1064 }
1065
1066 static void
1067 iscsi_conn_restore_callbacks(struct iscsi_tcp_conn *tcp_conn)
1068 {
1069         struct sock *sk = tcp_conn->sock->sk;
1070
1071         /* restore socket callbacks, see also: iscsi_conn_set_callbacks() */
1072         write_lock_bh(&sk->sk_callback_lock);
1073         sk->sk_user_data    = NULL;
1074         sk->sk_data_ready   = tcp_conn->old_data_ready;
1075         sk->sk_state_change = tcp_conn->old_state_change;
1076         sk->sk_write_space  = tcp_conn->old_write_space;
1077         sk->sk_no_check  = 0;
1078         write_unlock_bh(&sk->sk_callback_lock);
1079 }
1080
1081 /**
1082  * iscsi_xmit - TCP transmit
1083  **/
1084 static int
1085 iscsi_xmit(struct iscsi_conn *conn)
1086 {
1087         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1088         struct iscsi_segment *segment = &tcp_conn->out.segment;
1089         unsigned int consumed = 0;
1090         int rc = 0;
1091
1092         while (1) {
1093                 rc = iscsi_tcp_xmit_segment(tcp_conn, segment);
1094                 if (rc < 0)
1095                         goto error;
1096                 if (rc == 0)
1097                         break;
1098
1099                 consumed += rc;
1100
1101                 if (segment->total_copied >= segment->total_size) {
1102                         if (segment->done != NULL) {
1103                                 rc = segment->done(tcp_conn, segment);
1104                                 if (rc < 0)
1105                                         goto error;
1106                         }
1107                 }
1108         }
1109
1110         debug_tcp("xmit %d bytes\n", consumed);
1111
1112         conn->txdata_octets += consumed;
1113         return consumed;
1114
1115 error:
1116         /* Transmit error. We could initiate error recovery
1117          * here. */
1118         debug_tcp("Error sending PDU, errno=%d\n", rc);
1119         iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED);
1120         return rc;
1121 }
1122
1123 /**
1124  * iscsi_tcp_xmit_qlen - return the number of bytes queued for xmit
1125  */
1126 static inline int
1127 iscsi_tcp_xmit_qlen(struct iscsi_conn *conn)
1128 {
1129         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1130         struct iscsi_segment *segment = &tcp_conn->out.segment;
1131
1132         return segment->total_copied - segment->total_size;
1133 }
1134
1135 static inline int
1136 iscsi_tcp_flush(struct iscsi_conn *conn)
1137 {
1138         int rc;
1139
1140         while (iscsi_tcp_xmit_qlen(conn)) {
1141                 rc = iscsi_xmit(conn);
1142                 if (rc == 0)
1143                         return -EAGAIN;
1144                 if (rc < 0)
1145                         return rc;
1146         }
1147
1148         return 0;
1149 }
1150
1151 /*
1152  * This is called when we're done sending the header.
1153  * Simply copy the data_segment to the send segment, and return.
1154  */
1155 static int
1156 iscsi_tcp_send_hdr_done(struct iscsi_tcp_conn *tcp_conn,
1157                         struct iscsi_segment *segment)
1158 {
1159         tcp_conn->out.segment = tcp_conn->out.data_segment;
1160         debug_tcp("Header done. Next segment size %u total_size %u\n",
1161                   tcp_conn->out.segment.size, tcp_conn->out.segment.total_size);
1162         return 0;
1163 }
1164
1165 static void
1166 iscsi_tcp_send_hdr_prep(struct iscsi_conn *conn, void *hdr, size_t hdrlen)
1167 {
1168         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1169
1170         debug_tcp("%s(%p%s)\n", __FUNCTION__, tcp_conn,
1171                         conn->hdrdgst_en? ", digest enabled" : "");
1172
1173         /* Clear the data segment - needs to be filled in by the
1174          * caller using iscsi_tcp_send_data_prep() */
1175         memset(&tcp_conn->out.data_segment, 0, sizeof(struct iscsi_segment));
1176
1177         /* If header digest is enabled, compute the CRC and
1178          * place the digest into the same buffer. We make
1179          * sure that both iscsi_tcp_ctask and mtask have
1180          * sufficient room.
1181          */
1182         if (conn->hdrdgst_en) {
1183                 iscsi_tcp_dgst_header(&tcp_conn->tx_hash, hdr, hdrlen,
1184                                       hdr + hdrlen);
1185                 hdrlen += ISCSI_DIGEST_SIZE;
1186         }
1187
1188         /* Remember header pointer for later, when we need
1189          * to decide whether there's a payload to go along
1190          * with the header. */
1191         tcp_conn->out.hdr = hdr;
1192
1193         iscsi_segment_init_linear(&tcp_conn->out.segment, hdr, hdrlen,
1194                                 iscsi_tcp_send_hdr_done, NULL);
1195 }
1196
1197 /*
1198  * Prepare the send buffer for the payload data.
1199  * Padding and checksumming will all be taken care
1200  * of by the iscsi_segment routines.
1201  */
1202 static int
1203 iscsi_tcp_send_data_prep(struct iscsi_conn *conn, struct scatterlist *sg,
1204                          unsigned int count, unsigned int offset,
1205                          unsigned int len)
1206 {
1207         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1208         struct hash_desc *tx_hash = NULL;
1209         unsigned int hdr_spec_len;
1210
1211         debug_tcp("%s(%p, offset=%d, datalen=%d%s)\n", __FUNCTION__,
1212                         tcp_conn, offset, len,
1213                         conn->datadgst_en? ", digest enabled" : "");
1214
1215         /* Make sure the datalen matches what the caller
1216            said he would send. */
1217         hdr_spec_len = ntoh24(tcp_conn->out.hdr->dlength);
1218         WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
1219
1220         if (conn->datadgst_en)
1221                 tx_hash = &tcp_conn->tx_hash;
1222
1223         return iscsi_segment_seek_sg(&tcp_conn->out.data_segment,
1224                                    sg, count, offset, len,
1225                                    NULL, tx_hash);
1226 }
1227
1228 static void
1229 iscsi_tcp_send_linear_data_prepare(struct iscsi_conn *conn, void *data,
1230                                    size_t len)
1231 {
1232         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1233         struct hash_desc *tx_hash = NULL;
1234         unsigned int hdr_spec_len;
1235
1236         debug_tcp("%s(%p, datalen=%d%s)\n", __FUNCTION__, tcp_conn, len,
1237                   conn->datadgst_en? ", digest enabled" : "");
1238
1239         /* Make sure the datalen matches what the caller
1240            said he would send. */
1241         hdr_spec_len = ntoh24(tcp_conn->out.hdr->dlength);
1242         WARN_ON(iscsi_padded(len) != iscsi_padded(hdr_spec_len));
1243
1244         if (conn->datadgst_en)
1245                 tx_hash = &tcp_conn->tx_hash;
1246
1247         iscsi_segment_init_linear(&tcp_conn->out.data_segment,
1248                                 data, len, NULL, tx_hash);
1249 }
1250
1251 /**
1252  * iscsi_solicit_data_cont - initialize next Data-Out
1253  * @conn: iscsi connection
1254  * @ctask: scsi command task
1255  * @r2t: R2T info
1256  * @left: bytes left to transfer
1257  *
1258  * Notes:
1259  *      Initialize next Data-Out within this R2T sequence and continue
1260  *      to process next Scatter-Gather element(if any) of this SCSI command.
1261  *
1262  *      Called under connection lock.
1263  **/
1264 static int
1265 iscsi_solicit_data_cont(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask,
1266                         struct iscsi_r2t_info *r2t)
1267 {
1268         struct iscsi_data *hdr;
1269         int new_offset, left;
1270
1271         BUG_ON(r2t->data_length - r2t->sent < 0);
1272         left = r2t->data_length - r2t->sent;
1273         if (left == 0)
1274                 return 0;
1275
1276         hdr = &r2t->dtask.hdr;
1277         memset(hdr, 0, sizeof(struct iscsi_data));
1278         hdr->ttt = r2t->ttt;
1279         hdr->datasn = cpu_to_be32(r2t->solicit_datasn);
1280         r2t->solicit_datasn++;
1281         hdr->opcode = ISCSI_OP_SCSI_DATA_OUT;
1282         memcpy(hdr->lun, ctask->hdr->lun, sizeof(hdr->lun));
1283         hdr->itt = ctask->hdr->itt;
1284         hdr->exp_statsn = r2t->exp_statsn;
1285         new_offset = r2t->data_offset + r2t->sent;
1286         hdr->offset = cpu_to_be32(new_offset);
1287         if (left > conn->max_xmit_dlength) {
1288                 hton24(hdr->dlength, conn->max_xmit_dlength);
1289                 r2t->data_count = conn->max_xmit_dlength;
1290         } else {
1291                 hton24(hdr->dlength, left);
1292                 r2t->data_count = left;
1293                 hdr->flags = ISCSI_FLAG_CMD_FINAL;
1294         }
1295
1296         conn->dataout_pdus_cnt++;
1297         return 1;
1298 }
1299
1300 /**
1301  * iscsi_tcp_ctask - Initialize iSCSI SCSI_READ or SCSI_WRITE commands
1302  * @conn: iscsi connection
1303  * @ctask: scsi command task
1304  * @sc: scsi command
1305  **/
1306 static int
1307 iscsi_tcp_ctask_init(struct iscsi_cmd_task *ctask)
1308 {
1309         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1310         struct iscsi_conn *conn = ctask->conn;
1311         struct scsi_cmnd *sc = ctask->sc;
1312         int err;
1313
1314         BUG_ON(__kfifo_len(tcp_ctask->r2tqueue));
1315         tcp_ctask->sent = 0;
1316         tcp_ctask->exp_datasn = 0;
1317
1318         /* Prepare PDU, optionally w/ immediate data */
1319         debug_scsi("ctask deq [cid %d itt 0x%x imm %d unsol %d]\n",
1320                     conn->id, ctask->itt, ctask->imm_count,
1321                     ctask->unsol_count);
1322         iscsi_tcp_send_hdr_prep(conn, ctask->hdr, ctask->hdr_len);
1323
1324         if (!ctask->imm_count)
1325                 return 0;
1326
1327         /* If we have immediate data, attach a payload */
1328         err = iscsi_tcp_send_data_prep(conn, scsi_sglist(sc), scsi_sg_count(sc),
1329                                        0, ctask->imm_count);
1330         if (err)
1331                 return err;
1332         tcp_ctask->sent += ctask->imm_count;
1333         ctask->imm_count = 0;
1334         return 0;
1335 }
1336
1337 /**
1338  * iscsi_tcp_mtask_xmit - xmit management(immediate) task
1339  * @conn: iscsi connection
1340  * @mtask: task management task
1341  *
1342  * Notes:
1343  *      The function can return -EAGAIN in which case caller must
1344  *      call it again later, or recover. '0' return code means successful
1345  *      xmit.
1346  **/
1347 static int
1348 iscsi_tcp_mtask_xmit(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1349 {
1350         int rc;
1351
1352         /* Flush any pending data first. */
1353         rc = iscsi_tcp_flush(conn);
1354         if (rc < 0)
1355                 return rc;
1356
1357         if (mtask->hdr->itt == RESERVED_ITT) {
1358                 struct iscsi_session *session = conn->session;
1359
1360                 spin_lock_bh(&session->lock);
1361                 iscsi_free_mgmt_task(conn, mtask);
1362                 spin_unlock_bh(&session->lock);
1363         }
1364
1365         return 0;
1366 }
1367
1368 /*
1369  * iscsi_tcp_ctask_xmit - xmit normal PDU task
1370  * @conn: iscsi connection
1371  * @ctask: iscsi command task
1372  *
1373  * We're expected to return 0 when everything was transmitted succesfully,
1374  * -EAGAIN if there's still data in the queue, or != 0 for any other kind
1375  * of error.
1376  */
1377 static int
1378 iscsi_tcp_ctask_xmit(struct iscsi_conn *conn, struct iscsi_cmd_task *ctask)
1379 {
1380         struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1381         struct scsi_cmnd *sc = ctask->sc;
1382         int rc = 0;
1383
1384 flush:
1385         /* Flush any pending data first. */
1386         rc = iscsi_tcp_flush(conn);
1387         if (rc < 0)
1388                 return rc;
1389
1390         /* Are we done already? */
1391         if (sc->sc_data_direction != DMA_TO_DEVICE)
1392                 return 0;
1393
1394         if (ctask->unsol_count != 0) {
1395                 struct iscsi_data *hdr = &tcp_ctask->unsol_dtask.hdr;
1396
1397                 /* Prepare a header for the unsolicited PDU.
1398                  * The amount of data we want to send will be
1399                  * in ctask->data_count.
1400                  * FIXME: return the data count instead.
1401                  */
1402                 iscsi_prep_unsolicit_data_pdu(ctask, hdr);
1403
1404                 debug_tcp("unsol dout [itt 0x%x doff %d dlen %d]\n",
1405                                 ctask->itt, tcp_ctask->sent, ctask->data_count);
1406
1407                 iscsi_tcp_send_hdr_prep(conn, hdr, sizeof(*hdr));
1408                 rc = iscsi_tcp_send_data_prep(conn, scsi_sglist(sc),
1409                                               scsi_sg_count(sc),
1410                                               tcp_ctask->sent,
1411                                               ctask->data_count);
1412                 if (rc)
1413                         goto fail;
1414                 tcp_ctask->sent += ctask->data_count;
1415                 ctask->unsol_count -= ctask->data_count;
1416                 goto flush;
1417         } else {
1418                 struct iscsi_session *session = conn->session;
1419                 struct iscsi_r2t_info *r2t;
1420
1421                 /* All unsolicited PDUs sent. Check for solicited PDUs.
1422                  */
1423                 spin_lock_bh(&session->lock);
1424                 r2t = tcp_ctask->r2t;
1425                 if (r2t != NULL) {
1426                         /* Continue with this R2T? */
1427                         if (!iscsi_solicit_data_cont(conn, ctask, r2t)) {
1428                                 debug_scsi("  done with r2t %p\n", r2t);
1429
1430                                 __kfifo_put(tcp_ctask->r2tpool.queue,
1431                                             (void*)&r2t, sizeof(void*));
1432                                 tcp_ctask->r2t = r2t = NULL;
1433                         }
1434                 }
1435
1436                 if (r2t == NULL) {
1437                         __kfifo_get(tcp_ctask->r2tqueue, (void*)&tcp_ctask->r2t,
1438                                     sizeof(void*));
1439                         r2t = tcp_ctask->r2t;
1440                 }
1441                 spin_unlock_bh(&session->lock);
1442
1443                 /* Waiting for more R2Ts to arrive. */
1444                 if (r2t == NULL) {
1445                         debug_tcp("no R2Ts yet\n");
1446                         return 0;
1447                 }
1448
1449                 debug_scsi("sol dout %p [dsn %d itt 0x%x doff %d dlen %d]\n",
1450                         r2t, r2t->solicit_datasn - 1, ctask->itt,
1451                         r2t->data_offset + r2t->sent, r2t->data_count);
1452
1453                 iscsi_tcp_send_hdr_prep(conn, &r2t->dtask.hdr,
1454                                         sizeof(struct iscsi_hdr));
1455
1456                 rc = iscsi_tcp_send_data_prep(conn, scsi_sglist(sc),
1457                                               scsi_sg_count(sc),
1458                                               r2t->data_offset + r2t->sent,
1459                                               r2t->data_count);
1460                 if (rc)
1461                         goto fail;
1462                 tcp_ctask->sent += r2t->data_count;
1463                 r2t->sent += r2t->data_count;
1464                 goto flush;
1465         }
1466         return 0;
1467 fail:
1468         iscsi_conn_failure(conn, rc);
1469         return -EIO;
1470 }
1471
1472 static struct iscsi_cls_conn *
1473 iscsi_tcp_conn_create(struct iscsi_cls_session *cls_session, uint32_t conn_idx)
1474 {
1475         struct iscsi_conn *conn;
1476         struct iscsi_cls_conn *cls_conn;
1477         struct iscsi_tcp_conn *tcp_conn;
1478
1479         cls_conn = iscsi_conn_setup(cls_session, conn_idx);
1480         if (!cls_conn)
1481                 return NULL;
1482         conn = cls_conn->dd_data;
1483         /*
1484          * due to strange issues with iser these are not set
1485          * in iscsi_conn_setup
1486          */
1487         conn->max_recv_dlength = ISCSI_DEF_MAX_RECV_SEG_LEN;
1488
1489         tcp_conn = kzalloc(sizeof(*tcp_conn), GFP_KERNEL);
1490         if (!tcp_conn)
1491                 goto tcp_conn_alloc_fail;
1492
1493         conn->dd_data = tcp_conn;
1494         tcp_conn->iscsi_conn = conn;
1495
1496         tcp_conn->tx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1497                                                   CRYPTO_ALG_ASYNC);
1498         tcp_conn->tx_hash.flags = 0;
1499         if (IS_ERR(tcp_conn->tx_hash.tfm)) {
1500                 printk(KERN_ERR "Could not create connection due to crc32c "
1501                        "loading error %ld. Make sure the crc32c module is "
1502                        "built as a module or into the kernel\n",
1503                         PTR_ERR(tcp_conn->tx_hash.tfm));
1504                 goto free_tcp_conn;
1505         }
1506
1507         tcp_conn->rx_hash.tfm = crypto_alloc_hash("crc32c", 0,
1508                                                   CRYPTO_ALG_ASYNC);
1509         tcp_conn->rx_hash.flags = 0;
1510         if (IS_ERR(tcp_conn->rx_hash.tfm)) {
1511                 printk(KERN_ERR "Could not create connection due to crc32c "
1512                        "loading error %ld. Make sure the crc32c module is "
1513                        "built as a module or into the kernel\n",
1514                         PTR_ERR(tcp_conn->rx_hash.tfm));
1515                 goto free_tx_tfm;
1516         }
1517
1518         return cls_conn;
1519
1520 free_tx_tfm:
1521         crypto_free_hash(tcp_conn->tx_hash.tfm);
1522 free_tcp_conn:
1523         kfree(tcp_conn);
1524 tcp_conn_alloc_fail:
1525         iscsi_conn_teardown(cls_conn);
1526         return NULL;
1527 }
1528
1529 static void
1530 iscsi_tcp_release_conn(struct iscsi_conn *conn)
1531 {
1532         struct iscsi_session *session = conn->session;
1533         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1534         struct socket *sock = tcp_conn->sock;
1535
1536         if (!sock)
1537                 return;
1538
1539         sock_hold(sock->sk);
1540         iscsi_conn_restore_callbacks(tcp_conn);
1541         sock_put(sock->sk);
1542
1543         spin_lock_bh(&session->lock);
1544         tcp_conn->sock = NULL;
1545         conn->recv_lock = NULL;
1546         spin_unlock_bh(&session->lock);
1547         sockfd_put(sock);
1548 }
1549
1550 static void
1551 iscsi_tcp_conn_destroy(struct iscsi_cls_conn *cls_conn)
1552 {
1553         struct iscsi_conn *conn = cls_conn->dd_data;
1554         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1555
1556         iscsi_tcp_release_conn(conn);
1557         iscsi_conn_teardown(cls_conn);
1558
1559         if (tcp_conn->tx_hash.tfm)
1560                 crypto_free_hash(tcp_conn->tx_hash.tfm);
1561         if (tcp_conn->rx_hash.tfm)
1562                 crypto_free_hash(tcp_conn->rx_hash.tfm);
1563
1564         kfree(tcp_conn);
1565 }
1566
1567 static void
1568 iscsi_tcp_conn_stop(struct iscsi_cls_conn *cls_conn, int flag)
1569 {
1570         struct iscsi_conn *conn = cls_conn->dd_data;
1571
1572         iscsi_conn_stop(cls_conn, flag);
1573         iscsi_tcp_release_conn(conn);
1574 }
1575
1576 static int iscsi_tcp_get_addr(struct iscsi_conn *conn, struct socket *sock,
1577                               char *buf, int *port,
1578                               int (*getname)(struct socket *, struct sockaddr *,
1579                                         int *addrlen))
1580 {
1581         struct sockaddr_storage *addr;
1582         struct sockaddr_in6 *sin6;
1583         struct sockaddr_in *sin;
1584         int rc = 0, len;
1585
1586         addr = kmalloc(sizeof(*addr), GFP_KERNEL);
1587         if (!addr)
1588                 return -ENOMEM;
1589
1590         if (getname(sock, (struct sockaddr *) addr, &len)) {
1591                 rc = -ENODEV;
1592                 goto free_addr;
1593         }
1594
1595         switch (addr->ss_family) {
1596         case AF_INET:
1597                 sin = (struct sockaddr_in *)addr;
1598                 spin_lock_bh(&conn->session->lock);
1599                 sprintf(buf, NIPQUAD_FMT, NIPQUAD(sin->sin_addr.s_addr));
1600                 *port = be16_to_cpu(sin->sin_port);
1601                 spin_unlock_bh(&conn->session->lock);
1602                 break;
1603         case AF_INET6:
1604                 sin6 = (struct sockaddr_in6 *)addr;
1605                 spin_lock_bh(&conn->session->lock);
1606                 sprintf(buf, NIP6_FMT, NIP6(sin6->sin6_addr));
1607                 *port = be16_to_cpu(sin6->sin6_port);
1608                 spin_unlock_bh(&conn->session->lock);
1609                 break;
1610         }
1611 free_addr:
1612         kfree(addr);
1613         return rc;
1614 }
1615
1616 static int
1617 iscsi_tcp_conn_bind(struct iscsi_cls_session *cls_session,
1618                     struct iscsi_cls_conn *cls_conn, uint64_t transport_eph,
1619                     int is_leading)
1620 {
1621         struct iscsi_conn *conn = cls_conn->dd_data;
1622         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1623         struct sock *sk;
1624         struct socket *sock;
1625         int err;
1626
1627         /* lookup for existing socket */
1628         sock = sockfd_lookup((int)transport_eph, &err);
1629         if (!sock) {
1630                 printk(KERN_ERR "iscsi_tcp: sockfd_lookup failed %d\n", err);
1631                 return -EEXIST;
1632         }
1633         /*
1634          * copy these values now because if we drop the session
1635          * userspace may still want to query the values since we will
1636          * be using them for the reconnect
1637          */
1638         err = iscsi_tcp_get_addr(conn, sock, conn->portal_address,
1639                                  &conn->portal_port, kernel_getpeername);
1640         if (err)
1641                 goto free_socket;
1642
1643         err = iscsi_tcp_get_addr(conn, sock, conn->local_address,
1644                                 &conn->local_port, kernel_getsockname);
1645         if (err)
1646                 goto free_socket;
1647
1648         err = iscsi_conn_bind(cls_session, cls_conn, is_leading);
1649         if (err)
1650                 goto free_socket;
1651
1652         /* bind iSCSI connection and socket */
1653         tcp_conn->sock = sock;
1654
1655         /* setup Socket parameters */
1656         sk = sock->sk;
1657         sk->sk_reuse = 1;
1658         sk->sk_sndtimeo = 15 * HZ; /* FIXME: make it configurable */
1659         sk->sk_allocation = GFP_ATOMIC;
1660
1661         /* FIXME: disable Nagle's algorithm */
1662
1663         /*
1664          * Intercept TCP callbacks for sendfile like receive
1665          * processing.
1666          */
1667         conn->recv_lock = &sk->sk_callback_lock;
1668         iscsi_conn_set_callbacks(conn);
1669         tcp_conn->sendpage = tcp_conn->sock->ops->sendpage;
1670         /*
1671          * set receive state machine into initial state
1672          */
1673         iscsi_tcp_hdr_recv_prep(tcp_conn);
1674         return 0;
1675
1676 free_socket:
1677         sockfd_put(sock);
1678         return err;
1679 }
1680
1681 /* called with host lock */
1682 static void
1683 iscsi_tcp_mtask_init(struct iscsi_conn *conn, struct iscsi_mgmt_task *mtask)
1684 {
1685         debug_scsi("mtask deq [cid %d itt 0x%x]\n", conn->id, mtask->itt);
1686
1687         /* Prepare PDU, optionally w/ immediate data */
1688         iscsi_tcp_send_hdr_prep(conn, mtask->hdr, sizeof(*mtask->hdr));
1689
1690         /* If we have immediate data, attach a payload */
1691         if (mtask->data_count)
1692                 iscsi_tcp_send_linear_data_prepare(conn, mtask->data,
1693                                                    mtask->data_count);
1694 }
1695
1696 static int
1697 iscsi_r2tpool_alloc(struct iscsi_session *session)
1698 {
1699         int i;
1700         int cmd_i;
1701
1702         /*
1703          * initialize per-task: R2T pool and xmit queue
1704          */
1705         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1706                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
1707                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1708
1709                 /*
1710                  * pre-allocated x4 as much r2ts to handle race when
1711                  * target acks DataOut faster than we data_xmit() queues
1712                  * could replenish r2tqueue.
1713                  */
1714
1715                 /* R2T pool */
1716                 if (iscsi_pool_init(&tcp_ctask->r2tpool, session->max_r2t * 4, NULL,
1717                                     sizeof(struct iscsi_r2t_info))) {
1718                         goto r2t_alloc_fail;
1719                 }
1720
1721                 /* R2T xmit queue */
1722                 tcp_ctask->r2tqueue = kfifo_alloc(
1723                       session->max_r2t * 4 * sizeof(void*), GFP_KERNEL, NULL);
1724                 if (tcp_ctask->r2tqueue == ERR_PTR(-ENOMEM)) {
1725                         iscsi_pool_free(&tcp_ctask->r2tpool);
1726                         goto r2t_alloc_fail;
1727                 }
1728         }
1729
1730         return 0;
1731
1732 r2t_alloc_fail:
1733         for (i = 0; i < cmd_i; i++) {
1734                 struct iscsi_cmd_task *ctask = session->cmds[i];
1735                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1736
1737                 kfifo_free(tcp_ctask->r2tqueue);
1738                 iscsi_pool_free(&tcp_ctask->r2tpool);
1739         }
1740         return -ENOMEM;
1741 }
1742
1743 static void
1744 iscsi_r2tpool_free(struct iscsi_session *session)
1745 {
1746         int i;
1747
1748         for (i = 0; i < session->cmds_max; i++) {
1749                 struct iscsi_cmd_task *ctask = session->cmds[i];
1750                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1751
1752                 kfifo_free(tcp_ctask->r2tqueue);
1753                 iscsi_pool_free(&tcp_ctask->r2tpool);
1754         }
1755 }
1756
1757 static int
1758 iscsi_conn_set_param(struct iscsi_cls_conn *cls_conn, enum iscsi_param param,
1759                      char *buf, int buflen)
1760 {
1761         struct iscsi_conn *conn = cls_conn->dd_data;
1762         struct iscsi_session *session = conn->session;
1763         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1764         int value;
1765
1766         switch(param) {
1767         case ISCSI_PARAM_HDRDGST_EN:
1768                 iscsi_set_param(cls_conn, param, buf, buflen);
1769                 break;
1770         case ISCSI_PARAM_DATADGST_EN:
1771                 iscsi_set_param(cls_conn, param, buf, buflen);
1772                 tcp_conn->sendpage = conn->datadgst_en ?
1773                         sock_no_sendpage : tcp_conn->sock->ops->sendpage;
1774                 break;
1775         case ISCSI_PARAM_MAX_R2T:
1776                 sscanf(buf, "%d", &value);
1777                 if (session->max_r2t == roundup_pow_of_two(value))
1778                         break;
1779                 iscsi_r2tpool_free(session);
1780                 iscsi_set_param(cls_conn, param, buf, buflen);
1781                 if (session->max_r2t & (session->max_r2t - 1))
1782                         session->max_r2t = roundup_pow_of_two(session->max_r2t);
1783                 if (iscsi_r2tpool_alloc(session))
1784                         return -ENOMEM;
1785                 break;
1786         default:
1787                 return iscsi_set_param(cls_conn, param, buf, buflen);
1788         }
1789
1790         return 0;
1791 }
1792
1793 static int
1794 iscsi_tcp_conn_get_param(struct iscsi_cls_conn *cls_conn,
1795                          enum iscsi_param param, char *buf)
1796 {
1797         struct iscsi_conn *conn = cls_conn->dd_data;
1798         int len;
1799
1800         switch(param) {
1801         case ISCSI_PARAM_CONN_PORT:
1802                 spin_lock_bh(&conn->session->lock);
1803                 len = sprintf(buf, "%hu\n", conn->portal_port);
1804                 spin_unlock_bh(&conn->session->lock);
1805                 break;
1806         case ISCSI_PARAM_CONN_ADDRESS:
1807                 spin_lock_bh(&conn->session->lock);
1808                 len = sprintf(buf, "%s\n", conn->portal_address);
1809                 spin_unlock_bh(&conn->session->lock);
1810                 break;
1811         default:
1812                 return iscsi_conn_get_param(cls_conn, param, buf);
1813         }
1814
1815         return len;
1816 }
1817
1818 static int
1819 iscsi_tcp_host_get_param(struct Scsi_Host *shost, enum iscsi_host_param param,
1820                          char *buf)
1821 {
1822         struct iscsi_session *session = iscsi_hostdata(shost->hostdata);
1823         int len;
1824
1825         switch (param) {
1826         case ISCSI_HOST_PARAM_IPADDRESS:
1827                 spin_lock_bh(&session->lock);
1828                 if (!session->leadconn)
1829                         len = -ENODEV;
1830                 else
1831                         len = sprintf(buf, "%s\n",
1832                                      session->leadconn->local_address);
1833                 spin_unlock_bh(&session->lock);
1834                 break;
1835         default:
1836                 return iscsi_host_get_param(shost, param, buf);
1837         }
1838         return len;
1839 }
1840
1841 static void
1842 iscsi_conn_get_stats(struct iscsi_cls_conn *cls_conn, struct iscsi_stats *stats)
1843 {
1844         struct iscsi_conn *conn = cls_conn->dd_data;
1845         struct iscsi_tcp_conn *tcp_conn = conn->dd_data;
1846
1847         stats->txdata_octets = conn->txdata_octets;
1848         stats->rxdata_octets = conn->rxdata_octets;
1849         stats->scsicmd_pdus = conn->scsicmd_pdus_cnt;
1850         stats->dataout_pdus = conn->dataout_pdus_cnt;
1851         stats->scsirsp_pdus = conn->scsirsp_pdus_cnt;
1852         stats->datain_pdus = conn->datain_pdus_cnt;
1853         stats->r2t_pdus = conn->r2t_pdus_cnt;
1854         stats->tmfcmd_pdus = conn->tmfcmd_pdus_cnt;
1855         stats->tmfrsp_pdus = conn->tmfrsp_pdus_cnt;
1856         stats->custom_length = 3;
1857         strcpy(stats->custom[0].desc, "tx_sendpage_failures");
1858         stats->custom[0].value = tcp_conn->sendpage_failures_cnt;
1859         strcpy(stats->custom[1].desc, "rx_discontiguous_hdr");
1860         stats->custom[1].value = tcp_conn->discontiguous_hdr_cnt;
1861         strcpy(stats->custom[2].desc, "eh_abort_cnt");
1862         stats->custom[2].value = conn->eh_abort_cnt;
1863 }
1864
1865 static struct iscsi_cls_session *
1866 iscsi_tcp_session_create(struct iscsi_transport *iscsit,
1867                          struct scsi_transport_template *scsit,
1868                          uint16_t cmds_max, uint16_t qdepth,
1869                          uint32_t initial_cmdsn, uint32_t *hostno)
1870 {
1871         struct iscsi_cls_session *cls_session;
1872         struct iscsi_session *session;
1873         uint32_t hn;
1874         int cmd_i;
1875
1876         cls_session = iscsi_session_setup(iscsit, scsit, cmds_max, qdepth,
1877                                          sizeof(struct iscsi_tcp_cmd_task),
1878                                          sizeof(struct iscsi_tcp_mgmt_task),
1879                                          initial_cmdsn, &hn);
1880         if (!cls_session)
1881                 return NULL;
1882         *hostno = hn;
1883
1884         session = class_to_transport_session(cls_session);
1885         for (cmd_i = 0; cmd_i < session->cmds_max; cmd_i++) {
1886                 struct iscsi_cmd_task *ctask = session->cmds[cmd_i];
1887                 struct iscsi_tcp_cmd_task *tcp_ctask = ctask->dd_data;
1888
1889                 ctask->hdr = &tcp_ctask->hdr.cmd_hdr;
1890                 ctask->hdr_max = sizeof(tcp_ctask->hdr) - ISCSI_DIGEST_SIZE;
1891         }
1892
1893         for (cmd_i = 0; cmd_i < session->mgmtpool_max; cmd_i++) {
1894                 struct iscsi_mgmt_task *mtask = session->mgmt_cmds[cmd_i];
1895                 struct iscsi_tcp_mgmt_task *tcp_mtask = mtask->dd_data;
1896
1897                 mtask->hdr = (struct iscsi_hdr *) &tcp_mtask->hdr;
1898         }
1899
1900         if (iscsi_r2tpool_alloc(class_to_transport_session(cls_session)))
1901                 goto r2tpool_alloc_fail;
1902
1903         return cls_session;
1904
1905 r2tpool_alloc_fail:
1906         iscsi_session_teardown(cls_session);
1907         return NULL;
1908 }
1909
1910 static void iscsi_tcp_session_destroy(struct iscsi_cls_session *cls_session)
1911 {
1912         iscsi_r2tpool_free(class_to_transport_session(cls_session));
1913         iscsi_session_teardown(cls_session);
1914 }
1915
1916 static int iscsi_tcp_slave_configure(struct scsi_device *sdev)
1917 {
1918         blk_queue_bounce_limit(sdev->request_queue, BLK_BOUNCE_ANY);
1919         blk_queue_dma_alignment(sdev->request_queue, 0);
1920         return 0;
1921 }
1922
1923 static struct scsi_host_template iscsi_sht = {
1924         .module                 = THIS_MODULE,
1925         .name                   = "iSCSI Initiator over TCP/IP",
1926         .queuecommand           = iscsi_queuecommand,
1927         .change_queue_depth     = iscsi_change_queue_depth,
1928         .can_queue              = ISCSI_DEF_XMIT_CMDS_MAX - 1,
1929         .sg_tablesize           = 4096,
1930         .max_sectors            = 0xFFFF,
1931         .cmd_per_lun            = ISCSI_DEF_CMD_PER_LUN,
1932         .eh_abort_handler       = iscsi_eh_abort,
1933         .eh_device_reset_handler= iscsi_eh_device_reset,
1934         .eh_host_reset_handler  = iscsi_eh_host_reset,
1935         .use_clustering         = DISABLE_CLUSTERING,
1936         .use_sg_chaining        = ENABLE_SG_CHAINING,
1937         .slave_configure        = iscsi_tcp_slave_configure,
1938         .proc_name              = "iscsi_tcp",
1939         .this_id                = -1,
1940 };
1941
1942 static struct iscsi_transport iscsi_tcp_transport = {
1943         .owner                  = THIS_MODULE,
1944         .name                   = "tcp",
1945         .caps                   = CAP_RECOVERY_L0 | CAP_MULTI_R2T | CAP_HDRDGST
1946                                   | CAP_DATADGST,
1947         .param_mask             = ISCSI_MAX_RECV_DLENGTH |
1948                                   ISCSI_MAX_XMIT_DLENGTH |
1949                                   ISCSI_HDRDGST_EN |
1950                                   ISCSI_DATADGST_EN |
1951                                   ISCSI_INITIAL_R2T_EN |
1952                                   ISCSI_MAX_R2T |
1953                                   ISCSI_IMM_DATA_EN |
1954                                   ISCSI_FIRST_BURST |
1955                                   ISCSI_MAX_BURST |
1956                                   ISCSI_PDU_INORDER_EN |
1957                                   ISCSI_DATASEQ_INORDER_EN |
1958                                   ISCSI_ERL |
1959                                   ISCSI_CONN_PORT |
1960                                   ISCSI_CONN_ADDRESS |
1961                                   ISCSI_EXP_STATSN |
1962                                   ISCSI_PERSISTENT_PORT |
1963                                   ISCSI_PERSISTENT_ADDRESS |
1964                                   ISCSI_TARGET_NAME | ISCSI_TPGT |
1965                                   ISCSI_USERNAME | ISCSI_PASSWORD |
1966                                   ISCSI_USERNAME_IN | ISCSI_PASSWORD_IN |
1967                                   ISCSI_FAST_ABORT | ISCSI_ABORT_TMO |
1968                                   ISCSI_LU_RESET_TMO |
1969                                   ISCSI_PING_TMO | ISCSI_RECV_TMO,
1970         .host_param_mask        = ISCSI_HOST_HWADDRESS | ISCSI_HOST_IPADDRESS |
1971                                   ISCSI_HOST_INITIATOR_NAME |
1972                                   ISCSI_HOST_NETDEV_NAME,
1973         .host_template          = &iscsi_sht,
1974         .conndata_size          = sizeof(struct iscsi_conn),
1975         .max_conn               = 1,
1976         .max_cmd_len            = 16,
1977         /* session management */
1978         .create_session         = iscsi_tcp_session_create,
1979         .destroy_session        = iscsi_tcp_session_destroy,
1980         /* connection management */
1981         .create_conn            = iscsi_tcp_conn_create,
1982         .bind_conn              = iscsi_tcp_conn_bind,
1983         .destroy_conn           = iscsi_tcp_conn_destroy,
1984         .set_param              = iscsi_conn_set_param,
1985         .get_conn_param         = iscsi_tcp_conn_get_param,
1986         .get_session_param      = iscsi_session_get_param,
1987         .start_conn             = iscsi_conn_start,
1988         .stop_conn              = iscsi_tcp_conn_stop,
1989         /* iscsi host params */
1990         .get_host_param         = iscsi_tcp_host_get_param,
1991         .set_host_param         = iscsi_host_set_param,
1992         /* IO */
1993         .send_pdu               = iscsi_conn_send_pdu,
1994         .get_stats              = iscsi_conn_get_stats,
1995         .init_cmd_task          = iscsi_tcp_ctask_init,
1996         .init_mgmt_task         = iscsi_tcp_mtask_init,
1997         .xmit_cmd_task          = iscsi_tcp_ctask_xmit,
1998         .xmit_mgmt_task         = iscsi_tcp_mtask_xmit,
1999         .cleanup_cmd_task       = iscsi_tcp_cleanup_ctask,
2000         /* recovery */
2001         .session_recovery_timedout = iscsi_session_recovery_timedout,
2002 };
2003
2004 static int __init
2005 iscsi_tcp_init(void)
2006 {
2007         if (iscsi_max_lun < 1) {
2008                 printk(KERN_ERR "iscsi_tcp: Invalid max_lun value of %u\n",
2009                        iscsi_max_lun);
2010                 return -EINVAL;
2011         }
2012         iscsi_tcp_transport.max_lun = iscsi_max_lun;
2013
2014         if (!iscsi_register_transport(&iscsi_tcp_transport))
2015                 return -ENODEV;
2016
2017         return 0;
2018 }
2019
2020 static void __exit
2021 iscsi_tcp_exit(void)
2022 {
2023         iscsi_unregister_transport(&iscsi_tcp_transport);
2024 }
2025
2026 module_init(iscsi_tcp_init);
2027 module_exit(iscsi_tcp_exit);